Key result
Perioperative data improves nonhome discharge prediction to ~0.86 AUC across logistic regression and XGBoost.
Why the study?
Nonhome discharge after lower extremity bypass is associated with adverse outcomes and delays, and accurate risk assessment tools are needed to improve decision-making and discharge planning.
Can multivariable logistic regression and XGBoost machine learning models accurately predict nonhome discharge in adult patients undergoing lower extremity bypass?
Cohort (n=9,789)
Yes
Can multivariable logistic regression and XGBoost machine learning models accurately predict nonhome discharge in adult patients undergoing lower extremity bypass?
Effect estimate: AUC 0.760-0.855 (logistic regression); AUC 0.755-0.854 (XGBoost)
p-value: p=0.308
Both logistic regression and machine learning models can accurately predict nonhome discharge after lower extremity bypass, and their performance is significantly improved by including perioperative data.
May aid discharge planning after vascular procedures; leaves open prospective validation before clinical use.
BACKGROUND: Nonhome discharge (NHD) after lower extremity bypass is associated with discharge delays, increased postdischarge complications, and reduced patient quality of life. Accurate risk assessment would enhance shared decision-making, preoperative optimization, and discharge planning. METHODS: Adult patients in a statewide vascular surgery registry who underwent lower extremity bypass from January 2016 to December 2023 were analyzed. The primary outcome was NHD, defined as discharge to rehabilitation, acute care facility, nursing home/extended care, hospice, assisted living, or leaving against medical advice. Two models were developed-multivariable logistic regression and XGBoost machine learning (ML)-using predictors such as age, gender, race, comorbidities, smoking status, tissue loss, procedure urgency, and length of stay. Each model was built using (a) preoperative variables only and (b) both preoperative and perioperative variables, with data split into training and testing sets. Performance was compared using area under the receiver operating characteristic curve (AUC) and calibration plots. RESULTS: Among 9,789 patients, those experiencing NHD were more likely to be older, female, Black, not currently smoking, have tissue loss, require a vein graft, present urgently or emergently, have longer hospital stay, higher frailty scores, and multiple assessed comorbidities. Model discrimination improved with perioperative variables (logistic regression AUC 0.760-0.855; XGBoost 0.755-0.854). No difference in predictive accuracy was found between model types (P = 0.308). Models with only preoperative variables underpredicted risk in low-risk patients. CONCLUSION: Both modeling approaches accurately predicted NHD, with perioperative data improving performance. The findings informed a risk calculator to support preoperative planning and discharge efficiency.
No takes yet. Share an insight, caveat, or question.
Cunha et al. (2026) conducted a cohort in Lower extremity bypass (n=9,789). Risk prediction models (logistic regression and XGBoost) vs. Models with preoperative variables only was evaluated on Nonhome discharge (NHD) (AUC 0.760-0.855 (logistic regression); AUC 0.755-0.854 (XGBoost), p=0.308). Logistic regression and XGBoost models accurately predicted nonhome discharge (AUC up to 0.855), with perioperative data improving performance and no difference between model types (P=0.308).
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: