Predicting and explaining inflammation in Crohn’s disease patients using predictive analytics methods and electronic medical record data

Abstract

Crohn's disease is among the chronic inflammatory bowel diseases that impact the gastrointestinal tract. Understanding and predicting the severity of inflammation in real-time settings is critical to disease management. Extant literature has primarily focused on studies that are conducted in clinical trial settings to investigate the impact of a drug treatment on the remission status of the disease. This research proposes an analytics methodology where three different types of prediction models are developed to predict and to explain the severity of inflammation in patients diagnosed with Crohn's disease. The results show that machine-learning-based analytic methods such as gradient boosting machines can predict the inflammation severity with a very high accuracy (area under the curve = 92.82%), followed by regularized regression and logistic regression. According to the findings, a combination of baseline laboratory parameters, patient demographic characteristics, and disease location are among the strongest predictors of inflammation severity in Crohn's disease patients.

A Machine Learning-Based Approach for Predicting Patient Punctuality in Ambulatory Care Centers

“…Recent studies have leveraged ML models to enable patient-specific predictions and effective delivery of care [19,[29][30][31][32]. While most of these prior studies focused on estimating the disease risk of individuals [33][34][35], some researchers have also focused on predicting clinical uncertainties such as no-shows and demand [36,37].…”

Big data in IBD: big progress for clinical practice

“…Extraction of information from routinely collected electronic medical records (EMRs), such as physician’s clinical observations and endoscopy reports, will allow researchers to perform prognostic research on longitudinal data. A machine learning model trained on codified information (International Classification of Diseases, Ninth Revision (ICD-9)) retrieved from EMRs, including a set of baseline laboratory parameters, patient demographics and clinical characteristics, accurately (AUC= 0.93) predicted disease severity in patients with CD 95. Similarly, Waljee et al constructed a random forests machine learning model to predict IBD-related hospitalisation and outpatient steroid use, as surrogate markers of disease flares (AUC=0.87, 95% CI 0.87 to 0.88).…”

“…A machine learning model trained on codified information (International Classification of Diseases, Ninth Revision (ICD-9)) retrieved from EMRs, including a set of baseline laboratory parameters, patient demographics and clinical characteristics, accurately (AUC= 0.93) predicted disease severity in patients with CD. 95 Similarly, Waljee et al constructed a random forests machine learning model to predict IBD-related hospitalisation and outpatient steroid use, as surrogate markers of disease flares (AUC=0.87, 95% CI 0.87 to 0.88). The authors pointed out older age, high serum albumin, platelet counts, immunosuppressive medication, history of corticosteroid use and hospitalisation as risk predictors.…”

Predicting rural patients� use of eHealth through supervised machine learning algorithms: A study on Portable Health Clinic in Bangladesh (Preprint)

“…Ahmad et al [27] applied decision tree, support vector machine, and artificial neural network to predict breast cancer recurrence. Reddy et al [28] used gradient boosting machine, regularized regression, and logistic regression to predict inflammation in Crohn's disease patients. Lynch et al [29] predicted lung cancer patient survival via supervised machine learning algorithms.…”

“…Gupta et al applied logistic regression algorithm to predict online customers' purchase and found an accuracy rate of 88.75%. Reddy et al [28] found gradient boosting machine more efficient than regularized regression and logistic regression in predicting inflammation in Crohn's disease patients with AUC = 92.82%. However, in this study, we found logistic regression more effective with accuracy = 85.9%, precision = 86.4%, recall = 90.5%, F-score = 88.1%, and AUC = 91.5% than decision tree, support vector machine, and artificial neural network in predicting rural patients' use of eHealth with a sample of 292 and 12 predictors.…”