Asian Journal of Research in Medicine and Medical Science

Background: Diphtheria remains clinically important where immunity gaps and delayed recognition permit severe disease. This exploratory study compared linear discriminant analysis (LDA) and logistic regression (LR) for classifying survival outcomes using age and recorded final diphtheria classification.

Methods: The analytic dataset contained 58 complete observations. Outcome was coded 0 for death and 1 for survival; final classification was coded 0 for negative and 1 for positive diphtheria classification, and age was recorded in months. Analyses were performed in Python. Model performance was examined using a single held-out test set, five-fold cross-validation, probability plots, and confusion matrices.

Results: Thirty-nine of 58 patients (67.2%) survived and 19 (32.8%) died; 18 of 58 records (31.0%) had a positive final classification. In the 18-observation held-out evaluation, LDA had higher accuracy (77.8% vs. 72.2%), precision (75.0% vs. 70.6%), F1 score (85.7% vs. 82.8%), and ROC-AUC (0.750 vs. 0.736), while both models had recall of 100% for survival. In contrast, five-fold cross-validation yielded slightly higher mean accuracy and lower variability for LR (0.670 ± 0.124) than for LDA (0.652 ± 0.156). The full-sample confusion matrices likewise gave descriptive accuracies of 67.2% for LR and 65.5% for LDA.

Conclusion: The ranking of LDA and LR depended on the evaluation scheme. The single held-out split favoured LDA, whereas cross-validation and the full-sample classification summary slightly favoured LR. The present sample therefore does not establish stable superiority of either method. Larger, clinically richer datasets and external validation are required before either model is considered for clinical decision support.