Predictive value of the 5-item modified frailty index in adverse outcomes in total elbow arthroplasty: A machine-learning model analysis.

INTRODUCTION

The modified frailty index (mFI-5) is a five factor risk stratification tool predicated on functional status and key medical comorbidities. mFI-5 scores have already demonstrated potential for predicting adverse outcomes after common orthopaedic procedures. The aim of this study was to capitalize upon this potential and leverage machine learning analysis to (1) further interrogate the utility of the mFI-5 as a risk stratification tool, and (2) develop an algorithm with predictive value for adverse outcomes after total elbow arthroplasty (TEA).

METHODS

A retrospective review of patients who underwent TEA from 2010 to 2020 was conducted using the American College of Surgeons National Surgery Quality Improvement Program (NSQIP) database. Postoperative complications were analyzed as summative binary variables representing the rate of complications such as mortality, readmission, reoperation, extended hospital length of stay (LOS), and discharge to a non-home destination. Univariate and multivariate analysis were performed to determine the relationship between mFI-5 and postoperative complications at the p < 0.05 level. An XGBoost binary classifier was trained to predict significant associations identified in the multivariate regression. SHAP model explainability determined the relative importance of each mFI-5 component.

RESULTS

A total of 725 patients (mean age 65.7 13.1 years, 78 % female) were included. Higher mFI-5 scores were associated with longer hospital stays ( < 0.001) and non-home discharge ( < 0.001). The machine learning model receiver operating characteristic area under the curve was 0.85 for LOS and 0.78 for non-home discharge. SHAP analysis revealed hypertension as the primary driver of mFI-5 predictive power, whereas congestive heart failure was found to be the least important component.

CONCLUSION

mFI-5 scores have high predictive value for longer hospital stays and non-home discharge after TEA. There is growing evidence for using frailty indices to stratify risk and the model developed in this study may serve as a prototype for future clinical decision-making tools.