This document is submitted by the Massachusetts Health Data Consortium (MHDC) and its Data Governance Collaborative (DGC) in response to the CMS CMMI Episodic-Based Payment Model RFI (CMS–5540–NC) posted in the Federal Register on July 18, 2023 and found here: https://www.federalregister.gov/documents/2023/07/18/2023-15169/request-for-information-episode-based-payment-model
Founded in 1978, MHDC, a not-for-profit corporation, convenes the Massachusetts’s health information community in advancing multi-stakeholder health data collaborations. MHDC’s members include payers, providers, industry associations, state and federal agencies, technology and services companies, and consumers. The Consortium is the oldest organization of its kind in the country.
MHDC provides a variety of services to its members including educational and networking opportunities, analytics services on both the administrative and clinical side (Spotlight), and data governance and standardization efforts for both clinical and administrative data (the Data Governance Collaborative/DGC and the New England Healthcare Exchange Network, respectively).
The DGC is a collaboration between payer and provider organizations convened to discuss, design, and implement data sharing and interoperability among payers, providers, patients/members, and other interested parties who need health data. It is a one stop interoperability resource. The DGC primarily focuses on three areas:
This section general comments that cross multiple questions asked by CMS or that do not have specific questions in the RFI.
Our Data Governance Collaborative had a lively discussion on this topic and concluded that in most cases all providers of services throughout an episode of care must be in-network providers. As soon as an out of network provider is used the entire bolus of care should be kicked out of the episodic payment model into more traditional fee for service. Our participants feel out of network providers cannot be forced to participate as they did not sign any contracts with the payer. One possible exception to this is if there is no in-network provider able to provide services for the patient, but it is unclear exactly how this should work.
We note that No Surprises Act requirements that services provided at an in-network facility by an out of network provider be treated as in-network services may help, but only if that is extended to the reimbursement expectations as well as to patient responsibilities. As an aside, we would welcome specific regulation regarding the exact rules/extent of how this clause applies to allowed charges and payment responsibilities.
Related to the immediately previous comment, our Data Governance Collaborative also feels strongly that the network of available providers for services deemed essential to an episode of care be a factor in whether the episode of care can be enacted. If a particular specialist is required for services and the network has no specialists of that exact type, then patients in that particular plan should not be eligible for treatment under the episode of care (unless, per some of our thoughts in the previous comment, they can be required to accept payment within the strictures of the episode of care rules).
Participants in our Data Governance Collaborative note that there can be several reasons for a delay of services that could make a timeboxed episode of care problematic. Availability and staffing are major issues. If an episode of care is initiated by an ED visit that requires follow up in-patient care, there could be a delay of days before such a bed is secured for the patient. The patient remains in the ED and continues to receive care, but it is not at the level of treatment intended to be covered by the episode of care. In cases of this sort, our DGC believes the clock on the episode of care should not start until a bed is located and the patient is transferred to begin that phase of care.
Similarly, there is an ongoing staffing shortage in a wide swath of healthcare. This may affect appointment availability for a variety of services. If a patient is supposed to get two appointments for X as part of their episode of care but the providers of X don’t have any availability for three months, does the episode get stretched out or does the bundle of services fall out of the episode of care and get charged via more tradition means?
In both of these cases, there is or may be interim care that’s both less helpful to the patient and more expensive than the intended services. The question of how the additional days in an ED or extra ED visits made in lieu of seeing the expected specialist who had no appointments available is one that needs to be addressed. Sadly, we do not have the answers to these problems or how they do or should affect how bundles of care billed under an episode of care model for a patient experiencing them.
Our Data Governance Collaborative notes some frustration with being required to send mostly the same data in similar but not identical exchange mechanisms over and over again to different partners for different programs. We recognize that using USCDI/US Core-based FHIR exchange should significantly help this issue, but alignment of equity data collection, quality measures data, and other relevant data across different CMS programs (and with NCQA and other industry partners) would be extremely helpful.
This section will list specific questions asked in the care delivery/incentive structure section of the RFI and provide our responses to them.
Our Data Governance Collaborative believes some or all of the following would be helpful:
Several payer participants in our Data Governance Collaborative noted that the current payment models for home care and community-based interventions are structured completely differently from more traditional care models and it would be difficult to incorporate them into a combined episode of care payment model. We do not have any specific suggestions but agree that it is important to determine some mechanism to incorporate these services if one of the goals is equity as underserved and disadvantaged populations are more likely to use, need, and benefit from these types of services.
This section will list specific questions asked in the clinical episodes section of the RFI and provide our responses to them.
Our DGC participants have seen 30 day, 60 day, and 90 day episodes of care be implemented. In many cases, the length is dependent on the types of services included in the episode and the triggering event. For example, an orthopedic surgery will likely require more than 30 days of post-surgical physical therapy. If these services are part of the episode, the episode must be longer or suffer from an artificial time-based division of how services are compensated which adds considerable burden on the PT.
We favor having some flexibility on the episode length based on the type of triggering event and previous data showing the typical length of care needed afterward (plus some buffer given that individuals are different and will not all fall within even a standard deviation of the mean). If episodes are designed to be specific to a particular trigger, gathering and analyzing the necessary data to determine this time frame should be part of the preparation of the model. If not, allowing for flexibility in length provided it can be backed up by data should be part of the model.
We also note that network adequacy and industry issues such as staffing shortages may cause delays in care (see comments above); CMS should carefully consider how these are addressed within payment models so as to ensure equitable distribution of funds and that services are not being paid twice – once because they were anticipated to be part of an episode and once later when they’re actually provided.
The specific trigger may also play a role in how long an episode of care should last. For example, a patient may be under observation but some event triggers the need for more active care. We posit that any episode of care for that event should start at the triggering event when the more active care begins and not at the start of the observation period even though those may be part of the same visit/hospitalization/billable event from the patient perspective. If CMS chooses to start the event at the start of the observation period, a longer episode may be needed to ensure the patient receives the expected standard of care for that episode.
We recommend CMS look at data around how long more intensive post-hospital/post-surgery/post-event care happens for the event triggering the episode and build some guidelines around that (while remembering that patients are individuals and many will need more time than the average needed across a population). For example:
We urge CMS to think about and consider other situations that might be similar or involve similar patterns of care.
Perhaps the episode could be tied to a specific number or type of services rather than a timeframe for services that may need to continue over time (such as physical therapy). For example, if the typical patient needs 12-15 PT visits after a specific type of orthopedic surgery with 90% of patients dismissed by visit 18, the episode of care could cover up to 18 visits. This would also help address issues with staffing and appointment availability noted in a comment above.
This section will list specific questions asked in the participants section of the RFI and provide our responses to them.
Our DGC participants note that when taking on risk the size of a PGP matters a great deal. Do there need to be different programs for different types of practices to account for this? We did not come up with any specific suggestions around this, but want to make sure that CMS is considering it during its program design discussions.
This section will list specific questions asked in the equity section of the RFI and provide our responses to them.
Providers that typically serve “sicker” patients must be given more leeway on the financial benchmarks. One suggestion our Data Governance Collaborative had was perhaps developing three models for risk calculations:
This type of program might also incentivize providers to collect SDOH information routinely and reliably. We note we believe that USCDI-level collection including not just Z codes on claims (which are extremely limited) but full collection of concerns, assessments, goals, and interventions should be prioritized. At the same time, collecting the data just to have the data does not lead to a good patient experience so tying it to getting better financial incentives if you have actual programs should lead to a better patient experience around SDOH.
A model that gives more leeway to providers with significant patient populations with both mental and physical health issues to address may also be appropriate. The general consensus among DGC participants is this combination can exponentially increase costs, so any provider with a disproportionately high percentage of such patients will likely be at a financial disadvantage. However, we are not suggesting creating this model at this time because there may be some difficulty getting sufficient access to behavioral health data to accurately make these assessments. In addition to restrictions on sharing behavioral health data, so much behavioral health data is not paid for by patient insurance and is difficult to uncover. HRSA and SAMHSA may have access to some of this data that could be leveraged; we suggest CMS consider partnering with them to at least explore the idea.
Although we noted the possibility of having a separate risk model for participants committed to addressing SDOH needs of their patient population, another option might be to just directly give additional funds to programs that address SDOH needs. This could take the form of having an onsite food pantry, providing transportation vouchers to patients, or other programmatic efforts offered directly by the provider.
We also recommend providing a financial bump of some sorts to providers likely to have higher than average language translation needs to better serve non-native English speakers and to not disadvantage those providers likely to encounter them within their patient populations.
We are not familiar with any (beyond perhaps having a disability specialist within the patient services department), but providers that offer specific programs to directly serve disabled patients should also be considered in similar ways to any more general SDOH-focused programs noted above.
Claims data is not particularly useful for tracking health equity in our experience. Z codes are limited and demographic data is rarely included. We strongly recommend including clinical data exchange meeting at least USCDI v3 (now that it’s officially supported by ONC under the SVAP program) as part of all programs.
We believe reporting sliced by demographic data (race, ethnicity, religion, language, disability, etc) and also by geography (comparisons among providers in the same physical area) as well as modality of care (phone, video, onsite, home, etc.) would all be helpful.
In addition to CMS analysis, we suggest de-identified data related to these areas would be extremely helpful to AHRQ and other research agencies and sharing with them may also greatly increase the level and type of analysis made available about the data.
In addition, some of our Data Governance Collaborative participants note that CMS can sometimes take some time to provide public reporting of other data (one example given is data related to Stars ratings). Anything CMS can do to reduce the time between data collection and public reporting using it would be helpful.
This section will list specific questions asked in the quality measures section of the RFI and provide our responses to them.
This is not quite the question asked, but our Data Governance Collaborative participants (all well versed in quality measures) felt the following quality measures should be collected for an episode-based model:
We note there was a lengthy discussion of the usefulness, accuracy, and low response rate of patient surveys and a reluctance of some participants to give it significant weight. Some of the concerns raised include:
We did not come up with any major solutions, just a caution to take patient survey data with a grain of salt. The participants with survey experience noted that the responses to their surveys tend to be toward the extremes – people who were very happy or very unhappy with their care.
This question did not make sense to our Data Governance Collaborative participants who felt that the same exact system and questions should work in any setting and just be used as currently designed in a hospital setting. This included one participant who serves on a CAHPS committee and is very familiar with its design and contents.
We recommend the following:
This section will list specific questions asked in the payments section of the RFI and provide our responses to them.
This is a really sticky area. Our Data Governance Collaborative participants noted that risk adjustment is extremely sensitive to coding patterns. On one hand, there needs to be safeguards to ensure that upcoding doesn't happen but at the same time providers must be able to code for the legitimate conditions without fear for being dinged about it. All of the norms and standards are based on averages and expectations that patients generally fall within a standard deviation or so from those averages, but when dealing with individuals there are always outliers. Some patients will legitimately have a lot of intensive conditions that affect treatments and should be included in coding related to it.
One mechanism that seems obvious is to use clinical data to support diagnoses and coding choices used for specific procedures or events. If a patient is coded as having diabetes and hypertension and colon cancer and asthma and whatever else, use clinical data collection to verify that the patient is being treated for all of those conditions, or at least that they are noted in the patient’s history if not under active treatment. Preferably this would be done using data collected via standard operational data collection means such as US Core data exchanges over FHIR or perhaps via USCDI-compliant C-CDA documents (until everyone is ready for FHIR exchange).
This section will list specific questions asked in the model overlap section of the RFI and provide our responses to them.
Our Data Governance Collaborative agrees that the best way to do this is to incentivize interoperability efforts and consistent sharing of standardized data that drives collaboration and reduces redundancies.