Results
Baseline Characteristics
A total of 368,514 hospitalizations for sepsis were identified from 325 hospitals for analysis in the HCUP SID database from 2009 to 2011. After excluding hospitalizations with age less than 18 (n = 6,163), missing length of stay (n = 55) and patient identifier number (n = 17,715), out-of-state residents (n = 4,368), admissions to rehabilitation or psychiatric facilities (n = 2,585), discharges in December (n = 34,157), transfers to another acute care hospital (n = 17,174), and in-hospital death (n = 58,928), there were 240,198 index hospitalizations available for analysis. The baseline characteristics of index hospitalizations for sepsis, CHF, and AMI are shown in the Supplemental Table 2 (Supplemental Digital Content 1, http://links.lww.com/CCM/B362). For sepsis, median age was 72 years (interquartile range [IQR], 57–82 yr). The percentage of men and women was similar (47.6% vs 52.4%, respectively). Whites were the most prevalent racial/ethnic group in the study (59.1%), followed by Hispanics (20.5%), Blacks (9.3%), and Asians (9.2%). Most patients had multiple medical comorbidities as defined by the Charlson comorbidity index (Supplemental Table 2, Supplemental Digital Content 1, http://links.lww.com/CCM/B362). A similar proportion of hospitalizations for sepsis resulted in routine discharges to home (40.1%) or skilled nursing facility (39.9%). Approximately 18% of patients were discharged with home healthcare. The baseline characteristics of index hospitalizations for sepsis were similar between 2009, 2010, and 2011 (Supplemental Table 3, Supplemental Digital Content 1, http://links.lww.com/CCM/B362).
There were some notable differences in baseline characteristics of index hospitalizations for sepsis versus CHF and AMI (Supplemental Table 2, Supplemental Digital Content 1, http://links.lww.com/CCM/B362). Sepsis hospitalizations had a higher proportion of patients with dementia and malignancy as comorbidities, resulted more frequently in hospital discharge to a skilled nursing facility (39.9% for sepsis vs 17.2% and 14.0% for CHF and AMI, respectively), and had longer length of stay (6 d; IQR, 3–10 d vs 3 d; IQR, 2–6 d for both CHF and AMI).
Readmission Rates and Costs
The all-cause 30-day readmission rate for sepsis during the study period was 20.4% (Table 1). The 7-, 14-, 30-, and 90-day readmission rates for sepsis during 2009–2011 are shown in Figure 1. The average rate of readmission per day was 1.1% for days 1–7, 0.8% for days 7–14, 0.5% for days 14–30, and 0.2% for days 30–90. The 30-day readmission rates were 12.6%, 19.7%, and 29.2% among hospitalizations, resulting in a routine discharge home, a discharge with home healthcare, and discharge to a skilled nursing facility, respectively.
(Enlarge Image)
Figure 1.
Sepsis readmission rates for 2009–2011. The all-cause 30-day readmission rates for sepsis were 20.7%, 20.5%, and 20.1% for 2009, 2010, and 2011, respectively. Approximately 7% of patients were readmitted within 7 d after discharge from an index hospitalization for sepsis. Within 90 d of an index hospitalization, over 30% of patients were readmitted. The rate of 7-, 14-, 30-, and 90-day readmissions decreased slightly from 2009 to 2011.
The 30-day readmission rates following hospitalizations for CHF and AMI were 23.6% and 17.7%, respectively (Table 1). Median lengths of stay and in-hospital mortality were greater for sepsis than for CHF and AMI (Table 1). From 2009 through 2011, the total number of 30-day readmissions in California was 48,988 for sepsis, 45,651 for CHF, and 18,707 for AMI. Median cost of hospitalization was $15,462 (IQR, $8,443–30,180) for sepsis, $16,550 (IQR, $9,095–27,734) for AMI, and $8,870 (IQR, $5,455–15,655) for CHF (Table 1). Median cost of a readmission within 30 days was $18,794 (IQR, $10,260–35,386) for sepsis, $14,075 (IQR, $7,036–26,200) for AMI, and $8,973 (IQR, $5,465–15,965) for CHF. Estimated annual costs of all 30-day readmissions in California during the study period were $500 million/yr for sepsis, $229 million/yr for CHF, and $142 million/yr for AMI. The trends in 30-day readmission rates, lengths of stay, mortality, and costs by year are shown in Supplemental Table 4 (Supplemental Digital Content 1, http://links.lww.com/CCM/B362).
Hospital-level Variation in 30-Day Readmissions
There were 346 California hospitals eligible for the study in the dataset. After excluding hospitals with less than 10 annual hospitalizations for each diagnosis, 325 hospitals for sepsis, 328 hospitals for CHF, and 287 hospitals for AMI were included in the variability analysis. The risk- and reliability-adjusted readmission rates for sepsis ranged from 11.0% to 39.8% (median, 19.9%; IQR, 16.1–26.0%) and varied significantly across hospitals (p < 0.001). As an example of the magnitude of variation in readmissions across hospitals, the adjusted rates for hospitals 1 SD above versus below the mean were 33.7% (95% CI, 33.0–34.5) versus 13.0% (95% CI, 12.7–13.3), respectively. Hospitals in the 90th percentile were 2.5 times more likely to have a 30-day readmission for sepsis than those in the lowest 10th percentile (34.0% vs 13.8%). This variation is shown in the caterpillar plot where each hospital was ranked according to their risk-adjusted rate of 30-day readmissions following hospitalization for sepsis, CHF, and AMI (Fig. 2). The overall variation between hospitals for sepsis and CHF were similar (Fig. 2). Hospitals had higher variability in 30-day readmission rates for AMI compared with sepsis and CHF (Fig. 2).
(Enlarge Image)
Figure 2.
Variation of adjusted 30-day readmission rates in California hospitals for sepsis. The risk- and reliability-adjusted proportion of 30-day readmissions at each hospital ranged from 11.0% to 39.8% (median, 19.9%; interquartile range [IQR], 16.1–26.0%) for sepsis (black triangles), 11.3% to 38.4% (median, 22.9%; IQR, 19.2–26.6%) for congestive heart failure (CHF) (light gray diamonds), and 3.6% to 40.8% (median, 17.0%; IQR, 12.2–20.0%) for acute myocardial infarction (AMI) (dark gray squares). The 30-day readmission rates were adjusted for age and the number of medical comorbidities. The error bars indicate 95% CIs. *n = 287 for AMI, 328 for CHF, and 325 for sepsis.
Diagnoses on Readmissions
Septicemia was the most common diagnosis on 30-day readmission following sepsis and comprised 29.2% of the readmissions (Supplemental Table 5, Supplemental Digital Content 1, http://links.lww.com/CCM/B362). Pneumonia (4.8%) and respiratory failure (4.1%) were the next most common diagnoses (Supplemental Table 5, Supplemental Digital Content 1, http://links.lww.com/CCM/B362). The distribution of the most common clinical diagnosis categories for 30-day readmissions following sepsis is summarized in Figure 3. Infections accounted for 59.3% of the primary diagnoses on readmission at 30 days following an index hospitalization (Fig. 3). Frequencies of the diagnoses and clinical categories for 7-, 14-, 30-, and 90-day readmissions are shown in Supplemental Table 1 (Supplemental Digital Content 1, http://links.lww.com/CCM/B362) and Supplemental Table 5 (Supplemental Digital Content 1, http://links.lww.com/CCM/B362).
(Enlarge Image)
Figure 3.
Sepsis readmission diagnosis categories. Infections were the most common diagnosis category on 7-, 14-, 30-, and 90-day readmissions and comprised nearly 60% of the cases. Pulmonary diagnoses were the second most common cause, comprising 12–16% of cases. The proportion of readmissions caused by infections remained stable over the first 90 d after the index hospitalization, whereas pulmonary causes decreased over that time. GU = genitourinary.
Patient- and Hospital-level Factors Associated With 30-Day Readmission
Among patient-level factors, younger age (OR, 1.34; 95% CI, 1.29–1.39; between the youngest and oldest age categories), Black (OR, 1.29; 95% CI, 1.24–1.33; compared with White) and Native American (OR, 2.39; 95% CI, 1.79–3.19; compared with White) race, lower income (OR, 1.13; 95% CI, 1.10–1.16; between the lowest and highest income quartiles), residence in metropolitan areas, and greater burden of medical comorbidities were associated with higher odds of 30-day readmission following hospitalization for sepsis (Supplemental Table 6, Supplemental Digital Content 1, http://links.lww.com/CCM/B362; Supplemental Table 7, Supplemental Digital Content 1, http://links.lww.com/CCM/B362). Female gender was associated with lower odds of 30-day readmission (OR, 0.87; 95% CI, 0.86–0.89; compared with male). Hospital-level characteristics most strongly associated with 30-day readmissions included hospitalizations at an institution delivering healthcare to the highest proportion of minorities (OR, 1.28; 95% CI, 1.23–1.34; for quintile of hospitals that had the highest proportion of minorities compared with the lowest quintile) and for-profit (OR, 1.34; 95% CI, 1.31–1.38) or university (OR, 1.35; 95% CI, 1.26–1.44) hospitals. By contrast, hospitalizations for sepsis in public hospitals were less likely to have 30-day readmissions compared with nonprofit hospitals (OR, 0.85; 95% CI, 0.82–0.89). The c-statistic of the sepsis model was 0.68.
The patient- and hospital-level variables that were associated with higher odds of 30-day readmissions were generally similar between sepsis and CHF (Supplemental Table 6, Supplemental Digital Content 1, http://links.lww.com/CCM/B362). By contrast, there was a stronger association between hospital-level characteristics, including hospital size, teaching status, and proportion of minority patients, and 30-day readmissions for AMI than for sepsis or CHF (Supplemental Table 6, Supplemental Digital Content 1, http://links.lww.com/CCM/B362).
Sensitivity Analysis
Using the Angus implementation to identify hospitalizations for severe sepsis, the all-cause 30-day readmission rate in 2011 was 23.3% (Supplemental Table 8, Supplemental Digital Content 1, http://links.lww.com/CCM/B362). The number of sepsis cases in 2011 that were identified by the Angus implementation was greater than by the Martin implementation (236,932 vs 119,342 cases, respectively). As a result, the annual cost of sepsis readmissions was greater using the Angus implementation compared with the Martin implementation ($1.6 billion vs $612 million, respectively). The patient- and hospital-level factors that were associated with 30-day readmissions were similar between the Martin and Angus implementations (Supplemental Table 9, Supplemental Digital Content 1, http://links.lww.com/CCM/B362).