Truace
Subscribe

Predicting gallstones risk with body composition analysis and machine learning: a dual-center cohort study

Background To compare the predictive performance of computed tomography (CT) body composition indices, anthropometric indices, and laboratory indices for gallstones occurrence, and to construct machine-learning models to improve performance. Methods The dual-center retrospective cohort enrolled patients who underwent initial abdominal CT between January 2017 and January 2023, had no gallstones detected, and completed at least 3 years of follow-up. The data analysis was performed in April 2026. They were divided…

By the Truvace desk2 min read1 sourcePeer-reviewedTRV-2026-1775· cite & verify
Predicting gallstones risk with body composition analysis and machine learning: a dual-center cohort study
Case of transposition of the abdominal viscera by Campbell William, 1788-1848 Royal College of Physicians of Edinburgh. Public domain

In brief

Researchers at two centers followed adults who had an initial abdominal CT showing no gallstones and at least three years of follow-up. A deep-learning tool measured fat, muscle and bone at the third lumbar vertebra, and these measures were compared with body measurements and blood tests for predicting who later developed gallstones.

Models that combined all three kinds of indicators did better than any single marker, and the share of subcutaneous fat relative to total fat mattered most. The work is retrospective and the best result is moderate, so it does not show that the approach would change screening or care in practice.

Main points

  1. The cohort had 1,944 patients: 1,437 in the training cohort at Center 1 and 507 in the independent test cohort at Center 2.
  2. Neutrophil-to-lymphocyte ratio was the best single predictor, with an AUC of 0.627, but had limited standalone utility.
  3. The extreme trees model performed best with a test-set AUC of 0.772 (95% CI 0.714-0.822).
  4. SHAP analysis ranked the area ratio of subcutaneous to total fat as the most important feature.

The gain

Combining CT body composition, anthropometric and laboratory indices in an extreme trees model reached a test-set AUC of 0.772, outperforming the best single predictor at 0.627.

The rundown

Eligible patients had their first abdominal CT between January 2017 and January 2023, showed no gallstones, and completed at least three years of follow-up. They were grouped by follow-up findings into gallstone and non-gallstone groups.

The Body and Organ Analysis deep-learning tool quantified fat, muscle and bone at the third lumbar vertebra. Predictive models were built in the training cohort and checked with fivefold cross-validation and the independent test cohort.

Median age was 63 years in both cohorts. The training cohort included 699 males and the test cohort 266 males.

Sources

  1. Peer-reviewedAnnals of Medicine2026-10-07

The debate