Marin H. Kollef, MD

Rapid response systems (RRSs) have been developed to identify and treat deteriorating patients on general hospital units.[1] The most commonly proposed approach to the problem of identifying and stabilizing deteriorating hospitalized patients includes some combination of an early warning system to detect the deterioration and an RRS to deal with it. We previously demonstrated that a relatively simple hospital‐specific prediction model employing routine laboratory values and vital sign data is capable of predicting clinical deterioration, the need for intensive care unit (ICU) transfer, and hospital mortality in patients admitted to general medicine units.[2, 3, 4, 5, 6]

Hospital readmissions within 30 days of hospital discharge occur often and are difficult to predict. Starting in 2013, readmission penalties have been applied to specific conditions in the United States (acute myocardial infarction, heart failure, and pneumonia), with the expectation that additional conditions will be added to this group in years to come.[7, 8] Unfortunately, interventions developed to date have not been universally successful in preventing hospital readmissions for various medical conditions and patient types.[9] One potential explanation for this is the inability to reliably predict which patients are at risk for readmission to better target preventative interventions. Predictors of hospital readmission can be disease specific, such as the presence of multivessel disease in patients hospitalized with myocardial infarction,[10] or more general, such as lack of available medical follow‐up postdischarge.[11] Therefore, we performed a study to determine whether the occurrence of automated clinical deterioration alerts (CDAs) predicted 30‐day hospital readmission.

METHODS

RESULTS

The final cohort had 3015 patients with a mean age of 57.5 17.5 years and 47.8% males. The most common reasons for hospital admission were infection or sepsis syndrome including pneumonia and urinary tract infections (23.6%), congestive heart failure or other cardiac conditions (18.4%), respiratory distress including chronic obstructive pulmonary disease (16.2%), acute or chronic renal failure (9.7%), gastrointestinal disorders (8.4%), and diabetes mellitus management (7.4%). Overall, there were 567 (18.8%) patients who were readmitted within 30 days of their hospital discharge date.

Table 2 shows the characteristics of patients readmitted within 30 days and of patients not requiring hospital readmission within 30 days. Patients requiring hospital readmission within 30 days were younger and had significantly more comorbidities as manifested by significantly greater Charlson scores and individual comorbidities including coronary artery disease, congestive heart disease, peripheral vascular disease, connective tissue disease, cirrhosis, diabetes mellitus with end‐organ complications, renal failure, and metastatic cancer. Patients with a 30‐day readmission had significantly longer duration of hospitalization, more emergency department visits in the 6 months prior to the index hospitalization, lower minimum hemoglobin measurements, higher minimum serum creatinine values, and were more likely to have Medicare or Medicaid insurance compared to patients without a 30‐day readmission.

There were 1141 (34.4%) patients that triggered a CDA. Patients triggering a CDA were significantly more likely to have a 30‐day readmission compared to those who did not trigger a CDA (23.6% vs 15.9%; P < 0.001). Patients triggering a CDA were also significantly more likely to be readmitted within 60 days (31.7% vs 22.1%; P < 0.001) and 90 days (35.8% vs 26.2%; P < 0.001) compared to patients who did not trigger a CDA. Multiple logistic regression identified the triggering of a CDA to be independently associated with 30‐day readmission (OR: 1.40; 95% CI: 1.26‐1.55; P = 0.001) (Table 3). Other independent predictors of 30‐day readmission were: an emergency department visit in the previous 6 months, increasing age in 1‐year increments, presence of connective tissue disease, diabetes mellitus with end‐organ complications, chronic renal disease, cirrhosis, and metastatic cancer (Hosmer‐Lemeshow goodness of fit test, 0.363). Figure 1 reveals the ROC curves for the logistic regression model (Table 3) with and without the CDA variable. As the ROC curves document, the 2 models had similar sensitivity for the entire range of specificities. Reflecting this, the area under the ROC curve for the model inclusive of the CDA variable equaled 0.675 (95% CI: 0.649‐0.700), whereas the area under the ROC curve for the model excluding the CDA variable equaled 0.658 (95% CI: 0.632‐0.684).

Figure 1

DISCUSSION

We demonstrated that the occurrence of an automated CDA is associated with increased risk for 30‐day hospital readmission. However, the addition of the CDA variable to the other variables identified to be independently associated with 30‐day readmission (Table 3) did not significantly add to the overall predictive accuracy of the derived logistic regression model. Other investigators have previously attempted to develop automated predictors of hospital readmission. Amarasingham et al. developed a real‐time electronic predictive model that identifies hospitalized heart failure patients at high risk for readmission or death from clinical and nonclinical risk factors present on admission.[13] Their electronic model demonstrated good discrimination for 30‐day mortality and readmission and performed as well, or better than, models developed by the Center for Medicaid and Medicare Services and the Acute Decompensated Heart Failure Registry. Similarly, Baillie et al. developed an automated prediction model that was effectively integrated into an existing EMR and identified patients on admission who were at risk for readmission within 30 days of discharge.[14] Our automated CDA differs from these previous risk predictors by surveying patients throughout their hospital stay as opposed to identifying risk for readmission at a single time point.

Several limitations of our study should be recognized. First, this was a noninterventional study aimed at examining the ability of CDAs to predict hospital readmission. Future studies are needed to assess whether the use of enhanced readmission prediction algorithms can be utilized to avert hospital readmissions. Second, the data derive from a single center, and this necessarily limits the generalizability of our findings. As such, our results may not reflect what one might see at other institutions. For example, Barnes‐Jewish Hospital has a regional referral pattern that includes community hospitals, regional long‐term acute care hospitals, nursing homes, and chronic wound, dialysis, and infusion clinics. This may explain, in part, the relatively high rate of hospital readmission observed in our cohort. Third, there is the possibility that CDAs were associated with readmission by chance given the number of potential predictor variables examined. The importance of CDAs as a determinant of rehospitalization requires confirmation in other independent populations. Fourth, it is likely that we did not capture all hospital readmissions, primarily those occurring outside of our hospital system. Therefore, we may have underestimated the actual rates of readmission for this cohort. Finally, we cannot be certain that all important predictors of hospital readmission were captured in this study.

The development of an accurate real‐time early warning system has the potential to identify patients at risk for various adverse outcomes including clinical deterioration, hospital death, and postdischarge readmission. By identifying patients at greatest risk for readmission, valuable healthcare resources can be better targeted to such populations. Our findings suggest that existing readmission predictors may suboptimally risk‐stratify patients, and it may be important to include additional clinical variables if pay for performance and other across‐institution comparisons are to be fair to institutions that care for more seriously ill patients. The variables identified as predictors of 30‐day hospital readmission in our study, with the exception of a CDA, are all readily identifiable clinical characteristics. The modest incremental value of a CDA to these clinical characteristics suggests that they would suffice for the identification of patients at high risk for hospital readmission. This is especially important for safety‐net institutions not routinely employing automated CDAs. These safety‐net hospitals provide a disproportionate level of care for patients who otherwise would have difficulty obtaining inpatient medical care and disproportionately carry the greatest burden of hospital readmissions.[15]

Rapid response systems (RRSs) have been developed to identify and treat deteriorating patients on general hospital units.[1] The most commonly proposed approach to the problem of identifying and stabilizing deteriorating hospitalized patients includes some combination of an early warning system to detect the deterioration and an RRS to deal with it. We previously demonstrated that a relatively simple hospital‐specific prediction model employing routine laboratory values and vital sign data is capable of predicting clinical deterioration, the need for intensive care unit (ICU) transfer, and hospital mortality in patients admitted to general medicine units.[2, 3, 4, 5, 6]

Hospital readmissions within 30 days of hospital discharge occur often and are difficult to predict. Starting in 2013, readmission penalties have been applied to specific conditions in the United States (acute myocardial infarction, heart failure, and pneumonia), with the expectation that additional conditions will be added to this group in years to come.[7, 8] Unfortunately, interventions developed to date have not been universally successful in preventing hospital readmissions for various medical conditions and patient types.[9] One potential explanation for this is the inability to reliably predict which patients are at risk for readmission to better target preventative interventions. Predictors of hospital readmission can be disease specific, such as the presence of multivessel disease in patients hospitalized with myocardial infarction,[10] or more general, such as lack of available medical follow‐up postdischarge.[11] Therefore, we performed a study to determine whether the occurrence of automated clinical deterioration alerts (CDAs) predicted 30‐day hospital readmission.

METHODS

RESULTS

The final cohort had 3015 patients with a mean age of 57.5 17.5 years and 47.8% males. The most common reasons for hospital admission were infection or sepsis syndrome including pneumonia and urinary tract infections (23.6%), congestive heart failure or other cardiac conditions (18.4%), respiratory distress including chronic obstructive pulmonary disease (16.2%), acute or chronic renal failure (9.7%), gastrointestinal disorders (8.4%), and diabetes mellitus management (7.4%). Overall, there were 567 (18.8%) patients who were readmitted within 30 days of their hospital discharge date.

Table 2 shows the characteristics of patients readmitted within 30 days and of patients not requiring hospital readmission within 30 days. Patients requiring hospital readmission within 30 days were younger and had significantly more comorbidities as manifested by significantly greater Charlson scores and individual comorbidities including coronary artery disease, congestive heart disease, peripheral vascular disease, connective tissue disease, cirrhosis, diabetes mellitus with end‐organ complications, renal failure, and metastatic cancer. Patients with a 30‐day readmission had significantly longer duration of hospitalization, more emergency department visits in the 6 months prior to the index hospitalization, lower minimum hemoglobin measurements, higher minimum serum creatinine values, and were more likely to have Medicare or Medicaid insurance compared to patients without a 30‐day readmission.

There were 1141 (34.4%) patients that triggered a CDA. Patients triggering a CDA were significantly more likely to have a 30‐day readmission compared to those who did not trigger a CDA (23.6% vs 15.9%; P < 0.001). Patients triggering a CDA were also significantly more likely to be readmitted within 60 days (31.7% vs 22.1%; P < 0.001) and 90 days (35.8% vs 26.2%; P < 0.001) compared to patients who did not trigger a CDA. Multiple logistic regression identified the triggering of a CDA to be independently associated with 30‐day readmission (OR: 1.40; 95% CI: 1.26‐1.55; P = 0.001) (Table 3). Other independent predictors of 30‐day readmission were: an emergency department visit in the previous 6 months, increasing age in 1‐year increments, presence of connective tissue disease, diabetes mellitus with end‐organ complications, chronic renal disease, cirrhosis, and metastatic cancer (Hosmer‐Lemeshow goodness of fit test, 0.363). Figure 1 reveals the ROC curves for the logistic regression model (Table 3) with and without the CDA variable. As the ROC curves document, the 2 models had similar sensitivity for the entire range of specificities. Reflecting this, the area under the ROC curve for the model inclusive of the CDA variable equaled 0.675 (95% CI: 0.649‐0.700), whereas the area under the ROC curve for the model excluding the CDA variable equaled 0.658 (95% CI: 0.632‐0.684).

Figure 1

DISCUSSION

We demonstrated that the occurrence of an automated CDA is associated with increased risk for 30‐day hospital readmission. However, the addition of the CDA variable to the other variables identified to be independently associated with 30‐day readmission (Table 3) did not significantly add to the overall predictive accuracy of the derived logistic regression model. Other investigators have previously attempted to develop automated predictors of hospital readmission. Amarasingham et al. developed a real‐time electronic predictive model that identifies hospitalized heart failure patients at high risk for readmission or death from clinical and nonclinical risk factors present on admission.[13] Their electronic model demonstrated good discrimination for 30‐day mortality and readmission and performed as well, or better than, models developed by the Center for Medicaid and Medicare Services and the Acute Decompensated Heart Failure Registry. Similarly, Baillie et al. developed an automated prediction model that was effectively integrated into an existing EMR and identified patients on admission who were at risk for readmission within 30 days of discharge.[14] Our automated CDA differs from these previous risk predictors by surveying patients throughout their hospital stay as opposed to identifying risk for readmission at a single time point.

Several limitations of our study should be recognized. First, this was a noninterventional study aimed at examining the ability of CDAs to predict hospital readmission. Future studies are needed to assess whether the use of enhanced readmission prediction algorithms can be utilized to avert hospital readmissions. Second, the data derive from a single center, and this necessarily limits the generalizability of our findings. As such, our results may not reflect what one might see at other institutions. For example, Barnes‐Jewish Hospital has a regional referral pattern that includes community hospitals, regional long‐term acute care hospitals, nursing homes, and chronic wound, dialysis, and infusion clinics. This may explain, in part, the relatively high rate of hospital readmission observed in our cohort. Third, there is the possibility that CDAs were associated with readmission by chance given the number of potential predictor variables examined. The importance of CDAs as a determinant of rehospitalization requires confirmation in other independent populations. Fourth, it is likely that we did not capture all hospital readmissions, primarily those occurring outside of our hospital system. Therefore, we may have underestimated the actual rates of readmission for this cohort. Finally, we cannot be certain that all important predictors of hospital readmission were captured in this study.

The development of an accurate real‐time early warning system has the potential to identify patients at risk for various adverse outcomes including clinical deterioration, hospital death, and postdischarge readmission. By identifying patients at greatest risk for readmission, valuable healthcare resources can be better targeted to such populations. Our findings suggest that existing readmission predictors may suboptimally risk‐stratify patients, and it may be important to include additional clinical variables if pay for performance and other across‐institution comparisons are to be fair to institutions that care for more seriously ill patients. The variables identified as predictors of 30‐day hospital readmission in our study, with the exception of a CDA, are all readily identifiable clinical characteristics. The modest incremental value of a CDA to these clinical characteristics suggests that they would suffice for the identification of patients at high risk for hospital readmission. This is especially important for safety‐net institutions not routinely employing automated CDAs. These safety‐net hospitals provide a disproportionate level of care for patients who otherwise would have difficulty obtaining inpatient medical care and disproportionately carry the greatest burden of hospital readmissions.[15]

Rapid response systems (RRSs) have been developed to identify and treat deteriorating patients on general hospital units.[1] The most commonly proposed approach to the problem of identifying and stabilizing deteriorating hospitalized patients includes some combination of an early warning system to detect the deterioration and an RRS to deal with it. We previously demonstrated that a relatively simple hospital‐specific prediction model employing routine laboratory values and vital sign data is capable of predicting clinical deterioration, the need for intensive care unit (ICU) transfer, and hospital mortality in patients admitted to general medicine units.[2, 3, 4, 5, 6]

Hospital readmissions within 30 days of hospital discharge occur often and are difficult to predict. Starting in 2013, readmission penalties have been applied to specific conditions in the United States (acute myocardial infarction, heart failure, and pneumonia), with the expectation that additional conditions will be added to this group in years to come.[7, 8] Unfortunately, interventions developed to date have not been universally successful in preventing hospital readmissions for various medical conditions and patient types.[9] One potential explanation for this is the inability to reliably predict which patients are at risk for readmission to better target preventative interventions. Predictors of hospital readmission can be disease specific, such as the presence of multivessel disease in patients hospitalized with myocardial infarction,[10] or more general, such as lack of available medical follow‐up postdischarge.[11] Therefore, we performed a study to determine whether the occurrence of automated clinical deterioration alerts (CDAs) predicted 30‐day hospital readmission.

METHODS

RESULTS

The final cohort had 3015 patients with a mean age of 57.5 17.5 years and 47.8% males. The most common reasons for hospital admission were infection or sepsis syndrome including pneumonia and urinary tract infections (23.6%), congestive heart failure or other cardiac conditions (18.4%), respiratory distress including chronic obstructive pulmonary disease (16.2%), acute or chronic renal failure (9.7%), gastrointestinal disorders (8.4%), and diabetes mellitus management (7.4%). Overall, there were 567 (18.8%) patients who were readmitted within 30 days of their hospital discharge date.

Table 2 shows the characteristics of patients readmitted within 30 days and of patients not requiring hospital readmission within 30 days. Patients requiring hospital readmission within 30 days were younger and had significantly more comorbidities as manifested by significantly greater Charlson scores and individual comorbidities including coronary artery disease, congestive heart disease, peripheral vascular disease, connective tissue disease, cirrhosis, diabetes mellitus with end‐organ complications, renal failure, and metastatic cancer. Patients with a 30‐day readmission had significantly longer duration of hospitalization, more emergency department visits in the 6 months prior to the index hospitalization, lower minimum hemoglobin measurements, higher minimum serum creatinine values, and were more likely to have Medicare or Medicaid insurance compared to patients without a 30‐day readmission.

There were 1141 (34.4%) patients that triggered a CDA. Patients triggering a CDA were significantly more likely to have a 30‐day readmission compared to those who did not trigger a CDA (23.6% vs 15.9%; P < 0.001). Patients triggering a CDA were also significantly more likely to be readmitted within 60 days (31.7% vs 22.1%; P < 0.001) and 90 days (35.8% vs 26.2%; P < 0.001) compared to patients who did not trigger a CDA. Multiple logistic regression identified the triggering of a CDA to be independently associated with 30‐day readmission (OR: 1.40; 95% CI: 1.26‐1.55; P = 0.001) (Table 3). Other independent predictors of 30‐day readmission were: an emergency department visit in the previous 6 months, increasing age in 1‐year increments, presence of connective tissue disease, diabetes mellitus with end‐organ complications, chronic renal disease, cirrhosis, and metastatic cancer (Hosmer‐Lemeshow goodness of fit test, 0.363). Figure 1 reveals the ROC curves for the logistic regression model (Table 3) with and without the CDA variable. As the ROC curves document, the 2 models had similar sensitivity for the entire range of specificities. Reflecting this, the area under the ROC curve for the model inclusive of the CDA variable equaled 0.675 (95% CI: 0.649‐0.700), whereas the area under the ROC curve for the model excluding the CDA variable equaled 0.658 (95% CI: 0.632‐0.684).

Figure 1

DISCUSSION

We demonstrated that the occurrence of an automated CDA is associated with increased risk for 30‐day hospital readmission. However, the addition of the CDA variable to the other variables identified to be independently associated with 30‐day readmission (Table 3) did not significantly add to the overall predictive accuracy of the derived logistic regression model. Other investigators have previously attempted to develop automated predictors of hospital readmission. Amarasingham et al. developed a real‐time electronic predictive model that identifies hospitalized heart failure patients at high risk for readmission or death from clinical and nonclinical risk factors present on admission.[13] Their electronic model demonstrated good discrimination for 30‐day mortality and readmission and performed as well, or better than, models developed by the Center for Medicaid and Medicare Services and the Acute Decompensated Heart Failure Registry. Similarly, Baillie et al. developed an automated prediction model that was effectively integrated into an existing EMR and identified patients on admission who were at risk for readmission within 30 days of discharge.[14] Our automated CDA differs from these previous risk predictors by surveying patients throughout their hospital stay as opposed to identifying risk for readmission at a single time point.

Several limitations of our study should be recognized. First, this was a noninterventional study aimed at examining the ability of CDAs to predict hospital readmission. Future studies are needed to assess whether the use of enhanced readmission prediction algorithms can be utilized to avert hospital readmissions. Second, the data derive from a single center, and this necessarily limits the generalizability of our findings. As such, our results may not reflect what one might see at other institutions. For example, Barnes‐Jewish Hospital has a regional referral pattern that includes community hospitals, regional long‐term acute care hospitals, nursing homes, and chronic wound, dialysis, and infusion clinics. This may explain, in part, the relatively high rate of hospital readmission observed in our cohort. Third, there is the possibility that CDAs were associated with readmission by chance given the number of potential predictor variables examined. The importance of CDAs as a determinant of rehospitalization requires confirmation in other independent populations. Fourth, it is likely that we did not capture all hospital readmissions, primarily those occurring outside of our hospital system. Therefore, we may have underestimated the actual rates of readmission for this cohort. Finally, we cannot be certain that all important predictors of hospital readmission were captured in this study.

The development of an accurate real‐time early warning system has the potential to identify patients at risk for various adverse outcomes including clinical deterioration, hospital death, and postdischarge readmission. By identifying patients at greatest risk for readmission, valuable healthcare resources can be better targeted to such populations. Our findings suggest that existing readmission predictors may suboptimally risk‐stratify patients, and it may be important to include additional clinical variables if pay for performance and other across‐institution comparisons are to be fair to institutions that care for more seriously ill patients. The variables identified as predictors of 30‐day hospital readmission in our study, with the exception of a CDA, are all readily identifiable clinical characteristics. The modest incremental value of a CDA to these clinical characteristics suggests that they would suffice for the identification of patients at high risk for hospital readmission. This is especially important for safety‐net institutions not routinely employing automated CDAs. These safety‐net hospitals provide a disproportionate level of care for patients who otherwise would have difficulty obtaining inpatient medical care and disproportionately carry the greatest burden of hospital readmissions.[15]

Among hospitalized patients with serious infections, the choice of empiric therapy plays a key role in outcomes.[1, 2, 3, 4, 5, 6, 7, 8, 9] Rising rates and variable patterns of antimicrobial resistance, however, complicate selecting appropriate empiric therapy. Amidst this shifting landscape of resistance to antimicrobials, gram‐negative bacteria and specifically Acinetobacter baumannii (AB), remain a considerable challenge.[10] On the one hand, AB is a less‐frequent cause of serious infections than organisms like Pseudomonas aeruginosa or Enterobacteriaceae in severely ill hospitalized patients.[11, 12] On the other, AB has evolved a variety of resistance mechanisms and exhibits unpredictable susceptibility patterns.[13] These factors combine to increase the likelihood of administering inappropriate empiric therapy when faced with an infection caused by AB and, thereby, raising the risk of death.[14] The fact that clinicians may not routinely consider AB as the potential culprit pathogen in the patient they are treating along with this organism's highly in vitro resistant nature, may result in routine gram‐negative coverage being frequently inadequate for AB infections.

To address the poor outcomes related to inappropriate empiric therapy in the setting of AB, one requires an appreciation of the longitudinal changes and geographic differences in the susceptibility of this pathogen. Thus, we aimed to examine secular trends in the resistance of AB to antimicrobial agents whose effectiveness against this microorganism was well supported in the literature during the study timeframe.[15]

METHODS

To determine the prevalence of predefined resistance patterns among AB in respiratory and blood stream infection (BSI) specimens, we examined The Surveillance Network (TSN) database from Eurofins. We explored data collected between years 2003 and 2012. The database has been used extensively for surveillance purposes since 1994, and has previously been described in detail.[16, 17, 18, 19, 20] Briefly, TSN is a warehouse of routine clinical microbiology data collected from a nationally representative sample of microbiology laboratories in 217 hospitals in the United States. To minimize selection bias, laboratories are included based on their geography and the demographics of the populations they serve.[18] Only clinically significant samples are reported. No personal identifying information for source patients is available in this database. Only source laboratories that perform antimicrobial susceptibility testing according standard Food and Drug Administrationapproved testing methods and that interpret susceptibility in accordance with the Clinical Laboratory Standards Institute breakpoints are included.[21] (See Supporting Table 4 in the online version of this article for minimum inhibitory concentration (MIC) changes over the course of the studycurrent colistin and polymyxin breakpoints applied retrospectively). All enrolled laboratories undergo a pre‐enrollment site visit. Logical filters are used for routine quality control to detect unusual susceptibility profiles and to ensure appropriate testing methods. Repeat testing and reporting are done as necessary.[18]

Laboratory samples are reported as susceptible, intermediate, or resistant. We grouped isolates with intermediate MICs together with the resistant ones for the purposes of the current analysis. Duplicate isolates were excluded. Only samples representing 1 of the 2 infections of interest, respiratory or BSI, were included.

We examined 3 time periods2003 to 2005, 2006 to 2008, and 2009 to 2012for the prevalence of AB's resistance to the following antibiotics: carbapenems (imipenem, meropenem, doripenem), aminoglycosides (tobramycin, amikacin), tetracyclines (minocycline, doxycycline), polymyxins (colistin, polymyxin B), ampicillin‐sulbactam, and trimethoprim‐sulfamethoxazole. Antimicrobial resistance was defined by the designation of intermediate or resistant in the susceptibility category. Resistance to a class of antibiotics was defined as resistance to all drugs within the class for which testing was available. The organism was multidrug resistant (MDR) if it was resistant to at least 1 antimicrobial in at least 3 drug classes examined.[22] Resistance to a combination of 2 drugs was present if the specimen was resistant to both of the drugs in the combination for which testing was available. We examined the data by infection type, time period, the 9 US Census divisions, and location of origin of the sample.

All categorical variables are reported as percentages. Continuous variables are reported as meansstandard deviations and/or medians with the interquartile range (IQR). We did not pursue hypothesis testing due to a high risk of type I error in this large dataset. Therefore, only clinically important trends are highlighted.

RESULTS

Among the 39,320 AB specimens, 81.1% were derived from a respiratory source and 18.9% represented BSI. Demographics of source patients are listed in Table 1. Notably, the median age of those with respiratory infection (58 years; IQR 38, 73) was higher than among patients with BSI (54.5 years; IQR 36, 71), and there were proportionally fewer females among respiratory patients (39.9%) than those with BSI (46.0%). Though only 24.3% of all BSI samples originated from the intensive are unit (ICU), 40.5% of respiratory specimens came from that location. The plurality of all specimens was collected in the 2003 to 2005 time interval (41.3%), followed by 2006 to 2008 (34.7%), with a minority coming from years 2009 to 2012 (24.0%). The proportions of collected specimens from respiratory and BSI sources were similar in all time periods examined (Table 1). Geographically, the South Atlantic division contributed the most samples (24.1%) and East South Central the fewest (2.6%) (Figure 1). The vast majority of all samples came from hospital wards (78.6%), where roughly one‐half originated in the ICU (37.5%). Fewer still came from outpatient sources (18.3%), and a small minority (2.5%) from nursing homes.

Figure 1

Figure 2 depicts overall resistance patterns by individual drugs, drug classes, and frequently used combinations of agents. Although doripenem had the highest rate of resistance numerically (90.3%), its susceptibility was tested only in a small minority of specimens (n=31, 0.08%). Resistance to trimethoprim‐sulfamethoxazole was high (55.3%) based on a large number of samples tested (n=33,031). Conversely, colistin as an agent and polymyxins as a class exhibited the highest susceptibility rates of over 90%, though the numbers of samples tested for susceptibility to these drugs were also small (colistin n=2,086, 5.3%; polymyxins n=3,120, 7.9%) (Figure 2). Among commonly used drug combinations, carbapenem+aminoglycoside (18.0%) had the lowest resistance rates, and nearly 30% of all AB specimens tested met the criteria for MDR.

Figure 2

Over time, resistance to carbapenems more‐than doubled, from 21.0% in 2003 to 2005 to 47.9% in 2009 to 2012 (Table 2). Although relatively few samples were tested for colistin susceptibility (n=2,086, 5.3%), resistance to this drug also more than doubled from 2.8% (95% confidence interval: 1.9‐4.2) in 2006 to 2008 to 6.9% (95% confidence interval: 5.7‐8.2) in 2009 to 2012. As a class, however, polymyxins exhibited stable resistance rates over the time frame of the study (Table 2). Prevalence of MDR AB rose from 21.4% in 2003 to 2005 to 33.7% in 2006 to 2008, and remained stable at 35.2% in 2009 to 2012. Resistance to even such broad combinations as carbapenem+ampicillin/sulbactam nearly tripled from 13.2% in 2003 to 2005 to 35.5% in 2009 to 2012. Notably, between 2003 and 2012, although resistance rates either rose or remained stable to all other agents, those to minocycline diminished from 56.5% in 2003 to 2005 to 36.6% in 2006 to 2008 to 30.5% in 2009 to 2012. (See Supporting Table 1 in the online version of this article for time trends based on whether they represented respiratory or BSI specimens, with directionally similar trends in both.)

Regionally, examining resistance by classes and combinations of antibiotics, trimethoprim‐sulfamethoxazole exhibited consistently the highest rates of resistance, ranging from the lowest in the New England (28.8%) to the highest in the East North Central (69.9%) Census divisions (See Supporting Table 2 in the online version of this article). The rates of resistance to tetracyclines ranged from 0.0% in New England to 52.6% in the Mountain division, and to polymyxins from 0.0% in the East South Central division to 23.4% in New England. Generally, New England enjoyed the lowest rates of resistance (0.0% to tetracyclines to 28.8% to trimethoprim‐sulfamethoxazole), and the Mountain division the highest (0.9% to polymyxins to 52.6% to tetracyclines). The rates of MDR AB ranged from 8.0% in New England to 50.4% in the Mountain division (see Supporting Table 2 in the online version of this article).

Examining resistances to drug classes and combinations by the location of the source specimen revealed that trimethoprim‐sulfamethoxazole once again exhibited the highest rate of resistance across all locations (see Supporting Table 3 in the online version of this article). Despite their modest contribution to the overall sample pool (n=967, 2.5%), organisms from nursing home subjects had the highest prevalence of resistance to aminoglycosides (36.3%), tetracyclines (57.1%), and carbapenems (47.1%). This pattern held true for combination regimens examined. Nursing homes also vastly surpassed other locations in the rates of MDR AB (46.5%). Interestingly, the rates of MDR did not differ substantially among regular inpatient wards (29.2%), the ICU (28.7%), and outpatient locations (26.2%) (see Supporting Table 3 in the online version of this article).

DISCUSSION

In this large multicenter survey we have documented the rising rates of AB resistance to clinically important antimicrobials in the United States. On the whole, all antimicrobials, except for minocycline, exhibited either large or small increases in resistance. Alarmingly, even colistin, a true last resort AB treatment, lost a considerable amount of activity against AB, with the resistance rate rising from 2.8% in 2006 to 2008 to 6.9% in 2009 to 2012. The single encouraging trend that we observed was that resistance to minocycline appeared to diminish substantially, going from over one‐half of all AB tested in 2003 to 2005 to just under one‐third in 2009 to 2012.

Although we did note a rise in the MDR AB, our data suggest a lower percentage of all AB that meets the MDR phenotype criteria compared to reports by other groups. For example, the Center for Disease Dynamics and Economic Policy (CDDEP), analyzing the same data as our study, reports a rise in MDR AB from 32.1% in 1999 to 51.0% in 2010.[23] This discrepancy is easily explained by the fact that we included polymyxins, tetracyclines, and trimethoprim‐sulfamethoxazole in our evaluation, whereas the CDDEP did not examine these agents. Furthermore, we omitted fluoroquinolones, a drug class with high rates of resistance, from our study, because we were interested in focusing only on antimicrobials with clinical data in AB infections.[22] In addition, we limited our evaluation to specimens derived from respiratory or BSI sources, whereas the CDDEP data reflect any AB isolate present in TSN.

We additionally confirm that there is substantial geographic variation in resistance patterns. Thus, despite different definitions, our data agree with those from the CDDEP that the MDR prevalence is highest in the Mountain and East North Central divisions, and lowest in New England overall.[23] The wide variations underscore the fact that it is not valid to speak of national rates of resistance, but rather it is important to concentrate on the local patterns. This information, though important from the macroepidemiologic standpoint, is likely still not granular enough to help clinicians make empiric treatment decisions. In fact, what is needed for that is real‐time antibiogram data specific to each center and even each unit within each center.

The latter point is further illustrated by our analysis of locations of origin of the specimens. In this analysis, we discovered that, contrary to the common presumption that the ICU has the highest rate of resistant organisms, specimens derived from nursing homes represent perhaps the most intensely resistant organisms. In other words, the nursing home is the setting most likely to harbor patients with respiratory infections and BSIs caused by resistant AB. These data are in agreement with several other recent investigations. In a period‐prevalence survey conducted in the state of Maryland in 2009 by Thom and colleagues, long‐term care facilities were found to have the highest prevalence of any AB, and also those resistant to imipenem, MDR, and extensively drug‐resistant organisms.[24] Mortensen and coworkers confirmed the high prevalence of AB and AB resistance in long‐term care facilities, and extended this finding to suggest that there is evidence for intra‐ and interhospital spread of these pathogens.[25] Our data confirm this concerning finding at the national level, and point to a potential area of intervention for infection prevention.

An additional finding of some concern is that the highest proportion of colistin resistance among those specimens, whose location of origin was reported in the database, was the outpatient setting (6.6% compared to 5.4% in the ICU specimens, for example). Although these infections would likely meet the definition for healthcare‐associated infection, AB as a community‐acquired respiratory pathogen is not unprecedented either in the United States or abroad.[26, 27, 28, 29, 30] It is, however, reassuring that most other antimicrobials examined in our study exhibit higher rates of susceptibility in the specimens derived from the outpatient settings than either from the hospital or the nursing home.

Our study has a number of strengths. As a large multicenter survey, it is representative of AB susceptibility patterns across the United States, which makes it highly generalizable. We focused on antibiotics for which clinical evidence is available, thus adding a practical dimension to the results. Another pragmatic consideration is examining the data by geographic distributions, allowing an additional layer of granularity for clinical decisions. At the same time it suffers from some limitations. The TSN database consists of microbiology samples from hospital laboratories. Although we attempted to reduce the risk of duplication, because of how samples are numbered in the database, repeat sampling remains a possibility. Despite having stratified the data by geography and the location of origin of the specimen, it is likely not granular enough for local risk stratification decisions clinicians make daily about the choices of empiric therapy. Some of the MIC breakpoints have changed over the period of the study (see Supporting Table 4 in the online version of this article). Because these changes occurred in the last year of data collection (2012), they should have had only a minimal, if any, impact on the observed rates of resistance in the time frame examined. Additionally, because resistance rates evolve rapidly, more current data are required for effective clinical decision making.

In summary, we have demonstrated that the last decade has seen an alarming increase in the rate of resistance of AB to multiple clinically important antimicrobial agents and classes. We have further emphasized the importance of granularity in susceptibility data to help clinicians make sensible decisions about empiric therapy in hospitalized patients with serious infections. Finally, and potentially most disturbingly, the nursing home as a location appears to be a robust reservoir for spread for resistant AB. All of these observations highlight the urgent need to develop novel antibiotics and nontraditional agents, such as antibodies and vaccines, to combat AB infections, in addition to having important infection prevention implications if we are to contain the looming threat of the end of antibiotics.[31]

Among hospitalized patients with serious infections, the choice of empiric therapy plays a key role in outcomes.[1, 2, 3, 4, 5, 6, 7, 8, 9] Rising rates and variable patterns of antimicrobial resistance, however, complicate selecting appropriate empiric therapy. Amidst this shifting landscape of resistance to antimicrobials, gram‐negative bacteria and specifically Acinetobacter baumannii (AB), remain a considerable challenge.[10] On the one hand, AB is a less‐frequent cause of serious infections than organisms like Pseudomonas aeruginosa or Enterobacteriaceae in severely ill hospitalized patients.[11, 12] On the other, AB has evolved a variety of resistance mechanisms and exhibits unpredictable susceptibility patterns.[13] These factors combine to increase the likelihood of administering inappropriate empiric therapy when faced with an infection caused by AB and, thereby, raising the risk of death.[14] The fact that clinicians may not routinely consider AB as the potential culprit pathogen in the patient they are treating along with this organism's highly in vitro resistant nature, may result in routine gram‐negative coverage being frequently inadequate for AB infections.

To address the poor outcomes related to inappropriate empiric therapy in the setting of AB, one requires an appreciation of the longitudinal changes and geographic differences in the susceptibility of this pathogen. Thus, we aimed to examine secular trends in the resistance of AB to antimicrobial agents whose effectiveness against this microorganism was well supported in the literature during the study timeframe.[15]

METHODS

To determine the prevalence of predefined resistance patterns among AB in respiratory and blood stream infection (BSI) specimens, we examined The Surveillance Network (TSN) database from Eurofins. We explored data collected between years 2003 and 2012. The database has been used extensively for surveillance purposes since 1994, and has previously been described in detail.[16, 17, 18, 19, 20] Briefly, TSN is a warehouse of routine clinical microbiology data collected from a nationally representative sample of microbiology laboratories in 217 hospitals in the United States. To minimize selection bias, laboratories are included based on their geography and the demographics of the populations they serve.[18] Only clinically significant samples are reported. No personal identifying information for source patients is available in this database. Only source laboratories that perform antimicrobial susceptibility testing according standard Food and Drug Administrationapproved testing methods and that interpret susceptibility in accordance with the Clinical Laboratory Standards Institute breakpoints are included.[21] (See Supporting Table 4 in the online version of this article for minimum inhibitory concentration (MIC) changes over the course of the studycurrent colistin and polymyxin breakpoints applied retrospectively). All enrolled laboratories undergo a pre‐enrollment site visit. Logical filters are used for routine quality control to detect unusual susceptibility profiles and to ensure appropriate testing methods. Repeat testing and reporting are done as necessary.[18]

Laboratory samples are reported as susceptible, intermediate, or resistant. We grouped isolates with intermediate MICs together with the resistant ones for the purposes of the current analysis. Duplicate isolates were excluded. Only samples representing 1 of the 2 infections of interest, respiratory or BSI, were included.

We examined 3 time periods2003 to 2005, 2006 to 2008, and 2009 to 2012for the prevalence of AB's resistance to the following antibiotics: carbapenems (imipenem, meropenem, doripenem), aminoglycosides (tobramycin, amikacin), tetracyclines (minocycline, doxycycline), polymyxins (colistin, polymyxin B), ampicillin‐sulbactam, and trimethoprim‐sulfamethoxazole. Antimicrobial resistance was defined by the designation of intermediate or resistant in the susceptibility category. Resistance to a class of antibiotics was defined as resistance to all drugs within the class for which testing was available. The organism was multidrug resistant (MDR) if it was resistant to at least 1 antimicrobial in at least 3 drug classes examined.[22] Resistance to a combination of 2 drugs was present if the specimen was resistant to both of the drugs in the combination for which testing was available. We examined the data by infection type, time period, the 9 US Census divisions, and location of origin of the sample.

All categorical variables are reported as percentages. Continuous variables are reported as meansstandard deviations and/or medians with the interquartile range (IQR). We did not pursue hypothesis testing due to a high risk of type I error in this large dataset. Therefore, only clinically important trends are highlighted.

RESULTS

Among the 39,320 AB specimens, 81.1% were derived from a respiratory source and 18.9% represented BSI. Demographics of source patients are listed in Table 1. Notably, the median age of those with respiratory infection (58 years; IQR 38, 73) was higher than among patients with BSI (54.5 years; IQR 36, 71), and there were proportionally fewer females among respiratory patients (39.9%) than those with BSI (46.0%). Though only 24.3% of all BSI samples originated from the intensive are unit (ICU), 40.5% of respiratory specimens came from that location. The plurality of all specimens was collected in the 2003 to 2005 time interval (41.3%), followed by 2006 to 2008 (34.7%), with a minority coming from years 2009 to 2012 (24.0%). The proportions of collected specimens from respiratory and BSI sources were similar in all time periods examined (Table 1). Geographically, the South Atlantic division contributed the most samples (24.1%) and East South Central the fewest (2.6%) (Figure 1). The vast majority of all samples came from hospital wards (78.6%), where roughly one‐half originated in the ICU (37.5%). Fewer still came from outpatient sources (18.3%), and a small minority (2.5%) from nursing homes.

Figure 1

Figure 2 depicts overall resistance patterns by individual drugs, drug classes, and frequently used combinations of agents. Although doripenem had the highest rate of resistance numerically (90.3%), its susceptibility was tested only in a small minority of specimens (n=31, 0.08%). Resistance to trimethoprim‐sulfamethoxazole was high (55.3%) based on a large number of samples tested (n=33,031). Conversely, colistin as an agent and polymyxins as a class exhibited the highest susceptibility rates of over 90%, though the numbers of samples tested for susceptibility to these drugs were also small (colistin n=2,086, 5.3%; polymyxins n=3,120, 7.9%) (Figure 2). Among commonly used drug combinations, carbapenem+aminoglycoside (18.0%) had the lowest resistance rates, and nearly 30% of all AB specimens tested met the criteria for MDR.

Figure 2

Over time, resistance to carbapenems more‐than doubled, from 21.0% in 2003 to 2005 to 47.9% in 2009 to 2012 (Table 2). Although relatively few samples were tested for colistin susceptibility (n=2,086, 5.3%), resistance to this drug also more than doubled from 2.8% (95% confidence interval: 1.9‐4.2) in 2006 to 2008 to 6.9% (95% confidence interval: 5.7‐8.2) in 2009 to 2012. As a class, however, polymyxins exhibited stable resistance rates over the time frame of the study (Table 2). Prevalence of MDR AB rose from 21.4% in 2003 to 2005 to 33.7% in 2006 to 2008, and remained stable at 35.2% in 2009 to 2012. Resistance to even such broad combinations as carbapenem+ampicillin/sulbactam nearly tripled from 13.2% in 2003 to 2005 to 35.5% in 2009 to 2012. Notably, between 2003 and 2012, although resistance rates either rose or remained stable to all other agents, those to minocycline diminished from 56.5% in 2003 to 2005 to 36.6% in 2006 to 2008 to 30.5% in 2009 to 2012. (See Supporting Table 1 in the online version of this article for time trends based on whether they represented respiratory or BSI specimens, with directionally similar trends in both.)

Regionally, examining resistance by classes and combinations of antibiotics, trimethoprim‐sulfamethoxazole exhibited consistently the highest rates of resistance, ranging from the lowest in the New England (28.8%) to the highest in the East North Central (69.9%) Census divisions (See Supporting Table 2 in the online version of this article). The rates of resistance to tetracyclines ranged from 0.0% in New England to 52.6% in the Mountain division, and to polymyxins from 0.0% in the East South Central division to 23.4% in New England. Generally, New England enjoyed the lowest rates of resistance (0.0% to tetracyclines to 28.8% to trimethoprim‐sulfamethoxazole), and the Mountain division the highest (0.9% to polymyxins to 52.6% to tetracyclines). The rates of MDR AB ranged from 8.0% in New England to 50.4% in the Mountain division (see Supporting Table 2 in the online version of this article).

Examining resistances to drug classes and combinations by the location of the source specimen revealed that trimethoprim‐sulfamethoxazole once again exhibited the highest rate of resistance across all locations (see Supporting Table 3 in the online version of this article). Despite their modest contribution to the overall sample pool (n=967, 2.5%), organisms from nursing home subjects had the highest prevalence of resistance to aminoglycosides (36.3%), tetracyclines (57.1%), and carbapenems (47.1%). This pattern held true for combination regimens examined. Nursing homes also vastly surpassed other locations in the rates of MDR AB (46.5%). Interestingly, the rates of MDR did not differ substantially among regular inpatient wards (29.2%), the ICU (28.7%), and outpatient locations (26.2%) (see Supporting Table 3 in the online version of this article).

DISCUSSION

In this large multicenter survey we have documented the rising rates of AB resistance to clinically important antimicrobials in the United States. On the whole, all antimicrobials, except for minocycline, exhibited either large or small increases in resistance. Alarmingly, even colistin, a true last resort AB treatment, lost a considerable amount of activity against AB, with the resistance rate rising from 2.8% in 2006 to 2008 to 6.9% in 2009 to 2012. The single encouraging trend that we observed was that resistance to minocycline appeared to diminish substantially, going from over one‐half of all AB tested in 2003 to 2005 to just under one‐third in 2009 to 2012.

Although we did note a rise in the MDR AB, our data suggest a lower percentage of all AB that meets the MDR phenotype criteria compared to reports by other groups. For example, the Center for Disease Dynamics and Economic Policy (CDDEP), analyzing the same data as our study, reports a rise in MDR AB from 32.1% in 1999 to 51.0% in 2010.[23] This discrepancy is easily explained by the fact that we included polymyxins, tetracyclines, and trimethoprim‐sulfamethoxazole in our evaluation, whereas the CDDEP did not examine these agents. Furthermore, we omitted fluoroquinolones, a drug class with high rates of resistance, from our study, because we were interested in focusing only on antimicrobials with clinical data in AB infections.[22] In addition, we limited our evaluation to specimens derived from respiratory or BSI sources, whereas the CDDEP data reflect any AB isolate present in TSN.

We additionally confirm that there is substantial geographic variation in resistance patterns. Thus, despite different definitions, our data agree with those from the CDDEP that the MDR prevalence is highest in the Mountain and East North Central divisions, and lowest in New England overall.[23] The wide variations underscore the fact that it is not valid to speak of national rates of resistance, but rather it is important to concentrate on the local patterns. This information, though important from the macroepidemiologic standpoint, is likely still not granular enough to help clinicians make empiric treatment decisions. In fact, what is needed for that is real‐time antibiogram data specific to each center and even each unit within each center.

The latter point is further illustrated by our analysis of locations of origin of the specimens. In this analysis, we discovered that, contrary to the common presumption that the ICU has the highest rate of resistant organisms, specimens derived from nursing homes represent perhaps the most intensely resistant organisms. In other words, the nursing home is the setting most likely to harbor patients with respiratory infections and BSIs caused by resistant AB. These data are in agreement with several other recent investigations. In a period‐prevalence survey conducted in the state of Maryland in 2009 by Thom and colleagues, long‐term care facilities were found to have the highest prevalence of any AB, and also those resistant to imipenem, MDR, and extensively drug‐resistant organisms.[24] Mortensen and coworkers confirmed the high prevalence of AB and AB resistance in long‐term care facilities, and extended this finding to suggest that there is evidence for intra‐ and interhospital spread of these pathogens.[25] Our data confirm this concerning finding at the national level, and point to a potential area of intervention for infection prevention.

An additional finding of some concern is that the highest proportion of colistin resistance among those specimens, whose location of origin was reported in the database, was the outpatient setting (6.6% compared to 5.4% in the ICU specimens, for example). Although these infections would likely meet the definition for healthcare‐associated infection, AB as a community‐acquired respiratory pathogen is not unprecedented either in the United States or abroad.[26, 27, 28, 29, 30] It is, however, reassuring that most other antimicrobials examined in our study exhibit higher rates of susceptibility in the specimens derived from the outpatient settings than either from the hospital or the nursing home.

Our study has a number of strengths. As a large multicenter survey, it is representative of AB susceptibility patterns across the United States, which makes it highly generalizable. We focused on antibiotics for which clinical evidence is available, thus adding a practical dimension to the results. Another pragmatic consideration is examining the data by geographic distributions, allowing an additional layer of granularity for clinical decisions. At the same time it suffers from some limitations. The TSN database consists of microbiology samples from hospital laboratories. Although we attempted to reduce the risk of duplication, because of how samples are numbered in the database, repeat sampling remains a possibility. Despite having stratified the data by geography and the location of origin of the specimen, it is likely not granular enough for local risk stratification decisions clinicians make daily about the choices of empiric therapy. Some of the MIC breakpoints have changed over the period of the study (see Supporting Table 4 in the online version of this article). Because these changes occurred in the last year of data collection (2012), they should have had only a minimal, if any, impact on the observed rates of resistance in the time frame examined. Additionally, because resistance rates evolve rapidly, more current data are required for effective clinical decision making.

In summary, we have demonstrated that the last decade has seen an alarming increase in the rate of resistance of AB to multiple clinically important antimicrobial agents and classes. We have further emphasized the importance of granularity in susceptibility data to help clinicians make sensible decisions about empiric therapy in hospitalized patients with serious infections. Finally, and potentially most disturbingly, the nursing home as a location appears to be a robust reservoir for spread for resistant AB. All of these observations highlight the urgent need to develop novel antibiotics and nontraditional agents, such as antibodies and vaccines, to combat AB infections, in addition to having important infection prevention implications if we are to contain the looming threat of the end of antibiotics.[31]

Among hospitalized patients with serious infections, the choice of empiric therapy plays a key role in outcomes.[1, 2, 3, 4, 5, 6, 7, 8, 9] Rising rates and variable patterns of antimicrobial resistance, however, complicate selecting appropriate empiric therapy. Amidst this shifting landscape of resistance to antimicrobials, gram‐negative bacteria and specifically Acinetobacter baumannii (AB), remain a considerable challenge.[10] On the one hand, AB is a less‐frequent cause of serious infections than organisms like Pseudomonas aeruginosa or Enterobacteriaceae in severely ill hospitalized patients.[11, 12] On the other, AB has evolved a variety of resistance mechanisms and exhibits unpredictable susceptibility patterns.[13] These factors combine to increase the likelihood of administering inappropriate empiric therapy when faced with an infection caused by AB and, thereby, raising the risk of death.[14] The fact that clinicians may not routinely consider AB as the potential culprit pathogen in the patient they are treating along with this organism's highly in vitro resistant nature, may result in routine gram‐negative coverage being frequently inadequate for AB infections.

To address the poor outcomes related to inappropriate empiric therapy in the setting of AB, one requires an appreciation of the longitudinal changes and geographic differences in the susceptibility of this pathogen. Thus, we aimed to examine secular trends in the resistance of AB to antimicrobial agents whose effectiveness against this microorganism was well supported in the literature during the study timeframe.[15]

METHODS

To determine the prevalence of predefined resistance patterns among AB in respiratory and blood stream infection (BSI) specimens, we examined The Surveillance Network (TSN) database from Eurofins. We explored data collected between years 2003 and 2012. The database has been used extensively for surveillance purposes since 1994, and has previously been described in detail.[16, 17, 18, 19, 20] Briefly, TSN is a warehouse of routine clinical microbiology data collected from a nationally representative sample of microbiology laboratories in 217 hospitals in the United States. To minimize selection bias, laboratories are included based on their geography and the demographics of the populations they serve.[18] Only clinically significant samples are reported. No personal identifying information for source patients is available in this database. Only source laboratories that perform antimicrobial susceptibility testing according standard Food and Drug Administrationapproved testing methods and that interpret susceptibility in accordance with the Clinical Laboratory Standards Institute breakpoints are included.[21] (See Supporting Table 4 in the online version of this article for minimum inhibitory concentration (MIC) changes over the course of the studycurrent colistin and polymyxin breakpoints applied retrospectively). All enrolled laboratories undergo a pre‐enrollment site visit. Logical filters are used for routine quality control to detect unusual susceptibility profiles and to ensure appropriate testing methods. Repeat testing and reporting are done as necessary.[18]

Laboratory samples are reported as susceptible, intermediate, or resistant. We grouped isolates with intermediate MICs together with the resistant ones for the purposes of the current analysis. Duplicate isolates were excluded. Only samples representing 1 of the 2 infections of interest, respiratory or BSI, were included.

We examined 3 time periods2003 to 2005, 2006 to 2008, and 2009 to 2012for the prevalence of AB's resistance to the following antibiotics: carbapenems (imipenem, meropenem, doripenem), aminoglycosides (tobramycin, amikacin), tetracyclines (minocycline, doxycycline), polymyxins (colistin, polymyxin B), ampicillin‐sulbactam, and trimethoprim‐sulfamethoxazole. Antimicrobial resistance was defined by the designation of intermediate or resistant in the susceptibility category. Resistance to a class of antibiotics was defined as resistance to all drugs within the class for which testing was available. The organism was multidrug resistant (MDR) if it was resistant to at least 1 antimicrobial in at least 3 drug classes examined.[22] Resistance to a combination of 2 drugs was present if the specimen was resistant to both of the drugs in the combination for which testing was available. We examined the data by infection type, time period, the 9 US Census divisions, and location of origin of the sample.

All categorical variables are reported as percentages. Continuous variables are reported as meansstandard deviations and/or medians with the interquartile range (IQR). We did not pursue hypothesis testing due to a high risk of type I error in this large dataset. Therefore, only clinically important trends are highlighted.

RESULTS

Among the 39,320 AB specimens, 81.1% were derived from a respiratory source and 18.9% represented BSI. Demographics of source patients are listed in Table 1. Notably, the median age of those with respiratory infection (58 years; IQR 38, 73) was higher than among patients with BSI (54.5 years; IQR 36, 71), and there were proportionally fewer females among respiratory patients (39.9%) than those with BSI (46.0%). Though only 24.3% of all BSI samples originated from the intensive are unit (ICU), 40.5% of respiratory specimens came from that location. The plurality of all specimens was collected in the 2003 to 2005 time interval (41.3%), followed by 2006 to 2008 (34.7%), with a minority coming from years 2009 to 2012 (24.0%). The proportions of collected specimens from respiratory and BSI sources were similar in all time periods examined (Table 1). Geographically, the South Atlantic division contributed the most samples (24.1%) and East South Central the fewest (2.6%) (Figure 1). The vast majority of all samples came from hospital wards (78.6%), where roughly one‐half originated in the ICU (37.5%). Fewer still came from outpatient sources (18.3%), and a small minority (2.5%) from nursing homes.

Figure 1

Figure 2 depicts overall resistance patterns by individual drugs, drug classes, and frequently used combinations of agents. Although doripenem had the highest rate of resistance numerically (90.3%), its susceptibility was tested only in a small minority of specimens (n=31, 0.08%). Resistance to trimethoprim‐sulfamethoxazole was high (55.3%) based on a large number of samples tested (n=33,031). Conversely, colistin as an agent and polymyxins as a class exhibited the highest susceptibility rates of over 90%, though the numbers of samples tested for susceptibility to these drugs were also small (colistin n=2,086, 5.3%; polymyxins n=3,120, 7.9%) (Figure 2). Among commonly used drug combinations, carbapenem+aminoglycoside (18.0%) had the lowest resistance rates, and nearly 30% of all AB specimens tested met the criteria for MDR.

Figure 2

Over time, resistance to carbapenems more‐than doubled, from 21.0% in 2003 to 2005 to 47.9% in 2009 to 2012 (Table 2). Although relatively few samples were tested for colistin susceptibility (n=2,086, 5.3%), resistance to this drug also more than doubled from 2.8% (95% confidence interval: 1.9‐4.2) in 2006 to 2008 to 6.9% (95% confidence interval: 5.7‐8.2) in 2009 to 2012. As a class, however, polymyxins exhibited stable resistance rates over the time frame of the study (Table 2). Prevalence of MDR AB rose from 21.4% in 2003 to 2005 to 33.7% in 2006 to 2008, and remained stable at 35.2% in 2009 to 2012. Resistance to even such broad combinations as carbapenem+ampicillin/sulbactam nearly tripled from 13.2% in 2003 to 2005 to 35.5% in 2009 to 2012. Notably, between 2003 and 2012, although resistance rates either rose or remained stable to all other agents, those to minocycline diminished from 56.5% in 2003 to 2005 to 36.6% in 2006 to 2008 to 30.5% in 2009 to 2012. (See Supporting Table 1 in the online version of this article for time trends based on whether they represented respiratory or BSI specimens, with directionally similar trends in both.)

Regionally, examining resistance by classes and combinations of antibiotics, trimethoprim‐sulfamethoxazole exhibited consistently the highest rates of resistance, ranging from the lowest in the New England (28.8%) to the highest in the East North Central (69.9%) Census divisions (See Supporting Table 2 in the online version of this article). The rates of resistance to tetracyclines ranged from 0.0% in New England to 52.6% in the Mountain division, and to polymyxins from 0.0% in the East South Central division to 23.4% in New England. Generally, New England enjoyed the lowest rates of resistance (0.0% to tetracyclines to 28.8% to trimethoprim‐sulfamethoxazole), and the Mountain division the highest (0.9% to polymyxins to 52.6% to tetracyclines). The rates of MDR AB ranged from 8.0% in New England to 50.4% in the Mountain division (see Supporting Table 2 in the online version of this article).

Examining resistances to drug classes and combinations by the location of the source specimen revealed that trimethoprim‐sulfamethoxazole once again exhibited the highest rate of resistance across all locations (see Supporting Table 3 in the online version of this article). Despite their modest contribution to the overall sample pool (n=967, 2.5%), organisms from nursing home subjects had the highest prevalence of resistance to aminoglycosides (36.3%), tetracyclines (57.1%), and carbapenems (47.1%). This pattern held true for combination regimens examined. Nursing homes also vastly surpassed other locations in the rates of MDR AB (46.5%). Interestingly, the rates of MDR did not differ substantially among regular inpatient wards (29.2%), the ICU (28.7%), and outpatient locations (26.2%) (see Supporting Table 3 in the online version of this article).

DISCUSSION

In this large multicenter survey we have documented the rising rates of AB resistance to clinically important antimicrobials in the United States. On the whole, all antimicrobials, except for minocycline, exhibited either large or small increases in resistance. Alarmingly, even colistin, a true last resort AB treatment, lost a considerable amount of activity against AB, with the resistance rate rising from 2.8% in 2006 to 2008 to 6.9% in 2009 to 2012. The single encouraging trend that we observed was that resistance to minocycline appeared to diminish substantially, going from over one‐half of all AB tested in 2003 to 2005 to just under one‐third in 2009 to 2012.

Although we did note a rise in the MDR AB, our data suggest a lower percentage of all AB that meets the MDR phenotype criteria compared to reports by other groups. For example, the Center for Disease Dynamics and Economic Policy (CDDEP), analyzing the same data as our study, reports a rise in MDR AB from 32.1% in 1999 to 51.0% in 2010.[23] This discrepancy is easily explained by the fact that we included polymyxins, tetracyclines, and trimethoprim‐sulfamethoxazole in our evaluation, whereas the CDDEP did not examine these agents. Furthermore, we omitted fluoroquinolones, a drug class with high rates of resistance, from our study, because we were interested in focusing only on antimicrobials with clinical data in AB infections.[22] In addition, we limited our evaluation to specimens derived from respiratory or BSI sources, whereas the CDDEP data reflect any AB isolate present in TSN.

We additionally confirm that there is substantial geographic variation in resistance patterns. Thus, despite different definitions, our data agree with those from the CDDEP that the MDR prevalence is highest in the Mountain and East North Central divisions, and lowest in New England overall.[23] The wide variations underscore the fact that it is not valid to speak of national rates of resistance, but rather it is important to concentrate on the local patterns. This information, though important from the macroepidemiologic standpoint, is likely still not granular enough to help clinicians make empiric treatment decisions. In fact, what is needed for that is real‐time antibiogram data specific to each center and even each unit within each center.

The latter point is further illustrated by our analysis of locations of origin of the specimens. In this analysis, we discovered that, contrary to the common presumption that the ICU has the highest rate of resistant organisms, specimens derived from nursing homes represent perhaps the most intensely resistant organisms. In other words, the nursing home is the setting most likely to harbor patients with respiratory infections and BSIs caused by resistant AB. These data are in agreement with several other recent investigations. In a period‐prevalence survey conducted in the state of Maryland in 2009 by Thom and colleagues, long‐term care facilities were found to have the highest prevalence of any AB, and also those resistant to imipenem, MDR, and extensively drug‐resistant organisms.[24] Mortensen and coworkers confirmed the high prevalence of AB and AB resistance in long‐term care facilities, and extended this finding to suggest that there is evidence for intra‐ and interhospital spread of these pathogens.[25] Our data confirm this concerning finding at the national level, and point to a potential area of intervention for infection prevention.

An additional finding of some concern is that the highest proportion of colistin resistance among those specimens, whose location of origin was reported in the database, was the outpatient setting (6.6% compared to 5.4% in the ICU specimens, for example). Although these infections would likely meet the definition for healthcare‐associated infection, AB as a community‐acquired respiratory pathogen is not unprecedented either in the United States or abroad.[26, 27, 28, 29, 30] It is, however, reassuring that most other antimicrobials examined in our study exhibit higher rates of susceptibility in the specimens derived from the outpatient settings than either from the hospital or the nursing home.

Our study has a number of strengths. As a large multicenter survey, it is representative of AB susceptibility patterns across the United States, which makes it highly generalizable. We focused on antibiotics for which clinical evidence is available, thus adding a practical dimension to the results. Another pragmatic consideration is examining the data by geographic distributions, allowing an additional layer of granularity for clinical decisions. At the same time it suffers from some limitations. The TSN database consists of microbiology samples from hospital laboratories. Although we attempted to reduce the risk of duplication, because of how samples are numbered in the database, repeat sampling remains a possibility. Despite having stratified the data by geography and the location of origin of the specimen, it is likely not granular enough for local risk stratification decisions clinicians make daily about the choices of empiric therapy. Some of the MIC breakpoints have changed over the period of the study (see Supporting Table 4 in the online version of this article). Because these changes occurred in the last year of data collection (2012), they should have had only a minimal, if any, impact on the observed rates of resistance in the time frame examined. Additionally, because resistance rates evolve rapidly, more current data are required for effective clinical decision making.

In summary, we have demonstrated that the last decade has seen an alarming increase in the rate of resistance of AB to multiple clinically important antimicrobial agents and classes. We have further emphasized the importance of granularity in susceptibility data to help clinicians make sensible decisions about empiric therapy in hospitalized patients with serious infections. Finally, and potentially most disturbingly, the nursing home as a location appears to be a robust reservoir for spread for resistant AB. All of these observations highlight the urgent need to develop novel antibiotics and nontraditional agents, such as antibodies and vaccines, to combat AB infections, in addition to having important infection prevention implications if we are to contain the looming threat of the end of antibiotics.[31]

Despite its decreasing mortality, sepsis remains a leading reason for intensive care unit (ICU) admission and is associated with crude mortality in excess of 25%.[1, 2] In the United States there are between 660,000 and 750,000 sepsis hospitalizations annually, with the direct costs surpassing $24 billion.[3, 4, 5] As mortality rates have begun to fall, attention has shifted to issues of morbidity and recovery, the intermediate and longer‐term consequences associated with survivorship, and how interventions made while the patient is acutely ill in the ICU alter later health outcomes.[3, 5, 6, 7, 8]

One area of particular interest is the need for healthcare utilization following an acute admission for sepsis, and specifically rehospitalization within 30 days of discharge. This outcome is important not just from the perspective of the patient's well‐being, but also from the point of view of healthcare financing. Through the establishment of Hospital Readmission Reduction Program, the Centers for Medicare and Medicaid Services have sharply reduced reimbursement to hospitals for excessive rates of 30‐day readmissions.[9]

For sepsis, little is known about such readmissions, and even less about how to prevent them. A handful of studies suggest that this rate is between 5% and 26%.[10, 11, 12, 13] Whereas some of these studies looked at some of the factors that impact readmissions,[11, 12] none examined the potential contribution of microbiology of sepsis to this outcome.

To explore these questions, we conducted a single‐center retrospective cohort study among critically ill patients admitted to the ICU with severe culture‐positive sepsis and/or septic shock and determined the rate of early posthospital discharge readmission. In addition, we sought to elucidate predictors of subsequent readmission.

METHODS

RESULTS

Among the 1697 patients with severe sepsis or septic shock who were discharged alive from the hospital, 543 (32.0%) required a rehospitalization within 30 days. There were no differences in age or gender distribution between the groups (Table 1). All comorbidities examined were more prevalent among those with a 30‐day readmission than among those without, with the median Charlson comorbidity score reflecting this imbalance (5 vs 4, P<0.001). Similarly, most of the HCA risk factors were more prevalent among the readmitted group than the comparator group, with HCA sepsis among 94.2% of the former and 90.7% of the latter (P = 0.014).

During the index hospitalization, 589 patients (34.7%) suffered from septic shock requiring vasopressors; this did not impact the 30‐day readmission risk (Table 1). Commensurately, markers of severity of acute illness (APACHE II score, mechanical ventilation, peak white blood cell count) did not differ between the groups. With respect to the primary source of sepsis, urine was less, whereas central nervous system was more likely among those readmitted within 30 days. Similarly, there was a significant imbalance between the groups in the prevalence of AKI (Table 1). Specifically, those who did require a readmission were slightly less likely to have sustained no AKI (RIFLE: None; 14.9% vs 18.5%, P = 0.073). Those requiring readmission were also less likely to be in the category RIFLE: Risk (20.6% vs 26.5%, P = 0.009). The direction of this disparity was reversed for the Injury and Failure categories. No differences between groups were seen among those with categories Loss and end‐stage kidney disease (ESKD) (Table 1).

The microbiology of sepsis did not differ in most respects between the 30‐day readmission groups, save for several organisms (Table 2). Most strikingly, those who required a readmission were more likely than those who did not to be infected with Bacteroides spp, Candida spp, an MDR or an ESBL organism (Table 2). As for the outcomes of the index hospitalization, those with a repeat admission had a longer overall and postonset of sepsis initial hospital length of stay, and were less likely to be discharged either home without home health care or transferred to another hospital at the end of their index hospitalization (Table 3).

In a logistic regression model, 5 factors emerged as predictors of 30‐day readmission (Table 4). Having RIFLE: Injury or RIFLE: Failure carried an approximately 2‐fold increase in the odds of 30‐day rehospitalization (odds ratio: 1.95, 95% confidence interval: 1.302.93, P = 0.001) relative to having a RIFLE: None or RIFLE: Risk. Although having strong association with this outcome, harboring an ESBL organism or Bacteroides spp were both relatively infrequent events (3.3% ESBL and 3.2% Bacteroides spp). Infection with Escherichia coli and urine as the source of sepsis both appeared to be significantly protective against a readmission (Table 4). The model's discrimination was moderate (AUROC = 0.653) and its calibration adequate (Hosmer‐Lemeshow P = 0.907). (See Supporting Information, Appendix 1, in the online version of this article for the steps in the development of the final model.)

DISCUSSION

In this single‐center retrospective cohort study, nearly one‐third of survivors of culture‐positive severe sepsis or septic shock required a rehospitalization within 30 days of discharge from their index admission. Factors that contributed to a higher odds of rehospitalization were having mild‐to‐moderate AKI (RIFLE: Injury or RIFLE: Failure) and infection with ESBL organisms or Bacteroides spp, whereas urine as the source of sepsis and E coli as the pathogen appeared to be protective.

A recent study by Hua and colleagues examining the New York Statewide Planning and Research Cooperative System for the years 2008 to 2010 noted a 16.2% overall rate of 30‐day rehospitalization among survivors of initial critical illness.[11] Just as we observed, Hua et al. concluded that development of AKI correlated with readmission. Because they relied on administrative data for their analysis, AKI was diagnosed when hemodialysis was utilized. Examining AKI using SCr changes, our findings add a layer of granularity to the relationship between AKI stages and early readmission. Specifically, we failed to detect any rise in the odds of rehospitalization when either very mild (RIFLE: Risk) or severe (RIFLE: Loss or RIFLE: ESKD) AKI was present. Only when either RIFLE: Injury or RIFLE: Failure developed did the odds of readmission rise. In addition to diverging definitions between our studies, differences in populations also likely yielded different results.[11] Although Hua et al. examined all admissions to the ICU regardless of the diagnosis or illness severity, our cohort consisted of only those ICU patients who survived culture‐positive severe sepsis/septic shock. Because AKI is a known risk factor for mortality in sepsis,[19] the potential for immortal time bias leaves a smaller pool of surviving patients with ESKD at risk for readmission. Regardless of the explanation, it may be prudent to focus on preventing AKI not only to improve survival, but also from the standpoint of diminishing the risk of an early readmission.

Four additional studies have examined the frequency of early readmissions among survivors of critical illness. Liu et al. noted 17.9% 30‐day rehospitalization rate among sepsis survivors.[12] Factors associated with the risk of early readmission included acute and chronic diseases burdens, index hospital LOS, and the need for the ICU in the index sepsis admission. In contrast to our cohort, all of whom were in the ICU during their index episode, less than two‐thirds of the entire population studied by Liu had required an ICU admission. Additionally, Liu's study did not specifically examine the potential impact of AKI or of microbiology on this outcome.

Prescott and coworkers examined healthcare utilization following an episode of severe sepsis.[13] Among other findings, they reported a 30‐day readmission rate of 26.5% among survivors. Although closer to our estimate, this study included all patients surviving a severe sepsis hospitalization, and not only those with a positive culture. These investigators did not examine predictors of readmission.[13]

Horkan et al. examined specifically whether there was an association between AKI and postdischarge outcomes, including 30‐day readmission risk, in a large cohort of patients who survived their critical illness.[20] In it they found that readmission risk ranged from 19% to 21%, depending on the extent of the AKI. Moreover, similar to our findings, they reported that in an adjusted analysis RIFLE: Injury and RIFLE: Failure were associated with a rise in the odds of a 30‐day rehospitalizaiton. In contrast to our study, Horkan et al. did detect an increase in the odds of this outcome associated with RIFLE: Risk. There are likely at least 3 reasons for this difference. First, we focused only on patients with severe sepsis or septic shock, whereas Horkan and colleagues included all critical illness survivors. Second, we were able to explore the impact of microbiology on this outcome. Third, Horkan's study included an order of magnitude more patients than did ours, thus making it more likely either to have the power to detect a true association that we may have lacked or to be more susceptible to type I error.

Finally, Goodwin and colleagues utilized 3 states' databases included in the Health Care and Utilization Project (HCUP) from the Agency for Healthcare Research and Quality to study frequency and risk factors for 30‐day readmission among survivors of severe sepsis.[21] Patients were identified based on the use of the severe sepsis (995.92) and septic shock (785.52). These authors found a 30‐day readmission rate of 26%. Although chronic renal disease, among several other factors, was associated with an increase in this risk, the data source did not permit these investigators to examine the impact of AKI on the outcomes. Similarly, HCUP data do not contain microbiology, a distinct difference from our analysis.

If clinicians are to pursue strategies to reduce the risk of an all‐cause 30‐day readmission, the key goal is not simply to identify all variables associated with readmission, but to focus on factors that are potentially modifiable. Although neither Hua nor Liu and their teams identified any additional factors that are potentially modifiable,[11, 12] in the present study, among the 5 factors we identified, the development of mild to moderate AKI during the index hospitalization may deserve stronger consideration for efforts at prevention. Although one cannot conclude automatically that preventing AKI in this population could mitigate some of the early rehospitalization risk, critically ill patients are frequently exposed to a multitude of nephrotoxic agents. Those caring for subjects with sepsis should reevaluate the risk‐benefit equation of these factors more cautiously and apply guideline‐recommended AKI prevention strategies more aggressively, particularly because a relatively minor change in SCr resulted in an excess risk of readmission.[22]

In addition to AKI, which is potentially modifiable, we identified several other clinical factors predictive of 30‐day readmission, which are admittedly not preventable. Thus, microbiology was predictive of this outcome, with E coli engendering fewer and Bacteroides spp and ESBL organisms more early rehospitalizations. Similarly, urine as the source of sepsis was associated with a lower risk for this endpoint.

Our study has a number of limitations. As a retrospective cohort, it is subject to bias, most notably a selection bias. Specifically, because the flagship hospital of the BJC HealthCare system is a referral center, it is possible that we did not capture all readmissions. However, generally, if a patient who receives healthcare within 1 of the BJC hospitals presents to a nonsystem hospital, that patient is nearly always transferred back into the integrated system because of issues of insurance coverage. Analysis of certain diagnosis‐related groups has indicated that 73% of all patients overall discharged from 4 of the large BJC system institutions who require a readmission within 30 days of discharge return to a BJC hospital (personal communication, Financial Analysis and Decision Support Department at BJC to Dr. Kollef May 12, 2015). Therefore, we may have misclassified the outcome in as many as 180 patients. The fact that our readmission rate was fully double that seen in Hua et al.'s and Liu et al.'s studies, and somewhat higher than that reported by Prescott et al., attests not only to the population differences, but also to the fact that we are unlikely to have missed a substantial percentage of readmissions.[11, 12, 13] Furthermore, to mitigate biases, we enrolled all consecutive patients meeting the predetermined criteria. Missing from our analysis are events that occurred between the index discharge and the readmission. Likewise, we were unable to obtain such potentially important variables as code status or outpatient mortality following discharge. These intervening factors, if included in subsequent studies, may increase the predictive power of the model. Because we relied on administrative coding to identify cases of severe sepsis and septic shock, it is possible that there is misclassification within our cohort. Recent studies indicate, however, that the Angus definition, used in our study, has high negative and positive predictive values for severe sepsis identification.[23] It is still possible that our cohort is skewed toward a more severely ill population, making our results less generalizable to the less severely ill septic patients.[24] The study was performed at a single healthcare system and included only cases of severe sepsis or septic shock that had a positive blood culture, and thus the findings may not be broadly generalizable either to patients without a positive blood culture or to institutions that do not resemble it.

In summary, we have demonstrated that survivors of culture‐positive severe sepsis or septic shock have a high rate of 30‐day rehospitalization. Because the US federal government's initiatives deem 30‐day readmissions to be a quality metric and penalize institutions with higher‐than average readmission rates, a high volume of critically ill patients with culture‐positive severe sepsis and septic shock may disproportionately put an institution at risk for such penalties. Unfortunately, not many of the determinants of readmission are amenable to prevention. As sepsis survival continues to improve, hospitals will need to concentrate their resources on coordinating care of these complex patients so as to improve both individual quality of life and the quality of care that they provide.

Despite its decreasing mortality, sepsis remains a leading reason for intensive care unit (ICU) admission and is associated with crude mortality in excess of 25%.[1, 2] In the United States there are between 660,000 and 750,000 sepsis hospitalizations annually, with the direct costs surpassing $24 billion.[3, 4, 5] As mortality rates have begun to fall, attention has shifted to issues of morbidity and recovery, the intermediate and longer‐term consequences associated with survivorship, and how interventions made while the patient is acutely ill in the ICU alter later health outcomes.[3, 5, 6, 7, 8]

One area of particular interest is the need for healthcare utilization following an acute admission for sepsis, and specifically rehospitalization within 30 days of discharge. This outcome is important not just from the perspective of the patient's well‐being, but also from the point of view of healthcare financing. Through the establishment of Hospital Readmission Reduction Program, the Centers for Medicare and Medicaid Services have sharply reduced reimbursement to hospitals for excessive rates of 30‐day readmissions.[9]

For sepsis, little is known about such readmissions, and even less about how to prevent them. A handful of studies suggest that this rate is between 5% and 26%.[10, 11, 12, 13] Whereas some of these studies looked at some of the factors that impact readmissions,[11, 12] none examined the potential contribution of microbiology of sepsis to this outcome.

To explore these questions, we conducted a single‐center retrospective cohort study among critically ill patients admitted to the ICU with severe culture‐positive sepsis and/or septic shock and determined the rate of early posthospital discharge readmission. In addition, we sought to elucidate predictors of subsequent readmission.

METHODS

RESULTS

Among the 1697 patients with severe sepsis or septic shock who were discharged alive from the hospital, 543 (32.0%) required a rehospitalization within 30 days. There were no differences in age or gender distribution between the groups (Table 1). All comorbidities examined were more prevalent among those with a 30‐day readmission than among those without, with the median Charlson comorbidity score reflecting this imbalance (5 vs 4, P<0.001). Similarly, most of the HCA risk factors were more prevalent among the readmitted group than the comparator group, with HCA sepsis among 94.2% of the former and 90.7% of the latter (P = 0.014).

During the index hospitalization, 589 patients (34.7%) suffered from septic shock requiring vasopressors; this did not impact the 30‐day readmission risk (Table 1). Commensurately, markers of severity of acute illness (APACHE II score, mechanical ventilation, peak white blood cell count) did not differ between the groups. With respect to the primary source of sepsis, urine was less, whereas central nervous system was more likely among those readmitted within 30 days. Similarly, there was a significant imbalance between the groups in the prevalence of AKI (Table 1). Specifically, those who did require a readmission were slightly less likely to have sustained no AKI (RIFLE: None; 14.9% vs 18.5%, P = 0.073). Those requiring readmission were also less likely to be in the category RIFLE: Risk (20.6% vs 26.5%, P = 0.009). The direction of this disparity was reversed for the Injury and Failure categories. No differences between groups were seen among those with categories Loss and end‐stage kidney disease (ESKD) (Table 1).

The microbiology of sepsis did not differ in most respects between the 30‐day readmission groups, save for several organisms (Table 2). Most strikingly, those who required a readmission were more likely than those who did not to be infected with Bacteroides spp, Candida spp, an MDR or an ESBL organism (Table 2). As for the outcomes of the index hospitalization, those with a repeat admission had a longer overall and postonset of sepsis initial hospital length of stay, and were less likely to be discharged either home without home health care or transferred to another hospital at the end of their index hospitalization (Table 3).

In a logistic regression model, 5 factors emerged as predictors of 30‐day readmission (Table 4). Having RIFLE: Injury or RIFLE: Failure carried an approximately 2‐fold increase in the odds of 30‐day rehospitalization (odds ratio: 1.95, 95% confidence interval: 1.302.93, P = 0.001) relative to having a RIFLE: None or RIFLE: Risk. Although having strong association with this outcome, harboring an ESBL organism or Bacteroides spp were both relatively infrequent events (3.3% ESBL and 3.2% Bacteroides spp). Infection with Escherichia coli and urine as the source of sepsis both appeared to be significantly protective against a readmission (Table 4). The model's discrimination was moderate (AUROC = 0.653) and its calibration adequate (Hosmer‐Lemeshow P = 0.907). (See Supporting Information, Appendix 1, in the online version of this article for the steps in the development of the final model.)

DISCUSSION

In this single‐center retrospective cohort study, nearly one‐third of survivors of culture‐positive severe sepsis or septic shock required a rehospitalization within 30 days of discharge from their index admission. Factors that contributed to a higher odds of rehospitalization were having mild‐to‐moderate AKI (RIFLE: Injury or RIFLE: Failure) and infection with ESBL organisms or Bacteroides spp, whereas urine as the source of sepsis and E coli as the pathogen appeared to be protective.

A recent study by Hua and colleagues examining the New York Statewide Planning and Research Cooperative System for the years 2008 to 2010 noted a 16.2% overall rate of 30‐day rehospitalization among survivors of initial critical illness.[11] Just as we observed, Hua et al. concluded that development of AKI correlated with readmission. Because they relied on administrative data for their analysis, AKI was diagnosed when hemodialysis was utilized. Examining AKI using SCr changes, our findings add a layer of granularity to the relationship between AKI stages and early readmission. Specifically, we failed to detect any rise in the odds of rehospitalization when either very mild (RIFLE: Risk) or severe (RIFLE: Loss or RIFLE: ESKD) AKI was present. Only when either RIFLE: Injury or RIFLE: Failure developed did the odds of readmission rise. In addition to diverging definitions between our studies, differences in populations also likely yielded different results.[11] Although Hua et al. examined all admissions to the ICU regardless of the diagnosis or illness severity, our cohort consisted of only those ICU patients who survived culture‐positive severe sepsis/septic shock. Because AKI is a known risk factor for mortality in sepsis,[19] the potential for immortal time bias leaves a smaller pool of surviving patients with ESKD at risk for readmission. Regardless of the explanation, it may be prudent to focus on preventing AKI not only to improve survival, but also from the standpoint of diminishing the risk of an early readmission.

Four additional studies have examined the frequency of early readmissions among survivors of critical illness. Liu et al. noted 17.9% 30‐day rehospitalization rate among sepsis survivors.[12] Factors associated with the risk of early readmission included acute and chronic diseases burdens, index hospital LOS, and the need for the ICU in the index sepsis admission. In contrast to our cohort, all of whom were in the ICU during their index episode, less than two‐thirds of the entire population studied by Liu had required an ICU admission. Additionally, Liu's study did not specifically examine the potential impact of AKI or of microbiology on this outcome.

Prescott and coworkers examined healthcare utilization following an episode of severe sepsis.[13] Among other findings, they reported a 30‐day readmission rate of 26.5% among survivors. Although closer to our estimate, this study included all patients surviving a severe sepsis hospitalization, and not only those with a positive culture. These investigators did not examine predictors of readmission.[13]

Horkan et al. examined specifically whether there was an association between AKI and postdischarge outcomes, including 30‐day readmission risk, in a large cohort of patients who survived their critical illness.[20] In it they found that readmission risk ranged from 19% to 21%, depending on the extent of the AKI. Moreover, similar to our findings, they reported that in an adjusted analysis RIFLE: Injury and RIFLE: Failure were associated with a rise in the odds of a 30‐day rehospitalizaiton. In contrast to our study, Horkan et al. did detect an increase in the odds of this outcome associated with RIFLE: Risk. There are likely at least 3 reasons for this difference. First, we focused only on patients with severe sepsis or septic shock, whereas Horkan and colleagues included all critical illness survivors. Second, we were able to explore the impact of microbiology on this outcome. Third, Horkan's study included an order of magnitude more patients than did ours, thus making it more likely either to have the power to detect a true association that we may have lacked or to be more susceptible to type I error.

Finally, Goodwin and colleagues utilized 3 states' databases included in the Health Care and Utilization Project (HCUP) from the Agency for Healthcare Research and Quality to study frequency and risk factors for 30‐day readmission among survivors of severe sepsis.[21] Patients were identified based on the use of the severe sepsis (995.92) and septic shock (785.52). These authors found a 30‐day readmission rate of 26%. Although chronic renal disease, among several other factors, was associated with an increase in this risk, the data source did not permit these investigators to examine the impact of AKI on the outcomes. Similarly, HCUP data do not contain microbiology, a distinct difference from our analysis.

If clinicians are to pursue strategies to reduce the risk of an all‐cause 30‐day readmission, the key goal is not simply to identify all variables associated with readmission, but to focus on factors that are potentially modifiable. Although neither Hua nor Liu and their teams identified any additional factors that are potentially modifiable,[11, 12] in the present study, among the 5 factors we identified, the development of mild to moderate AKI during the index hospitalization may deserve stronger consideration for efforts at prevention. Although one cannot conclude automatically that preventing AKI in this population could mitigate some of the early rehospitalization risk, critically ill patients are frequently exposed to a multitude of nephrotoxic agents. Those caring for subjects with sepsis should reevaluate the risk‐benefit equation of these factors more cautiously and apply guideline‐recommended AKI prevention strategies more aggressively, particularly because a relatively minor change in SCr resulted in an excess risk of readmission.[22]

In addition to AKI, which is potentially modifiable, we identified several other clinical factors predictive of 30‐day readmission, which are admittedly not preventable. Thus, microbiology was predictive of this outcome, with E coli engendering fewer and Bacteroides spp and ESBL organisms more early rehospitalizations. Similarly, urine as the source of sepsis was associated with a lower risk for this endpoint.

Our study has a number of limitations. As a retrospective cohort, it is subject to bias, most notably a selection bias. Specifically, because the flagship hospital of the BJC HealthCare system is a referral center, it is possible that we did not capture all readmissions. However, generally, if a patient who receives healthcare within 1 of the BJC hospitals presents to a nonsystem hospital, that patient is nearly always transferred back into the integrated system because of issues of insurance coverage. Analysis of certain diagnosis‐related groups has indicated that 73% of all patients overall discharged from 4 of the large BJC system institutions who require a readmission within 30 days of discharge return to a BJC hospital (personal communication, Financial Analysis and Decision Support Department at BJC to Dr. Kollef May 12, 2015). Therefore, we may have misclassified the outcome in as many as 180 patients. The fact that our readmission rate was fully double that seen in Hua et al.'s and Liu et al.'s studies, and somewhat higher than that reported by Prescott et al., attests not only to the population differences, but also to the fact that we are unlikely to have missed a substantial percentage of readmissions.[11, 12, 13] Furthermore, to mitigate biases, we enrolled all consecutive patients meeting the predetermined criteria. Missing from our analysis are events that occurred between the index discharge and the readmission. Likewise, we were unable to obtain such potentially important variables as code status or outpatient mortality following discharge. These intervening factors, if included in subsequent studies, may increase the predictive power of the model. Because we relied on administrative coding to identify cases of severe sepsis and septic shock, it is possible that there is misclassification within our cohort. Recent studies indicate, however, that the Angus definition, used in our study, has high negative and positive predictive values for severe sepsis identification.[23] It is still possible that our cohort is skewed toward a more severely ill population, making our results less generalizable to the less severely ill septic patients.[24] The study was performed at a single healthcare system and included only cases of severe sepsis or septic shock that had a positive blood culture, and thus the findings may not be broadly generalizable either to patients without a positive blood culture or to institutions that do not resemble it.

In summary, we have demonstrated that survivors of culture‐positive severe sepsis or septic shock have a high rate of 30‐day rehospitalization. Because the US federal government's initiatives deem 30‐day readmissions to be a quality metric and penalize institutions with higher‐than average readmission rates, a high volume of critically ill patients with culture‐positive severe sepsis and septic shock may disproportionately put an institution at risk for such penalties. Unfortunately, not many of the determinants of readmission are amenable to prevention. As sepsis survival continues to improve, hospitals will need to concentrate their resources on coordinating care of these complex patients so as to improve both individual quality of life and the quality of care that they provide.

Despite its decreasing mortality, sepsis remains a leading reason for intensive care unit (ICU) admission and is associated with crude mortality in excess of 25%.[1, 2] In the United States there are between 660,000 and 750,000 sepsis hospitalizations annually, with the direct costs surpassing $24 billion.[3, 4, 5] As mortality rates have begun to fall, attention has shifted to issues of morbidity and recovery, the intermediate and longer‐term consequences associated with survivorship, and how interventions made while the patient is acutely ill in the ICU alter later health outcomes.[3, 5, 6, 7, 8]

One area of particular interest is the need for healthcare utilization following an acute admission for sepsis, and specifically rehospitalization within 30 days of discharge. This outcome is important not just from the perspective of the patient's well‐being, but also from the point of view of healthcare financing. Through the establishment of Hospital Readmission Reduction Program, the Centers for Medicare and Medicaid Services have sharply reduced reimbursement to hospitals for excessive rates of 30‐day readmissions.[9]

For sepsis, little is known about such readmissions, and even less about how to prevent them. A handful of studies suggest that this rate is between 5% and 26%.[10, 11, 12, 13] Whereas some of these studies looked at some of the factors that impact readmissions,[11, 12] none examined the potential contribution of microbiology of sepsis to this outcome.

To explore these questions, we conducted a single‐center retrospective cohort study among critically ill patients admitted to the ICU with severe culture‐positive sepsis and/or septic shock and determined the rate of early posthospital discharge readmission. In addition, we sought to elucidate predictors of subsequent readmission.

METHODS

RESULTS

Among the 1697 patients with severe sepsis or septic shock who were discharged alive from the hospital, 543 (32.0%) required a rehospitalization within 30 days. There were no differences in age or gender distribution between the groups (Table 1). All comorbidities examined were more prevalent among those with a 30‐day readmission than among those without, with the median Charlson comorbidity score reflecting this imbalance (5 vs 4, P<0.001). Similarly, most of the HCA risk factors were more prevalent among the readmitted group than the comparator group, with HCA sepsis among 94.2% of the former and 90.7% of the latter (P = 0.014).