Predictive Modeling of Thyroid Status in Diabetic Patients: Insights from Glycemic Parameters via Penalized Multinomial Logistic Regression"

Abstract

Objective :This study aimed to model the relationship between glycemic control indicators (Fasting Blood Glucose [FBG] and Glycated Hemoglobin [HbA1c]) and the three-category thyroid status (euthyroid, hypothyroid, hyperthyroid) in diabetic patients (N=114). The primary focus was on identifying statistically significant predictors and their associated effects.

Results :Descriptive analysis revealed a hypothyroidism prevalence of 36% and hyperthyroidism of 7%. Independent samples t-tests indicated statistically significant differences (p < 0.05) in mean FBG, HbA1c, TSH, and T3 between diabetic and control groups. Pearson's correlation analysis in the diabetic cohort showed significant positive correlations between HbA1c and TSH (r = 0.537, p < 0.001) and between HbA1c and T3 (r = 0.361, p = 0.004).

To model the multiple dependent variable categories, Penalized Multinomial Logistic Regression was employed to address the complete separation issue arising from the rarity of hyperthyroid cases. The model exhibited excellent goodness-of-fit (Pearson Chi-Square p = 1.000) and high explanatory power (Nagelkerke R2 = 0.962). For predicting hypothyroidism versus euthyroidism, HbA1c was a highly statistically significant predictor (Wald χ2, p = 0.001, Exp(B) = 4.661×107) and TSH was also statistically significant (Wald χ2, p = 0.012, Exp(B) = 158.861). For predicting hyperthyroidism versus euthyroidism, TSH was a statistically significant negative predictor (Wald χ2, p = 0.002, Exp(B) = 0.078) while T4 was a statistically significant positive predictor (Wald χ2, p = 0.041, Exp(B) = 6.050). The overall classification accuracy of the model was 96.5%, with 100% accuracy for hyperthyroid cases.

Conclusion : The findings underscore the statistically significant role of HbA1c as a potent predictor of hypothyroidism in diabetic patients. Furthermore, TSH and T4 were identified as statistically significant predictors for hyperthyroidism, attributable to the advanced statistical methodology employed. This study emphasizes the statistical and clinical imperative of glycemic control and thyroid function monitoring in this patient population.