Key result
EHR-based machine learning outperforms RAPT with ~90% accuracy for predicting home discharge after total joint arthroplasty.
Why the study?
The Risk Assessment Prediction Tool predicts discharge disposition after total joint arthroplasty with only 75% accuracy, raising the question of whether higher accuracy can be achieved using basic electronic health record data with machine learning algorithms.
Does machine learning with basic EHR data improve the accuracy of predicting discharge disposition after total joint arthroplasty compared to the Risk Assessment Prediction Tool?
Observational (n=1,405)
Does machine learning with basic EHR data improve the accuracy of predicting discharge disposition after total joint arthroplasty compared to the Risk Assessment Prediction Tool?
Machine learning using basic electronic health record data outperforms the standard Risk Assessment Prediction Tool in predicting discharge disposition after total joint arthroplasty.
RAPT's 75% accuracy warrants caution for discharge planning post-arthroplasty; leaves open whether ML-EHR models improve predictions.
The Risk Assessment Prediction Tool (RAPT) predicts discharge disposition after total joint arthroplasty with only 75% accuracy. The goal of this study was to evaluate whether higher accuracy can be achieved with basic electronic health record (EHR) data combined with machine learning (ML) algorithms. Three ML analysis models were developed: model 1 (M1) evaluated the accuracy of predicted discharge disposition in concordance with the RAPT; model 2 (M2) used the RAPT questionnaire to develop an ML algorithm to predict the likelihood of discharge to home vs facility; and model 3 (M3) was developed with non-RAPT data (age, surgeon, and discharge preference) with the same ML training process as M2. Evaluation metrics included accuracy for home discharge (HD), positive predictive value for HD (PPV-HD), negative predictive value for HD (NPV-HD), sensitivity, specificity, and area under the receiver operating curve (AUROC). A total of 1405 patients were included. With M1, the overall accuracy for HD was 83.5%, PPVHD was 92.1%, NPV-HD was 45%, sensitivity was 0.88, and specificity was 0.56. With M2, the overall accuracy for HD decreased to 82.8%, PPV-HD was 91.7%, NPV-HD was 43.1%, sensitivity was 0.87, specificity was 0.53, and mean AUROC was 0.87±0.03. With M3, overall accuracy for HD increased to 90.3%, PPV-HD was 95.2%, NPV-HD was 68.6%, sensitivity was 0.93, specificity was 0.76, and AUROC was 0.91±0.02. The use of basic EHR data combined with ML can exceed the accuracy of the RAPT. Applying big data on an individual level for this purpose may allow for safer and more appropriate discharge planning. [ Orthopedics . 2022;45(4):e211–e215.]
No takes yet. Share an insight, caveat, or question.
Gabor et al. (2022) conducted an observational in Total joint arthroplasty (n=1,405). Machine learning with basic EHR data vs. Risk Assessment Prediction Tool (RAPT) was evaluated on Accuracy for home discharge. Machine learning using basic EHR data achieved 90.3% accuracy for predicting home discharge after total joint arthroplasty, outperforming the RAPT tool (83.5%).
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: