Scientific Programme
Abstract
Register
Abstract
Year 2021
October 2021

SHBC1316

Abstract Title
A cost prediction model for capitation in Singapore
Authors

P.DEORANI1, RACHEL JL LIM1, ETHAN EH BEH1, JAMES WL YIP2

Institutions

MOH Office for Healthcare Transformation Pte Ltd1, National University Heart Centre2

Background & Hypothesis

Healthcare costs in Singapore have been rising significantly. A capitation based payment system can optimize the utilization of healthcare resources for maximum improvement in clinical outcomes for a fixed budget. This payment amount is adjusted for each patient’s expected utilization based on the expected progression of their clinical complexity using a relative risk prediction model.

Methods

Relative risk prediction models were developed using ICD 10 AM diagnosis codes. Various machine learning techniques were used and compared with the CMS-HCC (Center for Medicaid services – Hierarchical clinical categories) used in the US. All the models were analysed based on accuracy as well as possible biases in certain population subgroups. Shapley values were used for interpret the outcome of the models.

Results

Machine learning models outperformed the CMS-HCC models significantly in terms of both accuracy and the biases with the best model utilizing a 5 year look back of all diagnosis codes using a machine learning Bayesian Ridge Regression method. The R2 score was 7.6% with 90% of medians of a random test subgroup within 10% difference of actual predicted cost in the following year. Since capitation estimates are performed for groups of patients, these models should be sufficiently accurate for financial planning or block reimbursement purposes. This model has also been shown to have acceptably low biases for sub-populations.

Discussion & Conclusion

The R2 scores and biases in machine learning models were suitable for  developing a relative risk model for capitation in Singapore.

Top