The Probability of Specific Diagnoses for Patients Presenting with Common Symptoms to Dutch Family Physicians

Estimating the probability of disease in unselected patients lies at the heart of the clinical competence of family physicians (FPs).^1,2 This probability is called the prior or pretest probability, because it precedes any diagnostic testing, including the history and physical examination. Based on this knowledge, FPs often decide that since the probability of a serious disease is low, the best thing to do is watchful waiting, thus preventing unnecessary harm and cost to the patient.^3,4 Moderate probabilities may trigger a diagnostic evaluation, and high probabilities may warrant empiric therapy without further diagnostic confirmation. For example, knowing that the probability of gastric carcinoma is exceedingly rare for a dyspeptic patient younger than 40 years supports a diagnostic approach that does not include initial endoscopy. A moderate probability of strep throat is a situation where a rapid strep test may be helpful.⁵ Understanding the high prior probability of a urinary tract infection in young healthy women with dysuria might allow an FP to confidently institute a telephone-based protocol, including empiric treatment for selected patients.

The development of a prior probabilities database requires access to large diverse practice populations with adequate continuity of care and good documentation of all episodes of care.^2,6,7 An “episode of care” is defined as a health problem from the first encounter with a health care provider through the completion of the last encounter related to that particular problem.^2,8.10,11 Early in an episode of care, FPs will often assign a symptom diagnosis such as chest pain, gastric pain, or otalgia; a substantial number of these diagnoses will change over time. Therefore, the observation period should allow for the most important modifications of diagnoses to become visible. Because 1-year incidence and prevalence rates are usually calculated, at least a 1-year study period is required; however, a longer period is preferable, especially for information on chronic diseases.¹²

Episodes of care are based on the relationship between the patient’s reasons for encounter (RFE), the physician’s diagnostic interpretation, and the related interventions over time.¹³ Episodes of care are clearly distinguished from episodes of disease and episodes of illness. An episode of disease begins with its onset and continues until its resolution or the patient’s death, while an episode of illness refers to the period that someone suffers from symptoms or complaints experienced as illness. Not every disease, and certainly not every illness, results in an episode of care.^2,13,14 Most episodes of care, however, are part of an episode of disease and/or illness. Health maintenance episodes can be considered a special form of episodes of care.^10,11 For example, screening for breast cancer (an episode of health maintenance) may prove the existence of the episode of disease well before the patient has symptoms (an episode of illness); an episode of care will follow.

The development of a prior probabilities database also requires a primary care–specific system of classification. The International Classification of Primary Care (ICPC) includes approximately 200 symptoms/complaints and 300 diagnoses common in family practice. Almost all have an incidence of at least 1 per 1000 patients per year.^13,15

Since 1985, the members of the Transition Project of the University of Amsterdam Department of Family Practice have been contributing to episodeoriented epidemiology in family practice. We have developed a large prospective database that provides reliable probabilities of diagnoses for symptoms or complaints in a patient of a specific age and sex, suitable for the support of family physicians’ decision making.^2,7,9,16 This article describes the database in more detail and presents data for 4 common symptoms.*

Methods

From 1985 to 1995, 54 Dutch FPs in 23 locations routinely coded episode data for all direct (face-to-face) encounters with their listed patients; within the Dutch health care system, all citizens are listed (registered) with an FP. Each participating FP collected data during a period of at least 1 year; the registration period for patients ranged from 1 to 10 years (mean 2.4).^2,9,16 For each encounter, the patient’s reasons for the encounter, the diagnoses, and the interventions ordered by the physician were coded according to the ICPC. Data were entered on a special encounter form with a copy for a central data entry and included 93,297 patient years, with 236,027 episodes of care and 267,897 direct patient encounters.

Since the FPs had a well-defined practice with listed patients, a precise denominator could be established for the calculation of rates. In the Netherlands, patients can, in principle, not seek specialist care without a referral by their FP. Therefore, especially in a longer observation period, an FP will document a close approximation of the distribution of episodes of care in the Dutch population. In 1 year, 73% of listed patients have a direct encounter with their family physician; in a 2-year observation period this is 92%. It is therefore unlikely that a substantial group of listed patients receive specialist care without their FP’s being aware of it.¹²

A diagnosis could be modified during the course of an episode. If that occurred, the modified diagnosis was applied to all episode data in the analysis. New and ongoing episodes of care were included in a registration year if dealt with at least once; in case of a follow-up encounter in a later registration year, the episode was included again as an ongoing episode. As a consequence, an episode of a chronic disease (eg, diabetes, hypertension) coded in 2 or more registration years was included more than once in the annual prevalence.¹⁷ The average yearly practice population served as the denominator.

As is the case in all time-consuming morbidity studies, the participating FPs were selected, highly motivated, and in this respect, not representative of the average Dutch FP. The database was used in numerous studies, however, and its reliability consistently proved to be high: Approximately 2% of all episodes appeared to be missing in the paper record, and another 2% were erroneously included in the database. The complete reference database is available in Dutch on a CD-ROM attached to a family practice textbook. It includes all combinations of an RFE, a diagnosis, and an intervention for 7 standard age groups at the start of episodes and during follow-up, together with data on comorbidity.⁹

Our paper focuses on prior probabilities, expressed as the top 20 final diagnoses for 4 common reasons for encounter presented at the start of an episode of care^18,19: cough, shortness of breath, general weakness/tiredness, and low back pain without radiation. All probabilities are presented for the total population and for 7 standard age groups, as percentages with 95% symmetric confidence intervals.²⁰ Cells with fewer than 10 observations were excluded. Incidences standardized for the 1995 Dutch population were provided.

Results

(Table 1) shows that for the RFE “cough,” the patient’s age had a substantial impact on the probabilities. The diagnosis of acute bronchitis was common overall but especially in the very young and the very old. This table illustrates the relationships between a common symptom and several diseases with a relatively high incidence (the last column). The prior probabilities were well distributed over the standard table: empty cells occur infrequently.

“Shortness of breath/dyspnoea” as an RFE is associated with a very different distribution of diagnoses than found with cough, especially in the very young and the very old (Table 2). Asthma and acute laryngitis typically occur in the young, while chronic obstructive pulmonary disease, ischemic heart disease, and heart failure occur in the old. Hyperventilation had a peak in young adults. In this table, the relation between a less common symptom and several less common diseases is illustrated. In this case, more empty cells are found. Both cough and shortness of breath mainly relate to respiratory and cardiovascular diseases.

The RFE “tiredness” (Table 3) was associated with several common diseases, but a much wider range of clinical possibilities is apparent. Quite often, this RFE resulted in the symptom diagnosis “tiredness.” Major age differences existed for several diagnoses.

(Table 4) shows that the RFE “low back symptoms/complaints without radiation” quite often led to the same symptom diagnosis. Also, the rather skewed age distribution of low back complaints shows: most cells were insufficiently filled, while age differences for the most common diagnoses appeared to be relatively small.

TABLE 1
FINAL DIAGNOSES FOR EPISODES OF CARE STARTING WITH THE REASON FOR ENCOUNTER COUGH (R05) N=11092

TABLE 2
FINAL DIAGNOSES FOR EPISODES OF CARE STARTING WITH THE REASON FOR ENCOUNTER SHORTNESS OF BREATH (R02); N=2505

TABLE 3
FINAL DIAGNOSES FOR EPISODES OF CARE STARTING WITH THE REASON FOR ENCOUNTER GENERAL WEAKNESS/TIREDNESS (A04); N=5915

TABLE 4
FINAL DIAGNOSES FOR EPISODES OF CARE STARTING WITH THE REASON FOR ENCOUNTER LOW BACK SYMPTOMS/COMPLAINTS WITHOUT RADIATION (L03); N=4238

Discussion

Family practice can be characterized by the specific distribution of health problems and disease in its population, as distinct from the distributions in the general population and in specialists’ populations. Increasingly, empiric data indicate how morbidity patterns and the distribution of reasons for visit in family practice differ from those in hospitals. The availability of age-specific prior probabilities of common symptoms and complaints for diagnoses in family practice has great potential. Diagnostic labels often have the disadvantage of a relative uncertainty, caused by a more or less arbitrary attribution of different symptoms and signs (eg, syndrome diagnoses, psychiatric diagnoses). Symptoms and complaints on the other hand have the advantage of relative certainty, because they represent the patient’s ill health irrespective of the diagnostic label they are given. In the daily work of FPs the importance of the absence or presence of a symptom must be considered in light of the distribution of disease in the family practice setting.

Therefore, such data would not only seem to be crucial for the further development of family practice as an academic discipline or for the design of intervention studies but also has direct practical consequences for clinicians. This information can directly support FPs’ medical decision making and improve communication with patients. The process of finding common ground about diagnosis and management could especially profit from a realistic estimate of probabilities, bridging the gap between the patient’s perspective, as reflected in the presenting symptoms, and the clinical perspective of the FP who wants to provide optimal care.²¹

Several important types of distributions are shown in (Table1) through 4: a very common symptom in relation to highly incident diagnoses, a less common symptom leading to less incident diagnoses, and symptoms primarily resulting in a symptom diagnosis with the same label. Also, it is evident that the range of clinical considerations resulting from a presenting symptom can vary from a relatively limited to a very wide morbidity spectrum.

Sex-specific symptoms and complaints (eg, menstrual problems) typically result in rather specific distributions, as is illustrated in the database available on the JFP Web site. The distribution of diagnoses for symptoms that occur in both sexes may be different not only for age groups but also for sex /age groups.

The value of prior probabilities increases with the availability of data on incidence of diseases in the same population, allowing an estimation of the positive and negative predictive values. Since 1995, data collection has occurred electronically. Later in 2001, a database twice the size will become available that allows more precise estimations for finer age/sex distributions and symptom combinations. Although from Dutch family practice, these data have a high face validity for clinicians wherever they work.⁸ Nevertheless, it would seem that FPs in the United States and other countries should give priority to collecting their own reliable probability databases.^4,22,23

In the Netherlands, Japan, and Poland, an international comparative study has taken place with an electronic patient record, using ICPC. (See page xxx in this issue for the abstract of this article.) Based on a comparison of these databases with US NAMCS data (1995-1997), tables similar to those presented in this paper for the most frequent symptoms and complaints have been made available on the JFP Web site.^24.25

Estimating the probability of disease in unselected patients lies at the heart of the clinical competence of family physicians (FPs).^1,2 This probability is called the prior or pretest probability, because it precedes any diagnostic testing, including the history and physical examination. Based on this knowledge, FPs often decide that since the probability of a serious disease is low, the best thing to do is watchful waiting, thus preventing unnecessary harm and cost to the patient.^3,4 Moderate probabilities may trigger a diagnostic evaluation, and high probabilities may warrant empiric therapy without further diagnostic confirmation. For example, knowing that the probability of gastric carcinoma is exceedingly rare for a dyspeptic patient younger than 40 years supports a diagnostic approach that does not include initial endoscopy. A moderate probability of strep throat is a situation where a rapid strep test may be helpful.⁵ Understanding the high prior probability of a urinary tract infection in young healthy women with dysuria might allow an FP to confidently institute a telephone-based protocol, including empiric treatment for selected patients.

The development of a prior probabilities database requires access to large diverse practice populations with adequate continuity of care and good documentation of all episodes of care.^2,6,7 An “episode of care” is defined as a health problem from the first encounter with a health care provider through the completion of the last encounter related to that particular problem.^2,8.10,11 Early in an episode of care, FPs will often assign a symptom diagnosis such as chest pain, gastric pain, or otalgia; a substantial number of these diagnoses will change over time. Therefore, the observation period should allow for the most important modifications of diagnoses to become visible. Because 1-year incidence and prevalence rates are usually calculated, at least a 1-year study period is required; however, a longer period is preferable, especially for information on chronic diseases.¹²

Episodes of care are based on the relationship between the patient’s reasons for encounter (RFE), the physician’s diagnostic interpretation, and the related interventions over time.¹³ Episodes of care are clearly distinguished from episodes of disease and episodes of illness. An episode of disease begins with its onset and continues until its resolution or the patient’s death, while an episode of illness refers to the period that someone suffers from symptoms or complaints experienced as illness. Not every disease, and certainly not every illness, results in an episode of care.^2,13,14 Most episodes of care, however, are part of an episode of disease and/or illness. Health maintenance episodes can be considered a special form of episodes of care.^10,11 For example, screening for breast cancer (an episode of health maintenance) may prove the existence of the episode of disease well before the patient has symptoms (an episode of illness); an episode of care will follow.

The development of a prior probabilities database also requires a primary care–specific system of classification. The International Classification of Primary Care (ICPC) includes approximately 200 symptoms/complaints and 300 diagnoses common in family practice. Almost all have an incidence of at least 1 per 1000 patients per year.^13,15

Since 1985, the members of the Transition Project of the University of Amsterdam Department of Family Practice have been contributing to episodeoriented epidemiology in family practice. We have developed a large prospective database that provides reliable probabilities of diagnoses for symptoms or complaints in a patient of a specific age and sex, suitable for the support of family physicians’ decision making.^2,7,9,16 This article describes the database in more detail and presents data for 4 common symptoms.*

Methods

From 1985 to 1995, 54 Dutch FPs in 23 locations routinely coded episode data for all direct (face-to-face) encounters with their listed patients; within the Dutch health care system, all citizens are listed (registered) with an FP. Each participating FP collected data during a period of at least 1 year; the registration period for patients ranged from 1 to 10 years (mean 2.4).^2,9,16 For each encounter, the patient’s reasons for the encounter, the diagnoses, and the interventions ordered by the physician were coded according to the ICPC. Data were entered on a special encounter form with a copy for a central data entry and included 93,297 patient years, with 236,027 episodes of care and 267,897 direct patient encounters.

Since the FPs had a well-defined practice with listed patients, a precise denominator could be established for the calculation of rates. In the Netherlands, patients can, in principle, not seek specialist care without a referral by their FP. Therefore, especially in a longer observation period, an FP will document a close approximation of the distribution of episodes of care in the Dutch population. In 1 year, 73% of listed patients have a direct encounter with their family physician; in a 2-year observation period this is 92%. It is therefore unlikely that a substantial group of listed patients receive specialist care without their FP’s being aware of it.¹²

A diagnosis could be modified during the course of an episode. If that occurred, the modified diagnosis was applied to all episode data in the analysis. New and ongoing episodes of care were included in a registration year if dealt with at least once; in case of a follow-up encounter in a later registration year, the episode was included again as an ongoing episode. As a consequence, an episode of a chronic disease (eg, diabetes, hypertension) coded in 2 or more registration years was included more than once in the annual prevalence.¹⁷ The average yearly practice population served as the denominator.

As is the case in all time-consuming morbidity studies, the participating FPs were selected, highly motivated, and in this respect, not representative of the average Dutch FP. The database was used in numerous studies, however, and its reliability consistently proved to be high: Approximately 2% of all episodes appeared to be missing in the paper record, and another 2% were erroneously included in the database. The complete reference database is available in Dutch on a CD-ROM attached to a family practice textbook. It includes all combinations of an RFE, a diagnosis, and an intervention for 7 standard age groups at the start of episodes and during follow-up, together with data on comorbidity.⁹

Our paper focuses on prior probabilities, expressed as the top 20 final diagnoses for 4 common reasons for encounter presented at the start of an episode of care^18,19: cough, shortness of breath, general weakness/tiredness, and low back pain without radiation. All probabilities are presented for the total population and for 7 standard age groups, as percentages with 95% symmetric confidence intervals.²⁰ Cells with fewer than 10 observations were excluded. Incidences standardized for the 1995 Dutch population were provided.

Results

(Table 1) shows that for the RFE “cough,” the patient’s age had a substantial impact on the probabilities. The diagnosis of acute bronchitis was common overall but especially in the very young and the very old. This table illustrates the relationships between a common symptom and several diseases with a relatively high incidence (the last column). The prior probabilities were well distributed over the standard table: empty cells occur infrequently.

“Shortness of breath/dyspnoea” as an RFE is associated with a very different distribution of diagnoses than found with cough, especially in the very young and the very old (Table 2). Asthma and acute laryngitis typically occur in the young, while chronic obstructive pulmonary disease, ischemic heart disease, and heart failure occur in the old. Hyperventilation had a peak in young adults. In this table, the relation between a less common symptom and several less common diseases is illustrated. In this case, more empty cells are found. Both cough and shortness of breath mainly relate to respiratory and cardiovascular diseases.

The RFE “tiredness” (Table 3) was associated with several common diseases, but a much wider range of clinical possibilities is apparent. Quite often, this RFE resulted in the symptom diagnosis “tiredness.” Major age differences existed for several diagnoses.

(Table 4) shows that the RFE “low back symptoms/complaints without radiation” quite often led to the same symptom diagnosis. Also, the rather skewed age distribution of low back complaints shows: most cells were insufficiently filled, while age differences for the most common diagnoses appeared to be relatively small.

TABLE 1
FINAL DIAGNOSES FOR EPISODES OF CARE STARTING WITH THE REASON FOR ENCOUNTER COUGH (R05) N=11092

TABLE 2
FINAL DIAGNOSES FOR EPISODES OF CARE STARTING WITH THE REASON FOR ENCOUNTER SHORTNESS OF BREATH (R02); N=2505

TABLE 3
FINAL DIAGNOSES FOR EPISODES OF CARE STARTING WITH THE REASON FOR ENCOUNTER GENERAL WEAKNESS/TIREDNESS (A04); N=5915

TABLE 4
FINAL DIAGNOSES FOR EPISODES OF CARE STARTING WITH THE REASON FOR ENCOUNTER LOW BACK SYMPTOMS/COMPLAINTS WITHOUT RADIATION (L03); N=4238

Discussion

Family practice can be characterized by the specific distribution of health problems and disease in its population, as distinct from the distributions in the general population and in specialists’ populations. Increasingly, empiric data indicate how morbidity patterns and the distribution of reasons for visit in family practice differ from those in hospitals. The availability of age-specific prior probabilities of common symptoms and complaints for diagnoses in family practice has great potential. Diagnostic labels often have the disadvantage of a relative uncertainty, caused by a more or less arbitrary attribution of different symptoms and signs (eg, syndrome diagnoses, psychiatric diagnoses). Symptoms and complaints on the other hand have the advantage of relative certainty, because they represent the patient’s ill health irrespective of the diagnostic label they are given. In the daily work of FPs the importance of the absence or presence of a symptom must be considered in light of the distribution of disease in the family practice setting.

Therefore, such data would not only seem to be crucial for the further development of family practice as an academic discipline or for the design of intervention studies but also has direct practical consequences for clinicians. This information can directly support FPs’ medical decision making and improve communication with patients. The process of finding common ground about diagnosis and management could especially profit from a realistic estimate of probabilities, bridging the gap between the patient’s perspective, as reflected in the presenting symptoms, and the clinical perspective of the FP who wants to provide optimal care.²¹

Several important types of distributions are shown in (Table1) through 4: a very common symptom in relation to highly incident diagnoses, a less common symptom leading to less incident diagnoses, and symptoms primarily resulting in a symptom diagnosis with the same label. Also, it is evident that the range of clinical considerations resulting from a presenting symptom can vary from a relatively limited to a very wide morbidity spectrum.

Sex-specific symptoms and complaints (eg, menstrual problems) typically result in rather specific distributions, as is illustrated in the database available on the JFP Web site. The distribution of diagnoses for symptoms that occur in both sexes may be different not only for age groups but also for sex /age groups.

The value of prior probabilities increases with the availability of data on incidence of diseases in the same population, allowing an estimation of the positive and negative predictive values. Since 1995, data collection has occurred electronically. Later in 2001, a database twice the size will become available that allows more precise estimations for finer age/sex distributions and symptom combinations. Although from Dutch family practice, these data have a high face validity for clinicians wherever they work.⁸ Nevertheless, it would seem that FPs in the United States and other countries should give priority to collecting their own reliable probability databases.^4,22,23

In the Netherlands, Japan, and Poland, an international comparative study has taken place with an electronic patient record, using ICPC. (See page xxx in this issue for the abstract of this article.) Based on a comparison of these databases with US NAMCS data (1995-1997), tables similar to those presented in this paper for the most frequent symptoms and complaints have been made available on the JFP Web site.^24.25

Estimating the probability of disease in unselected patients lies at the heart of the clinical competence of family physicians (FPs).^1,2 This probability is called the prior or pretest probability, because it precedes any diagnostic testing, including the history and physical examination. Based on this knowledge, FPs often decide that since the probability of a serious disease is low, the best thing to do is watchful waiting, thus preventing unnecessary harm and cost to the patient.^3,4 Moderate probabilities may trigger a diagnostic evaluation, and high probabilities may warrant empiric therapy without further diagnostic confirmation. For example, knowing that the probability of gastric carcinoma is exceedingly rare for a dyspeptic patient younger than 40 years supports a diagnostic approach that does not include initial endoscopy. A moderate probability of strep throat is a situation where a rapid strep test may be helpful.⁵ Understanding the high prior probability of a urinary tract infection in young healthy women with dysuria might allow an FP to confidently institute a telephone-based protocol, including empiric treatment for selected patients.

The development of a prior probabilities database requires access to large diverse practice populations with adequate continuity of care and good documentation of all episodes of care.^2,6,7 An “episode of care” is defined as a health problem from the first encounter with a health care provider through the completion of the last encounter related to that particular problem.^2,8.10,11 Early in an episode of care, FPs will often assign a symptom diagnosis such as chest pain, gastric pain, or otalgia; a substantial number of these diagnoses will change over time. Therefore, the observation period should allow for the most important modifications of diagnoses to become visible. Because 1-year incidence and prevalence rates are usually calculated, at least a 1-year study period is required; however, a longer period is preferable, especially for information on chronic diseases.¹²

Episodes of care are based on the relationship between the patient’s reasons for encounter (RFE), the physician’s diagnostic interpretation, and the related interventions over time.¹³ Episodes of care are clearly distinguished from episodes of disease and episodes of illness. An episode of disease begins with its onset and continues until its resolution or the patient’s death, while an episode of illness refers to the period that someone suffers from symptoms or complaints experienced as illness. Not every disease, and certainly not every illness, results in an episode of care.^2,13,14 Most episodes of care, however, are part of an episode of disease and/or illness. Health maintenance episodes can be considered a special form of episodes of care.^10,11 For example, screening for breast cancer (an episode of health maintenance) may prove the existence of the episode of disease well before the patient has symptoms (an episode of illness); an episode of care will follow.

The development of a prior probabilities database also requires a primary care–specific system of classification. The International Classification of Primary Care (ICPC) includes approximately 200 symptoms/complaints and 300 diagnoses common in family practice. Almost all have an incidence of at least 1 per 1000 patients per year.^13,15

Since 1985, the members of the Transition Project of the University of Amsterdam Department of Family Practice have been contributing to episodeoriented epidemiology in family practice. We have developed a large prospective database that provides reliable probabilities of diagnoses for symptoms or complaints in a patient of a specific age and sex, suitable for the support of family physicians’ decision making.^2,7,9,16 This article describes the database in more detail and presents data for 4 common symptoms.*

Methods

From 1985 to 1995, 54 Dutch FPs in 23 locations routinely coded episode data for all direct (face-to-face) encounters with their listed patients; within the Dutch health care system, all citizens are listed (registered) with an FP. Each participating FP collected data during a period of at least 1 year; the registration period for patients ranged from 1 to 10 years (mean 2.4).^2,9,16 For each encounter, the patient’s reasons for the encounter, the diagnoses, and the interventions ordered by the physician were coded according to the ICPC. Data were entered on a special encounter form with a copy for a central data entry and included 93,297 patient years, with 236,027 episodes of care and 267,897 direct patient encounters.

Since the FPs had a well-defined practice with listed patients, a precise denominator could be established for the calculation of rates. In the Netherlands, patients can, in principle, not seek specialist care without a referral by their FP. Therefore, especially in a longer observation period, an FP will document a close approximation of the distribution of episodes of care in the Dutch population. In 1 year, 73% of listed patients have a direct encounter with their family physician; in a 2-year observation period this is 92%. It is therefore unlikely that a substantial group of listed patients receive specialist care without their FP’s being aware of it.¹²

A diagnosis could be modified during the course of an episode. If that occurred, the modified diagnosis was applied to all episode data in the analysis. New and ongoing episodes of care were included in a registration year if dealt with at least once; in case of a follow-up encounter in a later registration year, the episode was included again as an ongoing episode. As a consequence, an episode of a chronic disease (eg, diabetes, hypertension) coded in 2 or more registration years was included more than once in the annual prevalence.¹⁷ The average yearly practice population served as the denominator.

As is the case in all time-consuming morbidity studies, the participating FPs were selected, highly motivated, and in this respect, not representative of the average Dutch FP. The database was used in numerous studies, however, and its reliability consistently proved to be high: Approximately 2% of all episodes appeared to be missing in the paper record, and another 2% were erroneously included in the database. The complete reference database is available in Dutch on a CD-ROM attached to a family practice textbook. It includes all combinations of an RFE, a diagnosis, and an intervention for 7 standard age groups at the start of episodes and during follow-up, together with data on comorbidity.⁹

Our paper focuses on prior probabilities, expressed as the top 20 final diagnoses for 4 common reasons for encounter presented at the start of an episode of care^18,19: cough, shortness of breath, general weakness/tiredness, and low back pain without radiation. All probabilities are presented for the total population and for 7 standard age groups, as percentages with 95% symmetric confidence intervals.²⁰ Cells with fewer than 10 observations were excluded. Incidences standardized for the 1995 Dutch population were provided.

Results

(Table 1) shows that for the RFE “cough,” the patient’s age had a substantial impact on the probabilities. The diagnosis of acute bronchitis was common overall but especially in the very young and the very old. This table illustrates the relationships between a common symptom and several diseases with a relatively high incidence (the last column). The prior probabilities were well distributed over the standard table: empty cells occur infrequently.

“Shortness of breath/dyspnoea” as an RFE is associated with a very different distribution of diagnoses than found with cough, especially in the very young and the very old (Table 2). Asthma and acute laryngitis typically occur in the young, while chronic obstructive pulmonary disease, ischemic heart disease, and heart failure occur in the old. Hyperventilation had a peak in young adults. In this table, the relation between a less common symptom and several less common diseases is illustrated. In this case, more empty cells are found. Both cough and shortness of breath mainly relate to respiratory and cardiovascular diseases.

The RFE “tiredness” (Table 3) was associated with several common diseases, but a much wider range of clinical possibilities is apparent. Quite often, this RFE resulted in the symptom diagnosis “tiredness.” Major age differences existed for several diagnoses.

(Table 4) shows that the RFE “low back symptoms/complaints without radiation” quite often led to the same symptom diagnosis. Also, the rather skewed age distribution of low back complaints shows: most cells were insufficiently filled, while age differences for the most common diagnoses appeared to be relatively small.

TABLE 1
FINAL DIAGNOSES FOR EPISODES OF CARE STARTING WITH THE REASON FOR ENCOUNTER COUGH (R05) N=11092

TABLE 2
FINAL DIAGNOSES FOR EPISODES OF CARE STARTING WITH THE REASON FOR ENCOUNTER SHORTNESS OF BREATH (R02); N=2505

TABLE 3
FINAL DIAGNOSES FOR EPISODES OF CARE STARTING WITH THE REASON FOR ENCOUNTER GENERAL WEAKNESS/TIREDNESS (A04); N=5915

TABLE 4
FINAL DIAGNOSES FOR EPISODES OF CARE STARTING WITH THE REASON FOR ENCOUNTER LOW BACK SYMPTOMS/COMPLAINTS WITHOUT RADIATION (L03); N=4238

Discussion

Family practice can be characterized by the specific distribution of health problems and disease in its population, as distinct from the distributions in the general population and in specialists’ populations. Increasingly, empiric data indicate how morbidity patterns and the distribution of reasons for visit in family practice differ from those in hospitals. The availability of age-specific prior probabilities of common symptoms and complaints for diagnoses in family practice has great potential. Diagnostic labels often have the disadvantage of a relative uncertainty, caused by a more or less arbitrary attribution of different symptoms and signs (eg, syndrome diagnoses, psychiatric diagnoses). Symptoms and complaints on the other hand have the advantage of relative certainty, because they represent the patient’s ill health irrespective of the diagnostic label they are given. In the daily work of FPs the importance of the absence or presence of a symptom must be considered in light of the distribution of disease in the family practice setting.

Therefore, such data would not only seem to be crucial for the further development of family practice as an academic discipline or for the design of intervention studies but also has direct practical consequences for clinicians. This information can directly support FPs’ medical decision making and improve communication with patients. The process of finding common ground about diagnosis and management could especially profit from a realistic estimate of probabilities, bridging the gap between the patient’s perspective, as reflected in the presenting symptoms, and the clinical perspective of the FP who wants to provide optimal care.²¹

Several important types of distributions are shown in (Table1) through 4: a very common symptom in relation to highly incident diagnoses, a less common symptom leading to less incident diagnoses, and symptoms primarily resulting in a symptom diagnosis with the same label. Also, it is evident that the range of clinical considerations resulting from a presenting symptom can vary from a relatively limited to a very wide morbidity spectrum.

Sex-specific symptoms and complaints (eg, menstrual problems) typically result in rather specific distributions, as is illustrated in the database available on the JFP Web site. The distribution of diagnoses for symptoms that occur in both sexes may be different not only for age groups but also for sex /age groups.

The value of prior probabilities increases with the availability of data on incidence of diseases in the same population, allowing an estimation of the positive and negative predictive values. Since 1995, data collection has occurred electronically. Later in 2001, a database twice the size will become available that allows more precise estimations for finer age/sex distributions and symptom combinations. Although from Dutch family practice, these data have a high face validity for clinicians wherever they work.⁸ Nevertheless, it would seem that FPs in the United States and other countries should give priority to collecting their own reliable probability databases.^4,22,23

In the Netherlands, Japan, and Poland, an international comparative study has taken place with an electronic patient record, using ICPC. (See page xxx in this issue for the abstract of this article.) Based on a comparison of these databases with US NAMCS data (1995-1997), tables similar to those presented in this paper for the most frequent symptoms and complaints have been made available on the JFP Web site.^24.25