The appendix is located in the lower right abdomen and can become inflamed due to the accumulation of food debris and bacteria.1,2,3Acute appendicitis is the most common surgical emergency worldwide, with a lifetime risk of 7-8%.FourThe social impact and medical burden are enormousFiveIt is also one of the common causes of acute surgical abdominal pain in the elderly. Acute appendicitis can be classified into two types of UA, defined as cellulitic appendicitis without signs of necrosis or perforation;6CA may then be accompanied by focal or total wall necrosis and ultimately lead to perforation.7It is important to distinguish between these two conditions because urinary tract infections can be treated conservatively with antibiotics without surgery.8,9 It may heal spontaneously without antibiotic therapy.10,11Patients with CA require emergency appendectomy.
Elderly people often suffer from a variety of underlying diseases, including respiratory, circulatory, endocrine, and metabolic disorders, and weakened immune function. Conservative treatment may also be attempted in elderly UA patients with mild clinical symptoms and a strong desire to avoid surgery and accept the risk of recurrence. However, the probability of appendiceal perforation and death is significantly higher in elderly patients than in those without perforation.12Furthermore, delayed diagnosis of complicated appendicitis can lead to complications such as perforation and peritonitis, which can result in significant morbidity and mortality, especially in elderly patient populations with comorbidities.13Longer hospital stays, job losses, increased costs for additional tests, psychosocial problems, etc.14.
Therefore, it is important to correctly diagnose acute appendicitis, and a “two-stage” approach is commonly adopted in clinical practice. The first stage is to diagnose “acute appendicitis.” For patients who do not have acute appendicitis, the cause of abdominal pain should be identified and promptly treated. After the diagnosis of acute appendicitis is confirmed, the second stage is to distinguish between UA and CA and adopt different treatment protocols for other pathologies.
Currently, medical history and examination findings are still considered the basis for the diagnosis of acute appendicitis. However, there is a large interobserver variability and the accuracy has room for improvement. It has been shown that physicians cannot make a correct clinical diagnosis for all patients with acute abdominal pain based only on medical history and routine examination findings.15.
In addition to laboratory tests, complete blood count parameters (leukocytes, neutrophils, lymphocytes, platelets, platelet derivatives), which are part of the usual blood biochemistry parameters, as well as markers such as total bilirubin (TBil), C-reactive protein (CRP), and procalcitonin, are widely used as the next step in the diagnosis of acute appendicitis, as they change depending on the presence and severity of inflammation.16These inflammatory markers individually have poor discriminatory power, but when combined, have a higher discriminatory power in diagnosing acute appendicitis from nonappendicitis.17However, a prospective data study of 1,024 patients with clinical suspicion of acute appendicitis found that this combination did not rule out the possibility of appendicitis.18 Enough.
Scoring systems such as the AIRS and Alvarado, which consist of medical history, physical examination findings, blood biochemical parameters, radiological instruments, and combinations thereof, can distinguish between uncomplicated and complicated appendicitis.19None of these studies mentioned measures of diagnostic accuracy, so sensitivity and specificity could not be calculated.Two other studies reported on the design of a scoring system that included clinical and biochemical features but did not report measures of diagnostic accuracy.20,21Imaging is essential to distinguish between uncomplicated and complicated appendicitis, with ultrasound and computed tomography (CT) imaging improving diagnostic sensitivity and specificity. Nevertheless, higher sensitivity for all parameters is needed, and these tools have the disadvantages of being highly operator-dependent and exposing to radiation, respectively.twenty two.
From the above, we can see that the diagnosis of appendicitis relies on clinical evaluation, laboratory tests, and imaging tests such as ultrasound and computed tomography (CT) scans. However, these methods have limitations such as diagnostic inaccuracies and time-consuming procedures, which may lead to severe complications such as appendiceal perforation and sepsis.
To overcome these challenges, artificial intelligence (AI) is widely used in clinical settings to assist doctors in making diagnoses. Artificial intelligence refers to the ability of machines to mimic human cognitive processes to perform tasks autonomously. Related literature has shown that AI techniques are advantageous in diagnosing acute appendicitis. Alramadhan et al. found that an artificial neural network (ANN) could accurately predict the risk of intra-abdominal abscess (IAA) after appendectomy with an accuracy of 89.84%, sensitivity of 70%, and specificity of 93.61% on the test set.twenty three.
Xia et al. constructed a diagnostic model using a support vector mechanism based on an improved Grasshopper optimization algorithm to distinguish between complicated and simple appendicitis. The optimal model yielded an average accuracy of 83.56%, sensitivity of 81.71%, specificity of 85.33%, and a Matthews correlation coefficient of 0.6732.twenty fourKim et al. developed a model using multivariate logistic regression and Bayesian information criterion and assessed model performance in the validation dataset through calibration plots and area under the curve (AUC), respectively. The model's calibration and discrimination performance allowed them to identify patients with appendicitis without obvious complications who could benefit from non-surgical treatment with a low risk of failure.twenty five.
Shahmoradi et al. compared the output of an optimized SVM artificial neural network with pathological findings. The sensitivity, specificity and accuracy of the network in diagnosing acute appendicitis were 91.7%, 96.2% and 95%, respectively.26Erkent et al. used a decision tree approach to determine the severity of acute appendicitis (AA) without the use of imaging modalities.27.
The article, “Effectiveness of Machine Learning to Detect Complicated Appendicitis in Resource-Limited Settings: Findings from a Vietnam Study,” builds and validates a machine learning model to facilitate the detection of complicated appendicitis.28Phan–Mai et al. used several machine learning techniques, including SVM, to classify patients with complicated appendicitis from those with uncomplicated appendicitis. Experimental results show that the GB model has high validity. Both papers are about detecting and determining complicated appendicitis from a machine learning perspective. The differences are: (1) different datasets; (2) different numbers of machine learning models used; (3) in our paper, after determining GB as the optimal model, we used SHAP technology to visualize the weights of each parameter; (4) to maximize generalization and facilitate translation to real clinical practice, we developed a Shiny app for diagnosing complicated appendicitis to diagnose UA and CA.
The clinical data for this study were taken from the article “Development and validation of a clinical prediction model for complicated appendicitis in the elderly”. In this article, we used SPSS 26.0 and R 4.0.2 software to create a CA prediction model to help clinicians quickly determine the type of acute appendicitis. We found that three parameters based on the duration of abdominal pain, peritonitis, and total bilirubin can help doctors quickly and effectively determine UA or CA. We used machine learning methods to review the study based on the clinical data provided in this article. We used nine machine learning methods to build the most appropriate clinical prediction model, and further used SHAP technique to demonstrate the importance of each parameter. Finally, we developed a Shiny application for complicated appendicitis diagnosis to help clinicians quickly and effectively recognize patients with CA and UA.
Unlike traditional medical statistical methods, machine learning techniques predict new observations by learning based on existing data. However, a major problem with many state-of-the-art machine learning models is the need for greater transparency and interpretability. To make the predictions and judgments of machine learning models interpretable, explainable artificial intelligence (XAI) techniques have been applied to clinical research. Among them, SHAP technology is one of the XAI techniques, which determines values that indicate the direction and magnitude of the contribution of variables to the estimation of ML models and visualizes the contribution of variables.30.
In this study, we predicted UA and CA by ML modeling using patients' clinical and biochemical test values, and interpreted the model results using SHAP techniques. The main findings and contributions of this paper are as follows:
(1) An ML model was created to accurately predict UA and CA patients.
(2) The GBM model showed excellent performance in patient differentiation.
(3) To explain the model, we utilized the SHAP technique to show the importance of different parameters, and developed a CA diagnosis Shiny application to help clinicians diagnose UA and CA.
