SHBC1022

H.F.R.LIM¹, J.T.KUAN¹, X.ZHANG¹, D.Y.CHENG¹, I.Y.O.LEONG¹, C.S.WONG¹, J.TEY¹, S.C.LOH¹, E.F.SOH¹, W.Y.LIM¹

Tan Tock Seng Hospital¹

Risk adjustment and patient segmentation are key components of population health management. Many population segmentation tools require comprehensive health information spanning the full care continuum. We assessed the utility of applying the ACG® System, a well-established risk adjustment tool, as a risk segmentation tool using only hospital data in a hospital patient population.

Hospital encounters and diagnoses codes of, and medications prescribed to 100,000 randomly selected Tan Tock Seng Hospital patients from 1 January to 31 December 2017 were used to obtain ACG® outputs such as Resource Utilization Bands (RUB). Hospital costs, admission episodes and mortality in the subsequent year were used to assess the utility of ACG® System as a risk stratification system.

Patients placed in higher RUBs had higher prospective (2018) healthcare costs, and greater likelihood of having healthcare costs in the top 5th percentile, having 3 or more hospital admissions, and dying in the subsequent year. A combination of RUBs, ACG® System– generated rank probability of high healthcare costs, age and gender had good discriminatory ability for all 3 outcomes, with Area under the Receiver-Operator Characteristic curve (AUC) values of 0.827, 0.889 and 0.876 respectively. Using machine-learning methods improved the AUCs for predicting top 5th percentile of healthcare costs and death marginally, by about 0.02. Machine learning methods could predict actual prospective healthcare costs with an adjusted R2 of 0.385.

The ACG® System can be used to stratify risks in a hospital population with hospital-only data.