Prospective Study of an Explainable AI Tool to Support Biliopancreatic Limb Selection in One-Anastomosis Gastric Bypass

N/ANot Yet RecruitingNCT07818317

The Society of Bariatric and Metabolic Surgeons of Kazakhstan120 enrolled

Overview

This prospective observational cohort study will develop and externally validate a human-supervised, explainable artificial-intelligence decision-support model for analysing factors associated with biliopancreatic limb length selection during one-anastomosis gastric bypass (OAGB). The study will record three prespecified factors: the percentage of the laparoscopic instrument segment J-I covered by visceral or omental fat (P\_JI = 100 × FI/JI), the measured total small-bowel length (TSBL), and the abdominal integral index (AII = (CD/AC) + (CG/EG) + (JI/FI)). The study will evaluate weight-loss effectiveness, nutritional and malabsorptive outcomes, bile/acid reflux, major complications, and technical feasibility. During model development, the AI system will not autonomously assign a BPL length. The operating bariatric surgeon will make the clinical decision, and the model will be evaluated for calibration, external validity, factor contribution, safety constraints, and abstention.

One-anastomosis gastric bypass is a metabolic-bariatric procedure in which a long gastric pouch is connected to the small bowel by a single gastrojejunostomy. The length of the biliopancreatic limb (BPL) changes the length of bowel exposed to biliopancreatic secretions before mixing with ingested nutrients and may therefore influence weight loss, metabolic response, nutritional risk, and bile/acid reflux. Existing clinical practice includes different BPL lengths, and the available evidence does not establish one universally optimal length for every patient. This study is designed as a prospective, multicentre, observational cohort for development and external validation of an explainable clinical decision-support model. The study will enrol adults undergoing primary laparoscopic or robot-assisted OAGB as part of routine clinical care. The BPL length will not be assigned by the AI system during model development. The actual clinically selected BPL length will be recorded as an observed exposure together with operative configuration, centre, surgeon, and relevant patient characteristics. The primary prespecified factor is P\_JI, calculated as 100 × FI/JI. JI is the full measured distance from the internal left-hypochondrial trocar point J to the ligament of Treitz at point I. FI is the portion of the J-I instrument segment covered or occupied by the displaced visceral/omental fat layer. P\_JI ranges from 0% to 100% and is a laparoscopic geometric coverage metric, not a direct volumetric measurement of total visceral fat. The second factor is TSBL, measured in centimetres from the ligament of Treitz to the ileocecal junction. TSBL will be used to estimate residual absorptive bowel for each candidate BPL length. A 4-10 m interval may be used as an operational study range, but it is not a universal biological normal range; measurements outside this range will trigger quality review rather than automatic truncation. The third factor is AII, calculated as (CD/AC) + (CG/EG) + (JI/FI). AII is a secondary geometric and technical feature. When FI equals zero, JI/FI is structurally undefined and must be handled by a prespecified missingness and sensitivity-analysis strategy rather than by substituting an arbitrary value. The statistical baseline will be multivariable regression appropriate to each endpoint, with centre and surgeon effects handled explicitly. Nested models will compare the incremental contribution of P\_JI, TSBL, and AII. Explainable AI will be used as a calibrated prediction and visualization layer, not as an autonomous decision maker. Candidate BPL actions of 0.5, 1, 2, and 3 m will be represented as research actions for analysis. Longer candidates will be subject to residual-bowel, nutritional, anatomical, uncertainty, governance, and follow-up safety constraints. The proposed hierarchy P\_JI greater than TSBL greater than AII is a falsifiable hypothesis. It may be supported, rejected, or revised after out-of-sample validation. The study will report effectiveness and harm endpoints separately and will not use an unvalidated composite outcome as the sole definition of the best BPL length. The final surgeon decision, any model override, and the reason for abstention will be documented.

Interventions

One-anastomosis gastric bypass with observed biliopancreatic limb lengthPROCEDURE

Participants undergo primary laparoscopic or robot-assisted one-anastomosis gastric bypass as part of routine clinical care. The observed biliopancreatic limb length, operative configuration, surgical platform, centre, and surgeon are recorded. During model development, the study does not assign a BPL length and the AI system does not replace the surgeon's clinical judgment. Candidate lengths of 0.5, 1, 2, and 3 m are research-action labels used to evaluate predictions and safety constraints, not automatically assigned treatment arms. Additional exposure variables: * P\_JI (%) = 100 × FI/JI * TSBL in centimetres * AII = (CD/AC) + (CG/EG) + (JI/FI) * Residual absorptive bowel after observed BPL length * Surgical platform: laparoscopic or robot-assisted * Centre and surgeon identifier * Operative configuration and reflux-prevention technique

Eligibility

Sex: ALLMin age: 18 YearsMax age: 65 Years

Medical Language ↔ Plain English

Inclusion Criteria: 1. Adults aged 18 years or older at the time of consent. 2. Patients who meet the institution's independent clinical criteria for primary one-anastomosis gastric bypass. 3. Planned laparoscopic or robot-assisted OAGB at a participating centre. 4. Ability to undergo standardized intraoperative measurement of the J-I segment, P\_JI components, and total small-bowel length when clinically and ethically appropriate. 5. Ability to provide informed consent according to local requirements. 6. Willingness and reasonable ability to complete the prespecified postoperative follow-up. Exclusion Criteria: 1. Revisional, conversion, or non-primary gastric bypass surgery. 2. Operative anatomy that prevents reliable identification or measurement of the ligament of Treitz, the J-I segment, or the ileocecal junction. 3. Unaddressed altered gastrointestinal anatomy that is outside the development protocol. 4. Severe pre-existing malnutrition or another condition for which experimental variation in BPL length would be considered clinically inappropriate by the treating team. 5. Inability to provide informed consent when consent is required. 6. Inability to complete the minimum follow-up required by the approved protocol, unless the participant is retained for an approved missing-data or safety analysis. 7. Any intraoperative or postoperative circumstance in which the treating surgeon determines that protocol measurement would create unacceptable risk.

Outcomes

Primary Outcomes

Percent total weight loss at 12 months after OAGB

Percent total weight loss will be calculated as 100 × (baseline body weight - body weight at 12 months) / baseline body weight. Baseline weight is measured before OAGB using the site-approved calibrated scale. The measure is continuous and will be analysed separately from nutritional harm, reflux, major complications, and technical outcomes.

Time frame: 12 months after OAGB

Incremental out-of-sample predictive contribution of P_JI for 12-month percent total weight loss

The incremental contribution of P\_JI will be evaluated by comparing a prespecified baseline model without the three decision factors with a nested model adding P\_JI. The analysis will report out-of-sample change in Brier score, partial R² or an endpoint-appropriate explained-variation measure, calibration, discrimination, bootstrap confidence intervals, and decision-curve net benefit. P\_JI will be considered the leading factor only if its contribution is reproducibly larger than the incremental contributions of TSBL and AII under the prespecified analysis.

Time frame: From model development through external validation; primary clinical endpoint assessed at 12 months after OAGB

Secondary Outcomes

Percent total weight loss at 24 months after OAGB

Percent total weight loss will be calculated using baseline weight and weight at 24 months. The measure will be analysed by observed BPL length and in outcome models containing P\_JI, TSBL, AII, clinical covariates, centre, surgeon, operative configuration, and candidate-length interactions when supported by the design.

Time frame: 24 months after OAGB

Incremental predictive contribution of total small-bowel length

The incremental contribution of measured TSBL will be evaluated by comparing the model containing P\_JI with a nested model adding TSBL. TSBL will be analysed continuously in centimetres or after prespecified standardization. Residual absorptive bowel will be calculated for each observed BPL length. Performance will be reported with out-of-sample Brier score, calibration, discrimination, partial R² or an endpoint-appropriate explained-variation measure, and confidence intervals.

Time frame: From model development through external validation; clinical outcomes assessed at 12 and 24 months after OAGB

Incremental predictive contribution of the abdominal integral index

The incremental contribution of AII will be evaluated by comparing the model containing P\_JI and TSBL with a nested model adding AII. AII will be calculated as (CD/AC) + (CG/EG) + (JI/FI). Cases in which FI equals zero will be handled as structural missingness for the JI/FI component and analysed using the prespecified missingness and sensitivity-analysis strategy.

Time frame: From model development through external validation; clinical outcomes assessed at 12 and 24 months after OAGB

Nutritional and malabsorptive safety outcomes

Nutritional and malabsorptive outcomes will be reported separately and will include protein-energy malnutrition, hypoalbuminemia, anaemia, iron or ferritin deficiency, folate deficiency, calcium or vitamin D abnormalities, vitamin B12 deficiency, fat-soluble vitamin deficiency, diarrhoea, steatorrhoea, and need for intensive or parenteral replacement therapy. Definitions, laboratory thresholds, and clinically significant-event criteria will be prespecified in the statistical analysis plan before database lock.

Time frame: Baseline, 3, 6, 12, and 24 months after OAGB, with additional clinically indicated assessments

Bile and acid reflux outcomes

Bile and acid reflux outcomes will include patient-reported reflux symptoms, need for escalation of medical therapy, endoscopic findings when clinically indicated, marginal ulcer, and revision for clinically significant reflux. The operative configuration and reflux-prevention technique will be recorded because reflux cannot be attributed to BPL length alone.

Time frame: Baseline and 3, 6, 12, and 24 months after OAGB, with event-based assessment throughout follow-up

Major clinical complications and reoperation

Major complications will include anastomotic leak, reoperation, readmission, venous thromboembolism, hospitalization, mortality, and other protocol-defined serious adverse events. Events will be adjudicated according to the approved study definitions and reported separately from weight-loss effectiveness.

Time frame: From OAGB through 24 months after surgery

Technical feasibility and measurement reproducibility

Technical outcomes will include operative time, conversion from laparoscopic or robot-assisted surgery, inability to complete the planned reconstruction, revision of the planned BPL length, intraoperative bowel injury, blood loss, instrument exchanges, and surgeon-rated workload. Measurement reproducibility will include repeated AC, CD, CG, EG, KI, JI, FI, and TSBL measurements, intraclass correlation coefficient, coefficient of variation, and frequency of measurement-related abstention.

Time frame: Intraoperative period and immediately after the operation

Safety-gated candidate BPL eligibility and algorithm abstention

For candidate BPL lengths of 0.5, 1, 2, and 3 m, the study will record whether the candidate passes prespecified residual-bowel, nutrition, anatomy, follow-up, model-applicability, and uncertainty checks. The study will report the frequency and reasons for algorithm abstention, surgeon override, and suppression of longer research-only candidates. This outcome evaluates workflow safety and does not represent proof of clinical benefit.

Time frame: At intraoperative decision-support evaluation and through 24-month follow-up