Meet the Author
| Thread Title | Posted By | Contents |
|---|---|---|
| Welcome from the MMRR Editor-in-Chief | David Bott | Welcome to our first MMRR Meet the Author event! We look forward to an interesting and informative discussion. The mission for MMRR is to inform the future of the Medicare, Medicaid, and Childrens' Health Insurance programs. Our goal in this event is to offer an opportunity to interact with the lead author to learn more than is possible simply by reading the article. I want to thank Professor DeLia for taking time out of his busy schedule to respond to your questions and comments. For those not familiar with the Medicare Shared Savings Program, the official CMS Web page. ProfessorDeLia's research paper under discussion this week can be accessed at: download. Let the learning begin! David Bott Editor-in-Chief, Medicare & Medicaid Research Review If you have any questions about how to participate, please feel free to email the MMRR staff at MMRR-Editors@cms.hhs.gov. CMS staff will be monitoring the site and moderating posts to comply with HHS policy. Please note that all posts in this MMRR Meet the Author event are the opinions of individuals and do not represent the official views and policies of CMS, HHS, or the U.S. federal government. |
| Welcome to the CMS Meet the Author discussion forum | Derek DeLia | Hello and welcome to the CMS Meet the Author discussion forum. In this forum, we hope to use the recent publication of Statistical Uncertainty in the Medicare Shared Savings Program; in Medicare & Medicaid Research Review to generate discussion around Medicare ACOs, ACOs outside of Medicare, and the broader role of shared savings arrangements in healthcare payment and delivery reform. Please submit any questions or comments you have about the publication or related topics. Forum posts regarding policy, practice, and/or research methodology are all welcome. Along with my co-authors, Don Hoover and Joel Cantor,I thank you for your interest in our article. Sincerely, Derek DeLia (Rutgers Center for State Health Policy). |
| Implications for the CMMI ACO Program | Francois de Brantes | Hi ,Derek your paper suggests, appropriately, that insurance risk is difficult to manage when there are small numbers of patients (something that PHOs and small HMOs discovered to their detriment a couple of decades ago), and that as a result, the confidence intervals around a specific ACO's performance will be wide. So here are my questions: 1. Doesn't this imply pretty clearly that the min threshold of patients for inclusion in the ACO program should be raised? And if so, how high should a reasonable min be set? 2. Much of the variability that stems from insurance risk is often caused by the small number of patients that have conditions for which there is a known underlying uncontrollable variability in costs, which cannot be adjusted for in case-mix or severity. Would your conclusions and analysis change if the cohort of Medicare patients were more tightly defined? For example, what if cancer cases, ESRD patients and others were excluded? Thanks. |
| Implications for the CMMI ACO Program | Derek DeLia | Francois, yes, the direct implication is that you need larger patient groups to measure cost performance reliably given the parameters that have been set in the MSSP. Our calculations suggest that statistical uncertainty falls pretty rapidly after 20,000 patients. Of course, that depends a bit on the model assumptions, which I discussed further in the previous post. The question about targeted patient groups is interesting and also timely given CMSs recent announcement about shared savings in the ESRD program. Clearly, smaller groups (all else equal) present much greater challenges from a statistical perspective. If the smaller group is more homogeneous, then smaller variance around the mean would reduce the level of uncertainty. For a very specific patient group like those with ESRD, the greater concern may be controllable versus uncontrollable/normal variation. The controllable part is really deterministic and not an issue of statistical uncertainty at all. In very specified patient groups, shared savings might be best used as a gradual transition to a new (lower) bundled payment for that specific diagnosis. While the broader ACO shared savings could similarly lead to a reasonable global/population-based payment, this is a much harder and longer process for a broad and constantly changing patient population. |
| Implications for the CMMI ACO Program | Francois de Brantes | What I'm suggesting is not to hone in on a small population, for the reasons you describe, but rather excluding patients that have certain conditions or procedures at very low frequencies, and for which there is high variability. So, for example, the rate of amputations in the Medicare population is low, but the variability in costs for those episodes is high -- prosthesis, length of rehab/post acute, potential infections, etc... Similarly, certain cancers are rare but have high costs and high variability. It seems to me that removing these patients from the ACO calculations could help reduce the inherent variability caused by these statistical uncertainties and therefore make the formula less subject to insurance risk and more focused on performance risk. Thoughts? |
| Implications for the CMMI ACO Program | Derek DeLia | Yes, it does make sense to remove some conditions that are subject to great variability due to insurance risk, while holding ACOs accountable for performance risk. This would reduce some of the statistical uncertainty while holding ACOs accountable for what they can control. Perhaps major trauma from car accidents or criminal violence would be good examples. I am not sure about amputations since these are sometimes done for diabetics whose condition is poorly managed. I suppose a small list of clear conditions could be developed for removal from the spending calculations. My concern is that if you remove too much, there is not a large enough financial base to find population-based savings from care coordination. There would also be some gray areas about what is truly manageable versus random. Interestingly, CMS has taken a very different approach by excluding expenditures above the 99th percentile. While the logic is fairly clear, this approach does not distinguish between the types of conditions you have raised. It actually limits opportunities for providers who focus on the so-called super-users; who generate huge expenditures from a combination of poorly managed medical and social conditions. For these reasons, condition-based exclusions (even if limited to a small set) seem more appropriate than a dollar cutoff. |
| Question: next steps? | David Bott | Derek, We seem to have a lot of registrants, but few comments or questions, so let me begin by posing a question of my own to you: Your research examines the role of statistical uncertainty regardingthe Medicare Shared Savings Program (MSSP)using models to predict the range of outcomes in determination of savings across a range of ACO characteristics. In the discussion of the study's limitations, you list a number of simplifying assumptions that can affect our understanding of the role of uncertainty in the MSSP. Which of these simplifying assumptions are the most critical to address? And, in your view, what steps would be necessary to obtain the data or information necessary to verify or modify those critical assumptions in your model? |
| Question: next steps? | Derek DeLia | Dave, shared savings formulas are based primarily on baseline spending, expected trend in spending, and performance year spending. The parameters in the MSSP shared savings formulas contain an implicit assumption that the only source of statistical variability (i.e., normal variation;) is in the performance year spending. But in reality, baseline spending is subject to this same variability and the secular trend in spending growth is similarly subject to random factors that cannot be controlled or known in advance. If each ACO knew at the outset what their baseline spending level is, they could eliminate one dimension of uncertainty and work form there. This might be possible with prospective patient assignment (as done for some Pioneer ACOs) and very timely analysis of Medicare claims data. But in the MSSP, assignment is retrospective making it impossible to know what the baseline is before entering a shared savings agreement. Similarly, an ACO would need to make a forecast of the secular spending trend to determine whether their care coordination strategies are adequate to beat the trend.; This layer of uncertainty could be removed if a fixed target were set for the performance year spending; e.g., baseline spending plus 3%. As for data/information required, it would be ideal for each ACO group to have a timely set of Medicare claims data to assess what their baseline spending looks like and the year-to-year variability in that spending. Clearly, this would require input from CMS, since providers do not have information about their patients who use services outside of the ACO. And even if they did, pulling together all of that data so quickly is a huge challenge. Like everything else in healthcare, access to complete real-time data makes everything easier. Without it, we need to move ahead as best we can and use the rear view mirror, so to speak, to monitor how well its going. |
| Question: next steps? | David Bott | Is there some middle ground between retrospective “rear view mirror” models and real time data? I’m wondering if it may be possible to define and model “classes” of ACOs based on some observations of the existing participants, and then using those models to improve projections. |
| Question: next steps? | Derek DeLia | Yes, I think there is. Despite the natural turnover, most patients in a potential ACO service area will remain in that area for many years. You could get a rough idea about what the patient base looks like before full data become available. The MSSP addresses this to some extent by having CMS provide ACOs with early assessments of their expected assigned population; based on baseline data and early ACO experience. A more nuanced modeling approach (based on ACO type, geography, etc.) might make the process more precise though it would take some work to figure it out. |
| Question for registrants | David Bott | One of the goals of MMRR is to inform the future of Medicare. We have introduced this question and answer forum as a means of learning how published papers might be understood and/or used by our stakeholders, which include health care executives, policy makers, providers, the press, and beneficiaries. Do you have questions that you'd like to ask, but feel they are not appropriate? Email us directly at MMRR-Editors@cms.hhs.gov if you have a question that you would like to ask, but will not post it, and tell us the reason why you are hesitant to post it. |
| Thank you for a successful trial event | David Bott | Dear Meet the Author participants, I would like to thank everyone who participated in this trial event: registrants, guests, commenters, and especially Dr. DeLia. This event was a great first step in creating more constructive interactions between authors reporting their research and stakeholders who are trying to use the results of that research. CMS and MMRR staff will be evaluating this event to find ways to enhance future MMRR author-audience interactions. If you have any questions or comments, please let us know at MMRR-Editors@cms.hhs.gov. Thanks again for your participation. Stay tuned for future events, news, and the announcements of published papers by subscribing to MMRR Updates Sincerely, David Bott Editor-in-Chief Medicare & Medicaid Research Review |
To read the original article associated with this event, download "Statistical Uncertainty in the Medicare Shared Savings Program" in pdf format.
To stay informed about new MMRR activities, publications, calls for papers, and data/research opportunities related to Medicare, Medicaid, and the Child Health Insurance programs, please subscribe to MMRR Updates.
Downloads
- Page last Modified: 04/05/2013 3:42 PM
- Help with File Formats and Plug-Ins
