In adult patients undergoing surgery for acute appendicitis, stratified Alvarado risk groups show a significant association with final histopathological severity, but augmenting this clinical assessment with machine learning algorithms incorporating routine preoperative laboratory data does not significantly improve the detection of complicated appendicitis. A retrospective study conducted by researchers İbrahim Özalp and Cüneyt Özdemir at a single center examined whether computational modeling could resolve long-standing limitations of the conventional scoring system. While advanced algorithms yielded a slight discrimination advantage for detecting intraoperative abscesses, the models showed no net diagnostic advantage over standard clinical grouping for identifying complicated tissue pathology.
The Alvarado score is widely utilized by clinicians evaluating suspected acute appendicitis, yet its capacity to reliably identify complicated presentations remains limited. To evaluate whether algorithmic integration of routine laboratory markers could enhance diagnostic accuracy, investigators reviewed consecutive adult cases to examine the direct relationship between Alvarado risk classifications and histopathologically verified disease severity. Concurrently, the team designed and tested machine learning models that combined risk tiers with common preoperative parameters to establish whether data-driven approaches could more accurately predict complicated appendicitis and intraoperative abscess formation.
The study cohort comprised 420 consecutive adult patients who underwent appendectomy with histopathologically confirmed acute appendicitis at a single clinical institution. For predictive modeling, the researchers extracted six routine preoperative clinical and laboratory variables, consisting of patient age, biological sex, total leukocyte count, absolute neutrophil count, C-reactive protein levels, and the designated Alvarado risk group. Patients were classified across three recognized Alvarado risk categories based on standard scoring tiers, specifically scores 5 to 6, 7 to 8, and 9 to 10.
Clinical outcomes were strictly defined based on postoperative findings. The primary investigative outcome was complicated appendicitis, defined as histopathological grade 3 to 4 disease, while the secondary outcome was the identification of an intraoperative abscess. Histopathological assessment confirmed complicated appendicitis in 104 individuals, representing 24.8 percent of the 420 patients studied. In addition, intraoperative abscesses were documented in 90 individuals, accounting for 21.4 percent of the surgical cohort.
To evaluate predictive capabilities, the researchers implemented a primary random forest model alongside nine alternative machine learning algorithms and an ensemble architecture. Model validation was conducted using a five-fold nested cross-validation scheme repeated five times, calculating each patient’s out-of-fold prediction by averaging outputs across repetitions. Discriminative capacity was measured using the area under the receiver operating characteristic curve, which was compared against the standard Alvarado risk group using DeLong tests with Holm correction for multiple comparisons, complemented by calibration assessments and decision curve analyses.
Statistical evaluation demonstrated a strong relationship between the Alvarado risk groups and histopathological severity grades, yielding a chi-square value of 79.5 with statistical significance at p < 0.001. Patients in the highest Alvarado risk group, designated by scores 9 to 10, exhibited an odds ratio of 14.3 for complicated appendicitis compared to those in the lowest risk group of scores 5 to 6, with a 95 percent confidence interval spanning 6.6 to 31.0.
Despite this marked association, machine learning algorithms failed to enhance discrimination for complicated disease over standard risk grouping alone. The primary random forest model achieved an area under the curve of 0.759 with a 95 percent confidence interval of 0.707 to 0.810, which did not differ significantly from the Alvarado risk group alone, which attained an area of 0.725 with a confidence interval of 0.673 to 0.775, resulting in a Holm-adjusted p-value of 0.089. The nine remaining machine learning algorithms exhibited similar performance profiles, and none demonstrated statistically superior discrimination over standard Alvarado risk groups following Holm correction.
For intraoperative abscess, the random forest model demonstrated higher discrimination than the Alvarado risk group alone, achieving an area under the curve of 0.808 compared to 0.739, reflecting a difference of 0.069 with a 95 percent confidence interval of 0.013 to 0.129 and a Holm-adjusted p-value of 0.043. However, this statistical difference reached significance at p < 0.05 in only two of the five cross-validation repetitions. Following within-fold isotonic recalibration, calibration slopes for the models were 0.85 and 0.86, and decision curve analyses revealed no consistent net-benefit advantage over baseline Alvarado scoring tiers.
The study authors concluded that although Alvarado risk groups correlate significantly with histopathological tissue severity, their ability to discriminate complicated disease remains moderate. Adding routine laboratory variables through machine learning did not improve prediction for complicated appendicitis and provided only modest, marginally significant gains for intraoperative abscess. Consequently, the authors emphasized that these internally validated findings remain hypothesis-generating and cannot support clinical adoption without rigorous external validation in broader cohorts.
Subject of Research: Machine learning and Alvarado risk scoring in acute appendicitis severity prediction
Article Title: The relationship between the Alvarado risk group and final histopathological findings in acute appendicitis, with internally validated machine-learning models for complicated appendicitis and intraoperative abscess
Article References: Özalp, İ., & Özdemir, C. (2026). The relationship between the Alvarado risk group and final histopathological findings in acute appendicitis, with internally validated machine-learning models for complicated appendicitis and intraoperative abscess. BMC Surgery. https://doi.org/10.1186/s12893-026-04355-4
Image Credits: AI Generated
DOI: 10.1186/s12893-026-04355-4
Keywords: acute appendicitis, Alvarado score, machine learning, random forest, complicated appendicitis, intraoperative abscess, histopathology, DeLong test, decision curves, C-reactive protein, leukocyte count, neutrophil count, appendectomy, cross-validation, risk stratification, clinical prediction models, internal validation
News Source: Teresa Odom. (November 13, 2026). Machine Learning Fails to Improve Alvarado Prediction for Complicated Appendicitis. Scienmag.



