A 49‐year‐old retired air‐force officer presented with productive cough lasting 1 week. He was in good health except for back pain after a motor vehicle accident, 2 years ago, for which he took unspecified pain medications. His condition rapidly deteriorated necessitating mechanical ventilation. Chest x‐ray showed multilobar pneumonia (Figure 1). Blood and sputum cultures grew Serratia marcescens. Chest computed tomography (CT)‐scan showed multilobar involvement with debris in his airways that was thought to be respiratory secretions (Figure 2, arrows). He remained intubated for 2 weeks before his clinical status improved enough to permit extubation. Immediately following extubation, he coughed up a ball of crumpled, plastic material. His condition subsequently improved dramatically with complete radiological resolution of his pneumonia. On analyzing the foreign body, it turned out to be a fenestrated fentanyl patch (Figure 3). The patient then reported that he often cut‐up used fentanyl patches and sucked on them for an extra high. In retrospect, the endobronchial debris was actually the patch straddling the carina and had prevented rapid recovery. Aspiration of unusual foreign bodies have been reported in the literature. This foreign body masqueraded as respiratory secretions on imaging preventing early detection. This clinical encounter brings to light yet another innovative, but potentially deadly, form of drug abuse. It also emphasizes the importance of early bronchoscopy in nonresolving pneumonias.

Figure 1

Figure 2

Figure 3

A 49‐year‐old retired air‐force officer presented with productive cough lasting 1 week. He was in good health except for back pain after a motor vehicle accident, 2 years ago, for which he took unspecified pain medications. His condition rapidly deteriorated necessitating mechanical ventilation. Chest x‐ray showed multilobar pneumonia (Figure 1). Blood and sputum cultures grew Serratia marcescens. Chest computed tomography (CT)‐scan showed multilobar involvement with debris in his airways that was thought to be respiratory secretions (Figure 2, arrows). He remained intubated for 2 weeks before his clinical status improved enough to permit extubation. Immediately following extubation, he coughed up a ball of crumpled, plastic material. His condition subsequently improved dramatically with complete radiological resolution of his pneumonia. On analyzing the foreign body, it turned out to be a fenestrated fentanyl patch (Figure 3). The patient then reported that he often cut‐up used fentanyl patches and sucked on them for an extra high. In retrospect, the endobronchial debris was actually the patch straddling the carina and had prevented rapid recovery. Aspiration of unusual foreign bodies have been reported in the literature. This foreign body masqueraded as respiratory secretions on imaging preventing early detection. This clinical encounter brings to light yet another innovative, but potentially deadly, form of drug abuse. It also emphasizes the importance of early bronchoscopy in nonresolving pneumonias.

Figure 1

Figure 2

Figure 3

A 49‐year‐old retired air‐force officer presented with productive cough lasting 1 week. He was in good health except for back pain after a motor vehicle accident, 2 years ago, for which he took unspecified pain medications. His condition rapidly deteriorated necessitating mechanical ventilation. Chest x‐ray showed multilobar pneumonia (Figure 1). Blood and sputum cultures grew Serratia marcescens. Chest computed tomography (CT)‐scan showed multilobar involvement with debris in his airways that was thought to be respiratory secretions (Figure 2, arrows). He remained intubated for 2 weeks before his clinical status improved enough to permit extubation. Immediately following extubation, he coughed up a ball of crumpled, plastic material. His condition subsequently improved dramatically with complete radiological resolution of his pneumonia. On analyzing the foreign body, it turned out to be a fenestrated fentanyl patch (Figure 3). The patient then reported that he often cut‐up used fentanyl patches and sucked on them for an extra high. In retrospect, the endobronchial debris was actually the patch straddling the carina and had prevented rapid recovery. Aspiration of unusual foreign bodies have been reported in the literature. This foreign body masqueraded as respiratory secretions on imaging preventing early detection. This clinical encounter brings to light yet another innovative, but potentially deadly, form of drug abuse. It also emphasizes the importance of early bronchoscopy in nonresolving pneumonias.

Figure 1

Figure 2

Figure 3

Forty‐five‐million Americans speak a language other than English and more than 19 million of these speak English less than very wellor are limited English proficient (LEP).1 The number of non‐English‐speaking and LEP people in the US has risen in recent decades, presenting a challenge to healthcare systems to provide high‐quality, patient‐centered care for these patients.2

For outpatients, language barriers are a fundamental contributor to gaps in health care. In the clinic setting, patients who do not speak English well have less access to a usual source of care and lower rates of physician visits and preventive services.36 Even when patients with language barriers do have access to care, they have poorer adherence, decreased comprehension of their diagnoses, decreased satisfaction with care, and increased medication complications.710

Few studies, however, have examined how language influences outcomes of hospital care. Compared to English‐speakers, patients who do not speak English well may experience longer lengths of stay,11 and have more adverse events while in the hospital.12 However, these previous studies have not investigated outcomes immediately post‐hospitalization, such as readmission rates and mortality, nor have they directly addressed the interaction between ethnicity and language.

To understand these questions, we analyzed data collected from a university‐based teaching hospital which cares for patients of diverse cultural and language backgrounds. Using these data, we examined how patients' primary language influenced hospital costs, length of stay (LOS), 30‐day readmission, and 30‐day mortality risk.

Patients and Methods

Results

Conclusion/Discussion

Our results indicate that language barriers may contribute to higher readmission rates for non‐English speakers, but that they have less impact on care efficiency or mortality. This finding of an association between language and readmission, without a similar association with efficiency, suggests a potentially communication‐critical step in care.20, 21 Patients with language barriers are more likely to experience adverse events, and those events are often caused by errors in communication.12 It is conceivable that higher readmission risk for Chinese and Spanish speakers in our study was, at least in part, due to gaps in communication that are present in all patient groups, and exacerbated by the presence of a language barrier. This barrier is likely present during hospitalization but magnified at discharge, limiting caregivers' ability to understand patients' needs for home care, while simultaneously limiting patients' understanding of the discharge plan. After discharge, it is also possible that non‐English speakers are less able to communicate their needs as they arise, ormore subtlyfeel less supported by a primarily English speaking healthcare system. As in other clinical arenas,22 it is quite possible that increased access to professional interpreters in the hospital setting, and particularly at the time of discharge, would enhance communication and outcomes for LEP patients. Our interpreter services data showed that the patients in our study had quite limited access to staff professional interpreters.

Our findings differ somewhat from those of John‐Baptiste et al.,11 who found that language barriers contributed to increased LOS for patients with cardiac and major surgical diagnoses. Our study's findings are akin to recent research suggesting that being a monolingual Spanish speaker or receiving interpreter services may not significantly impact LOS or cost of hospitalization,23 and that LOS and in‐hospital mortality do not differ for non‐English speakers and English speakers after acute myocardial infarction.24 These studies, along with our results, suggest that that care efficiency in the hospital may be driven much more by clinical acuity (eg, the need to respond rapidly to urgent clinical signs such as hypotension, fever and respiratory distress) than by adequacy of communication. For example, elderly LEP patients may be even more likely than English speakers to have vigilant family members at the bedside throughout their hospitalization due to their need for communication assistance; these family members can quickly alert hospital staff to concerning changes in the patient's condition.

Our results also suggest the possibility that language and ethnicity are not monolithic concepts, and that even within language and ethnic groups there are potential differences in care pattern. For example, not speaking English may be a surrogate marker for unmeasured factors such as social supports and access to care. Language is intimately associated with culture; it remains plausible that cultural differences between highly acculturated and less acculturated members of a given ethno‐cultural group may have contributed to our observed differences in readmission rates. Differences in culture and associated factors, such as social support or use of multiple hospital systems, may account for lack of higher readmission risk in Russian speakers, while Chinese and Spanish speakers had higher readmission risk.

In addition, our finding that English‐speaking Latinos had lower readmission risk than any other group may be more consistent with their clinical characteristicseg, younger age, fewer comorbiditiesthan with cultural factors. Our finding that African American patients had the highest readmission risk in our hospital was both surprising and concerning. Some of this increased risk may be explained by clinical characteristics, such as higher comorbidities and higher rates of diagnoses leading to frequent admissions (eg, sickle cell disease); however, the reasons for this disparity deserves further investigation.

Our study has limitations. First, our data are administrative, and lack information about patients' educational attainment, social support, acculturation, utilization of other hospital systems, and usual source of care. Despite this, we were able to account for many significant covariates that might contribute to readmission rates, including age, insurance status, gender, comorbidities, and admission to the intensive care unit.2528 Second, our information about patients' English language proficiency is limited. While direct assessments of English proficiency are more accurate ways to determine a patient's ability to communicate with health care providers in English,29 our language validation work conducted in preparation for this study suggests that most of our patients recorded as having a non‐English primary language (87%) also have a low score on a language acculturation scale.

Third, only 14% of our non‐English speaking subjects utilized professional staff interpreters, and we had no information on the use of professional telephonic interpreters, or ad hoc interpretersfamily members, non‐interpreter staff membersand their impact on our results. It is well‐documented that ad hoc interpreters are used frequently in healthcare, particularly in the hospital setting, and thus we can assume this to be true in our study.30, 31 As noted above, it is likely that the advocacy of family members and friends at the bedside helped to minimize potential differences in care efficiency for patients with language barriers. Finally, our study was performed at a single university based hospital and may not produce results which are applicable to other care settings.

Our findings point to several avenues for future research on language barriers and hospitalized patients. First, the field would benefit from an examination of the impact of easy access to professional interpreters during hospitalization on outcomes of hospital care, in particular on readmission rates. Second, there is need for development and assessment of best practices for creating a culture of professional interpreter utilization in the hospital among physicians and nursing staff. Third, investigation of the role of caregiver presence in the hospital room and how this might differ by patient culture, age and language ability may further elucidate some of the differences across language groups observed in our study. Lastly, a more granular investigation of clinician‐patient communication and the importance of interpersonal processes of care on both patient satisfaction and understanding of and adherence to discharge instructions could lead to the development of detailed interventions to enhance this communication and these outcomes as it has done for communication‐sensitive outcomes in the outpatient arena.3234

In summary, our study suggests that higher risk for readmission can be added to the unfortunate list of outcomes which are worsened due to language barriers, pointing to transition from the hospital as a potentially communication‐critical step in care which may be amenable to intervention. Our findings also suggest that this risk can vary even between groups of patients who do not speak English primarily. Whether and to what degree language and communication barriers aloneincluding access to professional interpreters and patient‐centered communicationduring hospitalization, or differences in caregiver social support both during and after hospitalization as well as access to care post‐hospitalization contribute to these findings is a worthy subject of future research.

Forty‐five‐million Americans speak a language other than English and more than 19 million of these speak English less than very wellor are limited English proficient (LEP).1 The number of non‐English‐speaking and LEP people in the US has risen in recent decades, presenting a challenge to healthcare systems to provide high‐quality, patient‐centered care for these patients.2

For outpatients, language barriers are a fundamental contributor to gaps in health care. In the clinic setting, patients who do not speak English well have less access to a usual source of care and lower rates of physician visits and preventive services.36 Even when patients with language barriers do have access to care, they have poorer adherence, decreased comprehension of their diagnoses, decreased satisfaction with care, and increased medication complications.710

Few studies, however, have examined how language influences outcomes of hospital care. Compared to English‐speakers, patients who do not speak English well may experience longer lengths of stay,11 and have more adverse events while in the hospital.12 However, these previous studies have not investigated outcomes immediately post‐hospitalization, such as readmission rates and mortality, nor have they directly addressed the interaction between ethnicity and language.

To understand these questions, we analyzed data collected from a university‐based teaching hospital which cares for patients of diverse cultural and language backgrounds. Using these data, we examined how patients' primary language influenced hospital costs, length of stay (LOS), 30‐day readmission, and 30‐day mortality risk.

Patients and Methods

Results

Conclusion/Discussion

Our results indicate that language barriers may contribute to higher readmission rates for non‐English speakers, but that they have less impact on care efficiency or mortality. This finding of an association between language and readmission, without a similar association with efficiency, suggests a potentially communication‐critical step in care.20, 21 Patients with language barriers are more likely to experience adverse events, and those events are often caused by errors in communication.12 It is conceivable that higher readmission risk for Chinese and Spanish speakers in our study was, at least in part, due to gaps in communication that are present in all patient groups, and exacerbated by the presence of a language barrier. This barrier is likely present during hospitalization but magnified at discharge, limiting caregivers' ability to understand patients' needs for home care, while simultaneously limiting patients' understanding of the discharge plan. After discharge, it is also possible that non‐English speakers are less able to communicate their needs as they arise, ormore subtlyfeel less supported by a primarily English speaking healthcare system. As in other clinical arenas,22 it is quite possible that increased access to professional interpreters in the hospital setting, and particularly at the time of discharge, would enhance communication and outcomes for LEP patients. Our interpreter services data showed that the patients in our study had quite limited access to staff professional interpreters.

Our findings differ somewhat from those of John‐Baptiste et al.,11 who found that language barriers contributed to increased LOS for patients with cardiac and major surgical diagnoses. Our study's findings are akin to recent research suggesting that being a monolingual Spanish speaker or receiving interpreter services may not significantly impact LOS or cost of hospitalization,23 and that LOS and in‐hospital mortality do not differ for non‐English speakers and English speakers after acute myocardial infarction.24 These studies, along with our results, suggest that that care efficiency in the hospital may be driven much more by clinical acuity (eg, the need to respond rapidly to urgent clinical signs such as hypotension, fever and respiratory distress) than by adequacy of communication. For example, elderly LEP patients may be even more likely than English speakers to have vigilant family members at the bedside throughout their hospitalization due to their need for communication assistance; these family members can quickly alert hospital staff to concerning changes in the patient's condition.

Our results also suggest the possibility that language and ethnicity are not monolithic concepts, and that even within language and ethnic groups there are potential differences in care pattern. For example, not speaking English may be a surrogate marker for unmeasured factors such as social supports and access to care. Language is intimately associated with culture; it remains plausible that cultural differences between highly acculturated and less acculturated members of a given ethno‐cultural group may have contributed to our observed differences in readmission rates. Differences in culture and associated factors, such as social support or use of multiple hospital systems, may account for lack of higher readmission risk in Russian speakers, while Chinese and Spanish speakers had higher readmission risk.

In addition, our finding that English‐speaking Latinos had lower readmission risk than any other group may be more consistent with their clinical characteristicseg, younger age, fewer comorbiditiesthan with cultural factors. Our finding that African American patients had the highest readmission risk in our hospital was both surprising and concerning. Some of this increased risk may be explained by clinical characteristics, such as higher comorbidities and higher rates of diagnoses leading to frequent admissions (eg, sickle cell disease); however, the reasons for this disparity deserves further investigation.

Our study has limitations. First, our data are administrative, and lack information about patients' educational attainment, social support, acculturation, utilization of other hospital systems, and usual source of care. Despite this, we were able to account for many significant covariates that might contribute to readmission rates, including age, insurance status, gender, comorbidities, and admission to the intensive care unit.2528 Second, our information about patients' English language proficiency is limited. While direct assessments of English proficiency are more accurate ways to determine a patient's ability to communicate with health care providers in English,29 our language validation work conducted in preparation for this study suggests that most of our patients recorded as having a non‐English primary language (87%) also have a low score on a language acculturation scale.

Third, only 14% of our non‐English speaking subjects utilized professional staff interpreters, and we had no information on the use of professional telephonic interpreters, or ad hoc interpretersfamily members, non‐interpreter staff membersand their impact on our results. It is well‐documented that ad hoc interpreters are used frequently in healthcare, particularly in the hospital setting, and thus we can assume this to be true in our study.30, 31 As noted above, it is likely that the advocacy of family members and friends at the bedside helped to minimize potential differences in care efficiency for patients with language barriers. Finally, our study was performed at a single university based hospital and may not produce results which are applicable to other care settings.

Our findings point to several avenues for future research on language barriers and hospitalized patients. First, the field would benefit from an examination of the impact of easy access to professional interpreters during hospitalization on outcomes of hospital care, in particular on readmission rates. Second, there is need for development and assessment of best practices for creating a culture of professional interpreter utilization in the hospital among physicians and nursing staff. Third, investigation of the role of caregiver presence in the hospital room and how this might differ by patient culture, age and language ability may further elucidate some of the differences across language groups observed in our study. Lastly, a more granular investigation of clinician‐patient communication and the importance of interpersonal processes of care on both patient satisfaction and understanding of and adherence to discharge instructions could lead to the development of detailed interventions to enhance this communication and these outcomes as it has done for communication‐sensitive outcomes in the outpatient arena.3234

In summary, our study suggests that higher risk for readmission can be added to the unfortunate list of outcomes which are worsened due to language barriers, pointing to transition from the hospital as a potentially communication‐critical step in care which may be amenable to intervention. Our findings also suggest that this risk can vary even between groups of patients who do not speak English primarily. Whether and to what degree language and communication barriers aloneincluding access to professional interpreters and patient‐centered communicationduring hospitalization, or differences in caregiver social support both during and after hospitalization as well as access to care post‐hospitalization contribute to these findings is a worthy subject of future research.

Forty‐five‐million Americans speak a language other than English and more than 19 million of these speak English less than very wellor are limited English proficient (LEP).1 The number of non‐English‐speaking and LEP people in the US has risen in recent decades, presenting a challenge to healthcare systems to provide high‐quality, patient‐centered care for these patients.2

For outpatients, language barriers are a fundamental contributor to gaps in health care. In the clinic setting, patients who do not speak English well have less access to a usual source of care and lower rates of physician visits and preventive services.36 Even when patients with language barriers do have access to care, they have poorer adherence, decreased comprehension of their diagnoses, decreased satisfaction with care, and increased medication complications.710

Few studies, however, have examined how language influences outcomes of hospital care. Compared to English‐speakers, patients who do not speak English well may experience longer lengths of stay,11 and have more adverse events while in the hospital.12 However, these previous studies have not investigated outcomes immediately post‐hospitalization, such as readmission rates and mortality, nor have they directly addressed the interaction between ethnicity and language.

To understand these questions, we analyzed data collected from a university‐based teaching hospital which cares for patients of diverse cultural and language backgrounds. Using these data, we examined how patients' primary language influenced hospital costs, length of stay (LOS), 30‐day readmission, and 30‐day mortality risk.

Patients and Methods

Results

Conclusion/Discussion

Our results indicate that language barriers may contribute to higher readmission rates for non‐English speakers, but that they have less impact on care efficiency or mortality. This finding of an association between language and readmission, without a similar association with efficiency, suggests a potentially communication‐critical step in care.20, 21 Patients with language barriers are more likely to experience adverse events, and those events are often caused by errors in communication.12 It is conceivable that higher readmission risk for Chinese and Spanish speakers in our study was, at least in part, due to gaps in communication that are present in all patient groups, and exacerbated by the presence of a language barrier. This barrier is likely present during hospitalization but magnified at discharge, limiting caregivers' ability to understand patients' needs for home care, while simultaneously limiting patients' understanding of the discharge plan. After discharge, it is also possible that non‐English speakers are less able to communicate their needs as they arise, ormore subtlyfeel less supported by a primarily English speaking healthcare system. As in other clinical arenas,22 it is quite possible that increased access to professional interpreters in the hospital setting, and particularly at the time of discharge, would enhance communication and outcomes for LEP patients. Our interpreter services data showed that the patients in our study had quite limited access to staff professional interpreters.

Our findings differ somewhat from those of John‐Baptiste et al.,11 who found that language barriers contributed to increased LOS for patients with cardiac and major surgical diagnoses. Our study's findings are akin to recent research suggesting that being a monolingual Spanish speaker or receiving interpreter services may not significantly impact LOS or cost of hospitalization,23 and that LOS and in‐hospital mortality do not differ for non‐English speakers and English speakers after acute myocardial infarction.24 These studies, along with our results, suggest that that care efficiency in the hospital may be driven much more by clinical acuity (eg, the need to respond rapidly to urgent clinical signs such as hypotension, fever and respiratory distress) than by adequacy of communication. For example, elderly LEP patients may be even more likely than English speakers to have vigilant family members at the bedside throughout their hospitalization due to their need for communication assistance; these family members can quickly alert hospital staff to concerning changes in the patient's condition.

Our results also suggest the possibility that language and ethnicity are not monolithic concepts, and that even within language and ethnic groups there are potential differences in care pattern. For example, not speaking English may be a surrogate marker for unmeasured factors such as social supports and access to care. Language is intimately associated with culture; it remains plausible that cultural differences between highly acculturated and less acculturated members of a given ethno‐cultural group may have contributed to our observed differences in readmission rates. Differences in culture and associated factors, such as social support or use of multiple hospital systems, may account for lack of higher readmission risk in Russian speakers, while Chinese and Spanish speakers had higher readmission risk.

In addition, our finding that English‐speaking Latinos had lower readmission risk than any other group may be more consistent with their clinical characteristicseg, younger age, fewer comorbiditiesthan with cultural factors. Our finding that African American patients had the highest readmission risk in our hospital was both surprising and concerning. Some of this increased risk may be explained by clinical characteristics, such as higher comorbidities and higher rates of diagnoses leading to frequent admissions (eg, sickle cell disease); however, the reasons for this disparity deserves further investigation.

Our study has limitations. First, our data are administrative, and lack information about patients' educational attainment, social support, acculturation, utilization of other hospital systems, and usual source of care. Despite this, we were able to account for many significant covariates that might contribute to readmission rates, including age, insurance status, gender, comorbidities, and admission to the intensive care unit.2528 Second, our information about patients' English language proficiency is limited. While direct assessments of English proficiency are more accurate ways to determine a patient's ability to communicate with health care providers in English,29 our language validation work conducted in preparation for this study suggests that most of our patients recorded as having a non‐English primary language (87%) also have a low score on a language acculturation scale.

Third, only 14% of our non‐English speaking subjects utilized professional staff interpreters, and we had no information on the use of professional telephonic interpreters, or ad hoc interpretersfamily members, non‐interpreter staff membersand their impact on our results. It is well‐documented that ad hoc interpreters are used frequently in healthcare, particularly in the hospital setting, and thus we can assume this to be true in our study.30, 31 As noted above, it is likely that the advocacy of family members and friends at the bedside helped to minimize potential differences in care efficiency for patients with language barriers. Finally, our study was performed at a single university based hospital and may not produce results which are applicable to other care settings.

Our findings point to several avenues for future research on language barriers and hospitalized patients. First, the field would benefit from an examination of the impact of easy access to professional interpreters during hospitalization on outcomes of hospital care, in particular on readmission rates. Second, there is need for development and assessment of best practices for creating a culture of professional interpreter utilization in the hospital among physicians and nursing staff. Third, investigation of the role of caregiver presence in the hospital room and how this might differ by patient culture, age and language ability may further elucidate some of the differences across language groups observed in our study. Lastly, a more granular investigation of clinician‐patient communication and the importance of interpersonal processes of care on both patient satisfaction and understanding of and adherence to discharge instructions could lead to the development of detailed interventions to enhance this communication and these outcomes as it has done for communication‐sensitive outcomes in the outpatient arena.3234

In summary, our study suggests that higher risk for readmission can be added to the unfortunate list of outcomes which are worsened due to language barriers, pointing to transition from the hospital as a potentially communication‐critical step in care which may be amenable to intervention. Our findings also suggest that this risk can vary even between groups of patients who do not speak English primarily. Whether and to what degree language and communication barriers aloneincluding access to professional interpreters and patient‐centered communicationduring hospitalization, or differences in caregiver social support both during and after hospitalization as well as access to care post‐hospitalization contribute to these findings is a worthy subject of future research.

Studies of the red blood cell mass performed in hospitalized patients were first done in the early 1970s.1 Thereafter, a decrease in the hemoglobin concentration ([Hb]) and its potential causes have been further reported, especially in critically ill patients.25 The decrease in [Hb] can occur as a result of blood draws for diagnostic testing; blood loss associated with invasive procedures and bleeding; occult gastrointestinal bleeding; hemolysis; shortening of red cell survival; iron, folic acid or cobalamine deficiencies; renal, liver or endocrine failure; hemodilution associated with fluid therapy; and the so‐called anemia of inflammation (AI).68 The latter would be a consequence of a blunted response of the bone marrow due to several factors such as inadequate secretion of erythropoietin, inhibition of the proliferation and differentiation of the erythroid precursors of the bone marrow and an hepcidin‐mediated functional iron deficiency.810

As mentioned, most studies were done in critically ill patients and scarce information is available about general ward admitted patients (GWAP) with less severe illnesses. In this scenario, some of the proposed mechanisms may have a less clear role. It is widely accepted that [Hb] may decrease without overt bleeding in GWAP. However, given the lack of information on this matter, laboratory controls and invasive procedures are undertaken to determine its potential causes.

The purpose of the present study is to describe [Hb] variation over time in nonbleeding GWAP, to estimate the proportion of patients with [Hb] decreases 1.5 g/dL, and to evaluate possible related variables.

Materials and Methods

A 16‐week (September 2004‐January 2005) prospective observational study was conducted in Internal Medicine GWAP at 2 Buenos Aires teaching hospitals.

All consecutive patients older than 16 years were evaluated. Patients admitted for the following reasons were excluded: bleeding, trauma, surgery, invasive procedures associated with blood loss (biopsies, biliary drainage, endovascular therapeutic procedures, and chest tubes), blood transfusions, anemia, chemotherapy, and acute renal failure.

Patients with a bleeding history, chemotherapy or radiation therapy within two months prior to admission, patients on dialysis and patients with current oncologic or hematologic disease were excluded, as well as those with length of stay less than 3 days, or with less than 2 [Hb] or hematocrit (HCT) measurements.

Patients were followed until discharge, death or transfer to a critical care unit. Patients were withdrawn from the study if they presented with bleeding or hemolysis, underwent red blood cell transfusion or therapy affecting hemoglobin levels (iron, chemotherapy or erythropoietin) or if either a surgical or invasive procedure associated with blood loss was performed.

Data were collected from patients' medical records and additional information was obtained from treating physicians and patients by the authors (AL, NC, SM, AN, MH). Standardized case report forms were completed during the hospitalization including: age, sex, admissions in the previous 3 months (readmissions) and whether or not the patient lived in a nursing home. Patients were categorized according to discharge diagnosis as reported by Nguyen et al.4 with modifications related to our general ward population. Upon admission Katz daily activity index (ADL),11, 12 APACHE II acute physiology score (APS),13, 14 and Charlson comorbidity score (Charlson)15, 16 were assessed. In all cases the [Hb] and HCT values were registered on admission and on days 3, 6, 10 and prior to discharge as well as any other determination required by the treating physician. Anemia was defined as [Hb] 13 g/dL for men and 12 g/dL for women.17 Based on previous reports a 1.5 g/dL decrease in [Hb] and of 4.5 points in HCT compared to admission values were considered a significant fall.1, 4 Acute renal failure was defined as an increase in creatinine level 0.5 mg/dL or a 50% increase from baseline.18, 19

Every procedure without significant blood loss (PWSBL) such as venous catheter placement, thoracentesis, lumbar punction, paracentesis, skin biopsy, arthrocentesis, and diagnostic angiogram was also recorded. The study was approved by the Hospital's ethical committee.

Data analysis and processing were performed with Excel 2000 and Stata version 8.0 (Stata Corp; USA). For continuous variables, results were expressed by the mean value its standard deviation (SD), and compared with Student's t‐test. The chi‐square test was used to compare categorical variables. A survival curve was developed with the Kaplan‐Meier method to analyze the time to a significant fall in [Hb]. Finally, Cox proportional hazard modeling was performed to assess the association between a significant fall in [Hb] and other variables. We accepted P < 0,05 as significant.

Results

A total of 338 patients were admitted to the Internal Medicine Inpatient Services in the study period. Thirty‐nine percent (131) of these patients were included. Exclusion criteria were: diagnosis at admission (n = 95, 45.9%), past medical history (n = 56, 27%) and length of stay less than 3 days, or less than 2 determinations of [Hb] or HCT (n = 56, 27%). Data collection was discontinued for the following reasons: discharge (81.7%), death (6.9%), surgery or procedures associated with blood loss (5.3%), transfer to a critical care unit (3%), transfusions (2.3%), and chemotherapy (0.8%). The baseline characteristics of the study patients are shown in Table 1.

Diagnoses at discharge were divided into the following categories: infections (25.2%), electrolyte disturbances (7.6%), cardiac diseases (9.9%), neurologic (19.1%), respiratory (16.8%), gastrointestinal (10.7%), and other diagnosis (10.7%).

No evidence of bleeding was found in the included patients. Bleeding was only observed in 4 of the initially evaluated patients who were excluded since they failed to meet the required number of [Hb] determinations before bleeding.

A mean decrease in [Hb] of 0.71 g/dL was found between admission and discharge day (P < 0.001; 95% CI, 0.47‐0.97). Mean nadir [Hb] was 1.45 g/dL lower than admission [Hb] (P < 0.001; 95% CI, 1.24‐1.67). Mean nadir day occurred between hospitalization days 3 and 4. Mean [Hb] at discharge was 11.8 1.8 g/dL. This value is higher than the mean concentration at nadir (0.74 g/dL P < 0.001; 95% CI, 0.60‐0.97) (Figure 1).

Figure 1

Table 2 shows the rate of patients with a decrease in the [Hb] for different cutoff levels. Forty‐five percent of the study population (59 patients) had a significant fall in [Hb] during hospitalization. This was observed on day 2 in 50% of the patients. Likewise, a significant fall in HCT value was found in 42.7% of patients. If [Hb] decrease during hospitalization is analyzed as a proportion of [Hb] at admission, 52.7% of patients had a 10% or greater [Hb] decrease during their hospital stay.

Using Kaplan‐Meier analysis, the estimated proportion showing no significant fall of [Hb] was 0.63 (95% CI, 0.55‐0.71) on day 3 and 0.48 (95% CI, 0.29‐0.67) on day 10. By day 15, only 3.8% of the initially included patients were still hospitalized and showed no decrease in [Hb] 1.5 g/dl. These patients maintained their [Hb] stable for the rest of the follow up period (Figure 2).

Figure 2

In the univariate analysis, comparing patients that experienced a decrease in [Hb] 1.5 g/dL with those who did not, significant differences were only found in the following variables: length of stay, APS, anemia at admission, [Hb] at admission, and infectious, gastrointestinal or cardiac diseases at discharge diagnosis (Table 3).

In the Cox proportional hazard model adjusting for the variables shown in Table 4, a significant independent direct association was found between a significant fall in [Hb] during hospital stay and APS, [Hb] at admission, and either diagnosis of infectious or gastrointestinal disease at discharge. Similar results were found if a significant fall was redefined as a 12% decrease in [Hb] or as a 4.5 point decrease in HCT from baseline.

Discussion

This study describes the variation of [Hb] and HCT values in GWAP without bleeding or other obvious medical conditions associated with a decrease in the red blood cell mass. As previously described,4, 20, 21 we found that [Hb] decreases during hospital stay are frequently observed. The mean [Hb] fall from admission was 1.45 g/dL, and was mainly recorded between hospitalization days 3 and 4. In approximately half of the study population, a 1.5 g/dL decrease in [Hb] was observed. In the survival analysis, 40% and 55% of patients are expected to present such a fall on day 4 and 12 respectively. In accordance with prior reports4, 21 a greater decrease in [Hb] occurred in the first days of hospitalization, and a high proportion of patients were already anemic at the time of admission.

The following variables were associated with a decrease in the [Hb] during hospitalization: higher APS score, higher [Hb] at admission, and diagnosis of infectious or gastrointestinal disease at discharge. Even though several mechanisms seem to contribute to the decrease in [Hb] during hospitalization our data suggest that 1 of these factors seems to be the severity of the disease, as previously proposed by Nguyen et al.4 This observation is supported by the association found in our study between [Hb] decrease and APS. No association was found with Charlson, ADL, and being a nursing home resident. These variables, which have not been previously analyzed, seem to indicate that patients with chronic illnesses are not more likely to have decreases in [Hb] during hospitalization.

Patients with higher [Hb] at admission had a greater [Hb] decrease during their hospital stay. This finding could suggest that the mechanism related to this decrease had less effect on patients with lower [Hb]. This is similar to that observed in the anemia of chronic diseases where [Hb] does not usually fall to extreme values. Our analysis reveals a greater decrease during the first days of hospitalization. Given the high rate of patients anemic at admission, it is possible that the decrease in [Hb] had begun prior to admission.

Previous papers describe a low prevalence of cyanocobalamine (2%), folic acid (2%), and iron (9%) deficiency in patients admitted to critical care units.3 In these patients an inadequate bone marrow response has been proposed as a mechanism for [Hb] decrease, a phenomenon that has been called AI or anemia of critical illness.8, 2228 Inflammatory response associated with acute disease causes this hypoproliferative anemia through 3 different pathways: relative erythropoietin deficit, a direct inhibition of the erythropoiesis in the bone marrow through different mediators (ie, interleukin [IL]‐1, IL‐6, tumor necrosis factor [TNF]) and a functional iron deficit. This relative iron deficiency is produced mainly by the IL‐6 induced overexpression of the hepcidin gene in hepatocytes. Hepcidin causes impaired intestinal iron absorption and an inadequate delivery of iron from the iron‐recycling macrophages to the erythroid precursors in the bone marrow.8, 9

AI could explain some of our findings, such as the greater decrease observed in the first hospital days when inflamatory mediators levels are expected to be higher. The association between APS and infectious disease diagnosis and a greater [Hb] decrease may also be explained by this mechanism. Nonetheless, the daily bone marrow production of red blood cells is small compared with the circulating red mass cell and therefore it is necessary to have a better explanation on how AI could be associated with this rapid decrease in the [Hb]. Rice et al.30, 31 have explored some mechanisms leading to a rapid decrease in the red mass cell related to acute variations in erytropoietin level. A role of this mechanism and other could be hypothesized.29

Overt bleeding was not found in our study population and procedures without significant blood loss (PWSBL) were infrequent, therefore it is unlikely that they could have had an impact on the decrease of [Hb]. The influence of other variables, such as blood volume drawn for diagnostic testing and occult blood losses (especially from the gastrointestinal tract) was not investigated. Parenteral hydration and hemodilution neither were evaluated in our study nor were extensively described in the literature. We think this last mentioned mechanism plays a role in the large [Hb] decrease during the first days of hospitalization. However a lack of association was mentioned in the only study that evaluated this issue in critically ill patients. Intravascular hemolysis has been cited as an infrequent event in these patients.2, 4

Only patients without a clear explanation for their [Hb] decrease were included because they represent a matter of concern for hospitalists and other treating physicians. Therefore 60% of the initially evaluated patients were excluded, since they were admitted for conditions likely to cause a decrease in [Hb]. Nevertheless, it seems likely that the mechanisms involved in the included patients may play a role in those with an obvious cause for [Hb] variation as well. Accordingly, the decrease of [Hb] expected in patients admitted with disorders known to cause a decrease in the red blood cell mass would be greater than the 1 observed in our study population.

This [Hb] drop during hospitalization may be clinically relevant in a number of ways: it could cause the attending physicians to order costly and invasive studies, it could have prognostic value as it occurs in patients admitted with myocardial infarction, and it could trigger an acute coronary syndrome.20 We observed an association between [Hb] drop and a higher APS in our patients. This score has been validated as a prognostic factor in previous studies. However, it is not possible to conclude that this [Hb] drop is associated with a worse prognosis.

Our study had several limitations. The sample's heterogeneity inherent to our general ward population could have altered our results and their generalization. To overcome this bias we used a categorization system based on discharge diagnosis. However, this particular system has several limitations. The small sample size and the absence of follow up data after discharge limited our capacity to detect any prognostic significance of [Hb] decrease. In the present study, the total decrease of [Hb] may have been underestimated because the onset of the acute illness precedes hospital admission by a variable time. Finally, a relatively small sample size limited our ability to detect other possible significant predictors of [Hb] decrease. However, this study was not designed to assess the mechanisms associated with the decrease of [Hb], but rather to establish its occurrence, measure it and explore related variables.

These results may be useful for further studies to evaluate [Hb] variation in admitted patients and its relation to other variables, such as bone marrow production, oxygen use, erythrocyte survival, nutritional deficiencies, and erythropoietin and inflammatory mediators levels.

In conclusion, our general medical inpatients had a mean 1.45 g/dL [Hb] decrease during hospitalization, which was greater in the first days of hospitalization, even though an evident cause was not present. These findings would help attending physicians in general wards make a rational and efficient approach when dealing with patients' decrease in [Hb].

Studies of the red blood cell mass performed in hospitalized patients were first done in the early 1970s.1 Thereafter, a decrease in the hemoglobin concentration ([Hb]) and its potential causes have been further reported, especially in critically ill patients.25 The decrease in [Hb] can occur as a result of blood draws for diagnostic testing; blood loss associated with invasive procedures and bleeding; occult gastrointestinal bleeding; hemolysis; shortening of red cell survival; iron, folic acid or cobalamine deficiencies; renal, liver or endocrine failure; hemodilution associated with fluid therapy; and the so‐called anemia of inflammation (AI).68 The latter would be a consequence of a blunted response of the bone marrow due to several factors such as inadequate secretion of erythropoietin, inhibition of the proliferation and differentiation of the erythroid precursors of the bone marrow and an hepcidin‐mediated functional iron deficiency.810

As mentioned, most studies were done in critically ill patients and scarce information is available about general ward admitted patients (GWAP) with less severe illnesses. In this scenario, some of the proposed mechanisms may have a less clear role. It is widely accepted that [Hb] may decrease without overt bleeding in GWAP. However, given the lack of information on this matter, laboratory controls and invasive procedures are undertaken to determine its potential causes.

The purpose of the present study is to describe [Hb] variation over time in nonbleeding GWAP, to estimate the proportion of patients with [Hb] decreases 1.5 g/dL, and to evaluate possible related variables.

Materials and Methods

A 16‐week (September 2004‐January 2005) prospective observational study was conducted in Internal Medicine GWAP at 2 Buenos Aires teaching hospitals.

All consecutive patients older than 16 years were evaluated. Patients admitted for the following reasons were excluded: bleeding, trauma, surgery, invasive procedures associated with blood loss (biopsies, biliary drainage, endovascular therapeutic procedures, and chest tubes), blood transfusions, anemia, chemotherapy, and acute renal failure.

Patients with a bleeding history, chemotherapy or radiation therapy within two months prior to admission, patients on dialysis and patients with current oncologic or hematologic disease were excluded, as well as those with length of stay less than 3 days, or with less than 2 [Hb] or hematocrit (HCT) measurements.

Patients were followed until discharge, death or transfer to a critical care unit. Patients were withdrawn from the study if they presented with bleeding or hemolysis, underwent red blood cell transfusion or therapy affecting hemoglobin levels (iron, chemotherapy or erythropoietin) or if either a surgical or invasive procedure associated with blood loss was performed.

Data were collected from patients' medical records and additional information was obtained from treating physicians and patients by the authors (AL, NC, SM, AN, MH). Standardized case report forms were completed during the hospitalization including: age, sex, admissions in the previous 3 months (readmissions) and whether or not the patient lived in a nursing home. Patients were categorized according to discharge diagnosis as reported by Nguyen et al.4 with modifications related to our general ward population. Upon admission Katz daily activity index (ADL),11, 12 APACHE II acute physiology score (APS),13, 14 and Charlson comorbidity score (Charlson)15, 16 were assessed. In all cases the [Hb] and HCT values were registered on admission and on days 3, 6, 10 and prior to discharge as well as any other determination required by the treating physician. Anemia was defined as [Hb] 13 g/dL for men and 12 g/dL for women.17 Based on previous reports a 1.5 g/dL decrease in [Hb] and of 4.5 points in HCT compared to admission values were considered a significant fall.1, 4 Acute renal failure was defined as an increase in creatinine level 0.5 mg/dL or a 50% increase from baseline.18, 19

Every procedure without significant blood loss (PWSBL) such as venous catheter placement, thoracentesis, lumbar punction, paracentesis, skin biopsy, arthrocentesis, and diagnostic angiogram was also recorded. The study was approved by the Hospital's ethical committee.

Data analysis and processing were performed with Excel 2000 and Stata version 8.0 (Stata Corp; USA). For continuous variables, results were expressed by the mean value its standard deviation (SD), and compared with Student's t‐test. The chi‐square test was used to compare categorical variables. A survival curve was developed with the Kaplan‐Meier method to analyze the time to a significant fall in [Hb]. Finally, Cox proportional hazard modeling was performed to assess the association between a significant fall in [Hb] and other variables. We accepted P < 0,05 as significant.

Results

A total of 338 patients were admitted to the Internal Medicine Inpatient Services in the study period. Thirty‐nine percent (131) of these patients were included. Exclusion criteria were: diagnosis at admission (n = 95, 45.9%), past medical history (n = 56, 27%) and length of stay less than 3 days, or less than 2 determinations of [Hb] or HCT (n = 56, 27%). Data collection was discontinued for the following reasons: discharge (81.7%), death (6.9%), surgery or procedures associated with blood loss (5.3%), transfer to a critical care unit (3%), transfusions (2.3%), and chemotherapy (0.8%). The baseline characteristics of the study patients are shown in Table 1.

Diagnoses at discharge were divided into the following categories: infections (25.2%), electrolyte disturbances (7.6%), cardiac diseases (9.9%), neurologic (19.1%), respiratory (16.8%), gastrointestinal (10.7%), and other diagnosis (10.7%).

No evidence of bleeding was found in the included patients. Bleeding was only observed in 4 of the initially evaluated patients who were excluded since they failed to meet the required number of [Hb] determinations before bleeding.

A mean decrease in [Hb] of 0.71 g/dL was found between admission and discharge day (P < 0.001; 95% CI, 0.47‐0.97). Mean nadir [Hb] was 1.45 g/dL lower than admission [Hb] (P < 0.001; 95% CI, 1.24‐1.67). Mean nadir day occurred between hospitalization days 3 and 4. Mean [Hb] at discharge was 11.8 1.8 g/dL. This value is higher than the mean concentration at nadir (0.74 g/dL P < 0.001; 95% CI, 0.60‐0.97) (Figure 1).

Figure 1

Table 2 shows the rate of patients with a decrease in the [Hb] for different cutoff levels. Forty‐five percent of the study population (59 patients) had a significant fall in [Hb] during hospitalization. This was observed on day 2 in 50% of the patients. Likewise, a significant fall in HCT value was found in 42.7% of patients. If [Hb] decrease during hospitalization is analyzed as a proportion of [Hb] at admission, 52.7% of patients had a 10% or greater [Hb] decrease during their hospital stay.

Using Kaplan‐Meier analysis, the estimated proportion showing no significant fall of [Hb] was 0.63 (95% CI, 0.55‐0.71) on day 3 and 0.48 (95% CI, 0.29‐0.67) on day 10. By day 15, only 3.8% of the initially included patients were still hospitalized and showed no decrease in [Hb] 1.5 g/dl. These patients maintained their [Hb] stable for the rest of the follow up period (Figure 2).

Figure 2

In the univariate analysis, comparing patients that experienced a decrease in [Hb] 1.5 g/dL with those who did not, significant differences were only found in the following variables: length of stay, APS, anemia at admission, [Hb] at admission, and infectious, gastrointestinal or cardiac diseases at discharge diagnosis (Table 3).

In the Cox proportional hazard model adjusting for the variables shown in Table 4, a significant independent direct association was found between a significant fall in [Hb] during hospital stay and APS, [Hb] at admission, and either diagnosis of infectious or gastrointestinal disease at discharge. Similar results were found if a significant fall was redefined as a 12% decrease in [Hb] or as a 4.5 point decrease in HCT from baseline.

Discussion

This study describes the variation of [Hb] and HCT values in GWAP without bleeding or other obvious medical conditions associated with a decrease in the red blood cell mass. As previously described,4, 20, 21 we found that [Hb] decreases during hospital stay are frequently observed. The mean [Hb] fall from admission was 1.45 g/dL, and was mainly recorded between hospitalization days 3 and 4. In approximately half of the study population, a 1.5 g/dL decrease in [Hb] was observed. In the survival analysis, 40% and 55% of patients are expected to present such a fall on day 4 and 12 respectively. In accordance with prior reports4, 21 a greater decrease in [Hb] occurred in the first days of hospitalization, and a high proportion of patients were already anemic at the time of admission.

The following variables were associated with a decrease in the [Hb] during hospitalization: higher APS score, higher [Hb] at admission, and diagnosis of infectious or gastrointestinal disease at discharge. Even though several mechanisms seem to contribute to the decrease in [Hb] during hospitalization our data suggest that 1 of these factors seems to be the severity of the disease, as previously proposed by Nguyen et al.4 This observation is supported by the association found in our study between [Hb] decrease and APS. No association was found with Charlson, ADL, and being a nursing home resident. These variables, which have not been previously analyzed, seem to indicate that patients with chronic illnesses are not more likely to have decreases in [Hb] during hospitalization.

Patients with higher [Hb] at admission had a greater [Hb] decrease during their hospital stay. This finding could suggest that the mechanism related to this decrease had less effect on patients with lower [Hb]. This is similar to that observed in the anemia of chronic diseases where [Hb] does not usually fall to extreme values. Our analysis reveals a greater decrease during the first days of hospitalization. Given the high rate of patients anemic at admission, it is possible that the decrease in [Hb] had begun prior to admission.

Previous papers describe a low prevalence of cyanocobalamine (2%), folic acid (2%), and iron (9%) deficiency in patients admitted to critical care units.3 In these patients an inadequate bone marrow response has been proposed as a mechanism for [Hb] decrease, a phenomenon that has been called AI or anemia of critical illness.8, 2228 Inflammatory response associated with acute disease causes this hypoproliferative anemia through 3 different pathways: relative erythropoietin deficit, a direct inhibition of the erythropoiesis in the bone marrow through different mediators (ie, interleukin [IL]‐1, IL‐6, tumor necrosis factor [TNF]) and a functional iron deficit. This relative iron deficiency is produced mainly by the IL‐6 induced overexpression of the hepcidin gene in hepatocytes. Hepcidin causes impaired intestinal iron absorption and an inadequate delivery of iron from the iron‐recycling macrophages to the erythroid precursors in the bone marrow.8, 9

AI could explain some of our findings, such as the greater decrease observed in the first hospital days when inflamatory mediators levels are expected to be higher. The association between APS and infectious disease diagnosis and a greater [Hb] decrease may also be explained by this mechanism. Nonetheless, the daily bone marrow production of red blood cells is small compared with the circulating red mass cell and therefore it is necessary to have a better explanation on how AI could be associated with this rapid decrease in the [Hb]. Rice et al.30, 31 have explored some mechanisms leading to a rapid decrease in the red mass cell related to acute variations in erytropoietin level. A role of this mechanism and other could be hypothesized.29

Overt bleeding was not found in our study population and procedures without significant blood loss (PWSBL) were infrequent, therefore it is unlikely that they could have had an impact on the decrease of [Hb]. The influence of other variables, such as blood volume drawn for diagnostic testing and occult blood losses (especially from the gastrointestinal tract) was not investigated. Parenteral hydration and hemodilution neither were evaluated in our study nor were extensively described in the literature. We think this last mentioned mechanism plays a role in the large [Hb] decrease during the first days of hospitalization. However a lack of association was mentioned in the only study that evaluated this issue in critically ill patients. Intravascular hemolysis has been cited as an infrequent event in these patients.2, 4

Only patients without a clear explanation for their [Hb] decrease were included because they represent a matter of concern for hospitalists and other treating physicians. Therefore 60% of the initially evaluated patients were excluded, since they were admitted for conditions likely to cause a decrease in [Hb]. Nevertheless, it seems likely that the mechanisms involved in the included patients may play a role in those with an obvious cause for [Hb] variation as well. Accordingly, the decrease of [Hb] expected in patients admitted with disorders known to cause a decrease in the red blood cell mass would be greater than the 1 observed in our study population.

This [Hb] drop during hospitalization may be clinically relevant in a number of ways: it could cause the attending physicians to order costly and invasive studies, it could have prognostic value as it occurs in patients admitted with myocardial infarction, and it could trigger an acute coronary syndrome.20 We observed an association between [Hb] drop and a higher APS in our patients. This score has been validated as a prognostic factor in previous studies. However, it is not possible to conclude that this [Hb] drop is associated with a worse prognosis.

Our study had several limitations. The sample's heterogeneity inherent to our general ward population could have altered our results and their generalization. To overcome this bias we used a categorization system based on discharge diagnosis. However, this particular system has several limitations. The small sample size and the absence of follow up data after discharge limited our capacity to detect any prognostic significance of [Hb] decrease. In the present study, the total decrease of [Hb] may have been underestimated because the onset of the acute illness precedes hospital admission by a variable time. Finally, a relatively small sample size limited our ability to detect other possible significant predictors of [Hb] decrease. However, this study was not designed to assess the mechanisms associated with the decrease of [Hb], but rather to establish its occurrence, measure it and explore related variables.

These results may be useful for further studies to evaluate [Hb] variation in admitted patients and its relation to other variables, such as bone marrow production, oxygen use, erythrocyte survival, nutritional deficiencies, and erythropoietin and inflammatory mediators levels.

In conclusion, our general medical inpatients had a mean 1.45 g/dL [Hb] decrease during hospitalization, which was greater in the first days of hospitalization, even though an evident cause was not present. These findings would help attending physicians in general wards make a rational and efficient approach when dealing with patients' decrease in [Hb].

Studies of the red blood cell mass performed in hospitalized patients were first done in the early 1970s.1 Thereafter, a decrease in the hemoglobin concentration ([Hb]) and its potential causes have been further reported, especially in critically ill patients.25 The decrease in [Hb] can occur as a result of blood draws for diagnostic testing; blood loss associated with invasive procedures and bleeding; occult gastrointestinal bleeding; hemolysis; shortening of red cell survival; iron, folic acid or cobalamine deficiencies; renal, liver or endocrine failure; hemodilution associated with fluid therapy; and the so‐called anemia of inflammation (AI).68 The latter would be a consequence of a blunted response of the bone marrow due to several factors such as inadequate secretion of erythropoietin, inhibition of the proliferation and differentiation of the erythroid precursors of the bone marrow and an hepcidin‐mediated functional iron deficiency.810

As mentioned, most studies were done in critically ill patients and scarce information is available about general ward admitted patients (GWAP) with less severe illnesses. In this scenario, some of the proposed mechanisms may have a less clear role. It is widely accepted that [Hb] may decrease without overt bleeding in GWAP. However, given the lack of information on this matter, laboratory controls and invasive procedures are undertaken to determine its potential causes.

The purpose of the present study is to describe [Hb] variation over time in nonbleeding GWAP, to estimate the proportion of patients with [Hb] decreases 1.5 g/dL, and to evaluate possible related variables.

Materials and Methods

A 16‐week (September 2004‐January 2005) prospective observational study was conducted in Internal Medicine GWAP at 2 Buenos Aires teaching hospitals.

All consecutive patients older than 16 years were evaluated. Patients admitted for the following reasons were excluded: bleeding, trauma, surgery, invasive procedures associated with blood loss (biopsies, biliary drainage, endovascular therapeutic procedures, and chest tubes), blood transfusions, anemia, chemotherapy, and acute renal failure.

Patients with a bleeding history, chemotherapy or radiation therapy within two months prior to admission, patients on dialysis and patients with current oncologic or hematologic disease were excluded, as well as those with length of stay less than 3 days, or with less than 2 [Hb] or hematocrit (HCT) measurements.

Patients were followed until discharge, death or transfer to a critical care unit. Patients were withdrawn from the study if they presented with bleeding or hemolysis, underwent red blood cell transfusion or therapy affecting hemoglobin levels (iron, chemotherapy or erythropoietin) or if either a surgical or invasive procedure associated with blood loss was performed.

Data were collected from patients' medical records and additional information was obtained from treating physicians and patients by the authors (AL, NC, SM, AN, MH). Standardized case report forms were completed during the hospitalization including: age, sex, admissions in the previous 3 months (readmissions) and whether or not the patient lived in a nursing home. Patients were categorized according to discharge diagnosis as reported by Nguyen et al.4 with modifications related to our general ward population. Upon admission Katz daily activity index (ADL),11, 12 APACHE II acute physiology score (APS),13, 14 and Charlson comorbidity score (Charlson)15, 16 were assessed. In all cases the [Hb] and HCT values were registered on admission and on days 3, 6, 10 and prior to discharge as well as any other determination required by the treating physician. Anemia was defined as [Hb] 13 g/dL for men and 12 g/dL for women.17 Based on previous reports a 1.5 g/dL decrease in [Hb] and of 4.5 points in HCT compared to admission values were considered a significant fall.1, 4 Acute renal failure was defined as an increase in creatinine level 0.5 mg/dL or a 50% increase from baseline.18, 19

Every procedure without significant blood loss (PWSBL) such as venous catheter placement, thoracentesis, lumbar punction, paracentesis, skin biopsy, arthrocentesis, and diagnostic angiogram was also recorded. The study was approved by the Hospital's ethical committee.

Data analysis and processing were performed with Excel 2000 and Stata version 8.0 (Stata Corp; USA). For continuous variables, results were expressed by the mean value its standard deviation (SD), and compared with Student's t‐test. The chi‐square test was used to compare categorical variables. A survival curve was developed with the Kaplan‐Meier method to analyze the time to a significant fall in [Hb]. Finally, Cox proportional hazard modeling was performed to assess the association between a significant fall in [Hb] and other variables. We accepted P < 0,05 as significant.

Results

A total of 338 patients were admitted to the Internal Medicine Inpatient Services in the study period. Thirty‐nine percent (131) of these patients were included. Exclusion criteria were: diagnosis at admission (n = 95, 45.9%), past medical history (n = 56, 27%) and length of stay less than 3 days, or less than 2 determinations of [Hb] or HCT (n = 56, 27%). Data collection was discontinued for the following reasons: discharge (81.7%), death (6.9%), surgery or procedures associated with blood loss (5.3%), transfer to a critical care unit (3%), transfusions (2.3%), and chemotherapy (0.8%). The baseline characteristics of the study patients are shown in Table 1.

Diagnoses at discharge were divided into the following categories: infections (25.2%), electrolyte disturbances (7.6%), cardiac diseases (9.9%), neurologic (19.1%), respiratory (16.8%), gastrointestinal (10.7%), and other diagnosis (10.7%).

No evidence of bleeding was found in the included patients. Bleeding was only observed in 4 of the initially evaluated patients who were excluded since they failed to meet the required number of [Hb] determinations before bleeding.

A mean decrease in [Hb] of 0.71 g/dL was found between admission and discharge day (P < 0.001; 95% CI, 0.47‐0.97). Mean nadir [Hb] was 1.45 g/dL lower than admission [Hb] (P < 0.001; 95% CI, 1.24‐1.67). Mean nadir day occurred between hospitalization days 3 and 4. Mean [Hb] at discharge was 11.8 1.8 g/dL. This value is higher than the mean concentration at nadir (0.74 g/dL P < 0.001; 95% CI, 0.60‐0.97) (Figure 1).

Figure 1

Table 2 shows the rate of patients with a decrease in the [Hb] for different cutoff levels. Forty‐five percent of the study population (59 patients) had a significant fall in [Hb] during hospitalization. This was observed on day 2 in 50% of the patients. Likewise, a significant fall in HCT value was found in 42.7% of patients. If [Hb] decrease during hospitalization is analyzed as a proportion of [Hb] at admission, 52.7% of patients had a 10% or greater [Hb] decrease during their hospital stay.

Using Kaplan‐Meier analysis, the estimated proportion showing no significant fall of [Hb] was 0.63 (95% CI, 0.55‐0.71) on day 3 and 0.48 (95% CI, 0.29‐0.67) on day 10. By day 15, only 3.8% of the initially included patients were still hospitalized and showed no decrease in [Hb] 1.5 g/dl. These patients maintained their [Hb] stable for the rest of the follow up period (Figure 2).

Figure 2

In the univariate analysis, comparing patients that experienced a decrease in [Hb] 1.5 g/dL with those who did not, significant differences were only found in the following variables: length of stay, APS, anemia at admission, [Hb] at admission, and infectious, gastrointestinal or cardiac diseases at discharge diagnosis (Table 3).

In the Cox proportional hazard model adjusting for the variables shown in Table 4, a significant independent direct association was found between a significant fall in [Hb] during hospital stay and APS, [Hb] at admission, and either diagnosis of infectious or gastrointestinal disease at discharge. Similar results were found if a significant fall was redefined as a 12% decrease in [Hb] or as a 4.5 point decrease in HCT from baseline.

Discussion

This study describes the variation of [Hb] and HCT values in GWAP without bleeding or other obvious medical conditions associated with a decrease in the red blood cell mass. As previously described,4, 20, 21 we found that [Hb] decreases during hospital stay are frequently observed. The mean [Hb] fall from admission was 1.45 g/dL, and was mainly recorded between hospitalization days 3 and 4. In approximately half of the study population, a 1.5 g/dL decrease in [Hb] was observed. In the survival analysis, 40% and 55% of patients are expected to present such a fall on day 4 and 12 respectively. In accordance with prior reports4, 21 a greater decrease in [Hb] occurred in the first days of hospitalization, and a high proportion of patients were already anemic at the time of admission.

The following variables were associated with a decrease in the [Hb] during hospitalization: higher APS score, higher [Hb] at admission, and diagnosis of infectious or gastrointestinal disease at discharge. Even though several mechanisms seem to contribute to the decrease in [Hb] during hospitalization our data suggest that 1 of these factors seems to be the severity of the disease, as previously proposed by Nguyen et al.4 This observation is supported by the association found in our study between [Hb] decrease and APS. No association was found with Charlson, ADL, and being a nursing home resident. These variables, which have not been previously analyzed, seem to indicate that patients with chronic illnesses are not more likely to have decreases in [Hb] during hospitalization.

Patients with higher [Hb] at admission had a greater [Hb] decrease during their hospital stay. This finding could suggest that the mechanism related to this decrease had less effect on patients with lower [Hb]. This is similar to that observed in the anemia of chronic diseases where [Hb] does not usually fall to extreme values. Our analysis reveals a greater decrease during the first days of hospitalization. Given the high rate of patients anemic at admission, it is possible that the decrease in [Hb] had begun prior to admission.

Previous papers describe a low prevalence of cyanocobalamine (2%), folic acid (2%), and iron (9%) deficiency in patients admitted to critical care units.3 In these patients an inadequate bone marrow response has been proposed as a mechanism for [Hb] decrease, a phenomenon that has been called AI or anemia of critical illness.8, 2228 Inflammatory response associated with acute disease causes this hypoproliferative anemia through 3 different pathways: relative erythropoietin deficit, a direct inhibition of the erythropoiesis in the bone marrow through different mediators (ie, interleukin [IL]‐1, IL‐6, tumor necrosis factor [TNF]) and a functional iron deficit. This relative iron deficiency is produced mainly by the IL‐6 induced overexpression of the hepcidin gene in hepatocytes. Hepcidin causes impaired intestinal iron absorption and an inadequate delivery of iron from the iron‐recycling macrophages to the erythroid precursors in the bone marrow.8, 9

AI could explain some of our findings, such as the greater decrease observed in the first hospital days when inflamatory mediators levels are expected to be higher. The association between APS and infectious disease diagnosis and a greater [Hb] decrease may also be explained by this mechanism. Nonetheless, the daily bone marrow production of red blood cells is small compared with the circulating red mass cell and therefore it is necessary to have a better explanation on how AI could be associated with this rapid decrease in the [Hb]. Rice et al.30, 31 have explored some mechanisms leading to a rapid decrease in the red mass cell related to acute variations in erytropoietin level. A role of this mechanism and other could be hypothesized.29

Overt bleeding was not found in our study population and procedures without significant blood loss (PWSBL) were infrequent, therefore it is unlikely that they could have had an impact on the decrease of [Hb]. The influence of other variables, such as blood volume drawn for diagnostic testing and occult blood losses (especially from the gastrointestinal tract) was not investigated. Parenteral hydration and hemodilution neither were evaluated in our study nor were extensively described in the literature. We think this last mentioned mechanism plays a role in the large [Hb] decrease during the first days of hospitalization. However a lack of association was mentioned in the only study that evaluated this issue in critically ill patients. Intravascular hemolysis has been cited as an infrequent event in these patients.2, 4

Only patients without a clear explanation for their [Hb] decrease were included because they represent a matter of concern for hospitalists and other treating physicians. Therefore 60% of the initially evaluated patients were excluded, since they were admitted for conditions likely to cause a decrease in [Hb]. Nevertheless, it seems likely that the mechanisms involved in the included patients may play a role in those with an obvious cause for [Hb] variation as well. Accordingly, the decrease of [Hb] expected in patients admitted with disorders known to cause a decrease in the red blood cell mass would be greater than the 1 observed in our study population.

This [Hb] drop during hospitalization may be clinically relevant in a number of ways: it could cause the attending physicians to order costly and invasive studies, it could have prognostic value as it occurs in patients admitted with myocardial infarction, and it could trigger an acute coronary syndrome.20 We observed an association between [Hb] drop and a higher APS in our patients. This score has been validated as a prognostic factor in previous studies. However, it is not possible to conclude that this [Hb] drop is associated with a worse prognosis.

Our study had several limitations. The sample's heterogeneity inherent to our general ward population could have altered our results and their generalization. To overcome this bias we used a categorization system based on discharge diagnosis. However, this particular system has several limitations. The small sample size and the absence of follow up data after discharge limited our capacity to detect any prognostic significance of [Hb] decrease. In the present study, the total decrease of [Hb] may have been underestimated because the onset of the acute illness precedes hospital admission by a variable time. Finally, a relatively small sample size limited our ability to detect other possible significant predictors of [Hb] decrease. However, this study was not designed to assess the mechanisms associated with the decrease of [Hb], but rather to establish its occurrence, measure it and explore related variables.

These results may be useful for further studies to evaluate [Hb] variation in admitted patients and its relation to other variables, such as bone marrow production, oxygen use, erythrocyte survival, nutritional deficiencies, and erythropoietin and inflammatory mediators levels.

In conclusion, our general medical inpatients had a mean 1.45 g/dL [Hb] decrease during hospitalization, which was greater in the first days of hospitalization, even though an evident cause was not present. These findings would help attending physicians in general wards make a rational and efficient approach when dealing with patients' decrease in [Hb].

In the United States, candida now accounts for between 8% and 12% of all catheter‐associated blood stream infections (BSIs).1 Additionally, crude mortality rates in candidemia exceed 40%, and a recent systematic review demonstrated that the attributable mortality of candidemia ranges from 5% to 71%.2 Candidal BSIs also affect resource utilization. These infections independently increase length of stay and result in substantial excess costs.3 Most cases of candidemia arise in noncritically ill patients and thus may be managed by hospitalists.

Historically, the majority of candidal BSIs were caused by C. albicans. Presently, C. albicans accounts for only half of all yeast BSIs, and approximately 20% of these infections are caused by organisms such as C. glabrata and C. krusei.4 These 2 organisms have either variable or no susceptibility to agents, such as fluconazole, empirically employed against yeast. Parallel with the evolution in microbiology of candidemia has been recognition that inappropriate treatment of these infections independently increases mortality.5 These factors underscore the need for the clinician to treat suspected candidal BSIs aggressively in order to avoid the risks associated with inappropriate treatment.

Efforts to enhance rates of initial appropriate therapy for bacterial infections have encompassed the realization that health care‐associated infections (HAIs) represent a distinct syndrome.6, 7 Traditionally, infections were considered either community‐acquired or nosocomial in origin. However, with the spread of health care delivery beyond the hospital, multiple studies indicate that patients may now present to the emergency department with infections caused by pathogens such as Methicillin‐resistant Staphylococcus aureus (MRSA) and P. aeruginosaorganisms that were previously thought limited to hospital‐acquired processes.69 Furthermore, hospitalists often encounter subjects presenting to the hospital with suspected BSIs who have an active and ongoing interaction with the healthcare system.

The importance of candida as a health care‐associated pathogen in BSI remains unclear. We hypothesized that health care‐associated candidemia (HCAC) represented a distinct clinical entity. In order to confirm our theory, we conducted a retrospective analysis of all cases of candidal BSI at our institution over a 3‐year period.

Methods

We reviewed the records of all patients diagnosed with candidemia at our hospital between January 1, 2004 and December 31, 2006. Our institutional review board approved this study. We included adult patients diagnosed with candidemia. The diagnosis of candidemia was based on the isolation of yeast from the blood in at least one blood culture. We employ the BACTEC 9240 blood Culture System (Becton Dickinson Microbiology Systems, Sparks, MD). We excluded subjects who were admitted to the hospital within one month of a known diagnosis of candidemia.

We defined a nosocomial candidal BSI as the diagnosis of candidemia based on cultures drawn after the patient had been hospitalized for >48 hours. We considered HCAC to be present based on previously employed criteria for identifying HAI.69 Specifically, for patients with candidemia based on blood cultures obtained within 48 hours of hospitalization, a patient had to meet at least 1 of the following criteria: (1) receipt of intravenous therapy outside the hospital, (2) end stage renal disease necessitating hemodialysis (ESRD requiring HD), (3) hospitalization within previous 30 days, (4) residence in a nursing home or long term care facility, or (5) underwent an invasive procedure as an outpatient within 30 days of presentation. Community‐acquired candidemia was restricted to patients whose index culture was drawn within 48 hours of admission but who failed to meet the definition for HCAC.

The prevalence of the various forms of candidemia served as our primary endpoint. In addition, we compared patients with respect to demographic factors, comorbidities, and severity of illness. Severity of illness was calculated based on the Acute Physiology and Chronic Health Evaluation (APACHE) II score. We further noted rates of immune suppression in the cohort and defined this as treatment with corticosteroids (10 mg of prednisone or equivalent daily for more than 30 consecutive days), other immunosuppressants (eg, methotrexate), or chemotherapy. Those with acquired immune deficiency syndrome (AIDS) or another immunodeficiency syndrome were defined as immunosuppressed as well. We examined the distribution of yeast species across the 3 forms of candidemia. Finally, we assessed the prevalence of fluconazole resistance. Fluconazole susceptibilities were determined based on Etest (AB BIODISK, Solna, Sweden). An isolate was considered resistant to fluconazole if the minimum inhibitory concentration was >64 g/mL.

We compared categorical variables with the Fisher's exact test. Continuous variables were analyzed with either the Student's t‐test or a Mann‐Whitney test, as appropriate. All tests were 2 tailed and a P value of 0.05 was assumed to represent statistical significance. Analyses were performed with Stata 9.1 (Stata Corp., College Station, TX).

Results

The final cohort included 223 subjects. The mean age of the patients was 59.6 15.7 years and 49% were male. Nearly one quarter (n = 55) fulfilled our criteria for HCAC. The remainder met the definition for nosocomial candidemia. We observed no cases of community‐acquired candidemia. Most (n = 33) patients with HCAC had exposure to more than 1 health care‐related source and many were initially admitted to the medicine/hospitalist service as opposed to the intensive care unit (ICU). The most common criteria leading to categorization as HCAC was recent hospitalization (n = 30, 54.5% of all HCAC). The median time from recent hospitalization to admission was 17 days (Range: 5‐28 days). Other common reasons for classification as HCAC included ESRD requiring HD (30.9%), residence in a nursing home (25.5%), and undergoing an invasive outpatient procedure (16.4%). More than 75% of subjects with HCAC (n = 42) had central venous catheters in place at presentation. Between 2004 and 2006, the proportion of all candidemia due to HCAC increased from 20.9% to 26.9%, but this difference was not statistically significant.

Patients with HCAC were similar to those with nosocomial candidemia (Table 1). There was no difference in either severity of illness or the frequency of neutropenia. The prevalence of most comorbidities did not differ between those with nosocomial candidemia and persons with HCAC. However, immunosuppression was more prevalent among patients with HCAC (prevalence ratio, 1.67; 95% CI, 1.13‐3.08; P = 0.004). In part this finding is expected given that our definition of HCAC includes exposure to agents which may lead to immunosuppression, such as chemotherapy. Of patients with HCAC, the majority (n = 38, 69.1%) were initially admitted to the general medicine service and not to the ICU.

A multitude of various yeast species were recovered (Figure 1). Overall, nonalbicans candida were responsible for nearly 60% of all infections. Nonalbicans yeast were as likely to be recovered in HCAC as in nosocomial yeast infection. Among both types of Candidemia, C. krusei was a rare culprit accounting for fewer than 2% of infections. C. glabrata, however, occurred more often in HCAC. Specifically, C. glabrata represented 1 in 5 cases of HCAC as opposed to approximately 10% of all nosocomial yeast BSIs (P = 0.05). In part reflecting this, fluconazole resistance was noted more often in HCAC (18.2% of patients vs. 7.7% among nosocomial candidemia, P = 0.036). There was no difference in the eventual diagnosis of deep‐seeded yeast infections (ie, endocarditis, endopthlamitis, or osteomyelitis) between those with HCAC and persons with nosocomial candidemia (3 cases in each group).

Figure 1

Discussion

This analysis demonstrates that HCAC accounts for approximately a quarter of all candidemia. Our findings underscore that candidemia can present to the emergency department as an HAI and may potentially be initially cared for by a hospitalist. In addition, patients with HCAC and nosocomial candidemia share many attributes. Furthermore, nonalbicans yeast are as prevalent in HCAC as in nosocomial candidal infection. Nonetheless, there appear to be important differences in these syndromes. Immunosuppression appears to be more common in HCAC as does infection due to C. glabrata.

Others have explored the concept of HCAC. Kung et al.10 described community‐onset candidemia at a single center over a 10‐year period. They described 56 patients and noted that the majority had been recently hospitalized or had ongoing interaction with the healthcare system. Sofair et al.11 followed subjects presenting to emergency departments with candidemia. Overall, more than one‐third met criteria for community‐onset infection. In this analysis, though, Sofair et al.11 did not distinguish between community‐acquired processes and HCAC. From a population perspective, Chen et al.12 explored candidemia in Australia. Among over 1000 patients, the noted that 11.6% represented HCAC and, as we note, that select nonalbicans yeast occurred more often in HCAC than in nosocomial candidemia. Our project builds on and adds to these earlier efforts. First, we confirm the general observation that candidemia is no longer solely a nosocomial pathogen. Second, unlike several of these earlier reports we examined a larger cohort of candidemia. Third, beyond the observations of Chen et al.,12 we note that currently, the proportion of Candidal BSI classified as HACA relative to nosocomial candidemia seems larger than reported in the past. Finally, a unique aspect of our report is that we employed express criteria to define HAI.

Our findings have several implications. First, hospitalists and emergency department physicians, along with others, must remain vigilant when approaching patients presenting to the hospital with signs and symptoms of BSI and multiple risk factors for candidal BSI. The fact that the patient has not been hospitalized should not preclude consideration of and treatment for candidemia. The current evidence does not support broad, empiric use of antifungal agents, as this would lead to excessive costs and potentially expose many patients to unnecessary antifungal coverage. On the other hand, given the association between delayed antifungal therapy and the risk for death in candidemia, failure to consider this infection in at‐risk subjects may have adverse consequences. Second, our observations emphasize the need for clinical risk stratification schemes and rapid diagnostic modalities. Such tools are urgently needed if physicians hope to target antifungal therapies more appropriately. Third, if the clinician opts to initiate therapy for possible HCAC, reliance on fluconazole alone may prove inadequate. As the generalizability of our conclusions is necessarily limited, we recommend that infection control practitioners review local epidemiologic data regarding HCAC so that physicians can have the best available guidance.

Our study has several important limitations. Its retrospective nature exposes it to several forms of bias. The single center design limits the generalizability of our findings. Prospective, multicenter studies are needed to validate our results. Additionally, no universally accepted criteria exist to define HAI syndromes. Nonetheless, the criteria we employed have been used by others. We also lacked data on exposure to recent broad spectrum antimicrobials. Selection pressure via exposure to such agents is a risk factor for candidemia and without this data we cannot gauge the impact of this on our findings. Finally, we cannot control for the possibility that some patients were miscategorized. This could have arisen because of: (1) either limitations inherent in the definition of HCAC or (2) because the clinician delayed the decision to obtain blood cultures. Some patients classified as nosocomial may actually have had HCAC or community‐acquired diseasebut for some reason blood cultures were not drawn at time of admission but were deferred until later. Although a difficult issue to address in any study of the epidemiology of infection, the significance of this misclassification bias must be considered a significant concern.

In summary, Candidemia can be the cause of BSI presenting to the hospital. Moreover, HCAC represents a significant proportion of all Candidemia. Although patients with HCAC and nosocomial candidemia share select characteristics, there appear to be some differences in the microbiology of these syndromes.

In the United States, candida now accounts for between 8% and 12% of all catheter‐associated blood stream infections (BSIs).1 Additionally, crude mortality rates in candidemia exceed 40%, and a recent systematic review demonstrated that the attributable mortality of candidemia ranges from 5% to 71%.2 Candidal BSIs also affect resource utilization. These infections independently increase length of stay and result in substantial excess costs.3 Most cases of candidemia arise in noncritically ill patients and thus may be managed by hospitalists.

Historically, the majority of candidal BSIs were caused by C. albicans. Presently, C. albicans accounts for only half of all yeast BSIs, and approximately 20% of these infections are caused by organisms such as C. glabrata and C. krusei.4 These 2 organisms have either variable or no susceptibility to agents, such as fluconazole, empirically employed against yeast. Parallel with the evolution in microbiology of candidemia has been recognition that inappropriate treatment of these infections independently increases mortality.5 These factors underscore the need for the clinician to treat suspected candidal BSIs aggressively in order to avoid the risks associated with inappropriate treatment.

Efforts to enhance rates of initial appropriate therapy for bacterial infections have encompassed the realization that health care‐associated infections (HAIs) represent a distinct syndrome.6, 7 Traditionally, infections were considered either community‐acquired or nosocomial in origin. However, with the spread of health care delivery beyond the hospital, multiple studies indicate that patients may now present to the emergency department with infections caused by pathogens such as Methicillin‐resistant Staphylococcus aureus (MRSA) and P. aeruginosaorganisms that were previously thought limited to hospital‐acquired processes.69 Furthermore, hospitalists often encounter subjects presenting to the hospital with suspected BSIs who have an active and ongoing interaction with the healthcare system.

The importance of candida as a health care‐associated pathogen in BSI remains unclear. We hypothesized that health care‐associated candidemia (HCAC) represented a distinct clinical entity. In order to confirm our theory, we conducted a retrospective analysis of all cases of candidal BSI at our institution over a 3‐year period.

Methods

We reviewed the records of all patients diagnosed with candidemia at our hospital between January 1, 2004 and December 31, 2006. Our institutional review board approved this study. We included adult patients diagnosed with candidemia. The diagnosis of candidemia was based on the isolation of yeast from the blood in at least one blood culture. We employ the BACTEC 9240 blood Culture System (Becton Dickinson Microbiology Systems, Sparks, MD). We excluded subjects who were admitted to the hospital within one month of a known diagnosis of candidemia.

We defined a nosocomial candidal BSI as the diagnosis of candidemia based on cultures drawn after the patient had been hospitalized for >48 hours. We considered HCAC to be present based on previously employed criteria for identifying HAI.69 Specifically, for patients with candidemia based on blood cultures obtained within 48 hours of hospitalization, a patient had to meet at least 1 of the following criteria: (1) receipt of intravenous therapy outside the hospital, (2) end stage renal disease necessitating hemodialysis (ESRD requiring HD), (3) hospitalization within previous 30 days, (4) residence in a nursing home or long term care facility, or (5) underwent an invasive procedure as an outpatient within 30 days of presentation. Community‐acquired candidemia was restricted to patients whose index culture was drawn within 48 hours of admission but who failed to meet the definition for HCAC.

The prevalence of the various forms of candidemia served as our primary endpoint. In addition, we compared patients with respect to demographic factors, comorbidities, and severity of illness. Severity of illness was calculated based on the Acute Physiology and Chronic Health Evaluation (APACHE) II score. We further noted rates of immune suppression in the cohort and defined this as treatment with corticosteroids (10 mg of prednisone or equivalent daily for more than 30 consecutive days), other immunosuppressants (eg, methotrexate), or chemotherapy. Those with acquired immune deficiency syndrome (AIDS) or another immunodeficiency syndrome were defined as immunosuppressed as well. We examined the distribution of yeast species across the 3 forms of candidemia. Finally, we assessed the prevalence of fluconazole resistance. Fluconazole susceptibilities were determined based on Etest (AB BIODISK, Solna, Sweden). An isolate was considered resistant to fluconazole if the minimum inhibitory concentration was >64 g/mL.

We compared categorical variables with the Fisher's exact test. Continuous variables were analyzed with either the Student's t‐test or a Mann‐Whitney test, as appropriate. All tests were 2 tailed and a P value of 0.05 was assumed to represent statistical significance. Analyses were performed with Stata 9.1 (Stata Corp., College Station, TX).

Results

The final cohort included 223 subjects. The mean age of the patients was 59.6 15.7 years and 49% were male. Nearly one quarter (n = 55) fulfilled our criteria for HCAC. The remainder met the definition for nosocomial candidemia. We observed no cases of community‐acquired candidemia. Most (n = 33) patients with HCAC had exposure to more than 1 health care‐related source and many were initially admitted to the medicine/hospitalist service as opposed to the intensive care unit (ICU). The most common criteria leading to categorization as HCAC was recent hospitalization (n = 30, 54.5% of all HCAC). The median time from recent hospitalization to admission was 17 days (Range: 5‐28 days). Other common reasons for classification as HCAC included ESRD requiring HD (30.9%), residence in a nursing home (25.5%), and undergoing an invasive outpatient procedure (16.4%). More than 75% of subjects with HCAC (n = 42) had central venous catheters in place at presentation. Between 2004 and 2006, the proportion of all candidemia due to HCAC increased from 20.9% to 26.9%, but this difference was not statistically significant.

Patients with HCAC were similar to those with nosocomial candidemia (Table 1). There was no difference in either severity of illness or the frequency of neutropenia. The prevalence of most comorbidities did not differ between those with nosocomial candidemia and persons with HCAC. However, immunosuppression was more prevalent among patients with HCAC (prevalence ratio, 1.67; 95% CI, 1.13‐3.08; P = 0.004). In part this finding is expected given that our definition of HCAC includes exposure to agents which may lead to immunosuppression, such as chemotherapy. Of patients with HCAC, the majority (n = 38, 69.1%) were initially admitted to the general medicine service and not to the ICU.

A multitude of various yeast species were recovered (Figure 1). Overall, nonalbicans candida were responsible for nearly 60% of all infections. Nonalbicans yeast were as likely to be recovered in HCAC as in nosocomial yeast infection. Among both types of Candidemia, C. krusei was a rare culprit accounting for fewer than 2% of infections. C. glabrata, however, occurred more often in HCAC. Specifically, C. glabrata represented 1 in 5 cases of HCAC as opposed to approximately 10% of all nosocomial yeast BSIs (P = 0.05). In part reflecting this, fluconazole resistance was noted more often in HCAC (18.2% of patients vs. 7.7% among nosocomial candidemia, P = 0.036). There was no difference in the eventual diagnosis of deep‐seeded yeast infections (ie, endocarditis, endopthlamitis, or osteomyelitis) between those with HCAC and persons with nosocomial candidemia (3 cases in each group).

Figure 1

Discussion

This analysis demonstrates that HCAC accounts for approximately a quarter of all candidemia. Our findings underscore that candidemia can present to the emergency department as an HAI and may potentially be initially cared for by a hospitalist. In addition, patients with HCAC and nosocomial candidemia share many attributes. Furthermore, nonalbicans yeast are as prevalent in HCAC as in nosocomial candidal infection. Nonetheless, there appear to be important differences in these syndromes. Immunosuppression appears to be more common in HCAC as does infection due to C. glabrata.

Others have explored the concept of HCAC. Kung et al.10 described community‐onset candidemia at a single center over a 10‐year period. They described 56 patients and noted that the majority had been recently hospitalized or had ongoing interaction with the healthcare system. Sofair et al.11 followed subjects presenting to emergency departments with candidemia. Overall, more than one‐third met criteria for community‐onset infection. In this analysis, though, Sofair et al.11 did not distinguish between community‐acquired processes and HCAC. From a population perspective, Chen et al.12 explored candidemia in Australia. Among over 1000 patients, the noted that 11.6% represented HCAC and, as we note, that select nonalbicans yeast occurred more often in HCAC than in nosocomial candidemia. Our project builds on and adds to these earlier efforts. First, we confirm the general observation that candidemia is no longer solely a nosocomial pathogen. Second, unlike several of these earlier reports we examined a larger cohort of candidemia. Third, beyond the observations of Chen et al.,12 we note that currently, the proportion of Candidal BSI classified as HACA relative to nosocomial candidemia seems larger than reported in the past. Finally, a unique aspect of our report is that we employed express criteria to define HAI.

Our findings have several implications. First, hospitalists and emergency department physicians, along with others, must remain vigilant when approaching patients presenting to the hospital with signs and symptoms of BSI and multiple risk factors for candidal BSI. The fact that the patient has not been hospitalized should not preclude consideration of and treatment for candidemia. The current evidence does not support broad, empiric use of antifungal agents, as this would lead to excessive costs and potentially expose many patients to unnecessary antifungal coverage. On the other hand, given the association between delayed antifungal therapy and the risk for death in candidemia, failure to consider this infection in at‐risk subjects may have adverse consequences. Second, our observations emphasize the need for clinical risk stratification schemes and rapid diagnostic modalities. Such tools are urgently needed if physicians hope to target antifungal therapies more appropriately. Third, if the clinician opts to initiate therapy for possible HCAC, reliance on fluconazole alone may prove inadequate. As the generalizability of our conclusions is necessarily limited, we recommend that infection control practitioners review local epidemiologic data regarding HCAC so that physicians can have the best available guidance.

Our study has several important limitations. Its retrospective nature exposes it to several forms of bias. The single center design limits the generalizability of our findings. Prospective, multicenter studies are needed to validate our results. Additionally, no universally accepted criteria exist to define HAI syndromes. Nonetheless, the criteria we employed have been used by others. We also lacked data on exposure to recent broad spectrum antimicrobials. Selection pressure via exposure to such agents is a risk factor for candidemia and without this data we cannot gauge the impact of this on our findings. Finally, we cannot control for the possibility that some patients were miscategorized. This could have arisen because of: (1) either limitations inherent in the definition of HCAC or (2) because the clinician delayed the decision to obtain blood cultures. Some patients classified as nosocomial may actually have had HCAC or community‐acquired diseasebut for some reason blood cultures were not drawn at time of admission but were deferred until later. Although a difficult issue to address in any study of the epidemiology of infection, the significance of this misclassification bias must be considered a significant concern.

In summary, Candidemia can be the cause of BSI presenting to the hospital. Moreover, HCAC represents a significant proportion of all Candidemia. Although patients with HCAC and nosocomial candidemia share select characteristics, there appear to be some differences in the microbiology of these syndromes.

In the United States, candida now accounts for between 8% and 12% of all catheter‐associated blood stream infections (BSIs).1 Additionally, crude mortality rates in candidemia exceed 40%, and a recent systematic review demonstrated that the attributable mortality of candidemia ranges from 5% to 71%.2 Candidal BSIs also affect resource utilization. These infections independently increase length of stay and result in substantial excess costs.3 Most cases of candidemia arise in noncritically ill patients and thus may be managed by hospitalists.

Historically, the majority of candidal BSIs were caused by C. albicans. Presently, C. albicans accounts for only half of all yeast BSIs, and approximately 20% of these infections are caused by organisms such as C. glabrata and C. krusei.4 These 2 organisms have either variable or no susceptibility to agents, such as fluconazole, empirically employed against yeast. Parallel with the evolution in microbiology of candidemia has been recognition that inappropriate treatment of these infections independently increases mortality.5 These factors underscore the need for the clinician to treat suspected candidal BSIs aggressively in order to avoid the risks associated with inappropriate treatment.

Efforts to enhance rates of initial appropriate therapy for bacterial infections have encompassed the realization that health care‐associated infections (HAIs) represent a distinct syndrome.6, 7 Traditionally, infections were considered either community‐acquired or nosocomial in origin. However, with the spread of health care delivery beyond the hospital, multiple studies indicate that patients may now present to the emergency department with infections caused by pathogens such as Methicillin‐resistant Staphylococcus aureus (MRSA) and P. aeruginosaorganisms that were previously thought limited to hospital‐acquired processes.69 Furthermore, hospitalists often encounter subjects presenting to the hospital with suspected BSIs who have an active and ongoing interaction with the healthcare system.

The importance of candida as a health care‐associated pathogen in BSI remains unclear. We hypothesized that health care‐associated candidemia (HCAC) represented a distinct clinical entity. In order to confirm our theory, we conducted a retrospective analysis of all cases of candidal BSI at our institution over a 3‐year period.

Methods

We reviewed the records of all patients diagnosed with candidemia at our hospital between January 1, 2004 and December 31, 2006. Our institutional review board approved this study. We included adult patients diagnosed with candidemia. The diagnosis of candidemia was based on the isolation of yeast from the blood in at least one blood culture. We employ the BACTEC 9240 blood Culture System (Becton Dickinson Microbiology Systems, Sparks, MD). We excluded subjects who were admitted to the hospital within one month of a known diagnosis of candidemia.

We defined a nosocomial candidal BSI as the diagnosis of candidemia based on cultures drawn after the patient had been hospitalized for >48 hours. We considered HCAC to be present based on previously employed criteria for identifying HAI.69 Specifically, for patients with candidemia based on blood cultures obtained within 48 hours of hospitalization, a patient had to meet at least 1 of the following criteria: (1) receipt of intravenous therapy outside the hospital, (2) end stage renal disease necessitating hemodialysis (ESRD requiring HD), (3) hospitalization within previous 30 days, (4) residence in a nursing home or long term care facility, or (5) underwent an invasive procedure as an outpatient within 30 days of presentation. Community‐acquired candidemia was restricted to patients whose index culture was drawn within 48 hours of admission but who failed to meet the definition for HCAC.

The prevalence of the various forms of candidemia served as our primary endpoint. In addition, we compared patients with respect to demographic factors, comorbidities, and severity of illness. Severity of illness was calculated based on the Acute Physiology and Chronic Health Evaluation (APACHE) II score. We further noted rates of immune suppression in the cohort and defined this as treatment with corticosteroids (10 mg of prednisone or equivalent daily for more than 30 consecutive days), other immunosuppressants (eg, methotrexate), or chemotherapy. Those with acquired immune deficiency syndrome (AIDS) or another immunodeficiency syndrome were defined as immunosuppressed as well. We examined the distribution of yeast species across the 3 forms of candidemia. Finally, we assessed the prevalence of fluconazole resistance. Fluconazole susceptibilities were determined based on Etest (AB BIODISK, Solna, Sweden). An isolate was considered resistant to fluconazole if the minimum inhibitory concentration was >64 g/mL.

We compared categorical variables with the Fisher's exact test. Continuous variables were analyzed with either the Student's t‐test or a Mann‐Whitney test, as appropriate. All tests were 2 tailed and a P value of 0.05 was assumed to represent statistical significance. Analyses were performed with Stata 9.1 (Stata Corp., College Station, TX).

Results

The final cohort included 223 subjects. The mean age of the patients was 59.6 15.7 years and 49% were male. Nearly one quarter (n = 55) fulfilled our criteria for HCAC. The remainder met the definition for nosocomial candidemia. We observed no cases of community‐acquired candidemia. Most (n = 33) patients with HCAC had exposure to more than 1 health care‐related source and many were initially admitted to the medicine/hospitalist service as opposed to the intensive care unit (ICU). The most common criteria leading to categorization as HCAC was recent hospitalization (n = 30, 54.5% of all HCAC). The median time from recent hospitalization to admission was 17 days (Range: 5‐28 days). Other common reasons for classification as HCAC included ESRD requiring HD (30.9%), residence in a nursing home (25.5%), and undergoing an invasive outpatient procedure (16.4%). More than 75% of subjects with HCAC (n = 42) had central venous catheters in place at presentation. Between 2004 and 2006, the proportion of all candidemia due to HCAC increased from 20.9% to 26.9%, but this difference was not statistically significant.

Patients with HCAC were similar to those with nosocomial candidemia (Table 1). There was no difference in either severity of illness or the frequency of neutropenia. The prevalence of most comorbidities did not differ between those with nosocomial candidemia and persons with HCAC. However, immunosuppression was more prevalent among patients with HCAC (prevalence ratio, 1.67; 95% CI, 1.13‐3.08; P = 0.004). In part this finding is expected given that our definition of HCAC includes exposure to agents which may lead to immunosuppression, such as chemotherapy. Of patients with HCAC, the majority (n = 38, 69.1%) were initially admitted to the general medicine service and not to the ICU.

A multitude of various yeast species were recovered (Figure 1). Overall, nonalbicans candida were responsible for nearly 60% of all infections. Nonalbicans yeast were as likely to be recovered in HCAC as in nosocomial yeast infection. Among both types of Candidemia, C. krusei was a rare culprit accounting for fewer than 2% of infections. C. glabrata, however, occurred more often in HCAC. Specifically, C. glabrata represented 1 in 5 cases of HCAC as opposed to approximately 10% of all nosocomial yeast BSIs (P = 0.05). In part reflecting this, fluconazole resistance was noted more often in HCAC (18.2% of patients vs. 7.7% among nosocomial candidemia, P = 0.036). There was no difference in the eventual diagnosis of deep‐seeded yeast infections (ie, endocarditis, endopthlamitis, or osteomyelitis) between those with HCAC and persons with nosocomial candidemia (3 cases in each group).

Figure 1

Discussion

This analysis demonstrates that HCAC accounts for approximately a quarter of all candidemia. Our findings underscore that candidemia can present to the emergency department as an HAI and may potentially be initially cared for by a hospitalist. In addition, patients with HCAC and nosocomial candidemia share many attributes. Furthermore, nonalbicans yeast are as prevalent in HCAC as in nosocomial candidal infection. Nonetheless, there appear to be important differences in these syndromes. Immunosuppression appears to be more common in HCAC as does infection due to C. glabrata.

Others have explored the concept of HCAC. Kung et al.10 described community‐onset candidemia at a single center over a 10‐year period. They described 56 patients and noted that the majority had been recently hospitalized or had ongoing interaction with the healthcare system. Sofair et al.11 followed subjects presenting to emergency departments with candidemia. Overall, more than one‐third met criteria for community‐onset infection. In this analysis, though, Sofair et al.11 did not distinguish between community‐acquired processes and HCAC. From a population perspective, Chen et al.12 explored candidemia in Australia. Among over 1000 patients, the noted that 11.6% represented HCAC and, as we note, that select nonalbicans yeast occurred more often in HCAC than in nosocomial candidemia. Our project builds on and adds to these earlier efforts. First, we confirm the general observation that candidemia is no longer solely a nosocomial pathogen. Second, unlike several of these earlier reports we examined a larger cohort of candidemia. Third, beyond the observations of Chen et al.,12 we note that currently, the proportion of Candidal BSI classified as HACA relative to nosocomial candidemia seems larger than reported in the past. Finally, a unique aspect of our report is that we employed express criteria to define HAI.

Our findings have several implications. First, hospitalists and emergency department physicians, along with others, must remain vigilant when approaching patients presenting to the hospital with signs and symptoms of BSI and multiple risk factors for candidal BSI. The fact that the patient has not been hospitalized should not preclude consideration of and treatment for candidemia. The current evidence does not support broad, empiric use of antifungal agents, as this would lead to excessive costs and potentially expose many patients to unnecessary antifungal coverage. On the other hand, given the association between delayed antifungal therapy and the risk for death in candidemia, failure to consider this infection in at‐risk subjects may have adverse consequences. Second, our observations emphasize the need for clinical risk stratification schemes and rapid diagnostic modalities. Such tools are urgently needed if physicians hope to target antifungal therapies more appropriately. Third, if the clinician opts to initiate therapy for possible HCAC, reliance on fluconazole alone may prove inadequate. As the generalizability of our conclusions is necessarily limited, we recommend that infection control practitioners review local epidemiologic data regarding HCAC so that physicians can have the best available guidance.

Our study has several important limitations. Its retrospective nature exposes it to several forms of bias. The single center design limits the generalizability of our findings. Prospective, multicenter studies are needed to validate our results. Additionally, no universally accepted criteria exist to define HAI syndromes. Nonetheless, the criteria we employed have been used by others. We also lacked data on exposure to recent broad spectrum antimicrobials. Selection pressure via exposure to such agents is a risk factor for candidemia and without this data we cannot gauge the impact of this on our findings. Finally, we cannot control for the possibility that some patients were miscategorized. This could have arisen because of: (1) either limitations inherent in the definition of HCAC or (2) because the clinician delayed the decision to obtain blood cultures. Some patients classified as nosocomial may actually have had HCAC or community‐acquired diseasebut for some reason blood cultures were not drawn at time of admission but were deferred until later. Although a difficult issue to address in any study of the epidemiology of infection, the significance of this misclassification bias must be considered a significant concern.

In summary, Candidemia can be the cause of BSI presenting to the hospital. Moreover, HCAC represents a significant proportion of all Candidemia. Although patients with HCAC and nosocomial candidemia share select characteristics, there appear to be some differences in the microbiology of these syndromes.

Patients who leave a hospital against medical advice (AMA) have increased risks of negative clinical outcomes. Leaving a hospital AMA has been associated with increased risks of emergency department and hospital readmission, longer lengths of stay upon hospital readmission, increased risk of morbidities, and increased mortality.13 Ibrahim et al.4 estimated that 1.44% of all hospitalizations for adults in the United States end with the patient leaving AMA. Prior research, however, has shown that specific patient populations may experience greater AMA discharge rates, as up to 4.9% of asthma‐related hospitalizations,2 13% of human immunodeficiency virus (HIV)‐related hospitalizations,1 and 30% of psychiatric‐related hospitalizations have been shown to end with the patient leaving AMA.5

A small body of research examines the reasons that patients leave AMA. In prior studies, dissatisfaction with care and conflicts with medical staff were among the most commonly cited reasons given by patients.2, 6 In a study of healthcare provider reflections on recent patients who left AMA, providers identified patient mistrust, suboptimal physician‐patient communication, and physician‐patient conflict as important contributors to the patient leaving AMA.7

Sickle‐cell disease (SCD) is a painful genetic condition which in the United States affects mostly African Americans and leads to frequent hospital utilization. Patients with SCD frequently report having poor‐quality interpersonal relations with health care providers when the patient seeks treatment for his or her pain.8 Patients often report that their health care providers do not believe the patient's reports of pain, providers do not involve the patient to his or her satisfaction in setting the course of care, and providers typically stigmatize the patient as having a substance abuse problem.817 Indeed, a recent systematic review has shown that there is a high level of evidence that negative health care provider attitudes serve as a barrier to appropriate pain management in SCD.18

To our knowledge, the only study to date of hospital self‐discharge history among adults with SCD was conducted in the United Kingdom by Elander et al.19 These investigators found that 14% of their sample of patients reported ever having discharged themselves from a hospital. The most common reasons for this behavior given by these patients were that they grew tired of waiting for relief of their pain, there were other conflicts that occurred on the medical ward, and because they just simply wanted to go home.

Given the interpersonal conflicts and poor‐quality pain management often found in SCD care, additional examinations of hospital self‐discharge in this population are warranted. We hypothesized that SCD patient reports of interpersonal conflict during previous health care encounters would have an independent association with the likelihood of the patient having ever self‐discharged from a hospital. We also hypothesized that there would be an independent association between a patient's level of trust in the medical profession and their self‐discharge history. The aim of this study was to test these hypotheses.

Methods

Results

Discussion

In this study, we found that a high proportion of patients with SCD had a history of hospital self‐discharge. Patients with lower trust, and those who reported difficulty in persuading medical staff about sickle‐cell pain, were more likely to report having ever self‐discharged from a hospital, even after controlling for self‐reported hospital utilization for sickle‐cell pain, and the patient's 5‐year toxicology screen history. Because hospital self‐discharge is potentially dangerous,13 our study reveals an understudied aspect of how low trust and poor health care experiences may put patients with SCD at risk for poor outcomes.

In our study, 46.5% of our sample reported ever having self‐discharged from a hospital, which is much higher than the 14% found by Elander et al.19 in their United Kingdom (London‐based) sample. Other differences in our 2 patient populations may account for this discrepancy. Compared to the Elander sample, a much greater percentage of our patients reported ever having difficulty persuading medical staff about their pain (65% vs. 39%). As difficulty persuading medical staff about pain was independently associated with an increased hospital self‐discharge history in our study, one might expect that our sample, which had a higher percentage of patients reporting difficulty, would also be found to have a higher percentage of patients reporting a history of hospital self‐discharge. A second difference between the two patient samples is that our sample of patients experienced a greater number of hospital visits in the 12 months preceding the study compared to the Elander et al.19 sample. If this difference reflects an underlying difference in the overall hospital utilization experiences of the 2 groups, then our sample of adults would have greater opportunities, on average, than the Elander et al.19 sample to experience hospital self‐discharge. Other factors, such as patient behavioral or cultural differences between patients in the United Kingdom (with a national health system) and the United States (without a national health system), might be explored in future studies.

It is important to note that the wording of the hospital self‐discharge item as used both in our study and by Elander et al.19 would not only capture AMA discharges, but additionally may capture other sudden decisions about hospital discharge made by patients. A national‐level study of AMA discharges among adults with SCD in the United States that uses hospital records and/or chart review is needed in order to provide a more generalizable estimate of the prevalence of AMA discharge among this patient population in the United States.

Elander et al.19 suggest that while hospital self‐discharge among SCD patients may be interpreted by many as a sign that the patient engages in problematic use of opioids or other substances, it may be more appropriate to view this behavior as a sign that the patient has received inadequate management of their pain.19 In our study, hospital self‐discharge tended to be associated with having a history of substance abuse as operationalized by a positive toxicology screen for cocaine or marijuana use during any admission in the previous 5 years. Elander et al.19 found that hospital self‐discharge and other so‐called concern raising behaviors such as use of illicit substances were found to be significantly associated with patient attempts to obtain relief from their pain, but were not significantly associated with symptoms of actual substance dependence or addiction. For example, each instance of illicit substance use reported by the patients in the Elander et al.19 study described patient attempts to use marijuana in efforts to manage pain, to relax, or as alternatives to prescribed analgesics. Clinicians in the United States who observe positive toxicology screen results for SCD patients may see these results as casting doubt upon the legitimacy of the patient's pain reports, thus causing a reduction in the amount of pain medicine provided to the patient, when in fact, a substantial percentage of these results may reflect SCD patients attempts to manage their pain outside of a hospital setting. This potential discrepancy between clinician interpretations of the meaning of positive toxicology screen results for SCD patients, and the actual significance of these results for many patients as reflecting attempts to manage pain, could contribute to interpersonal conflicts between the clinicians and patients, and ultimately, patient self‐discharge and decreases in patient trust in clinicians. Further, to the extent that SCD patient positive toxicology screen results reflect use of illicit substances for reasons other than attempts to manage pain, this should signal for clinicians a need to refer the patient for substance abuse treatment and counseling in addition to (and not instead of) efforts to manage the patient's pain.

Our study is among the first to show empirically that persons with a history of hospital self‐discharge have lower levels of trust in the medical profession. Discharging himself or herself from a hospital could cause a patient to view future health care experiences in a more negative light, and cause the patient to have lower trust in the medical profession. Healthcare providers often label patients with a history of leaving AMA as challenging patients. Seeing in the medical record that a patient has left AMA before may bias the provider to view the patient in a more negative light, and consequently affect the quality of their communication with the patient, leading to lower patient trust. Alternatively, a patient could already possess lower trust in the medical profession due to poor‐quality interpersonal experiences, and thus be more likely to self‐discharge from a hospital during a future acute care visit due to a heightened wariness or greater level of anxiety.

The most consistent and robust predictors of trust found across studies in the literature are the quality of previous interactions with medical care.23 Poor physician communication, and experiences of conflict with staff have been associated with lower ratings of trust among a wide variety of patient populations.2427 Interestingly, we found a relationship between trust and hospital self‐discharge even after controlling for the quality of previous interpersonal experiences as measured by prior difficulty persuading staff about pain. Future studies should examine the relationship between trust and hospital self‐discharge history while controlling for other measures of previous interpersonal healthcare experiences among this patient population.

There are limitations to the current study that must be considered. First, as a single‐institution study, these results may not be generalizable to patients with SCD seeking care at other institutions. Also, we did not assess the actual reasons why patients chose to self‐discharge. Thus, while our data suggests that patient perceptions of poor‐quality care contributed to this behavior, we cannot state this definitively. The validity of a self‐reported annual hospital utilization measure as used in this study may be limited by inaccurate patient recall. However, we compared our patient's self‐report of annual hospital utilization with chart documented hospital utilization in the previous 12 months and found that the 2 measures were correlated in the appropriate direction, thereby giving us greater confidence in the validity of our self‐reported measure. Finally, the cross‐sectional nature of the data in this study makes it impossible to specify with certainty the causal directionality of the associations found here. Prospective research must be conducted to help tease apart the potentially complex relationships among trust, interpersonal experiences with care, and hospital self‐discharge.

Adults with SCD who have ever self‐discharged from a hospital have lower trust in the medical profession, and are more likely to report having had prior difficulty persuading medical staff about their sickle‐cell pain. The clinical consequences of hospital self‐discharge in this patient population must be examined, and the specific reasons behind this behavior must be further elucidated, so that clinicians and researchers may be able to design interventions to mitigate the occurrence of this potentially dangerous phenomenon.