Development of a Risk Score to Predict Post-Discharge Rehabilitation Care After Liver Metastasectomy

Abstract

IntroductionNeed for discharge to intermediate care (DCIC) can increase length of stay and be a source of stress to patients. Estimating risk of DCIC would allow earlier involvement of case managers, improve length of stay and patient satisfaction by setting realistic expectations. The aim was to use National Surgical Quality Improvement Program dataset to develop a prediction model for DCIC after undergoing liver metastasectomy. MethodsData were obtained from National Surgical Quality Improvement Program 2011-2018 covering liver metastasectomy. Recursive partitioning narrowed potential predictors and identified thresholds for categorization of continuous variables. Logistic regression identified a predictive model, internally validated by using 200 bootstrap samples with replacement. A risk score was derived using Framingham Study methodology by dividing all regression coefficients by the smallest model coefficient. Receiver operating characteristic analysis identified the score that maximized sensitivity/specificity, defining low/high risk. Finally, recursive partitioning identified categories low/medium/high. ResultsThe most parsimonious model predicting DCIC area under the curve (, 0.722, 95%CI: 0.705-0.739) identified five independent predictors including age >60, procedure type, hypertension requiring medication, albumin <3.5 mg/dL and hematocrit <30%. Internal validation resulted in expected bias-corrected area under the curve of 0.717, 95% CI: 0.698-0.732. The maximum score was 17.9 and 5.8 maximized sensitivity (sn) and specificity (sp) [sn = 81%, sp = 51%) predicting DCIC. Stratified into three groups, a score ≥9.5 identified highest risk (12.8%), ≥4.3 medium (6.1%) and <4.3 lowest risk (1.5%). ConclusionsDetermining risk of DCIC benefits shared decision making and patient care. This evidence may enhance discharge planning after liver metastasectomy expediting the process. Age >60 contributed the most weight to the score, but the use of additional variables in three groups allowed further discrimination between patients.