American Journal of Respiratory and Critical Care Medicine

Rationale: Intensive care unit (ICU)-acquired weakness is a common issue for sepsis survivors that is characterized by impaired muscle strength and causes functional disability. Although inpatient rehabilitation has not been found to reduce in-hospital mortality, the impact of postdischarge rehabilitation on sepsis survivors is uncertain.

Objectives: To investigate the benefit of postdischarge rehabilitation to long-term mortality in sepsis survivors.

Methods: We conducted a nationwide, population-based, high-dimensional propensity score–matched cohort study using Taiwan’s National Health Insurance Research Database. The rehabilitation cohort comprised 15,535 ICU patients who survived sepsis and received rehabilitation within 3 months after discharge between 2000 and 2010. The control cohort consisted of 15,535 high-dimensional propensity score–matched subjects who did not receive rehabilitation within 3 months after discharge. The endpoint was mortality during the 10-year follow-up period.

Measurements and Main Results: Compared with the control cohort, the rehabilitation cohort had a significantly lower risk of 10-year mortality (adjusted hazard ratio, 0.94; 95% confidence interval, 0.92–0.97; P < 0.001), with an absolute risk reduction of 1.4 per 100 person-years. The frequency of rehabilitation was inversely associated with 10-year mortality (≥3 vs. 1 course: adjusted hazard ratio, 0.82; P < 0.001). Compared with the control cohort, improved survival was observed in the rehabilitation cohort among ill patients who had more comorbidities, required more prolonged mechanical ventilation, and had longer ICU or hospital stays, but not among those with the opposite conditions (i.e., less ill patients).

Conclusions: Postdischarge rehabilitation may be associated with a reduced risk of 10-year mortality in the subset of patients with particularly long ICU courses.

Scientific Knowledge on the Subject

The inpatient rehabilitation during the course of sepsis has not been found to reduce in-hospital mortality. However, the impact of postdischarge rehabilitation on sepsis survivors is uncertain.

What This Study Adds to the Field

This comprehensive analysis demonstrates that postdischarge rehabilitation may be associated with a reduced risk of mortality in sepsis survivors, especially those with more comorbidities, more prolonged mechanical ventilation, and longer intensive care unit or hospital stays.

Sepsis is a common and serious medical disorder. Its annual incidence increased from 8.7% in 1979–2000 to 13.3% in 2004–2009 in the United States (1, 2). However, average annual mortality rates among patients with sepsis declined 1.3% between 2000 and 2010 in Australia and New Zealand (3) and 3% between 1993 and 2009 in the United States (4), resulting in a great number of sepsis survivors. Survivors of critical illness may experience muscle wasting, functional disability, and cognitive impairment for a period extending beyond several years after discharge (58). Moreover, muscle wasting and sarcopenia are strongly correlated with all-cause mortality (912). Recent studies have found that rehabilitation may maximize the functional activity of patients with stroke (13), improve the exercise capacity of patients with chronic obstructive pulmonary disease (14), and improve the prognosis of patients after major surgery (15, 16). A meta-analysis of 10 randomized controlled trials (RCTs) reported that inpatient rehabilitation of critically ill patients in intensive care units (ICUs) provided short-term benefits, such as improved physical activity, better respiratory muscle strength, and reduced hospital stays; however, no benefit to in-hospital mortality was observed (17).

Debate about the importance of postdischarge rehabilitation is ongoing. Relevant clinical trials have produced conflicting results regarding physical recovery, and small samples and short follow-up periods of those studies pose barriers to the exploration of the effects of postdischarge rehabilitation on hard endpoints, such as mortality (1821). Moreover, patient compliance with postdischarge rehabilitation is lower than compliance in the ICU or other hospital wards. A pilot study investigating muscle loss in critically ill patients documented the occurrence of muscle wasting within 1 week of admission (22). Furthermore, the effects of muscle wasting continued, even in patients with a greater chance of survival; hence, these patients developed functional disability and, eventually, an increased risk of postcritical mortality. Postdischarge rehabilitation may break this vicious cycle by improving muscle strength and functional activity, thereby reducing mortality (23, 24).

ICU survivors comprise a heterogeneous study population with different admitting diagnoses, rendering determination of the true survival benefit of rehabilitation difficult. These factors prompted us to focus on ICU survivors of sepsis. To reduce potential confounders stemming from inadequate control for comorbidities and underpowered sample sizes, we conducted a nationwide population-based high-dimensional propensity score (hdPS)-matching study to evaluate the association between postdischarge rehabilitation and long-term mortality in ICU survivors of sepsis using Taiwan’s National Health Insurance Research Database (NHIRD).

Data Source

Taiwan’s National Health Insurance (NHI) program, launched in 1995, currently covers 99% of the population of 23 million people. The NHI offers comprehensive medical care, including coverage for outpatient, inpatient, emergency, and dental care; traditional Chinese medicine services; and prescription drugs. The data source for this study was the NHIRD, which consists of deidentified secondary data released for research purposes by the National Health Research Institutes. The database includes all NHI registry and claims data, ranging from demographic data to detailed orders and prescriptions from ambulatory and inpatient care. Individuals’ identities have been encrypted using unique identification numbers, enabling claim linkage within the database. Diseases have been coded according to the International Classification of Disease, Ninth Revision, Clinical Modification (ICD-9-CM). The accuracy of diagnoses in the NHIRD has been validated for several diseases, such as acute kidney injury, chronic kidney disease, diabetes mellitus, and ischemic stroke (2529). Several published studies have been based on the NHIRD (3032). This study was exempted from full review by the Institutional Review Board of Taipei City Hospital because the dataset comprised deidentified secondary data.

Study Design and Sample

This population-based observational cohort study aimed to assess the association between postdischarge rehabilitation and mortality in ICU survivors of sepsis. The study population comprised two cohorts: a rehabilitation cohort and a control cohort that did not receive rehabilitation (Figure 1). ICD-9-CM codes have been used in several previous studies of sepsis (1, 33) and validated in a tertiary medical center in Taiwan (see the online supplement). We identified all patients hospitalized for sepsis for the first time between January 2000 and December 2010. The diagnosis of sepsis was defined by a primary discharge diagnosis of septicemia (ICD-9-CM code 038.x) plus the prescription of antibiotics. We extracted all demographic data, diagnosis and procedure codes, and drug prescription data for the period from January 1995 to December 2011 for included patients, and ensured that all individuals had available data for at least 5 years before inclusion. Baseline information was collected from the 5-year period before inclusion. Patients were excluded for the following reasons: age younger than 18 years, no ICU admission, in-hospital death, and death within 90 days after discharge. The index date was defined as 91 days after discharge from hospitalization for sepsis to avoid immortal time bias.

The rehabilitation cohort was defined as patients who had received rehabilitation within 90 days after discharge. If the patient’s first day of the rehabilitation course was within 90 days after discharge from sepsis, he or she would be included in the rehabilitation cohort. Information on recipients of postdischarge rehabilitation was retrieved from ambulatory and inpatient care data using NHI claims. In many countries, more educated patients and families and those with higher socioeconomic backgrounds or from geographic areas with better provision often have better access to rehabilitation. However, Taiwan's NHI provides low-cost insurance to facilitate access to rehabilitation in hospitals and satellite clinics.

The referral of critical illness survivors to postdischarge rehabilitation is based mainly on the advice of physicians in charge. After consultation, a physiatrist led each rehabilitation team, which included physical and/or occupational therapists, speech pathologists, social workers, and/or athletic trainers, according to the patient’s underlying medical problems and functional status. Rehabilitation protocols consisted of individualized structured exercise regimens, such as facilitating muscle strengthening and movement (e.g., active and passive shoulder elevation, respiratory muscle training, getting out of and/or sitting on the edge of bed, moving from a seated to a standing position with or without assistance, ergometric leg exercises, marching in place, treadmill walking, stair climbing), activities of daily living, cardiovascular capacity, functional ability, and occupational and communication therapy. A dose of rehabilitation was defined as an ambulatory visit to a physiatrist for one course of physiotherapist-led exercise rehabilitation. One rehabilitation course consisted of six sessions (30 min–1 h per session) in a 1-month period. Physiatrists determined the frequency of rehabilitation sessions based on patients’ medical histories, clinical status, and tolerance.

After a complete course of six treatment sessions, patients visited their physiatrists again for the next course of rehabilitation. In addition, patients were encouraged to perform home-based exercise alone and/or with the assistance of family members on days that they did not receive physical therapist–led rehabilitation. Physiatrists provided printed physical education worksheets to patients for this purpose. The control cohort comprised patients hospitalized for sepsis who did not receive rehabilitation within 90 days after discharge.

High-Dimensional Propensity Score Matching

Studies based on large health-care use databases have been criticized for the incompleteness of information on potential confounders, including body mass index, smoking status, lifestyle factors, clinical disease severity, mental status, and laboratory results. In the present study, such factors may have led to the selective prescription of rehabilitation in sepsis survivors, creating residual unmeasured confounders and, in turn, resulting in biased estimates of the association between postdischarge rehabilitation and mortality. The hdPS algorithm was developed to reduce residual confounding of treatment effects in studies based on claims databases (3438), and has been shown to produce results closely approximating expected findings based on RCTs (34).

HdPSs were calculated for all patients in our study using a previously described multistep process (see online supplement) (34, 39). The hdPS algorithm prioritized thousands of variables at the drug, diagnostic, and procedural levels for the periods of 180 days before hospitalization for sepsis and during hospitalization, and at the demographic level according to variables’ potential to cause multiplicative bias in the estimation of exposure-outcome associations (i.e., rate ratios). We selected the 500 empirical variables most likely to cause bias for inclusion in the propensity score model (see Table E1 in the online supplement). We also included the following covariates, which were strongly associated with mortality: age, sex, year and month of index date, monthly income, urbanization, Charlson Comorbidity Index (CCI) score (40), hypertension, dyslipidemia, heart failure, myocardial infarction, cerebrovascular disease, diabetes mellitus, plegia, dementia, and sepsis severity (see Table E2).

The severity of sepsis was determined based on the infection site, occurrence of acute organ dysfunction and shock or respiratory failure, need for hemodialysis, lengths of ICU and hospital stays, duration of mechanical ventilation, receipt of rehabilitation, and use of neuromuscular blockers or steroids during hospitalization. HdPSs were calculated using logistic regression to estimate the predicted probability (as measured by the propensity score) of receiving postdischarge rehabilitation, conditional on all included covariates. We then performed 1:1 matching using the nearest-neighbor method without replacement (caliper width = 0.0193583; 0.1 SD of hdPS logit); for each patient in the rehabilitation cohort, we identified one control patient who was matched according to age (±3 yr) and hdPS.


The primary endpoint was mortality during the 10-year follow-up period. The secondary endpoints were 1-, 3-, and 5-year mortality. Both cohorts were followed until mortality, the 10th year of the follow-up period, or the end of the study period (December 31, 2011).

Statistical Analysis

Descriptive statistics were used to describe the baseline characteristics of the cohorts. These characteristics were compared between groups using Pearson chi-squared tests for categorical variables and independent t tests and Mann-Whitney U tests for parametric and nonparametric continuous variables, respectively. HdPSs for the likelihood of receiving rehabilitation were calculated using a multistep process (34, 39). Incidence rates of mortality were calculated and compared between groups using Poisson regression models. The cumulative incidences of mortality in cohorts were calculated with the Kaplan-Meier method and were compared with the log-rank test. The relative risk of mortality in the rehabilitation cohort in comparison with the control cohort was determined by calculating the hazard ratio (HR) in a Cox regression model using a conditional approach with stratification adjusted for the hdPS (41).We also further stratified analysis by inpatient and outpatient rehabilitation.

Subgroup analyses in the Cox regression model were performed according to age, sex, CCI score, use of mechanical ventilation, lengths of ICU and hospital stays, and number of dysfunctional organs to examine the impact of these factors on the risk of mortality after postdischarge rehabilitation. Interaction was examined using the likelihood ratio test. Finally, we used a bootstrap approach for sensitivity analysis to validate our findings in the Cox regression models. Correspondence between bootstrap HRs was tested using 1,000 replications. We also used several models to validate our findings as described in Tables E3 and E4: (1) HR for mortality before hdPS matching in a Cox regression model with adjusted hdPS to take the effects of excluding subjects after matching into consideration; (2) a multivariable-adjusted Cox model including all sepsis survivors among which postdischarge rehabilitation was calculated as time-varying covariate, and testing the various time windows for rehabilitation after hospital discharge (e.g., 45, 90, 180, and 365 d); and (3) Cox model with sandwich estimator of hospital-level or with shared-frailty model of the individual hospital identifier to adjust for hospital-level variation in patterns of rehabilitation referral. Microsoft SQL Server 2012 (Microsoft Corporation, Redmond, WA) was used for data linkage, processing, and sampling. HdPSs were calculated with SAS software (version 9.2; SAS Institute Inc., Cary, NC) (39). All other statistical analyses were conducted using STATA statistical software (version 12.0; StataCorp, College Station, TX). Statistical significance was defined as P less than 0.05.

Characteristics of the Study Population

A total of 114,059 patients hospitalized for sepsis between January 2000 and December 2011 met the inclusion criteria; 16,239 of these patients received rehabilitation and 97,820 patients did not. Before matching, patients receiving rehabilitation were older and predominantly male, had lower economic status and more comorbid conditions, and received more concomitant medications than those who did not receive rehabilitation. After hdPS matching, 15,535 patients each in the rehabilitation and control cohorts were included in the study. The rehabilitation cohort was largely comparable with the control cohort (Table 1). Although the hdPS model of the present study included the information of 180 days before hospitalization for sepsis, total number of hospitalizations in days −545 to −181 before the sepsis hospitalization did not differ significantly between rehabilitation and control cohorts (rehabilitation cohort [mean, 0.89] vs. matched cohort [mean, 0.90]; P = 0.692).

Table 1. Demographics and Clinical Characteristics of Patients

 Before MatchingMatched
CharacteristicPatients Receiving RehabilitationPatients Not Receiving RehabilitationP ValueRehabilitation CohortControl CohortP Value
No. of patients16,23997,820 15,53515,535 
Male, n (%)9,665 (59.5)53,525 (54.7)<0.0019,194 (59.2)9,094 (59.2)0.249
Age, yr (SD)69.1 (14.8)66.8 (16.7)<0.00169.4 (14.7)69.5 (14.5)0.515
Monthly income, n (%)  <0.001  0.520
 Dependent5,897 (36.3)33,934 (34.7) 5,640 (36.3)5,731 (36.9) 
 NT$ 0–19,1009,255 (57.0)55,681 (56.9) 8,863 (57.1)8,826 (56.8) 
 NT$ 19,100–42,000882 (5.4)6,947 (7.1) 840 (5.4)797 (5.1) 
 >NT$ 42,000205 (1.3)1,258 (1.3) 192 (1.2)181 (1.2) 
Urbanization,* n (%)  <0.001  0.992
 110,631 (65.5)63,517 (64.9) 10,161 (65.4)10,150 (65.3) 
 25,202 (32.0)31,297 (32.0) 4,982 (32.1)4,999 (32.2) 
 3317 (2.0)2,074 (2.1) 304 (2.0)301 (1.9) 
 489 (0.5)932 (1.0) 88 (0.6)85 (0.5) 
Charlson Comorbidity Index score, n (%)  <0.001  0.165
 0 score465 (2.9)5,386 (5.5) 450 (2.9)405 (2.6) 
 1 score1,081 (6.7)8,766 (9.0) 1,043 (6.7)982 (6.3) 
 2 score1,442 (8.9)10,668 (10.9) 1,389 (8.9)1,350 (8.7) 
 3 score1,765 (10.9)11,274 (11.5) 1,712 (11.0)1,700 (10.9) 
 4 score1,939 (11.9)11,398 (11.6) 1,858 (12.0)1,796 (11.6) 
 5 score1,846 (11.3)11,287 (11.5) 1,765 (11.4)1,853 (11.9) 
 ≥6 score7,701 (47.4)39,041 (39.9) 7,318 (47.1)7,449 (47.9) 
Site of infection, n (%)      
 Respiratory10,075 (62.0)52,099 (53.3)<0.0019,595 (61.8)9,685 (62.3)0.293
 Bacteremia221 (1.4)1,027 (1.0)<0.001219 (1.4)204 (1.3)0.463
 GU tract5,743 (35.4)30,037 (30.7)<0.0015,446 (35.1)5,498 (35.4)0.537
 Intraabdomen2,350 (14.5)14,852 (15.2)<0.0012,250 (14.5)2,188 (14.1)0.315
 Wound683 (4.2)3,687 (3.8)<0.001652 (4.2)687 (4.4)0.328
 Device230 (1.4)1,293 (1.3)0.331215 (1.4)228 (1.5)0.534
 CNS212 (1.3)670 (0.7)<0.001190 (1.2)200 (1.3)0.610
 Endocarditis138 (0.8)693 (0.7)0.050134 (0.9)159 (1.0)0.142
 Others925 (5.7)5,887 (6.0)0.109886 (5.7)860 (5.5)0.522
Number of acute organ dysfunction, n (%)  <0.001  0.774
 06,128 (37.7)39,204 (40.1) 5,842 (37.6)5,830 (37.5) 
 16,974 (42.9)41,511 (42.4) 6,687 (43.0)6,638 (42.7) 
 22,726 (16.8)15,115 (15.5) 2,612 (16.8)2,658 (17.1) 
 3387 (2.4)1,892 (1.9) 373 (2.4)393 (2.5) 
 4+24 (0.1)98 (0.1) 21 (0.1)16 (0.1) 
Using steroids, n (%)8,983 (55.3)46,603 (47.6)<0.0018,582 (55.2)8,706 (56.0)0.157
Using neuromuscular blocker, n (%)1,190 (7.3)5,100 (5.2)<0.0011,119 (7.2)1,161 (7.5)0.361
Using inotropic agents, n (%)7,520 (46.3)44,077 (45.1)0.0037,245 (46.6)7,260 (46.7)0.865
Renal-replacement therapy, n (%)9,320 (57.4)53,266 (54.5)<0.0018,972 (57.8)9,142 (58.8)0.050
Using ventilator, n (%)8,539 (52.6)43,152 (44.1)<0.0018,143 (52.4)8,298 (53.4)0.078
Day of ventilator, d (IQR)10 (5–19)8 (4–15)<0.00110 (5-19)10 (5-20)0.097
Day of hospitalization, d (IQR)21 (12–36)16 (10–28)<0.00121 (12-36)22 (13-37)<0.001
Day of ICU, d (IQR)7 (3–14)5 (3–10)<0.0016 (3-14)7 (3-14)0.133
Receiving rehabilitation during hospitalization for sepsis, n (%)5,582 (34.4)7,677 (7.8)<0.0015,138 (33.1)5,138 (33.1)1.000
Comorbidities, n (%)      
 Hypertension12,781 (78.7)67,873 (69.4)<0.00112,185 (78.4)12,331 (79.4)0.042
 Dyslipidemia6,649 (40.9)34,936 (35.7)<0.0016,342 (40.8)6,421 (41.3)0.362
 Heart failure5,013 (30.9)27,381 (28.0)<0.0014,807 (30.9)4,952 (31.9)0.076
 Myocardial infarction1,675 (10.3)9,145 (9.3)<0.0011,600 (10.3)1,640 (10.6)0.458
 Cerebrovascular disease10,418 (64.2)45,004 (46.0)<0.0019,814 (63.2)9,916 (63.8)0.229
 Diabetes8,904 (54.8)48,981 (50.1)<0.0018,504 (54.7)8,632 (55.6)0.144
 Plegia2,841 (17.4)9,305 (9.5)<0.0012,573 (16.6)2,647 (17.0)0.261
 Dementia3,400 (20.9)16,515 (16.9)<0.0013,242 (20.9)3,347 (21.5)0.145
hdPS (SD)0.28 (0.21)0.12 (0.11)<0.0010.26 (0.19)0.26 (0.19)0.346

Definition of abbreviations: CNS = central nervous system; GU = genitourinary; hdPS = high dimensional propensity score; ICU = intensive care unit; IQR = interquartile range; NT$ = new Taiwan dollars.

*Urbanization levels in Taiwan are divided into four strata according to the Taiwan National Health Research Institute publications. Level 1 designates the most urbanized areas, and level 4 designates the least urbanized areas.

Charlson Comorbidity Index score is used to determine overall systemic health. With each increased level of Charlson Comorbidity Index score, there are stepwise increases in the cumulative mortality.

Risk of Mortality

During a mean follow-up period of 2.8 (SD, 2.4) years, 16,985 deaths occurred. The incidence rates of 10-year mortality were 20.6 and 22.0 per 100 person-years in the rehabilitation and control cohorts, respectively. Kaplan-Meier curves showed a survival benefit in the rehabilitation cohort relative to the control cohort (Figure 2). The risk of 10-year mortality was lower in the rehabilitation than in the control cohort (adjusted HR [aHR], 0.94; 95% confidence interval [CI], 0.92–0.97; P < 0.001) (Table 2). The risks of 1-year (aHR, 0.92; 95% CI, 0.88–0.96; P < 0.001), 3-year (aHR, 0.93; 95% CI, 0.90–0.96; P < 0.001), and 5-year (aHR, 0.94; 95% CI, 0.91–0.97; P < 0.001) mortality were also lower in the rehabilitation than in the control cohort.

Table 2. Incidence and Risk of Mortality among Postsepsis Patients

 No. of EventPerson-YearsIncidence RatePropensity Score–matched
HR (95% CI)P ValueHR (95% CI)P Value
1-yr mortality
 Control cohort4,23012,42534.04As ReferenceAs Reference
 Rehabilitation cohort3,97312,69031.310.92 (0.88–0.96)<0.0010.92 (0.88–0.96)<0.001
3-yr mortality
 Control cohort6,93726,76425.92As ReferenceAs Reference
 Rehabilitation cohort6,66127,81723.950.93 (0.90–0.96)<0.0010.93 (0.90–0.96)<0.001
5-yr mortality
 Control cohort7,92033,91123.36As ReferenceAs Reference
 Rehabilitation cohort7,75735,44821.880.94 (0.92–0.97)<0.0010.94 (0.91–0.97)<0.001
10-yr mortality
 Control cohort8,55438,90021.99As ReferenceAs Reference
 Rehabilitation cohort8,43140,86720.630.95 (0.92–0.98)<0.0010.94 (0.92–0.97)<0.001

Definition of abbreviations: CI = confidence interval; HR = hazard ratio.

*Adjusted for high dimensional propensity score.

Per 102 person-years.

Among patients receiving postdischarge rehabilitation, a greater frequency of rehabilitation courses was associated with a lower risk of mortality. Compared with one course, aHRs for two and three or more outpatient rehabilitation courses and one hospitalization for inpatient rehabilitation were 0.77 (95% CI, 0.71–0.83; P < 0.001), 0.71 (95% CI, 0.67–0.74; P < 0.001), and 0.54 (95% CI, 0.48–0.62; P < 0.001), respectively (Table 3).

Table 3. Risk of 10-Year Mortality in Intensive Care Unit Survivors of Sepsis Receiving Postdischarge Rehabilitation

Number of Courses for RehabilitationHR (95% CI)P ValueHR (95% CI)P Value
1As Reference As Reference 
20.83 (0.77–0.90)<0.0010.77 (0.71–0.83)<0.001
≥30.80 (0.76–0.84)<0.0010.71 (0.67–0.74)<0.001
Hospitalization for rehabilitation0.46 (0.39–0.51)<0.0010.54 (0.48–0.62)<0.001

Definition of abbreviations: CI = confidence interval; HR = hazard ratio.

*Adjusted for high dimensional propensity score.

One course consisting of six sessions of physiotherapist-led exercise rehabilitation within 1 month would be arranged after an ambulatory visit to a physiatrist.

Subgroup Analyses of Mortality Risk

As shown in Figure 3, interaction test results for postdischarge rehabilitation were significant for CCI score (Pinteraction = 0.049), duration of mechanical ventilation, and lengths of ICU and hospital stays (all Pinteraction < 0.001). However, higher-order interactions between these factors were not calculated. Subgroup analyses showed that postdischarge rehabilitation significantly decreased the risk of mortality relative to the control cohort in patients with CCI scores greater than or equal to 3 (aHR, 0.94; 95% CI, 0.91–0.97), those who used mechanical ventilation for 29–35 days (aHR, 0.82; 95% CI, 0.68–0.99) and 36–42 days (aHR, 0.72; 95% CI, 0.57–0.89), those with ICU stays greater than or equal to 7 days (aHR, 0.89; 95% CI, 0.86–0.93), and patients with hospital stays greater than or equal to 21 days (aHR, 0.90; 95% CI, 0.86–0.93).

Sensitivity Analysis

Sensitivity analysis confirmed that postdischarge rehabilitation reduced the risk of mortality in ICU survivors of sepsis when assessed by different time periods of diagnosis, excluding control cohort patients who received rehabilitation at any time during the follow-up period, and further excluding patients in both cohorts who were followed for less than 1 year (Table 4). Similar results were obtained in different Cox model before hdPS matching (see Table E3). The results remained consistent in sensitivity analyses including patients in both cohorts who died within 90 days of discharge after sepsis, those receiving rehabilitation within various time windows after discharge, and adjusted for hospital-level variation in patterns of rehabilitation referral (see Figure E4 and Table E4).

Table 4. Bootstrap Sensitivity Analysis of Cox Regression Model for Mortality in Intensive Care Unit Survivors of Sepsis

 No. Event/No. CaseNo. Event/No. ControlAdjusted HR* (95% CI)P Value
Primary analysis8,431/15,5358,554/15,5350.94 (0.92–0.97)<0.001
Index year, 2000–20054,245/5,8924,227/5,8680.95 (0.91–0.99)0.021
Index year, 2006–20114,206/9,6434,338/9,6670.94 (0.90–0.98)0.003
Excluding control cohort receiving rehabilitation during whole follow-up period6,708/11,5508,431/15,5350.78 (0.75–0.80)<0.001
Excluding both cohorts followed-up for less than 1 year and control cohort receiving rehabilitation during whole follow-up period4,458/10,5012,866/6,6990.85 (0.81–0.90)<0.001

Definition of abbreviations: CI = confidence interval; HR = hazard ratio.

*Adjusted for high dimensional propensity score.

Index date was refined as the date 1 year after discharge to avoid immortal time bias.

To our knowledge, this study is the first nationwide cohort study to investigate whether postdischarge rehabilitation is associated with improved long-term survival in ICU survivors of sepsis. The results showed that postdischarge rehabilitation was associated with a 5.6% risk reduction in mortality during the 10-year follow-up period. Furthermore, this association seemed to be dose dependent, but no survival benefit of postdischarge rehabilitation among less ill survivors after sepsis was observed in subgroup analyses.

Most studies of critically ill patients have examined the outcomes of inpatient, but not postdischarge rehabilitation (17, 4246). Because of financial limitations, the long-term effects of rehabilitation after discharge can be difficult to document. In the intensive care setting, the provision of rehabilitation in the acute phase can be challenging because of concerns about wasting patients’ energy needed to fight disease, increasing ICU staff workload, and frequent interruption of rehabilitation programs (47). For many patients, discharge from critical care is the beginning of a journey to recovery from loss of energy, weakness, and functional disability (6, 7). Thus, the impact of postdischarge rehabilitation on long-term prognosis is an important but overlooked issue.

Three RCTs investigating the effects of rehabilitation following critical illness showed no difference in physical activity or cognitive function after a 6- to 12-month follow-up period (18, 19, 21), whereas another RCT found that postdischarge rehabilitation improved physical recovery (20). However, previous studies have not fully examined hard clinical endpoints, such as mortality. Moreover, the real effect of postdischarge rehabilitation may have been misestimated in analyses that were underpowered because of small and/or heterogeneous samples.

No previous study has addressed the clinical impact of rehabilitation on ICU survivors after sepsis. The current study documented 8% and 7.3% reductions in the mortality rate in the rehabilitation cohort compared with the control cohort during the first and third years after discharge, respectively. The reduction in mortality persisted over 5- and 10-year follow-up periods. Because medical practices and rehabilitation protocols may have varied with time, we also analyzed the effect of rehabilitation on mortality in different time cohorts, and obtained consistent results.

In subgroup analyses, mortality rates were lower among patients receiving postdischarge rehabilitation that required mechanical ventilation (for longer periods), had higher CCI scores, and/or had longer ICU or hospital stays. In other words, the impact of postsepsis rehabilitation was greater in patients at greater risk of muscle wasting. However, no interaction was noted between mortality and other variables, including age, sex, and the number of acute organ failures, possibly because those factors did not affect muscle loss directly.

Our study has several strengths. First, it is the largest cohort study of its kind conducted to date, with a 10-year follow-up period. Second, because ICU survivors comprise a heterogeneous group, with different diagnosed and undiagnosed diseases and different treatments and prognoses, we focused only on patients with diagnoses of sepsis drawn from a national population-based claims database. We believe that this specific group of patients served as an ideal group to evaluate the effect of postdischarge rehabilitation because of relative similar clinical pictures. Furthermore, we used hdPS matching to reduce potential biases, such as differences in age, economic status, and number of comorbid conditions between the rehabilitation and control cohorts.

This study also has some limitations. First, the initiation of rehabilitation relied on physician referrals or recommendations, which may have caused selection bias. However, this issue could be partly overcome by the use of hdPS matching. Moreover, the comprehensiveness of NHI coverage eliminates many financial barriers to postdischarge rehabilitation for ICU survivors of sepsis. Second, discharge dates did not necessarily reflect recovery dates in our sample. To avoid survival bias, we conducted a sensitivity analysis to exclude patients who died within 1 year of discharge, which produced consistent results. Third, because the study endpoint was examined based on the intention-to-treat principle, potential exposure contamination in the control group (i.e., postdischarge rehabilitation after 3 mo) may have interfered with the results. We thus performed another bootstrap sensitivity analysis in which patients who received rehabilitation were excluded from the control group, and the results remained consistent. Finally, data on several potentially confounding factors, including obesity, smoking, alcohol use, and psychosocial status, were not available in our dataset. A large RCT is required to validate our findings regarding the impact of postdischarge rehabilitation on long-term mortality, and examining its effects on quality of life and functional outcomes. Based on our data, the sample size required to perform an RCT evaluating the effect of postdischarge rehabilitation in sepsis survivors receiving mechanical ventilation for greater than 7 days (aHR, 0.89; 95% CI, 0.83–0.82, in our study) would be 453 per arm if a two-sided P value of 0.05 and a power of 0.80.

In conclusion, the results of the current study may provide further support for the long-term survival benefit of postdischarge rehabilitation in ICU survivors of sepsis. Sepsis survivors with more comorbidities who require prolonged mechanical ventilation and/or lengthy ICU or hospital stays may be a target subpopulation likely to benefit most from postdischarge rehabilitation.

1. Martin GS, Mannino DM, Eaton S, Moss M. The epidemiology of sepsis in the United States from 1979 through 2000. N Engl J Med 2003;348:15461554.
2. Gaieski DF, Edwards JM, Kallan MJ, Carr BG. Benchmarking the incidence and mortality of severe sepsis in the United States. Crit Care Med 2013;41:11671174.
3. Kaukonen KM, Bailey M, Suzuki S, Pilcher D, Bellomo R. Mortality related to severe sepsis and septic shock among critically ill patients in Australia and New Zealand, 2000-2012. JAMA 2014;311:13081316.
4. Stevenson EK, Rubenstein AR, Radin GT, Wiener RS, Walkey AJ. Two decades of mortality trends among patients with severe sepsis: a comparative meta-analysis*. Crit Care Med 2014;42:625631.
5. Desai SV, Law TJ, Needham DM. Long-term complications of critical care. Crit Care Med 2011;39:371379.
6. Herridge MS, Tansey CM, Matté A, Tomlinson G, Diaz-Granados N, Cooper A, Guest CB, Mazer CD, Mehta S, Stewart TE, et al.; Canadian Critical Care Trials Group. Functional disability 5 years after acute respiratory distress syndrome. N Engl J Med 2011;364:12931304.
7. Iwashyna TJ, Ely EW, Smith DM, Langa KM. Long-term cognitive impairment and functional disability among survivors of severe sepsis. JAMA 2010;304:17871794.
8. Hopkins RO, Jackson JC. Long-term neurocognitive function after critical illness. Chest 2006;130:869878.
9. Strandberg TE, Stenholm S, Strandberg AY, Salomaa VV, Pitkälä KH, Tilvis RS. The “obesity paradox,” frailty, disability, and mortality in older men: a prospective, longitudinal cohort study. Am J Epidemiol 2013;178:14521460.
10. Roshanravan B, Robinson-Cohen C, Patel KV, Ayers E, Littman AJ, de Boer IH, Ikizler TA, Himmelfarb J, Katzel LI, Kestenbaum B, et al. Association between physical performance and all-cause mortality in CKD. J Am Soc Nephrol 2013;24:822830.
11. Arango-Lopera VE, Arroyo P, Gutiérrez-Robledo LM, Pérez-Zepeda MU, Cesari M. Mortality as an adverse outcome of sarcopenia. J Nutr Health Aging 2013;17:259262.
12. Montano-Loza AJ, Meza-Junco J, Prado CM, Lieffers JR, Baracos VE, Bain VG, Sawyer MB. Muscle wasting is associated with mortality in patients with cirrhosis. Clin Gastroenterol Hepatol 2012;10:166173.
13. Saunders DH, Sanderson M, Brazzelli M, Greig CA, Mead GE. Physical fitness training for stroke patients. Cochrane Database Syst Rev 2013;10:CD003316.
14. Beauchamp MK, Evans R, Janaudis-Ferreira T, Goldstein RS, Brooks D. Systematic review of supervised exercise programs after pulmonary rehabilitation in individuals with COPD. Chest 2013;144:11241133.
15. Pack QR, Goel K, Lahr BD, Greason KL, Squires RW, Lopez-Jimenez F, Zhang Z, Thomas RJ. Participation in cardiac rehabilitation and survival after coronary artery bypass graft surgery: a community-based study. Circulation 2013;128:590597.
16. Rideout A, Lindsay G, Godwin J. Patient mortality in the 12 years following enrolment into a pre-surgical cardiac rehabilitation programme. Clin Rehabil 2012;26:642647.
17. Kayambu G, Boots R, Paratz J. Physical therapy for the critically ill in the ICU: a systematic review and meta-analysis. Crit Care Med 2013;41:15431554.
18. Denehy L, Skinner EH, Edbrooke L, Haines K, Warrillow S, Hawthorne G, Gough K, Hoorn SV, Morris ME, Berney S. Exercise rehabilitation for patients with critical illness: a randomized controlled trial with 12 months of follow-up. Crit Care 2013;17:R156.
19. Cuthbertson BH, Rattray J, Campbell MK, Gager M, Roughton S, Smith A, Hull A, Breeman S, Norrie J, Jenkinson D, et al.; PRaCTICaL study group. The PRaCTICaL study of nurse led, intensive care follow-up programmes for improving long term outcomes from critical illness: a pragmatic randomised controlled trial. BMJ 2009;339:b3723.
20. Jones C, Skirrow P, Griffiths RD, Humphris GH, Ingleby S, Eddleston J, Waldmann C, Gager M. Rehabilitation after critical illness: a randomized, controlled trial. Crit Care Med 2003;31:24562461.
21. Elliott D, McKinley S, Alison J, Aitken LM, King M, Leslie GD, Kenny P, Taylor P, Foley R, Burmeister E. Health-related quality of life and physical recovery after a critical illness: a multi-centre randomised controlled trial of a home-based physical rehabilitation program. Crit Care 2011;15:R142.
22. Puthucheary ZA, Rawal J, McPhail M, Connolly B, Ratnayake G, Chan P, Hopkinson NS, Phadke R, Dew T, Sidhu PS, et al. Acute skeletal muscle wasting in critical illness. JAMA 2013;310:15911600.
23. Vivodtzev I, Pépin JL, Vottero G, Mayer V, Porsin B, Lévy P, Wuyam B. Improvement in quadriceps strength and dyspnea in daily tasks after 1 month of electrical stimulation in severely deconditioned and malnourished COPD. Chest 2006;129:15401548.
24. Nuhr MJ, Pette D, Berger R, Quittan M, Crevenna R, Huelsman M, Wiesinger GF, Moser P, Fialka-Moser V, Pacher R. Beneficial effects of chronic low-frequency stimulation of thigh muscles in patients with advanced chronic heart failure. Eur Heart J 2004;25:136143.
25. Lindblad U, Råstam L, Ranstam J, Peterson M. Validity of register data on acute myocardial infarction and acute stroke: the Skaraborg Hypertension Project. Scand J Soc Med 1993;21:39.
26. Lin CC, Lai MS, Syu CY, Chang SC, Tseng FY. Accuracy of diabetes diagnosis in health insurance claims data in Taiwan. J Formos Med Assoc 2005;104:157163.
27. Wu CY, Chan FK, Wu MS, Kuo KN, Wang CB, Tsao CR, Lin JT. Histamine2-receptor antagonists are an alternative to proton pump inhibitor in patients receiving clopidogrel. Gastroenterology 2010;139:11651171.
28. Cheng CL, Kao YH, Lin SJ, Lee CH, Lai ML. Validation of the National Health Insurance Research Database with ischemic stroke cases in Taiwan. Pharmacoepidemiol Drug Saf 2011;20:236242.
29. Wu VC, Wu CH, Huang TM, Wang CY, Lai CF, Shiao CC, Chang CH, Lin SL, Chen YY, Chen YM, et al.; NSARF Group. Long-term risk of coronary events after AKI. J Am Soc Nephrol 2014;25:595605.
30. Wang KL, Liu CJ, Chao TF, Huang CM, Wu CH, Chen SJ, Chen TJ, Lin SJ, Chiang CE. Statins, risk of diabetes, and implications on outcomes in the general population. J Am Coll Cardiol 2012;60:12311238.
31. Wu CY, Chen YJ, Ho HJ, Hsu YC, Kuo KN, Wu MS, Lin JT. Association between nucleoside analogues and risk of hepatitis B virus–related hepatocellular carcinoma recurrence following liver resection. JAMA 2012;308:19061914.
32. Ou SM, Chen YT, Chao PW, Lee YJ, Liu CJ, Yeh CM, Chen TJ, Chen TW, Yang WC, Li SY. Nonsteroidal anti-inflammatory drug use is associated with cancer risk reduction in chronic dialysis patients. Kidney Int 2013;84:198205.
33. Iwashyna TJ, Odden A, Rohde J, Bonham C, Kuhn L, Malani P, Chen L, Flanders S. Identifying patients with severe sepsis using administrative claims: patient-level validation of the angus implementation of the international consensus conference definition of severe sepsis. Med Care 2014;52:e39e43.
34. Schneeweiss S, Rassen JA, Glynn RJ, Avorn J, Mogun H, Brookhart MA. High-dimensional propensity score adjustment in studies of treatment effects using health care claims data. Epidemiology 2009;20:512522.
35. Schneeweiss S, Gagne JJ, Glynn RJ, Ruhl M, Rassen JA. Assessing the comparative effectiveness of newly marketed medications: methodological challenges and implications for drug development. Clin Pharmacol Ther 2011;90:777790.
36. Rassen JA, Avorn J, Schneeweiss S. Multivariate-adjusted pharmacoepidemiologic analyses of confidential information pooled from multiple health care utilization databases. Pharmacoepidemiol Drug Saf 2010;19:848857.
37. Toh S, García Rodríguez LA, Hernán MA. Confounding adjustment via a semi-automated high-dimensional propensity score algorithm: an application to electronic medical records. Pharmacoepidemiol Drug Saf 2011;20:849857.
38. Dormuth CR, Hemmelgarn BR, Paterson JM, James MT, Teare GF, Raymond CB, Lafrance JP, Levy A, Garg AX, Ernst P; Canadian Network for Observational Drug Effect Studies. Use of high potency statins and rates of admission for acute kidney injury: multicenter, retrospective observational analysis of administrative databases. BMJ 2013;346:f880.
39. Rassen JA, Doherty M, Huang W, Schneeweiss S. Pharmacoepidemiology toolbox [accessed 2014 Mar 1]. Boston, MA. Available from:
40. Charlson ME, Pompei P, Ales KL, MacKenzie CR. A new method of classifying prognostic comorbidity in longitudinal studies: development and validation. J Chronic Dis 1987;40:373383.
41. Austin PC. A critical appraisal of propensity-score matching in the medical literature between 1996 and 2003. Stat Med 2008;27:20372049.
42. Kayambu G, Boots RJ, Paratz JD. Early rehabilitation in sepsis: a prospective randomised controlled trial investigating functional and physiological outcomes The i-PERFORM Trial (Protocol Article). BMC Anesthesiol 2011;11:21.
43. Parry SM, Berney S, Koopman R, Bryant A, El-Ansary D, Puthucheary Z, Hart N, Warrillow S, Denehy L. Early rehabilitation in critical care (eRiCC): functional electrical stimulation with cycling protocol for a randomised controlled trial. BMJ Open 2012;2:2.
44. Hough CL. Improving physical function during and after critical care. Curr Opin Crit Care 2013;19:488495.
45. Schweickert WD, Pohlman MC, Pohlman AS, Nigos C, Pawlik AJ, Esbrook CL, Spears L, Miller M, Franczyk M, Deprizio D, et al. Early physical and occupational therapy in mechanically ventilated, critically ill patients: a randomised controlled trial. Lancet 2009;373:18741882.
46. Mendez-Tellez PA, Nusr R, Feldman D, Needham DM. Early physical rehabilitation in the ICU: a review for the neurohospitalist. Neurohospitalist 2012;2:96105.
47. Pawlik AJ, Kress JP. Issues affecting the delivery of physical therapy services for individuals with critical illness. Phys Ther 2013;93:256265.
Correspondence and requests for reprints should be addressed to Yung-Tai Chen, M.D., Department of Nephrology, Taipei City Hospital Heping Fuyou Branch, Taipei, 112, Taiwan. E-mail: ; or Shuo-Ming Ou, M.D., Department of Nephrology, Taipei Veterans General Hospital, Taipei, 11217, Taiwan. E-mail:

*These authors contributed equally.

Author Contributions: Study concept and design, P.-W.C., C.-J.S., Y.-J.L., C.-M.T., S.-C.K., Y.-N.S., S.-M.O., and Y.-T.C. Acquisition of data, P.-W.C., Y.-T.C., and C.-J.S. Analysis and interpretation of data, P.-W.C., C.-J.S., Y.-J.L, C.-M.T., S.-C.K., Y.-N.S., D.-C.T., S.-M.O., and Y.-T.C. Drafting of the manuscript, P.-W.C., C.-J.S., D.-C.T., S.-M.O., and Y.-T.C. Statistical analysis, Y.-T.C. and S.-M.O. Administrative, technical, or material support, P.-W.C., C.-J.S., K.-T.C., S.-Y.L. D.-C.T., S.-M.O., and Y.-T.C. Study supervision, Y.-T.C. and S.-M.O.

Originally Published in Press as DOI: 10.1164/rccm.201406-1170OC on September 11, 2014

This article has an online supplement, which is accessible from this issue's table of contents at

Author disclosures are available with the text of this article at

Comments Post a Comment

New User Registration

Not Yet Registered?
Benefits of Registration Include:
 •  A Unique User Profile that will allow you to manage your current subscriptions (including online access)
 •  The ability to create favorites lists down to the article level
 •  The ability to customize email alerts to receive specific notifications about the topics you care most about and special offers
American Journal of Respiratory and Critical Care Medicine

Click to see any corrections or updates and to confirm this is the authentic version of record