IDEAS-AAP System Diagnoses Acute Abdominal Pain

Overview

This is a study to validate the effect of the intelligent diagnostic evidence-based analytic system in acute abdominal pain augmentation. Included physicians were randomly assigned into control or AI-assisted group. In this experiment, the whole electronic health record of each acute abdominal pain patient was divided into two parts, signs and symptoms recording (including chief complaint, present history, physical examination, past medical history, trauma surgery history, personal history, family history, obstetrical history, menstrual history, blood transfusion history, drug allergy history) and auxiliary examination recording (including laboratory examination and radiology report). For each case, the control group readers will first read the signs and symptoms recording of electronic health record and make a clinical diagnosis. Then the readers have to decide to either order a list of auxiliary examinations or confirm the clinical diagnosis without further examination. If the readers choose to order examinations, the corresponding examination results will be feedback to the readers, and the readers can then decide to either continue to order a list of auxiliary examinations or make a confirming diagnosis. Such cycle will last until the reader make a confirming diagnosis. For the AI-assisted readers, the physicians were additionally provided with the feature extracted by IDEAS-AAP, a list of suspicious diagnoses predicted by IDEAS-AAP, and corresponding diagnostic criteria according to guidelines. After the readers get the examination results, the IDEAS-AAP will renew its diagnosis prediction

In recent years, with the continuous development of science and technology, the range of diagnostic tests and biomarkers for disease and treatment modalities has increased exponentially, and medical information has become increasingly complex. This requires the clinician to comprehensively evaluate the patient's condition, so as to choose the best examination and treatment. However, for the complex symptoms in the actual clinical environment, the corresponding diseases are numerous; In the face of complex and heavy clinical work, how to extract the important characteristics of patients' diseases faster and more accurately to achieve high-quality and accurate diagnosis and treatment is the key problem to be solved at present. For example, in the field of digestion, the chief complaint of abdominal pain is one of the most common clinical symptoms of patients seeking medical treatment, and some acute abdominal pain, such as gastrointestinal ulcer perforation, strangulated intestinal obstruction, acute obstructive suppurative cholangitis and other urgent onset, narrow treatment time window, high mortality. Clinicians must make a quick diagnosis and distinguish between those that require emergency intervention and those that do not in order to manage patients in a timely manner and avoid catastrophic events. However, the causes of abdominal pain are many and the mechanisms are complex. In addition, since pain is a subjective sensation and is greatly influenced by subjective factors, there are no clear objective indicators to determine whether or not and the degree of pain, and it is extremely challenging to correctly diagnose and interpret abdominal pain. To this end, the clinician must take a detailed history and perform a thorough physical examination when evaluating a patient's abdominal pain. In recent years, artificial intelligence technology has developed rapidly, especially in the field of medicine has been widely applied research, mainly reflected in the diagnosis and differential diagnosis of diseases, prognosis judgment and clinical decision analysis. Some studies have shown that in terms of auxiliary pathology and imaging diagnosis, AI has reached or even exceeded the average diagnostic level of corresponding specialists. Most of these studies focus on pattern recognition based on images, and the logical judgment based on natural language using medical records information is still in the preliminary development stage. There are no relevant reports on integrating comprehensive information of large medical records to make intelligent prediction of digestive tract diseases.