Evaluating machine learning algorithms to Predict 30-day Unplanned REadmission (PURE) in Urology patients.

BACKGROUND: Unplanned hospital readmissions are serious medical adverse events, stressful to patients, and expensive for hospitals. This study aims to develop a probability calculator to predict unplanned readmissions (PURE) within 30-days after discharge from the department of Urology, and evaluate the respective diagnostic performance characteristics of the PURE probability calculator developed with machine learning (ML) algorithms comparing regression versus classification algorithms. METHODS: Eight ML models (i.e. logistic regression, LASSO regression, RIDGE regression, decision tree, bagged trees, boosted trees, XGBoost trees, RandomForest) were trained on 5.323 unique patients with 52 different features, and evaluated on diagnostic performance of PURE within 30 days of discharge from the department of Urology. RESULTS: Our main findings were that performances from classification to regression algorithms had good AUC scores (0.62-0.82), and classification algorithms showed a stronger overall performance as compared to models trained with regression algorithms. Tuning the best model, XGBoost, resulted in an accuracy of 0.83, sensitivity of 0.86, specificity of 0.57, AUC of 0.81, PPV of 0.95, and a NPV of 0.31. CONCLUSIONS: Classification models showed stronger performance than regression models with reliable prediction for patients with high probability of readmission, and should be considered as first choice. The tuned XGBoost model shows performance that indicates safe clinical appliance for discharge management in order to prevent an unplanned readmission at the department of Urology.

STUDY NEED AND IMPORTANCE: Unplanned readmissions form a consistent problem for many hospitals. Unplanned readmission rates can go up as high as to 35%, and may differ significantly between respective hospital departments. In addition, in the field of Urology readmission rates can be greatly influenced by type of surgery performed and unplanned readmissions in patients can go up as high as 26%. Although predicting unplanned readmissions for individual patients is often complex, due to multiple factors that need to be taken into account (e.g. functional disability, poor overall condition), there is evidence that these can be prevented when discharge management is evaluated with an objective measuring tool that facilitate such risk stratification between high and low risk patients. However, to the best of our knowledge, the latter risk stratification using ML driven probability calculators in the field of Urology have not been evaluated to date. Using ML, calculated risk scores based on analysing complex data patterns on patient level can support safe discharge and inform concerning the risk of having an unplanned readmission. WHAT WE FOUND: Eight ML models were trained on 5.323 unique patients with 52 different features, and evaluated on diagnostic performance. Classification models showed stronger performance than regression models with reliable prediction for patients with high probability of readmission, and should be considered as first choice. The tuned XGBoost model shows performance that indicates safe clinical appliance for discharge management in order to prevent an unplanned readmission at the department of Urology. Limitations of our study were the quality and presence of patient data on features, and how to implement these findings in clinical setting to transition from predicting to preventing unplanned readmissions. INTERPRETATION FOR CLINICIANS: ML models based on classification should be first choice to predict unplanned readmissions, and the XGBoost model showed the strongest results.

Implications of resampling data to address the class imbalance problem (IRCIP): an evaluation of impact on performance between classification algorithms in medical data.

Objective: When correcting for the "class imbalance" problem in medical data, the effects of resampling applied on classifier algorithms remain unclear. We examined the effect on performance over several combinations of classifiers and resampling ratios. Materials and Methods: Multiple classification algorithms were trained on 7 resampled datasets: no correction, random undersampling, 4 ratios of Synthetic Minority Oversampling Technique (SMOTE), and random oversampling with the Adaptive Synthetic algorithm (ADASYN). Performance was evaluated in Area Under the Curve (AUC), precision, recall, Brier score, and calibration metrics. A case study on prediction modeling for 30-day unplanned readmissions in previously admitted Urology patients was presented. Results: For most algorithms, using resampled data showed a significant increase in AUC and precision, ranging from 0.74 (CI: 0.69-0.79) to 0.93 (CI: 0.92-0.94), and 0.35 (CI: 0.12-0.58) to 0.86 (CI: 0.81-0.92) respectively. All classification algorithms showed significant increases in recall, and significant decreases in Brier score with distorted calibration overestimating positives. Discussion: Imbalance correction resulted in an overall improved performance, yet poorly calibrated models. There can still be clinical utility due to a strong discriminating performance, specifically when predicting only low and high risk cases is clinically more relevant. Conclusion: Resampling data resulted in increased performances in classification algorithms, yet produced an overestimation of positive predictions. Based on the findings from our case study, a thoughtful predefinition of the clinical prediction task may guide the use of resampling techniques in future studies aiming to improve clinical decision support tools.