Shared Frailty Model in Survival Analysis of Time to Discharge Dynamics for Myocardial Infarction Adult Patients at Ayder Comprehensive Specialized Referral Hospital (Jan 1, 2018 - Dec 31, 2020)

Abstract

Background: MI, commonly known as heart attack, happens when a blood clot obstructs the coronary arteries, resulting in decreased oxygen and nutrient supply to the heart muscle. MI continues to be a significant cause of morbidity and mortality globally, with variations in the time-to-discharge dynamics among patients. Understanding the time to discharge is crucial for optimizing patient care and resource allocation, particularly in settings with limited healthcare resources like Ayder Comprehensive Specialized Referral Hospital. Objective: The overall objective of this study was to investigate and gain a comprehensive understanding of the time to discharge dynamics in MI patients at Ayder Comprehensive Specialized Referral Hospital, using a survival analysis with a shared frailty model. Methods: To fulfill the study goal, secondary data from Ayder Comprehensive Specialized Referral Hospital was collected from 206 MI patients who initiated their follow-up between January 2018 and December 2020. K-M curves used to compare the survival curve for categorical variables and the univariable analysed used Cox regression model to select variable which were included in the multivariable analysis. The Cox PH model with a parametric shared frailty distribution was utilized, with the follow-up site where treatment was administered serving as a clustering effect in the models. The study employed gamma and inverse Gaussian shared frailty distributions alongside Exponential, Weibull, and log-logistic baseline models to analyze the risk factors associated with survival time until discharge, considering socioeconomic and demographic factors. All fitted models were compared using the AIC and BIC values derived from the actual dataset. Results: Of the 206 patients were seen at Ayder Comprehensive Specialized Referral Hospital during the study period. The results revealed that approximately 54% experienced the event, while 46% did not experience it by the end of the follow-up period. The AIC values for the three baseline distributions (Exponential, Weibull, and Gompertz) of the PH model were found to be 370.8967, 90.3539, and 96.2921, respectively. The corresponding BIC values for those baseline distributions were 424.1427, 146.9278, and 152.8661, respectively. The AIC values for the three baseline distributions for the Gamma shared frailty model were found to be 375.4022, 89.2839, and 98.5131, with the BIC values for the same model found to be 431.9761, 148.9278, and ix | P a g e By: G/wahd T. 155.4568, respectively. Based on the AIC and BIC values from the simulation experiment and graphical evidence, the Gamma shared frailty model with the Weibull baseline was preferred when compared to other models. The clustering effect (follow-up site) was found to be significant for modeling the risk factors of time-to-discharge patients with MI. The estimated value of theta (θ), which measures the contribution of a frailty component to the model, was 1.1056. A chi-square value of 0.00372 with one degree of freedom resulted in a p-value of 0.0031. Based on the results of the Gamma shared frailty model with the Weibull baseline, the follow-up site at the medical ward, obesity (BMI > 30), age (in years), weight, diabetes mellitus, family history of MI, uncontrolled blood pressure, high cholesterol levels, and male gender were identified as the most significant risk factors for the outcome variable, survival time to discharge. The hazard ratio and 95% confidence interval for patient age and weight were 0.9844 (CI [1.0105, 1.5116]) and 1.0101 (CI [0.0193, 0.9879]) with p-values of 0.002 and 0.019, respectively. The covariates, including follow-up site at the Medical ward, BMI with obesity (BMI>30), diabetes mellitus, family history of MI, uncontrolled blood pressure, high cholesterol levels, and male gender, exhibited hazard ratios of 2.4868 (CI [0.4281, 1.4369]), 2.3445 (CI [0.3901, 2.1253]), 2.7563 (CI [0.582, 1.5858]), 3.7139 (CI [0.0152, 1.3031]), 1.0726 (CI [0.3823, 1.8024]), 1.7318 (CI [0.3385, 1.3835]), and 4.1012 (CI [0.0110, 1.1967]), respectively, with associated p-values of 0.018, 0.001, 0.003, 0.021, 0.026, 0.013, and 0.001, indicating their respective impacts on the study outcomes. Conclusions & Recommendation: The model suggested that there is a strong evidence of heterogeneity among follow up sites where the MI patients were treated. From the candidate models, Weibull-gamma shared frailty model was an appropriate model for the MI dataset. There was a frailty effect on the survival of the MI patients that arises due to differences in the distribution of follow up sites. The risk factors follow-up site at the medical ward, being obese, age(in years), weight, diabetes mellitus, family history of MI, uncontrolled blood pressure, high cholesterol levels, and male gender were statistically significant for the survival of MI patients whereas the other risk factors were not statistically significant. Health care providers must focus on high-risk MI patients, considering factors like age, obesity, uncontrolled blood pressure, diabetes, high cholesterol level, gender, and family history of MI.