Healthy Lifestyle Linked to Lower Risk of Death and New Cancers in People with Metabolic Syndrome

This study obtained data from the UK Biobank. The UK Biobank, an ongoing cohort, has recruited more than 0.5 million individuals between 37 and 70 years across the UK. Participants’ demographic characteristics, biological samples, and potential health-related information were investigated between 2006 and 2010, and their health and medical information were continually followed up [18]. The inclusion criteria for this study were: (1) individuals with MetS at recruitment and the baseline age ≥40 years. The exclusion criteria were: (1) individuals diagnosed with cancer at baseline, (2) who died or had cancer within the first three years after recruitment, and (3) lost to follow-up or incomplete baseline information. Among about 0.5 million participants, 137,591 participants with MetS were included. We further excluded those with cancers at baseline (n = 12,317), lost to follow-up (n = 296), and those aged <40 years (n = 2). In addition, a total of 30,041 participants with missing information and 7593 participants who died or had cancer within the first three years after recruitment were also excluded. Overall, 87,342 participants with MetS at baseline were included in this study (Supplementary Material Figure S1).

The UK Biobank was approved by the Research Ethics Committees of the Northwest Multi-Centre (reference no. 16/NW/0274). All the study participants signed an informed consent form. This study was performed under the approval of the UK Biobank (Application number: 78563).

MetS was defined as the occurrence of three or more of the following risk factors according to the most recent joint interim statement of the International Diabetes Federation (IDF) and National Heart, Lung and Blood Institute (NHLBI)/American Heart Association (AHA): central obesity, elevated BP, increased fasting blood glucose (FBG), reduced high-density lipoprotein cholesterol (HDL-C), and increased triglycerides (TG) [19]. Waist circumference thresholds, which determined central obesity, were measured according to sex and population specifications, such as ≥85 cm for men and ≥80 cm for women in China, and ≥102 cm for men and ≥88 cm for women in Europe [19]. An elevated FBG level was defined as FBG ≥ 100 mg/dL, or a diagnosis of diabetes mellitus (DM), or the use of DM medication. Elevated BP was defined as systolic BP ≥ 130 mmHg and/or a diastolic BP ≥ 85 mmHg, a diagnosis of hypertension, or a history of antihypertensive drug treatment. Lower HDL-C level was defined as HDL-C < 1.0 mmol/L for men and < 1.3 mmol/L for women or with a history of medical treatment for reduced HDL-C. Raised TG was defined as TG ≥ 1.7 mmol/L or treatment history.

Healthy lifestyle scores were created based on seven recommended lifestyle behaviours (moderate alcohol consumption, never/quit smoking, good sleep, healthy diet, sufficient physical activity, social support, and less sedentary behaviour) [9,20], obtained at recruitment through touchscreen questionnaires. Moderate alcohol consumption was defined as alcohol consumption of ≤14 g/day and ≤28 g/day for women and men, respectively [21]. Never smoked and quitting smoking were classified as healthy levels [22]. Good sleep was defined as a sleep duration of 7–8 h/night without snoring [23]. A healthy diet was defined as containing at least four of the recommended food intakes as follows: adequate intake of fruits, vegetables, fish, and whole grains; lower intake of processed meats, unprocessed red meats, and refined grains [24]. Sufficient exercise was defined as moderate exercise ≥150 min/week or ≥5 d/week, vigorous exercise ≥75 min/week or at least once a week, or an equivalent combination [25,26]. Social support was described as having at least two of the following social connections: friend/family visits at least once a week, being able to confide in someone close, having a close relationship with family, or having leisure/social activities [27,28]. Less sedentary behaviour was defined as the sum of time spent driving, watching TV, and using a computer ≤4 h/day [29,30]. Participants were given one point for each of the recommended behaviours. The total score possible ranged from 0 to 7 points, with a higher score indicating a healthier lifestyle (Supplementary Material Table S1).

The weighted healthy lifestyle scores were constructed by using the original binary lifestyle factors as below: weighted healthy lifestyle scores = (β₋₁ × factor₋₁ + β₋₂ × factor₋₂ + …+ β₋₇ × factor₋₇) ÷ (β₋₁ + β₋₂ +…+ β₋₇) × 7 [24]. The βcoefficients represented the association of each binary lifestyle factor with the outcome of all-cause mortality or overall cancer incidence that were assessed by using Cox regression with adjustment for model three as below. The weighted healthy lifestyle scores were subsequently categorised as: unfavourable (lowest quintile), intermediate (quintiles 2–4), and favourable (highest quintile) lifestyles.

The main outcome of this study was all-cause mortality. Mortality information was recorded by the death registry, which is linked to the National Health Service (NHS) Digital Centre (participants from England/Wales) and NHS Central Register (participants from Scotland) [31]. The cause of death was based on the 10th revision of the International Classification of Diseases (ICD-10), which included cancer (C00-C97), CVD (I05-I89), respiratory disease (J09-J99), digestive disease (K20-K93), and others. The follow-up time on the outcome of mortality, which was measured in person-years, was calculated from baseline to the date of death, or until 30 September 2021 (England/Wales participants) or 31 October 2021 (Scotland participants).

The secondary outcome was the first occurrence of incident cancer. The details of cancer incidence were obtained from the cancer registry, which is linked to the Medical Research Information Service of the NHS (participants from England/Wales) and the Information Services Division of the NHS Scotland (participants from Scotland). Death and cancer registries of UK Biobank have been proven to have high integrity and reliability [32,33]. Incident cancers were recorded using the ICD-10 (C00–C97), ICD-9 (140–208), or self-reported cancer (−1 to 9999). The follow-up of the outcome of incident cancer, as measured in person-years, was calculated from baseline to the first occurrence of cancer, death for any reason, or until 31 October 2015 (participants in Scotland) or 29 February 2020 (England/Wales participants), whichever occurred first (Supplementary Material Table S2).

Covariates were collected through touchscreen questionnaires at recruitment. These included sociodemographic information, such as age (years), sex (male or female), ethnicity (white, Asian, black, or mixed), education (college/university degree, others), employment status (employed, unemployed), Townsend deprivation index (TDI) (lowest quintile, quintiles 2–4, highest quintile); health status, such as BMI, medication treatment history for anti-cholesterol drug (yes, no), antihypertensive drug (yes, no), insulin (yes, no), family history of hypertension (yes, no), DM (yes, no), or cancer (yes, no); and the number of MetS component traits (3, 4, or 5). TDI is a measure of socioeconomic status that combines information on unemployment, non-car ownership, non-homeownership, and household overcrowding at baseline, with a higher score indicating a lower socioeconomic status [34]. BMI, an international standard for measuring body fat, was calculated using the formula: BMI = weight (kg)/height (m)², and divided into three groups: <25 kg/m² (underweight or normal weight), 25–30 kg/m² (overweight), and 30 kg/m² (obese), according to the World Health Organization [35]. The participants’ weight and height were measured by well-trained nurses.

The baseline characteristics of the participants were summarised based on all-cause mortality and cancer incidence. Continuous variables were presented as mean ± standard deviation (normally distributed) or median (interquartile range) (non-normally distributed) and were examined using Student’s t-test or Wilcoxon rank test. Categorical variables were presented as frequencies (%) and analysed using Pearson’s χ² test. It should be mentioned that even minor differences can be statistically significant in a large sample size. The median follow-up person-years were calculated from the baseline survey to the date of each endpoint using the reverse Kaplan–Meier method.

Cox proportional hazard regressions were conducted to examine correlations between lifestyle factors and categories, all-cause mortality, cause-specific mortality, and incident cancer. The results were presented as hazard ratios (HRs) and 95% confidence intervals (CIs). We included age at baseline (years), sex (male, female), and ethnicity (white, Asian, black, or mixed) in model one and additionally adjusted for education (college/university degree, others), employment status (employed, unemployed), and TDI (lowest quintile, quintiles 2–4, highest quintile) in model two, and further adjusted for BMI category, medication treatment history for anti-cholesterol drug (yes, no), antihypertensive drug (yes, no), and insulin (yes, no), family history of hypertension (yes, no), DM (yes, no), cancer (yes, no), and the number of MetS components (3, 4, and 5) in model three. We performed the Schoenfeld residuals test to examine the proportional hazards assumption, and no evidence of disproportionality was found (Supplementary Material Figures S2–S7) [36]. Stratified and crossover analyses were performed according to age at baseline; sex; education; employment status; TDI; BMI; medication use history of anti-cholesterol drugs, antihypertensive drugs, and insulin; and family history of hypertension, DM, and cancer. Interactions were examined using the interaction term beta in the entire model. The population-attributable risk percent (PAR%) was calculated using the following formula [37]:

where P_e is the proportion of the population exposed to the non-low-risk group, and HR is the multivariate-adjusted Cox regression hazard ratio of those participants in the full models.

Sensitivity analyses were performed to examine the robustness of the estimates. First, given the large sample size, we excluded participants with missing values of subjective data to avoid imputations affecting the authenticity of the subjective data. Second, we excluded participants with baseline cancer, those who died, or those who were diagnosed with cancer in the first three years of follow-up to rule out potential reverse causation. Third, the unweighted healthy lifestyle scores were classified into three groups: scoring 0–2, 3–5, and 6–7, to investigate all-cause mortality, cause-specific mortality, and incident cancer risk. Fourth, we reconstructed lifestyle scores by defining 18.5 ≤ BMI < 25 kg/m² as the recommended level. Fifth, never smoking was redefined as recommended behaviour. All statistical analyses were performed using Stata (version 15.0; Stata Corp, College Station, TX, USA), and graphs were plotted using R (version 4.1.3; R Core Team (2022). R Foundation for Statistical Computing, Vienna, Austria). All p-values were two-sided, and p-values < 0.05 were considered statistically significant.

The baseline characteristics of participants, according to all-cause mortality and overall cancer incidence, are presented in Table 1. Among the 87,342 participants with MetS, 6739 deaths with a median follow-up period of 12.54 years (1,074,630 total person-years), and 10,802 new cases of cancer with a median follow-up period of 10.69 years (882,973 total person-years) were documented. Compared to surviving participants, those who died were more likely to be older, male, unemployed, have lower educational levels, higher BMI, and lower socioeconomic status (p < 0.05). The characteristics of participants with MetS excluded from the analysis are shown in Supplementary Material Table S3. Lifestyle scores were normally distributed (Supplementary Material Figure S8). The baseline characteristics of MetS according to weighted and unweighted lifestyle scores are presented in Supplementary Material Tables S4 and S5.

As shown in Figure 1, after total adjustment for potential confounders, never/quit smoking (HR: 0.51; 95% CI: 0.48–0.54), moderate alcohol consumption (HR: 0.89; 95% CI: 0.84–0.94), good sleep (HR: 0.88; 95% CI: 0.84–0.93), healthy diet (HR: 0.89; 95% CI: 0.85–0.94), sufficient exercise (HR: 0.84; 95% CI: 0.80–0.88), social support (HR: 0.93; 95% CI: 0.89–0.98), and less sedentary behaviour (HR: 0.94; 95% CI: 0.89–0.99) were all associated with a lower all-cause mortality risk compared to the corresponding unrecommended behaviour (p-value < 0.05). Never/quit smoking (HR: 0.82; 95% CI: 0.77–0.87) was associated with a lower risk of overall cancer incidence (Supplementary Material Table S6).

When integrating these seven factors, by considering each β lifestyle factor coefficient into a weighted lifestyle score and categorising them into three groups, it was found that participants with a favourable lifestyle (highest quintile) had HRs of 0.57 (95% CI: 0.53–0.62) for all-cause mortality and 0.84 (95% CI: 0.79–0.90) for overall cancer incidence compared to those with an unfavourable lifestyle (lowest quintile). Notably, a favourable lifestyle was associated with lower site-specific cancer risk in the digestive, respiratory, and intrathoracic organs (p-trend < 0.05) (Supplementary Material Table S7).

Among 6739 deaths, 2712 (40.2%) were from cancer, 1731 (25.7%) from CVD, 535 (7.9%) from respiratory disease, 321 (4.8%) from digestive disease, and 1440 (21.4%) from all other causes. Based on the Cox proportional hazards regression curves of cumulative risk, all-cause mortality risk decreased monotonically across healthy lifestyle scores, and a higher healthy lifestyle score was associated with a lower risk of cause-specific mortality from respiratory disease, cancer, digestive disease, CVD, and other causes (p-value for trend < 0.001) (Figure 2 and Supplementary Material Table S8). The same pattern of effects was observed when lifestyle categories (weighted lifestyle scores) were used instead of lifestyle scores (Supplementary Material Table S9). The multivariable-adjusted PAR% (95% CI) suggested that 20.62% (15.57–26.03%) of all-cause mortality, 14.96% (6.32–23.18%) deaths from cancer, 26.94% (15.57–36.97%) deaths from CVD, 37.96% (17.89–53.63%) deaths from respiratory disease, and 46.81% (22.17–64.31%) deaths from the digestive disease were theoretically attributable to nonadherence to six or seven recommended lifestyle behaviours (Figure 3). Similar results were found in weighted lifestyle scores, where 18.57% of deaths were attributable to nonadherence to a favourable lifestyle (Supplementary Material Figure S9).

Further analyses stratified by age at baseline, sex, employment status, education level, socioeconomic status, BMI categories, cholesterol medication, hypertension medication, insulin medication, family history of hypertension, DM, and cancer, with unfavourable lifestyle as the reference group, yielded a similar pattern of associations; favourable and intermediate lifestyles were inversely associated with all-cause mortality risk. Notably, crossover analysis also indicated that associations of favourable and intermediate lifestyles with all-cause mortality incidence were consistently stronger in participants with MetS who were women, employed, and had lower BMI (p for interaction < 0.001) (Figure 4 and Supplementary Material Table S10). Stronger associations for favourable and intermediate lifestyles were also identified in participants without a family history of DM (p-interaction < 0.001).

Sensitivity analyses provided strong evidence for the robustness of the results. First, the unweighted lifestyle scores, which were categorised as scoring 0–2, 3–5, and 6–7, were inversely associated with all-cause and cause-specific mortality, as well as the overall and site-specific cancer risk (Supplementary Material Tables S11 and S12). Additionally, a similar pattern of results was observed when rebuilding the lifestyle categories by further including 18.5 ≤ BMI < 25 kg/m² or redefining never smoking as a recommended behaviour (Supplementary Material Tables S13–S16).

The UK Biobank datasets are openly available by submitting a data request proposal from https://www.ukbiobank.ac.uk/↗ (accessed on 9 April 2022). We are authorised to access the database through the Access Management System (AMS) (Application number: 78563).