Development and validation of a LASSO-based nomogram for predicting anastomotic leakage in elderly patients after laparoscopic gastrectomy

Abstract

Background: Anastomotic leakage (AL), a major postoperative complication following laparoscopic gastrectomy (LG), presents a critical diagnostic challenge in elderly patients, often resulting in life-threatening outcomes. This study aimed to develop and validate a risk prediction model to facilitate the early identification of AL in this population.

Methods: Retrospective data from 884 elderly patients diagnosed with gastric cancer who underwent LG were analyzed. The patients were randomly divided into training and validation cohorts in a 7:3 ratio. Clinically relevant predictors of AL were identified using least absolute shrinkage and selection operator (LASSO) regression and multivariate logistic regression analyses. A nomogram model was subsequently developed using these predictors. Model performance was evaluated and validated using the area under the curve (AUC) for discrimination, the Hosmer-Lemeshow test and calibration curve for accuracy, and decision curve analysis (DCA) for clinical applicability.

Results: The incidence rate of AL in the cohort was 13.6% (120/884). Five variables emerged as independent predictors of AL, including age, American Society of Anesthesiologists (ASA), diabetes, intraoperative blood loss, and prognostic nutritional index (PNI). The nomogram exhibited robust predictive accuracy, with AUC values of 0.870 [95% confidence interval (CI): 0.826-0.913] and 0.890 (95% CI: 0.828-0.952) in the training and validation cohorts, respectively. Calibration curves demonstrated a strong concordance between predicted and observed outcomes. DCA further indicated favorable clinical utility across a wide range of risk thresholds.

Conclusions: This study developed a LASSO-derived nomogram that incorporates five routinely assessed perioperative variables (age, ASA score, diabetes, intraoperative blood loss, and PNI) as a reliable tool for predicting AL risk in elderly patients undergoing LG. The model demonstrated satisfactory accuracy, discrimination, and clinical efficacy, thus enabling early risk identification to guide targeted preventive interventions.

Keywords: Anastomotic leakage (AL); elderly; laparoscopic gastrectomy (LG); nomogram; prediction model.

Figure 1

Predictor selection using a LASSO regression model. (A) Coefficient path plot of LASSO regression model. Each line represents the coefficient of a predictor variable, starting from the full model (left) and shrinking to zero as λ increases (right). (B) 10-fold cross-validation plot for optimal λ selection. The left dotted vertical line indicates the λ_min and the right dotted vertical line shows the λ_1−SE. The predictors in this study were selected using the λ_1−SE criterion. λ, lambda; LASSO, least absolute shrinkage and selection operator; SE, standard error.

Figure 2

Forest plot of multivariate logistic regression analysis of the risk factors for AL in elderly patients following LG. AL, anastomotic leakage; ASA, American Society of Anesthesiologists; CI, confidence interval; LG, laparoscopic gastrectomy; OR, odds ratio; PNI, prognostic nutritional index.

Figure 3

Nomogram for individualized predicting AL in elderly patients following LG. (A) Nomogram prediction model to evaluate the risk of AL based on age, ASA, diabetes, blood loss, and PNI. (B) Clinical application example: for an 88-year-old patient (ASA III, non-diabetes, blood loss <200 mL, and PNI <50), the total score [257] corresponds to an estimated AL risk of 0.814 (red arrow). Red dots represent the parameter-specific values of the patient in the example; the asterisks (***) besides each variable denote the predictive factors derived from multivariable regression analysis. AL, anastomotic leakage; ASA, American Society of Anesthesiologists; LG, laparoscopic gastrectomy; PNI, prognostic nutritional index.

Figure 4

Predictive performance and clinical utility of the nomogram. (A) ROC curve demonstrating discriminative ability in training [AUC =0.870, 95% CI: 0.826–0.913; optimal threshold =0.130 (sensitivity =0.774, specificity =0.845)] and validation cohorts [AUC =0.890, 95% CI: 0.828–0.952; optimal threshold =0.226 (sensitivity =0.899, specificity =0.750)]. (B) Calibration plots comparing predicted probabilities of AL and observed outcomes. (C) DCA quantifying net benefit across threshold probabilities, affirming the model’s clinical applicability. AL, anastomotic leakage; AUC, area under the curve; CI, confidence interval; DCA, decision curve analysis; ROC, receiver operating characteristic.