That striking gap motivated the researchers to build a diagnostic prediction model from things almost any hospital already has on hand: endoscopic appearance and routine blood counts. The study, published in Frontiers in Medicine, was retrospective. The team included 655 patients from four centers and analyzed complete cases for modeling, which came to 455 patients, of whom 103 had ulcerative colitis.

They prespecified four logistic regression models and compared their performance using standard measures: discrimination (how well the model separates cases from non-cases, expressed as the area under the receiver operating characteristic curve, or AUC), the Brier score, calibration, and decision-curve analysis. Internal validation relied on 1,000 bootstrap resamples, and the authors tested transportability across sites using leave-one-center-out cross-validation. As a sensitivity check, they also ran multiple imputation to confirm the robustness of the four primary models.

The winning model combined two endoscopic findings, mucosal granularity and mucosal edema, with two blood count-derived inflammatory indices: the neutrophil-to-lymphocyte ratio and the systemic immune-inflammation index. This combined model achieved an apparent AUC of 0.739 (95 percent confidence interval 0.684 to 0.795), and after optimism correction for overfitting, an AUC of 0.733. In clinical prediction terms, that is moderate discrimination, meaning the model is meaningfully better than a coin flip but far from definitive on its own.

The comparisons were revealing. The combined model clearly outperformed a model built from blood-derived indices alone, which managed an AUC of just 0.594, a statistically significant difference (P < 0.001). It also beat a simplified endoscopic model (AUC = 0.711, P = 0.024).

However, it did not differ significantly from a three-item endoscopic model composed of vascular pattern abnormality, bleeding, and erosions or ulcers, which reached an AUC of 0.758 (P = 0.521). In other words, the blood indices added only modest incremental value on top of what the endoscopist could already see. An exploratory analysis using LASSO, a statistical technique that shrinks and selects variables, picked out vascular pattern abnormality, bleeding, and mucosal granularity as the most informative features, yielding an optimism-corrected AUC of 0.746.

That result reinforces the study’s central message: the visual texture of the mucosa and its bleeding pattern carry most of the diagnostic signal, while routine blood inflammation markers contribute a smaller, though measurable, increment. One clinically important number emerged from the threshold analysis. At a cutoff of 0.155 on the model’s predicted probability scale, the negative predictive value was 0.901.

In practical terms, when the model scores a patient below that threshold, there is roughly a nine-in-ten chance the patient does not have ulcerative colitis. A high negative predictive value is useful in settings where the goal is to rule out ulcerative colitis, though the study did not evaluate how such scores would be used in downstream diagnostic decisions. The context matters.

But the new study’s documentation rate of 0.9 percent shows that in these four Chinese centers, the test was almost never recorded at the moment of assessment. Whether that reflects cost, turnaround time, or local practice patterns is beyond what the data show, but the practical implication is clear: clinicians frequently must decide without it. The study has limits worth stating plainly.

It was retrospective, which means the models were fit to existing records rather than tested prospectively on new patients. The complete-case analysis dropped 200 of the original 655 patients, and although multiple imputation supported the robustness of the primary models, missing data remain a concern in any retrospective design. The optimism-corrected AUCs come from bootstrap resampling within the same dataset, and while leave-one-center-out cross-validation addressed transportability across the four sites, external validation in entirely new populations and health systems has not yet been reported.

The authors also characterize the LASSO-based analyses as exploratory. What the study does establish is a modest but useful benchmark. A handful of endoscopic features, granularity, edema, vascular pattern, bleeding, and erosions or ulcers, combined with two ratios calculated from a routine blood count, can predict ulcerative colitis with an AUC in the low-to-mid 0.7 range and a negative predictive value above 0.9 at a defined cutoff. The authors frame the combined model as showing moderate discrimination with modest incremental value from the blood-derived indices, an honest summary of a tool that may help triage patients when the standard stool marker is simply not available.

Subject of Research: Diagnostic prediction of ulcerative colitis using simplified endoscopic features and blood count-derived inflammatory indices in a retrospective multicenter study

Article Title: Diagnostic prediction of ulcerative colitis using simplified endoscopic features and blood count-derived indices: a multicenter study

Article References: Sun, Y., Cheng, D., Huang, H., Lai, D., Xu, R., Wang, D., Li, H., Shen, L., & Zhang, T. (2026). Diagnostic prediction of ulcerative colitis using simplified endoscopic features and blood count-derived indices: a multicenter study. Frontiers in Medicine, 13, Article 1991253. https://doi.org/10.3389/fmed.2026.1991253

Image Credits: AI Generated

DOI: 10.3389/fmed.2026.1991253

Keywords: ulcerative colitis, diagnostic prediction model, fecal calprotectin, endoscopy, neutrophil-to-lymphocyte ratio, systemic immune-inflammation index, mucosal granularity, retrospective study, multicenter study, gastroenterology, AUC, logistic regression