Development and internal validation of a logistic regression model for predicting in-hospital mortality among hospitalized patients with type 2 diabetes and COVID-19: a retrospective cohort study

Abstract

Patients with type 2 diabetes mellitus (T2DM) are at increased risk of adverse outcomes following COVID-19 infection. This study aimed to develop and internally validate a prediction model for in-hospital mortality among hospitalized patients with T2DM and COVID-19. This retrospective cohort study included 925 adults with T2DM and PCR-confirmed COVID-19 hospitalized at Shahid Sadoughi Hospital, Yazd, Iran, between March 2020 and March 2021. Demographic, clinical, vaccination, and laboratory variables obtained at hospital admission and before initiation of COVID-19 treatment were analyzed using backward likelihood-ratio logistic regression. Calcium was modeled using centered linear and quadratic terms. Model discrimination and calibration were assessed, and internal validation was performed using 1,000 bootstrap resamples. Among 925 patients, 143 (15.5%) died during hospitalization. The final model included ESR, LDH, AST, hypertension, malignancy, neutrophil percentage, WBC count, RBC count, and linear and quadratic calcium terms. Higher ESR (aOR = 1.079, 95% CI 1.058–1.100), LDH (aOR = 1.009, 95% CI 1.007–1.011), AST (aOR = 1.030, 95% CI 1.014–1.046), WBC count (aOR = 1.109, 95% CI 1.044–1.178), hypertension (aOR = 1.992, 95% CI 1.238–3.205), and malignancy (aOR = 2.924, 95% CI 1.026–8.333) were independently associated with higher odds of in-hospital mortality. Calcium demonstrated a nonlinear association with mortality. The apparent AUC was 0.8707 (95% CI 0.838–0.904), with a bootstrap optimism-corrected AUC of 0.8645. Bootstrap analysis demonstrated generally stable coefficient estimates. The model demonstrated good discrimination with limited optimism after internal bootstrap validation. Because the model was developed and internally validated in a single retrospective cohort, external validation in independent populations is required before clinical implementation.