Robustness of Joint Over Separate Models for Investigating Predictors of Blood Sugar Level and Time to First...

Diabetes mellitus (DM) is a major public health problem that is responsible for morbidity and mortality. Blood sugar levels in DM patients fluctuate based on self-care management, influencing survival biomarkers such as death, complications, and recovery. A joint modeling approach was used to evaluate the relationship between these biomarkers, including their longitudinal trajectories and the corresponding survival times. The study aimed to identify factors affecting longitudinal blood sugar level measurements and time to first remission in T1DM patients at Debre Tabor General Hospital, Northwest Ethiopia.

A retrospective study was conducted on 217 randomly selected T1DM patients from January 2018 to January 2020. The linear mixed model for the longitudinal part, the Cox PH model for the survival part, and the joint model for their association were used. The Kaplan-Meier survival estimate and Log-Rank test were utilized to assess and compare the survival times.

In the current study, about 67.7% of the patients had their first remission, and the rest 32.3% were censored. The estimate of the unobserved association parameter ( $\alpha$ ) in the joint model was -1.7914 (p < 0.001), indicating a strong negative correlation between the two sub-models. Older age [AHR = 0.9746, p < 0.001], male gender [AHR = 0.1706, p < 0.001], comorbidities [AHR = 0.0783, p < 0.001], family history of diabetes mellitus [AHR = -2.661, p = 0.008], and anemia [AHR = 2.1833, p = 0.02] were associated with increased blood sugar levels and delayed remission.

The joint model outperforms the separate model in terms of variability, goodness of fit, and statistical significance. Future studies should use joint modeling for longitudinal and survival data analysis. Targeted interventions for risk factors like age, gender, comorbidities, family history, and anemia may enhance remission and glycemic control.