Cross-Sectional Mixture Modeling for Cardio-Metabolic Risk Factors: A Mixture of Categorical and Continuous Latent Class Indicators

Abstract

Introduction: Cardiometabolic disease is used to describe a range of diseases including cardiovascular disease, metabolic syndrome, stroke and type 2 diabetes. This disease is one of the most common diseases all over the world and in Iran, and currently has the heaviest economic burden for developing countries. High blood pressure, overweight and obesity, dyslipidemia and triglycerides are known risk factors for cardiometabolic disease. Cardiometabolic risk factors differ among people based on age, sexual maturity, family history and high levels of risk factors. Therefore, predicting the risk factors of cardiometabolic disease can be effective in the treatment process of patients. In this way, it is important to choose a model that can make a good prediction of risk factors. There are various statistical methods to investigate the risk factors of cardiometabolic diseases. Latent profile analysis (LPA) is used to determine whether defined risk profiles provide sufficient cardiometabolic detail. In addition, information about risk factors can provide a basis for better identifying high-risk profiles and deal with cardiometabolic diseases. Methods and materials: This cross-sectional study was carried out on healthcare workers in 2020 as a part of the Azar cohort study, which was conducted by the liver and gastrointestinal diseases research center of Tabriz University of Medical Sciences (TBZMED). A total of 500 persons full-time and long-term contract employees participated in this study. In the current study, the combination of serial and continuous indicator variables was used to identify latent classes, which was a generalization of two approaches of latent profile analysis (LPA) and latent class analysis (LCA), which is mixed modeling. It is called growth mixture model (GMM). LPA is a categorical latent variable approach that identified latent risk profiles in the present population based on a specific set of variables. Then LPA placed people in groups with different degrees of probability that had different profiles of personal or environmental characteristics. In addition, the fit indices provided in LPA provide the possibility of comparing different models and making informed decisions about the number of basic classes. Results: The total number of subjects in the initial evaluation was 500, of which 493 (98.6%) remained in the study. The final sample included 493 participants and 63.8% were male. The mean age of patients was 43 (SD 7.2). Goodness of fit and entropy indices were compared to select the best model from 1-7 classes. Also, statistical tests were calculated for a better fit of the model. The prevalence percentage of people in the classes and the change percentage of the models compared to the models with a lower class were checked for information criteria. Finally, a model with two latent profiles was selected that obtained the highest percentage of prevalence, the highest change percent and the best fit for lower BIC (1.37575) and higher entropy (0.910). The individuals were classified into two LPA-driven classes, including a low-risk profile (n = 210), and a high-risk profile (n = 290). The risk factors in the high-risk profile were significantly more than the low-risk profile. The results of logistic regression showed that the high-risk profile is significantly related to BMI (OR=3.309, 95% CI: 0.01-0.95), triglyceride (OR=1.02, 95% CI: 0.1-0.05), HDL cholesterol (OR=0.88, 95% CI: 0.79-0.99), past or current smoking (OR=0.05, 95% CI: 0.01-0.95) and hematocrit (OR = 12.07, 95% CI: 3.46-42-11). Also, the diagnostic indices and the area under the ROC curve were calculated to check the adequacy of the regression-logistic model, and the results showed that the model has the necessary adequacy. Conclusion: Our study with the LPA approach clearly demonstrated the pattern of latent shared characteristics of the health care employees, a low-risk and a high-risk profile. Specifically, we found that higher BMI levels, higher triglyceride values, higher hematocrit, and lower HDL values were the main predictors of a cardiometabolic high-risk profile. Furthermore, the LPA-derived latent risk profiles and specific predictors of the profiles, definitely help find ways to prevent the occurrence of cardio metabolic disease.