Purpose: Differentiating complicated appendicitis has become important, as multiple trials showed that non-operative management of uncomplicated appendicitis is feasible. We developed and validated a diagnostic model to differentiate complicated from uncomplicated appendicitis.
Methods: This retrospective study included 1153 patients (mean age ± standard deviation, 30 ± 8 years) with appendicitis on CT (804 patients for development, and 349 for validation). Complicated appendicitis was confirmed in 300 and 121 patients in the development and validation datasets, respectively. The reference standard was surgical or pathological report except in 7 patients who underwent percutaneous abscess drainage. We developed a model using multivariable logistic regression and Bayesian information criterion. We assessed calibration and discriminatory performance of the model in the validation dataset via calibration plot and the area under the curve (AUC), respectively. We measured sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and proportion of false- and true-negatives of the model in the validation dataset, targeting 95% sensitivity.
Results: Five CT features (contrast-enhancement defect of the appendiceal wall, abscess, moderate or severe periappendiceal fat stranding, appendiceal diameter, and extraluminal air) and percentage of segmented neutrophil were included in our model. The calibration slope was 1.03, and AUC was 0.81 (95% CI 0.77-0.85) in the validation dataset. The sensitivity, specificity, PPV, NPV, and proportion of false- and true-negatives were 93.4% (91.8-99.1), 28.1% (13.6-24.1), 40.8% (35.0-46.8), 88.9% (79.3-95.1), 2.3%, and 18.3%, respectively.
Conclusion: Our model may identify patients with unequivocally uncomplicated appendicitis, who may benefit from non-operative management with low risk of failure.
Keywords: Appendicitis; Conservative treatment; Sensitivity and specificity; Tomography, X-ray computed.