Original Research

Hemochromatosis is an autosomal recessive abnormality of iron regulation that results in excessive intestinal absorption and cellular deposition of iron.¹ Although hemochromatosis was once thought to be rare, many screening studies have established that it is among the most common inherited metabolic abnormalities.^2-6 The College of American Pathologists has recommended population-based screening for hemochromatosis with the use of the serum transferrin saturation level.⁷ Although prevalent in the population, hemochromatosis is rarely diagnosed.⁸ The pathologic iron accumulation resulting in hemochromatosis affects many organs including the liver, pancreas, and heart.^9-12 Because primary hemochromatosis is a common comorbid condition with diabetes,^13-15 most work on the relation between hemochromatosis and diabetes has focused on screening patients with diabetes for hemochromatosis.^16,17

Clinical reviews have stated that, because diabetes is a serious complication of hemochromatosis, screening patients without diabetes for hemochromatosis might be a useful strategy to decrease the likelihood that they will develop diabetes.^15,18-20 However, there are few primary data to support this contention. There is some evidence to indicate that hemochromatosis has the pathogenic features of impaired insulin secretion and insulin resistance due to iron accumulation in the liver.²¹ One study indicated that, in individuals with hemochromatosis but neither cirrhosis nor diabetes (n = 7), phlebotomy treatment normalizes serum ferritin levels, acute insulin response to glucose, and glucose tolerance.²² In patients with hemochromatosis and newly diagnosed diabetes, phlebotomy did not affect glucose tolerance or insulin resistance. In a nationally representative cohort, we examined the likelihood that patients with an elevated serum transferrin saturation rate but no current diagnosis of diabetes would develop diabetes during 20 years of follow-up.

Methods

This retrospective cohort study followed individuals without a diagnosis of diabetes, aged 25 to 74 years at the time of the index interview. We used the National Health and Nutrition Examination Survey I (1971–1974; NHANES I) merged with the NHANES I Epidemiologic Followup Study (1992; NHEFS).

The NHANES I was conducted between 1971 and 1975 and allowed for representative estimates of the non-institutionalized civilian US population. The NHEFS is a national longitudinal study of individuals assessed at the NHANES I baseline. The NHEFS initial population included the 14,407 participants who were 25 to 74 years of age when first examined in NHANES I. More than 98% of the individuals in the initial NHANES I cohort were traced and supplied data for the NHEFS.

The follow-up information was gathered in 3 ways. Surviving subjects were interviewed. If the subject was deceased or alive but incapacitated, a slightly modified version of the subject questionnaire was administered to a proxy respondent. For individuals who had died in the period between the NHANES I index interview and the follow-up interview, information from death certificates was recorded. A total of 1,681 proxy respondents was interviewed in the NHEFS.

Serum transferrin saturation was measured in the original NHANES I. We defined elevated serum transferrin saturation as greater than 45%, greater than 50%, greater than 55%, greater than 60%, or greater than 62%. All of these cutoff values had previously been proposed or used in population-based studies of elevated serum transferrin saturation.^4,5,23

Diabetes was operationalized as a positive response to the question, Has a doctor ever told you that you have diabetes? This question was asked in the original NHANES I and in each wave of the follow-up survey (1982–1984, 1986, 1987, and 1992). For individuals who could not participate, proxy respondents were queried. In terms of individuals who died before the follow-up survey, we operationalized the development of diabetes as an ICD-9 diagnosis of 250.XX for underlying cause of death or any of the 20 other diagnoses listed on the death certificate.

We also assessed risk factors for diabetes available in the NHANES I, including obesity represented by a body mass index greater than 27 kg/m 2 , race, sex, age, physician diagnosis of hypertension, and total serum cholesterol above 240 mg/dL as a way to increase our understanding of diabetes mellitus as a consequence of hemochromatosis.

Our index sample was limited to men and women 25 to 74 years of age in the NHANES I, who had a serum transferrin saturation rate recorded in the NHANES I, did not have diabetes at the initial index interview, and had information on the development of diabetes (n = 9724).

Data analysis

We used sampling weights to calculate prevalence estimates for the civilian noninstitutionalized US population. Because of the complex sampling design of the survey, we performed all analyses with SUDAAN.²⁴

We initially computed unadjusted estimates of the likelihood of development of diabetes for different levels of elevated serum transferrin saturation between 1971 and 1974. We attempted to compute analyses for serum transferrin saturation levels of 60% and 62%, but the number of people who developed diabetes with those levels was so small (n

Because we could not determine whether individuals with elevated serum transferrin saturation received treatment for hemochromatosis during the time of the study, we computed a series of analyses assuming that different proportions of individuals received treatment during the time frame. Some national evidence has suggested that few individuals are diagnosed with hemochromatosis. In fact, in the 1996, 1997, and 1998 National Ambulatory Medical Care Surveys, there were 7 visits for hemochromatosis of 64,001 total evaluated visits (0.01% of visits).⁸ We recomputed the adjusted odds ratios for the samples of people with transferrin saturation rates greater than 45% and 50% after randomly selecting 10% of each group and treating those individuals as though they were undergoing therapeutic phlebotomies.

Results

Table 1 presents the characteristics of the population with baseline characteristics and characteristics measured in the follow-up data collection. A substantial proportion of the adult population had elevated serum transferrin saturation greater than 45%, 50%, and 55%. The incidence of diagnosed diabetes in the cohort was 10.2%.

Among individuals with serum transferrin saturation levels greater than 45% at the NHANES I baseline, 8.9% developed diagnosed diabetes compared with 10.3% who did not have elevated serum transferrin saturation (P = .44). Similarly, among individuals with serum transferrin saturation levels greater than 50% at the NHANES I baseline, 8.1% developed diagnosed diabetes compared with 10.3% who did not have elevated serum transferrin saturation (P = .34); of individuals with transferrin saturation levels greater than 55%, 7.5% developed diabetes compared with 10.2% of those without elevated serum transferrin saturation (P = .38). Table 2 indicates that individuals with elevated transferrin saturation levels are not significantly more likely to develop diagnosed diabetes than individuals without elevated serum transferrin saturation. The lack of a significant relation is present in unadjusted and adjusted analyses.

When we reestimated the models assuming that 10% of the population with elevated serum transferrin saturation (> 45%) were successfully treated, individuals with elevated serum transferrin saturation were not significantly more likely to develop diagnosed diabetes than individuals without elevated serum transferrin saturation. This relation held in unadjusted and adjusted analyses. When we assumed that 10% of the population with elevated serum transferrin saturation at greater than 50% and greater than 55% were successfully treated, individuals with elevated serum transferrin saturation were not significantly more likely to develop diagnosed diabetes than individuals without elevated serum transferrin saturation. These relations remained consistent in unadjusted and adjusted analyses.

TABLE 1

Baseline characteristics of the population collected in National Health and Nutrition Examination Survey I

TABLE 2

Unadjusted and adjusted odds of developing diabetes with an elevated serum transferrin saturation level

Discussion

The findings of this study call into question the commonly held assumption that there is a causative relation between the presence of hemochromatosis and the subsequent development of diabetes mellitus. Although diabetes is a common comorbid condition with hemochromatosis,^13,14 this may be due to the fact that both conditions are relatively common, not that one disease leads to the development of the other. In this longitudinal analysis, even when examining the likelihood of developing diabetes at different levels of transferrin saturation, the findings suggested that hemochromatosis does not lead to diabetes.

Could the findings of the current study be explained by the fact that people were treated for hemochromatosis, thus reducing the subsequent development of complications such as diabetes? This seems unlikely because few people with hemochromatosis are routinely identified, and even fewer are treated on a chronic basis. On the contrary, the phenomenon that few people with hemochromatosis are diagnosed and treated is the rationale for recent recommendations for screening asymptomatic persons. Further, in unadjusted and adjusted analyses of the current study, people with elevated transferrin saturation were no more likely to develop diabetes than people without elevated transferrin saturation, even after assuming that 10% of the population with elevated transferrin saturation (> 45%, > 50%, and > 55%) were successfully treated.

The findings of this study have implications for whether screening for hemochromatosis is worthwhile, assuming that prevention of diabetes is a goal. Hemochromatosis has many characteristics that make it attractive for screening: the disorder is common, it posseses a long asymptomatic phase, a simple screening test is available, and treatment is effective. To be a reasonable candidate for screening, the condition also needs to cause substantial morbidity or mortality, and treatment in the asymptomatic phase should be more effective than treatment initiated after the onset of symptoms.²⁵ The findings of this study suggested that screening for and treatment of hemochromatosis are not worthwhile as a way to prevent diabetes. However, the relation between hemochromatosis, treatment, and the development of cirrhosis or hepatocellular carcinoma may warrant screening for hemochromatosis. There is preliminary evidence from an observational study that diagnosing patients with hemochromatosis in the precirrhotic stage and treating them with phlebotomy results in a normal life expectancy, whereas those diagnosed with hemochromatosis and cirrhosis have a shortened life expectancy and a high risk of liver cancer, even when iron depletion has been achieved.¹⁰

Our study had several limitations. First, the estimate from the NHANES I was based on an elevated serum transferrin saturation level. This is an appropriate first step in a diagnosis of hemochromatosis. Some investigators have recommended that elevated serum transferrin levels should be confirmed with a second fasting level or an elevated ferritin level.^26,27 Further, we did not have access to liver biopsy data, which is considered the gold standard for diagnosing hemochromatosis. Thus, a single elevated transferrin saturation level may have resulted in overestimates of the prevalence of hemochromatosis in the study population. Second, the estimate also might have been affected by the use of the lower levels of serum transferrin saturation (> 45%, > 50%, or > 55%). Although using a more stringent level, eg, greater than 60%, might have strengthened the conclusions, so few people with this level developed diabetes that we could not accurately make a population estimate.

In summary, diabetes does not seem to be a likely complication of hemochromatosis as indicated by the presence of an elevated serum transferrin saturation. Consequently, cost-effectiveness models of screening for hereditary hemochromatosis may need to be reevaluated.

Hemochromatosis is an autosomal recessive abnormality of iron regulation that results in excessive intestinal absorption and cellular deposition of iron.¹ Although hemochromatosis was once thought to be rare, many screening studies have established that it is among the most common inherited metabolic abnormalities.^2-6 The College of American Pathologists has recommended population-based screening for hemochromatosis with the use of the serum transferrin saturation level.⁷ Although prevalent in the population, hemochromatosis is rarely diagnosed.⁸ The pathologic iron accumulation resulting in hemochromatosis affects many organs including the liver, pancreas, and heart.^9-12 Because primary hemochromatosis is a common comorbid condition with diabetes,^13-15 most work on the relation between hemochromatosis and diabetes has focused on screening patients with diabetes for hemochromatosis.^16,17

Clinical reviews have stated that, because diabetes is a serious complication of hemochromatosis, screening patients without diabetes for hemochromatosis might be a useful strategy to decrease the likelihood that they will develop diabetes.^15,18-20 However, there are few primary data to support this contention. There is some evidence to indicate that hemochromatosis has the pathogenic features of impaired insulin secretion and insulin resistance due to iron accumulation in the liver.²¹ One study indicated that, in individuals with hemochromatosis but neither cirrhosis nor diabetes (n = 7), phlebotomy treatment normalizes serum ferritin levels, acute insulin response to glucose, and glucose tolerance.²² In patients with hemochromatosis and newly diagnosed diabetes, phlebotomy did not affect glucose tolerance or insulin resistance. In a nationally representative cohort, we examined the likelihood that patients with an elevated serum transferrin saturation rate but no current diagnosis of diabetes would develop diabetes during 20 years of follow-up.

Methods

This retrospective cohort study followed individuals without a diagnosis of diabetes, aged 25 to 74 years at the time of the index interview. We used the National Health and Nutrition Examination Survey I (1971–1974; NHANES I) merged with the NHANES I Epidemiologic Followup Study (1992; NHEFS).

The NHANES I was conducted between 1971 and 1975 and allowed for representative estimates of the non-institutionalized civilian US population. The NHEFS is a national longitudinal study of individuals assessed at the NHANES I baseline. The NHEFS initial population included the 14,407 participants who were 25 to 74 years of age when first examined in NHANES I. More than 98% of the individuals in the initial NHANES I cohort were traced and supplied data for the NHEFS.

The follow-up information was gathered in 3 ways. Surviving subjects were interviewed. If the subject was deceased or alive but incapacitated, a slightly modified version of the subject questionnaire was administered to a proxy respondent. For individuals who had died in the period between the NHANES I index interview and the follow-up interview, information from death certificates was recorded. A total of 1,681 proxy respondents was interviewed in the NHEFS.

Serum transferrin saturation was measured in the original NHANES I. We defined elevated serum transferrin saturation as greater than 45%, greater than 50%, greater than 55%, greater than 60%, or greater than 62%. All of these cutoff values had previously been proposed or used in population-based studies of elevated serum transferrin saturation.^4,5,23

Diabetes was operationalized as a positive response to the question, Has a doctor ever told you that you have diabetes? This question was asked in the original NHANES I and in each wave of the follow-up survey (1982–1984, 1986, 1987, and 1992). For individuals who could not participate, proxy respondents were queried. In terms of individuals who died before the follow-up survey, we operationalized the development of diabetes as an ICD-9 diagnosis of 250.XX for underlying cause of death or any of the 20 other diagnoses listed on the death certificate.

We also assessed risk factors for diabetes available in the NHANES I, including obesity represented by a body mass index greater than 27 kg/m 2 , race, sex, age, physician diagnosis of hypertension, and total serum cholesterol above 240 mg/dL as a way to increase our understanding of diabetes mellitus as a consequence of hemochromatosis.

Our index sample was limited to men and women 25 to 74 years of age in the NHANES I, who had a serum transferrin saturation rate recorded in the NHANES I, did not have diabetes at the initial index interview, and had information on the development of diabetes (n = 9724).

Data analysis

We used sampling weights to calculate prevalence estimates for the civilian noninstitutionalized US population. Because of the complex sampling design of the survey, we performed all analyses with SUDAAN.²⁴

We initially computed unadjusted estimates of the likelihood of development of diabetes for different levels of elevated serum transferrin saturation between 1971 and 1974. We attempted to compute analyses for serum transferrin saturation levels of 60% and 62%, but the number of people who developed diabetes with those levels was so small (n

Because we could not determine whether individuals with elevated serum transferrin saturation received treatment for hemochromatosis during the time of the study, we computed a series of analyses assuming that different proportions of individuals received treatment during the time frame. Some national evidence has suggested that few individuals are diagnosed with hemochromatosis. In fact, in the 1996, 1997, and 1998 National Ambulatory Medical Care Surveys, there were 7 visits for hemochromatosis of 64,001 total evaluated visits (0.01% of visits).⁸ We recomputed the adjusted odds ratios for the samples of people with transferrin saturation rates greater than 45% and 50% after randomly selecting 10% of each group and treating those individuals as though they were undergoing therapeutic phlebotomies.

Results

Table 1 presents the characteristics of the population with baseline characteristics and characteristics measured in the follow-up data collection. A substantial proportion of the adult population had elevated serum transferrin saturation greater than 45%, 50%, and 55%. The incidence of diagnosed diabetes in the cohort was 10.2%.

Among individuals with serum transferrin saturation levels greater than 45% at the NHANES I baseline, 8.9% developed diagnosed diabetes compared with 10.3% who did not have elevated serum transferrin saturation (P = .44). Similarly, among individuals with serum transferrin saturation levels greater than 50% at the NHANES I baseline, 8.1% developed diagnosed diabetes compared with 10.3% who did not have elevated serum transferrin saturation (P = .34); of individuals with transferrin saturation levels greater than 55%, 7.5% developed diabetes compared with 10.2% of those without elevated serum transferrin saturation (P = .38). Table 2 indicates that individuals with elevated transferrin saturation levels are not significantly more likely to develop diagnosed diabetes than individuals without elevated serum transferrin saturation. The lack of a significant relation is present in unadjusted and adjusted analyses.

When we reestimated the models assuming that 10% of the population with elevated serum transferrin saturation (> 45%) were successfully treated, individuals with elevated serum transferrin saturation were not significantly more likely to develop diagnosed diabetes than individuals without elevated serum transferrin saturation. This relation held in unadjusted and adjusted analyses. When we assumed that 10% of the population with elevated serum transferrin saturation at greater than 50% and greater than 55% were successfully treated, individuals with elevated serum transferrin saturation were not significantly more likely to develop diagnosed diabetes than individuals without elevated serum transferrin saturation. These relations remained consistent in unadjusted and adjusted analyses.

TABLE 1

Baseline characteristics of the population collected in National Health and Nutrition Examination Survey I

TABLE 2

Unadjusted and adjusted odds of developing diabetes with an elevated serum transferrin saturation level

Discussion

The findings of this study call into question the commonly held assumption that there is a causative relation between the presence of hemochromatosis and the subsequent development of diabetes mellitus. Although diabetes is a common comorbid condition with hemochromatosis,^13,14 this may be due to the fact that both conditions are relatively common, not that one disease leads to the development of the other. In this longitudinal analysis, even when examining the likelihood of developing diabetes at different levels of transferrin saturation, the findings suggested that hemochromatosis does not lead to diabetes.

Could the findings of the current study be explained by the fact that people were treated for hemochromatosis, thus reducing the subsequent development of complications such as diabetes? This seems unlikely because few people with hemochromatosis are routinely identified, and even fewer are treated on a chronic basis. On the contrary, the phenomenon that few people with hemochromatosis are diagnosed and treated is the rationale for recent recommendations for screening asymptomatic persons. Further, in unadjusted and adjusted analyses of the current study, people with elevated transferrin saturation were no more likely to develop diabetes than people without elevated transferrin saturation, even after assuming that 10% of the population with elevated transferrin saturation (> 45%, > 50%, and > 55%) were successfully treated.

The findings of this study have implications for whether screening for hemochromatosis is worthwhile, assuming that prevention of diabetes is a goal. Hemochromatosis has many characteristics that make it attractive for screening: the disorder is common, it posseses a long asymptomatic phase, a simple screening test is available, and treatment is effective. To be a reasonable candidate for screening, the condition also needs to cause substantial morbidity or mortality, and treatment in the asymptomatic phase should be more effective than treatment initiated after the onset of symptoms.²⁵ The findings of this study suggested that screening for and treatment of hemochromatosis are not worthwhile as a way to prevent diabetes. However, the relation between hemochromatosis, treatment, and the development of cirrhosis or hepatocellular carcinoma may warrant screening for hemochromatosis. There is preliminary evidence from an observational study that diagnosing patients with hemochromatosis in the precirrhotic stage and treating them with phlebotomy results in a normal life expectancy, whereas those diagnosed with hemochromatosis and cirrhosis have a shortened life expectancy and a high risk of liver cancer, even when iron depletion has been achieved.¹⁰

Our study had several limitations. First, the estimate from the NHANES I was based on an elevated serum transferrin saturation level. This is an appropriate first step in a diagnosis of hemochromatosis. Some investigators have recommended that elevated serum transferrin levels should be confirmed with a second fasting level or an elevated ferritin level.^26,27 Further, we did not have access to liver biopsy data, which is considered the gold standard for diagnosing hemochromatosis. Thus, a single elevated transferrin saturation level may have resulted in overestimates of the prevalence of hemochromatosis in the study population. Second, the estimate also might have been affected by the use of the lower levels of serum transferrin saturation (> 45%, > 50%, or > 55%). Although using a more stringent level, eg, greater than 60%, might have strengthened the conclusions, so few people with this level developed diabetes that we could not accurately make a population estimate.

In summary, diabetes does not seem to be a likely complication of hemochromatosis as indicated by the presence of an elevated serum transferrin saturation. Consequently, cost-effectiveness models of screening for hereditary hemochromatosis may need to be reevaluated.

Hemochromatosis is an autosomal recessive abnormality of iron regulation that results in excessive intestinal absorption and cellular deposition of iron.¹ Although hemochromatosis was once thought to be rare, many screening studies have established that it is among the most common inherited metabolic abnormalities.^2-6 The College of American Pathologists has recommended population-based screening for hemochromatosis with the use of the serum transferrin saturation level.⁷ Although prevalent in the population, hemochromatosis is rarely diagnosed.⁸ The pathologic iron accumulation resulting in hemochromatosis affects many organs including the liver, pancreas, and heart.^9-12 Because primary hemochromatosis is a common comorbid condition with diabetes,^13-15 most work on the relation between hemochromatosis and diabetes has focused on screening patients with diabetes for hemochromatosis.^16,17

Clinical reviews have stated that, because diabetes is a serious complication of hemochromatosis, screening patients without diabetes for hemochromatosis might be a useful strategy to decrease the likelihood that they will develop diabetes.^15,18-20 However, there are few primary data to support this contention. There is some evidence to indicate that hemochromatosis has the pathogenic features of impaired insulin secretion and insulin resistance due to iron accumulation in the liver.²¹ One study indicated that, in individuals with hemochromatosis but neither cirrhosis nor diabetes (n = 7), phlebotomy treatment normalizes serum ferritin levels, acute insulin response to glucose, and glucose tolerance.²² In patients with hemochromatosis and newly diagnosed diabetes, phlebotomy did not affect glucose tolerance or insulin resistance. In a nationally representative cohort, we examined the likelihood that patients with an elevated serum transferrin saturation rate but no current diagnosis of diabetes would develop diabetes during 20 years of follow-up.

Methods

This retrospective cohort study followed individuals without a diagnosis of diabetes, aged 25 to 74 years at the time of the index interview. We used the National Health and Nutrition Examination Survey I (1971–1974; NHANES I) merged with the NHANES I Epidemiologic Followup Study (1992; NHEFS).

The NHANES I was conducted between 1971 and 1975 and allowed for representative estimates of the non-institutionalized civilian US population. The NHEFS is a national longitudinal study of individuals assessed at the NHANES I baseline. The NHEFS initial population included the 14,407 participants who were 25 to 74 years of age when first examined in NHANES I. More than 98% of the individuals in the initial NHANES I cohort were traced and supplied data for the NHEFS.

The follow-up information was gathered in 3 ways. Surviving subjects were interviewed. If the subject was deceased or alive but incapacitated, a slightly modified version of the subject questionnaire was administered to a proxy respondent. For individuals who had died in the period between the NHANES I index interview and the follow-up interview, information from death certificates was recorded. A total of 1,681 proxy respondents was interviewed in the NHEFS.

Serum transferrin saturation was measured in the original NHANES I. We defined elevated serum transferrin saturation as greater than 45%, greater than 50%, greater than 55%, greater than 60%, or greater than 62%. All of these cutoff values had previously been proposed or used in population-based studies of elevated serum transferrin saturation.^4,5,23

Diabetes was operationalized as a positive response to the question, Has a doctor ever told you that you have diabetes? This question was asked in the original NHANES I and in each wave of the follow-up survey (1982–1984, 1986, 1987, and 1992). For individuals who could not participate, proxy respondents were queried. In terms of individuals who died before the follow-up survey, we operationalized the development of diabetes as an ICD-9 diagnosis of 250.XX for underlying cause of death or any of the 20 other diagnoses listed on the death certificate.

We also assessed risk factors for diabetes available in the NHANES I, including obesity represented by a body mass index greater than 27 kg/m 2 , race, sex, age, physician diagnosis of hypertension, and total serum cholesterol above 240 mg/dL as a way to increase our understanding of diabetes mellitus as a consequence of hemochromatosis.

Our index sample was limited to men and women 25 to 74 years of age in the NHANES I, who had a serum transferrin saturation rate recorded in the NHANES I, did not have diabetes at the initial index interview, and had information on the development of diabetes (n = 9724).

Data analysis

We used sampling weights to calculate prevalence estimates for the civilian noninstitutionalized US population. Because of the complex sampling design of the survey, we performed all analyses with SUDAAN.²⁴

We initially computed unadjusted estimates of the likelihood of development of diabetes for different levels of elevated serum transferrin saturation between 1971 and 1974. We attempted to compute analyses for serum transferrin saturation levels of 60% and 62%, but the number of people who developed diabetes with those levels was so small (n

Because we could not determine whether individuals with elevated serum transferrin saturation received treatment for hemochromatosis during the time of the study, we computed a series of analyses assuming that different proportions of individuals received treatment during the time frame. Some national evidence has suggested that few individuals are diagnosed with hemochromatosis. In fact, in the 1996, 1997, and 1998 National Ambulatory Medical Care Surveys, there were 7 visits for hemochromatosis of 64,001 total evaluated visits (0.01% of visits).⁸ We recomputed the adjusted odds ratios for the samples of people with transferrin saturation rates greater than 45% and 50% after randomly selecting 10% of each group and treating those individuals as though they were undergoing therapeutic phlebotomies.

Results

Table 1 presents the characteristics of the population with baseline characteristics and characteristics measured in the follow-up data collection. A substantial proportion of the adult population had elevated serum transferrin saturation greater than 45%, 50%, and 55%. The incidence of diagnosed diabetes in the cohort was 10.2%.

Among individuals with serum transferrin saturation levels greater than 45% at the NHANES I baseline, 8.9% developed diagnosed diabetes compared with 10.3% who did not have elevated serum transferrin saturation (P = .44). Similarly, among individuals with serum transferrin saturation levels greater than 50% at the NHANES I baseline, 8.1% developed diagnosed diabetes compared with 10.3% who did not have elevated serum transferrin saturation (P = .34); of individuals with transferrin saturation levels greater than 55%, 7.5% developed diabetes compared with 10.2% of those without elevated serum transferrin saturation (P = .38). Table 2 indicates that individuals with elevated transferrin saturation levels are not significantly more likely to develop diagnosed diabetes than individuals without elevated serum transferrin saturation. The lack of a significant relation is present in unadjusted and adjusted analyses.

When we reestimated the models assuming that 10% of the population with elevated serum transferrin saturation (> 45%) were successfully treated, individuals with elevated serum transferrin saturation were not significantly more likely to develop diagnosed diabetes than individuals without elevated serum transferrin saturation. This relation held in unadjusted and adjusted analyses. When we assumed that 10% of the population with elevated serum transferrin saturation at greater than 50% and greater than 55% were successfully treated, individuals with elevated serum transferrin saturation were not significantly more likely to develop diagnosed diabetes than individuals without elevated serum transferrin saturation. These relations remained consistent in unadjusted and adjusted analyses.

TABLE 1

Baseline characteristics of the population collected in National Health and Nutrition Examination Survey I

TABLE 2

Unadjusted and adjusted odds of developing diabetes with an elevated serum transferrin saturation level

Discussion

The findings of this study call into question the commonly held assumption that there is a causative relation between the presence of hemochromatosis and the subsequent development of diabetes mellitus. Although diabetes is a common comorbid condition with hemochromatosis,^13,14 this may be due to the fact that both conditions are relatively common, not that one disease leads to the development of the other. In this longitudinal analysis, even when examining the likelihood of developing diabetes at different levels of transferrin saturation, the findings suggested that hemochromatosis does not lead to diabetes.

Could the findings of the current study be explained by the fact that people were treated for hemochromatosis, thus reducing the subsequent development of complications such as diabetes? This seems unlikely because few people with hemochromatosis are routinely identified, and even fewer are treated on a chronic basis. On the contrary, the phenomenon that few people with hemochromatosis are diagnosed and treated is the rationale for recent recommendations for screening asymptomatic persons. Further, in unadjusted and adjusted analyses of the current study, people with elevated transferrin saturation were no more likely to develop diabetes than people without elevated transferrin saturation, even after assuming that 10% of the population with elevated transferrin saturation (> 45%, > 50%, and > 55%) were successfully treated.

The findings of this study have implications for whether screening for hemochromatosis is worthwhile, assuming that prevention of diabetes is a goal. Hemochromatosis has many characteristics that make it attractive for screening: the disorder is common, it posseses a long asymptomatic phase, a simple screening test is available, and treatment is effective. To be a reasonable candidate for screening, the condition also needs to cause substantial morbidity or mortality, and treatment in the asymptomatic phase should be more effective than treatment initiated after the onset of symptoms.²⁵ The findings of this study suggested that screening for and treatment of hemochromatosis are not worthwhile as a way to prevent diabetes. However, the relation between hemochromatosis, treatment, and the development of cirrhosis or hepatocellular carcinoma may warrant screening for hemochromatosis. There is preliminary evidence from an observational study that diagnosing patients with hemochromatosis in the precirrhotic stage and treating them with phlebotomy results in a normal life expectancy, whereas those diagnosed with hemochromatosis and cirrhosis have a shortened life expectancy and a high risk of liver cancer, even when iron depletion has been achieved.¹⁰

Our study had several limitations. First, the estimate from the NHANES I was based on an elevated serum transferrin saturation level. This is an appropriate first step in a diagnosis of hemochromatosis. Some investigators have recommended that elevated serum transferrin levels should be confirmed with a second fasting level or an elevated ferritin level.^26,27 Further, we did not have access to liver biopsy data, which is considered the gold standard for diagnosing hemochromatosis. Thus, a single elevated transferrin saturation level may have resulted in overestimates of the prevalence of hemochromatosis in the study population. Second, the estimate also might have been affected by the use of the lower levels of serum transferrin saturation (> 45%, > 50%, or > 55%). Although using a more stringent level, eg, greater than 60%, might have strengthened the conclusions, so few people with this level developed diabetes that we could not accurately make a population estimate.

In summary, diabetes does not seem to be a likely complication of hemochromatosis as indicated by the presence of an elevated serum transferrin saturation. Consequently, cost-effectiveness models of screening for hereditary hemochromatosis may need to be reevaluated.

Urinary incontinence is a common problem among primary care patients.¹ In recent years, patient perceptions of quality of life have become increasingly important in the evaluation of health conditions and their treatment.² Specific instruments have been developed for the evaluation of the health-related quality of life of women reporting urinary incontinence.^3,4 Wagner and colleagues developed a self-report quality of life measure specific to urinary incontinence (the Incontinence Quality of Life Instrument; I-QOL) that could be used as an outcome measure in clinical trials and patient care centers.⁵ The developers tested the instrument on a sample of 62 subjects and reported that the I-QOL was more sensitive than a generic instrument, such as the Short Form 36 (SF-36), in detecting differences between levels of self-perceived incontinence severity. In a follow-up study,⁶ incontinent women (141 with stress and 147 with mixed urinary incontinence) completed the I-QOL and comparative instruments at screening, pretreatment, and 4 follow-up visits during participation in a randomized trial assessing the efficacy of a medication for incontinence. Those investigators reported that, in the clinical trial, the I-QOL proved to be valid, reproducible, and responsive to treatment for urinary incontinence in women.

The aim of our study was to measure the prevalence of urinary incontinence and its impact on quality of life, in a population of community dwelling women. We selected the I-QOL and the Short Form 12 (SF-12)⁷ as specific and generic instruments, respectively, for the measurement of quality of life among incontinent women. In this report we describe the performance of the I-QOL in the community setting.

METHODS

Subjects

The subjects were women 45 years or older who attended either of 2 participating family medicine clinics in the city of Hamilton, Ontario. This was a postal survey using a modified Dillman method.⁸ The Dillman method calls for 3 mailings and a reminder postcard. Because of budgetary constraints we planned an initial mailing, followed by a reminder postcard and a second mailing to nonrespondents. Our budget permitted a sample size of about 1000, so that questionnaires were sent to all eligible patients attending the smaller clinic and to a random sample drawn from the roster of the larger clinic.

The initial mailing was sent to 1082 women. Ninety envelopes were returned undeliverable; in addition, 115 women returned the survey but selected the option of nonparticipation. The final response rate was 605 (61%) of 992.

Survey questions

Two questions inquired about the presence of incontinence: (1) “During the past month have you ever experienced urine loss (wet yourself) when coughing, laughing, or doing some other activity?” (2) “During the past month have you ever had to pee and then wet yourself before getting to the toilet?” We classified incontinence as “stress incontinence” if respondents replied “yes” to question 1; as “urge incontinence” if they replied “yes” to question 2; and as “mixed incontinence” if they responded “yes” to both. In addition, we asked, “Is wetting yourself a problem that interferes with your day-to-day activities or bothers you in other ways?” We also inquired about the number of daytime and nighttime leaking episodes in an average week and the amount of wetness (underwear or pad only, outer clothing, urine runs down legs, or pools on floor). The survey included 2 health-related quality of life instruments, the I-QOL and the SF-12,⁷ a generic instrument.

Statistical analysis

Handling missing responses. Missing data is a common problem in survey research. Until recently, the only methods widely available for analyzing incomplete data focused on “removing” the missing values by ignoring subjects with incomplete information or by substituting plausible values (eg, means or regression predictions) for the missing items. These ad hoc methods, although simple to implement, have serious drawbacks,⁹ including the potential introduction of bias. In the past 2 decades, substantial progress has been made in developing statistical procedures for missing data. In an incomplete data set, the observed values provide indirect evidence about the likely values of the unobserved data and one can use the available data to make estimates of the values of the missing data. Because any one estimate is uncertain, one may repeat the process a number of times and then average over these estimates for the missing values in the statistical analysis. Rubin developed the paradigm of multiple imputation, which carries out the averaging via simulation; each missing value is replaced by plausible values drawn from their predictive distribution.¹⁰ The variation among the number of imputations reflects the uncertainty with which the missing values can be predicted from the observed ones. After performing identical analyses on each data set, the results are combined according to simple rules to produce overall estimates and standard errors that reflect missing-data uncertainty. We used the publicly available software program, Amelia,¹¹ to generate 5 data sets containing imputed values for those subjects with missing values for I-QOL. Variables in the imputation model included age, education, type of incontinence, number of incontinent episodes, I-QOL, and scores on the physical and mental components of the SF-12.

Regression analysis. Logistic regression was used to investigate factors associated with incomplete responses to the I-QOL instrument. The relations between the I-QOL scores and predictor variables were modeled with multiple linear regression. Before the regression, we used generalized additive models,12 a method that uses data smoothers to graphically display the pattern of relationships, to explore the shape of nonlinear relations, and suggest linearizing transformations. Model checking included an analysis of residuals.¹³

An important measure of the impact of incontinence is whether or not subjects consider their incontinence to be a “problem.” To compare the performance of the I-QOL and generic quality of life measures in discriminating between women who found and did not find their incontinence to be a problem, we computed the area under receiver operating characteristic curves.¹⁴ Computations were done with the Stata statistical package.¹⁵

RESULTS

Prevalence of incontinence

Of the 605 respondents, 310 (51%) reported urinary incontinence in the month before the survey. Table 1 shows the distribution of the respondents by age and incontinence status. Among our respondents, the prevalence of incontinence decreased slightly with age, a trend of borderline statistical significance (P = .08). Most surveys have reported that the prevalence of incontinence increases with age. We have no explanation for why this was not the case in our survey.

Incomplete responses to the I-QOL

The I-QOL is a questionnaire instrument with 22 items and the score is computed from all items. In such a situation, missing responses might be an important problem, leading to the reduction of sample size and study power. In our survey, 11 subjects (3.5%) replied to none of the I-QOL items; however, at least 1 item was missing for 49 other subjects. Thirty subjects missed 1 item, 4 missed 2 items, 3 missed 3 items, and 12 each missed 4 to 20 items. The number of missing responses for the 22 questions ranged from 5% to 7%, with the exception of the statement, “I worry about having sex,” for which the missing rate was 13%. Even though the missing rate for individual items was no higher than 13%, only 250 (80.7%) of I-QOL scores were complete.

Table 2 shows those variables significantly associated with incomplete response in a logistic regression model. Older women and women who had not graduated from high school were less likely to return a completed instrument, as were women who reported only urge incontinence. In addition, women who reported that incontinence was a problem were less likely to complete all the questions. These associations suggested that omitting women with incomplete responses from the analyses of I-QOL might introduce selection bias. We managed this problem by making use of the methods of multiple imputation.

Associations of I-QOL with incontinence factors

The I-QOL is scored in the range of 0 to 100, with lower scores indicating lower quality of life. The mean value of I-QOL among respondents was 83 (range, 15-100). We anticipated that the I-QOL scores would vary systematically with incontinence-related factors and planned to investigate these relationships with a linear regression model. We suspected, however, that the relationships between I-QOL and the number of daytime and nighttime episodes of incontinence might be nonlinear. We thus explored these relations with nonparametric regression methods, and Figure 1 shows the relations as estimated with a generalized additive model.¹² The method of generalized additive models is a computer-intensive one that makes no prior assumptions about the shape of the relations between the outcome and the explanatory variables. It fits smooth, arbitrarily shaped functions to the data by a method that is a generalization of the “moving average.” Figure 1 shows that the I-QOL decreased as the number of incontinence episodes increased, but then reached a plateau when the frequency of occurrence was about 1 episode per day. In other words, quality of life worsened as the number of incontinence episodes increased from less than once per week to once per day. However, once incontinence was occurring daily, there was no further decrease in the perceived quality of life as incontinence episodes became more frequent than once per day. We found that these plateau relations could be satisfactorily modeled with logarithmic transformations of the number of incontinence episodes; ie, log (number of incontinence episodes per week + 1), where the 1 has been added to avoid log(0), which is undefined mathematically.

The following candidate variables were chosen for inclusion in an initial regression model for the I-QOL scores: age, type of incontinence (stress, urge, mixed), incontinence is perceived as a “problem” (yes/no), log-transformed number of daytime and nighttime incontinence episodes, amount of wetting (wets outer clothing or runs down leg/wets pad or underpants only), self-reported health status (excellent to poor), and education. In this initial model, there was no significant association between I-QOL scores and age, type of incontinence, or education. These variables thus were not retained in the final model, which is shown in Table 3. The model provided a good fit to the data (F_8,190 = 43.1; P < .0001; R² = .64). In addition to presenting the results of the model that excluded subjects with incomplete I-QOL scores, Table 3 shows the results of the model that included these subjects by using imputed values. There were only minor differences between the models. The I-QOL score decreased as the number of daytime and nighttime incontinence episodes increased and as the amount of urine loss increased. After controlling for other variables, the I-QOL averaged 12 points fewer among those women who considered their incontinence problematic. Further, there was a strong relation between the I-QOL and self-reported general health status.

DISCUSSION

Urinary incontinence is common among women in the community, but loss of bladder control is perceived quite differently by various respondents. In a survey of 36,000 Americans with incontinence, Jeter and Wagner reported that 17% described their incontinence as a major problem with important social implications,¹ but the rest described it as a relatively minor problem with limited impact on their respective lifestyles. Self-administered quality of life instruments are valuable in measuring the impact of incontinence on the lives of subjects, of identifying subjects among whom interventions might have a beneficial impact on quality of life, and in following the natural history of incontinence and its treatment.

One of these instruments, the I-QOL, has been used in the clinical trial setting. We were impressed with reports of its performance and selected it for use in a postal survey of community dwelling patients of 2 family medicine clinics. It was immediately apparent that the 22-item length of the instrument posed problems in a postal survey because, even though the missing rate for individual items was no higher than 13%, only 80.7% of I-QOL scores were complete. We found that older, less educated subjects were more likely to return incomplete questionnaires. We did not contact subjects to obtain responses to the omitted questions. Rather, we chose to make use of responses to other questions to impute the missing I-QOL scores. Although the differential response rates suggested the possibility of selection bias, there was in fact little difference in the coefficients of regression models when using complete versus complete plus imputed data.

As reported by Patrick and colleagues in the clinical trial context,⁶ we found that the I-QOL score correlates strongly with physical measures of the extent of incontinence, including the number of incontinence episodes and the amount of wetting. Similar to those researchers, we found no relation between the I-QOL and age, education, or the type of incontinence. These are desirable properties for a condition-specific quality of life instrument; it is responsive to the impact of incontinence on the quality of life and not to “nuisance” variables. Another desirable characteristic was that the I-QOL correlates strongly with the statement, “wetting is a problem.” Figure 2 shows the relations between the I-QOL score and the probability that a respondent would state that incontinence is a problem. To understand Figure 2, consider that each subject will agree or disagree with the statement, “wetting is a problem.” Each subject thus appears on the graph at a probability level of 0 or 1. Because there is considerable overlap of the data points on the display, the point for each subject has been “jittered,” ie, given a random amount of vertical displacement to better display the density of data points in relation to the I-QOL. The curve on the plot is the fit of a “loess” smoother to the data points.16 This smoother is, in essence, a running weighted average of the proportion of subjects who reported that wetting is a problem. When the I-QOL score was low, there was a high probability that incontinence would be seen as a problem. The I-QOL score at which 50% of women found incontinence to be a problem was about 77, and the probability declined rapidly after that.

There were significant correlations between the I-QOL score and the subscales of the SF-12 generic quality of life instrument (r = .49 for the mental component of the SF-12 and r = .33 for the physical component). We found, however, that the I-QOL was more sensitive than the generic instruments in identifying subjects who considered incontinence to be a problem. This is shown in the receiver operating characteristic curves displayed in Figure 3. The area under the I-QOL receiver operating characteristic curve was 0.84, compared with 0.63 for the medical component scale and 0.58 for the physical component scale of the SF-12. It is reassuring that a condition-specific instrument performs better than a generic one.

We note some limitations to this study. Estimates of prevalence in the published epidemiologic data on urinary incontinence often vary widely. These large variations are derived mainly from the wide range in definitions used, differences in survey setting, and methodology used. The overall prevalence rate of 51% reported in our survey is high. Further, the prevalence of incontinence among our respondents appeared to decrease slightly with age. One possible explanation and limitation of our study is that respondents self-selected to participate. The other limitation is that older, incontinent women likely were underrepresented in our sample due to institutionalization and survivor bias. The end result is that the prevalence of incontinence might have been overestimated for younger respondents (incontinent younger women more likely to respond) and underestimated for the older respondents (incontinent older women less likely to dwell in communities). The choice of our sampling frame and thus our ability to generalize the findings to other settings also might be viewed as problematic. This was not intended to be a community sample. We were interested in the impact of urinary incontinence on women attending family medicine clinics because these women are our patients and we have the opportunity to intervene to improve their quality of life.

Another limitation was that our questionnaire was available only in English. This might have affected our response rate and generalizability of these findings.

Despite these limitations, we believe the I-QOL is a useful instrument for the investigation of incontinence-related quality of life in the community and the clinical trial setting. We have begun to use the I-QOL among patients attending an incontinence clinic. We have found it to be well received and plan to report on its performance in this setting in the near future.

Urinary incontinence is a common problem among primary care patients.¹ In recent years, patient perceptions of quality of life have become increasingly important in the evaluation of health conditions and their treatment.² Specific instruments have been developed for the evaluation of the health-related quality of life of women reporting urinary incontinence.^3,4 Wagner and colleagues developed a self-report quality of life measure specific to urinary incontinence (the Incontinence Quality of Life Instrument; I-QOL) that could be used as an outcome measure in clinical trials and patient care centers.⁵ The developers tested the instrument on a sample of 62 subjects and reported that the I-QOL was more sensitive than a generic instrument, such as the Short Form 36 (SF-36), in detecting differences between levels of self-perceived incontinence severity. In a follow-up study,⁶ incontinent women (141 with stress and 147 with mixed urinary incontinence) completed the I-QOL and comparative instruments at screening, pretreatment, and 4 follow-up visits during participation in a randomized trial assessing the efficacy of a medication for incontinence. Those investigators reported that, in the clinical trial, the I-QOL proved to be valid, reproducible, and responsive to treatment for urinary incontinence in women.

The aim of our study was to measure the prevalence of urinary incontinence and its impact on quality of life, in a population of community dwelling women. We selected the I-QOL and the Short Form 12 (SF-12)⁷ as specific and generic instruments, respectively, for the measurement of quality of life among incontinent women. In this report we describe the performance of the I-QOL in the community setting.

METHODS

Subjects

The subjects were women 45 years or older who attended either of 2 participating family medicine clinics in the city of Hamilton, Ontario. This was a postal survey using a modified Dillman method.⁸ The Dillman method calls for 3 mailings and a reminder postcard. Because of budgetary constraints we planned an initial mailing, followed by a reminder postcard and a second mailing to nonrespondents. Our budget permitted a sample size of about 1000, so that questionnaires were sent to all eligible patients attending the smaller clinic and to a random sample drawn from the roster of the larger clinic.

The initial mailing was sent to 1082 women. Ninety envelopes were returned undeliverable; in addition, 115 women returned the survey but selected the option of nonparticipation. The final response rate was 605 (61%) of 992.

Survey questions

Two questions inquired about the presence of incontinence: (1) “During the past month have you ever experienced urine loss (wet yourself) when coughing, laughing, or doing some other activity?” (2) “During the past month have you ever had to pee and then wet yourself before getting to the toilet?” We classified incontinence as “stress incontinence” if respondents replied “yes” to question 1; as “urge incontinence” if they replied “yes” to question 2; and as “mixed incontinence” if they responded “yes” to both. In addition, we asked, “Is wetting yourself a problem that interferes with your day-to-day activities or bothers you in other ways?” We also inquired about the number of daytime and nighttime leaking episodes in an average week and the amount of wetness (underwear or pad only, outer clothing, urine runs down legs, or pools on floor). The survey included 2 health-related quality of life instruments, the I-QOL and the SF-12,⁷ a generic instrument.

Statistical analysis

Handling missing responses. Missing data is a common problem in survey research. Until recently, the only methods widely available for analyzing incomplete data focused on “removing” the missing values by ignoring subjects with incomplete information or by substituting plausible values (eg, means or regression predictions) for the missing items. These ad hoc methods, although simple to implement, have serious drawbacks,⁹ including the potential introduction of bias. In the past 2 decades, substantial progress has been made in developing statistical procedures for missing data. In an incomplete data set, the observed values provide indirect evidence about the likely values of the unobserved data and one can use the available data to make estimates of the values of the missing data. Because any one estimate is uncertain, one may repeat the process a number of times and then average over these estimates for the missing values in the statistical analysis. Rubin developed the paradigm of multiple imputation, which carries out the averaging via simulation; each missing value is replaced by plausible values drawn from their predictive distribution.¹⁰ The variation among the number of imputations reflects the uncertainty with which the missing values can be predicted from the observed ones. After performing identical analyses on each data set, the results are combined according to simple rules to produce overall estimates and standard errors that reflect missing-data uncertainty. We used the publicly available software program, Amelia,¹¹ to generate 5 data sets containing imputed values for those subjects with missing values for I-QOL. Variables in the imputation model included age, education, type of incontinence, number of incontinent episodes, I-QOL, and scores on the physical and mental components of the SF-12.

Regression analysis. Logistic regression was used to investigate factors associated with incomplete responses to the I-QOL instrument. The relations between the I-QOL scores and predictor variables were modeled with multiple linear regression. Before the regression, we used generalized additive models,12 a method that uses data smoothers to graphically display the pattern of relationships, to explore the shape of nonlinear relations, and suggest linearizing transformations. Model checking included an analysis of residuals.¹³

An important measure of the impact of incontinence is whether or not subjects consider their incontinence to be a “problem.” To compare the performance of the I-QOL and generic quality of life measures in discriminating between women who found and did not find their incontinence to be a problem, we computed the area under receiver operating characteristic curves.¹⁴ Computations were done with the Stata statistical package.¹⁵

RESULTS

Prevalence of incontinence

Of the 605 respondents, 310 (51%) reported urinary incontinence in the month before the survey. Table 1 shows the distribution of the respondents by age and incontinence status. Among our respondents, the prevalence of incontinence decreased slightly with age, a trend of borderline statistical significance (P = .08). Most surveys have reported that the prevalence of incontinence increases with age. We have no explanation for why this was not the case in our survey.

Incomplete responses to the I-QOL

The I-QOL is a questionnaire instrument with 22 items and the score is computed from all items. In such a situation, missing responses might be an important problem, leading to the reduction of sample size and study power. In our survey, 11 subjects (3.5%) replied to none of the I-QOL items; however, at least 1 item was missing for 49 other subjects. Thirty subjects missed 1 item, 4 missed 2 items, 3 missed 3 items, and 12 each missed 4 to 20 items. The number of missing responses for the 22 questions ranged from 5% to 7%, with the exception of the statement, “I worry about having sex,” for which the missing rate was 13%. Even though the missing rate for individual items was no higher than 13%, only 250 (80.7%) of I-QOL scores were complete.

Table 2 shows those variables significantly associated with incomplete response in a logistic regression model. Older women and women who had not graduated from high school were less likely to return a completed instrument, as were women who reported only urge incontinence. In addition, women who reported that incontinence was a problem were less likely to complete all the questions. These associations suggested that omitting women with incomplete responses from the analyses of I-QOL might introduce selection bias. We managed this problem by making use of the methods of multiple imputation.

Associations of I-QOL with incontinence factors

The I-QOL is scored in the range of 0 to 100, with lower scores indicating lower quality of life. The mean value of I-QOL among respondents was 83 (range, 15-100). We anticipated that the I-QOL scores would vary systematically with incontinence-related factors and planned to investigate these relationships with a linear regression model. We suspected, however, that the relationships between I-QOL and the number of daytime and nighttime episodes of incontinence might be nonlinear. We thus explored these relations with nonparametric regression methods, and Figure 1 shows the relations as estimated with a generalized additive model.¹² The method of generalized additive models is a computer-intensive one that makes no prior assumptions about the shape of the relations between the outcome and the explanatory variables. It fits smooth, arbitrarily shaped functions to the data by a method that is a generalization of the “moving average.” Figure 1 shows that the I-QOL decreased as the number of incontinence episodes increased, but then reached a plateau when the frequency of occurrence was about 1 episode per day. In other words, quality of life worsened as the number of incontinence episodes increased from less than once per week to once per day. However, once incontinence was occurring daily, there was no further decrease in the perceived quality of life as incontinence episodes became more frequent than once per day. We found that these plateau relations could be satisfactorily modeled with logarithmic transformations of the number of incontinence episodes; ie, log (number of incontinence episodes per week + 1), where the 1 has been added to avoid log(0), which is undefined mathematically.

The following candidate variables were chosen for inclusion in an initial regression model for the I-QOL scores: age, type of incontinence (stress, urge, mixed), incontinence is perceived as a “problem” (yes/no), log-transformed number of daytime and nighttime incontinence episodes, amount of wetting (wets outer clothing or runs down leg/wets pad or underpants only), self-reported health status (excellent to poor), and education. In this initial model, there was no significant association between I-QOL scores and age, type of incontinence, or education. These variables thus were not retained in the final model, which is shown in Table 3. The model provided a good fit to the data (F_8,190 = 43.1; P < .0001; R² = .64). In addition to presenting the results of the model that excluded subjects with incomplete I-QOL scores, Table 3 shows the results of the model that included these subjects by using imputed values. There were only minor differences between the models. The I-QOL score decreased as the number of daytime and nighttime incontinence episodes increased and as the amount of urine loss increased. After controlling for other variables, the I-QOL averaged 12 points fewer among those women who considered their incontinence problematic. Further, there was a strong relation between the I-QOL and self-reported general health status.

DISCUSSION

Urinary incontinence is common among women in the community, but loss of bladder control is perceived quite differently by various respondents. In a survey of 36,000 Americans with incontinence, Jeter and Wagner reported that 17% described their incontinence as a major problem with important social implications,¹ but the rest described it as a relatively minor problem with limited impact on their respective lifestyles. Self-administered quality of life instruments are valuable in measuring the impact of incontinence on the lives of subjects, of identifying subjects among whom interventions might have a beneficial impact on quality of life, and in following the natural history of incontinence and its treatment.

One of these instruments, the I-QOL, has been used in the clinical trial setting. We were impressed with reports of its performance and selected it for use in a postal survey of community dwelling patients of 2 family medicine clinics. It was immediately apparent that the 22-item length of the instrument posed problems in a postal survey because, even though the missing rate for individual items was no higher than 13%, only 80.7% of I-QOL scores were complete. We found that older, less educated subjects were more likely to return incomplete questionnaires. We did not contact subjects to obtain responses to the omitted questions. Rather, we chose to make use of responses to other questions to impute the missing I-QOL scores. Although the differential response rates suggested the possibility of selection bias, there was in fact little difference in the coefficients of regression models when using complete versus complete plus imputed data.

As reported by Patrick and colleagues in the clinical trial context,⁶ we found that the I-QOL score correlates strongly with physical measures of the extent of incontinence, including the number of incontinence episodes and the amount of wetting. Similar to those researchers, we found no relation between the I-QOL and age, education, or the type of incontinence. These are desirable properties for a condition-specific quality of life instrument; it is responsive to the impact of incontinence on the quality of life and not to “nuisance” variables. Another desirable characteristic was that the I-QOL correlates strongly with the statement, “wetting is a problem.” Figure 2 shows the relations between the I-QOL score and the probability that a respondent would state that incontinence is a problem. To understand Figure 2, consider that each subject will agree or disagree with the statement, “wetting is a problem.” Each subject thus appears on the graph at a probability level of 0 or 1. Because there is considerable overlap of the data points on the display, the point for each subject has been “jittered,” ie, given a random amount of vertical displacement to better display the density of data points in relation to the I-QOL. The curve on the plot is the fit of a “loess” smoother to the data points.16 This smoother is, in essence, a running weighted average of the proportion of subjects who reported that wetting is a problem. When the I-QOL score was low, there was a high probability that incontinence would be seen as a problem. The I-QOL score at which 50% of women found incontinence to be a problem was about 77, and the probability declined rapidly after that.

There were significant correlations between the I-QOL score and the subscales of the SF-12 generic quality of life instrument (r = .49 for the mental component of the SF-12 and r = .33 for the physical component). We found, however, that the I-QOL was more sensitive than the generic instruments in identifying subjects who considered incontinence to be a problem. This is shown in the receiver operating characteristic curves displayed in Figure 3. The area under the I-QOL receiver operating characteristic curve was 0.84, compared with 0.63 for the medical component scale and 0.58 for the physical component scale of the SF-12. It is reassuring that a condition-specific instrument performs better than a generic one.

We note some limitations to this study. Estimates of prevalence in the published epidemiologic data on urinary incontinence often vary widely. These large variations are derived mainly from the wide range in definitions used, differences in survey setting, and methodology used. The overall prevalence rate of 51% reported in our survey is high. Further, the prevalence of incontinence among our respondents appeared to decrease slightly with age. One possible explanation and limitation of our study is that respondents self-selected to participate. The other limitation is that older, incontinent women likely were underrepresented in our sample due to institutionalization and survivor bias. The end result is that the prevalence of incontinence might have been overestimated for younger respondents (incontinent younger women more likely to respond) and underestimated for the older respondents (incontinent older women less likely to dwell in communities). The choice of our sampling frame and thus our ability to generalize the findings to other settings also might be viewed as problematic. This was not intended to be a community sample. We were interested in the impact of urinary incontinence on women attending family medicine clinics because these women are our patients and we have the opportunity to intervene to improve their quality of life.

Another limitation was that our questionnaire was available only in English. This might have affected our response rate and generalizability of these findings.

Despite these limitations, we believe the I-QOL is a useful instrument for the investigation of incontinence-related quality of life in the community and the clinical trial setting. We have begun to use the I-QOL among patients attending an incontinence clinic. We have found it to be well received and plan to report on its performance in this setting in the near future.

Urinary incontinence is a common problem among primary care patients.¹ In recent years, patient perceptions of quality of life have become increasingly important in the evaluation of health conditions and their treatment.² Specific instruments have been developed for the evaluation of the health-related quality of life of women reporting urinary incontinence.^3,4 Wagner and colleagues developed a self-report quality of life measure specific to urinary incontinence (the Incontinence Quality of Life Instrument; I-QOL) that could be used as an outcome measure in clinical trials and patient care centers.⁵ The developers tested the instrument on a sample of 62 subjects and reported that the I-QOL was more sensitive than a generic instrument, such as the Short Form 36 (SF-36), in detecting differences between levels of self-perceived incontinence severity. In a follow-up study,⁶ incontinent women (141 with stress and 147 with mixed urinary incontinence) completed the I-QOL and comparative instruments at screening, pretreatment, and 4 follow-up visits during participation in a randomized trial assessing the efficacy of a medication for incontinence. Those investigators reported that, in the clinical trial, the I-QOL proved to be valid, reproducible, and responsive to treatment for urinary incontinence in women.

The aim of our study was to measure the prevalence of urinary incontinence and its impact on quality of life, in a population of community dwelling women. We selected the I-QOL and the Short Form 12 (SF-12)⁷ as specific and generic instruments, respectively, for the measurement of quality of life among incontinent women. In this report we describe the performance of the I-QOL in the community setting.

METHODS

Subjects

The subjects were women 45 years or older who attended either of 2 participating family medicine clinics in the city of Hamilton, Ontario. This was a postal survey using a modified Dillman method.⁸ The Dillman method calls for 3 mailings and a reminder postcard. Because of budgetary constraints we planned an initial mailing, followed by a reminder postcard and a second mailing to nonrespondents. Our budget permitted a sample size of about 1000, so that questionnaires were sent to all eligible patients attending the smaller clinic and to a random sample drawn from the roster of the larger clinic.

The initial mailing was sent to 1082 women. Ninety envelopes were returned undeliverable; in addition, 115 women returned the survey but selected the option of nonparticipation. The final response rate was 605 (61%) of 992.

Survey questions

Two questions inquired about the presence of incontinence: (1) “During the past month have you ever experienced urine loss (wet yourself) when coughing, laughing, or doing some other activity?” (2) “During the past month have you ever had to pee and then wet yourself before getting to the toilet?” We classified incontinence as “stress incontinence” if respondents replied “yes” to question 1; as “urge incontinence” if they replied “yes” to question 2; and as “mixed incontinence” if they responded “yes” to both. In addition, we asked, “Is wetting yourself a problem that interferes with your day-to-day activities or bothers you in other ways?” We also inquired about the number of daytime and nighttime leaking episodes in an average week and the amount of wetness (underwear or pad only, outer clothing, urine runs down legs, or pools on floor). The survey included 2 health-related quality of life instruments, the I-QOL and the SF-12,⁷ a generic instrument.

Statistical analysis

Handling missing responses. Missing data is a common problem in survey research. Until recently, the only methods widely available for analyzing incomplete data focused on “removing” the missing values by ignoring subjects with incomplete information or by substituting plausible values (eg, means or regression predictions) for the missing items. These ad hoc methods, although simple to implement, have serious drawbacks,⁹ including the potential introduction of bias. In the past 2 decades, substantial progress has been made in developing statistical procedures for missing data. In an incomplete data set, the observed values provide indirect evidence about the likely values of the unobserved data and one can use the available data to make estimates of the values of the missing data. Because any one estimate is uncertain, one may repeat the process a number of times and then average over these estimates for the missing values in the statistical analysis. Rubin developed the paradigm of multiple imputation, which carries out the averaging via simulation; each missing value is replaced by plausible values drawn from their predictive distribution.¹⁰ The variation among the number of imputations reflects the uncertainty with which the missing values can be predicted from the observed ones. After performing identical analyses on each data set, the results are combined according to simple rules to produce overall estimates and standard errors that reflect missing-data uncertainty. We used the publicly available software program, Amelia,¹¹ to generate 5 data sets containing imputed values for those subjects with missing values for I-QOL. Variables in the imputation model included age, education, type of incontinence, number of incontinent episodes, I-QOL, and scores on the physical and mental components of the SF-12.

Regression analysis. Logistic regression was used to investigate factors associated with incomplete responses to the I-QOL instrument. The relations between the I-QOL scores and predictor variables were modeled with multiple linear regression. Before the regression, we used generalized additive models,12 a method that uses data smoothers to graphically display the pattern of relationships, to explore the shape of nonlinear relations, and suggest linearizing transformations. Model checking included an analysis of residuals.¹³

An important measure of the impact of incontinence is whether or not subjects consider their incontinence to be a “problem.” To compare the performance of the I-QOL and generic quality of life measures in discriminating between women who found and did not find their incontinence to be a problem, we computed the area under receiver operating characteristic curves.¹⁴ Computations were done with the Stata statistical package.¹⁵

RESULTS

Prevalence of incontinence

Of the 605 respondents, 310 (51%) reported urinary incontinence in the month before the survey. Table 1 shows the distribution of the respondents by age and incontinence status. Among our respondents, the prevalence of incontinence decreased slightly with age, a trend of borderline statistical significance (P = .08). Most surveys have reported that the prevalence of incontinence increases with age. We have no explanation for why this was not the case in our survey.

Incomplete responses to the I-QOL

The I-QOL is a questionnaire instrument with 22 items and the score is computed from all items. In such a situation, missing responses might be an important problem, leading to the reduction of sample size and study power. In our survey, 11 subjects (3.5%) replied to none of the I-QOL items; however, at least 1 item was missing for 49 other subjects. Thirty subjects missed 1 item, 4 missed 2 items, 3 missed 3 items, and 12 each missed 4 to 20 items. The number of missing responses for the 22 questions ranged from 5% to 7%, with the exception of the statement, “I worry about having sex,” for which the missing rate was 13%. Even though the missing rate for individual items was no higher than 13%, only 250 (80.7%) of I-QOL scores were complete.

Table 2 shows those variables significantly associated with incomplete response in a logistic regression model. Older women and women who had not graduated from high school were less likely to return a completed instrument, as were women who reported only urge incontinence. In addition, women who reported that incontinence was a problem were less likely to complete all the questions. These associations suggested that omitting women with incomplete responses from the analyses of I-QOL might introduce selection bias. We managed this problem by making use of the methods of multiple imputation.

Associations of I-QOL with incontinence factors

The I-QOL is scored in the range of 0 to 100, with lower scores indicating lower quality of life. The mean value of I-QOL among respondents was 83 (range, 15-100). We anticipated that the I-QOL scores would vary systematically with incontinence-related factors and planned to investigate these relationships with a linear regression model. We suspected, however, that the relationships between I-QOL and the number of daytime and nighttime episodes of incontinence might be nonlinear. We thus explored these relations with nonparametric regression methods, and Figure 1 shows the relations as estimated with a generalized additive model.¹² The method of generalized additive models is a computer-intensive one that makes no prior assumptions about the shape of the relations between the outcome and the explanatory variables. It fits smooth, arbitrarily shaped functions to the data by a method that is a generalization of the “moving average.” Figure 1 shows that the I-QOL decreased as the number of incontinence episodes increased, but then reached a plateau when the frequency of occurrence was about 1 episode per day. In other words, quality of life worsened as the number of incontinence episodes increased from less than once per week to once per day. However, once incontinence was occurring daily, there was no further decrease in the perceived quality of life as incontinence episodes became more frequent than once per day. We found that these plateau relations could be satisfactorily modeled with logarithmic transformations of the number of incontinence episodes; ie, log (number of incontinence episodes per week + 1), where the 1 has been added to avoid log(0), which is undefined mathematically.

The following candidate variables were chosen for inclusion in an initial regression model for the I-QOL scores: age, type of incontinence (stress, urge, mixed), incontinence is perceived as a “problem” (yes/no), log-transformed number of daytime and nighttime incontinence episodes, amount of wetting (wets outer clothing or runs down leg/wets pad or underpants only), self-reported health status (excellent to poor), and education. In this initial model, there was no significant association between I-QOL scores and age, type of incontinence, or education. These variables thus were not retained in the final model, which is shown in Table 3. The model provided a good fit to the data (F_8,190 = 43.1; P < .0001; R² = .64). In addition to presenting the results of the model that excluded subjects with incomplete I-QOL scores, Table 3 shows the results of the model that included these subjects by using imputed values. There were only minor differences between the models. The I-QOL score decreased as the number of daytime and nighttime incontinence episodes increased and as the amount of urine loss increased. After controlling for other variables, the I-QOL averaged 12 points fewer among those women who considered their incontinence problematic. Further, there was a strong relation between the I-QOL and self-reported general health status.

DISCUSSION

Urinary incontinence is common among women in the community, but loss of bladder control is perceived quite differently by various respondents. In a survey of 36,000 Americans with incontinence, Jeter and Wagner reported that 17% described their incontinence as a major problem with important social implications,¹ but the rest described it as a relatively minor problem with limited impact on their respective lifestyles. Self-administered quality of life instruments are valuable in measuring the impact of incontinence on the lives of subjects, of identifying subjects among whom interventions might have a beneficial impact on quality of life, and in following the natural history of incontinence and its treatment.

One of these instruments, the I-QOL, has been used in the clinical trial setting. We were impressed with reports of its performance and selected it for use in a postal survey of community dwelling patients of 2 family medicine clinics. It was immediately apparent that the 22-item length of the instrument posed problems in a postal survey because, even though the missing rate for individual items was no higher than 13%, only 80.7% of I-QOL scores were complete. We found that older, less educated subjects were more likely to return incomplete questionnaires. We did not contact subjects to obtain responses to the omitted questions. Rather, we chose to make use of responses to other questions to impute the missing I-QOL scores. Although the differential response rates suggested the possibility of selection bias, there was in fact little difference in the coefficients of regression models when using complete versus complete plus imputed data.

As reported by Patrick and colleagues in the clinical trial context,⁶ we found that the I-QOL score correlates strongly with physical measures of the extent of incontinence, including the number of incontinence episodes and the amount of wetting. Similar to those researchers, we found no relation between the I-QOL and age, education, or the type of incontinence. These are desirable properties for a condition-specific quality of life instrument; it is responsive to the impact of incontinence on the quality of life and not to “nuisance” variables. Another desirable characteristic was that the I-QOL correlates strongly with the statement, “wetting is a problem.” Figure 2 shows the relations between the I-QOL score and the probability that a respondent would state that incontinence is a problem. To understand Figure 2, consider that each subject will agree or disagree with the statement, “wetting is a problem.” Each subject thus appears on the graph at a probability level of 0 or 1. Because there is considerable overlap of the data points on the display, the point for each subject has been “jittered,” ie, given a random amount of vertical displacement to better display the density of data points in relation to the I-QOL. The curve on the plot is the fit of a “loess” smoother to the data points.16 This smoother is, in essence, a running weighted average of the proportion of subjects who reported that wetting is a problem. When the I-QOL score was low, there was a high probability that incontinence would be seen as a problem. The I-QOL score at which 50% of women found incontinence to be a problem was about 77, and the probability declined rapidly after that.

There were significant correlations between the I-QOL score and the subscales of the SF-12 generic quality of life instrument (r = .49 for the mental component of the SF-12 and r = .33 for the physical component). We found, however, that the I-QOL was more sensitive than the generic instruments in identifying subjects who considered incontinence to be a problem. This is shown in the receiver operating characteristic curves displayed in Figure 3. The area under the I-QOL receiver operating characteristic curve was 0.84, compared with 0.63 for the medical component scale and 0.58 for the physical component scale of the SF-12. It is reassuring that a condition-specific instrument performs better than a generic one.

We note some limitations to this study. Estimates of prevalence in the published epidemiologic data on urinary incontinence often vary widely. These large variations are derived mainly from the wide range in definitions used, differences in survey setting, and methodology used. The overall prevalence rate of 51% reported in our survey is high. Further, the prevalence of incontinence among our respondents appeared to decrease slightly with age. One possible explanation and limitation of our study is that respondents self-selected to participate. The other limitation is that older, incontinent women likely were underrepresented in our sample due to institutionalization and survivor bias. The end result is that the prevalence of incontinence might have been overestimated for younger respondents (incontinent younger women more likely to respond) and underestimated for the older respondents (incontinent older women less likely to dwell in communities). The choice of our sampling frame and thus our ability to generalize the findings to other settings also might be viewed as problematic. This was not intended to be a community sample. We were interested in the impact of urinary incontinence on women attending family medicine clinics because these women are our patients and we have the opportunity to intervene to improve their quality of life.

Another limitation was that our questionnaire was available only in English. This might have affected our response rate and generalizability of these findings.

Despite these limitations, we believe the I-QOL is a useful instrument for the investigation of incontinence-related quality of life in the community and the clinical trial setting. We have begun to use the I-QOL among patients attending an incontinence clinic. We have found it to be well received and plan to report on its performance in this setting in the near future.

The National Cholesterol Education Program Adult Treatment Panel III Report (NCEP-ATP III) has identified low-density lipoprotein cholesterol (LDL-C) as the primary target of therapy.^1,2The Friedewald calculated LDL-C (C-LDL-C) is the preferred method^2,3 and is calculated with the following equation:

LDL-C =TC – HDL-C – TG/5

where TC is total cholesterol concentration, HDL-C is high-density lipoprotein cholesterol concentration, and TG is triglyceride concentration. The complete NCEP-ATP III report has indicated that methods to directly measure LDL-C (D-LDL-C) in the non-fasting state have been developed and will grow in use but require careful quality control.²Our VA hospital clinical laboratory is 1 of 10 hospitals in the South Central VA Health Care Network that routinely reports D-LDL-C rather than C-LDL-C levels to clinicians. Telephone calls to 4 other research and clinical laboratories found that all are using D-LDL-C to some extent.

D-LDL-C assays correlate variably with C-LDL-C measurements used in research studies.^4-17The purported advantages of such measurements are that fasting is not required and that D-LDL-C may be determined in patients with serum triglyceride levels greater than 400 mg/dL when the C-HDL-C and are less reliable. However, clinical trials demonstrating benefit of lowering LDL-C with drug therapy used the C-LDL-C.^18-22Only the recently reported Heart Protection Study used a non-fasting D-LDL-C.²³Thus, it is important in practicing evidence-based medicine to demonstrate that the D-LDL-C measurements are comparable to those of the C-LDL-C. The present study determined how the D-LDL-C correlated with C-LDL-C and how such a correlation would affect treatment decisions based on the NCEP-ATP III guidelines.

Methods

Data from all patients with a lipid panel during a single week were analyzed. Patients with triglyceride levels above 1000 mg/dL were excluded. Thirty-four patients with triglyceride levels between 400 and 1000 mg/dL were analyzed separately. A C-LDL-C was determined and compared with the D-LDL-C in all 464 patients. Total cholesterol, triglyceride, and HDL-C measurements were done with an autoanalyzer. D-LDL-C was measured with Sigma Diagnostics EZ LDL Cholesterol, procedure 358 (Sigma, St. Louis, MO). Linear regression was performed using Microsoft Excel (Microsoft Corporation, Redmond, WA).

Results

The samples in this study represented the expected distribution of LDL-C concentrations seen in a clinical practice of predominantly male veterans. Of the 464 patient samples with triglyceride levels below 400 mg/dL, the mean C-LDL-C was 123 mg/dL. Twenty-eight percent had a C-LDL-C below 100 mg/dL, 32% had a C-LDL-C of 100 to 129.9 mg/dL, 24% had a C-LDL-C of 130 to 159.9 mg/dL, 12% had a C-LDL-C of 160 to 189.9 mg/dL, and 4% had a C-LDL-C above 190 mg/dL.

The Figure shows the correlation between the C-LDL-C and D-LDL-C in all patients with triglyceride levels below 400 mg/dL. Although there is a strong correlation between the C-LDL-C and D-LDL-C (r = .86), the regression line does not go through 0. A C-LDL-C of 100 mg/dL or lower is the NCEP-ATP III goal for patients with known coronary heart disease (CHD) and other clinical forms of atherosclerotic disease, diabetes, or multiple risk factors that confer a 10-year risk for CHD greater than 20%.¹At this cutoff for C-LDL-C, the D-LDL-C derived from the regression line is 118 mg/dL. At a C-LDL-C of 160 mg/dL, the 2 values are comparable; at a C-LDL-C of 190 mg/dL, the D-LDL-C is slightly lower at 182 mg/dL. This is demonstrated graphically in the Figure by a dashed line indicating a perfect correlation between the 2 methods. The Figure also displays vertical and horizontal lines through an LDL-C of 100 mg/dL, the level above which drug therapy is likely to be started or increased in patients with CHD or CHD risk equivalents. This partition illustrates those patients who would require treatment when using the 100 mg/dL treatment goal by the C-LDL-C, the D-LDLC, neither, or both. This is also shown in the (Table, which shows the number of patients who would be treated with the LDL-C cutoffs for treatment recommended by NCEP-ATP III. At an LDL cutoff of 100 mg/dL, 60 patients (13% of total) would be treated with the D-LDL-C and not the C-LDL-C, whereas only 2 patients (<1%) would be treated with the C-LDL-C and not with the D-LDL-C. The results are similar when using a 130 mg/dL cutoff for treatment. Thus, treatment decisions based on the D-LDL-C results in many patients being treated who would not have been treated when using the C-LDL-C.

To determine whether triglyceride concentration influences treatment decisions by either method of LDL-C measurement, similar correlations and analyses were done on the data according to the following triglyceride groupings: <100 mg/dL, 100 to 199 mg/dL, 200 to 299 mg/dL, 300 to 399 mg/dL, and >400 mg/dL. This was further evaluated by plotting triglyceride vs D-LDL-C and triglyceride vs C-LDL-C (data not shown). Whereas the C-LDL-C showed no correlation with triglyceride, the DLDL-C showed a statistically significant correlation with triglyceride concentrations (r = .27), indicating that D-LDL-C increases at higher triglyceride levels. This suggested an influence of triglyceride on the D-LDL-C assay. This has been reported by others in 3 of 4 different D-LDL-C assays including the Sigma assay.¹⁵ However, alterations in treatment possibilities when using the D-LDL-C are present at all triglyceride concentrations.

FIGURE 1
Direct vs calculated LDL-C (mg/dL)

TABLE
Effect of LDL assay by LDL treatment cutoff*

Discussion

LDL-C has been identified in the NCEP-ATP III as the primary target of therapy. Treatment recommendations for high LDL-C are based on low, moderate, or high risk for CHD, with treatment goals of 160, 130, and 100 mg/dL, respectively.^1,2

These evidence-based recommendations rely on data from clinical trials demonstrating prevention of CHD events by lowering LDL-C, all of which, with the exception of the recently reported Heart Protection Study, used the CLDL-C.^18-23 Thus, important treatment decisions depend on this estimated LDL-C, and systematic deviations from the C-LDL-C will affect treatment decisions and cost.

The Sigma EZ LDL D-LDL-C assay in our hospital produces higher LDL-C levels than the C-LDL-C in a range of 100 to 160 mg/dL, the range of most common concern to clinicians. This results in inappropriate treatment or intensification in treatment according to the NCEP-ATP III guidelines. The DLDL-C was higher than the C-LDL-C at all triglyceride levels, but the error was greater for hypertriglyceridemia, the very situation for which it has been advocated.

Previous publications using a D-LDL-C assay have emphasized the correlation between the DLDL-C assay and research LDL-C determinations rather than the correlation with the C-LDL-C.^5-8 Other investigators have observed a similar tendency for higher D-LDL-C than C-LDL-C measurements at an LDL-C of 100 mg/dL¹⁷ and a positive bias at higher triglyceride levels.¹⁴ Although C-LDLC was often performed, data similar to those shown in the Figure, ie, the simple correlation between the D-LDL-C and C-LDL-C, have not been presented. Two very recent reviews have suggested caution in routinely implementing the D-LDL-C assays and pointed out the considerable variation from one assay to another.^14,15 Laboratories often change their assay method; in fact, our hospital laboratory has recently changed to a different DLDL-C method.

Physicians and institutions should be cautious about using a D-LDL-C method as a substitute for the C-LDL-C. First, it has not been standardized in large populations and, with the exception of the recent Heart Protection Study,²³ has not been used in large clinical trials demonstrating the benefits of lowering LDL-C. Although the C-LDL-C has been recommended by the NCEP-ATP III,² the Executive Summary of these guidelines did not address the method for measuring LDL-C.¹ Second, cost is increased from the additional therapy and performing the D-LDL-C assay. Third, the major reasons proposed for using a D-LDL-C assay (lack of need for a fasting specimen and usefulness at triglyceride > 400 mg/dL) may not be valid or relevant. Variation in the LDL-C due to hypertriglyceridemia occurs with the D-LDL-C. In addition, the NCEP-ATP III report emphasized triglyceride and recommended a fasting lipid panel including total cholesterol, triglycerides, and HDL-C.^1,2 One limitation of this study is the inclusion of predominantly male veterans. There may be populations, not considered in this study, that have an abnormal lipoprotein composition that significantly affects the C-LDL-C.

Conclusions

The C-LDL-C should remain the method of choice for LDL-C determinations because (1) this assay was used in clinical trials documenting the benefits of cholesterol-lowering therapy and (2) use of the D-LDL-C increases cost without evidence of benefit. Further studies are needed to standardize the direct LDL-C assays, and outcome trials using these assays need to be performed.

ACKNOWLEDGMENTS

This work was supported in part by the Biomedical Research Foundation of Arkansas and the Central Arkansas Veterans Healthcare System.

The National Cholesterol Education Program Adult Treatment Panel III Report (NCEP-ATP III) has identified low-density lipoprotein cholesterol (LDL-C) as the primary target of therapy.^1,2The Friedewald calculated LDL-C (C-LDL-C) is the preferred method^2,3 and is calculated with the following equation:

LDL-C =TC – HDL-C – TG/5

where TC is total cholesterol concentration, HDL-C is high-density lipoprotein cholesterol concentration, and TG is triglyceride concentration. The complete NCEP-ATP III report has indicated that methods to directly measure LDL-C (D-LDL-C) in the non-fasting state have been developed and will grow in use but require careful quality control.²Our VA hospital clinical laboratory is 1 of 10 hospitals in the South Central VA Health Care Network that routinely reports D-LDL-C rather than C-LDL-C levels to clinicians. Telephone calls to 4 other research and clinical laboratories found that all are using D-LDL-C to some extent.

D-LDL-C assays correlate variably with C-LDL-C measurements used in research studies.^4-17The purported advantages of such measurements are that fasting is not required and that D-LDL-C may be determined in patients with serum triglyceride levels greater than 400 mg/dL when the C-HDL-C and are less reliable. However, clinical trials demonstrating benefit of lowering LDL-C with drug therapy used the C-LDL-C.^18-22Only the recently reported Heart Protection Study used a non-fasting D-LDL-C.²³Thus, it is important in practicing evidence-based medicine to demonstrate that the D-LDL-C measurements are comparable to those of the C-LDL-C. The present study determined how the D-LDL-C correlated with C-LDL-C and how such a correlation would affect treatment decisions based on the NCEP-ATP III guidelines.

Methods

Data from all patients with a lipid panel during a single week were analyzed. Patients with triglyceride levels above 1000 mg/dL were excluded. Thirty-four patients with triglyceride levels between 400 and 1000 mg/dL were analyzed separately. A C-LDL-C was determined and compared with the D-LDL-C in all 464 patients. Total cholesterol, triglyceride, and HDL-C measurements were done with an autoanalyzer. D-LDL-C was measured with Sigma Diagnostics EZ LDL Cholesterol, procedure 358 (Sigma, St. Louis, MO). Linear regression was performed using Microsoft Excel (Microsoft Corporation, Redmond, WA).

Results

The samples in this study represented the expected distribution of LDL-C concentrations seen in a clinical practice of predominantly male veterans. Of the 464 patient samples with triglyceride levels below 400 mg/dL, the mean C-LDL-C was 123 mg/dL. Twenty-eight percent had a C-LDL-C below 100 mg/dL, 32% had a C-LDL-C of 100 to 129.9 mg/dL, 24% had a C-LDL-C of 130 to 159.9 mg/dL, 12% had a C-LDL-C of 160 to 189.9 mg/dL, and 4% had a C-LDL-C above 190 mg/dL.

The Figure shows the correlation between the C-LDL-C and D-LDL-C in all patients with triglyceride levels below 400 mg/dL. Although there is a strong correlation between the C-LDL-C and D-LDL-C (r = .86), the regression line does not go through 0. A C-LDL-C of 100 mg/dL or lower is the NCEP-ATP III goal for patients with known coronary heart disease (CHD) and other clinical forms of atherosclerotic disease, diabetes, or multiple risk factors that confer a 10-year risk for CHD greater than 20%.¹At this cutoff for C-LDL-C, the D-LDL-C derived from the regression line is 118 mg/dL. At a C-LDL-C of 160 mg/dL, the 2 values are comparable; at a C-LDL-C of 190 mg/dL, the D-LDL-C is slightly lower at 182 mg/dL. This is demonstrated graphically in the Figure by a dashed line indicating a perfect correlation between the 2 methods. The Figure also displays vertical and horizontal lines through an LDL-C of 100 mg/dL, the level above which drug therapy is likely to be started or increased in patients with CHD or CHD risk equivalents. This partition illustrates those patients who would require treatment when using the 100 mg/dL treatment goal by the C-LDL-C, the D-LDLC, neither, or both. This is also shown in the (Table, which shows the number of patients who would be treated with the LDL-C cutoffs for treatment recommended by NCEP-ATP III. At an LDL cutoff of 100 mg/dL, 60 patients (13% of total) would be treated with the D-LDL-C and not the C-LDL-C, whereas only 2 patients (<1%) would be treated with the C-LDL-C and not with the D-LDL-C. The results are similar when using a 130 mg/dL cutoff for treatment. Thus, treatment decisions based on the D-LDL-C results in many patients being treated who would not have been treated when using the C-LDL-C.

To determine whether triglyceride concentration influences treatment decisions by either method of LDL-C measurement, similar correlations and analyses were done on the data according to the following triglyceride groupings: <100 mg/dL, 100 to 199 mg/dL, 200 to 299 mg/dL, 300 to 399 mg/dL, and >400 mg/dL. This was further evaluated by plotting triglyceride vs D-LDL-C and triglyceride vs C-LDL-C (data not shown). Whereas the C-LDL-C showed no correlation with triglyceride, the DLDL-C showed a statistically significant correlation with triglyceride concentrations (r = .27), indicating that D-LDL-C increases at higher triglyceride levels. This suggested an influence of triglyceride on the D-LDL-C assay. This has been reported by others in 3 of 4 different D-LDL-C assays including the Sigma assay.¹⁵ However, alterations in treatment possibilities when using the D-LDL-C are present at all triglyceride concentrations.

FIGURE 1
Direct vs calculated LDL-C (mg/dL)

TABLE
Effect of LDL assay by LDL treatment cutoff*

Discussion

LDL-C has been identified in the NCEP-ATP III as the primary target of therapy. Treatment recommendations for high LDL-C are based on low, moderate, or high risk for CHD, with treatment goals of 160, 130, and 100 mg/dL, respectively.^1,2

These evidence-based recommendations rely on data from clinical trials demonstrating prevention of CHD events by lowering LDL-C, all of which, with the exception of the recently reported Heart Protection Study, used the CLDL-C.^18-23 Thus, important treatment decisions depend on this estimated LDL-C, and systematic deviations from the C-LDL-C will affect treatment decisions and cost.

The Sigma EZ LDL D-LDL-C assay in our hospital produces higher LDL-C levels than the C-LDL-C in a range of 100 to 160 mg/dL, the range of most common concern to clinicians. This results in inappropriate treatment or intensification in treatment according to the NCEP-ATP III guidelines. The DLDL-C was higher than the C-LDL-C at all triglyceride levels, but the error was greater for hypertriglyceridemia, the very situation for which it has been advocated.

Previous publications using a D-LDL-C assay have emphasized the correlation between the DLDL-C assay and research LDL-C determinations rather than the correlation with the C-LDL-C.^5-8 Other investigators have observed a similar tendency for higher D-LDL-C than C-LDL-C measurements at an LDL-C of 100 mg/dL¹⁷ and a positive bias at higher triglyceride levels.¹⁴ Although C-LDLC was often performed, data similar to those shown in the Figure, ie, the simple correlation between the D-LDL-C and C-LDL-C, have not been presented. Two very recent reviews have suggested caution in routinely implementing the D-LDL-C assays and pointed out the considerable variation from one assay to another.^14,15 Laboratories often change their assay method; in fact, our hospital laboratory has recently changed to a different DLDL-C method.

Physicians and institutions should be cautious about using a D-LDL-C method as a substitute for the C-LDL-C. First, it has not been standardized in large populations and, with the exception of the recent Heart Protection Study,²³ has not been used in large clinical trials demonstrating the benefits of lowering LDL-C. Although the C-LDL-C has been recommended by the NCEP-ATP III,² the Executive Summary of these guidelines did not address the method for measuring LDL-C.¹ Second, cost is increased from the additional therapy and performing the D-LDL-C assay. Third, the major reasons proposed for using a D-LDL-C assay (lack of need for a fasting specimen and usefulness at triglyceride > 400 mg/dL) may not be valid or relevant. Variation in the LDL-C due to hypertriglyceridemia occurs with the D-LDL-C. In addition, the NCEP-ATP III report emphasized triglyceride and recommended a fasting lipid panel including total cholesterol, triglycerides, and HDL-C.^1,2 One limitation of this study is the inclusion of predominantly male veterans. There may be populations, not considered in this study, that have an abnormal lipoprotein composition that significantly affects the C-LDL-C.

Conclusions

The C-LDL-C should remain the method of choice for LDL-C determinations because (1) this assay was used in clinical trials documenting the benefits of cholesterol-lowering therapy and (2) use of the D-LDL-C increases cost without evidence of benefit. Further studies are needed to standardize the direct LDL-C assays, and outcome trials using these assays need to be performed.

ACKNOWLEDGMENTS

This work was supported in part by the Biomedical Research Foundation of Arkansas and the Central Arkansas Veterans Healthcare System.

The National Cholesterol Education Program Adult Treatment Panel III Report (NCEP-ATP III) has identified low-density lipoprotein cholesterol (LDL-C) as the primary target of therapy.^1,2The Friedewald calculated LDL-C (C-LDL-C) is the preferred method^2,3 and is calculated with the following equation:

LDL-C =TC – HDL-C – TG/5

where TC is total cholesterol concentration, HDL-C is high-density lipoprotein cholesterol concentration, and TG is triglyceride concentration. The complete NCEP-ATP III report has indicated that methods to directly measure LDL-C (D-LDL-C) in the non-fasting state have been developed and will grow in use but require careful quality control.²Our VA hospital clinical laboratory is 1 of 10 hospitals in the South Central VA Health Care Network that routinely reports D-LDL-C rather than C-LDL-C levels to clinicians. Telephone calls to 4 other research and clinical laboratories found that all are using D-LDL-C to some extent.

D-LDL-C assays correlate variably with C-LDL-C measurements used in research studies.^4-17The purported advantages of such measurements are that fasting is not required and that D-LDL-C may be determined in patients with serum triglyceride levels greater than 400 mg/dL when the C-HDL-C and are less reliable. However, clinical trials demonstrating benefit of lowering LDL-C with drug therapy used the C-LDL-C.^18-22Only the recently reported Heart Protection Study used a non-fasting D-LDL-C.²³Thus, it is important in practicing evidence-based medicine to demonstrate that the D-LDL-C measurements are comparable to those of the C-LDL-C. The present study determined how the D-LDL-C correlated with C-LDL-C and how such a correlation would affect treatment decisions based on the NCEP-ATP III guidelines.

Methods

Data from all patients with a lipid panel during a single week were analyzed. Patients with triglyceride levels above 1000 mg/dL were excluded. Thirty-four patients with triglyceride levels between 400 and 1000 mg/dL were analyzed separately. A C-LDL-C was determined and compared with the D-LDL-C in all 464 patients. Total cholesterol, triglyceride, and HDL-C measurements were done with an autoanalyzer. D-LDL-C was measured with Sigma Diagnostics EZ LDL Cholesterol, procedure 358 (Sigma, St. Louis, MO). Linear regression was performed using Microsoft Excel (Microsoft Corporation, Redmond, WA).

Results

The samples in this study represented the expected distribution of LDL-C concentrations seen in a clinical practice of predominantly male veterans. Of the 464 patient samples with triglyceride levels below 400 mg/dL, the mean C-LDL-C was 123 mg/dL. Twenty-eight percent had a C-LDL-C below 100 mg/dL, 32% had a C-LDL-C of 100 to 129.9 mg/dL, 24% had a C-LDL-C of 130 to 159.9 mg/dL, 12% had a C-LDL-C of 160 to 189.9 mg/dL, and 4% had a C-LDL-C above 190 mg/dL.

The Figure shows the correlation between the C-LDL-C and D-LDL-C in all patients with triglyceride levels below 400 mg/dL. Although there is a strong correlation between the C-LDL-C and D-LDL-C (r = .86), the regression line does not go through 0. A C-LDL-C of 100 mg/dL or lower is the NCEP-ATP III goal for patients with known coronary heart disease (CHD) and other clinical forms of atherosclerotic disease, diabetes, or multiple risk factors that confer a 10-year risk for CHD greater than 20%.¹At this cutoff for C-LDL-C, the D-LDL-C derived from the regression line is 118 mg/dL. At a C-LDL-C of 160 mg/dL, the 2 values are comparable; at a C-LDL-C of 190 mg/dL, the D-LDL-C is slightly lower at 182 mg/dL. This is demonstrated graphically in the Figure by a dashed line indicating a perfect correlation between the 2 methods. The Figure also displays vertical and horizontal lines through an LDL-C of 100 mg/dL, the level above which drug therapy is likely to be started or increased in patients with CHD or CHD risk equivalents. This partition illustrates those patients who would require treatment when using the 100 mg/dL treatment goal by the C-LDL-C, the D-LDLC, neither, or both. This is also shown in the (Table, which shows the number of patients who would be treated with the LDL-C cutoffs for treatment recommended by NCEP-ATP III. At an LDL cutoff of 100 mg/dL, 60 patients (13% of total) would be treated with the D-LDL-C and not the C-LDL-C, whereas only 2 patients (<1%) would be treated with the C-LDL-C and not with the D-LDL-C. The results are similar when using a 130 mg/dL cutoff for treatment. Thus, treatment decisions based on the D-LDL-C results in many patients being treated who would not have been treated when using the C-LDL-C.

To determine whether triglyceride concentration influences treatment decisions by either method of LDL-C measurement, similar correlations and analyses were done on the data according to the following triglyceride groupings: <100 mg/dL, 100 to 199 mg/dL, 200 to 299 mg/dL, 300 to 399 mg/dL, and >400 mg/dL. This was further evaluated by plotting triglyceride vs D-LDL-C and triglyceride vs C-LDL-C (data not shown). Whereas the C-LDL-C showed no correlation with triglyceride, the DLDL-C showed a statistically significant correlation with triglyceride concentrations (r = .27), indicating that D-LDL-C increases at higher triglyceride levels. This suggested an influence of triglyceride on the D-LDL-C assay. This has been reported by others in 3 of 4 different D-LDL-C assays including the Sigma assay.¹⁵ However, alterations in treatment possibilities when using the D-LDL-C are present at all triglyceride concentrations.

FIGURE 1
Direct vs calculated LDL-C (mg/dL)

TABLE
Effect of LDL assay by LDL treatment cutoff*

Discussion

LDL-C has been identified in the NCEP-ATP III as the primary target of therapy. Treatment recommendations for high LDL-C are based on low, moderate, or high risk for CHD, with treatment goals of 160, 130, and 100 mg/dL, respectively.^1,2

These evidence-based recommendations rely on data from clinical trials demonstrating prevention of CHD events by lowering LDL-C, all of which, with the exception of the recently reported Heart Protection Study, used the CLDL-C.^18-23 Thus, important treatment decisions depend on this estimated LDL-C, and systematic deviations from the C-LDL-C will affect treatment decisions and cost.

The Sigma EZ LDL D-LDL-C assay in our hospital produces higher LDL-C levels than the C-LDL-C in a range of 100 to 160 mg/dL, the range of most common concern to clinicians. This results in inappropriate treatment or intensification in treatment according to the NCEP-ATP III guidelines. The DLDL-C was higher than the C-LDL-C at all triglyceride levels, but the error was greater for hypertriglyceridemia, the very situation for which it has been advocated.

Previous publications using a D-LDL-C assay have emphasized the correlation between the DLDL-C assay and research LDL-C determinations rather than the correlation with the C-LDL-C.^5-8 Other investigators have observed a similar tendency for higher D-LDL-C than C-LDL-C measurements at an LDL-C of 100 mg/dL¹⁷ and a positive bias at higher triglyceride levels.¹⁴ Although C-LDLC was often performed, data similar to those shown in the Figure, ie, the simple correlation between the D-LDL-C and C-LDL-C, have not been presented. Two very recent reviews have suggested caution in routinely implementing the D-LDL-C assays and pointed out the considerable variation from one assay to another.^14,15 Laboratories often change their assay method; in fact, our hospital laboratory has recently changed to a different DLDL-C method.

Physicians and institutions should be cautious about using a D-LDL-C method as a substitute for the C-LDL-C. First, it has not been standardized in large populations and, with the exception of the recent Heart Protection Study,²³ has not been used in large clinical trials demonstrating the benefits of lowering LDL-C. Although the C-LDL-C has been recommended by the NCEP-ATP III,² the Executive Summary of these guidelines did not address the method for measuring LDL-C.¹ Second, cost is increased from the additional therapy and performing the D-LDL-C assay. Third, the major reasons proposed for using a D-LDL-C assay (lack of need for a fasting specimen and usefulness at triglyceride > 400 mg/dL) may not be valid or relevant. Variation in the LDL-C due to hypertriglyceridemia occurs with the D-LDL-C. In addition, the NCEP-ATP III report emphasized triglyceride and recommended a fasting lipid panel including total cholesterol, triglycerides, and HDL-C.^1,2 One limitation of this study is the inclusion of predominantly male veterans. There may be populations, not considered in this study, that have an abnormal lipoprotein composition that significantly affects the C-LDL-C.

Conclusions

The C-LDL-C should remain the method of choice for LDL-C determinations because (1) this assay was used in clinical trials documenting the benefits of cholesterol-lowering therapy and (2) use of the D-LDL-C increases cost without evidence of benefit. Further studies are needed to standardize the direct LDL-C assays, and outcome trials using these assays need to be performed.

ACKNOWLEDGMENTS

This work was supported in part by the Biomedical Research Foundation of Arkansas and the Central Arkansas Veterans Healthcare System.

Taking a family history is a significant component of providing comprehensive primary care because family history provides key psychosocial and medical risk information.^1,2 As our understanding of the genetic basis for disease grows, obtaining an accurate and complete family history is likely to gain increasing relevance as a vital source of data to guide counseling and testing. Patients who have a first-degree relative with a colon neoplasm or prostate cancer are advised to screen differently from those who do not.³ Failure to gather accurate or complete family data prevents the clinician from providing advice that is consistent with screening guidelines. Understanding how primary care clinicians gather family history data is necessary to identify gaps in current performance and to develop strategies to bridge these gaps.

Several studies have used physician self-report to assess the current level of taking a family history in primary care. In 1 study, 90% of surveyed physicians stated that they obtain a family history of cancer from their patients, with 77% to 80% inquiring about a family history of colorectal cancer in their patients who are at least 40 years of age.³ In another study, 63% to 85% of responding physicians reported obtaining a family history of cancer from 76% to 100% of their patients.⁵ Respondents in 1 study reported obtaining family histories of colorectal cancer in only 30.7% of patients, breast cancer in 48.4% of patients, and coronary disease and hypertension in 94.3% of patients.⁶ However, data on actual performance of family history taking are sparse.^7,8 The physicians in the Direct Observation of Primary Care study obtained a family history during 51% of new visits and 22% of established visits. A genogram was present in 11% of charts, and documentation of a family history of breast cancer or colorectal cancer was found in 40% of charts. Further analysis from this study showed that providers who more frequently obtained and recorded family history information performed more preventive care services for their patients.⁹

First-degree relatives of cancer patients appear to be interested in and receptive to information about their risk and in the possibility of genetic counseling.^10-14 In fact, many patients overestimate their likelihood of getting cancer based on family history¹⁰; primary care providers thus have the opportunity to counsel and relieve anxiety in their patients. Family history is an important tool to define risk and guide referral, counseling, and testing.

This article presents the findings of a descriptive study of the review of 500 charts from 3 different family practice offices. We documented general family history components and completeness related to a family history of cancer, including whether enough family history information was collected to appropriately identify patients at increased risk for cancer.

Methods

Data were collected from 500 patient charts from 3 family practice offices in the greater Philadelphia area. Patients who were in the practice for at least 1 year, made a minimum of 2 visits between June 1, 1997 and June 1, 2000, and were between the ages of 40 and 60 years at 1 of those visits were included. Fifty charts were selected by using a random starting point at each of 2 small (1 to 3 providers) private practices, and 400 charts were randomly selected from all eligible charts in a large academic practice (more than 60 providers). The large practice had nearly 1500 patients who fit the selection criteria, and the total population did not differ from the random sample in mean age, sex, or race (P < .05). Total population characteristics were not available for the 2 smaller practices.

Family history data were collected from progress notes and designated family history spaces on flow sheets or chart covers. Family histories consisting only of “none” or “noncontributory” were counted only when they clearly referred to a specific condition. Family births or deaths recorded in a psychosocial context also were not counted as part of the family history unless death from a specific disease was mentioned. The first dated family history was considered to be first for the purpose of this study. Date of birth, race, sex, current primary care provider, if any, and the date first seen in the practice were recorded. The current primary care provider was determined by the physician seen for the majority of recent visits and/or notations in the chart at acute visits. Data collected for the primary care providers included practice site, sex, level of training, and years in practice. Variables collected from family history information included: date of first family history; date of most recent family history; presence of a genogram, presence of a patient-completed family history self-questionnaire; whether any mention was made, positive or negative, of cancer or colon polyps; and whether there was a positive family history of cancer or colon polyps. For individuals with a positive family history of cancer or colon polyps, all details recorded in the chart were abstracted; these included site of cancer or polyp, relationship to patient, age at diagnosis, and age at death. The data were entered into an Access 97 database and stored separately from chart number identifiers. All analyses and tests were done in SAS version 6.12.

Results

Demographic data from the 500 patients whose charts were audited are presented in Table 1. Ninety-seven percent of patients had a primary care provider, which included 60 physicians and 3 nurse practitioners. No significant associations were seen with practice site, sex, or level of training of the provider and the presence of family history information in the chart.

Most patients (89%) had some family history information in the chart and 63% had a genogram. This did not differ by sex or race of the patient. Fifty-seven percent of patients supplied family history information at the first visit to the office; 59% of these patients had no family history data recorded on subsequent visits. Only 31% of charts had updated family history information Table 2. For patients who had been in the same practice for at least 5 years and had some family history in the chart, 20% had some updated information within the past 3 years.

Of the 500 charts, 276 (55%) recorded a family history of cancer, positive or negative. Two hundred fifteen patients (43%) had a positive family history of cancer, with a total of 356 relatives affected. The site of cancer was listed for 88% of all family member cancers, with breast, colon, lung, and prostate being the most common cancer locations. The specific relative was identified in 92% of cases, with most being first (51%) or second (37%) degree. Although degree of relative and cancer location were usually recorded, age at diagnosis was listed for only 8% of affected relatives, and age at death was identified for 19% of relatives with cancer.

For patients with affected first-degree relatives, the group with the greatest clinical significance, primary care providers identified the location of the cancer in 93% of cases but listed the ages at diagnosis and death in only 7% and 31%, respectively Table 3. Only 7 medical records (1.4%) had any mention of a family history of polyps; of these, 5 (1%) were positive. None listed an age at diagnosis. Five patients in our study met the American Society of Clinical Oncology criteria to be evaluated for genetic breast and ovarian cancer syndrome, and no patients met the criteria for hereditary nonpolyposis colon cancer.

The 2 community practices intermittently used patient self-administered medical intake questionnaires. In our sample, 31 of 500 patients (6%) had a questionnaire in the chart. All patients who completed questionnaires had family history data in their charts. Use of a questionnaire was associated with a greater likelihood that the physician recorded the age of diagnosis for a relative with cancer, although this did not reach significance (20% vs 7%).

Discussion

Despite our finding that providers are documenting family histories in most charts, very few are recording the age of diagnosis in relatives diagnosed with cancer. Age of diagnosis plays a critical role in determining screening recommendations and identifying patients with possible genetic syndromes. For example, the Amsterdam criteria used to identify families with hereditary nonpolyposis colon cancer include knowing whether 1 of 3 relatives with colorectal cancer was diagnosed at younger than 50 years. Breast and ovarian cancer syndrome should be suspected when breast and/or ovarian cancer is diagnosed in 2 first-degree relatives younger than 50 years.¹⁵

Most physicians obtain family histories at initial visits. If the patient’s initial visit is for symptom- or disease-related care, an opportunity to gather family history data may be lost. New tools to consistently capture comprehensive family history data at this first visit may be beneficial. Patient self-administered intake questionnaires may prove valuable in this respect, but only 6% of charts in our study contained such a questionnaire, so we cannot draw conclusions about its impact. We did observe a trend toward gathering more complete family history data in patients who used a questionnaire.

There are no clear guidelines regarding when to update a family history. We arbitrarily chose 3 years as a reasonable period for primary care providers to explore changes in family history status. Updating at any subsequent visit was recorded for 35% of patient charts in which a family history was initially taken. It is not clear that primary care providers are documenting changes in family history in any systematic way. Opportunistic updating likely occurs when a new diagnosis of serious disease in a close family member produces anxiety, stress, or concern in the patient. The value of updating family history and the ideal interval to reexamine family history are unknown.

Several conditions would need to be met for family history updating to have value. (1) A close relative must have developed an important illness in the interval since the last family history was recorded or the update must discover family information that was previously missed. (2) The illness must have a familial component that affects the estimate of the risk of the identified patient. (3) The clinician would need to be aware of the updated information. (4) The clinician must change recommendations to the patient based on this new information. Discovering a new family history of colonic neoplasm satisfies these conditions. Process measures that have the potential to improve updating include adding an update of family history as an item on a preventive care flow sheet or using periodic self-administered patient questionnaires. Whether adequate improvements in health care would occur to justify these changes in process will need to be studied. If any updating has value, determining the appropriate intervals for systematic updates deserves attention.

Charts in this study included a genogram 63% of the time, a significant increase over the 11% noted in the Direct Observation of Primary Care study.⁹ This discrepancy may be explained by differences in practice types because 1 study suggested a higher genogram use in academic medical centers than in community practices.¹⁶ The genogram has been cited as an attractive and efficient way to document family history,^17,18 but over one fourth of the charts that contained family history in our study used the more cumbersome narrative form. Many geneticists predict that our ability to apply genetic testing will grow dramatically over the next decade. Optimal application of this new knowledge will rely on the health care system’s capacity to accurately identify risk based on assessment of family history. The 3-generation pedigree is likely to be a key tool in finding individuals who may benefit from testing. However, there is currently no standardized education in family history taking in many undergraduate and graduate medical education programs.¹⁹

Although patients with a first-degree relative with a history of polyps diagnosed at younger than 60 years are considered to be at increased risk for colorectal cancer,³ providers infrequently asked about a family history of polyps. This may reflect a recent finding that only 36% of primary care providers recommend screening at the age of 40 years for their patients with a family history of polyps in relatives younger than 60 years.²⁰ In fact, family history data do not consistently influence behavior: in the same study, gastroenterologists asked about a family history of polyps 93% of the time, but only 37% recommended earlier screening in those with such a history.

The study is limited in its use of only 3 primary care practices, 1 of which was a large academic family practice. However, because the charts of 63 different clinicians were represented, a range of educational backgrounds and personal philosophies toward family history taking was included. Most patients (97%) had a clearly identified primary care provider and patients had been members of the practice for an average of 7.6 years. The sample was specifically chosen to review the charts of individuals who had been enrolled in the same practice for at least 1 year. This may explain the higher rates of family history taking found in this study compared with previously published studies. Given that the vast majority of charts did contain some family history, it is even more compelling that age at diagnosis of cancer was inadequately recorded. This study reflected only what was documented in the patient chart and not direct observation of physician behavior regarding the family history. It is likely that physicians are not recording all responses to inquiries about family history, although the extent of this underreporting is unknown.

CONCLUSION

Findings in this chart review study are consistent with previous work showing that the quantity and type of family history currently being recorded in primary care charts are not adequate to fully assess familial risk. Bridging the gap between recommendations and actual practice will demand interventions to alter primary care practice or the introduction of new models to gather and analyze family data. Further research is also needed to evaluate the impact of improved family history taking on health care costs and outcomes.

Acknowledgments

The authors thank Howard Rabinowitz, MD, for providing helpful suggestions in the development and execution of this project and Aliza Mansolino for the preparation of the manuscript.

Taking a family history is a significant component of providing comprehensive primary care because family history provides key psychosocial and medical risk information.^1,2 As our understanding of the genetic basis for disease grows, obtaining an accurate and complete family history is likely to gain increasing relevance as a vital source of data to guide counseling and testing. Patients who have a first-degree relative with a colon neoplasm or prostate cancer are advised to screen differently from those who do not.³ Failure to gather accurate or complete family data prevents the clinician from providing advice that is consistent with screening guidelines. Understanding how primary care clinicians gather family history data is necessary to identify gaps in current performance and to develop strategies to bridge these gaps.

Several studies have used physician self-report to assess the current level of taking a family history in primary care. In 1 study, 90% of surveyed physicians stated that they obtain a family history of cancer from their patients, with 77% to 80% inquiring about a family history of colorectal cancer in their patients who are at least 40 years of age.³ In another study, 63% to 85% of responding physicians reported obtaining a family history of cancer from 76% to 100% of their patients.⁵ Respondents in 1 study reported obtaining family histories of colorectal cancer in only 30.7% of patients, breast cancer in 48.4% of patients, and coronary disease and hypertension in 94.3% of patients.⁶ However, data on actual performance of family history taking are sparse.^7,8 The physicians in the Direct Observation of Primary Care study obtained a family history during 51% of new visits and 22% of established visits. A genogram was present in 11% of charts, and documentation of a family history of breast cancer or colorectal cancer was found in 40% of charts. Further analysis from this study showed that providers who more frequently obtained and recorded family history information performed more preventive care services for their patients.⁹

First-degree relatives of cancer patients appear to be interested in and receptive to information about their risk and in the possibility of genetic counseling.^10-14 In fact, many patients overestimate their likelihood of getting cancer based on family history¹⁰; primary care providers thus have the opportunity to counsel and relieve anxiety in their patients. Family history is an important tool to define risk and guide referral, counseling, and testing.

This article presents the findings of a descriptive study of the review of 500 charts from 3 different family practice offices. We documented general family history components and completeness related to a family history of cancer, including whether enough family history information was collected to appropriately identify patients at increased risk for cancer.

Methods

Data were collected from 500 patient charts from 3 family practice offices in the greater Philadelphia area. Patients who were in the practice for at least 1 year, made a minimum of 2 visits between June 1, 1997 and June 1, 2000, and were between the ages of 40 and 60 years at 1 of those visits were included. Fifty charts were selected by using a random starting point at each of 2 small (1 to 3 providers) private practices, and 400 charts were randomly selected from all eligible charts in a large academic practice (more than 60 providers). The large practice had nearly 1500 patients who fit the selection criteria, and the total population did not differ from the random sample in mean age, sex, or race (P < .05). Total population characteristics were not available for the 2 smaller practices.

Family history data were collected from progress notes and designated family history spaces on flow sheets or chart covers. Family histories consisting only of “none” or “noncontributory” were counted only when they clearly referred to a specific condition. Family births or deaths recorded in a psychosocial context also were not counted as part of the family history unless death from a specific disease was mentioned. The first dated family history was considered to be first for the purpose of this study. Date of birth, race, sex, current primary care provider, if any, and the date first seen in the practice were recorded. The current primary care provider was determined by the physician seen for the majority of recent visits and/or notations in the chart at acute visits. Data collected for the primary care providers included practice site, sex, level of training, and years in practice. Variables collected from family history information included: date of first family history; date of most recent family history; presence of a genogram, presence of a patient-completed family history self-questionnaire; whether any mention was made, positive or negative, of cancer or colon polyps; and whether there was a positive family history of cancer or colon polyps. For individuals with a positive family history of cancer or colon polyps, all details recorded in the chart were abstracted; these included site of cancer or polyp, relationship to patient, age at diagnosis, and age at death. The data were entered into an Access 97 database and stored separately from chart number identifiers. All analyses and tests were done in SAS version 6.12.

Results

Demographic data from the 500 patients whose charts were audited are presented in Table 1. Ninety-seven percent of patients had a primary care provider, which included 60 physicians and 3 nurse practitioners. No significant associations were seen with practice site, sex, or level of training of the provider and the presence of family history information in the chart.

Most patients (89%) had some family history information in the chart and 63% had a genogram. This did not differ by sex or race of the patient. Fifty-seven percent of patients supplied family history information at the first visit to the office; 59% of these patients had no family history data recorded on subsequent visits. Only 31% of charts had updated family history information Table 2. For patients who had been in the same practice for at least 5 years and had some family history in the chart, 20% had some updated information within the past 3 years.

Of the 500 charts, 276 (55%) recorded a family history of cancer, positive or negative. Two hundred fifteen patients (43%) had a positive family history of cancer, with a total of 356 relatives affected. The site of cancer was listed for 88% of all family member cancers, with breast, colon, lung, and prostate being the most common cancer locations. The specific relative was identified in 92% of cases, with most being first (51%) or second (37%) degree. Although degree of relative and cancer location were usually recorded, age at diagnosis was listed for only 8% of affected relatives, and age at death was identified for 19% of relatives with cancer.

For patients with affected first-degree relatives, the group with the greatest clinical significance, primary care providers identified the location of the cancer in 93% of cases but listed the ages at diagnosis and death in only 7% and 31%, respectively Table 3. Only 7 medical records (1.4%) had any mention of a family history of polyps; of these, 5 (1%) were positive. None listed an age at diagnosis. Five patients in our study met the American Society of Clinical Oncology criteria to be evaluated for genetic breast and ovarian cancer syndrome, and no patients met the criteria for hereditary nonpolyposis colon cancer.

The 2 community practices intermittently used patient self-administered medical intake questionnaires. In our sample, 31 of 500 patients (6%) had a questionnaire in the chart. All patients who completed questionnaires had family history data in their charts. Use of a questionnaire was associated with a greater likelihood that the physician recorded the age of diagnosis for a relative with cancer, although this did not reach significance (20% vs 7%).

Discussion

Despite our finding that providers are documenting family histories in most charts, very few are recording the age of diagnosis in relatives diagnosed with cancer. Age of diagnosis plays a critical role in determining screening recommendations and identifying patients with possible genetic syndromes. For example, the Amsterdam criteria used to identify families with hereditary nonpolyposis colon cancer include knowing whether 1 of 3 relatives with colorectal cancer was diagnosed at younger than 50 years. Breast and ovarian cancer syndrome should be suspected when breast and/or ovarian cancer is diagnosed in 2 first-degree relatives younger than 50 years.¹⁵

Most physicians obtain family histories at initial visits. If the patient’s initial visit is for symptom- or disease-related care, an opportunity to gather family history data may be lost. New tools to consistently capture comprehensive family history data at this first visit may be beneficial. Patient self-administered intake questionnaires may prove valuable in this respect, but only 6% of charts in our study contained such a questionnaire, so we cannot draw conclusions about its impact. We did observe a trend toward gathering more complete family history data in patients who used a questionnaire.

There are no clear guidelines regarding when to update a family history. We arbitrarily chose 3 years as a reasonable period for primary care providers to explore changes in family history status. Updating at any subsequent visit was recorded for 35% of patient charts in which a family history was initially taken. It is not clear that primary care providers are documenting changes in family history in any systematic way. Opportunistic updating likely occurs when a new diagnosis of serious disease in a close family member produces anxiety, stress, or concern in the patient. The value of updating family history and the ideal interval to reexamine family history are unknown.

Several conditions would need to be met for family history updating to have value. (1) A close relative must have developed an important illness in the interval since the last family history was recorded or the update must discover family information that was previously missed. (2) The illness must have a familial component that affects the estimate of the risk of the identified patient. (3) The clinician would need to be aware of the updated information. (4) The clinician must change recommendations to the patient based on this new information. Discovering a new family history of colonic neoplasm satisfies these conditions. Process measures that have the potential to improve updating include adding an update of family history as an item on a preventive care flow sheet or using periodic self-administered patient questionnaires. Whether adequate improvements in health care would occur to justify these changes in process will need to be studied. If any updating has value, determining the appropriate intervals for systematic updates deserves attention.

Charts in this study included a genogram 63% of the time, a significant increase over the 11% noted in the Direct Observation of Primary Care study.⁹ This discrepancy may be explained by differences in practice types because 1 study suggested a higher genogram use in academic medical centers than in community practices.¹⁶ The genogram has been cited as an attractive and efficient way to document family history,^17,18 but over one fourth of the charts that contained family history in our study used the more cumbersome narrative form. Many geneticists predict that our ability to apply genetic testing will grow dramatically over the next decade. Optimal application of this new knowledge will rely on the health care system’s capacity to accurately identify risk based on assessment of family history. The 3-generation pedigree is likely to be a key tool in finding individuals who may benefit from testing. However, there is currently no standardized education in family history taking in many undergraduate and graduate medical education programs.¹⁹

Although patients with a first-degree relative with a history of polyps diagnosed at younger than 60 years are considered to be at increased risk for colorectal cancer,³ providers infrequently asked about a family history of polyps. This may reflect a recent finding that only 36% of primary care providers recommend screening at the age of 40 years for their patients with a family history of polyps in relatives younger than 60 years.²⁰ In fact, family history data do not consistently influence behavior: in the same study, gastroenterologists asked about a family history of polyps 93% of the time, but only 37% recommended earlier screening in those with such a history.

The study is limited in its use of only 3 primary care practices, 1 of which was a large academic family practice. However, because the charts of 63 different clinicians were represented, a range of educational backgrounds and personal philosophies toward family history taking was included. Most patients (97%) had a clearly identified primary care provider and patients had been members of the practice for an average of 7.6 years. The sample was specifically chosen to review the charts of individuals who had been enrolled in the same practice for at least 1 year. This may explain the higher rates of family history taking found in this study compared with previously published studies. Given that the vast majority of charts did contain some family history, it is even more compelling that age at diagnosis of cancer was inadequately recorded. This study reflected only what was documented in the patient chart and not direct observation of physician behavior regarding the family history. It is likely that physicians are not recording all responses to inquiries about family history, although the extent of this underreporting is unknown.

CONCLUSION

Findings in this chart review study are consistent with previous work showing that the quantity and type of family history currently being recorded in primary care charts are not adequate to fully assess familial risk. Bridging the gap between recommendations and actual practice will demand interventions to alter primary care practice or the introduction of new models to gather and analyze family data. Further research is also needed to evaluate the impact of improved family history taking on health care costs and outcomes.

Acknowledgments

The authors thank Howard Rabinowitz, MD, for providing helpful suggestions in the development and execution of this project and Aliza Mansolino for the preparation of the manuscript.

Taking a family history is a significant component of providing comprehensive primary care because family history provides key psychosocial and medical risk information.^1,2 As our understanding of the genetic basis for disease grows, obtaining an accurate and complete family history is likely to gain increasing relevance as a vital source of data to guide counseling and testing. Patients who have a first-degree relative with a colon neoplasm or prostate cancer are advised to screen differently from those who do not.³ Failure to gather accurate or complete family data prevents the clinician from providing advice that is consistent with screening guidelines. Understanding how primary care clinicians gather family history data is necessary to identify gaps in current performance and to develop strategies to bridge these gaps.

Several studies have used physician self-report to assess the current level of taking a family history in primary care. In 1 study, 90% of surveyed physicians stated that they obtain a family history of cancer from their patients, with 77% to 80% inquiring about a family history of colorectal cancer in their patients who are at least 40 years of age.³ In another study, 63% to 85% of responding physicians reported obtaining a family history of cancer from 76% to 100% of their patients.⁵ Respondents in 1 study reported obtaining family histories of colorectal cancer in only 30.7% of patients, breast cancer in 48.4% of patients, and coronary disease and hypertension in 94.3% of patients.⁶ However, data on actual performance of family history taking are sparse.^7,8 The physicians in the Direct Observation of Primary Care study obtained a family history during 51% of new visits and 22% of established visits. A genogram was present in 11% of charts, and documentation of a family history of breast cancer or colorectal cancer was found in 40% of charts. Further analysis from this study showed that providers who more frequently obtained and recorded family history information performed more preventive care services for their patients.⁹

First-degree relatives of cancer patients appear to be interested in and receptive to information about their risk and in the possibility of genetic counseling.^10-14 In fact, many patients overestimate their likelihood of getting cancer based on family history¹⁰; primary care providers thus have the opportunity to counsel and relieve anxiety in their patients. Family history is an important tool to define risk and guide referral, counseling, and testing.

This article presents the findings of a descriptive study of the review of 500 charts from 3 different family practice offices. We documented general family history components and completeness related to a family history of cancer, including whether enough family history information was collected to appropriately identify patients at increased risk for cancer.

Methods

Data were collected from 500 patient charts from 3 family practice offices in the greater Philadelphia area. Patients who were in the practice for at least 1 year, made a minimum of 2 visits between June 1, 1997 and June 1, 2000, and were between the ages of 40 and 60 years at 1 of those visits were included. Fifty charts were selected by using a random starting point at each of 2 small (1 to 3 providers) private practices, and 400 charts were randomly selected from all eligible charts in a large academic practice (more than 60 providers). The large practice had nearly 1500 patients who fit the selection criteria, and the total population did not differ from the random sample in mean age, sex, or race (P < .05). Total population characteristics were not available for the 2 smaller practices.

Family history data were collected from progress notes and designated family history spaces on flow sheets or chart covers. Family histories consisting only of “none” or “noncontributory” were counted only when they clearly referred to a specific condition. Family births or deaths recorded in a psychosocial context also were not counted as part of the family history unless death from a specific disease was mentioned. The first dated family history was considered to be first for the purpose of this study. Date of birth, race, sex, current primary care provider, if any, and the date first seen in the practice were recorded. The current primary care provider was determined by the physician seen for the majority of recent visits and/or notations in the chart at acute visits. Data collected for the primary care providers included practice site, sex, level of training, and years in practice. Variables collected from family history information included: date of first family history; date of most recent family history; presence of a genogram, presence of a patient-completed family history self-questionnaire; whether any mention was made, positive or negative, of cancer or colon polyps; and whether there was a positive family history of cancer or colon polyps. For individuals with a positive family history of cancer or colon polyps, all details recorded in the chart were abstracted; these included site of cancer or polyp, relationship to patient, age at diagnosis, and age at death. The data were entered into an Access 97 database and stored separately from chart number identifiers. All analyses and tests were done in SAS version 6.12.

Results

Demographic data from the 500 patients whose charts were audited are presented in Table 1. Ninety-seven percent of patients had a primary care provider, which included 60 physicians and 3 nurse practitioners. No significant associations were seen with practice site, sex, or level of training of the provider and the presence of family history information in the chart.

Most patients (89%) had some family history information in the chart and 63% had a genogram. This did not differ by sex or race of the patient. Fifty-seven percent of patients supplied family history information at the first visit to the office; 59% of these patients had no family history data recorded on subsequent visits. Only 31% of charts had updated family history information Table 2. For patients who had been in the same practice for at least 5 years and had some family history in the chart, 20% had some updated information within the past 3 years.

Of the 500 charts, 276 (55%) recorded a family history of cancer, positive or negative. Two hundred fifteen patients (43%) had a positive family history of cancer, with a total of 356 relatives affected. The site of cancer was listed for 88% of all family member cancers, with breast, colon, lung, and prostate being the most common cancer locations. The specific relative was identified in 92% of cases, with most being first (51%) or second (37%) degree. Although degree of relative and cancer location were usually recorded, age at diagnosis was listed for only 8% of affected relatives, and age at death was identified for 19% of relatives with cancer.

For patients with affected first-degree relatives, the group with the greatest clinical significance, primary care providers identified the location of the cancer in 93% of cases but listed the ages at diagnosis and death in only 7% and 31%, respectively Table 3. Only 7 medical records (1.4%) had any mention of a family history of polyps; of these, 5 (1%) were positive. None listed an age at diagnosis. Five patients in our study met the American Society of Clinical Oncology criteria to be evaluated for genetic breast and ovarian cancer syndrome, and no patients met the criteria for hereditary nonpolyposis colon cancer.

The 2 community practices intermittently used patient self-administered medical intake questionnaires. In our sample, 31 of 500 patients (6%) had a questionnaire in the chart. All patients who completed questionnaires had family history data in their charts. Use of a questionnaire was associated with a greater likelihood that the physician recorded the age of diagnosis for a relative with cancer, although this did not reach significance (20% vs 7%).

Discussion

Despite our finding that providers are documenting family histories in most charts, very few are recording the age of diagnosis in relatives diagnosed with cancer. Age of diagnosis plays a critical role in determining screening recommendations and identifying patients with possible genetic syndromes. For example, the Amsterdam criteria used to identify families with hereditary nonpolyposis colon cancer include knowing whether 1 of 3 relatives with colorectal cancer was diagnosed at younger than 50 years. Breast and ovarian cancer syndrome should be suspected when breast and/or ovarian cancer is diagnosed in 2 first-degree relatives younger than 50 years.¹⁵

Most physicians obtain family histories at initial visits. If the patient’s initial visit is for symptom- or disease-related care, an opportunity to gather family history data may be lost. New tools to consistently capture comprehensive family history data at this first visit may be beneficial. Patient self-administered intake questionnaires may prove valuable in this respect, but only 6% of charts in our study contained such a questionnaire, so we cannot draw conclusions about its impact. We did observe a trend toward gathering more complete family history data in patients who used a questionnaire.

There are no clear guidelines regarding when to update a family history. We arbitrarily chose 3 years as a reasonable period for primary care providers to explore changes in family history status. Updating at any subsequent visit was recorded for 35% of patient charts in which a family history was initially taken. It is not clear that primary care providers are documenting changes in family history in any systematic way. Opportunistic updating likely occurs when a new diagnosis of serious disease in a close family member produces anxiety, stress, or concern in the patient. The value of updating family history and the ideal interval to reexamine family history are unknown.

Several conditions would need to be met for family history updating to have value. (1) A close relative must have developed an important illness in the interval since the last family history was recorded or the update must discover family information that was previously missed. (2) The illness must have a familial component that affects the estimate of the risk of the identified patient. (3) The clinician would need to be aware of the updated information. (4) The clinician must change recommendations to the patient based on this new information. Discovering a new family history of colonic neoplasm satisfies these conditions. Process measures that have the potential to improve updating include adding an update of family history as an item on a preventive care flow sheet or using periodic self-administered patient questionnaires. Whether adequate improvements in health care would occur to justify these changes in process will need to be studied. If any updating has value, determining the appropriate intervals for systematic updates deserves attention.

Charts in this study included a genogram 63% of the time, a significant increase over the 11% noted in the Direct Observation of Primary Care study.⁹ This discrepancy may be explained by differences in practice types because 1 study suggested a higher genogram use in academic medical centers than in community practices.¹⁶ The genogram has been cited as an attractive and efficient way to document family history,^17,18 but over one fourth of the charts that contained family history in our study used the more cumbersome narrative form. Many geneticists predict that our ability to apply genetic testing will grow dramatically over the next decade. Optimal application of this new knowledge will rely on the health care system’s capacity to accurately identify risk based on assessment of family history. The 3-generation pedigree is likely to be a key tool in finding individuals who may benefit from testing. However, there is currently no standardized education in family history taking in many undergraduate and graduate medical education programs.¹⁹

Although patients with a first-degree relative with a history of polyps diagnosed at younger than 60 years are considered to be at increased risk for colorectal cancer,³ providers infrequently asked about a family history of polyps. This may reflect a recent finding that only 36% of primary care providers recommend screening at the age of 40 years for their patients with a family history of polyps in relatives younger than 60 years.²⁰ In fact, family history data do not consistently influence behavior: in the same study, gastroenterologists asked about a family history of polyps 93% of the time, but only 37% recommended earlier screening in those with such a history.

The study is limited in its use of only 3 primary care practices, 1 of which was a large academic family practice. However, because the charts of 63 different clinicians were represented, a range of educational backgrounds and personal philosophies toward family history taking was included. Most patients (97%) had a clearly identified primary care provider and patients had been members of the practice for an average of 7.6 years. The sample was specifically chosen to review the charts of individuals who had been enrolled in the same practice for at least 1 year. This may explain the higher rates of family history taking found in this study compared with previously published studies. Given that the vast majority of charts did contain some family history, it is even more compelling that age at diagnosis of cancer was inadequately recorded. This study reflected only what was documented in the patient chart and not direct observation of physician behavior regarding the family history. It is likely that physicians are not recording all responses to inquiries about family history, although the extent of this underreporting is unknown.

CONCLUSION

Findings in this chart review study are consistent with previous work showing that the quantity and type of family history currently being recorded in primary care charts are not adequate to fully assess familial risk. Bridging the gap between recommendations and actual practice will demand interventions to alter primary care practice or the introduction of new models to gather and analyze family data. Further research is also needed to evaluate the impact of improved family history taking on health care costs and outcomes.

Acknowledgments

The authors thank Howard Rabinowitz, MD, for providing helpful suggestions in the development and execution of this project and Aliza Mansolino for the preparation of the manuscript.

The management of women who have low-grade cytologic abnormalities—including atypical squamous cells (ASC) and low-grade squamous intraepithelial lesions (LSIL)—is controversial.^1-4 Without strong evidence favoring a single approach, some clinicians recommend immediate colposcopy to obtain a definitive diagnosis and to exclude the presence of a high-grade lesion, while others recommend observation with serial Pap smears, given the tendency for many low-grade lesions to regress spontaneously.^5,6 Immediate colposcopy has the advantage of giving a patient a relatively rapid assessment of the nature and extent of her cervical dysplasia; however, the procedure is uncomfortable, and overall management may not be affected. Observation with serial Pap smears may avoid an invasive procedure, but it may also cause anxiety as time passes without a definitive diagnosis.

Eliciting and understanding patient preferences is an important part of clinical decision making. The clinician provides the best available information on the probability of clinical outcomes and the implications of each for the patient’s health. But only the patient knows what these outcomes mean to her well-being (also called “utility”).

Given the clinical disagreement over how to proceed with abnormal ASC and LSIL Pap smear results, the decision should be influenced by a patient’s preference, informed by knowledge of outcomes and costs of alternative approaches. It is unclear which approach women prefer, and whether women’s preferences for specific protocols are associated with sociodemographic characteristics. To understand better how women weigh these trade-offs, we evaluated the preferences of a diverse group of women for contrasting management approaches to the evaluation of a hypothetical low-grade abnormal Pap smear result.

Methods

Study population

Study participants were recruited from the waiting rooms of 5 family planning clinics in Northern California’s Central Valley. Women were eligible for the study if they were 18 years of age or older, or, if minors, they were emancipated and could thus provide informed consent. Potential subjects were excluded if they spoke neither English nor Spanish or if they had never had a Pap smear. The study protocol and informed consent procedures were reviewed and approved by the University of California, Davis, Human Subjects Committee.

Instruments and outcome measures

Interviews were conducted in English or Spanish. Information regarding demographic characteristics, level of education, past experiences with abnormal Pap smears and cervical cancer, and self-rated religiosity was collected with a self-administered questionnaire. The primary outcome measures were utilities (quantified preferences for specific health states) for 6 different scenarios. These were assessed by the standard gamble (SG) method, described in more detail below.⁷

Possible utility scores range from 0 to 1. A score of 0 represents immediate death; a score of 1 represents full (or ideal) health for the rest of one’s life. Because the scenarios under consideration in this study did not involve any meaningful level of risk of death, we expected utility scores for the scenarios to cluster toward the upper end of the scale. As a result, a measurement instrument based on an “immediate death” versus “full health” scale would be unable to discriminate between different scenarios. To avoid this problem, a scale was used in which the lower end point was a non-death state unambiguously less preferred than each of the scenarios under consideration.⁸ We used “invasive cervical cancer requiring hysterectomy” as the lower end point (utility of 0) contrasted with “full health with all normal Pap smears” (utility of 1) to generate the original score (SG Dys). In a separate standard gamble, subjects rated invasive cervical cancer versus immediate death (SG Ca), so that all utilities could be converted to the standard scale, using the formula: (1 – SG Ca) (SG Dys) + SG Ca.

The 6 scenarios rated in the study are shown in Table 1. The scenarios represent 3 sets of progressively more invasive interventions for a low-grade abnormal Pap smear: (1) resolution, representing spontaneous regression with treatment not required; (2) a low-grade abnormality requiring treatment with cryotherapy; (3) a more severe abnormality requiring a cervical cone biopsy. Following either spontaneous resolution or treatment, all scenarios assumed the abnormality was resolved. For each of the 3 results, a management strategy based on observation with serial Pap smears was applied in 1 instance, and a strategy of early colposcopy was applied in the other instance, resulting in 6 different pathways to the ultimate outcome; a normal Pap smear. The time frame for these scenarios was 18–36 months.

Trained interviewers used a standardized approach to elicit preferences from each subject. Subjects were read a description of all the procedures involved in the scenarios. Descriptions were accompanied by cards summarizing each procedure in pictures and words, and included information about the possibility of progression and spontaneous regression of the Pap smear abnormality. Subjects were encouraged to ask questions at any point during the interview. Procedure descriptions are available from the authors on request.

TABLE 1

Clinical scenarios classified by management approach and required treatment^*

Standard gamble

Subjects were asked their preference between the certainty of the scenario under consideration and an uncertain prospect of either having cervical cancer treated by hysterectomy or full health. A probability wheel was used as visual aid.⁹ The probability of cervical cancer was altered until the subject was indifferent between the certain scenario and the uncertain prospect. Once all 6 scenarios had been scored, each subject was asked about her preference between the certainty of cervical cancer treated by hysterectomy and the uncertain prospect of immediate death or full health, using the same method.

At the end of the interview, both the subject and the interviewer completed evaluation forms including ratings of how well the subject understood the standard gamble rating exercises. Subject confusion was also defined a priori as those placing a higher utility on scenario 3 (observation for a long period followed by cone biopsy), which represented the longest period of uncertainty followed by the most invasive procedure, than on scenario 1 (a single mildly abnormal Pap smear evaluated by observation which then resolved spontaneously), which represented the absence of any invasive procedure.

Statistical analysis

Descriptive statistics were generated for ratings of each scenario for the entire group and with the confused subjects removed. Confused subjects included those who reported they found the interview “very confusing,” those who were recorded by the interviewer as finding the interview “very confusing,” and those whose rankings met the criteria for subject confusion, as described above. Means, standard deviations, medians, and percentiles were calculated for each scenario. The mean differences in adjusted standard gamble ratings between paired scenarios was evaluated using a t distribution. Multiple regression analyses were used to explore how much between-subject variation in the standard gamble scores was explained by the variables listed above.

A simple decision tree (Figure 1) was constructed to contrast preferences for an observational approach vs early colposcopy. Outcome probabilities were derived from meta-analyses of the medical literature,⁵ from observational data obtained at the same Northern California family planning clinics,¹⁰ and, for cone biopsy outcomes, from expert opinion obtained using a modified Delphi process.¹¹ Utilities were assigned to the decision tree based on the standard gamble results. Women having 2 consecutive low-grade abnormal Pap results followed by a normal Pap result were assigned the same utility value as that for women with a single abnormal result. Analysis of the tree, including 1-way and 2-way sensitivity analysis of key variables, was conducted with Data 3.5.

Results

One hundred seventy interviews were completed. Characteristics of the interview subjects are shown in Table 2. A total of 22 subjects were designated “confused.” Analyses including the confused subjects did not alter the pattern of results, but the range in responses was larger. All analyses are presented here with confused subjects removed (n = 148).

Median ratings with 25th–75th percentiles for the paired scenarios rated by the standard gamble are shown as box plots in Figure 2. Mean adjusted scores, standard deviations, and mean differences in scores between paired scenarios are shown in Table 3. Notable findings include the following. (1) For each scenario, the range of responses by either rating method was very large. (2) Mean differences in utilities for observation vs early colposcopy were small. (3) For the paired scenarios in which the outcome was spontaneous resolution, observation was preferred (P = .01); in the paired scenarios in which the outcome was cryotherapy, early colposcopy was preferred (P = .02). (4) In the multiple regression analyses for each scenario, age, education, ethnicity, religiosity, and having known someone with cervical cancer together explained only a small amount of the variability between subjects (range for R², .09–.22).

The decision model with baseline probabilities is shown in Figure 1. The model was simplified to exclude the outcome of cervical cancer, which is a very rare outcome for women with ASC or LSIL cervical smears who have adequate follow-up.⁵ In the baseline analysis, the overall utility of early colposcopy was slightly favored over the overall utility of the observation approach (utility of observation = 0.932; utility of early colposcopy = 0.940).

Sensitivity analysis examines the effect of varying elements of the model on the outcome. In sensitivity analyses of probabilities, the early colposcopy branch was favored, but the differences were small. The maximum difference in utilities between branches was 0.012 in these sensitivity analyses. In 1-way sensitivity analysis of branch utilities, threshold utility values to favor the observation branch were 0.986 for spontaneous resolution after observation and 0.898 for early colposcopy. Threshold values for cryotherapy were 0.938 for observation and 0.938 for early colposcopy.

TABLE 2

Characteristics of study subjects (n = 170)

TABLE 3

Adjusted standard gamble values and paired differences* (n=148)

FIGURE 1Decision model comparing observation with early colposcopy *

FIGURE 2Distribution of individual utilities as assessed by the standard gamble*

Discussion

We found wide variation in women’s preferences for management approaches to a low-grade abnormal Pap smear result. The range of responses was very large and the variation between individuals rating the same scenario was substantially greater than the variation in mean ratings between different scenarios. Measured subject characteristics explained only a small proportion of the observed variation, indicating that other unmeasured factors contributed substantially to the variation. Although 25% of subjects stated they knew someone with cervical cancer, this high percentage seems improbable and more likely reflects knowledge of someone who had an abnormal Pap smear.

The decision model displayed a small preference for immediate colposcopy. This may be related to preference for quicker resolution of the concern about cancer, although it involves more procedures. Small changes in utilities for spontaneous resolution and cryotherapy influenced the model to prefer observation. For cryotherapy, these utility values were within 1 standard deviation of the mean.

Our finding of a wide variation in preferences is supported by other patient preference studies,^12-14 including 2 on this subject. Ferris et al assessed triage preferences for the evaluation and management ASC and LSIL.¹³ They used a questionnaire with a sample of 968 women who presented for care at obstetrics and gynecology and family practice clinics. They found that more women preferred repeat Pap smear when the index smear was ASC, and more women preferred colposcopy when the index smear was LSIL. Among a group of 136 Canadian women with atypia or LSIL referred for colposcopy, Meana et al found that 64% preferred early colposcopy, while 17% preferred observation and 17% had no strong preference.¹⁴

The factors contributing to patient preferences are complex. Differences in preferences may be influenced by knowledge and understanding of the disease and possible interventions, risk aversion, access to services, socioeconomics, cultural background, and other factors. While 1 patient may be most interested in establishing a definitive diagnosis and undergoing treatment as soon as possible, another may place priority on avoiding invasive or uncomfortable procedures. How differences in patient preferences influence clinical choices is highlighted by the work of Kuppermann et al.¹⁵ These investigators found that utilities for outcomes of prenatal diagnostic testing predicted subsequent testing behavior.

Our findings are limited by our use of a convenience sample of women attending family planning clinics. They may not be representative of women’s preferences in general, or even those of women attending family planning clinics. Outcomes in our study were specified during the preference assessment process; in real decision making, the outcome is always unknown at the time the decision is made. We did not include HPV typing as an option in our clinical scenarios. While HPV typing may have a role for triage of ASC,^6,16 it appears not to be useful in management of LSIL.¹⁷

Cost-effectiveness analysis would offer important information about which management approach might be favored in the context of resource allocation. For decision making by individual patients and doctors, however, decision analysis is often more relevant. In this case, the “preferred” decision is very sensitive to patient utilities, emphasizing the need for clear physician-patient communication.

Strengths of our study include the diversity of the subjects, the formal process for preference assessment, and the paired scenarios, which allow assessment of preferences for a single management decision, in which 2 separate paths lead to an equivalent ultimate outcome. Our findings are consistent with previous work showing that the sequence of events leading to an outcome will influence utilities for the outcome.¹⁸

Application to clinical practice

How might our findings be translated into clinical practice? In clinical situations where different approaches to management are unlikely to result in substantial outcome differences (a “toss-up”), patient preferences are a key aspect of the decision-making process.¹⁹ For women with lowgrade Pap abnormalities, several diagnostic options are available and no single option is strongly supported by evidence to offer better outcomes. Our study indicates that no single option is preferred by most women. Under these conditions, engaging the patient in the decision-making process may produce better health outcomes.²⁰ Clinicians should anticipate highly varied preferences, and will need to adopt a flexible approach. Not all patients will want to be actively involved in the decision process, but the desire for information is nearly universal. Flexible use of the questions in Table 4 may help patients to define their preferences and will likely improve their satisfaction and adherence to the treatment plan.

TABLE 4

Questions for patients with an abnormal Pap smear

· Acknowledgments ·

The authors thank the staff of Planned Parenthood Mar Monte East for their assistance with subject recruitment and interviews.

The management of women who have low-grade cytologic abnormalities—including atypical squamous cells (ASC) and low-grade squamous intraepithelial lesions (LSIL)—is controversial.^1-4 Without strong evidence favoring a single approach, some clinicians recommend immediate colposcopy to obtain a definitive diagnosis and to exclude the presence of a high-grade lesion, while others recommend observation with serial Pap smears, given the tendency for many low-grade lesions to regress spontaneously.^5,6 Immediate colposcopy has the advantage of giving a patient a relatively rapid assessment of the nature and extent of her cervical dysplasia; however, the procedure is uncomfortable, and overall management may not be affected. Observation with serial Pap smears may avoid an invasive procedure, but it may also cause anxiety as time passes without a definitive diagnosis.

Eliciting and understanding patient preferences is an important part of clinical decision making. The clinician provides the best available information on the probability of clinical outcomes and the implications of each for the patient’s health. But only the patient knows what these outcomes mean to her well-being (also called “utility”).

Given the clinical disagreement over how to proceed with abnormal ASC and LSIL Pap smear results, the decision should be influenced by a patient’s preference, informed by knowledge of outcomes and costs of alternative approaches. It is unclear which approach women prefer, and whether women’s preferences for specific protocols are associated with sociodemographic characteristics. To understand better how women weigh these trade-offs, we evaluated the preferences of a diverse group of women for contrasting management approaches to the evaluation of a hypothetical low-grade abnormal Pap smear result.

Methods

Study population

Study participants were recruited from the waiting rooms of 5 family planning clinics in Northern California’s Central Valley. Women were eligible for the study if they were 18 years of age or older, or, if minors, they were emancipated and could thus provide informed consent. Potential subjects were excluded if they spoke neither English nor Spanish or if they had never had a Pap smear. The study protocol and informed consent procedures were reviewed and approved by the University of California, Davis, Human Subjects Committee.

Instruments and outcome measures

Interviews were conducted in English or Spanish. Information regarding demographic characteristics, level of education, past experiences with abnormal Pap smears and cervical cancer, and self-rated religiosity was collected with a self-administered questionnaire. The primary outcome measures were utilities (quantified preferences for specific health states) for 6 different scenarios. These were assessed by the standard gamble (SG) method, described in more detail below.⁷

Possible utility scores range from 0 to 1. A score of 0 represents immediate death; a score of 1 represents full (or ideal) health for the rest of one’s life. Because the scenarios under consideration in this study did not involve any meaningful level of risk of death, we expected utility scores for the scenarios to cluster toward the upper end of the scale. As a result, a measurement instrument based on an “immediate death” versus “full health” scale would be unable to discriminate between different scenarios. To avoid this problem, a scale was used in which the lower end point was a non-death state unambiguously less preferred than each of the scenarios under consideration.⁸ We used “invasive cervical cancer requiring hysterectomy” as the lower end point (utility of 0) contrasted with “full health with all normal Pap smears” (utility of 1) to generate the original score (SG Dys). In a separate standard gamble, subjects rated invasive cervical cancer versus immediate death (SG Ca), so that all utilities could be converted to the standard scale, using the formula: (1 – SG Ca) (SG Dys) + SG Ca.

The 6 scenarios rated in the study are shown in Table 1. The scenarios represent 3 sets of progressively more invasive interventions for a low-grade abnormal Pap smear: (1) resolution, representing spontaneous regression with treatment not required; (2) a low-grade abnormality requiring treatment with cryotherapy; (3) a more severe abnormality requiring a cervical cone biopsy. Following either spontaneous resolution or treatment, all scenarios assumed the abnormality was resolved. For each of the 3 results, a management strategy based on observation with serial Pap smears was applied in 1 instance, and a strategy of early colposcopy was applied in the other instance, resulting in 6 different pathways to the ultimate outcome; a normal Pap smear. The time frame for these scenarios was 18–36 months.

Trained interviewers used a standardized approach to elicit preferences from each subject. Subjects were read a description of all the procedures involved in the scenarios. Descriptions were accompanied by cards summarizing each procedure in pictures and words, and included information about the possibility of progression and spontaneous regression of the Pap smear abnormality. Subjects were encouraged to ask questions at any point during the interview. Procedure descriptions are available from the authors on request.

TABLE 1

Clinical scenarios classified by management approach and required treatment^*

Standard gamble

Subjects were asked their preference between the certainty of the scenario under consideration and an uncertain prospect of either having cervical cancer treated by hysterectomy or full health. A probability wheel was used as visual aid.⁹ The probability of cervical cancer was altered until the subject was indifferent between the certain scenario and the uncertain prospect. Once all 6 scenarios had been scored, each subject was asked about her preference between the certainty of cervical cancer treated by hysterectomy and the uncertain prospect of immediate death or full health, using the same method.

At the end of the interview, both the subject and the interviewer completed evaluation forms including ratings of how well the subject understood the standard gamble rating exercises. Subject confusion was also defined a priori as those placing a higher utility on scenario 3 (observation for a long period followed by cone biopsy), which represented the longest period of uncertainty followed by the most invasive procedure, than on scenario 1 (a single mildly abnormal Pap smear evaluated by observation which then resolved spontaneously), which represented the absence of any invasive procedure.

Statistical analysis

Descriptive statistics were generated for ratings of each scenario for the entire group and with the confused subjects removed. Confused subjects included those who reported they found the interview “very confusing,” those who were recorded by the interviewer as finding the interview “very confusing,” and those whose rankings met the criteria for subject confusion, as described above. Means, standard deviations, medians, and percentiles were calculated for each scenario. The mean differences in adjusted standard gamble ratings between paired scenarios was evaluated using a t distribution. Multiple regression analyses were used to explore how much between-subject variation in the standard gamble scores was explained by the variables listed above.

A simple decision tree (Figure 1) was constructed to contrast preferences for an observational approach vs early colposcopy. Outcome probabilities were derived from meta-analyses of the medical literature,⁵ from observational data obtained at the same Northern California family planning clinics,¹⁰ and, for cone biopsy outcomes, from expert opinion obtained using a modified Delphi process.¹¹ Utilities were assigned to the decision tree based on the standard gamble results. Women having 2 consecutive low-grade abnormal Pap results followed by a normal Pap result were assigned the same utility value as that for women with a single abnormal result. Analysis of the tree, including 1-way and 2-way sensitivity analysis of key variables, was conducted with Data 3.5.

Results

One hundred seventy interviews were completed. Characteristics of the interview subjects are shown in Table 2. A total of 22 subjects were designated “confused.” Analyses including the confused subjects did not alter the pattern of results, but the range in responses was larger. All analyses are presented here with confused subjects removed (n = 148).

Median ratings with 25th–75th percentiles for the paired scenarios rated by the standard gamble are shown as box plots in Figure 2. Mean adjusted scores, standard deviations, and mean differences in scores between paired scenarios are shown in Table 3. Notable findings include the following. (1) For each scenario, the range of responses by either rating method was very large. (2) Mean differences in utilities for observation vs early colposcopy were small. (3) For the paired scenarios in which the outcome was spontaneous resolution, observation was preferred (P = .01); in the paired scenarios in which the outcome was cryotherapy, early colposcopy was preferred (P = .02). (4) In the multiple regression analyses for each scenario, age, education, ethnicity, religiosity, and having known someone with cervical cancer together explained only a small amount of the variability between subjects (range for R², .09–.22).

The decision model with baseline probabilities is shown in Figure 1. The model was simplified to exclude the outcome of cervical cancer, which is a very rare outcome for women with ASC or LSIL cervical smears who have adequate follow-up.⁵ In the baseline analysis, the overall utility of early colposcopy was slightly favored over the overall utility of the observation approach (utility of observation = 0.932; utility of early colposcopy = 0.940).

Sensitivity analysis examines the effect of varying elements of the model on the outcome. In sensitivity analyses of probabilities, the early colposcopy branch was favored, but the differences were small. The maximum difference in utilities between branches was 0.012 in these sensitivity analyses. In 1-way sensitivity analysis of branch utilities, threshold utility values to favor the observation branch were 0.986 for spontaneous resolution after observation and 0.898 for early colposcopy. Threshold values for cryotherapy were 0.938 for observation and 0.938 for early colposcopy.

TABLE 2

Characteristics of study subjects (n = 170)

TABLE 3

Adjusted standard gamble values and paired differences* (n=148)

FIGURE 1Decision model comparing observation with early colposcopy *

FIGURE 2Distribution of individual utilities as assessed by the standard gamble*

Discussion

We found wide variation in women’s preferences for management approaches to a low-grade abnormal Pap smear result. The range of responses was very large and the variation between individuals rating the same scenario was substantially greater than the variation in mean ratings between different scenarios. Measured subject characteristics explained only a small proportion of the observed variation, indicating that other unmeasured factors contributed substantially to the variation. Although 25% of subjects stated they knew someone with cervical cancer, this high percentage seems improbable and more likely reflects knowledge of someone who had an abnormal Pap smear.

The decision model displayed a small preference for immediate colposcopy. This may be related to preference for quicker resolution of the concern about cancer, although it involves more procedures. Small changes in utilities for spontaneous resolution and cryotherapy influenced the model to prefer observation. For cryotherapy, these utility values were within 1 standard deviation of the mean.

Our finding of a wide variation in preferences is supported by other patient preference studies,^12-14 including 2 on this subject. Ferris et al assessed triage preferences for the evaluation and management ASC and LSIL.¹³ They used a questionnaire with a sample of 968 women who presented for care at obstetrics and gynecology and family practice clinics. They found that more women preferred repeat Pap smear when the index smear was ASC, and more women preferred colposcopy when the index smear was LSIL. Among a group of 136 Canadian women with atypia or LSIL referred for colposcopy, Meana et al found that 64% preferred early colposcopy, while 17% preferred observation and 17% had no strong preference.¹⁴

The factors contributing to patient preferences are complex. Differences in preferences may be influenced by knowledge and understanding of the disease and possible interventions, risk aversion, access to services, socioeconomics, cultural background, and other factors. While 1 patient may be most interested in establishing a definitive diagnosis and undergoing treatment as soon as possible, another may place priority on avoiding invasive or uncomfortable procedures. How differences in patient preferences influence clinical choices is highlighted by the work of Kuppermann et al.¹⁵ These investigators found that utilities for outcomes of prenatal diagnostic testing predicted subsequent testing behavior.

Our findings are limited by our use of a convenience sample of women attending family planning clinics. They may not be representative of women’s preferences in general, or even those of women attending family planning clinics. Outcomes in our study were specified during the preference assessment process; in real decision making, the outcome is always unknown at the time the decision is made. We did not include HPV typing as an option in our clinical scenarios. While HPV typing may have a role for triage of ASC,^6,16 it appears not to be useful in management of LSIL.¹⁷

Cost-effectiveness analysis would offer important information about which management approach might be favored in the context of resource allocation. For decision making by individual patients and doctors, however, decision analysis is often more relevant. In this case, the “preferred” decision is very sensitive to patient utilities, emphasizing the need for clear physician-patient communication.

Strengths of our study include the diversity of the subjects, the formal process for preference assessment, and the paired scenarios, which allow assessment of preferences for a single management decision, in which 2 separate paths lead to an equivalent ultimate outcome. Our findings are consistent with previous work showing that the sequence of events leading to an outcome will influence utilities for the outcome.¹⁸

Application to clinical practice

How might our findings be translated into clinical practice? In clinical situations where different approaches to management are unlikely to result in substantial outcome differences (a “toss-up”), patient preferences are a key aspect of the decision-making process.¹⁹ For women with lowgrade Pap abnormalities, several diagnostic options are available and no single option is strongly supported by evidence to offer better outcomes. Our study indicates that no single option is preferred by most women. Under these conditions, engaging the patient in the decision-making process may produce better health outcomes.²⁰ Clinicians should anticipate highly varied preferences, and will need to adopt a flexible approach. Not all patients will want to be actively involved in the decision process, but the desire for information is nearly universal. Flexible use of the questions in Table 4 may help patients to define their preferences and will likely improve their satisfaction and adherence to the treatment plan.

TABLE 4

Questions for patients with an abnormal Pap smear

· Acknowledgments ·

The authors thank the staff of Planned Parenthood Mar Monte East for their assistance with subject recruitment and interviews.

The management of women who have low-grade cytologic abnormalities—including atypical squamous cells (ASC) and low-grade squamous intraepithelial lesions (LSIL)—is controversial.^1-4 Without strong evidence favoring a single approach, some clinicians recommend immediate colposcopy to obtain a definitive diagnosis and to exclude the presence of a high-grade lesion, while others recommend observation with serial Pap smears, given the tendency for many low-grade lesions to regress spontaneously.^5,6 Immediate colposcopy has the advantage of giving a patient a relatively rapid assessment of the nature and extent of her cervical dysplasia; however, the procedure is uncomfortable, and overall management may not be affected. Observation with serial Pap smears may avoid an invasive procedure, but it may also cause anxiety as time passes without a definitive diagnosis.

Eliciting and understanding patient preferences is an important part of clinical decision making. The clinician provides the best available information on the probability of clinical outcomes and the implications of each for the patient’s health. But only the patient knows what these outcomes mean to her well-being (also called “utility”).

Given the clinical disagreement over how to proceed with abnormal ASC and LSIL Pap smear results, the decision should be influenced by a patient’s preference, informed by knowledge of outcomes and costs of alternative approaches. It is unclear which approach women prefer, and whether women’s preferences for specific protocols are associated with sociodemographic characteristics. To understand better how women weigh these trade-offs, we evaluated the preferences of a diverse group of women for contrasting management approaches to the evaluation of a hypothetical low-grade abnormal Pap smear result.

Methods

Study population

Study participants were recruited from the waiting rooms of 5 family planning clinics in Northern California’s Central Valley. Women were eligible for the study if they were 18 years of age or older, or, if minors, they were emancipated and could thus provide informed consent. Potential subjects were excluded if they spoke neither English nor Spanish or if they had never had a Pap smear. The study protocol and informed consent procedures were reviewed and approved by the University of California, Davis, Human Subjects Committee.

Instruments and outcome measures

Interviews were conducted in English or Spanish. Information regarding demographic characteristics, level of education, past experiences with abnormal Pap smears and cervical cancer, and self-rated religiosity was collected with a self-administered questionnaire. The primary outcome measures were utilities (quantified preferences for specific health states) for 6 different scenarios. These were assessed by the standard gamble (SG) method, described in more detail below.⁷

Possible utility scores range from 0 to 1. A score of 0 represents immediate death; a score of 1 represents full (or ideal) health for the rest of one’s life. Because the scenarios under consideration in this study did not involve any meaningful level of risk of death, we expected utility scores for the scenarios to cluster toward the upper end of the scale. As a result, a measurement instrument based on an “immediate death” versus “full health” scale would be unable to discriminate between different scenarios. To avoid this problem, a scale was used in which the lower end point was a non-death state unambiguously less preferred than each of the scenarios under consideration.⁸ We used “invasive cervical cancer requiring hysterectomy” as the lower end point (utility of 0) contrasted with “full health with all normal Pap smears” (utility of 1) to generate the original score (SG Dys). In a separate standard gamble, subjects rated invasive cervical cancer versus immediate death (SG Ca), so that all utilities could be converted to the standard scale, using the formula: (1 – SG Ca) (SG Dys) + SG Ca.

The 6 scenarios rated in the study are shown in Table 1. The scenarios represent 3 sets of progressively more invasive interventions for a low-grade abnormal Pap smear: (1) resolution, representing spontaneous regression with treatment not required; (2) a low-grade abnormality requiring treatment with cryotherapy; (3) a more severe abnormality requiring a cervical cone biopsy. Following either spontaneous resolution or treatment, all scenarios assumed the abnormality was resolved. For each of the 3 results, a management strategy based on observation with serial Pap smears was applied in 1 instance, and a strategy of early colposcopy was applied in the other instance, resulting in 6 different pathways to the ultimate outcome; a normal Pap smear. The time frame for these scenarios was 18–36 months.

Trained interviewers used a standardized approach to elicit preferences from each subject. Subjects were read a description of all the procedures involved in the scenarios. Descriptions were accompanied by cards summarizing each procedure in pictures and words, and included information about the possibility of progression and spontaneous regression of the Pap smear abnormality. Subjects were encouraged to ask questions at any point during the interview. Procedure descriptions are available from the authors on request.

TABLE 1

Clinical scenarios classified by management approach and required treatment^*

Standard gamble

Subjects were asked their preference between the certainty of the scenario under consideration and an uncertain prospect of either having cervical cancer treated by hysterectomy or full health. A probability wheel was used as visual aid.⁹ The probability of cervical cancer was altered until the subject was indifferent between the certain scenario and the uncertain prospect. Once all 6 scenarios had been scored, each subject was asked about her preference between the certainty of cervical cancer treated by hysterectomy and the uncertain prospect of immediate death or full health, using the same method.

At the end of the interview, both the subject and the interviewer completed evaluation forms including ratings of how well the subject understood the standard gamble rating exercises. Subject confusion was also defined a priori as those placing a higher utility on scenario 3 (observation for a long period followed by cone biopsy), which represented the longest period of uncertainty followed by the most invasive procedure, than on scenario 1 (a single mildly abnormal Pap smear evaluated by observation which then resolved spontaneously), which represented the absence of any invasive procedure.

Statistical analysis

Descriptive statistics were generated for ratings of each scenario for the entire group and with the confused subjects removed. Confused subjects included those who reported they found the interview “very confusing,” those who were recorded by the interviewer as finding the interview “very confusing,” and those whose rankings met the criteria for subject confusion, as described above. Means, standard deviations, medians, and percentiles were calculated for each scenario. The mean differences in adjusted standard gamble ratings between paired scenarios was evaluated using a t distribution. Multiple regression analyses were used to explore how much between-subject variation in the standard gamble scores was explained by the variables listed above.

A simple decision tree (Figure 1) was constructed to contrast preferences for an observational approach vs early colposcopy. Outcome probabilities were derived from meta-analyses of the medical literature,⁵ from observational data obtained at the same Northern California family planning clinics,¹⁰ and, for cone biopsy outcomes, from expert opinion obtained using a modified Delphi process.¹¹ Utilities were assigned to the decision tree based on the standard gamble results. Women having 2 consecutive low-grade abnormal Pap results followed by a normal Pap result were assigned the same utility value as that for women with a single abnormal result. Analysis of the tree, including 1-way and 2-way sensitivity analysis of key variables, was conducted with Data 3.5.

Results

One hundred seventy interviews were completed. Characteristics of the interview subjects are shown in Table 2. A total of 22 subjects were designated “confused.” Analyses including the confused subjects did not alter the pattern of results, but the range in responses was larger. All analyses are presented here with confused subjects removed (n = 148).

Median ratings with 25th–75th percentiles for the paired scenarios rated by the standard gamble are shown as box plots in Figure 2. Mean adjusted scores, standard deviations, and mean differences in scores between paired scenarios are shown in Table 3. Notable findings include the following. (1) For each scenario, the range of responses by either rating method was very large. (2) Mean differences in utilities for observation vs early colposcopy were small. (3) For the paired scenarios in which the outcome was spontaneous resolution, observation was preferred (P = .01); in the paired scenarios in which the outcome was cryotherapy, early colposcopy was preferred (P = .02). (4) In the multiple regression analyses for each scenario, age, education, ethnicity, religiosity, and having known someone with cervical cancer together explained only a small amount of the variability between subjects (range for R², .09–.22).

The decision model with baseline probabilities is shown in Figure 1. The model was simplified to exclude the outcome of cervical cancer, which is a very rare outcome for women with ASC or LSIL cervical smears who have adequate follow-up.⁵ In the baseline analysis, the overall utility of early colposcopy was slightly favored over the overall utility of the observation approach (utility of observation = 0.932; utility of early colposcopy = 0.940).

Sensitivity analysis examines the effect of varying elements of the model on the outcome. In sensitivity analyses of probabilities, the early colposcopy branch was favored, but the differences were small. The maximum difference in utilities between branches was 0.012 in these sensitivity analyses. In 1-way sensitivity analysis of branch utilities, threshold utility values to favor the observation branch were 0.986 for spontaneous resolution after observation and 0.898 for early colposcopy. Threshold values for cryotherapy were 0.938 for observation and 0.938 for early colposcopy.

TABLE 2

Characteristics of study subjects (n = 170)

TABLE 3

Adjusted standard gamble values and paired differences* (n=148)

FIGURE 1Decision model comparing observation with early colposcopy *

FIGURE 2Distribution of individual utilities as assessed by the standard gamble*

Discussion

We found wide variation in women’s preferences for management approaches to a low-grade abnormal Pap smear result. The range of responses was very large and the variation between individuals rating the same scenario was substantially greater than the variation in mean ratings between different scenarios. Measured subject characteristics explained only a small proportion of the observed variation, indicating that other unmeasured factors contributed substantially to the variation. Although 25% of subjects stated they knew someone with cervical cancer, this high percentage seems improbable and more likely reflects knowledge of someone who had an abnormal Pap smear.

The decision model displayed a small preference for immediate colposcopy. This may be related to preference for quicker resolution of the concern about cancer, although it involves more procedures. Small changes in utilities for spontaneous resolution and cryotherapy influenced the model to prefer observation. For cryotherapy, these utility values were within 1 standard deviation of the mean.

Our finding of a wide variation in preferences is supported by other patient preference studies,^12-14 including 2 on this subject. Ferris et al assessed triage preferences for the evaluation and management ASC and LSIL.¹³ They used a questionnaire with a sample of 968 women who presented for care at obstetrics and gynecology and family practice clinics. They found that more women preferred repeat Pap smear when the index smear was ASC, and more women preferred colposcopy when the index smear was LSIL. Among a group of 136 Canadian women with atypia or LSIL referred for colposcopy, Meana et al found that 64% preferred early colposcopy, while 17% preferred observation and 17% had no strong preference.¹⁴

The factors contributing to patient preferences are complex. Differences in preferences may be influenced by knowledge and understanding of the disease and possible interventions, risk aversion, access to services, socioeconomics, cultural background, and other factors. While 1 patient may be most interested in establishing a definitive diagnosis and undergoing treatment as soon as possible, another may place priority on avoiding invasive or uncomfortable procedures. How differences in patient preferences influence clinical choices is highlighted by the work of Kuppermann et al.¹⁵ These investigators found that utilities for outcomes of prenatal diagnostic testing predicted subsequent testing behavior.

Our findings are limited by our use of a convenience sample of women attending family planning clinics. They may not be representative of women’s preferences in general, or even those of women attending family planning clinics. Outcomes in our study were specified during the preference assessment process; in real decision making, the outcome is always unknown at the time the decision is made. We did not include HPV typing as an option in our clinical scenarios. While HPV typing may have a role for triage of ASC,^6,16 it appears not to be useful in management of LSIL.¹⁷

Cost-effectiveness analysis would offer important information about which management approach might be favored in the context of resource allocation. For decision making by individual patients and doctors, however, decision analysis is often more relevant. In this case, the “preferred” decision is very sensitive to patient utilities, emphasizing the need for clear physician-patient communication.

Strengths of our study include the diversity of the subjects, the formal process for preference assessment, and the paired scenarios, which allow assessment of preferences for a single management decision, in which 2 separate paths lead to an equivalent ultimate outcome. Our findings are consistent with previous work showing that the sequence of events leading to an outcome will influence utilities for the outcome.¹⁸

Application to clinical practice

How might our findings be translated into clinical practice? In clinical situations where different approaches to management are unlikely to result in substantial outcome differences (a “toss-up”), patient preferences are a key aspect of the decision-making process.¹⁹ For women with lowgrade Pap abnormalities, several diagnostic options are available and no single option is strongly supported by evidence to offer better outcomes. Our study indicates that no single option is preferred by most women. Under these conditions, engaging the patient in the decision-making process may produce better health outcomes.²⁰ Clinicians should anticipate highly varied preferences, and will need to adopt a flexible approach. Not all patients will want to be actively involved in the decision process, but the desire for information is nearly universal. Flexible use of the questions in Table 4 may help patients to define their preferences and will likely improve their satisfaction and adherence to the treatment plan.

TABLE 4

Questions for patients with an abnormal Pap smear

· Acknowledgments ·

The authors thank the staff of Planned Parenthood Mar Monte East for their assistance with subject recruitment and interviews.

Over the past half century, changing medical technology, law, education, ethics, and research have influenced the current shape of physician-patient interactions.⁹ In 1956, the traditional model of Activity-Passivity (physician does something to the patient) was challenged with the revolutionary concept of active patient participation.¹⁰ The models of Guidance and Cooperation (physician tells patient what to do, patient cooperates) and Mutual Participation (physician enables patient to help him/herself, patient is a partner) were proposed¹⁰ and are reflected in modern theoretically-based interaction models. Numerous models have been proposed as variants of the Guidance/Cooperation model (eg, paternalistic model,¹¹ priestly model,¹² contractual model¹³) and the Mutual Participation model (eg, ethnographic model,¹⁴ consumerist model,^11,15 family systems model¹⁶). Few of these models, though, have been empirically evaluated. The best-developed and most-studied mutual participation model is the patient-centered method.^5,17-20

When data have been collected using quantitative or qualitative approaches, significant strides have been made in understanding physician-patient interaction^{3, 21-23} and the effect of such interactions on patient outcomes,^5,24,25 primarily patient satisfaction.^1,26-29 However, many studies have been limited by their focus on a narrow aspect of physician-patient communication, studying a small number of physicians or patients, and using medical students, residents, and hospital faculty as study subjects.

The purpose of this study was not to develop a new model of physician-patient interaction. Rather, variables characterizing physician style grounded by the direct observation of thousands of encounters for 138 community practicing family physicians were used to empirically cluster physicians into groups that represent distinct interaction styles. Because interaction style may be manifested in all phases of a patient encounter, we used as a guiding framework the 3 primary functions of an interview:^30,31gathering information, enhancing a healing relationship, and making and implementing decisions. The importance of each of these functions varies depending on the nature of the encounter, but our overall approach provides a practical way of conceptualizing physician-patient interaction style. The association of the empirically derived and theoretically-based physician styles are tested with 3 outcomes: 1) patient report of delivery of attributes of primary care measured using the Components of Primary Care Instrument (CPCI), 2) patient satisfaction with the visit, and 3) the duration of the visit.

Methods

This study was part of the larger Direct Observation of Primary Care (DOPC) study, a cross sectional observational study that examined the content of 4454 outpatient visits to family physicians in northeast Ohio. Details of the methods of the DOPC study have been described extensively elsewhere.^32,34 Briefly, 4 teams of 2 research nurses directly observed consecutive patient visits to 138 participating physicians in 84 practices between October 1994 and August 1995. The research nurses collected data on the content and context of consecutive office visits using the following methods: direct observation of the patient visit, patient exit questionnaire, medical record review, and collection of ethnographic field notes.^33,34

Measures

Patients’ perception of the delivery 5 attributes of primary care was measured by the Components of Primary Care Instrument (CPCI). Interpersonal communication was an evaluation of the ease of exchange of information between patient and physician. The physician’s accumulated knowledge about the patient refers to the physician’s understanding of the patient’s medical history, health care needs, and values. Coordination of care refers to the information received from referrals to specialists and previous health care visits, and its incorporation into the current and future care of the patient. Preference to see usual physician refers to the degree to which patients believed and valued that they could go to their regular physician for almost all problems. Scale scores demonstrate good internal consistency reliability (Cronbach’s alpha: .68–.79).³⁵ Continuity of care is measured by the Usual Provider Continuity index (UPC), which is the proportion of visits to the patient’s regular doctor in the past year out of the total number of physician visits in the past year.

Patient satisfaction was measured using the 4 physician-specific items from the MOS 9 Item Visit Rating Form³⁶ (Cronbach’s alpha = .89).³³ Also included on the patient survey was a single item assessing the degree to which patients’ expectations with the visit were met. Duration of the visit was the total face-to-face time the physician spent with the patient and was measured by direct observation.

Each physician’s interaction style was determined through a 2-step process. In the first step, ethnographic field notes were used to gather information that helps define core features of physician style. The field notes from 4 days of observation of 138 family physicians in 84 practices were transcribed and imported into FolioVIEWS37 for data management and coding. Analysis was conducted with an immersion-crystallization approach³⁸ involving repetitive reading and summarization of the text data. Case summaries were constructed from a sample of practices selected to maximize variation among practice characteristics such as size, physician sex, and practice location. The case summaries were independently reviewed, and important features were identified. These features were cross-checked against the original data. This process, and the resulting 30 features, are described in detail elsewhere.³²

Six of the features that emerged from the qualitative analyses pertain to physician style and are listed in Table 1. Each of the 3 primary interview functions³⁰ is represented by at least 1 feature, ensuring good coverage of the core aspects of the interaction. Gathering information is shaped by physician orientation and the clinical information allowed or elicited in the visits. Enhancing healing relationships is realized in part through affective connection with patients. The final function, making and implementing decisions, is influenced by the level of control or shared power with patients, the physician’s openness to patients’ agendas, and the physician’s willingness to negotiate options with patients.

The second step involved a cluster analysis of the 6 variables. First a hierarchical approach was used to estimate the number of clusters. Then a non-hierarchical clustering approach was used to determine physician classification among the clusters and the features that distinguish the clusters.³⁹ Analysis of variance was used to confirm that variables included in the cluster analysis significantly differed between at least 2 of the identified clusters, and thus were contributing to defining interaction style.

TABLE 1

Physician style variables

Analyses

The association of physician and patient characteristics with interaction style was assessed by chi square for categorical variables and by analysis of variance for continuous variables. The association of physician style with each of the 5 attributes of primary care measured by the CPCI, the indicators of patient satisfaction, and duration of the visit were tested using multilevel modeling,⁴⁰ to account for the hierarchical nature of data (ie, patients nested within physicians).

Results

Of the 4994 patients presenting for care by their family physicians, 4454 (89%) agreed to participate in the DOPC study. Physicians participating in the DOPC study were similar in age to national samples of family physicians, but over-represented female and residency-trained physicians.³⁴ Patient age, sex, and race were similar to the population of patients seeing family physicians and general practitioners nationally as reported in the National Ambulatory Medical Care Survey.³⁴ Patient questionnaires were returned by 3283 (74%) of the patients. Of those respondents, 2881 satisfactorily completed the CPCI, representing 88% of those returning a patient questionnaire and 65% of the total sample. The patients who completed the CPCI were more likely to be white, have private health care insurance, and be somewhat older than patients who did not complete the CPCI.³⁵

The cluster analysis identified 4 distinct groups of physicians. Each of the 138 physicians was classified into 1 group. Each of the 6 variables in the analysis contributed to defining the 4 groups by significantly (P

Forty-nine percent of physicians were classified as person focused. These physicians were more focused on the person than the disease, were perceived as personable and friendly, were open to the patient’s agenda, and frequently negotiated options with the patient. Physicians classified as biopsychosocial (16%) were more focused on the patient’s disease, but elicited psychosocial clinical information. Physicians classified as biomedical (20%) were also more focused on the patient’s disease and were unlikely to elicit psychosocial information. These physicians also demonstrated a low level of friendliness and were unlikely to negotiate options with the patient. The high physician control group’s major characteristics were domination of the encounter and disregard of the patient’s agenda (14%).

Association of physician characteristics with the interaction styles is presented in Table 2. The percent of male and female physicians differed greatly among the 4 style groups. The proportion of female physicians in the person-focused group was almost 4 times that of the biopsychosocial group and the high physician control group (P

As reported in Table 3, physician style is significantly associated with 3 of the 5 patient reports of the attributes of primary care. Physicians classified as having a person-focused approach have the highest mean score of communication; the other 3 styles score lower, with the high-physician-control style scoring the lowest. Person-focused and biopsychosocial physicians scored highest on patient reports of accumulated knowledge; those in the biomedical group scored the lowest. Coordination of care was highest among the person-focused group and lowest among the high-control group Across the different types of physician style, there was no difference in patient report of preference for his or her regular physician or the measure of continuity of care.

The associations of physician style with 2 indicators of patient satisfaction are displayed in Table 4. The highest group mean of patient satisfaction is for the person-focused style, and the lowest is for the high-physician-control group. The indicator of the degree to which patient expectations were met also follows this pattern. Also displayed in Table 4, the person-focused style demonstrated the longest average duration of visit, at 11.5 minutes; the high-physician-control group visits were the shortest in duration, at about 9.5 minutes.

TABLE 2

Physician and patient characteristics associated with interaction style

Association of physician style with attributes of primary care¹

Association of physician interaction style with patient satisfaction and duration of visit¹

Discussion

These data indicate that a person-focused approach is actively used in community practice, and is the style most congruent with patient-reported quality of primary care and satisfaction with care. Our data, in concert with data reported by others,^5,24 indicate strong support for the feasibility and value of the person-focused model. We found that, of the 4 distinct interaction styles, physicians with the person-focused style scored highest across all measures of the attributes of primary care and on the indicators of patient satisfaction, with the exception of continuity of care. In contrast, physicians with the high-control style were generally lowest on the primary care and satisfaction indicators.

It is important to emphasize that, even though the vast majority of patients in this sample are likely to have self-selected their primary care physician, patient rating of some attributes of primary care differed across the 4 physician styles. Patients of physicians with different styles equally valued seeing their regular physician, as reported by the preference-for-their-regular-doctor score; they exhibited similar proportions of continuity visits in the past year; and their satisfaction scores were all generally high. Patients appear to want to see their regular physician, regardless of interaction approach, even though some approaches—particularly the high-physician-control style—were rated poorer for communication, coordination of care, and accumulated knowledge.

There may be several explanations as to why a particular physician style is associated with specific patient reports of communication, accumulated knowledge, and coordination of care. Openness to the patient’s agenda and willingness to negotiate options—as was characteristic of the person-focused physicians—may facilitate good communication and convey an understanding of patient preferences and values regarding health. It is interesting to note that different groups scored lowest on some of the attributes of primary care. The high-physician-control group was the lowest on interpersonal communication and coordination of care. High-control physicians were more likely to dominate the agenda and the verbal exchanges. Patients may have felt they could not ask questions or that the physician did not listen to what they tried to say. The biomedical group of physicians were given the lowest scores by patients on accumulated knowledge, suggesting that patients thought these physicians were less likely to know their preferences and values regarding health care, know less about them as persons, and know less about their family and medical histories.

As others have proposed, we concur that interaction style is not a dichotomy or even a continuum of patient versus physician control, but is multidimensional, cutting across the main functions of the patient encounter (ie, information gathering, relationship building, and making and implementing decisions). These data provide some confirmation for the original scheme proposed by Szasz and Hollander,¹⁰ with the Mutual Participation model most represented by the person-focused approach and the Activity-Passivity model most represented by the high-physician-control group. The biopsychosocial and biomedical approaches represent different versions of the Guidance and Cooperative model.

The 4 types of physician style empirically derived from our data are similar to communication pattern types found by Roter et al,²⁷ in a study with similar aims but different methods. Of the 5 types reported, narrowly biomedical and expanded biomedical accounted for 65% of visits, and biopsychosocial accounted for 20%. Psychosocial and consumerist (distinguished by a high degree of patient questions) accounted for only 8% each. It is interesting that in our data, we found the person-focused style was by far the most common approach (49%) among this group of family physicians. These differences in use of particular interaction styles may have several explanations. First, these data were collected more recently.²⁷ Thus our data may reflect trends in a movement away from a paternalistic style and toward an increased patient participatory style. Second, our sample consisted entirely of family physicians practicing in the community, where the model of person-focused care may have a longer history of support and endorsement or be of greater importance to community family physicians, whose emphasis is on a breadth of care based on patient needs.^6,7,18

Physicians with a person-focused style granted the longest visits, while high-control-physicians granted the shortest—a difference of more than 2 minutes per visit on average. The associations were not explained away by accounting for patient or physician characteristics, suggesting that a person-focused style may require more time. However, others have found that physicians engaging in a patient participatory style had office visits that were of similar duration as found with other approaches,^{23, 27} although the average duration of visit for both of these studies were considerably longer than the office visits among our sample.

This study has several strengths. The use of community practicing physicians in real world conditions for whom visits were similar in content to the visits reported by NAMCS³⁴ adds to the generalizability of the findings. We have used an integration of qualitative and quantitative approaches to empirically derive categories of physician interaction style. Our data are based on nurse observation of an average of 32 encounters per physician and documented in rich and comprehensive qualitative fieldnotes. And finally, by using multilevel modeling, we have reported an honest estimate of the association of physician style and patient report of primary care by appropriately modeling the nested data structure.

The findings must be interpreted in light of potential study limitations. First, the patients who did not complete the patient questionnaire are somewhat different demographically than those patients who did complete it. However, non-completion of the questionnaire was not associated with physician style; therefore, it is unlikely that the associations would change, had these individuals been included. Second, because the study was cross-sectional we cannot control for patient self-selection of physicians. Nonetheless, since patients dissatisfied with the quality of care are likely to seek another physician, we would expect patient self-selection of physicians to bias the study toward the null, thus making our results even more remarkable.

These findings, in combination with the literature on the person-focused,²⁴ patient-centered^{5,17,19,20,41} and relationship-centered approaches,⁴² provide strong evidence to support the widespread implementation of this physician-patient interaction approach. Further investigation in community practice may lead to identification of ways to support and encourage person-focused care and the time needed to provide such care.

· Acknowledgments ·

The authors are indebted to the physicians, office staff members, and patients without whose participation this study would not have been possible. This paper was improved by helpful suggestions on an earlier draft by Kurt C. Stange, MD, PhD. This study was supported by a grant from the National Cancer Institute (1R01 CA60862) and in part by the Center for Research in Family Practice and Primary Care and the American Academy of Family Practice.