Incorporating risk adjustment into an alternative provider payment arrangement can help payers and providers share risk, align financial incentives, and reduce health plan costs. There are many risk adjustment models on the market for payers and providers to choose from. However, both parties need to know the predictive abilities of a risk adjuster and its implications on projected reimbursement levels. Milliman consultant Ksenia Whittal provides some perspective in her article “Provider payment: What does risk adjustment have to do with it?”
Not all risk adjusters are created equal. There are multiple models available on the market and they vary in their predictive abilities, the populations they are calibrated to assess, and the time periods for which analyses can be conducted. It is no surprise that the results of any risk-adjustment analysis will be more reliable using a model with stronger predictive power. In large part, the predictive power will be driven by the algorithms underlying the model, but consideration should also be given to the model’s intended use. For provider payment specifically, it is critical to ensure that the variables used for risk score development are resistant to manipulation by providers and do not create perverse incentives. Examples of potentially problematic variables include incurred cost in a prior period, procedures, or diagnostic testing, because these items could lead to model exploitation and distort true morbidity levels.
Arrangements should also specify whether risk adjustment will be applied to actual experience at the end of a contract period (a concurrent analysis), and result in a retrospective adjustment to prior payments to account for the risk level actually encountered, or if the rate will be set prospectively, using current experience to project the appropriate rate for the next contract period. Concurrent risk adjustment is far more accurate than prospective because it seeks to explain what already happened rather than predict what will happen. However, concurrent models can introduce uncertainty during the payment year as to what the retrospective settlement may be. This uncertainty can create challenges for insurers and providers in their budgeting and financial reporting processes. Additionally, a prospective approach should exclude the use of prior cost levels to project future risk levels in order to avoid incentivizing activities that would artificially inflate costs and result in exaggerated prospective risk scores.
Beyond ensuring that the model used is a strong and accurate predictor, it is also important to choose a risk adjuster that will most closely model results for the population in question or the payment arrangement in place. Risk-adjustment models should ideally be calibrated for a population reasonably similar to the one being analyzed. For instance, a model calibrated to a commercial population will not generally be appropriate for risk-adjusting a population of Medicare enrollees, and vice versa. Broad population categories such as these have different morbidity profiles, and risk adjusters should target and be calibrated to capture these nuances, or at least a study should be done to check if a model is performing adequately if it is used on a population that differs significantly from the one used in calibration.
For more Milliman perspective on risk adjustment, click here.