Study design and data collection
Surviving patients who were firstly diagnosed with cancer from 2010 to 2017 were identified from the Guangzhou Cancer Registry (GCR) of the Guangzhou Center for Disease Control and Prevention (GZCDC) and included in the current study. Information on the diagnosis of cancer was obtained from the electronic medical records in hospitals in Guangzhou. The Guangzhou Cancer Registry was launched in 2008, and the surveillance and follow-up system were established in 2010, which covered residents from all districts of Guangzhou. Data of this study were derived from the GCR. The GCR was approved by the Ministry of Finance of the People’s Republic of China, National Health Commission of the People’s Republic of China, Guangzhou Municipal Finance Bureau and Guangzhou Municipal Health Commission. Ethical approval of this study was obtained from the ethical committee in the GZCDC.
All types of cancer were included in this study. Local surviving patients who were discharged from local hospitals were referred to primary care centers 1 month after discharge from the hospital and completed a validated brief questionnaire-based survey. Information on demographic characteristics, lifestyle factors including smoking status, alcohol use, physical activity, and sleep duration in the past 30 days, and disease history were collected. Anthropometric measurements such as height and weight were measured. Cancer-related information such as date of diagnosis, diagnosis methods and hospital, types of cancer, and treatment history was derived from medical records. Karnofsky Performance Status (KPS) was used to assess the general functional capacity of the cancer survivors [15].
Lifestyle variables
Current smoking was defined by at least 1 cigarette/day or 7 cigarettes/week in the past 30 days [16]. Patients were classified as current smokers if they answered “yes,” former smokers if answered “yes in the past, but have quitted smoking now,” and non-smokers if answered “no.” Besides, alcohol use was defined as the use of alcohol at least 10g/day in the past 30 days [17]. Alcohol use was assessed based on the choices of questions about drinking habits and categorized into three groups: never, former, and current alcohol users. Average time spent in physical activity in the past 30 days was also assessed and categorized into four groups: ≤1 h/week, 2–4 h/week, 5–7 h/week, and >7 h/week [18]. Sleep duration was categorized into three groups: ≤5 h/day, 6–8 h/day, and ≥ 9 h/day [19]. Body mass index (BMI) was calculated based on measured height and weight and was categorized into four groups according to the National Health Commission of the People’s Republic of China: <18.5kg/m2, 18.5–23.9 kg/m2, 24.0–27.9 kg/m2, and ≥28.0 kg/m2 [20].
Assessment of healthy lifestyle score
According to previous studies, the healthy lifestyle score was created by combining the most important lifestyle factors relevant to outcome based on a priori knowledge in a binary point system [7,8,9]. Therefore, a healthy lifestyle score was derived based on five factors associated with cancer mortality, included smoking [21], alcohol use [22], physical activity [23], sleep duration [24], and BMI [25]. Smoking status was categorized into non-smoking and ever-smoking, and alcohol use was categorized into limited alcohol use and alcohol use. Survivors who reported physical activity of ≥ 2 h/week were classified as regular physical activity; otherwise, were classified as inactivity. Sleep duration was classified into 2 categories including insufficient sleep (≤ 5 h/day) and sufficient sleep (≥ 6 h/day), and BMI was classified into 2 categories (<18.5 kg/m2 and ≥18.5 kg/m2). Participants received 1 point for each respective lifestyle factor: nonsmoking, limited alcohol use, regular physical activity, sufficient sleep, or BMI ≥18.5 kg/m2. A combined score (0–5 points) was calculated by summing the scores of these 5 factors. We also categorized the score into four groups (0–2, 3,4, and 5).
Ascertainment of outcomes
Outcomes included all-cause mortality in all survivors and by diagnosis. Overall survival was analyzed as the time from diagnosis to death during the follow-up [26]. Information on vital status was collected from the death registration system in the GZCDC. In the present study, we analyzed the mortality data until December 31, 2019.
Statistical analysis
Person-years of follow-up were calculated from the date of baseline enrollment to death, or the end of the study on December 31, 2019, whichever came first. We used Cox proportional hazards regression models to assess the association of healthy lifestyle score with all-cause mortality risk, giving hazard ratios (HRs) and 95% confidence intervals (95% CIs). Potential confounders such as sex, age, education, treatment (surgery, chemotherapy, radiation therapy, traditional Chinese medicine, biotherapy, intervention, and other treatment) and employment status were adjusted. The proportional hazard assumption was tested by the Schoenfeld residuals method [27], and no significant violation of the assumption was found. We also conducted subgroup analyses to examine the potential effect modification by sex and age groups (<65/≥65 years). Whether the association was modified by sex and age was assessed by likelihood ratio test comparing models with and without interaction terms. Moreover, we also checked for interactions between the lifestyle score and sex or age by using interaction plots.
We used the life table method to calculate each participant’s life expectancy according to different healthy lifestyle scores. The life tables were constructed using three estimates: (1) total number of different healthy lifestyle score in each age group (nPx), (2) the censored number of different healthy lifestyle score in each age group (nCx), and (3) the death toll of different healthy lifestyle score in each age group (nDx). These estimates were used to assess life expectancy for different age intervals using the following methods. Firstly, age-specific all-cause mortality rates (nmx) of different score were calculated as follows [28]: nmx = nDx / (nPx - nCx /2). Secondly, probability of dying was set of 0 at age 55 and set of 1 at more than age 81. The probability of dying (nqx) between age t and t+4 was estimated as [28]: nqx =2n× nmx / (2+n× nmx), where n refers to the age interval. Thirdly, our study applied the predicted survival probabilities(lx) on a hypothetical cohort of 100,000 55-year-old participants to obtain the expected number of deaths in each age interval [t, t+4] [28]. The number of person-years of survival (nLx) within [t, t+4] was estimated as follows [28]: nLx= (lx+ lx+n) ×n/2. The life expectancy at each age group was then calculated by dividing the total person-years that would be lived beyond age t by the number of persons who survived to that age interval [28].
In sensitivity analyses, we further explored whether the associations varied by sex and age groups in survivors of type-specific cancer (breast cancer, colorectal cancer, lung cancer, liver cancer, nasopharynx cancer, gastric cancer, and kidney cancer). In addition, we conducted leave-one-out analyses excluding single lifestyle factor respectively from the combined healthy lifestyle. We also estimated the association between each lifestyle factor and the life expectancy. As both lifestyle factors and mortality could be influenced by demographic factors (sex, age, education), treatment (surgery, chemotherapy, radiation therapy, traditional Chinese medicine, biotherapy, intervention, and other treatment) and employment [29, 30], these variables were considered as potential confounders. Statistical analysis was done using Stata (STATA Corp LP, version 15). Two-sided P values < 0.05 were considered as statistically significant.