Clostridium difficile is the single most common cause of nosocomial diarrhea in both adults and children.[1, 2] C difficile infections (CDIs) can range from self‐limited diarrhea to severe pseudomembranous colitis. Though widely distributed in the environment, hospitals and child care facilities are major reservoirs for C difficile. Traditionally, hospitalization and antibiotic use have been the 2 major risk factors for acquiring CDI.

Recent studies suggest C difficile epidemiology is shifting. In 2005, the Centers for Disease Control and Prevention (CDC) reported CDIs in 33 otherwise low‐risk patients, 6 of whom were children.[3] Other studies have noted increasing incidence of pediatric CDIs,[4, 5, 6, 7] 1 identifying 43% with no prior antibiotic use.[4] This emerging data led to the recent American Academy of Pediatrics policy statement on pediatric CDIs.[8] Data regarding associated clinical risk factors of CDIs in pediatric patients in light of the changing epidemiology are limited. Only 1 recent study looked at 6 clinical factors and found that antibiotic use, history of solid organ transplantation, gastrointestinal (GI) devices, and acid suppressing medications increased risk for CDIs.[9]

Data regarding the source of these infections are also limited. Three pediatric studies evaluating source found a significant amount of community‐acquired disease (59%, 25%, and 19% of the study population, respectively).[4, 9, 10] However, only 1 of these studies provided clinical comparisons between community and hospital‐acquired cases.[10] To date, no study has examined a comprehensive list of potential risk factors that might differentiate hospitalized pediatric patients with CDIs from those with acute gastroenteritis (AGE).

PATIENTS AND METHODS

We conducted an investigator‐initiated, retrospective, case‐control study examining risk factors associated with CDIs in a hospitalized pediatric population at Rady Children's Hospital San Diego (RCHSD). Rady Children's is a tertiary‐care pediatric healthcare system and the sole pediatric referral center for San Diego, with a catchment of 850,000 children. RCHSD posts over 71,000 emergency department (ED) and 30,000 urgent care (UC) visits at 4 sites and over 15,000 admissions yearly. All system information is archived in 1 electronic database. We reviewed patient records for a 2‐year period from June 1, 2008 through May 31, 2010. The study protocol was reviewed and approved by the institutional review board at the University of California San Diego.

Cases of C difficile (CDs) included pediatric patients 18 years of age with all of the following: International Classification of Diseases, 9th Revision (ICD‐9) code for C difficile infection (08.45), a positive C difficile toxin A or B by enzyme immunoassay (EIA) (Meridian Bioscience, Inc., Cincinnati, OH), and the presence of diarrhea and/or abdominal pain. Randomly selected age‐matched controls from the same time period with a discharge diagnosis of AGE (APR‐DRG 249) and the presence of diarrhea served as controls (CTLs). In the 1 year age group, any patient with a positive C difficile toxin assay but no diagnosis of CDI was excluded from the CTL group to avoid potential confounding.

Records were reviewed for multiple potential risk factors based on limited past studies and other factors associated with CDI pathogenesis including age, race, ethnicity, antibiotic use within the previous 90 days (type, route, and duration), diarrhea type, abdominal pain, fever, proton pump inhibitor (PPI) use, sick contacts (diarrheal illness), recent travel, and hospitalization within the last 6 months. Diarrhea was defined as increase in stool frequency or volume. Past medical/surgical history abstracted included GI disease, past CDIs, abdominal surgery, immunodeficiency, renal disease, cardiac disease, nutritional deficiencies, and number of past hospitalizations (all cause). In addition, multiple factors during the hospital course were reviewed: length of stay (LOS), antibiotic therapy, diarrhea type, abdominal pain, fever, electrolyte levels, need for stool replacement fluid, and altered diet recommendations. Thirty‐day return to ED/UC or readmission and cause for the return were also retrieved on all patients. An objective data collection form was used, and all records were reviewed by 1 researcher (W.S.) with a second reviewer (E.F.) reviewing 20% of the charts, with 90% initial concordance. Consensus was reached on all elements abstracted.

Three additional subanalyses were completed. The first subanalysis compared antibiotic prophylaxis (defined as daily use of an antibiotic for >28 days) in CDs versus CTLs. We reviewed charts to ensure extended antibiotic use was for prophylaxis and not treatment. The second subanalysis compared CDs to those CTLs with a negative C difficile toxin assay. This was done to evaluate whether using this control group would highlight a different set of risk factors. The third subanalysis separated CDs into community‐acquired CD (CA‐CD) and hospital‐acquired CD (HA‐CD). We defined CA‐CD as any patient with symptoms either prior to or within the first 48 hours of the index admission and no past hospitalizations or with the last hospitalization >4 weeks prior to the index admission. Patients who developed symptoms at home or within 48 hours of the index admission, but had been hospitalized within the past 4 weeks, were defined as community‐onset HA‐CD. Patients who developed CDIs after 48 hours of the index admission were defined as hospital‐onset HA‐CD. These groupings are consistent with the CDI surveillance recommendations.[11]

All statistical analyses were performed with SPSS statistical software version 21.0 (SPSS Inc., Chicago, IL). Initial comparisons between CDs and CTLs were conducted using t tests for continuous variables and [2] tests for categorical variables. As CDI in infants is controversial, we analyzed our data with and without this cohort to eliminate extraneous, age‐related differences. After confirming that there were no issues with tolerance among possibly related factors, a saturated multiple logistic regression model was used to determine which of the independent variables identified in the initial comparison were predictors of having C difficile when controlling for factors associated with chronic disease.

RESULTS

Descriptive characteristics of the 134 CDs and the 274 CTLs are provided in Table 1. CDs and CTLs were similar in gender and race. More CDs had recent hospitalization and antibiotic exposure, with 24% of CDs versus 3% of CTLs treated with 2 or more antibiotics. Watery stools were the most common type of diarrhea in both CDs and CTLs, and bloody stools did not differ significantly between the 2 groups. However, abdominal pain on admission was more common in CTLs. CDs were more likely to have a history of GI disease, abdominal surgery, and specifically GI surgery. Immunodeficiency and PPI use were far more frequent in CDs, whereas exposure to sick contacts was more common in CTLs. Although CDs had an overall higher rate of ED/UC return visits and readmissions, the rate of return due to GI symptoms was similar in both groups. Reanalysis of the data with the <1‐year cohort removed showed persistent statistically significant findings in these variables. Hospital course, including electrolyte levels, need for intravenous fluids, or modified diets, did not significantly differ between CDs and CTLs (data not shown).

Analysis of CDs without traditional risk factors was performed. To identify patients, we first selected the 46 (34%) without prior antibiotic exposure, then eliminated 19 who had been hospitalized within the past 6 months. Of the remaining 27 patients, 16 had a prolonged hospitalization (>5 days) at the time of CDI diagnosis. This left us with 11 patients (8% of CDs) without any common risk factors of antibiotic use, recent hospitalization, or prolonged hospitalization. None of these patients had a history of CDIs; 6 had significant medical histories. A detailed description of these 11 patients if provided in Table 2.

The first subanalysis evaluated antibiotic prophylaxis and found 51 (37%) in CDs versus 10 (4%) in CTLs. However, after controlling for immunodeficiency found in 40 of these CDs, we found no statistically significant difference. There were insufficient numbers of those on prophylaxis for other reasons (eg, vesicoureteral reflux) to analyze prophylaxis independently.

The second subanalysis compared controls with a negative C difficile toxin assay (21% of CTLs) to CDs on a number of clinical factors. Results were compared to the primary analysis. Many factors remained significant: antibiotic use in the past 90 days was still more frequent in CDs (66% vs 35%, P<0.001) as was immunodeficiency in CDs (46% vs 14%, P<0.001). However, immunodeficiency in this subset of the controls was represented over twice as often as that of the baseline CTLs (14% vs 6%), whereas GI disease was similar between the 2 groups (37% vs 31%, P<0.40). PPI use demonstrated a suggestive relationship (22% vs 11%, P<0.07).

Data for the third subanalysis between CA‐CD and HA‐CD are shown on Table 3. We initially compared CA‐CD, community‐onset HA‐CD, and hospital‐onset HA‐CD. However, when stratification was found to not be significant, we combined both categories of HA‐CD into 1 group. CA‐CD and HA‐CD did not demonstrate significant difference in antibiotic use, type, prophylaxis, history of abdominal surgery, immunodeficiency, or GI disease. Bloody stools were more common in CA‐CD.

Odds ratio (OR) was calculated for association of individual risk factors for disease between CDs and CTLs (Table 4). Our model controlled for antibiotics use in the past 90 days, PPI use, treatment with 2 or more antibiotics, recent hospitalization, past history of CDIs, history of GI disease, history of abdominal surgery, and being immunodeficient. Antibiotic use within the past 90 days (OR: 2.80, P=0.001), recent hospitalization (OR: 2.33, P=0.007), and immunodeficiency (OR: 6.02, P<0.001) were associated with having C difficile. A similar logistic regression was conducted using a model comparing community‐ and hospital‐acquired cases, but no difference was found among risk factors.

DISCUSSION/CONCLUSION

Our study shows that in addition to traditional risk factors of antibiotic use and recent hospitalization, immunodeficiency is a significant key factor associated with the diagnosis of CD. We found that traditional risk factors are not present in all hospitalized pediatric patients with CD. Our study does not support routine testing for C difficile in patients with diarrhea; however, it does suggest testing children with persistent or severe diarrheal symptoms even if traditional risk factors are absent, especially in the presence of immunodeficiency. The intervals we used for antibiotic exposure (past 90 days) and recent hospitalization (past 6 months) were longer compared to other studies,[9, 12] making our findings even more meaningful. Although some of the 11 patients without traditional risk factors had the presence of clinical factors shown in previous studies to be more common in patients with CDIs (GI disease, GI surgery, gastric tube/nasogastric feeding),[12, 13] we still find 4 patients >1 year of age with CDIs and no risk factors. This echoes the CDCs concerns of CDIs in low‐risk patients.[3]

Unlike clinical history, we found clinical symptoms and basic electrolyte testing may not help to distinguish CD from AGE patients. Although abdominal pain and diarrhea on admission were significantly more common in CTLs, when including abdominal pain and diarrhea during hospitalization, this finding was no longer valid. Additionally, although overall return rate was higher for CDs, the return rate for GI symptoms specifically was not different. The former was instead most often due to complications associated with comorbid conditions (GI disease, immunodeficiency). We did assess LOS for both CDs and CTLs; however, due to the high percentage of CDs with malignancy and other severe illnesses, it was difficult to ascertain the effect of CDIs on LOS. Severe CD is described as admission to the intensive care unit due to C difficile complications, colectomy, and death secondary to C difficile.[11] Although our study did not look at severe CDI as a direct outcome, we did not have any cases of colectomy or death secondary to CDI.

Two recent studies[9, 14] showed a high percentage of acid suppression medication use in patients with CDIs, with 1 study reporting 60% using PPIs and 21% using histamine blockers. Our study initially found similar high levels of PPI use among patients with CDIs; however, no significance was found when controlling for chronic disease. Prescriptions of PPIs for pediatric patients have risen dramatically recently,[15] as have reported all‐cause complications.[16] Further studies are needed to evaluate the independent risks of PPI use and CDIs in children. We were unable to analyze the influence of antibiotic use at prophylactic levels on CD rates, as the majority our CDs were on prophylaxis due to immunodeficiency.

Our study is unique in many ways. It is the first study to evaluate hospitalized pediatric patients with a comprehensive list of potential risk factors for CDIs, looking at clinical data on admission and during hospitalization. Additionally, as our site archives all clinical information in 1 database, we were able to identify ED/UC return and hospital readmissions. Although it is possible patients may have been evaluated outside of our healthcare system, this would be uncommon due to our referral patterns and UC sites. Our study used age‐matched patients with diarrheal symptoms and AGE discharge diagnosis as the control group. This differs from the 1 previous study looking at risk factors for CDIs in children.[9] In that study, researchers used patients with negative C difficile toxin testing as controls. Our subanalysis of CTLs with a negative toxin assay found much higher rates of underlying GI disease and immunodeficiency. Whereas previous studies compared patients already at high risk for CDI and assessed the differences between those with and without the infection, our study looked at what clinical factors distinguish CDI from AGE in a hospitalized population.

Similar to other pediatric studies, our study found a significant number of CA‐CD. However our study is 1 of the first to compare pediatric CA‐CD with HA‐CD based on clinical factors. Of the 9 demographic and clinical variables assessed, the only significant difference found was presence of bloody diarrhea. It may be that bloody diarrhea prompted the patients to be admitted as opposed to evaluated in the ambulatory setting.

Our study had some limitations. We used ICD‐9 discharge diagnosis codes to identify our patients; however, thorough chart review found clinical indices (diarrhea and abdominal pain) that correlated well with CDI diagnosis in addition to positive laboratory test. The EIA C difficile toxin assay was the standard of care during our study period. However, a recent study has shown false positives using EIA testing in pediatric populations.[17] In our primary analysis, we did not exclude patients with a past history of CDIs. Recurrent CDI is defined as having symptoms within 8 weeks after the primary infection. Of our patients with a history of CDIs, only 2 met this definition. Due to the small number, excluding these patients would not have changed our results significantly. Last, as with any retrospective study, we relied on caregiver reports regarding clinical history, especially in the CA‐CD cohort.

Based on our comprehensive analysis of pediatric patients, there should be increased suspicion for CDI in children with baseline immunodeficiency. Our study also supports testing children with persistent or severe GI symptoms even in the absence of traditional risk factors. These elements, coupled with history of antibiotic use, recent hospitalization, GI disease, and abdominal surgery could be used to create an assessment tool to assist clinicians in the diagnosis of CDIs in pediatric patients. A significant percentage of CDIs continues to be CA‐CD. HA‐CD and CA‐CD patients have similar clinical features. Further studies are needed to determine the effect of PPI use and prophylactic antibiotics on CDIs in children.

Clostridium difficile is the single most common cause of nosocomial diarrhea in both adults and children.[1, 2] C difficile infections (CDIs) can range from self‐limited diarrhea to severe pseudomembranous colitis. Though widely distributed in the environment, hospitals and child care facilities are major reservoirs for C difficile. Traditionally, hospitalization and antibiotic use have been the 2 major risk factors for acquiring CDI.

Recent studies suggest C difficile epidemiology is shifting. In 2005, the Centers for Disease Control and Prevention (CDC) reported CDIs in 33 otherwise low‐risk patients, 6 of whom were children.[3] Other studies have noted increasing incidence of pediatric CDIs,[4, 5, 6, 7] 1 identifying 43% with no prior antibiotic use.[4] This emerging data led to the recent American Academy of Pediatrics policy statement on pediatric CDIs.[8] Data regarding associated clinical risk factors of CDIs in pediatric patients in light of the changing epidemiology are limited. Only 1 recent study looked at 6 clinical factors and found that antibiotic use, history of solid organ transplantation, gastrointestinal (GI) devices, and acid suppressing medications increased risk for CDIs.[9]

Data regarding the source of these infections are also limited. Three pediatric studies evaluating source found a significant amount of community‐acquired disease (59%, 25%, and 19% of the study population, respectively).[4, 9, 10] However, only 1 of these studies provided clinical comparisons between community and hospital‐acquired cases.[10] To date, no study has examined a comprehensive list of potential risk factors that might differentiate hospitalized pediatric patients with CDIs from those with acute gastroenteritis (AGE).

PATIENTS AND METHODS

We conducted an investigator‐initiated, retrospective, case‐control study examining risk factors associated with CDIs in a hospitalized pediatric population at Rady Children's Hospital San Diego (RCHSD). Rady Children's is a tertiary‐care pediatric healthcare system and the sole pediatric referral center for San Diego, with a catchment of 850,000 children. RCHSD posts over 71,000 emergency department (ED) and 30,000 urgent care (UC) visits at 4 sites and over 15,000 admissions yearly. All system information is archived in 1 electronic database. We reviewed patient records for a 2‐year period from June 1, 2008 through May 31, 2010. The study protocol was reviewed and approved by the institutional review board at the University of California San Diego.

Cases of C difficile (CDs) included pediatric patients 18 years of age with all of the following: International Classification of Diseases, 9th Revision (ICD‐9) code for C difficile infection (08.45), a positive C difficile toxin A or B by enzyme immunoassay (EIA) (Meridian Bioscience, Inc., Cincinnati, OH), and the presence of diarrhea and/or abdominal pain. Randomly selected age‐matched controls from the same time period with a discharge diagnosis of AGE (APR‐DRG 249) and the presence of diarrhea served as controls (CTLs). In the 1 year age group, any patient with a positive C difficile toxin assay but no diagnosis of CDI was excluded from the CTL group to avoid potential confounding.

Records were reviewed for multiple potential risk factors based on limited past studies and other factors associated with CDI pathogenesis including age, race, ethnicity, antibiotic use within the previous 90 days (type, route, and duration), diarrhea type, abdominal pain, fever, proton pump inhibitor (PPI) use, sick contacts (diarrheal illness), recent travel, and hospitalization within the last 6 months. Diarrhea was defined as increase in stool frequency or volume. Past medical/surgical history abstracted included GI disease, past CDIs, abdominal surgery, immunodeficiency, renal disease, cardiac disease, nutritional deficiencies, and number of past hospitalizations (all cause). In addition, multiple factors during the hospital course were reviewed: length of stay (LOS), antibiotic therapy, diarrhea type, abdominal pain, fever, electrolyte levels, need for stool replacement fluid, and altered diet recommendations. Thirty‐day return to ED/UC or readmission and cause for the return were also retrieved on all patients. An objective data collection form was used, and all records were reviewed by 1 researcher (W.S.) with a second reviewer (E.F.) reviewing 20% of the charts, with 90% initial concordance. Consensus was reached on all elements abstracted.

Three additional subanalyses were completed. The first subanalysis compared antibiotic prophylaxis (defined as daily use of an antibiotic for >28 days) in CDs versus CTLs. We reviewed charts to ensure extended antibiotic use was for prophylaxis and not treatment. The second subanalysis compared CDs to those CTLs with a negative C difficile toxin assay. This was done to evaluate whether using this control group would highlight a different set of risk factors. The third subanalysis separated CDs into community‐acquired CD (CA‐CD) and hospital‐acquired CD (HA‐CD). We defined CA‐CD as any patient with symptoms either prior to or within the first 48 hours of the index admission and no past hospitalizations or with the last hospitalization >4 weeks prior to the index admission. Patients who developed symptoms at home or within 48 hours of the index admission, but had been hospitalized within the past 4 weeks, were defined as community‐onset HA‐CD. Patients who developed CDIs after 48 hours of the index admission were defined as hospital‐onset HA‐CD. These groupings are consistent with the CDI surveillance recommendations.[11]

All statistical analyses were performed with SPSS statistical software version 21.0 (SPSS Inc., Chicago, IL). Initial comparisons between CDs and CTLs were conducted using t tests for continuous variables and [2] tests for categorical variables. As CDI in infants is controversial, we analyzed our data with and without this cohort to eliminate extraneous, age‐related differences. After confirming that there were no issues with tolerance among possibly related factors, a saturated multiple logistic regression model was used to determine which of the independent variables identified in the initial comparison were predictors of having C difficile when controlling for factors associated with chronic disease.

RESULTS

Descriptive characteristics of the 134 CDs and the 274 CTLs are provided in Table 1. CDs and CTLs were similar in gender and race. More CDs had recent hospitalization and antibiotic exposure, with 24% of CDs versus 3% of CTLs treated with 2 or more antibiotics. Watery stools were the most common type of diarrhea in both CDs and CTLs, and bloody stools did not differ significantly between the 2 groups. However, abdominal pain on admission was more common in CTLs. CDs were more likely to have a history of GI disease, abdominal surgery, and specifically GI surgery. Immunodeficiency and PPI use were far more frequent in CDs, whereas exposure to sick contacts was more common in CTLs. Although CDs had an overall higher rate of ED/UC return visits and readmissions, the rate of return due to GI symptoms was similar in both groups. Reanalysis of the data with the <1‐year cohort removed showed persistent statistically significant findings in these variables. Hospital course, including electrolyte levels, need for intravenous fluids, or modified diets, did not significantly differ between CDs and CTLs (data not shown).

Analysis of CDs without traditional risk factors was performed. To identify patients, we first selected the 46 (34%) without prior antibiotic exposure, then eliminated 19 who had been hospitalized within the past 6 months. Of the remaining 27 patients, 16 had a prolonged hospitalization (>5 days) at the time of CDI diagnosis. This left us with 11 patients (8% of CDs) without any common risk factors of antibiotic use, recent hospitalization, or prolonged hospitalization. None of these patients had a history of CDIs; 6 had significant medical histories. A detailed description of these 11 patients if provided in Table 2.

The first subanalysis evaluated antibiotic prophylaxis and found 51 (37%) in CDs versus 10 (4%) in CTLs. However, after controlling for immunodeficiency found in 40 of these CDs, we found no statistically significant difference. There were insufficient numbers of those on prophylaxis for other reasons (eg, vesicoureteral reflux) to analyze prophylaxis independently.

The second subanalysis compared controls with a negative C difficile toxin assay (21% of CTLs) to CDs on a number of clinical factors. Results were compared to the primary analysis. Many factors remained significant: antibiotic use in the past 90 days was still more frequent in CDs (66% vs 35%, P<0.001) as was immunodeficiency in CDs (46% vs 14%, P<0.001). However, immunodeficiency in this subset of the controls was represented over twice as often as that of the baseline CTLs (14% vs 6%), whereas GI disease was similar between the 2 groups (37% vs 31%, P<0.40). PPI use demonstrated a suggestive relationship (22% vs 11%, P<0.07).

Data for the third subanalysis between CA‐CD and HA‐CD are shown on Table 3. We initially compared CA‐CD, community‐onset HA‐CD, and hospital‐onset HA‐CD. However, when stratification was found to not be significant, we combined both categories of HA‐CD into 1 group. CA‐CD and HA‐CD did not demonstrate significant difference in antibiotic use, type, prophylaxis, history of abdominal surgery, immunodeficiency, or GI disease. Bloody stools were more common in CA‐CD.

Odds ratio (OR) was calculated for association of individual risk factors for disease between CDs and CTLs (Table 4). Our model controlled for antibiotics use in the past 90 days, PPI use, treatment with 2 or more antibiotics, recent hospitalization, past history of CDIs, history of GI disease, history of abdominal surgery, and being immunodeficient. Antibiotic use within the past 90 days (OR: 2.80, P=0.001), recent hospitalization (OR: 2.33, P=0.007), and immunodeficiency (OR: 6.02, P<0.001) were associated with having C difficile. A similar logistic regression was conducted using a model comparing community‐ and hospital‐acquired cases, but no difference was found among risk factors.

DISCUSSION/CONCLUSION

Our study shows that in addition to traditional risk factors of antibiotic use and recent hospitalization, immunodeficiency is a significant key factor associated with the diagnosis of CD. We found that traditional risk factors are not present in all hospitalized pediatric patients with CD. Our study does not support routine testing for C difficile in patients with diarrhea; however, it does suggest testing children with persistent or severe diarrheal symptoms even if traditional risk factors are absent, especially in the presence of immunodeficiency. The intervals we used for antibiotic exposure (past 90 days) and recent hospitalization (past 6 months) were longer compared to other studies,[9, 12] making our findings even more meaningful. Although some of the 11 patients without traditional risk factors had the presence of clinical factors shown in previous studies to be more common in patients with CDIs (GI disease, GI surgery, gastric tube/nasogastric feeding),[12, 13] we still find 4 patients >1 year of age with CDIs and no risk factors. This echoes the CDCs concerns of CDIs in low‐risk patients.[3]

Unlike clinical history, we found clinical symptoms and basic electrolyte testing may not help to distinguish CD from AGE patients. Although abdominal pain and diarrhea on admission were significantly more common in CTLs, when including abdominal pain and diarrhea during hospitalization, this finding was no longer valid. Additionally, although overall return rate was higher for CDs, the return rate for GI symptoms specifically was not different. The former was instead most often due to complications associated with comorbid conditions (GI disease, immunodeficiency). We did assess LOS for both CDs and CTLs; however, due to the high percentage of CDs with malignancy and other severe illnesses, it was difficult to ascertain the effect of CDIs on LOS. Severe CD is described as admission to the intensive care unit due to C difficile complications, colectomy, and death secondary to C difficile.[11] Although our study did not look at severe CDI as a direct outcome, we did not have any cases of colectomy or death secondary to CDI.

Two recent studies[9, 14] showed a high percentage of acid suppression medication use in patients with CDIs, with 1 study reporting 60% using PPIs and 21% using histamine blockers. Our study initially found similar high levels of PPI use among patients with CDIs; however, no significance was found when controlling for chronic disease. Prescriptions of PPIs for pediatric patients have risen dramatically recently,[15] as have reported all‐cause complications.[16] Further studies are needed to evaluate the independent risks of PPI use and CDIs in children. We were unable to analyze the influence of antibiotic use at prophylactic levels on CD rates, as the majority our CDs were on prophylaxis due to immunodeficiency.

Our study is unique in many ways. It is the first study to evaluate hospitalized pediatric patients with a comprehensive list of potential risk factors for CDIs, looking at clinical data on admission and during hospitalization. Additionally, as our site archives all clinical information in 1 database, we were able to identify ED/UC return and hospital readmissions. Although it is possible patients may have been evaluated outside of our healthcare system, this would be uncommon due to our referral patterns and UC sites. Our study used age‐matched patients with diarrheal symptoms and AGE discharge diagnosis as the control group. This differs from the 1 previous study looking at risk factors for CDIs in children.[9] In that study, researchers used patients with negative C difficile toxin testing as controls. Our subanalysis of CTLs with a negative toxin assay found much higher rates of underlying GI disease and immunodeficiency. Whereas previous studies compared patients already at high risk for CDI and assessed the differences between those with and without the infection, our study looked at what clinical factors distinguish CDI from AGE in a hospitalized population.

Similar to other pediatric studies, our study found a significant number of CA‐CD. However our study is 1 of the first to compare pediatric CA‐CD with HA‐CD based on clinical factors. Of the 9 demographic and clinical variables assessed, the only significant difference found was presence of bloody diarrhea. It may be that bloody diarrhea prompted the patients to be admitted as opposed to evaluated in the ambulatory setting.

Our study had some limitations. We used ICD‐9 discharge diagnosis codes to identify our patients; however, thorough chart review found clinical indices (diarrhea and abdominal pain) that correlated well with CDI diagnosis in addition to positive laboratory test. The EIA C difficile toxin assay was the standard of care during our study period. However, a recent study has shown false positives using EIA testing in pediatric populations.[17] In our primary analysis, we did not exclude patients with a past history of CDIs. Recurrent CDI is defined as having symptoms within 8 weeks after the primary infection. Of our patients with a history of CDIs, only 2 met this definition. Due to the small number, excluding these patients would not have changed our results significantly. Last, as with any retrospective study, we relied on caregiver reports regarding clinical history, especially in the CA‐CD cohort.

Based on our comprehensive analysis of pediatric patients, there should be increased suspicion for CDI in children with baseline immunodeficiency. Our study also supports testing children with persistent or severe GI symptoms even in the absence of traditional risk factors. These elements, coupled with history of antibiotic use, recent hospitalization, GI disease, and abdominal surgery could be used to create an assessment tool to assist clinicians in the diagnosis of CDIs in pediatric patients. A significant percentage of CDIs continues to be CA‐CD. HA‐CD and CA‐CD patients have similar clinical features. Further studies are needed to determine the effect of PPI use and prophylactic antibiotics on CDIs in children.

Clostridium difficile is the single most common cause of nosocomial diarrhea in both adults and children.[1, 2] C difficile infections (CDIs) can range from self‐limited diarrhea to severe pseudomembranous colitis. Though widely distributed in the environment, hospitals and child care facilities are major reservoirs for C difficile. Traditionally, hospitalization and antibiotic use have been the 2 major risk factors for acquiring CDI.

Recent studies suggest C difficile epidemiology is shifting. In 2005, the Centers for Disease Control and Prevention (CDC) reported CDIs in 33 otherwise low‐risk patients, 6 of whom were children.[3] Other studies have noted increasing incidence of pediatric CDIs,[4, 5, 6, 7] 1 identifying 43% with no prior antibiotic use.[4] This emerging data led to the recent American Academy of Pediatrics policy statement on pediatric CDIs.[8] Data regarding associated clinical risk factors of CDIs in pediatric patients in light of the changing epidemiology are limited. Only 1 recent study looked at 6 clinical factors and found that antibiotic use, history of solid organ transplantation, gastrointestinal (GI) devices, and acid suppressing medications increased risk for CDIs.[9]

Data regarding the source of these infections are also limited. Three pediatric studies evaluating source found a significant amount of community‐acquired disease (59%, 25%, and 19% of the study population, respectively).[4, 9, 10] However, only 1 of these studies provided clinical comparisons between community and hospital‐acquired cases.[10] To date, no study has examined a comprehensive list of potential risk factors that might differentiate hospitalized pediatric patients with CDIs from those with acute gastroenteritis (AGE).

PATIENTS AND METHODS

We conducted an investigator‐initiated, retrospective, case‐control study examining risk factors associated with CDIs in a hospitalized pediatric population at Rady Children's Hospital San Diego (RCHSD). Rady Children's is a tertiary‐care pediatric healthcare system and the sole pediatric referral center for San Diego, with a catchment of 850,000 children. RCHSD posts over 71,000 emergency department (ED) and 30,000 urgent care (UC) visits at 4 sites and over 15,000 admissions yearly. All system information is archived in 1 electronic database. We reviewed patient records for a 2‐year period from June 1, 2008 through May 31, 2010. The study protocol was reviewed and approved by the institutional review board at the University of California San Diego.

Cases of C difficile (CDs) included pediatric patients 18 years of age with all of the following: International Classification of Diseases, 9th Revision (ICD‐9) code for C difficile infection (08.45), a positive C difficile toxin A or B by enzyme immunoassay (EIA) (Meridian Bioscience, Inc., Cincinnati, OH), and the presence of diarrhea and/or abdominal pain. Randomly selected age‐matched controls from the same time period with a discharge diagnosis of AGE (APR‐DRG 249) and the presence of diarrhea served as controls (CTLs). In the 1 year age group, any patient with a positive C difficile toxin assay but no diagnosis of CDI was excluded from the CTL group to avoid potential confounding.

Records were reviewed for multiple potential risk factors based on limited past studies and other factors associated with CDI pathogenesis including age, race, ethnicity, antibiotic use within the previous 90 days (type, route, and duration), diarrhea type, abdominal pain, fever, proton pump inhibitor (PPI) use, sick contacts (diarrheal illness), recent travel, and hospitalization within the last 6 months. Diarrhea was defined as increase in stool frequency or volume. Past medical/surgical history abstracted included GI disease, past CDIs, abdominal surgery, immunodeficiency, renal disease, cardiac disease, nutritional deficiencies, and number of past hospitalizations (all cause). In addition, multiple factors during the hospital course were reviewed: length of stay (LOS), antibiotic therapy, diarrhea type, abdominal pain, fever, electrolyte levels, need for stool replacement fluid, and altered diet recommendations. Thirty‐day return to ED/UC or readmission and cause for the return were also retrieved on all patients. An objective data collection form was used, and all records were reviewed by 1 researcher (W.S.) with a second reviewer (E.F.) reviewing 20% of the charts, with 90% initial concordance. Consensus was reached on all elements abstracted.

Three additional subanalyses were completed. The first subanalysis compared antibiotic prophylaxis (defined as daily use of an antibiotic for >28 days) in CDs versus CTLs. We reviewed charts to ensure extended antibiotic use was for prophylaxis and not treatment. The second subanalysis compared CDs to those CTLs with a negative C difficile toxin assay. This was done to evaluate whether using this control group would highlight a different set of risk factors. The third subanalysis separated CDs into community‐acquired CD (CA‐CD) and hospital‐acquired CD (HA‐CD). We defined CA‐CD as any patient with symptoms either prior to or within the first 48 hours of the index admission and no past hospitalizations or with the last hospitalization >4 weeks prior to the index admission. Patients who developed symptoms at home or within 48 hours of the index admission, but had been hospitalized within the past 4 weeks, were defined as community‐onset HA‐CD. Patients who developed CDIs after 48 hours of the index admission were defined as hospital‐onset HA‐CD. These groupings are consistent with the CDI surveillance recommendations.[11]

All statistical analyses were performed with SPSS statistical software version 21.0 (SPSS Inc., Chicago, IL). Initial comparisons between CDs and CTLs were conducted using t tests for continuous variables and [2] tests for categorical variables. As CDI in infants is controversial, we analyzed our data with and without this cohort to eliminate extraneous, age‐related differences. After confirming that there were no issues with tolerance among possibly related factors, a saturated multiple logistic regression model was used to determine which of the independent variables identified in the initial comparison were predictors of having C difficile when controlling for factors associated with chronic disease.

RESULTS

Descriptive characteristics of the 134 CDs and the 274 CTLs are provided in Table 1. CDs and CTLs were similar in gender and race. More CDs had recent hospitalization and antibiotic exposure, with 24% of CDs versus 3% of CTLs treated with 2 or more antibiotics. Watery stools were the most common type of diarrhea in both CDs and CTLs, and bloody stools did not differ significantly between the 2 groups. However, abdominal pain on admission was more common in CTLs. CDs were more likely to have a history of GI disease, abdominal surgery, and specifically GI surgery. Immunodeficiency and PPI use were far more frequent in CDs, whereas exposure to sick contacts was more common in CTLs. Although CDs had an overall higher rate of ED/UC return visits and readmissions, the rate of return due to GI symptoms was similar in both groups. Reanalysis of the data with the <1‐year cohort removed showed persistent statistically significant findings in these variables. Hospital course, including electrolyte levels, need for intravenous fluids, or modified diets, did not significantly differ between CDs and CTLs (data not shown).

Analysis of CDs without traditional risk factors was performed. To identify patients, we first selected the 46 (34%) without prior antibiotic exposure, then eliminated 19 who had been hospitalized within the past 6 months. Of the remaining 27 patients, 16 had a prolonged hospitalization (>5 days) at the time of CDI diagnosis. This left us with 11 patients (8% of CDs) without any common risk factors of antibiotic use, recent hospitalization, or prolonged hospitalization. None of these patients had a history of CDIs; 6 had significant medical histories. A detailed description of these 11 patients if provided in Table 2.

The first subanalysis evaluated antibiotic prophylaxis and found 51 (37%) in CDs versus 10 (4%) in CTLs. However, after controlling for immunodeficiency found in 40 of these CDs, we found no statistically significant difference. There were insufficient numbers of those on prophylaxis for other reasons (eg, vesicoureteral reflux) to analyze prophylaxis independently.

The second subanalysis compared controls with a negative C difficile toxin assay (21% of CTLs) to CDs on a number of clinical factors. Results were compared to the primary analysis. Many factors remained significant: antibiotic use in the past 90 days was still more frequent in CDs (66% vs 35%, P<0.001) as was immunodeficiency in CDs (46% vs 14%, P<0.001). However, immunodeficiency in this subset of the controls was represented over twice as often as that of the baseline CTLs (14% vs 6%), whereas GI disease was similar between the 2 groups (37% vs 31%, P<0.40). PPI use demonstrated a suggestive relationship (22% vs 11%, P<0.07).

Data for the third subanalysis between CA‐CD and HA‐CD are shown on Table 3. We initially compared CA‐CD, community‐onset HA‐CD, and hospital‐onset HA‐CD. However, when stratification was found to not be significant, we combined both categories of HA‐CD into 1 group. CA‐CD and HA‐CD did not demonstrate significant difference in antibiotic use, type, prophylaxis, history of abdominal surgery, immunodeficiency, or GI disease. Bloody stools were more common in CA‐CD.

Odds ratio (OR) was calculated for association of individual risk factors for disease between CDs and CTLs (Table 4). Our model controlled for antibiotics use in the past 90 days, PPI use, treatment with 2 or more antibiotics, recent hospitalization, past history of CDIs, history of GI disease, history of abdominal surgery, and being immunodeficient. Antibiotic use within the past 90 days (OR: 2.80, P=0.001), recent hospitalization (OR: 2.33, P=0.007), and immunodeficiency (OR: 6.02, P<0.001) were associated with having C difficile. A similar logistic regression was conducted using a model comparing community‐ and hospital‐acquired cases, but no difference was found among risk factors.

DISCUSSION/CONCLUSION

Our study shows that in addition to traditional risk factors of antibiotic use and recent hospitalization, immunodeficiency is a significant key factor associated with the diagnosis of CD. We found that traditional risk factors are not present in all hospitalized pediatric patients with CD. Our study does not support routine testing for C difficile in patients with diarrhea; however, it does suggest testing children with persistent or severe diarrheal symptoms even if traditional risk factors are absent, especially in the presence of immunodeficiency. The intervals we used for antibiotic exposure (past 90 days) and recent hospitalization (past 6 months) were longer compared to other studies,[9, 12] making our findings even more meaningful. Although some of the 11 patients without traditional risk factors had the presence of clinical factors shown in previous studies to be more common in patients with CDIs (GI disease, GI surgery, gastric tube/nasogastric feeding),[12, 13] we still find 4 patients >1 year of age with CDIs and no risk factors. This echoes the CDCs concerns of CDIs in low‐risk patients.[3]

Unlike clinical history, we found clinical symptoms and basic electrolyte testing may not help to distinguish CD from AGE patients. Although abdominal pain and diarrhea on admission were significantly more common in CTLs, when including abdominal pain and diarrhea during hospitalization, this finding was no longer valid. Additionally, although overall return rate was higher for CDs, the return rate for GI symptoms specifically was not different. The former was instead most often due to complications associated with comorbid conditions (GI disease, immunodeficiency). We did assess LOS for both CDs and CTLs; however, due to the high percentage of CDs with malignancy and other severe illnesses, it was difficult to ascertain the effect of CDIs on LOS. Severe CD is described as admission to the intensive care unit due to C difficile complications, colectomy, and death secondary to C difficile.[11] Although our study did not look at severe CDI as a direct outcome, we did not have any cases of colectomy or death secondary to CDI.

Two recent studies[9, 14] showed a high percentage of acid suppression medication use in patients with CDIs, with 1 study reporting 60% using PPIs and 21% using histamine blockers. Our study initially found similar high levels of PPI use among patients with CDIs; however, no significance was found when controlling for chronic disease. Prescriptions of PPIs for pediatric patients have risen dramatically recently,[15] as have reported all‐cause complications.[16] Further studies are needed to evaluate the independent risks of PPI use and CDIs in children. We were unable to analyze the influence of antibiotic use at prophylactic levels on CD rates, as the majority our CDs were on prophylaxis due to immunodeficiency.

Our study is unique in many ways. It is the first study to evaluate hospitalized pediatric patients with a comprehensive list of potential risk factors for CDIs, looking at clinical data on admission and during hospitalization. Additionally, as our site archives all clinical information in 1 database, we were able to identify ED/UC return and hospital readmissions. Although it is possible patients may have been evaluated outside of our healthcare system, this would be uncommon due to our referral patterns and UC sites. Our study used age‐matched patients with diarrheal symptoms and AGE discharge diagnosis as the control group. This differs from the 1 previous study looking at risk factors for CDIs in children.[9] In that study, researchers used patients with negative C difficile toxin testing as controls. Our subanalysis of CTLs with a negative toxin assay found much higher rates of underlying GI disease and immunodeficiency. Whereas previous studies compared patients already at high risk for CDI and assessed the differences between those with and without the infection, our study looked at what clinical factors distinguish CDI from AGE in a hospitalized population.

Similar to other pediatric studies, our study found a significant number of CA‐CD. However our study is 1 of the first to compare pediatric CA‐CD with HA‐CD based on clinical factors. Of the 9 demographic and clinical variables assessed, the only significant difference found was presence of bloody diarrhea. It may be that bloody diarrhea prompted the patients to be admitted as opposed to evaluated in the ambulatory setting.

Our study had some limitations. We used ICD‐9 discharge diagnosis codes to identify our patients; however, thorough chart review found clinical indices (diarrhea and abdominal pain) that correlated well with CDI diagnosis in addition to positive laboratory test. The EIA C difficile toxin assay was the standard of care during our study period. However, a recent study has shown false positives using EIA testing in pediatric populations.[17] In our primary analysis, we did not exclude patients with a past history of CDIs. Recurrent CDI is defined as having symptoms within 8 weeks after the primary infection. Of our patients with a history of CDIs, only 2 met this definition. Due to the small number, excluding these patients would not have changed our results significantly. Last, as with any retrospective study, we relied on caregiver reports regarding clinical history, especially in the CA‐CD cohort.

Based on our comprehensive analysis of pediatric patients, there should be increased suspicion for CDI in children with baseline immunodeficiency. Our study also supports testing children with persistent or severe GI symptoms even in the absence of traditional risk factors. These elements, coupled with history of antibiotic use, recent hospitalization, GI disease, and abdominal surgery could be used to create an assessment tool to assist clinicians in the diagnosis of CDIs in pediatric patients. A significant percentage of CDIs continues to be CA‐CD. HA‐CD and CA‐CD patients have similar clinical features. Further studies are needed to determine the effect of PPI use and prophylactic antibiotics on CDIs in children.

Healthcare delivery organizations are under increasing pressure to improve patient safety. The fundamental underpinning of efforts to improve safety has been the establishment of a no‐blame culture, one that focuses less on individual transgressions and more on system improvement.[1, 2] As evidence‐based practices to improve care have emerged, and the pressures to deliver tangible improvements in safety and quality have grown, providers, healthcare system leaders, and policymakers are struggling with how best to balance the need for accountability with this no‐blame paradigm.

In dealing with areas such as hand hygiene, where there is strong evidence for the value of the practice yet relatively poor adherence in many institutions, Wachter and Pronovost have argued that the scales need to tip more in the direction of accountability, including the imposition of penalties for clinicians who habitually fail to follow certain safety practices.[3] Although not obviating the critical importance of systems improvement, they argue that a failure to enforce such measures undermines trust in the system and invites external regulation. Chassin and colleagues made a similar point in arguing for the identification of certain accountability measures that could be used in public reporting and pay‐for‐performance programs.[4]

Few organizations have enacted robust systems to hold providers responsible for adhering to accountability measures.[4] Although many hospitals have policies to suspend clinical privileges for failing to sign discharge summaries or obtain a yearly purified protein derivative test, few have formal programs to identify and deal with clinicians whose behavior is persistently problematic.[3] Furthermore, existing modes of physician accountability, such as state licensing boards, only discipline physicians retroactively (and rarely) when healthcare organizations report poor performance. State boards typically do not consider prevention of injury, such as adherence to safety practices, to be part of their responsibility.[5] Similarly, credentialing boards (eg, the American Board of Internal Medicine) do not assess adherence to such practices in coming to their decisions.

It is estimated that strict adherence to infection control practices, such as hand hygiene, could prevent over 100,000 hospital deaths every year; adherence to other evidence‐based safety practices such as the use of a preoperative time‐out would likely prevent many more deaths and cases of medical injury.[3, 6] Although there are practical issues, such as how to audit individual clinician adherence in ways that are feasible and fair, that make enforcing individual provider accountability challenging, there seems little doubt that attitudes regarding the appropriateness of enacting penalties for safety transgressions will be key determinants of whether such measures are considered. Yet no study to date has assessed the opinions of different stakeholders (physicians, nurses, trainees, patients) regarding various strategies, including public reporting and penalties, to improve adherence to safety practices. We aimed to assess these attitudes across a variety of such stakeholders.

METHODS

RESULTS

A total of 287 individuals were given surveys, and 183 completed them: 22 attending physicians, 33 resident physicians, 61 nurses, 47 medical students, and 20 patients (overall response rate 64%). Response rate for attending and resident physicians was 73%, for nurses 59%, and for medical students 54%. Among patients who were approached and agreed to accept a survey, 87% returned completed surveys (Table 1). The average age of attending physicians was 35.8 years (standard deviation [SD]: 5.3), residents was 28.3 years (SD: 1.7), nurses was 43.6 years (SD: 11.1), medical students was 26.6 years (SD: 2.9), and inpatients was 48.2 years (SD: 15.9). Thirty‐two percent of attending physicians were female, 67% of resident physicians were female, 88% of nurses were female, 66% of medical students were female, and 47% of inpatients were female.

Perceived Harm

Out of the 3 scenarios presented in in the survey, participants believed that not conducting preoperative time‐outs in surgery presented the highest risk to patient safety, with 57% (residents) to 86% (nurses) rating the potential harm as high (Figure 1). Not conducting fall risk assessments was perceived as second most potentially harmful, and not properly practicing hand hygiene was perceived as least potentially harmful to patient safety. There were significant differences among groups in perceptions of potential harm for all 3 scenarios (P<0.001 for all).

Figure 1

Appropriateness of Public Reporting and Penalties

Public reporting was viewed as ever appropriate by 34% of all respondents for hand‐hygiene protocol lapses, 58% for surgical time‐out lapses, and 43% for fall risk assessment lapses. There were no significant differences among groups in endorsement of public reporting for individual scenarios (Figure 2). Penalties were endorsed more frequently than public reporting for all groups and all scenarios. The proportion of attending physicians and patients who rated penalties as ever appropriate were similar for each scenario. Residents, medical students, and nurses were less likely than patients and attending physicians to support penalties (P<0.05 for all differences).

Figure 2

The aggregated analysis revealed that nurses and medical students were significantly less likely than patients to endorse public reporting across scenarios. In terms of endorsement of penalties, we found no significant differences between attending physicians and patients, but residents (odds ratio [OR]: 0.09, 95% confidence interval [CI]: 0.03‐0.32), students (OR: 0.12, 95% CI: 0.04‐0.34), and nurses (OR: 0.17, 95% CI: 0.03‐0.41) had significantly lower odds of favoring penalties than did patients (Table 2).

Table 2.Likelihood of Endorsing Public Reporting or Penalties at Any Time by Group and Scenario

	Odds Ratio (95% CI)
	Public Reporting	Penalty
NOTE: Odds ratios and proportions were derived from logistic regression models including group, scenario, age, and sex adjusting for clustering within individuals. Abbreviations: CI, confidence interval.
Group, across all scenarios
Patients	Reference	Reference
Attending physicians	0.58 (0.172.01)	0.88 (0.203.84)
Resident physicians	0.42 (0.121.52)	0.09 (0.020.32)
Nurses	0.32 (0.120.88)	0.17 (0.030.41)
Medical students	0.22 (0.060.80)	0.12 (0.040.34)
Scenario, across all groups
Hand hygiene	Reference	Reference
Surgical time‐out	2.82 (2.033.91)	4.29 (2.976.20)
Fall assessment	1.47 (1.091.98)	1.74 (1.272.37)

Across all surveyed groups, public reporting was more often supported for lapses of surgical timeout (OR: 2.82, 95% CI: 2.03‐3.91) and fall risk assessment protocols (OR: 1.47, 95% CI: 1.09‐1.98) than for the referent, hand‐hygiene lapses. Across all groups, penalties were more likely to be supported for surgical timeout (OR: 4.29, 95% CI: 2.97‐6.20) and fall risk assessment protocol lapses (OR: 1.74, 95% CI: 1.27‐2.37) than for hand‐hygiene lapses.

Thresholds for Public Reporting and Penalties

The log‐rank test showed no significant differences among the surveyed groups in the number of transgressions at which public reporting was deemed appropriate in any of the 3 scenarios (P=0.37, P=0.71, and P=0.32 for hand hygiene, surgical time‐out, and fall risk assessment, respectively) (Figure 3). However, patients endorsed penalties after significantly fewer occurrences than residents, medical students, and nurses for all 3 scenarios (P<0.001 for all differences), and at a significantly lower threshold than attending physicians for surgical timeout and fall risk assessment (P<0.001 and P=0.03, respectively).

Figure 3

DISCUSSION

This survey assessed attitudes of healthcare professionals, trainees, and inpatients toward public reporting and penalties when clinicians do not follow basic safety protocols. Respondents tended to favor more aggressive measures when they deemed the safety risk from protocol violations to be higher. Almost all participants favored providing feedback after safety protocol lapses. Healthcare professionals tended to favor punitive measures, such as fines, suspension, and firing, more than public reporting of transgressions. Patients had a lower threshold than both providers and trainees for public reporting and punitive measures. In aggregate, our study suggests that after a decade of emphasis on a no‐blame response to patient safety hazards, both healthcare providers and patients now believe clinicians should be held accountable for following basic safety protocols, though their thresholds and triggers vary.

A surprising finding was that providers were more likely to favor penalties (such as fines, suspension, or firing) than public reporting of safety transgressions. Multiple studies have suggested that public reporting of hospital quality data has improved adherence to care processes and may improve patient outcomes.[15, 16, 17] Although our data do not tell us why clinicians appear to be more worried about public reporting than penalties, they do help explain why transparency has been a relatively powerful strategy to motivate changes in practice, even when it is unaccompanied by significant shifts in consumer choices.[18] It would be natural to consider public reporting to be a softer strategy than fines, suspension, or firing; however, our results indicate that many clinicians do not see it that way. Alternatively, the results could also suggest that clinicians prefer measures that provide more immediate feedback than public reporting generally provides. These attitudes should be considered when enacting public reporting strategies.

Another interesting finding was that patients and attending physicians tended to track together regarding their attitudes toward penalties for safety lapses. Although patients had a lower threshold for favoring penalties than attendings, similar proportions of patients and attending physicians believed that penalties should be enacted for safety transgressions, and both groups were more penal than physician trainees and nurses. We speculate that attendings and patients may have the most skin in the game, patients as the ones directly harmed by a preventable adverse event, and attending physicians as the most responsible clinicians, at least in the eyes of the malpractice system, licensing boards, and credentials committees.

Even though our study illustrates relatively high levels of endorsement for aggressive measures to deal with clinicians who fail to follow evidence‐based safety practices, a shift in this direction has risks and benefits. The no‐blame paradigm in patient safety grew out of a need to encourage open discussion about medical mistakes.[2] Whereas shifting away from a purely no‐ blame approach may lead to greater adherence with safety practices, and one hopes fewer cases of preventable harm, it also risks stifling the open discussions about medical errors that characterize learning organizations.[13, 19] Because of this, a movement in this direction should be undertaken carefully, starting first with a small number of well‐established safety practices, and ensuring that robust education and system improvements precede and accompany the imposition of penalties for nonadherence.

Our study has limitations. The survey was developed using convenience samples of UCSF faculty and medical students, so broader inclusion of physicians, nurses, trainees, and patients may have yielded a different survey instrument. As a survey, we cannot be certain that any of the groups' responses in real life (eg, in a vote of the medical staff on a given policy) would mirror their survey response. Additionally, the responses to protocol lapses did not include all possible administrative responses, such as mandatory training/remediation or rewards for positive behaviors. The responses could have also been different if participants were presented with different patient safety scenarios. The study population was limited in several ways. Attending and resident physicians were drawn from an academic department of internal medicine; it is possible that other specialties would have different attitudes. Patients were relatively young (likely due to the inclusion criteria), as were attending physicians (due to oversampling of hospitalist physicians). The relatively small number of participants could also limit statistical power to detect differences among groups. Additionally, the study population was limited to patients and healthcare professionals in academic medical centers in San Francisco. It is possible that attitudes would be different in other regions and practice settings.

The no‐blame approach to patient safety has been crucial in refocusing the lens on systems failures and in encouraging the active engagement by clinicians, particularly physicians.[2, 3] On the other hand, there are legitimate concerns that a unidimensional no‐blame approach has permitted, perhaps even promoted, nonadherence to evidence‐based safety practices that could prevent many cases of harm. Although it may not be surprising that patients favor harsher consequences for providers who do not follow basic safety protocols, our study demonstrates relatively widespread support for such consequences even among clinicians and trainees. However, all groups appear to recognize the nuances underlying this set of issues, with varying levels of enthusiasm for punitive responses based on perceived risk and number of transgressions. Future studies are needed to investigate how best to implement public reporting and penalties in ways that can maximize the patient safety benefits.

Healthcare delivery organizations are under increasing pressure to improve patient safety. The fundamental underpinning of efforts to improve safety has been the establishment of a no‐blame culture, one that focuses less on individual transgressions and more on system improvement.[1, 2] As evidence‐based practices to improve care have emerged, and the pressures to deliver tangible improvements in safety and quality have grown, providers, healthcare system leaders, and policymakers are struggling with how best to balance the need for accountability with this no‐blame paradigm.

In dealing with areas such as hand hygiene, where there is strong evidence for the value of the practice yet relatively poor adherence in many institutions, Wachter and Pronovost have argued that the scales need to tip more in the direction of accountability, including the imposition of penalties for clinicians who habitually fail to follow certain safety practices.[3] Although not obviating the critical importance of systems improvement, they argue that a failure to enforce such measures undermines trust in the system and invites external regulation. Chassin and colleagues made a similar point in arguing for the identification of certain accountability measures that could be used in public reporting and pay‐for‐performance programs.[4]

Few organizations have enacted robust systems to hold providers responsible for adhering to accountability measures.[4] Although many hospitals have policies to suspend clinical privileges for failing to sign discharge summaries or obtain a yearly purified protein derivative test, few have formal programs to identify and deal with clinicians whose behavior is persistently problematic.[3] Furthermore, existing modes of physician accountability, such as state licensing boards, only discipline physicians retroactively (and rarely) when healthcare organizations report poor performance. State boards typically do not consider prevention of injury, such as adherence to safety practices, to be part of their responsibility.[5] Similarly, credentialing boards (eg, the American Board of Internal Medicine) do not assess adherence to such practices in coming to their decisions.

It is estimated that strict adherence to infection control practices, such as hand hygiene, could prevent over 100,000 hospital deaths every year; adherence to other evidence‐based safety practices such as the use of a preoperative time‐out would likely prevent many more deaths and cases of medical injury.[3, 6] Although there are practical issues, such as how to audit individual clinician adherence in ways that are feasible and fair, that make enforcing individual provider accountability challenging, there seems little doubt that attitudes regarding the appropriateness of enacting penalties for safety transgressions will be key determinants of whether such measures are considered. Yet no study to date has assessed the opinions of different stakeholders (physicians, nurses, trainees, patients) regarding various strategies, including public reporting and penalties, to improve adherence to safety practices. We aimed to assess these attitudes across a variety of such stakeholders.

METHODS

RESULTS

A total of 287 individuals were given surveys, and 183 completed them: 22 attending physicians, 33 resident physicians, 61 nurses, 47 medical students, and 20 patients (overall response rate 64%). Response rate for attending and resident physicians was 73%, for nurses 59%, and for medical students 54%. Among patients who were approached and agreed to accept a survey, 87% returned completed surveys (Table 1). The average age of attending physicians was 35.8 years (standard deviation [SD]: 5.3), residents was 28.3 years (SD: 1.7), nurses was 43.6 years (SD: 11.1), medical students was 26.6 years (SD: 2.9), and inpatients was 48.2 years (SD: 15.9). Thirty‐two percent of attending physicians were female, 67% of resident physicians were female, 88% of nurses were female, 66% of medical students were female, and 47% of inpatients were female.

Perceived Harm

Out of the 3 scenarios presented in in the survey, participants believed that not conducting preoperative time‐outs in surgery presented the highest risk to patient safety, with 57% (residents) to 86% (nurses) rating the potential harm as high (Figure 1). Not conducting fall risk assessments was perceived as second most potentially harmful, and not properly practicing hand hygiene was perceived as least potentially harmful to patient safety. There were significant differences among groups in perceptions of potential harm for all 3 scenarios (P<0.001 for all).

Figure 1

Appropriateness of Public Reporting and Penalties

Public reporting was viewed as ever appropriate by 34% of all respondents for hand‐hygiene protocol lapses, 58% for surgical time‐out lapses, and 43% for fall risk assessment lapses. There were no significant differences among groups in endorsement of public reporting for individual scenarios (Figure 2). Penalties were endorsed more frequently than public reporting for all groups and all scenarios. The proportion of attending physicians and patients who rated penalties as ever appropriate were similar for each scenario. Residents, medical students, and nurses were less likely than patients and attending physicians to support penalties (P<0.05 for all differences).

Figure 2

The aggregated analysis revealed that nurses and medical students were significantly less likely than patients to endorse public reporting across scenarios. In terms of endorsement of penalties, we found no significant differences between attending physicians and patients, but residents (odds ratio [OR]: 0.09, 95% confidence interval [CI]: 0.03‐0.32), students (OR: 0.12, 95% CI: 0.04‐0.34), and nurses (OR: 0.17, 95% CI: 0.03‐0.41) had significantly lower odds of favoring penalties than did patients (Table 2).

Table 2.Likelihood of Endorsing Public Reporting or Penalties at Any Time by Group and Scenario

	Odds Ratio (95% CI)
	Public Reporting	Penalty
NOTE: Odds ratios and proportions were derived from logistic regression models including group, scenario, age, and sex adjusting for clustering within individuals. Abbreviations: CI, confidence interval.
Group, across all scenarios
Patients	Reference	Reference
Attending physicians	0.58 (0.172.01)	0.88 (0.203.84)
Resident physicians	0.42 (0.121.52)	0.09 (0.020.32)
Nurses	0.32 (0.120.88)	0.17 (0.030.41)
Medical students	0.22 (0.060.80)	0.12 (0.040.34)
Scenario, across all groups
Hand hygiene	Reference	Reference
Surgical time‐out	2.82 (2.033.91)	4.29 (2.976.20)
Fall assessment	1.47 (1.091.98)	1.74 (1.272.37)

Across all surveyed groups, public reporting was more often supported for lapses of surgical timeout (OR: 2.82, 95% CI: 2.03‐3.91) and fall risk assessment protocols (OR: 1.47, 95% CI: 1.09‐1.98) than for the referent, hand‐hygiene lapses. Across all groups, penalties were more likely to be supported for surgical timeout (OR: 4.29, 95% CI: 2.97‐6.20) and fall risk assessment protocol lapses (OR: 1.74, 95% CI: 1.27‐2.37) than for hand‐hygiene lapses.

Thresholds for Public Reporting and Penalties

The log‐rank test showed no significant differences among the surveyed groups in the number of transgressions at which public reporting was deemed appropriate in any of the 3 scenarios (P=0.37, P=0.71, and P=0.32 for hand hygiene, surgical time‐out, and fall risk assessment, respectively) (Figure 3). However, patients endorsed penalties after significantly fewer occurrences than residents, medical students, and nurses for all 3 scenarios (P<0.001 for all differences), and at a significantly lower threshold than attending physicians for surgical timeout and fall risk assessment (P<0.001 and P=0.03, respectively).

Figure 3

DISCUSSION

This survey assessed attitudes of healthcare professionals, trainees, and inpatients toward public reporting and penalties when clinicians do not follow basic safety protocols. Respondents tended to favor more aggressive measures when they deemed the safety risk from protocol violations to be higher. Almost all participants favored providing feedback after safety protocol lapses. Healthcare professionals tended to favor punitive measures, such as fines, suspension, and firing, more than public reporting of transgressions. Patients had a lower threshold than both providers and trainees for public reporting and punitive measures. In aggregate, our study suggests that after a decade of emphasis on a no‐blame response to patient safety hazards, both healthcare providers and patients now believe clinicians should be held accountable for following basic safety protocols, though their thresholds and triggers vary.

A surprising finding was that providers were more likely to favor penalties (such as fines, suspension, or firing) than public reporting of safety transgressions. Multiple studies have suggested that public reporting of hospital quality data has improved adherence to care processes and may improve patient outcomes.[15, 16, 17] Although our data do not tell us why clinicians appear to be more worried about public reporting than penalties, they do help explain why transparency has been a relatively powerful strategy to motivate changes in practice, even when it is unaccompanied by significant shifts in consumer choices.[18] It would be natural to consider public reporting to be a softer strategy than fines, suspension, or firing; however, our results indicate that many clinicians do not see it that way. Alternatively, the results could also suggest that clinicians prefer measures that provide more immediate feedback than public reporting generally provides. These attitudes should be considered when enacting public reporting strategies.

Another interesting finding was that patients and attending physicians tended to track together regarding their attitudes toward penalties for safety lapses. Although patients had a lower threshold for favoring penalties than attendings, similar proportions of patients and attending physicians believed that penalties should be enacted for safety transgressions, and both groups were more penal than physician trainees and nurses. We speculate that attendings and patients may have the most skin in the game, patients as the ones directly harmed by a preventable adverse event, and attending physicians as the most responsible clinicians, at least in the eyes of the malpractice system, licensing boards, and credentials committees.

Even though our study illustrates relatively high levels of endorsement for aggressive measures to deal with clinicians who fail to follow evidence‐based safety practices, a shift in this direction has risks and benefits. The no‐blame paradigm in patient safety grew out of a need to encourage open discussion about medical mistakes.[2] Whereas shifting away from a purely no‐ blame approach may lead to greater adherence with safety practices, and one hopes fewer cases of preventable harm, it also risks stifling the open discussions about medical errors that characterize learning organizations.[13, 19] Because of this, a movement in this direction should be undertaken carefully, starting first with a small number of well‐established safety practices, and ensuring that robust education and system improvements precede and accompany the imposition of penalties for nonadherence.

Our study has limitations. The survey was developed using convenience samples of UCSF faculty and medical students, so broader inclusion of physicians, nurses, trainees, and patients may have yielded a different survey instrument. As a survey, we cannot be certain that any of the groups' responses in real life (eg, in a vote of the medical staff on a given policy) would mirror their survey response. Additionally, the responses to protocol lapses did not include all possible administrative responses, such as mandatory training/remediation or rewards for positive behaviors. The responses could have also been different if participants were presented with different patient safety scenarios. The study population was limited in several ways. Attending and resident physicians were drawn from an academic department of internal medicine; it is possible that other specialties would have different attitudes. Patients were relatively young (likely due to the inclusion criteria), as were attending physicians (due to oversampling of hospitalist physicians). The relatively small number of participants could also limit statistical power to detect differences among groups. Additionally, the study population was limited to patients and healthcare professionals in academic medical centers in San Francisco. It is possible that attitudes would be different in other regions and practice settings.

The no‐blame approach to patient safety has been crucial in refocusing the lens on systems failures and in encouraging the active engagement by clinicians, particularly physicians.[2, 3] On the other hand, there are legitimate concerns that a unidimensional no‐blame approach has permitted, perhaps even promoted, nonadherence to evidence‐based safety practices that could prevent many cases of harm. Although it may not be surprising that patients favor harsher consequences for providers who do not follow basic safety protocols, our study demonstrates relatively widespread support for such consequences even among clinicians and trainees. However, all groups appear to recognize the nuances underlying this set of issues, with varying levels of enthusiasm for punitive responses based on perceived risk and number of transgressions. Future studies are needed to investigate how best to implement public reporting and penalties in ways that can maximize the patient safety benefits.

Healthcare delivery organizations are under increasing pressure to improve patient safety. The fundamental underpinning of efforts to improve safety has been the establishment of a no‐blame culture, one that focuses less on individual transgressions and more on system improvement.[1, 2] As evidence‐based practices to improve care have emerged, and the pressures to deliver tangible improvements in safety and quality have grown, providers, healthcare system leaders, and policymakers are struggling with how best to balance the need for accountability with this no‐blame paradigm.

In dealing with areas such as hand hygiene, where there is strong evidence for the value of the practice yet relatively poor adherence in many institutions, Wachter and Pronovost have argued that the scales need to tip more in the direction of accountability, including the imposition of penalties for clinicians who habitually fail to follow certain safety practices.[3] Although not obviating the critical importance of systems improvement, they argue that a failure to enforce such measures undermines trust in the system and invites external regulation. Chassin and colleagues made a similar point in arguing for the identification of certain accountability measures that could be used in public reporting and pay‐for‐performance programs.[4]

Few organizations have enacted robust systems to hold providers responsible for adhering to accountability measures.[4] Although many hospitals have policies to suspend clinical privileges for failing to sign discharge summaries or obtain a yearly purified protein derivative test, few have formal programs to identify and deal with clinicians whose behavior is persistently problematic.[3] Furthermore, existing modes of physician accountability, such as state licensing boards, only discipline physicians retroactively (and rarely) when healthcare organizations report poor performance. State boards typically do not consider prevention of injury, such as adherence to safety practices, to be part of their responsibility.[5] Similarly, credentialing boards (eg, the American Board of Internal Medicine) do not assess adherence to such practices in coming to their decisions.

It is estimated that strict adherence to infection control practices, such as hand hygiene, could prevent over 100,000 hospital deaths every year; adherence to other evidence‐based safety practices such as the use of a preoperative time‐out would likely prevent many more deaths and cases of medical injury.[3, 6] Although there are practical issues, such as how to audit individual clinician adherence in ways that are feasible and fair, that make enforcing individual provider accountability challenging, there seems little doubt that attitudes regarding the appropriateness of enacting penalties for safety transgressions will be key determinants of whether such measures are considered. Yet no study to date has assessed the opinions of different stakeholders (physicians, nurses, trainees, patients) regarding various strategies, including public reporting and penalties, to improve adherence to safety practices. We aimed to assess these attitudes across a variety of such stakeholders.

METHODS

RESULTS

A total of 287 individuals were given surveys, and 183 completed them: 22 attending physicians, 33 resident physicians, 61 nurses, 47 medical students, and 20 patients (overall response rate 64%). Response rate for attending and resident physicians was 73%, for nurses 59%, and for medical students 54%. Among patients who were approached and agreed to accept a survey, 87% returned completed surveys (Table 1). The average age of attending physicians was 35.8 years (standard deviation [SD]: 5.3), residents was 28.3 years (SD: 1.7), nurses was 43.6 years (SD: 11.1), medical students was 26.6 years (SD: 2.9), and inpatients was 48.2 years (SD: 15.9). Thirty‐two percent of attending physicians were female, 67% of resident physicians were female, 88% of nurses were female, 66% of medical students were female, and 47% of inpatients were female.

Perceived Harm

Out of the 3 scenarios presented in in the survey, participants believed that not conducting preoperative time‐outs in surgery presented the highest risk to patient safety, with 57% (residents) to 86% (nurses) rating the potential harm as high (Figure 1). Not conducting fall risk assessments was perceived as second most potentially harmful, and not properly practicing hand hygiene was perceived as least potentially harmful to patient safety. There were significant differences among groups in perceptions of potential harm for all 3 scenarios (P<0.001 for all).

Figure 1

Appropriateness of Public Reporting and Penalties

Public reporting was viewed as ever appropriate by 34% of all respondents for hand‐hygiene protocol lapses, 58% for surgical time‐out lapses, and 43% for fall risk assessment lapses. There were no significant differences among groups in endorsement of public reporting for individual scenarios (Figure 2). Penalties were endorsed more frequently than public reporting for all groups and all scenarios. The proportion of attending physicians and patients who rated penalties as ever appropriate were similar for each scenario. Residents, medical students, and nurses were less likely than patients and attending physicians to support penalties (P<0.05 for all differences).

Figure 2

The aggregated analysis revealed that nurses and medical students were significantly less likely than patients to endorse public reporting across scenarios. In terms of endorsement of penalties, we found no significant differences between attending physicians and patients, but residents (odds ratio [OR]: 0.09, 95% confidence interval [CI]: 0.03‐0.32), students (OR: 0.12, 95% CI: 0.04‐0.34), and nurses (OR: 0.17, 95% CI: 0.03‐0.41) had significantly lower odds of favoring penalties than did patients (Table 2).

Table 2.Likelihood of Endorsing Public Reporting or Penalties at Any Time by Group and Scenario

	Odds Ratio (95% CI)
	Public Reporting	Penalty
NOTE: Odds ratios and proportions were derived from logistic regression models including group, scenario, age, and sex adjusting for clustering within individuals. Abbreviations: CI, confidence interval.
Group, across all scenarios
Patients	Reference	Reference
Attending physicians	0.58 (0.172.01)	0.88 (0.203.84)
Resident physicians	0.42 (0.121.52)	0.09 (0.020.32)
Nurses	0.32 (0.120.88)	0.17 (0.030.41)
Medical students	0.22 (0.060.80)	0.12 (0.040.34)
Scenario, across all groups
Hand hygiene	Reference	Reference
Surgical time‐out	2.82 (2.033.91)	4.29 (2.976.20)
Fall assessment	1.47 (1.091.98)	1.74 (1.272.37)

Across all surveyed groups, public reporting was more often supported for lapses of surgical timeout (OR: 2.82, 95% CI: 2.03‐3.91) and fall risk assessment protocols (OR: 1.47, 95% CI: 1.09‐1.98) than for the referent, hand‐hygiene lapses. Across all groups, penalties were more likely to be supported for surgical timeout (OR: 4.29, 95% CI: 2.97‐6.20) and fall risk assessment protocol lapses (OR: 1.74, 95% CI: 1.27‐2.37) than for hand‐hygiene lapses.

Thresholds for Public Reporting and Penalties

The log‐rank test showed no significant differences among the surveyed groups in the number of transgressions at which public reporting was deemed appropriate in any of the 3 scenarios (P=0.37, P=0.71, and P=0.32 for hand hygiene, surgical time‐out, and fall risk assessment, respectively) (Figure 3). However, patients endorsed penalties after significantly fewer occurrences than residents, medical students, and nurses for all 3 scenarios (P<0.001 for all differences), and at a significantly lower threshold than attending physicians for surgical timeout and fall risk assessment (P<0.001 and P=0.03, respectively).

Figure 3

DISCUSSION

This survey assessed attitudes of healthcare professionals, trainees, and inpatients toward public reporting and penalties when clinicians do not follow basic safety protocols. Respondents tended to favor more aggressive measures when they deemed the safety risk from protocol violations to be higher. Almost all participants favored providing feedback after safety protocol lapses. Healthcare professionals tended to favor punitive measures, such as fines, suspension, and firing, more than public reporting of transgressions. Patients had a lower threshold than both providers and trainees for public reporting and punitive measures. In aggregate, our study suggests that after a decade of emphasis on a no‐blame response to patient safety hazards, both healthcare providers and patients now believe clinicians should be held accountable for following basic safety protocols, though their thresholds and triggers vary.

A surprising finding was that providers were more likely to favor penalties (such as fines, suspension, or firing) than public reporting of safety transgressions. Multiple studies have suggested that public reporting of hospital quality data has improved adherence to care processes and may improve patient outcomes.[15, 16, 17] Although our data do not tell us why clinicians appear to be more worried about public reporting than penalties, they do help explain why transparency has been a relatively powerful strategy to motivate changes in practice, even when it is unaccompanied by significant shifts in consumer choices.[18] It would be natural to consider public reporting to be a softer strategy than fines, suspension, or firing; however, our results indicate that many clinicians do not see it that way. Alternatively, the results could also suggest that clinicians prefer measures that provide more immediate feedback than public reporting generally provides. These attitudes should be considered when enacting public reporting strategies.

Another interesting finding was that patients and attending physicians tended to track together regarding their attitudes toward penalties for safety lapses. Although patients had a lower threshold for favoring penalties than attendings, similar proportions of patients and attending physicians believed that penalties should be enacted for safety transgressions, and both groups were more penal than physician trainees and nurses. We speculate that attendings and patients may have the most skin in the game, patients as the ones directly harmed by a preventable adverse event, and attending physicians as the most responsible clinicians, at least in the eyes of the malpractice system, licensing boards, and credentials committees.

Even though our study illustrates relatively high levels of endorsement for aggressive measures to deal with clinicians who fail to follow evidence‐based safety practices, a shift in this direction has risks and benefits. The no‐blame paradigm in patient safety grew out of a need to encourage open discussion about medical mistakes.[2] Whereas shifting away from a purely no‐ blame approach may lead to greater adherence with safety practices, and one hopes fewer cases of preventable harm, it also risks stifling the open discussions about medical errors that characterize learning organizations.[13, 19] Because of this, a movement in this direction should be undertaken carefully, starting first with a small number of well‐established safety practices, and ensuring that robust education and system improvements precede and accompany the imposition of penalties for nonadherence.

Our study has limitations. The survey was developed using convenience samples of UCSF faculty and medical students, so broader inclusion of physicians, nurses, trainees, and patients may have yielded a different survey instrument. As a survey, we cannot be certain that any of the groups' responses in real life (eg, in a vote of the medical staff on a given policy) would mirror their survey response. Additionally, the responses to protocol lapses did not include all possible administrative responses, such as mandatory training/remediation or rewards for positive behaviors. The responses could have also been different if participants were presented with different patient safety scenarios. The study population was limited in several ways. Attending and resident physicians were drawn from an academic department of internal medicine; it is possible that other specialties would have different attitudes. Patients were relatively young (likely due to the inclusion criteria), as were attending physicians (due to oversampling of hospitalist physicians). The relatively small number of participants could also limit statistical power to detect differences among groups. Additionally, the study population was limited to patients and healthcare professionals in academic medical centers in San Francisco. It is possible that attitudes would be different in other regions and practice settings.

The no‐blame approach to patient safety has been crucial in refocusing the lens on systems failures and in encouraging the active engagement by clinicians, particularly physicians.[2, 3] On the other hand, there are legitimate concerns that a unidimensional no‐blame approach has permitted, perhaps even promoted, nonadherence to evidence‐based safety practices that could prevent many cases of harm. Although it may not be surprising that patients favor harsher consequences for providers who do not follow basic safety protocols, our study demonstrates relatively widespread support for such consequences even among clinicians and trainees. However, all groups appear to recognize the nuances underlying this set of issues, with varying levels of enthusiasm for punitive responses based on perceived risk and number of transgressions. Future studies are needed to investigate how best to implement public reporting and penalties in ways that can maximize the patient safety benefits.

In their study of against medical advice (AMA) discharges, Edwards et al.[1] express surprise that prescriptions were given and follow‐up arranged at a much lower rate than the frequency of warning of impending AMA discharge. The authors assume that when doctors know a patient wants to leave AMA, they will and should, as a matter of course, write prescriptions and arrange follow‐up. Not considered is the possibility that doctors may decide for selected patients that the better response is not to prescribe and not to arrange follow‐up. Prescribing medications to a patient who has already shown disinterest in heeding doctors' advice may be considered dangerous. Similarly, making an appointment for a patient who has already demonstrated a lack of adherence, thereby depriving another patient of that appointment, may be considered an imprudent use of resources. The authors do not provide data on how many AMA discharges may have been averted by this approach. Attempts to minimize the negative impact of capable patients' decisions neglect that some patients do not categorically prioritize health, and that true autonomy entails not just decision making but bearing responsibility for those decisions' consequences. Medical risk reduction is not the only value at play in these complex situations.

In their study of against medical advice (AMA) discharges, Edwards et al.[1] express surprise that prescriptions were given and follow‐up arranged at a much lower rate than the frequency of warning of impending AMA discharge. The authors assume that when doctors know a patient wants to leave AMA, they will and should, as a matter of course, write prescriptions and arrange follow‐up. Not considered is the possibility that doctors may decide for selected patients that the better response is not to prescribe and not to arrange follow‐up. Prescribing medications to a patient who has already shown disinterest in heeding doctors' advice may be considered dangerous. Similarly, making an appointment for a patient who has already demonstrated a lack of adherence, thereby depriving another patient of that appointment, may be considered an imprudent use of resources. The authors do not provide data on how many AMA discharges may have been averted by this approach. Attempts to minimize the negative impact of capable patients' decisions neglect that some patients do not categorically prioritize health, and that true autonomy entails not just decision making but bearing responsibility for those decisions' consequences. Medical risk reduction is not the only value at play in these complex situations.

In their study of against medical advice (AMA) discharges, Edwards et al.[1] express surprise that prescriptions were given and follow‐up arranged at a much lower rate than the frequency of warning of impending AMA discharge. The authors assume that when doctors know a patient wants to leave AMA, they will and should, as a matter of course, write prescriptions and arrange follow‐up. Not considered is the possibility that doctors may decide for selected patients that the better response is not to prescribe and not to arrange follow‐up. Prescribing medications to a patient who has already shown disinterest in heeding doctors' advice may be considered dangerous. Similarly, making an appointment for a patient who has already demonstrated a lack of adherence, thereby depriving another patient of that appointment, may be considered an imprudent use of resources. The authors do not provide data on how many AMA discharges may have been averted by this approach. Attempts to minimize the negative impact of capable patients' decisions neglect that some patients do not categorically prioritize health, and that true autonomy entails not just decision making but bearing responsibility for those decisions' consequences. Medical risk reduction is not the only value at play in these complex situations.

We believe medications can safely be prescribed to most patients who leave against medical advice (AMA), and that follow‐up should be offered to most if not all such patients. Why should we do this? Consider a wheezing asthma patient who leaves AMA. She or he is probably more likely to return to the emergency department (somewhere) or be readmitted (somewhere) and cost more money (to the system) than if given an inhaler and steroid taper.

Dr. Querques et al. suggest that doctors should potentially not prescribe and should not offer follow‐up to certain patients who want to leave AMA, particularly those who show disinterest in heeding the doctor's advice and have already demonstrated a lack of adherence. How should doctors make those judgments? Patients leave AMA for a variety of reasons: for example to avoid cost, because they feel better, or poor communication. Certainly, not all patients who want to leave AMA are categorically nonadherent. Conversely, up to 50% of all continuity patients are not fully adherent to the lifestyle changes and medications their physicians prescribe,[1] yet they would rarely if ever threaten AMA. Is withholding treatments that are likely to be effective and have minimal risk worth the potential benefit of increasing a patient's priority on their own healthcare? As emphasized by Berger (2008),[2] interventions with low risk and high potential for efficacy (assistance with establishing a follow‐up) should be pursued, and those with potential risks (starting new long‐term medications) should be avoided. At minimum, considering these options is an ethical requirement in the care of patients. We maintain that this reasoning should be explained and documented, which often is not being done in healthcare today.

How many AMAs are avoided by truly collaborative relationships with patients (nonevents), and how many are fueled by a more paternalistic relationship? For example, if a patient truly has a sick family member or child to take care of or has financial problems or no insurance, then it seems reasonable, perhaps even responsible, to leave the hospital even if maximal benefits of care have not been reached. In a collaborative relationship, providers may then tailor treatment to the patient's circumstances, even if this means the patient is not getting the best possible care.

We believe medications can safely be prescribed to most patients who leave against medical advice (AMA), and that follow‐up should be offered to most if not all such patients. Why should we do this? Consider a wheezing asthma patient who leaves AMA. She or he is probably more likely to return to the emergency department (somewhere) or be readmitted (somewhere) and cost more money (to the system) than if given an inhaler and steroid taper.

Dr. Querques et al. suggest that doctors should potentially not prescribe and should not offer follow‐up to certain patients who want to leave AMA, particularly those who show disinterest in heeding the doctor's advice and have already demonstrated a lack of adherence. How should doctors make those judgments? Patients leave AMA for a variety of reasons: for example to avoid cost, because they feel better, or poor communication. Certainly, not all patients who want to leave AMA are categorically nonadherent. Conversely, up to 50% of all continuity patients are not fully adherent to the lifestyle changes and medications their physicians prescribe,[1] yet they would rarely if ever threaten AMA. Is withholding treatments that are likely to be effective and have minimal risk worth the potential benefit of increasing a patient's priority on their own healthcare? As emphasized by Berger (2008),[2] interventions with low risk and high potential for efficacy (assistance with establishing a follow‐up) should be pursued, and those with potential risks (starting new long‐term medications) should be avoided. At minimum, considering these options is an ethical requirement in the care of patients. We maintain that this reasoning should be explained and documented, which often is not being done in healthcare today.

How many AMAs are avoided by truly collaborative relationships with patients (nonevents), and how many are fueled by a more paternalistic relationship? For example, if a patient truly has a sick family member or child to take care of or has financial problems or no insurance, then it seems reasonable, perhaps even responsible, to leave the hospital even if maximal benefits of care have not been reached. In a collaborative relationship, providers may then tailor treatment to the patient's circumstances, even if this means the patient is not getting the best possible care.

We believe medications can safely be prescribed to most patients who leave against medical advice (AMA), and that follow‐up should be offered to most if not all such patients. Why should we do this? Consider a wheezing asthma patient who leaves AMA. She or he is probably more likely to return to the emergency department (somewhere) or be readmitted (somewhere) and cost more money (to the system) than if given an inhaler and steroid taper.

Dr. Querques et al. suggest that doctors should potentially not prescribe and should not offer follow‐up to certain patients who want to leave AMA, particularly those who show disinterest in heeding the doctor's advice and have already demonstrated a lack of adherence. How should doctors make those judgments? Patients leave AMA for a variety of reasons: for example to avoid cost, because they feel better, or poor communication. Certainly, not all patients who want to leave AMA are categorically nonadherent. Conversely, up to 50% of all continuity patients are not fully adherent to the lifestyle changes and medications their physicians prescribe,[1] yet they would rarely if ever threaten AMA. Is withholding treatments that are likely to be effective and have minimal risk worth the potential benefit of increasing a patient's priority on their own healthcare? As emphasized by Berger (2008),[2] interventions with low risk and high potential for efficacy (assistance with establishing a follow‐up) should be pursued, and those with potential risks (starting new long‐term medications) should be avoided. At minimum, considering these options is an ethical requirement in the care of patients. We maintain that this reasoning should be explained and documented, which often is not being done in healthcare today.

How many AMAs are avoided by truly collaborative relationships with patients (nonevents), and how many are fueled by a more paternalistic relationship? For example, if a patient truly has a sick family member or child to take care of or has financial problems or no insurance, then it seems reasonable, perhaps even responsible, to leave the hospital even if maximal benefits of care have not been reached. In a collaborative relationship, providers may then tailor treatment to the patient's circumstances, even if this means the patient is not getting the best possible care.