A machine learning algorithm called the Baltimore score (B score) can potentially help hospitals predict which discharged patients will be readmitted, according to a University of Maryland School of Medicine study that was published in JAMA.
Hospital readmissions occur in approximately 20% of patients hospitalized in the US and are associated with increased patient harm and expenses. As such, hospitals have put an emphasis on identifying which patients incur the greatest risk of being readmitted. “A significant proportion of readmissions may be preventable with better planning and follow-up for how the patient would transition back into the community,” said the study’s lead researcher Daniel Morgan, MD, MS, Associate Professor of Epidemiology and Public Health at the University of Maryland School in a press release.
Machine learning models have increasingly been used to make predictions in medical settings. A few existing machine learning readmission risk-assessment tools include LACE index, the HOSPITAL score and the Maxim/RightCare score. They evaluate a limited set of variables for each patient, such as length of stay in a hospital, type and severity of admission, and types and amounts of medications. The B score algorithm was individualized for three of the University of Maryland Medical System hospitals in different settings, after assessing more than 8,000 possible data variables from September 1, 2014 through August 31, 2016.
In this prognostic study, researchers identified 14,062 consecutive adult hospital patients with 16,649 discharges from the University of Maryland Medical Center (a tertiary care medical center), Saint Joseph Medical Center (a suburban community medical center), and Maryland Midtown Hospital (an urban safety-net hospital) from September to December 2016. Patients were excluded from this study if they were not considered eligible discharges by the Centers for Medicare & Medicaid Services (CMS) or the Chesapeake Regional Information System for Our Patients (CRISP), resulting in 10,732 unique patients (52.2% male, mean age, 54.56) with CMS-eligible discharges. Of that total, (n=6,214) came from hospital one, (n=3,440) came from hospital two, and (n=1,078) came from hospital three. Overall, (n=1,422) patients were readmitted. This study’s primary outcome was stipulated as all-cause hospital readmissions within 30 days of initial visit, excluding planned readmissions based on CMS and CRISP definitions and reporting.
Results May Help Solve ‘Readmission Puzzle’
According to the study results, the area under the receiver operating characteristic curve (AUROC) for individual rules was perceptibly lower for the HOSPITAL score, 0.63 (95% CI, 0.61 to 0.65) juxtaposed to the 0.66 for modified LACE score (95% CI, 0.64 to 0.68; P < .001). The B score machine learning score was notably higher than all others, scoring a 0.72 (95% CI, 0.70 to 0.73) 48 hours after admission, and subsequently increasing to 0.78 (95% CI, 0.77 to 0.79) at the time of discharge (all P < .001). Moreover, at the hospital using Maxim/RightCare score, the AUROC was 0.63 (95% CI, 0.59-0.69) for HOSPITAL, 0.64 (95% CI, 0.61-0.68) for Maxim/RightCare, and 0.66 (95% CI, 0.62-0.69) for modified LACE score. The B score was 0.72 (95% CI, 0.69-0.75) 48 hours after admission and 0.81 (95% CI, 0.79-0.84) at discharge. Overall, when directly comparing the B score with the values for modified LACE, HOSPITAL, and Maxim/RightCare scores, the B score was able to identify the same number of readmitted patients while flagging 25.5% to 54.9% fewer patients.
“The widespread use of electronic health records has enhanced information flow from all clinicians involved in a patient’s treatment,” said UMSOM Dean E. Albert Reece, MD, PhD, MBA, University Executive Vice President for Medical Affairs and the John Z. and Akiko K. Bowers Distinguished Professor. “This study underscores how patient data may also help solve the readmission puzzle and, ultimately, improve the quality of patient care.”
The study authors concluded by writing that “although the B score had access to some social determinants of health such as home zip code, insurance type, and homelessness, the validity of such data in the EHR is uncertain. Further work on predicting and targeting readmission prevention efforts needs to account for social determinants of health.”
— Medical Xpress (@medical_xpress) June 5, 2019
A University of Maryland School of Medicine study suggests new computer analytics may solve the hospital readmission puzzle https://t.co/0HcGsKJHK4
— Art Fridrich (@Ahighervision) June 5, 2019