Statistical Model for Synthesis Science Assessing the Effect of Telomere Length on Type 2 Diabetes Mellitus, Cardiovascular Disease, and Obesity

<Background> Leucocyte telomere length (LTL) has been recognized as a predictor for aging and age-related diseases, including T2DM, cardiovascular disease (CVD), obesity, and cancer. Numerous studies have been conducted to report the effect sizes regarding the influence of short LTL on the risk of each disease of interest with individual studies and with meta-analysis. Moreover, such a kind of associated study is faced with the interesting question: Is it adequate to investigate the influence of short LTL on single disease regardless of single study or meta-analysis? The rationale is that as LTL is genetically or epigenetically inherited the influence may be pervasive in the involvement of multiple diseases rather than single disease. It seems more attractive to study the effect of short LTL with multiple outcomes of interest. <Aims> The objectives of this thesis were: (1)to use Monte Carlo micro-simulation to generate empirical data from published articles with relevant covariates; (2)to estimate the effect size of short LTL on T2DM, CVD, and obesity with the Bayesian hierarchical random-effect model based on simulated data as indicated in (1); (3)to incorporate observed outcomes with different characteristics including both categorical and continuous variables with the Bayesian generalized linear model underpinning in the process of deriving synthesis evidence; and (4)to assess the evidence of integrated effect of LTL on the three diseases by extending the Bayesian hierarchical model for synthetic science based on the simulated data according to the published articles. <Material and methods> From the published articles on effect of LTL on the three diseases, sufficient statistics of relevant covariates for each study were abstracted, and used for micro-simulation to generate individual data. The effect of LTL on T2DM, CVD, and obesity was derived by using DerSimonian and Laird method and generalized linear model of fixed effect and random effect. The effect size was evaluated after considering individual heterogeneity and the heterogeneity at study level by the proposed Bayesian hierarchical model with random intercept and random slope parameters. The integrated effect of LTL on T2DM, CVD, and obesity was assessed by extending the proposed Bayesian hierarchical model for synthetic science to include the three category of disease and modelled by multivariate normal distribution. <Results> Considering the outcome of T2DM and CVD, a total of 8 and 7 articles were enrolled for data extraction. The association of LTL and obesity were explored by these 15 studies. This comprised a total study subjects of 18376, 7781, and 26157, respectively. By using DerSimonian and Laird method the cOR of short TL on T2DM, CVD and obesity were 1.38 (95%CI: 1.25, 1.52), 1.51 (95%CI: 1.28, 1.78), and 1.01 (95%CI: 0.96, 1.06), respectively. By using the Bayesian hierarchical model the aOR of LTL on T2DM, CVD, and obesity were 1.46 (95%CI: 1.36, 1.56), 1.59 (95%CI: 1.35, 1.84), and 1.06 (95%CI: 1.00, 1.13) respectively. The most appropriate model for T2DM, CVD, and obesity were random intercept model (DIC: 17605.1), random slope and random intercept model (DIC: 9031.7), and random intercept model (DIC: 2874.1) respectively. Using the random-effect model treating T2DM, CVD, obesity as multiple correlated outcome with Bayesian underpinning, the integrated effect of aOR of short LTL on T2DM, CVD, and obesity were 1.23 (95%CI:1.21, 1.24), 1.54 (95% CI: 1.51, 1.57), and 0.99 (95%CI: 0.99, 1.00), respectively. The number of DM and CVD cases and the number of all-cause deaths for the Taiwanese population aged 40 years and older from Keelung Community-based integrated screening (KCIS) study. The results show that there were10,438 extra all-cause deaths projected in the group with shorten LTL given relative rate of 1.021 (95% CI: 1.017-1.025). <Conclusion> Based on the principle of synthesis science, the current thesis made expedient use of Monte Carlo simulation that was used to generate individual empirical data in conjunction with Bayesian hierarchical random-effect model for modelling on the effect of LTL on three common chronic diseases, allowing for heterogeneity explained by relevant covariates and the unexplained variation due to target populations, study designs, the variation of measurement in LTL, and other variations. The conclusion based on the empirical findings is that most notable effect of LTL is seen in CVD (60% increased risk), followed by type 2 DM (46% increased risk) and the least (only 6% increased risk) for obesity among three chronic diseases. It is interesting to note that the magnitude of contribution still remained for the joint effect of three diseases but the effect sizes were reduced by 10 % for CVD (54% increased risk), 50% for Type 2 DM (23% increased risk) and almost lacking of elevated risk for obesity when correlation across three diseases has been considered.