Varghese, Chris

ABSTRACT Background Evaluating fairness in artificial intelligence (AI) is essential before clinical implementation to prevent unintended harm across diverse patient populations. This study assesses performance differences of an image-based surgical site infection (SSI) detection model across age, gender, geography, race, and skin color. Methods Model development/validation study using retrospective, multicenter data across surgical subspecialties from adults who submitted an image of their incision via the electronic health record within 30 days of their procedure between 2019-2024. Area under receiver operating curve (AUROC) was compared by age (<55 vs. ≥55 years old), gender (male vs. female), geography (Rochester, Minnesota vs. Jacksonville, Florida vs. Scottsdale, Arizona vs. regional Minnesota/Wisconsin areas), race (White/Caucasian vs. non-White/Caucasian), and skin color (Fitpatrick I-III vs. IV-VI and Monk A-E vs. F-J). Results Training data contained 13,702 images from 4,274 patients between 2019-2022 (50.1% ≥55 years old; 63.2% female; 91.6% White/Caucasian; 58.5% Rochester, 17.3% Florida, 10.5% Arizona; 96.3% light skin tone), of which 1,348 images (9.8%) had an SSI. No differences in SSI detection existed for age, gender, race, or skin color (AUROC 0.83-0.87, p>0.09), however geographical differences were significant: lowest AUROC 0.83 (95% CI 0.77-0.88) vs. highest AUROC 0.87 (95% CI 0.84-0.89, p=0.01). Temporal validation using 4,242 images from 1,353 patients between 2022-2023 showed intact fairness across subgroups (ΔAUROC ≤0.06) except geography (ΔAUROC ≤0.16). Conclusions The model behaved fairly across age, gender, race, and skin color, however geographical differences persisted. These findings have practical implications underscoring the importance of evaluating bias in clinical AI before deployment.

ABSTRACT Introduction Postoperative mortality commonly follows complications that are not prevented, recognised or managed effectively. Ward‐based nursing care is central to surveillance, escalation and rescue, but population‐level evidence using shift‐level staffing data is limited. This national population‐based cohort study examined the association between surgical ward nurse staffing and postoperative outcomes. Methods Adults aged 18 years or older undergoing inpatient surgery under general or neuraxial anaesthesia in public hospitals in Aotearoa New Zealand between 2022 and 2024 were included. Prospectively recorded TrendCare staffing data were linked to administrative health datasets. The exposure was shift‐level staffing variance, defined as the difference between required and available nursing care hours on surgical wards in the treating hospital during admission. The primary outcome was 90‐day postoperative mortality. Secondary outcomes were postoperative complications and failure to rescue. Results The cohort included 202 428 patients undergoing 223 415 surgical admissions. Within 90 days, 6296 admissions resulted in death (2.8%). Complications occurred in 41 722 admissions (18.7%), with failure to rescue in 15.1%. Of 281 147 surgical ward shifts, 122 186 (43.5%) were understaffed. After adjustment, each one‐nurse reduction in staffing per shift was associated with higher odds of mortality (aOR: 1.13, 95% CI: 1.07–1.18), complications (aOR: 1.07, 95% CI: 1.05–1.10) and failure to rescue (aOR: 1.07, 95% CI: 1.00–1.14). Modelling estimated that correcting observed staffing deficits was associated with 182 fewer deaths annually. Conclusion Lower surgical ward nurse staffing was associated with increased postoperative mortality, complications and failure to rescue. Nurse staffing is a modifiable system factor associated with postoperative safety and should be prioritised to improve surgical outcomes.

BACKGROUND: Surgeons and ostomy nurses receive a high volume of stoma photos from patients. OBJECTIVE: This study aimed to develop and validate an automated and scalable artificial intelligence pipeline for identification and triage of patient-submitted intestinal ostomy photos. DESIGN: Retrospective derivation and validation study with stakeholder engagement to guide model design. Clinical teams reviewed photos in duplicate and categorized them into: (a) obstructed view, (b) healthy stoma, (c) suitable for conservative management, or (d) requiring in-person review. Pre-trained neural networks including MobileNetV4, ResNet50, ViT and CLIP-ViT were fine-tuned with 5-fold cross-validation. SETTINGS: Nine Mayo Clinic hospitals (2019–2022). PATIENTS: Adult patients (≥18 years) undergoing surgery who submitted images within 30 days after surgery. MAIN OUTCOME MEASURES: Model performance was evaluated with area under the receiver operator curve, precision, recall, and F1-scores. RESULTS: In this study, 538 photos of abdominal ostomies were sent in by 191 patients (median age 52 years, IQR 40–63; 60.4% female). Expert consensus triaged 236 (43.9%) photos as having an obstructed view, 30 (5.6%) as healthy, 177 (32.9%) for conservative management, and 95 (17.7%) for in-person review. All models achieved area under the receiver operator curve >0.97 for identifying stomas. The CLIP-ViT model performed best at triage (macro- area under the receiver operator curve 0.94 ± 0.06; F1 0.77 ± 0.10). End-to-end detection and triage using CLIP-ViT achieved area under the receiver operator curve of 0.99 ± 0.01, precision of 0.87 ± 0.08, recall of 0.93 ± 0.04, and F1 of 0.89 ± 0.07. Attention maps showed that models focused on stomas to determine classification. LIMITATIONS: Low sample size and lack of prospective validation. CONCLUSIONS: A vision language model pipeline accurately detected and triaged patient-submitted ostomy photos. Prospective evaluation is now needed to support integration into multidisciplinary digital workflows. See Video Abstract .

Abstract Background Failure to rescue (FTR) is a key driver of postoperative mortality and reflects a hospital’s ability to recognize, escalate, and treat complications. Although FTR varies substantially between hospitals, the hospital characteristics that underpin this variation are inconsistently described. The aim of this study was to synthesize evidence on hospital-level factors associated with FTR after postoperative complications across surgical populations and health systems. Methods A systematic review of observational studies evaluating associations between hospital-level characteristics and FTR in surgical patients was undertaken. MEDLINE, EMBASE, and Scopus were systematically searched on 17 July 2024 and updated to 31 December 2025. Hospital factors were classified into structural, staffing, and system domains. Where three or more non-overlapping studies reported adjusted estimates in comparable formats, random-effects meta-analyses were performed; otherwise, findings were summarized narratively. Methodological quality of included studies was assessed using the ROBINS-I tool (version 2). Results A total of 111 studies encompassing over 137 million patients and 301 hospital factors were included. Meta-analyses demonstrated lower FTR rates in larger hospitals (odds ratio 0.85, 95% confidence interval 0.74 to 0.97), hospitals with greater intensive care unit capacity (odds ratio 0.80, 0.71 to 0.90), and teaching hospitals (odds ratio 0.88, 0.83 to 0.93). Better nursing resources were consistently associated with lower FTR, including patient-to-nurse ratios (odds ratio 1.07, 1.03 to 1.10), higher nurse education (odds ratio 0.95, 0.91 to 0.99), and more favourable work environments (odds ratio 0.93, 0.90 to 0.95). Safety-net hospital status was associated with higher FTR (odds ratio 1.22, 1.03 to 1.45). Substantial heterogeneity was observed across most analyses. Conclusion Hospital-level structural, staffing, and system characteristics are associated with variation in FTR. Postoperative mortality is closely linked to institutional capacity to detect and manage complications, underscoring the importance of adequate resourcing and system organization to improve surgical outcomes.

Abstract Body surface gastric mapping (BSGM) enables non-invasive measurement of gastric electrophysiology, but the signals are approximately 100 times weaker than cardiac potentials and overlap spectrally with motion artifacts, necessitating labor-intensive manual review that limits clinical scalability. We present an uncertainty-aware deep learning framework combining a signal reconstruction network with a parallel uncertainty estimation network to automate artifact correction in high-resolution BSGM. Models were trained on 2,398 multihour, 64-channel recordings from 27 international clinical sites using weak supervision, a physiology-aware loss function, and uncertainty-gated quality control. In an independent cohort of 127 patients, the system achieved relative reductions of 39% in signal reconstruction error, 9% in total data removed, and 23% in amplitude–movement correlation compared with the industry-standard Wiener filter. Improved signal fidelity altered automated clinical phenotyping in 7% of patients by recovering previously obscured gastric rhythms. Uncertainty-aware deep learning enables reliable automated artifact correction in body-surface gastric mapping, improving signal fidelity and enabling scalable clinical interpretation. The system is FDA-cleared (510(k) K252504) and deployed in clinical practice, demonstrating that data-driven artifact correction can meet regulatory requirements for medical devices and reduce dependence on specialist manual review.

ABSTRACT Introduction Chronic gastroduodenal disorders remain challenging to manage, and new diagnostic approaches are needed to better delineate underlying causes and guide therapeutic decisions. Body Surface Gastric Mapping (BSGM) technologies combine high‐resolution gastric myoelectrical activity measurements with symptom and psychological profiling to provide mechanistic insights into gastric motor and sensory dysfunction. An International Working Group convened to derive the first consensus classification of BSGM phenotypes (the “Auckland Classification”). Methods A Technical Group conducted a systematic literature and clinical database review to identify objective test biomarkers and candidate disease mechanisms. Evidence was synthesized across 50 studies (primarily in gastroparesis, chronic nausea and vomiting, and functional dyspepsia), and BSGM phenotypes were mapped to existing treatment guidelines. Subsequently, iterative review and development of consensus was performed by a Consensus Group composed of international clinical experts familiar with BSGM. Eleven statements underlying the classification were then derived and circulated as a final survey to establish agreement. Results Six BSGM phenotypes were endorsed: three defined by myoelectrical abnormalities (Dysrhythmic, High Frequency, and Low Meal Response) and three by characteristic symptom associations (Sensorimotor, Continuous, and Delayed Onset Symptoms). Published studies plausibly linked these phenotypes to mechanisms including interstitial cell of Cajal depletion, vagal impairment, hypomotility, visceral hypersensitivity, gut‐brain dysregulation, and small bowel dysfunction. Phenotypes were also mapped to existing mechanism‐based treatment guidelines. Ten out of the eleven statements had > 80% agreement. Conclusions The Auckland Classification, derived by international consensus, presents a structured framework for BSGM‐defined patient phenotypes. Evidence and mechanism‐based treatment options are suggested for each phenotype to provide a foundation for research, further validation, and a pathway for integrating BSGM into clinical care.

Objective: To develop a predictive model for pathological complete response (pCR) after total neoadjuvant therapy (TNT) to inform selection for watch-and-wait (W/W). Summary Background Data: Patient selection for W/W after TNT for locally advanced rectal cancer remains challenging. Methods: An ensemble of tabular foundation models was fine-tuned in adults with clinical stage II or III microsatellite stable primary rectal adenocarcinoma undergoing TNT and total mesorectal excision (TNT+TME) from 2018-2023 to predict pCR, using pre-TNT, post-TNT and pre-TME variables. This model was externally validated on patients having TNT and W/W (TNT+W/W) to predict persistent clinical complete response (pcCR; the absence of local regrowth, distant metastases, or persistent near-cCR). Area under the receiver operator curve (AUROC), area under the precision-recall curve (AUPRC), and Brier score were calculated with 95% confidence intervals (CI). Results: Among 308 patients that underwent TNT+TME (median age 56; 40% female), the model predicted pCR with an AUROC 0.71 (95% CI 0.65-0.77), AUPRC 0.44 (95% CI 0.35-0.57), and was well calibrated with a Brier score of 0.17 (95% CI 0.15-0.20). At external validation in a cohort of 83 patients that are being managed with TNT+W/W (median age 57; 37% female), the model predicted pcCR with an AUROC 0.69 (95% CI 0.57-0.82), AUPRC 0.90 (95% CI 0.82-0.96), and Brier score of 0.30 (95% CI 0.26-0.33), improving to 0.17 with recalibration. Conclusions: This novel predictive model demonstrated good discrimination and calibration for pCR after TNT+TME with utility in TNT+W/W for pcCR after appropriate recalibration, supporting its application for W/W patient selection.

ABSTRACT Background The assessment of gastrointestinal (GI) motility disorders is limited by the invasive, resource‐intensive, and subjective nature of current testing. Artificial Intelligence (AI) is being applied to address these challenges. This review aimed to comprehensively identify, appraise, and synthesize the spectrum of AI applications in GI motility diagnostics. Methods A systematic search of PubMed, Embase, Scopus, Web of Science, and Medline for original studies was conducted in March 2025. Results were narratively synthesized and grouped by anatomic region. Key Results Of 1383 articles, 90 primary studies met the inclusion criteria. In the esophagus ( n = 31), the primary focus was automation of High‐Resolution Esophageal Manometry (HREM), pH impedance, and Functional Luminal Imaging Probe (FLIP) panometry analysis. In the gastroduodenum ( n = 21), studies looked to improve diagnostic accuracy and interpretability of legacy electrogastrography (EGG). In the small and large intestine ( n = 19), AI was used to determine transit time and predict post‐operative ileus. In the anorectum ( n = 9), tools focused on standardizing manometry interpretation. Review of pan‐GI applications ( n = 10) highlighted AI's application to wireless capsule endoscopy and bowel sounds analysis. While promising, all areas need more rigorous research and external validation before widespread deployment. Conclusions and Inferences AI will play an essential role in improving and automating the interpretation of GI motility diagnostics. However, even the most promising models require large‐scale prospective validation before clinical implementation. Trial Registration PROSPERO (ID: 1128662)

Background Predictive models employing machine learning algorithms are increasingly being used in clinical decision making, and improperly calibrated models can result in systematic harm. We sought to investigate the impact of class imbalance correction, a commonly applied preprocessing step in machine learning model development, on calibration and modelled clinical decision making in a large real-world context. Methods A histogram boosted gradient classifier was trained on a highly imbalanced national dataset of >1.8 million patients undergoing surgery, to predict the risk of 90-day mortality and complications after surgery. Class imbalance correction strategies including random oversampling, synthetic minority oversampling technique, random under-sampling, and cost-sensitive learning were compared to the natural distribution ('natural'). Models were tested and compared with classification metrics, calibration plots, decision curve analysis, and simulated clinical impact analysis. Results The natural model demonstrated high performance (AUROC 0.94, 95% CI 0.94--0.95 for mortality; 0.84, 95% CI 0.84--0.85 for complications) and calibration (log loss 0.05, 95% CI 0.04--0.05 for mortality; 0.23, 95% CI 0.23--0.24 for complications). Class imbalance mitigation (CSL, ROS, RUS, and SMOTE) did not improve AUROC or AUPRC but increased recall and F1 scores at the expense of precision and accuracy. However, these methods severely compromised model calibration, leading to significant over-prediction of risks (up to a 62.8 % increase) as further evidenced by increased log loss across all mitigation techniques. Decision curve analysis and clinical scenario testing confirmed that the natural model provided the highest net benefit. Conclusion Class imbalance correction methods result in significant miscalibration, leading to possible harm whenused for clinical decision making.