Abstract

Hospital readmission among patients with diabetes remains an important clinical and economic challenge, yet predictive-modelling research often emphasises algorithmic solutions over clinically grounded feature derivation. This study developed and evaluated a systematic, interpretable feature-engineering framework for stratifying diabetes readmission risk using non-invasive administrative and clinical data extracted from electronic health records. The analysis used 40,000 encounters from the Diabetes 130-US Hospitals dataset. Twenty engineered predictors were derived across seven domains: visit history, medication burden, procedural utilisation, laboratory utilisation, composite clinical risk, medication-change patterns, and age-related indicators; diagnosis codes were mapped into 16 clinically coherent categories. A hybrid stacking ensemble combining XGBoost, LightGBM and Extra Trees with a logistic-regression meta-learner was benchmarked against baseline classifiers. Performance was evaluated using accuracy, precision, recall, F1-score, ROC-AUC and average precision (AP), while SHAP was used to interpret feature contributions. The hybrid stacking model achieved the highest ROC-AUC (0.7042), AP (0.6678), accuracy (0.6461) and precision (0.6377), whereas XGBoost achieved higher recall (0.6323) and F1-score (0.6246). The number of laboratory procedures, medication burden relative to length of stay (meds_per_day), total medication count, hospital duration, diagnostic categories and prior inpatient utilisation emerged as important predictors. The composite clinical_risk_score was higher among readmitted than non-readmitted encounters (mean 0.54 vs. 0.28), supporting its value as an interpretable risk representation. These findings indicate that clinically informed feature engineering can yield actionable and explainable predictors of diabetes readmission while showing that added ensemble complexity provides only marginal discrimination gains. External validation and prospective evaluation remain necessary before clinical deployment.

References

• American Diabetes Association, “Economic costs of diabetes in the U.S. in 2017,” Diabetes Care, vol. 41, no. 5, pp. 917-928, 2018, doi: 10.2337/dci18-0007.
• B. Strack, J. P. DeShazo, C. Gennings, J. L. Olmo, S. Ventura, K. J. Cios, and J. N. Clore, “Impact of HbA1c measurement on hospital readmission rates: analysis of 70,000 clinical database patient records,” BioMed Research International, vol. 2014, Art. no. 781670, 2014, doi: 10.1155/2014/781670.
• I. Kavakiotis, O. Tsave, A. Salifoglou, N. Maglaveras, I. Vlahavas, and I. Chouvarda, “Machine learning and data mining methods in diabetes research,” Computational and Structural Biotechnology Journal, vol. 15, pp. 104-116, 2017, doi: 10.1016/j.csbj.2016.12.005.
• T. Chen and C. Guestrin, “XGBoost: A scalable tree boosting system,” in Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, 2016, pp. 785-794.
• G. Ke, Q. Meng, T. Finley, T. Wang, W. Chen, W. Ma, Q. Ye, and T.-Y. Liu, “LightGBM: A highly efficient gradient boosting decision tree,” Advances in Neural Information Processing Systems, vol. 30, pp. 3146-3154, 2017.
• S. M. Lundberg and S.-I. Lee, “A unified approach to interpreting model predictions,” Advances in Neural Information Processing Systems, vol. 30, 2017.
• D. J. Rubin, N. Maliakkal, H. Zhao, and E. E. Miller, “Hospital readmission risk and risk factors of people with a primary or secondary discharge diagnosis of diabetes,” Journal of Clinical Medicine, vol. 12, no. 4, p. 1274, 2023, doi: 10.3390/jcm12041274.
• C. C. McDaniel and C. Chou, “Clinical risk factors and social needs of 30-day readmission among patients with diabetes: A retrospective study of the Deep South,” Frontiers in Clinical Diabetes and Healthcare, vol. 3, 1050579, 2022, doi: 10.3389/fcdhc.2022.1050579.
• Y. Shang, K. Jiang, L. Wang, Z. Zhang, S. Zhou, Y. Liu, J. Dong, and H. Wu, “The 30-days hospital readmission risk in diabetic patients: predictive modeling with machine learning classifiers,” BMC Medical Informatics and Decision Making, vol. 21, Suppl. 2, p. 57, 2021, doi: 10.1186/s12911-021-01423-y.
• A. Zarghani, “Comparative analysis of LSTM neural networks and traditional machine learning models for predicting diabetes patient readmission,” arXiv:2406.19980, 2024, doi: 10.48550/arXiv.2406.19980.
• O. G. Emi-Johnson and K. J. Nkrumah, “Predicting 30-day hospital readmission in patients with diabetes using machine learning on electronic health record data,” Cureus, vol. 17, no. 4, e82437, 2025, doi: 10.7759/cureus.82437.
• V. Mishra, M. R. Tanniru, and J. Sreedharan, “Prediction of 30-day readmission in diabetes management using machine learning,” Computers in Biology and Medicine, vol. 195, 110616, 2025, doi: 10.1016/j.compbiomed.2025.110616.
• Y. He, Y. Yuan, Q. Tan, X. Zhang, Y. Liu, and M. Xiao, “Development and validation of a risk prediction model for 30-day readmission in elderly type 2 diabetes patients complicated with heart failure: a multicenter, retrospective study,” Frontiers in Endocrinology, vol. 16, 1534516, 2025, doi: 10.3389/fendo.2025.1534516.
• E. Mahmoudi, N. Kamdar, N. Kim, G. Gonzales, K. Singh, and A. K. Waljee, “Use of electronic medical records in development and validation of risk prediction models of hospital readmission: systematic review,” BMJ, vol. 369, m958, 2020, doi: 10.1136/bmj.m958.
• M. Sharda, S. Sharma, S. Raikar, N. Verhagen, J. Wagle, R. Mathur, S. Gowda, S. Kommu, R. Prasad, S. Bhandari, and P. Jha, “The role of machine learning in predicting hospital readmissions among general internal medicine patients: A systematic review,” Cureus, vol. 17, no. 5, e84761, 2025, doi: 10.7759/cureus.84761.