2.1. Schizophrenia cohort

The research protocol was approved by the Ethical Committee for Human Research at the National Cheng Kung University. The Taiwanese National Health Insurance (NHI) program is a universal health program that was introduced in March 1995. By the end of 1999, approximately 99% of the Taiwanese population had enrolled in the NHI program. Therefore, the NHI database is representative of the entire Taiwan population. The NHI databases include a registry of beneficiaries, ambulatory care claims, inpatient claims, prescriptions dispensed at pharmacies, a registry of medical facilities, a registry of board-certified specialists and a registry of catastrophic illnesses. After careful review and approval by the Institutional Review Board, these databases were linked to individual personal identification numbers (PINs) to provide patient-level information on demographic characteristics, and PINs were then encrypted prior to data release for research purposes. The registry of catastrophic illnesses features up-to-date information regarding severe illnesses, including schizophrenia, dementia, end-stage renal disease, cancer, etc. In general, a patient must be evaluated and diagnosed by a board-certified physician in the specific field in order to write a certificate for an application, which then must be validated by another specialist before approval for registration as a victim of such illnesses. For patients with a catastrophic illness, co-payments for outpatient or inpatient care can be waived. We identified 52,299 patients who were diagnosed with schizophrenia (International Classification of Diseases, Ninth Revision, Clinical Modification [ICD-9-CM], code 295) from the catastrophic illness register of the NHI during the period of 2000–2008. The index date of follow-up was defined as the date of the first application for a catastrophic illness certificate. In order to match the cross-sectional survey data in terms of age (the patient age of diagnosis ranged from 15 to 50 years in the survey data), we restricted the cohort to within the age range of 15–50 years, resulting in a final cohort of 42,339 patients. The all-cause mortality was extracted from the catastrophic illness records, which noted the occurrence of death. The NHI administration regularly checks patient status (death or invalidity) and notes these checks in the catastrophic illness records every month.

2.2. Collection of data regarding functionality and functional levels

Functionality was measured by conducting a cross-sectional survey of patients with schizophrenia analyzed after the date of first diagnosis. The sample size was 230 patients from 2 medical centers and one mental hospital in Taiwan (day-care wards, community rehabilitation centers, outpatient clinics). The inclusion criteria were: patients meeting the DSM-IV criteria for schizophrenia, and currently aged 20–60 years (age of diagnosis from 15 to 50 years). The exclusion criteria were: a) a severe and unstable major medical disease or a history of neurological disease; b) a history of alcohol or substance dependence or abuse, except for nicotine dependence; c) a history of head injury; d) receipt of electroconvulsive therapy within the previous 6 months; e) an intelligence quotient <70; f) severe vision deficiency, e.g., color blindness or any corrected visual acuity <0.5; and g) illiteracy or disability, and inability to read traditional Chinese characters.

The tool applied for assessment of the patients was the Personal and Social Performance scale (PSP), administered by psychiatrists in each hospital during the one-year study period. The PSP was developed based on the social functioning component of the Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition (DSM-IV) and the social occupational functioning assessment scale (SOFAS). The reliability and validity of the PSP have been investigated in previous studies, and high test–retest reliability, a good inter-rater reliability [Reference Wu, Lin, Tseng, Liu, Chen and Huang8], and high levels of correlation with the Global Assessment of Functioning (GAF) scale, the Activities of Daily Living (ADL) scale, the SOFAS scale and the Positive and Negative Syndrome Scale (PANSS) were observed [Reference Nasrallah, Morosini and Gagnon9, Reference Hsieh, Huang, Wang, Liu, Bai and Chen10]. The PSP consists of four domains: socially-useful activities; personal and social relationships; self-care; and disturbing and aggressive behaviors. Each domain is rated on a six-point Likert scale from 1 to 6, representing absent, mild, manifest but not marked, marked, severe and very severe. Higher domain scores indicated poorer functioning in that particular domain. A global score, which reflected global functioning, was derived from the levels of impairment in each of the four domains, ranging from 1 to 100. Higher global scores indicated better functioning on the global level. In this study, functional disability was measured according to the scores for each domain (a higher score indicating poorer functioning) and the global score (a lower score indicating poorer functioning).

2.3. Survival analysis and extrapolation to estimate life expectancy and expected years of life lost

We applied the Kaplan-Meier method to estimate the time-to-death of the schizophrenia cohort (obtained from the NHI catastrophic illness database). Lifetime survival was estimated using the semi-parametric extrapolation method [Reference Hwang, Tsauo and Wang11, Reference Jackson, Stevens, Ren, Latimer, Bojke and Manca12]. The lifetime survival of the schizophrenia cohort was estimated for up to 900 months using a simple linear regression extrapolation of a logic transformation curve to assess the survival ratio between the schizophrenia cohort and a year-, age- and gender-matched reference population, which was generated by the Monto Carlo simulation method from life tables for the general population of Taiwan. The survival ratio between the schizophrenia cohort and the reference population assumed a constant excess hazard, the detailed method and mathematical proof of which have been described in a previous study [Reference Leng, Chou, Lin, Yang and Wang13]. The standard error of survival estimate was obtained through a bootstrap method of repeated sampling with replacement from a real dataset 100 times. The life expectancy (LE) of the schizophrenia cohort was estimated by summing the area under the estimated lifetime survival curve. The expected years of life lost (EYLL) was calculated as the difference in LE between the schizophrenia cohort and the age- and gender-matched reference population.

To validate the extrapolation method, the relative bias between the predictions and real values was estimated: the LE estimate (using extrapolation based on the first 5 years of follow-up) was compared with the Kaplan Meier estimate based on the actual 9-year follow-up period. The relative bias was less than 1% for the whole study population and for gender-stratified groups. The details are presented in Supplementary Table 1.

2.4. Estimations of proportions of patients with functional disabilities at different functional levels

Functional disability does not remain constant over time in patients with schizophrenia, and changes according to the duration of illness; therefore, a kernel-smoothing method was used to estimate the dynamic changes in functional disability [Reference Hwang, Tsauo and Wang11]. To synchronize with clinical significance, we converted a patient’s PSP score into two functional levels: less (mild) difficulty (global score: 71–100; domain score: 1–2) and manifested disability (global score ≤70; domain score: 3–6). We classified the indicator of manifested disability as 1; else, it would be 0. The proportion of patients of certain levels of functional disability at time t can be obtained by the function , where N(t) is the number of patients and G(t) is the size of the population at time t. The proportions of patients at certain levels of functional disability across different time periods can be obtained by a kernel-smoothing method. Curves of dynamic changes of the proportions of patients with functional disability at different functional levels were obtained by calculating the moving average of the nearby or neighboring 10% to estimate the proportions of patients at different times. Hung et al. published a more detailed method of calculation [Reference Hung, Sung, Chang, Hwang and Wang14]. In this study, the kernel-smoothed curves were obtained from the four domains of the PSP to illustrate dynamic changes in the proportions of patients with functional disability in particular domains over time.

2.5. Expected years of living with disability

To integrate the proportion of patients with functional disability and lifetime survival, we then multiplied the lifetime probability of survival at each time point by the proportions of patients with varying levels of different types of functional disability (four domains and global score) to obtain a lifetime functional disability-adjusted survival [Reference Hung, Sung, Chang, Hwang and Wang14]. The area under the curve was the functional disability-adjusted life expectancy of the patients with schizophrenia, as shown in Supplementary Fig. 2. The unit of the area under the curve was years; thus, the area under the curve was the expected years of living with disability (EYLD), one EYLD representing 1 year of mild or manifested functional disability depending on the functional level; hence, 20 EYLDs represented 20 years of living with manifested disability for each patient. The equation used to estimate the lifetime durations at different functional levels can be illustrated as follows [Reference Hung, Sung, Chang, Hwang and Wang14 ]: represents the estimated lifetime survival function, is the proportion of patients of a specific functional level, and i is the functional level.

2.6. Statistical analysis

Descriptive statistics were used to summarize the characteristics of the study subjects. Apart from gender (n [%]), all variables were described using the mean (standard deviation [SD]). For inferential statistics, Chi-square tests and Student t-tests were used for comparison of gender and age between the schizophrenia cohort obtained from the catastrophic illness data and the subjects enrolled in the cross-sectional survey. The proportions of patients with functional disability were obtained by a kernel-smoothing method with a moving average of the neighboring 10% of data points from the survey data. The Kaplan-Meier method was used to estimate the time-to-death of the schizophrenia cohort (obtained from the registry of catastrophic illnesses in the NHI database). The LE was calculated for the schizophrenia cohort by summing the area under the estimated lifetime survival curve, which was estimated by the Monto Carlo simulation method from life tables for the general population of Taiwan. The EYLL was calculated for the schizophrenia cohort as the difference in LE between the schizophrenia cohort and the age- and gender-matched general population (obtained from life tables). EYLD was the years of disability-adjusted survival expectancy, which was obtained according to the proportion of patients with functional disability (from the survey data) multiplied by the lifetime survival probability (from the schizophrenia cohort) over time throughout the patients' life. Detailed calculations for LE, EYLL and EYLD are presented in parts 2.4 and 2.5 of the Methods section. Regarding gender differences in the changes of functional disability (global and each domain) over time, we stratified the study groups by gender and compared the differences in LE, EYLL and EYLD in the female and male patients. In order to ensure internal consistency of estimations of LE, EYLL and EYLD, sensitivity analysis was performed on a random sample of 100 patients who had completed the cross-sectional survey. All significance levels were two-sided; P < 0.05. We performed statistical analyses using SAS software (version 9.3 for Windows; SAS Institute Inc., Cary, NC, USA). Estimates of LE, EYLL and EYLD were easily obtained using iSQoL software, developed by Hwang’s team and free to be downloaded from the following website: http://www.stat.sinica.edu.tw/isqol/.