TOPLINE:

A natural language processing (NLP) algorithm applied to CT reports accurately identified and classified diverticulitis complications, outperforming diagnostic codes. These severity classifications substantially improved the prediction of severe recurrence requiring hospitalization, with risk increasing stepwise from mild to severe to chronic complications.

METHODOLOGY:

• Because diagnostic codes lack the precision to identify specific complications of diverticulitis, researchers developed and validated an NLP algorithm to identify diverticulitis subtypes, classify specific complications, and predict long-term recurrences.
• They analyzed data from 16,349 patients in a large integrated US healthcare system who had diagnosis codes for any type of diverticular disease and a CT report mentioning diverticular disease terms from January 1979 to June 2024.
• The algorithm extracted 12 CT features and grouped them by severity: uncomplicated (wall thickening, fat stranding, and inflammatory changes), mild complications (microperforation and phlegmon), severe complications (abscess, perforation, free air, and peritonitis), and chronic complications (fistula, large bowel obstruction, and stricture).
• The algorithm was internally validated by four investigators blinded to the NLP results who manually reviewed 1220 randomly selected CT reports; the researchers then examined the relationship between NLP-defined initial severity (uncomplicated, mild, severe, or chronic) and the risk for severe diverticulitis recurrence.
• The primary endpoint was severe diverticulitis recurrence, defined as any inpatient encounter with a primary or admitting diagnosis of diverticulitis occurring more than 30 days after the initial episode. The median follow-up duration was 3.3 years.

TAKEAWAY:

• The NLP algorithm demonstrated strong diagnostic performance, with sensitivity and specificity ranging from 82.8% to 99.9% and positive and negative predictive values ranging from 85.8% to 99.8% for relevant diverticulitis features; it showed near-perfect agreement with blinded manual review and outperformed diagnostic codes in classifying complicated diverticulitis.
• During the follow-up period, 3192 patients experienced severe diverticulitis recurrence, with risk increasing stepwise with the severity of NLP-derived complications. Compared with uncomplicated diverticulitis, adjusted hazard ratios (aHRs) were 1.39 for mild complications, 3.02 for severe complications, and 5.41 for chronic complications.
• Models incorporating NLP-detected features significantly improved the prediction of severe diverticulitis recurrence (area under the receiver operating characteristic curve [AUC], 0.70) compared with models using diagnostic code-based severity plus covariates (AUC, 0.63) or covariates only (AUC, 0.52; DeLong P ≤ .0001 for all comparisons).
• Among participants with complicated diverticulitis, surgery at initial diagnosis was associated with a lower risk for severe recurrence (aHR, 0.78; 95% CI, 0.62-1.00), with the greatest reduction observed in those with chronic complications (aHR, 0.37; 95% CI, 0.21-0.63).

IN PRACTICE:

“NLP-detected features have the potential to be incorporated in a clinical decision stool to improve risk stratification and identify patients who are more susceptible to readmission after their initial episode, thus helping guide management and prevention of diverticular disease,” the authors of the study wrote.

SOURCE:

The study was led by Wenjie Ma, Clinical and Translational Epidemiology Unit and Division of Gastroenterology, Massachusetts General Hospital and Harvard Medical School, both in Boston. It was published online in Clinical Gastroenterology and Hepatology.

LIMITATIONS:

The NLP algorithm was developed and validated exclusively using data from a single integrated healthcare system, limiting the generalizability of the findings to other settings. The true recurrence rate may have been underestimated because recurrence cases managed at other hospitals or in outpatient settings were not captured. Additionally, the analysis relied solely on CT reports and did not include additional details such as abscess size or location.

DISCLOSURES:

The study received funding from the National Institute of Diabetes and Digestive and Kidney Diseases, National Institutes of Health; the American Gastroenterological Association Research Scholar Awards; and the Massachusetts General Hospital Claflin Distinguished Scholar Award. Two authors disclosed serving as consultants, one of whom also reported receiving industry grants.

This article was created using several editorial tools, including AI, as part of the process. Human editors reviewed this content before publication.