PD08.07. Multimodal AI-Driven Postoperative Surveillance for Oesophageal Cancer: Automated Patient-Reported Outcome Assessment and Early Complication Warning in a Prospective Study
Aug 2026· Diseases of the esophagus· Vol 39· 0 citations
TL;DR
An end-to-end multimodal AI system can accurately automate PRO assessment from natural patient conversations and provide clinically meaningful early warning for postoperative complications, with external validation confirming generalisability across cohorts.
Abstract
Esophageal Cancer: Surgical Treatment of Esophageal Cancer – early outcomes and complications
Postoperative complications following oesophagectomy occur in 25–60% of patients, yet traditional patient-reported outcome (PRO) assessment relies on standardised questionnaires with limited reliability, particularly among elderly patients with low health literacy. We aimed to develop and externally validate a multimodal artificial intelligence (AI) system that automates PRO assessment from natural patient conversations and provides early warning of major postoperative complications.
We conducted a prospective observational study at a tertiary cancer center, enrolling consecutive patients undergoing esophagectomy.The development cohort (Centre 1, n=196) and the temporally independent external validation cohort (Centre 2, n=103) were recruited sequentially. Using the validated Patient Symptom Assessment for Oesophageal Cancer (PSA-ESO) instrument, we collected trimodal recordings (video, audio, text) at up to 25 timepoints per patient, yielding 6,813 evaluable assessments. We fine-tuned the Qwen2.5-Omni-7B multimodal large language model with LoRA adaptation for two tasks: automated PRO symptom grading (Task A) and Temporal Transformer-based early warning of Clavien-Dindo grade II or higher complications (Task B). In a prospective implementation substudy (n=62), we evaluated the clinical impact of real-time alerts on time-to-intervention.
In the external validation cohort, the trimodal PRO assessment achieved a weighted kappa of 0.801 (95% CI 0.73–0.87) and ICC of 0.858 against expert consensus, significantly outperforming audio-text bimodal (kappa 0.754, p<0.0001) and text-only (kappa 0.689, p<0.0001) configurations. The early warning model achieved an AUROC of 0.873 (95% CI 0.82–0.93) with a mean detection lead time of 28.7 hours before clinical diagnosis. The system detected 76.1% of symptom under-reporting cases. In the implementation substudy, real-time alerts reduced median time-to-intervention from 14.2 hours to 6.8 hours (p=0.003) and were associated with shorter ICU stays (3.1 vs 5.4 days, p=0.028).
An end-to-end multimodal AI system can accurately automate PRO assessment from natural patient conversations and provide clinically meaningful early warning for postoperative complications, with external validation confirming generalisability across cohorts. This conversation-based approach represents a paradigm shift from questionnaire-based PRO evaluation in surgical oncology, with particular relevance for populations with limited health literacy.
A multimodal early warning system integrating PROs, laboratory trajectories, and large language model-extracted clinical text features to enable real-time, automated complication surveillance substantially outperforms single-modality approaches for detecting postoperative complications after esophagectomy.
Si-miao Lu, Yi Zhu, Yong-tao Han et al.· Diseases of the esophagus· 0 citations
Symptom persistence does not adversely affect overall survival; its association with improved DFS likely reflects the tumor-control benefit of adjuvant therapy.
Si-miao Lu, Yi Zhu, Yong-tao Han et al.· Diseases of the esophagus· 0 citations
Esophageal Cancer: Surgical Treatment of Esophageal Cancer – technique
Robot-assisted minimally invasive esophagectomy (RAMIE) is increasingly adopted, yet comparative patient-reported outcome (PRO) data against conventional minimally invasive esophagectomy (MIE) remains scarce. We aimed to compare perioperative PROs, including quality of life, nutritional status, cough-related quality of life, and anxiety, between RAMIE and MIE for esophageal squamous cell carcinoma (ESCC).
We enrolled 214 patients with ESCC who underwent RAMIE (n = 47) or conventional MIE (n = 167) from an ongoing prospective cohort study. PROs were assessed using the EQ-5D-5L, PG-SGA, Leicester Cough Questionnaire (LCQ), and Self-rating Anxiety Scale (SAS). A 1:1 propensity score-matched (PSM) analysis was performed to balance baseline covariates. Nutritional trajectory was tracked from admission through discharge using serial PG-SGA assessments.
Before PSM, the RAMIE group had significantly higher preoperative PG-SGA scores at admission (5.18 ± 3.32 vs 4.09 ± 3.01, P = 0.015) and pre-surgery (5.44 ± 3.69 vs 3.78 ± 2.88, P = 0.006), indicating worse baseline nutritional status. However, PG-SGA scores at discharge were comparable (4.56 ± 2.74 vs 4.42 ± 2.28, P = 0.931), demonstrating nutritional convergence. No significant differences were observed in EQ-5D-5L utility index (0.92 vs 0.93, P = 0.934), LCQ total score (121.6 vs 122.6, P = 0.751), or SAS standard score (45.4 vs 45.4, P = 0.551). The RAMIE group had a higher anastomotic leak rate (14.9% vs 4.8%, P = 0.025) but no Clavien-Dindo grade ≥III complications.
RAMIE and conventional MIE achieved comparable perioperative PROs in ESCC patients. Despite worse baseline nutritional status in the RAMIE group, nutritional recovery by discharge was equivalent. These findings support RAMIE as a viable alternative with comparable patient-centered outcomes, though the higher anastomotic leak rate warrants further investigation.
Si-miao Lu, Yi Zhu, Yong-tao Han et al.· Diseases of the esophagus· 0 citations
Standard preoperative variables are insufficient for individualized risk stratification of RLNP, anastomotic leak, or vocal cord palsy after esophagectomy, and improved prediction will require prospective integration of real-time intraoperative data.
Si-miao Lu, Yi Zhu, Yong-tao Han et al.· Diseases of the esophagus· 0 citations
The developed ML models for predicting AL after esophagectomy demonstrated acceptable discriminative ability and identified key risk factors including radiotherapy, hemoglobin drop, bleeding, and lymphocyte count and provide a data-driven foundation for risk stratification.
Chi Zhang, Yongde Liao· Diseases of the esophagus· 0 citations
INTRODUCTION & AIMS
Cancer anorexia (CA) is a prevalent and debilitating complication in postoperative patients with gastric cancer (GC). Standardized tools for the early prediction of CA risk in this specific population remain lacking. This study aimed to identify predictor domains retained in a TRIPOD-compliant multivariable model and to develop a nomogram for CA risk assessment in postoperative patients with GC.
METHODS
We performed a cross-sectional predictive modeling study of 440 postoperative patients with GC at a tertiary cancer center. Using a date-level temporal split by questionnaire assessment date (cut date: 2024-09-08), patients assessed on or before the cut (2023-04-07 to 2024-09-08; n = 309) formed the training cohort and those assessed after the cut (2024-09-09 to 2024-11-30; n = 131) formed the temporal validation cohort. CA was defined using the Anorexia/Cachexia Subscale-12 (A/CS-12) of the FAACT questionnaire, a validated screening instrument with a cut-off of total score 37. A pool of 30 candidate predictors spanning sociodemographic, lifestyle, AJCC 8th edition pathological staging (pT/pN/pM), surgical access, multi-organ resection, chemotherapy status at assessment, nutritional (GLIM), psychosocial (K-10, PSSS), and symptom (MDASI-GI) domains was screened by LASSO regression with 10-fold cross-validation, restricted to the training cohort in accordance with TRIPOD guidance. Surviving variables were entered into a multivariable logistic regression with AIC-based backward elimination. A nomogram was constructed from the final model. Discrimination, calibration (Hosmer-Lemeshow chi-square, calibration slope and intercept, Brier score), and clinical utility (decision curve analysis) were evaluated in the temporal validation cohort..
RESULTS
The prevalence of A/CS-12-defined CA was 39.55%. LASSO retained 15 variables on the training cohort; AIC backward elimination produced a final model with 11 predictor domains: sex, education level, commercial medical insurance, albumin level, scope of gastrectomy, multi-organ resection, malnutrition (GLIM), psychological distress (K-10 total), symptom severity (MDASI-GI mean), symptom interference (MDASI-GI mean), perceived social support (categorical). Domain-level likelihood-ratio tests are reported in Table 4b. The nomogram, constructed in one-to-one correspondence with the AIC-final model, achieved an area under the receiver operating characteristic curve (AUC) of 0.84 (0.793-0.881) in training and 0.75 (0.664-0.831) in temporal validation. Decision curve analysis showed net benefit over default strategies across a clinically meaningful range of threshold probabilities. Calibration in the validation cohort was imperfect (Hosmer-Lemeshow chi-square = 14.79, p = 0.063; calibration slope = 0.66; intercept = -0.42). Intercept + slope recalibration in the validation cohort improved the slope to 1.00 and Hosmer-Lemeshow p to 0.698.
CONCLUSIONS
The nomogram provides good discrimination for A/CS-12-defined cancer-anorexia risk in postoperative gastric-cancer patients and may be useful as a screening aid to direct early nutritional and psychosocial interventions toward the highest-risk patients. The model is best regarded as a potentially useful screening tool that requires recalibration and external multicenter validation before clinical implementation.
Xiao-xue Chen, Fang Xiao, Li-Li Kong et al.· Clinical Nutrition ESPEN· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
