The Synergistic Effect of Functional Status and Comorbidity Burden on Mortality: A 16-Year Survival Analysis

Objectives The relationship between disability and comorbidity on mortality is widely perceived as additive in clinical models of frailty. Design National data were retrospectively extracted from medical records of community hospital. Data Sources There were of 12,804 acutely-disabled patients admitted for inpatient rehabilitation in Singapore rehabilitation community hospitals from 1996 through 2005 were followed up for death till 31 December 2011. Outcome Measure Cox proportional-hazards regression to assess the interaction of comorbidity and disability at discharge on all-cause mortality. Results During a median follow-up of 10.9 years, there were 8,565 deaths (66.9%). The mean age was 73.0 (standard deviation: 11.5) years. Independent risk factors of mortality were higher comorbidity (p<0.001), severity of disability at discharge (p<0.001), being widowed (adjusted hazard ratio [aHR]: 1.38, 95% confidence interval [CI]:1.25–1.53), low socioeconomic status (aHR:1.40, 95%CI:1.29–1.53), discharge to nursing home (aHR:1.14, 95%CI:1.05–1.22) and re-admission into acute care (aHR:1.54, 95%CI:1.45–1.65). In the main effects model, those with high comorbidity had an aHR = 2.41 (95%CI:2.13–2.72) whereas those with total disability had an aHR = 2.28 (95%CI:2.12–2.46). In the interaction model, synergistic interaction existed between comorbidity and disability (p<0.001) where those with high comorbidity and total disability had much higher aHR = 6.57 (95%CI:5.15–8.37). Conclusions Patients with greater comorbidity and disability at discharge, discharge to nursing home or re-admission into acute care, lower socioeconomic status and being widowed had higher mortality risk. Our results identified predictive variables of mortality that map well onto the frailty cascade model. Increasing comorbidity and disability interacted synergistically to increase mortality risk.


Introduction
Frailty is an important clinical syndrome that increases an individual's vulnerability to adverse health outcomes and mortality. [1][2][3] The classic frailty cascade model describes how agerelated physiological deterioration, stressful events, functional decline, recovery, disability, hospitalization, and institutionalization ultimately contribute to premature death in older individuals. [4] Older patients admitted for acute illness may have multiple chronic conditions, [5] experience disability, [6] and become institutionalized following discharge from the acute care setting. [7] Development of physical frailty in older individuals is greatly impacted by age-related physiological deterioration and a wide range of diseases and medical conditions. [8,9] Comorbidity is the concurrent presence of two or more medically diagnosed diseases, [3] and is commonly experienced by older individuals. [8][9][10][11] Comorbidity has also been shown to increase the risk of premature death in a variety of patients such as breast cancer, [12] essential hypertension or diabetes [13] and spinal cord injury. [14] Disability can be characterized as dependency in performing activities of daily living [3] and are common in older individuals. Elderly patients are particularly vulnerable to disability and functional decline after acute hospitalizations due to enforced immobilization and deconditioning. [3,5,15] Development of disabilities in older patients results in not only substantial costs, but also increased long term morbidity and mortality. [16][17][18] For example, hip fracture may reduce life expectancy by as much as 25%. [16] Some authors have perceived comorbidity as an etiologic risk factor and disability as an outcome of frailty. [19,20] Other models of frailty such as the frailty cascade, [4] Frailty Index (FI) [21,22] and Frailty Scale (FS) [23,24] include both disability and comorbidity as risk factors of premature mortality outcome. [25,26] Although relationship between frailty and mortality has been shown previously, [27,28] the relative and combined impact of specific aspects of frailty (e.g. comorbidity and disability) remain unclear. The specific aims of this study were to identify predictors of mortality and to describe the combined effect of comorbidity and disability on mortality with reference to models of frailty.

Study Population
The study population was a historical national cohort taken from a database of all patients admitted to all community hospitals in Singapore between 1996 and 2005. Community hospitals are rehabilitation hospitals that cater to patients who are fit for discharge from acute hospitals but require inpatient convalescent or subacute rehabilitative care before returning to a final domiciliary site. [29] We excluded patients who died during their hospital stay in the community hospital (n = 24). Inclusion criteria for this study were:

Data Extraction
Trained research nurses performed data extraction from noncomputerized medical records between November 2005 and August 2008. Multiple iterations of data cleaning and verification were performed. An independent physician analyzed a 10% random sample of patients for data extraction accuracy and the error rate was 0.07%.

Covariates
A number of covariates were controlled to elicit the specific effects of comorbidity and disability including socio-demographic variables such as age group (18-64, $65 years), gender (female, male), ethnicity (Chinese, Malay, Indian, others), marital status (single, married, widowed, separated or divorced), caregiver availability (no, yes) and length of hospital stay. Hospitalization subsidy level served as a proxy measure for socioeconomic status (SES). In Singapore, 81% of public hospitals' beds (B2 and C class) are highly subsidized. [29] B2+ and B1 class lower subsidy and A class have no subsidy. For this study, SES was classified into three groups: high SES (A, B1 and B2+ class), moderate SES (B2 class) and low SES (C class). Discharge destinations of patients were home (n = 9,774), acute hospital (n = 1,583), nursing home (n = 1,251) and others (n = 196). As the numbers who were discharged to other destinations were low, they were excluded from multivariate models.

Exposures of interest
Comorbidity. Comorbidity was measured using the Charlson comorbidity index (CCI). [12] Comorbidity data was manually extracted by trained nurses by reviewing the patients' medical problem list and complete medical records. The CCI measures the burden of medical illnesses and comprises of 19 different categories. The overall CCI score is a summation of weighted scores. A higher score reflects greater cumulative disease burden. We categorized index scores into four groups: no comorbidity (0), low comorbidity (1-3), moderate comorbidity (4-6) and high comorbidity (7)(8)(9)(10)(11)(12)(13)(14)(15)(16). The Charlson comorbidity score also has been shown to have a high inter-rater reliability of a kappa score of 0.93. [30] Disability at discharge. As recommended by the Singapore Ministry of Health guidelines, disability was assessed in both community hospitals at admission and discharge [29] using the Shah modified Barthel Index (BI). This is usually done by qualified physiotherapists and/or occupational therapists in all community hospitals. The BI is a reliable and widely-used assessment for functional status [31] and disability. [32,33] It comprises of 10 items measuring distinct activities of daily living (ADL). The BI utilizes a five-point scale for each ADL and has a maximum score of 100. [34] BI scores are categorized by the severity of disability: 0-24 (total disability), 25-49 (severe disability), 50-74 (moderate disability), 75-90 (mild disability), 91-99 (minimal disability) and 100 (no disability). [34] For this study, the last three categories were collapsed into one category (75-100) and termed ''no or mild disability'' because of small sample sizes. The reliability of the BI has been demonstrated in numerous studies and test-retest, intrarater and inter-rater reliability have been shown to be high by correlation methods (r = 0.87, 0.71-0.99 and 0.75-0.99 respectively). [35,36] We chose to use discharge BI score of patients admitted for rehabilitation because this represented the best functional status after optimization with therapy.

Determination of Endpoint
The primary endpoint was mortality. Time-to-event was calculated from the time of discharge disability assessment to the time of death. Subjects who remained alive at study closure were censored at 31 December 2011. Survival status was obtained from national death databases.

Statistical Analyses
For bivariate analyses, Chi-square tests were performed for categorical variables, log rank test for survival time, Kruskal-Wallis tests for length of stay.
Kaplan-Meier overall survival curves were compared for exposures of interest, namely comorbidity and disability, using two-sided log rank tests. The Cox proportional-hazards regression was implemented to identify predictors of mortality within the cohort. Schoenfeld residuals were used to test the proportionalhazards (PH) assumption after model fitting. In the event of violation of PH assumption for a specific covariate, the same covariate was included as a stratifying factor in the Cox model. Likelihood ratio tests were used to compare nested models to identify significant predictors of mortality. In the synergistic model, the multiplicative interaction term of comorbidity and disability was included in the model. Potential confounders were adjusted in multiplicative interaction model. All analyses were performed using STATA version 11 (StataCorp LP, USA) with the significance level set at 0.05.

Results
19,360 community hospital patient admissions were initially considered. 2,314 (12.0%) non-rehabilitation were first excluded, followed by 2,604 (15.3%) patients with second or subsequent admissions. A total of 14,442 first rehabilitation admission patients were thus recruited. Among them, 1,638 patients (11.3%) had missing discharge disability status. The final cohort consisted of 12,804 patients ( Figure S1), of whom 4,239 (33.1%) were alive at the end of the study. The median follow-up time across the cohort was 10.9 years (interquartile range [IQR]: 8.5-13.3 years). The mean age of the cohort at admission was 73.0 (standard deviation [SD]: 11.5) years, mean discharge BI scores was 60.4 (SD = 29.0), median length of stay was 32.0 days (IQR: 20.0-47.0), and median Charlson comorbidity index score was 3 (IQR: 1-5). Patients who had total disability were more likely to be admitted for stroke (p, 0.001), moderate SES (p,0.001) and discharged to acute hospital (25.4%) compared to other disability groups. Length of stay was not significantly different across the four disability groups (p = 0.053) ( Table 1).

Multivariate analyses
Length of stay was dropped from final model as it became insignificant after adjusting for discharge destination (p = 0.662). Significant predictors of mortality in the final multivariate model were discharge destination, SES group, community hospital and marital status ( Table 3). Patients discharged to nursing homes and acute hospitals had 1.14 (95% CI: 1.05-1.22, p = 0.001) and 1.54 (95% CI: 1.45-1.65, p,0.001) times mortality risk respectively compared to those discharged to home. Patients from moderate or low SES group had 1.12 (95% CI: 1.03-1.22, p = 0.007) and 1.40 (95% CI: 1.29-1.53, p,0.001) times greater mortality risk respectively compared to those in high SES group. Patients who were widowed had 1.38 (95% CI: 1.25-1.53, p, 0.001) times mortality risk compared to singles.
The interaction between comorbidity and disability at discharge was significant in the final synergistic interaction model of mortality risk (p,0.001) (Figure 2, Figure S2). Patients with highest mortality risks had both comorbidity and disability compared to patients with either comorbidity or disability alone   Figure 2).

Discussion
A national cohort of patients admitted for step-down inpatient rehabilitation after acute hospitalization was followed-up for up to 16 years. Comorbidity, discharge disability, widowhood, low socioeconomic status, readmission into acute care and institutionalization were independent predictors of mortality, reflecting the main elements in models of frailty. [20][21][22][23][24] Our results also demonstrated a novel synergistic effect between comorbidity and discharge disability on long-term mortality. [23,24] Our results identified predictive variables of mortality that map well onto the frailty cascade model and quantified the increased risks of each of these factors. [4] For example, in the presence of physiological decline (age-related) across multiple physiological systems aggravated by disease processes (comorbidity), an acute physical event (hospitalization) may cause a negative cascade of events from disability, healthcare utilization (readmission and institutionalization) to eventually death. One other integrated  conceptual model of frailty postulated that life course determinants and diseases may lead to frailty and adverse outcomes. [37][38][39] This model of frailty can also be supported by our findings. For example, widowhood, low socioeconomic status (life course determinants) and comorbidity (diseases) result in disability (physical frailty), institutionalization and death (adverse outcomes). Hence, our results statistically validate earlier theoretical models identifying factors contributing to mortality in frail individuals. Current models of clinical frailty that combine disability and comorbidity on an additive scale may underestimate the mortality risk, especially in the high-risk groups. The interaction between comorbidity and disability is better considered as a ''comorbiditydisability complex''. This complex is a central component of many conceptual models of frailty. [3,4,17,40] However, in contrast with current models that assume an additive effect of comorbidity and disability on mortality risk, our findings provide evidence that the effect of this comorbidity-disability complex on mortality risk is synergistic. To our knowledge, no previous study has reported a combined synergistic effect of comorbidity and disability on longterm mortality risk. One approach to understanding the synergistic interaction effect is to tease out how comorbidity and disability impact each other. For example, it can be conceived that comorbidities increase the likelihood of hospitalizations, leading to disability, decreased mobility, sarcopenia, a higher level of frailty, and potentially further comorbidities. These factors are likely to have bidirectional influences and may continuously propagate in a vicious cycle that perpetuates itself until physical frailty results in premature death. Clinical management of disability, comorbidity and frailty each has its unique challenges. Disabled older patients are at greater risk of social isolation, institutionalization, and new chronic diseases and initiation of frailty and initially of frailty. [3] Hence, medical care for the disabled involves rehabilitation to maximize function and prevent further decline. Fragmented sub-specialized care focused on single disease leads to complications in patients with multiple comorbidities due to complex relationships between conditions and their treatments. [41] Frail patients also have additional needs beyond those of underlying comorbidity and disability as they are vulnerable to other stressors such as hospitalization, under-nutrition and falls. Additional care is therefore needed to treat pathologic causes of progressive weakness, prevent iatrogenesis and reduce risk factors that result in disability. [3]

Strengths and limitations
The strengths of our study are a large study sample with long follow-up of up to 16 years. These provided the opportunity to elicit complex long-term relationships such as the synergistic interaction between comorbidity and disability on mortality. The database tracked all patients admitted to community hospitals during the study period and hence was a nationally representative study population. The limitation of this study is the use of retrospective data and incompleteness of records which could have introduced data entry biases. Furthermore, the study was limited to investigating variables that the database included and did not consider other factors may also affect long-term outcomes (e.g. cognition, mental health, quality of life and healthcare decisionmaking). In addition, we excluded 24 patients who died during their hospital stay. As these patients had poorer admission functional scores (mean = 23.9 vs. 46.3) and greater comorbidity burden (mean = 4.5 vs. 2.9) compared to patients in our study, our current risk estimates would be slightly conservative. Finally, the study was completed in an Asian population within a developed economy and an advanced healthcare system. Generalization of these findings to Western and other populations should be done with care.

Conclusions
This study confirmed that comorbidity and disability are independent predictors of mortality risks in patients after discharge from acute hospitalizations. In addition to widowhood and institutionalization, we also found a novel synergistic interaction effect of the comorbidity-disability complex independent on mortality risk. Future research should consider the feasibility and value of replicating this prospective study in non-Asian populations. The mechanisms through which the comorbidity-disability complex impacts mortality also warrants further investigation.