In what aspects of value-based care is predictive analytics already making an impact? And what are some aspects in which predictive analytics has not yet significantly been deployed?
Mar. 17— Each month, Predictive Modeling News asks a panel of industry experts to discuss a topic suggested by a subscriber.
Q “In what aspects of value-based care is predictive analytics already making an impact? And what are some aspects in which predictive analytics has not yet significantly been deployed?”
“Predictive models are useful for four main purposes in value-based care:
Average risk scores can be used to characterize a population to support decisions about pricing and associated economic risk.
Individual risk scores can be used to determine which members need specific care management interventions and to prioritize and triage among multiple members needing services in limited supply based on objective criteria.
Individual risk scores can be directly used by clinical team members to support subjective professional judgement and clinical decision making.
And risk scores can be used to support risk-adjusted performance measures, including measures of quality of care, outcomes, utilization of services and cost.
“To date, risk scores have been employed successfully for all of these purposes in value-based care, but deployment has been limited and execution has often been poor. Barriers to deployment include data that is incomplete, inaccurate or insufficiently timely for the purpose and lack of training in analytic methods of the analysts who use predictive models — particularly lack of understanding of biostatistics, epidemiology and actuary science. Opportunities to improve the value of predictive models in this field include developing and using models that are designed to identify and prioritize candidates for specific care management interventions, rather than just using general models that predict future total cost of care or healthcare events.
“Use of predictive models for risk-adjusted measurement, where the intent is to make ‘level-playing-field’ comparisons of performance of different providers or comparisons of performance in different time periods should use a direct standardization method, rather than the ‘observed-over-expected’ (indirect standardization) method employed in most claims- based reporting applications.
“Finally, predictive models would add more value if the input data elements were captured proactively and consistently, including relevant clinical parameters, functional outcomes and patient preference variables, and if the risk scores were provided to members of the clinical team within EHR and care management applications where they can have the greatest impact on decision making.”
Comments