1. Prediction of 30-Day Hospital Readmissions for All-Cause Dental Conditions using Machine Learning
- Author
-
Hung M, Li W, Hon ES, Su S, Su W, He Y, Sheng X, Holubkov R, and Lipsky MS
- Subjects
machine learning ,dentistry ,quality improvement ,risk prediction ,healthcare policy ,precision medicine ,Public aspects of medicine ,RA1-1270 - Abstract
Man Hung,1– 5 Wei Li,2 Eric S Hon,6 Sharon Su,1 Weicong Su,7 Yao He,8 Xiaoming Sheng,9 Richard Holubkov,10 Martin S Lipsky1 1Roseman University of Health Sciences, College of Dental Medicine, South Jordan, UT, USA; 2University of Utah, Department of Family and Preventive Medicine, Salt Lake City, UT, USA; 3University of Utah, Department of Orthopaedics, Salt Lake City, UT, USA; 4University of Utah, School of Business, Salt Lake City, UT, USA; 5George E. Wahlen Veterans Affairs Medical Center, Salt Lake City, UT, USA; 6University of Chicago, Department of Economics, Chicago, IL, USA; 7University of Utah, Department of Mathematics, Salt Lake City, UT, USA; 8University of Utah Alzheimer’s Center, Salt Lake City, UT, USA; 9University of Utah, College of Nursing, Salt Lake City, UT, USA; 10University of Utah, Department of Pediatrics, Salt Lake City, UT, USACorrespondence: Man HungRoseman University of Health Sciences College of Dental Medicine, 10894 S. River Front Parkway, South Jordan, UT 84095, USATel +1801-878-1270Email mhung@roseman.eduIntroduction: It is unknown whether patients admitted for all-cause dental conditions (ACDC) are at high risk for hospital readmission, or what are the risk factors for dental hospital readmission.Objective: We examined the prevalence of, and risk factors associated with, 30-day hospital readmission for patients with an all-cause dental admission. We applied artificial intelligence to develop machine learning (ML) algorithms to predict patients at risk of 30-day hospital readmission.Methods: This study used data extracted from the 2013 Nationwide Readmissions Database (NRD). There were a total of 11,341 cases for all-cause index admission for dental patients admitted to the hospitals. Descriptive statistics were used to analyze patient characteristics. This study applied five techniques to build risk prediction models and to identify risk factors. Model performance was evaluated using area under the receiver operating characteristic curve (AUC), and accuracy, sensitivity, specificity and precision.Results: There were 11% of patients admitted for ACDC readmitted within 30 days of hospital discharge. On average, the total charge per patient was $131,004 for those with 30-day readmission (n=1254) and $69,750 for those without readmission (n=10,087). Factors significantly associated with 30-day hospital readmission included total charges, number of diagnoses, age, number of chronic conditions, length of hospital stays, number of procedures, Medicare insurance and Medicaid insurance, and severity of illness. Model performance from all methods was similar with the artificial neural network showing the highest AUC of 0.739.Conclusion: Our results demonstrate that readmission after hospitalization with ACDC is fairly common. If one-third of the 30-day readmission cases can be avoided, there is a potential annual saving of over $25 million among the twenty-one states represented in the NRD. The ML algorithms can predict hospital readmission in dental patients and should be further tested to aid the reduction of hospital readmission and enhancement of patient-centered care.Keywords: machine learning, dentistry, quality improvement, risk prediction, healthcare policy, precision medicine
- Published
- 2020