Novel Machine Learning Models Predict High-value Healthcare Outcomes Following Surgical Resection of Multifocal GBM

Introduction: Multifocal Glioblastoma (GBM) present with multiple, connected lesions and poor prognoses. Despite an increased risk of adverse postoperative outcomes, risk factors for high-value health care outcomes have not been well characterized in multifocal GBM cohorts. The purpose of the present study was to develop predictive models for extended length of stay (LOS) and nonroutine discharge disposition for multifocal GBM patients.

Methods: All multifocal GBM patients undergoing surgical resection at our institution between January 1, 2009 and December 31, 2020 were included. Predictive models for nonroutine discharge disposition and extended LOS were developed using multivariate logistic regression models. Optimism-corrected C-statistics for each model were calculated using 1000 bootstrapped samples and model calibration was assessed using the Hosmer-Lemeshow test. p < 0.05 was considered statistically significant.

Results: 134 multifocal GBM patients who underwent surgical resection were included. 43 patients (32.1%) experienced a nonroutine discharge and 34 patients (25.6%) had an extended LOS. In our multivariate models, ER admission (OR 3.50, p=0.0237), increasing mFI-5 scores (OR 1.77, p=0.0368) and decreasing KPS scores (OR 0.97, p=0.0417) independently and significantly predicted extended LOS while increasing age (OR 1.03, p=0.0454) and decreasing KPS score (OR 0.94, p< 0.001) significantly and independently predicted odds of nonroutine discharge. Our models for nonroutine discharge and extended LOS had optimism-correct c-statistics of 0.7372 and 0.7365 respectively and displayed adequate calibration.

Conclusion : Following external validation, our open-access, online calculators may assist with surgical decision making for multifocal GBM patients.