Pular para o conteúdo

Machine learning in prediction of individual patient readmissions for elective carotid endarterectomy, aortofemoral bypass/aortic aneurysm repair, and femoral-distal arterial bypass

machine_learning_vascular

Abstract
Objective:
Early hospital readmissions have been rising and are increasingly used for public reporting and pay-for-performance. The readmission problem is fundamentally different in surgical patients compared with medical patients. There is an opportunity to intervene preoperatively to decrease the risk of readmission postoperatively.

Methods:
A predictive model of 90-day hospital readmission for patients undergoing elective carotid endarterectomy, aortofemoral bypass/aortic aneurysm repair, and femoral-distal arterial bypass was developed using data from the Healthcare Cost and Utilization Project State Inpatient Database for Florida State. The model training followed a nested resampling method with subsampling to increase execution speed and reduce overfitting. The following predictors were used: age, gender, race, median household income, primary expected payer, patient location, admission type, Elixhauser–van Walraven Comorbidity Index, Charlson comorbidity score, main surgical procedure, length of stay, disposition of the patient at discharge, period of the year, hospital volume, and surgeon volume.

Results:
Our sample comprised data on 246,405 patients, of whom 30.3% were readmitted within 90 days. Readmitted patients were more likely to be admitted via emergency (47.2% vs 30%), included a higher percentage with a Charlson score greater than 3 (35.8% vs 18.7%), had a higher mean van Walraven score (8.32 vs 5.34), and had a higher mean length of hospital stay (6.59 vs 3.51). Endarterectomy was the most common procedure, accounting for 19.9% of all procedures. When predicting 90-day readmission, Shrinkage Discriminant Analysis was the best performing model (area under the curve = 0.68). Important variables for the best predictive model included length of stay in the hospital, comorbidity scores, endarterectomy procedure, and elective admission type. The survival analysis for the time to readmission after the surgical procedures demonstrated that the hazard ratios were higher for subjects who presented Charlson comorbidity score above three (2.29 (2.26, 2.33)), patients transferred to a short-term hospital (2.4 (2.23, 2.59)), home healthcare (1.64 (1.61, 1.68)), other type of facility (2.59 (2.54, 2.63)) or discharged against medical advice (2.06 (1.88, 2.26)), and those with greater length of stay (1.89 (1.86, 1.91)).

Conclusion:
The model stratifies readmission risk on the basis of vascular procedure type, which suggests that attempts to decrease vascular readmission should focus on emergency procedures. Given the current focus on readmissions and increasing pressure to prevent unplanned readmissions, this score stratifies patients by readmission risk, providing an additional resource to identify and prevent unnecessary readmissions.

https://journals.sagepub.com/doi/10.1177/2050312120909057

Machine learning in prediction of individual patient readmissions for elective carotid endarterectomy, aort… by Alexandre Amato on Scribd

Marcações: