Prediction of Vestibular Schwannoma Resection Outcomes Using Three Risk Stratification Scores and Development of a Machine Learning-Based Vestibular Schwannoma Risk Score

Introduction: Patient frailty has been shown to predict higher morbidity and mortality for several neurosurgical procedures. However, the comparative accuracy between different risk stratification measures for neurosurgical outcomes is poorly characterized. Consequently, we aimed to evaluate the association between three different scores and outcomes for nationwide vestibular schwannoma (VS) resection admissions as well as develop a custom risk stratification score for VS resection.

Methods: We identified all admissions for VS resection in the Nationwide Inpatient Sample from 2002 to 2017. Three risk stratification scores were analyzed: mFI-5 (range = 0–5), mFI-11 (range = 0–11), and Charlson Comorbidity Index (CCI) (range = 0–29). Receiver operating characteristic (ROC) curves were generated to compare accuracy in predicting routine discharge and determine the optimal cutoff of “high frailty” for each score (≥2 for all three scores). Analyzed endpoints included in-hospital mortality, routine discharge, complications, length of stay (LOS), and hospitalization costs. We performed survey-weighted multivariate regression to evaluate associations between frailty and outcomes, adjusting for ten confounding variables: patient demographics (age, sex, race, insurance source), hospital characteristics (ownership, teaching status, bed size), and disease severity on admission (presence of hydrocephalus, diagnosis of neurofibromatosis 2, preoperative cranial nerve palsies). Statistical significance was maintained at p < 0.05. Subsequently, we employed k-fold cross validation and Akaike Information Criterion–based machine learning model selection to create a custom VS risk score encompassing age and comorbidities used by existing frailty scores.

Results: We analyzed 32,465 VS resection admissions. Patients were 50.5 years old on average, and the study population trended female (54.4%), white (77.4%), and privately insured (73.2%). Only CCI was predictive of greater odds of in-hospital mortality (odds ratio [OR] = 4.75, p = 0.011). High frailty patients identified via the mFI-11 (OR = 1.27, p = 0.021) and CCI (OR = 1.72, p < 0.001) exhibited significantly higher odds of perioperative complications. All three severity scores were predictive of lower routine discharge rates and elevated LOS and costs (all p < 0.05). Only respiratory complications were associated with high frailty across all three scores (all p < 0.001). High frailty as determined by the CCI was further associated with higher odds of developing cardiovascular (OR = 3.23, p < 0.001) and neurologic complications (OR = 1.59, p = 0.005). Frailty scores were not predictive of surgical complications including cranial nerve injury, cerebrospinal fluid leak, meningitis, or hemorrhage.

Our custom VS risk score, the VS-5 (https://skullbaseresearch.shinyapps.io/vs-5_calculator/), featured five variables (age ≥60, hypertension, diabetes mellitus, hydrocephalus on admission, and preoperative cranial nerve palsies) and was predictive of higher hospital mortality (OR = 6.40, p = 0.001), decreased routine hospital discharge rates (OR = 0.28, p < 0.001), perioperative complications (OR = 1.59, p < 0.001), LOS (+48%, p < 0.001), and hospital costs (+23%, p = 0.001). On ROC analysis, our custom score significantly outperformed all three existing frailty metrics in predicting routine discharge (all p < 0.001).

Conclusion: Patient frailty was predictive of higher in-hospital mortality, complications, LOS, and costs following VS surgery. Moreover, high frailty status across all scores was predominantly associated with medical complications like respiratory and cardiovascular events, rather than surgical complications like perioperative hemorrhage and cranial nerve damage. Our custom VS risk score (https://skullbaseresearch.shinyapps.io/vs-5_calculator/) outperformed the modified frailty indices and CCI in predicting VS resection outcomes and will be validated using an institutional case series.

Fig. 1 Overview of Frailty Scoring Systems. Breakdown of comorbidities used for calculation of mFI-5, mFI-11, and Charlson Comorbidity Index.

Fig. 2 Nationwide Admissions for Vestibular Schwannoma Resection. Admissions in the National Inpatient Sample for vestibular schwannoma resection from 2002-2017. Annual totals for high frailty patients, as identified by the mFI-5, mFI-11, and CCI, were also visualized.

Fig. 3 Association Between Frailty Scores and Vestibular Schwannoma Resection Outcomes. All scores were dichotomized into low frailty and high frailty, with reported results corresponding to high frailty patients. Odds ratios were reported for binary outcomes and β-coefficients, representing percent changes, were reported for continuous outcomes. A: Associations between mFI-5 and inpatient outcomes. B: Associations between mFI-11 and Inpatient outcomes. C: Associations between CCI and inpatient outcomes. D: Associations between vestibular schwannoma risk score (V-5) and inpatient outcomes. The score was calculated as 2.65 * (1 if age over 60) + 10.49 * (1 if hydrocephalus present) + 1.61 * (1 if preoperative cranial nerve palsies present) + 1.66 * (1 if diabetes present) + 1.45 * (1 if hypertension present). High frailty patients were defined as having a score ≥2.

Fig. 4 ROC Analysis of Frailty Scores. Receiver operating characteristic (ROC) curves for prediction of routine hospital discharge using mFI-5, mFI-11, CCI, and the custom vestibular schwannoma risk score (VS-5). Area under the curve (AUC) for different ROC curves were compared using the DeLong test.

Publikationsverlauf

Artikel online veröffentlicht:
15. Februar 2022