Association between healthy lifestyle on life course and multimorbidity in adults: results from two national prospective cohort studies | BMC Public Health
Descriptive statistics
The baseline characteristics of participants and three lifestyle scores were shown in Table 1 and eTable 1 in Supplement. The average follow-up time was 14 and 12 years in the ELSA and HRS, respectively. In the SCORE2 scoring model, 2245 and 1251 participants were included from the ELSA and HRS, of whom 703 and 668 developed multimorbidity. The median SCORE2 scores were 4.55 (3.19, 6.39) and 7.18 (4.30, 10.64), respectively. The median age was 59 (56, 64) and 60 (55, 65), and the median BMI was 27.27 (24.66, 30.20) and 27.26 (24.39, 30.88). In the LE ‘8 scoring model, 2731 and 1156 participants were included from the ELSA and HRS, respectively, of whom 990 and 691 developed multimorbidity. The median LE ‘8 score was 53.33 (43.33, 61.67) and 65.83 (55.00, 75.83), and the median age was 62.00 (57.00, 69.00) and 63.00 (56.00, 70.00), respectively. In the HLS scoring model, we included 3299 and 1645 participants from the ELSA and HRS, of whom 1307 and 998 participants developed multimorbidity. The median HLS score was 3 (2, 4) and 2 (1, 3), respectively, and the median age was 62 (57, 69) and 63 (56, 70).
The three lifestyle scores and multimorbidity
The associations between three lifestyle scores and multimorbidity were shown in Table 2. After adjusting for covariates, the results were robust in two cohorts. In the SCORE2, the hazard ratio (HR) of multimorbidity in the ELSA and HRS was 1.208 [95% confidence interval (CI) (1.176,1.241)] and 1.085 [95% CI (1.070, 1.101)], respectively, for each one-point increase. On the other hand, high scores were correlated with a higher risk of multimorbidity, compared to the low scores [HR = 4.125, 95% CI (3.042, 5.594) in the ELSA and HR = 2.902, 95% CI (2.353, 3.579) in the HRS]. The E-values of the above models were 1.401–7.715 and 1.160–5.251. In the LE ‘8, every one-point increase in scores were correlated with a lower risk of multimorbidity [HR = 0.918, 95% CI (0.882, 0.956) in the ELSA and HR = 0.981, 95% CI (0.976, 0.986) in the HRS]. Using low scores as reference, high scores were linked with lower risk of multimorbidity [HR = 0.517, 95% CI (0.303, 0.880) in the ELSA and HR = 0.432, 95% CI (0.327, 0.571) in the HRS]. The E-values of the above models were 1.401–7.715 and 1.160–5.251. In HLS, every increase of one point was related to the reduction of multimorbidity risk. [HR = 0.971, 95% CI (0.966, 0.976) in the ELSA and HR = 0.897, 95% CI (0.847, 0.950) in the HRS]. The risk of multimorbidity in people with high scores was related to the decrease, taking the low score as a reference. [HR = 0.369, 95% CI (0.299, 0.456) in the ELSA and HR = 0.792, 95% CI (0.633, 0.991) in the HRS]. The E-values of the above models were 1.401–7.715 and 1.160–5.251. The FDR of the above models were all < 0.05. We also analyzed the crude models and subgroups of the scoring models, and the results were mostly consistent (eTable 2–3 in Supplement).
Marginal predictions of three lifestyle scores and multimorbidity
Marginal predictions of three lifestyle scores and multimorbidity were shown in Fig. 1. When other variables were controlled at the mean, in the SCORE2, the predicted risk of multiple disease increased for every point increase, with a coefficient of 0.052 and 0.033 in the ELSA and HRS (eTable 4 in Supplement). In the LE ‘8 and HLS, each one-point increase was negatively associated with the marginal prediction of multimorbidity.
The trajectories of different lifestyles changes and multimorbidity
Besides, we have determined different trajectories for three lifestyle changes (eFigure 2 in Supplement), and the trajectory associations between three lifestyle changes and multimorbidity were shown in Table 3. In the SCORE2, using the low pattern as reference, the risk of multimorbidity was increased in both the moderate and crest patterns in the ELSA, which was also found in the HRS. In the LE ‘8, the risk of multimorbidity in the slow high/ high pattern was correlated with the reduced risk of multimorbidity, compared to the slow low /low pattern. The HR was 0.410 in the ELSA cohort, with 95% CI (0.307, 0.548). The HR was 0.570 in the HRS, with 95% CI (0.438, 0.743). In the HLS, the HR of multimorbidity in the slow high/high model, compared to the low pattern, was 0.868 in the ELSA, with 95% CI (0.639, 1.180), and 0.734 in the HRS, with 95%CI (0.593, 0.908).
The RMST of three lifestyle scores and multimorbidity
On the other hand, some Cox regression models did not comply with the proportional hazards assumption in ELSA. We analyzed the RMST and survival matrix, and the results were shown in Fig. 2. At 4, 8, and 12 years of follow-up, the mean survival time was not proportional (Fig. 2A, C, and E). In the SCORE2, the probability of survival gradually decreased with longer follow-up, and the reduction in survival probability was much greater in the exposed group with a score of > 10 than in the group with a score of 5 (Fig. 2B). In LE ‘8 and HLS, the probability of survival gradually decreased with longer follow-up, and the probability of survival decreased in exposed groups with higher scores slowly (Fig. 2D and F). It shows that adhering to a long-term healthy lifestyle can improve the survival probability of the population.
Mediation analysis
The mediating analysis highlighted that triglyceride (TG), C-reaction protein (CRP), fibrinogen, and Cystatin C may partly explain the correlation between healthy lifestyle and multimorbidity (mediating percentages ranged from 0.06% to 12,76%, Fig. 3) in the ELSA and HRS.
Sensitivity analysis
First, we perform intercept correction on the SCORE2 model. By adjusting the intercept of the model, the calibrated predicted mean was close to the actual observed mean, and the calibration degree of the model is improved. The results are consistent with the main results (eTable 5 and eFigure 3 in Supplement). Second, the results were robust by removing baseline disease and those who developed multimorbidity within two years (eTable 6 in Supplement). Third, the population is re-analyzed by inverse probability weighting, and the result remains unchanged (eTable 7 in Supplement). Fourthly, the competitive risk model with mortality is used for analysis, and the results are robust (eTable 8 in Supplement).
link