Developing a Novel Artificial Intelligence Framework to Measure the Balance of Clinical Versus Nonclinical Influences on Posthepatectomy Length of Stay.

BACKGROUND

Length of stay (LOS) is a key indicator of posthepatectomy care quality. While clinical factors influencing LOS are identified, the balance between clinical and nonclinical influences remains unquantified. We developed an artificial intelligence (AI) framework to quantify clinical influences on LOS and infer the impact of hard-to-measure nonclinical factors.

METHODS

Patients from the 2017 to 2021 ACS NSQIP Hepatectomy-Targeted database were stratified into major and minor hepatectomy groups. A three-tiered model-multivariable linear regression (MLR), random forest (RF), and extreme gradient boosting (XGBoost)-was developed to evaluate the effect of 52 clinical variables on LOS. Models were fine-tuned to maximize clinical variables' explanatory power, with residual unexplained variability attributed to nonclinical factors. Model performance was measured using R and mean absolute error (MAE).

RESULTS

A total of 21,039 patients (mean age: 60 years; 51% male) were included: 70% underwent minor resection (mean LOS: 5.0 days), and 30% underwent major resection (mean LOS: 6.9 days). Random forest had the best performance, explaining 75% of LOS variability for both groups (R: 0.75). The significant improvement in R from MLR to RF suggests significant nonlinear interactions of clinical factors' impact on LOS. Mean absolute errors were 1.15 and 1.38 days for minor and major resections, indicating that clinical factors could not explain 1.15 to 1.38 days of LOS.

CONCLUSIONS

This study is the first to measure the true influence of clinical factors on posthepatectomy LOS, showing that they explain 75% of the variability. Furthermore, it indirectly evaluated the overall impact of hard-to-measure nonclinical factors, revealing that they account for 25% of LOS.