Predicting Protracted Recovery and Long-Term Loss of Independence after Major Surgery in Veterans

Surgery often has lasting impacts on quality of life, which requires outcome measurement beyond 30 days. Even when long-term outcomes are measured, point-of-care risk calculators have not been developed for long-term outcomes.

To develop a risk calculator for protracted recovery and loss of independence after surgery.

This retrospective cohort study used all Veterans Affairs (VA) Surgical Quality Improvement Program data (2016-2019) linked to a variety of VA data sources. We captured a broad range of candidate predictors and used LASSO regression (a statistical method for variable selection) to estimate a predictive model.

Surgical care in VA hospitals.

Trajectories of postoperative recovery, divided into two groups: (1) routine or slow recovery (0-60 day recovery) and (2) protracted recovery or loss of independence (6 months or more).

The strongest predictor of protracted recovery or loss of independence was the Care Assessment Needs score for 90-day hospitalization, followed by functional status. The model had high discriminative power across training and test samples (c-statistics .906-.908) and was highly sensitive (83.0-83.6%) and specific (82.4-82.5%). Gradient boosting (tree-based machine learning) models did not achieve substantially better performance (c-statistic 0.920).

We developed a risk calculator for long-term trajectories of postoperative recovery. Identifying patients at high risk for loss of independence or protracted recovery may allow for targeted interventions to improve outcomes for these patients. Developing risk calculators for long-term surgical outcomes is feasible and requires future work on risk communication for shared decision-making.