Integration of Artificial Intelligence Models and Logistic Regression to Increase the Diagnostic Accuracy of Acute...

In patients with acute abdominal pain (AAP), the accuracy of artificial intelligence (AI) models in the diagnosis of acute appendicitis (AA) has been rarely reported so far.

A cohort of 1333 AAP patients, including 697 women (52.3%) and 636 men (47.7%), was included in the study. The clinical symptoms (n=22), signs (n=14) and laboratory tests (n=3) were recorded in each pa tient. The most significant diagnostic predictors were used to construct diagnostic formulas (DFs) to be applied for AA diagnosis. Hierarchical summary receiver operating characteristics (HSROC) models were used to calculate the summary sensitivity and specificity estimates for each data set (history-taking, diagnostic findings and tests as well as DFs).

In HSROC analysis, the area under the curve (AUC) values for i) clinical history-taking, ii) diagnostic findings and tests, and iii) DFs were as follows: i) AUC=0.550 [95% confidence interval (CI)=0.500-0.590]; ii) AUC=0.731 (95% CI=0.690-0.770), and for iii) AUC=0.958 (95% CI=0.940-0.986). The differences between these AUC values (roc-comp) were all statistically significant: between i) and ii) p<0.0001; between i) and iii) p<0.0001; between ii) and iii) p<0.0001.

The diagnostic accuracy (DA) of DFs is superior to both the clinical history-taking and clinical findings (with tests), and therefore the use of these AI-based models should be an important part of the diagnostic algorithm of AA among patients presenting with AAP.