@nealemahoney @afrakt @JonSkinner17 @UpshotNYT First of all, this is a remarkable example of DiD/RCT health care economics. Well done.
@nealemahoney @afrakt @JonSkinner17 @UpshotNYT But I cannot support the claim that more health care policy analysis needs to be done through DiD methods. Equally so, I do not support RCT methods applied to health care payment demonstrations.
@nealemahoney @afrakt @JonSkinner17 @UpshotNYT First of all, I question how randomization is accomplished. If we are faithful to the "RCT as ideal," the treatment and control groups would be drawn from the same underlying distribution. All DiD analyses of Medicare policy I have reviewed, in fact, deviate from this ideal.
@nealemahoney @afrakt @JonSkinner17 @UpshotNYT This is concerning because matching will not correct for the imbalance created between treatment and control groups. Studies since the late 1960s show that unobservable causes (including "decision-making") drive variations. We shouldn't be surprised by RCTs with biased results.
@nealemahoney @afrakt @JonSkinner17 @UpshotNYT Moreover, the parameter of interest (difference in population means; Average Treatment Effect) is easily influenced by outliers, which are endemic to health care spending studies. Again, as with covariate imbalance, skewness of the ATE is not ideal.
@nealemahoney @afrakt @JonSkinner17 @UpshotNYT More to the point of health care payment reform, we should be focused on marginal effects, not effects averaged over populations. We should be thinking about how to find the top decile of performance and understand what drives it.
@nealemahoney @afrakt @JonSkinner17 @UpshotNYT These mandatory payment models -- designed to facilitate RCTs -- also cannot support the kind of delivery system reform envisioned by most CMS leaders, past and present. CJR makes assumptions about the approp. locus of control; assumptions that are likely not universally correct.
@nealemahoney @afrakt @JonSkinner17 @UpshotNYT Single-episode mandatory models (i) heighten the financial risk for participants; (ii) diminish the role of clinicians best positioned to improve patient outcomes; and (iii) replace market-based incentives with regulatory compliance.
@nealemahoney @afrakt @JonSkinner17 @UpshotNYT Only the largest providers are likely to have sufficient volume in single-episode programs to warrant the investments necessary to redesign care. By contrast, programs covering a majority of conditions across service lines are far more effective to support the investments.
@nealemahoney @afrakt @JonSkinner17 @UpshotNYT In a single-episode bundled payment model, the financial risks are also worse than those under multi-episode models. Programs with only a single episode type foist on providers uncontrollable performance volatility due to high cost patients in a small number of episodes.
@nealemahoney @afrakt @JonSkinner17 @UpshotNYT Outlier episodes are a feature of every program. But the impact falls disproportionately on smaller programs, which do not have sufficient volume of episodes to buffer the effect of one or two high-cost patients.
@nealemahoney @afrakt @JonSkinner17 @UpshotNYT Voluntary episode-based payment models, by contrast, encourage competition. In particular, the unique design of BPCI Advanced has not favored any single type of health care organization.
@nealemahoney @afrakt @JonSkinner17 @UpshotNYT In places where physician groups are more willing to accept the risk of episode-based payments, they have proven an ability to achieve success and integrate with hospitals and community providers. In others, hospitals have shown the ability to also achieve consistent success.
@nealemahoney @afrakt @JonSkinner17 @UpshotNYT Ideally, there should be dynamic competition in every community. Medicare should avoid ‘picking winners’ and let innovation spring from any organization willing to accept clinical and financial responsibility for patients in acute episodes.
