We modeled the impact on risk scores if services are avoided during a code capture period.
The dramatic and new effects of the ongoing COVID-19 pandemic are far-reaching and changing rapidly. While there is a great deal of focus on resource availability and handling a potential influx of severe inpatient cases resulting from COVID-19 infections, the majority of the United States saw a dramatic reduction in healthcare services around March and April 2020 and measurable reductions continue with great variation across the nation.1
As with many prospective risk adjustment models, Medicare Advantage (MA) and Part D (PD) risk scores are based on medical claims, more specifically diagnoses from face-to-face visits, from the year prior to the year in which the risk score drives revenue. For 2021 MA payments, 2020 diagnoses are the basis of the final risk scores. To the extent that beneficiaries delay or avoid care, there may be fewer face-to-face encounters with providers where diagnoses can be recorded and applied toward 2021 risk scores. While the Centers for Medicare and Medicaid Services (CMS) has announced additional flexibilities2 in including telehealth-based diagnoses in risk score calculations, a significant reduction in overall services is likely to result in a material reduction in both MA and PD risk scores. This article outlines the results of an analysis we performed to support 2021 MA and PD bids.
Our modeling spanned a variety of scenarios that apply in different degrees to different portions of the country. In general our results range from about a 1% reduction to risk scores (revenue) to a 9% reduction; scenarios that seem most plausible to us centered around roughly a 3% reduction. Each specific situation will vary, and the results are very dependent on the progression of the disease, the timing of stay-at-home orders, the government-enforced shutdowns, and the effects of public opinion and compliance with public health recommendations. Readers should engage their own competent experts to understand how these results apply to their specific situations and potentially to model different scenarios using emerging information.
Results
We present nine scenarios intended to illustrate a range of potential outcomes on 2021 MA3 and PD risk scores. These scenarios were selected based on information available in advance of the filing of MA-PD bids.
We used the 2017 and 2018 Medicare 5% Sample database (Limited Data Sets4) to model the exclusion of diagnoses from the baseline claims for a variety of time periods during a calendar year and at a variety of elimination rates. We intend to create a range of scenarios that will assist in understanding the impact on projected 2021 risk scores. The use of any scenario will require significant judgment and qualified experience, as with most things related to the evolving pandemic and healthcare finance. We do not identify a single “best” risk score impact due to the many considerations outlined below.
Figure 1 shows our scenarios along with brief descriptions of how we adjusted the diagnoses that drive risk scores in the 5% Sample. All scenario impacts are presented as multiplicative adjustments, not additive adjustments. That is, a -1.7% impact represents a multiplicative risk score reduction factor of 1 minus 1.7%, or a factor of 0.983 to be multiplied by a baseline risk score for a population.
Figure 1: Scenarios
Description | Scenario Number | Decrease from Baseline | |
---|---|---|---|
Part C | Part D | ||
Exclude all services from Mar 15 to May 14 | 1 | -3.6% | -2.9% |
Exclude all services from Mar 15 to July 14 | 2 | -8.0% | -6.5% |
Exclude all services from Mar 15 to May 14 and from Nov 1 to Dec 31 | 3 | -9.0% | -7.1% |
Exclude all services from Nov 1 to Dec 31 | 4 | -4.8% | -3.4% |
Exclude services from Mar 15 to June 6 at the following rates: 90% of preventive, 33% of inpatient, 40% of all other | 5 | -1.7% | -1.5% |
Exclude services from Mar 15 to Aug 29 at the following rates: 90% of preventive, 33% of inpatient, 40% of all other | 6 | -3.5% | -3.1% |
Exclude services from Mar 15 to June 6 at the following rates: 90% of preventive, 50% of inpatient, 60% of all other | 7 | -2.8% | -2.4% |
Exclude services from Mar 15 to June 6 at the following rates: 70% of preventive, 15% of inpatient, 20% of all other | 8 | -0.8% | -0.7% |
Exclude services from Mar 15 to June 6 at the following rates:
90% of preventive, 33% of inpatient, 40% of all other Exclude services from June 6 to Dec 31 at the following rates: 50% of preventive, 7% of inpatient, 11% of all other |
9 | -3.1% | -2.9% |
- Scenario 1 is intended to provide the impact of assuming two months of completely eliminated services, while scenario 5 is intended to provide a perspective where only a portion of services are eliminated over slightly more than two months. In practice, services would never be completely eliminated during a specific time period.
- Scenario 2 shows the impact over an extended initial period (four months) where all diagnoses for services are excluded.
- Scenario 3 also is a four-month full exclusion scenario; however, it models the second two-month reduction due to a second period of shutdown occurring at the end of the year.
- Scenario 4 quantifies the impact of a two-month full exclusion scenario in the winter. This scenario is provided to assist in making adjustments when assuming all, or effectively all, care deferred at the beginning of 2020 will be delivered in advance of a second shutdown. Further, a comparison of scenarios 1 and 4 illustrates a seasonality impact.
- Scenarios 5 through 9 model varying levels of service elimination by type of service. To simulate varying levels of service elimination, we exclude all claims and associated diagnoses incurred on a specific set of dates within the period. The proportion of excluded dates of service are consistent with the percentages described in each scenario in Figure 1.
It is worthwhile to note that the impact of scenario 3 is greater than the impact of scenarios 1 and 4 combined. This is because beneficiaries may have multiple encounters with a particular diagnosis, and when one diagnosis is in the March-May time period and a second is in the November-December time period, eliminating one of those time periods does not eliminate the diagnosis from calendar year, but eliminating both time periods does. To the extent that follow-up care is delivered a few months after initial care, or follow-ups with providers are scheduled on a periodic basis, the time elapsed between the first and potential second shutdown could have an impact on final 2021 risk scores.
Methodology and additional considerations
The CMS-Hierarchical Condition Category (HCC) risk score model5 does not consider the number of diagnoses for a particular condition; rather, once a condition is triggered by an eligible diagnosis code in a calendar year, the beneficiary’s risk score increases for that condition. As such, unlike other analyses measuring impacts to healthcare claims, for risk score purposes it is important to consider whether care is delivered before the end of the year or is delayed beyond the code capture period or eliminated entirely. It may be possible that delaying second-quarter medical care by three months will not have an effect on risk scores, as the diagnoses may still be coded within the calendar year.
We started our analysis by creating a baseline scenario from CMS’s Medicare 5% Sample, which we reconcile to CMS’s nationwide risk scores. We then summarized the baseline data and produced alternative scenarios by excluding diagnoses from selected claims, recalculating the Part C and Part D risk scores, and dividing the resulting risk score by the baseline risk score.
We exclude diagnoses based on admission date for inpatient claims and date of service for other claim types. We did not consider continuous episodes of care, nor did we consider the severity or specific type of diagnosis associated with the care. Based on recent research studies,6,7 typical expectations for which care would be avoided do not appear to apply and, as such, we have not attempted to establish relationships between the types of care that would tend to be avoided and the correlation of the diagnoses associated with such care. Our assumptions about elimination rates are as follows:
- Scenarios were selected to illustrate a range of outcomes by varying time periods, the inclusion of a second shutdown due to severe infection rates (scenario 3), higher and lower rates of eliminated medical care (scenarios 7 and 8), and a loosening of restrictions accompanied by an extended lower-intensity elimination of care (scenario 9).
- For our initial inpatient care elimination rates, we assumed 50% of care would be delayed and 35% of the delayed care would be delivered during a later time period in 2020, resulting in an inpatient and mental health reduction of about 33%.
- We assumed preventive care is highly elective and would be a low priority for the vast majority of beneficiaries and thus reduced preventive care by 90%.
- All other services were assumed to have a 60% deferral rate, with 35% of the deferred care delivered later in 2020, resulting in a service reduction rate of about 40%.
We created scenarios 7 and 8 to increase and decrease, respectively, the elimination ratios in order to provide sensitivity impacts assuming either higher or lower service elimination rates.
We assume telehealth will reduce the quantity of eliminated care, but did not explicitly model the potential impact of including increased telehealth claims either in total or by specialty. Historical data in the 5% Sample shows low telehealth utilization, where regulations were more stringent before this pandemic.8 As one exception, we assumed the pre-COVID-19 tendency for more telehealth in mental healthcare. In scenarios 5 through 9, we applied the inpatient eliminated service ratio to mental healthcare.
We tested for variations in risk score impacts by risk score model and by Risk Adjustment Processing System (RAPS) versus Encounter Data System (EDS)9 risk score filtering logic and found no significant differences. To avoid false precision, we include a single impact that represents both EDS and RAPS and their respective risk score models in our results. Further, we tested for variations by population type and generally found small variances. The notable exception was for institutionalized beneficiaries, whose Part D risk scores were impacted by approximately half as many noninstitutionalized beneficiaries as in scenario 1.
Finally, we tested for variations in risk score impacts by individual beneficiary 2017 allowed cost level. We intended to determine whether risk score adjustment factors should vary by the relative morbidity of the projected population. We did not find significant variations in risk score adjustment factors by morbidity level. Note that it may not be appropriate to use the results of this analysis to project risk scores for plans with significant portions of members who are new to Medicare (and therefore receive the flat new enrollee risk score, which would not be impacted by 2020 diagnoses). Given the wide range of results from the selected scenarios above, and the relatively small variations by population and claim levels, we did not include the additional variables in our results.
Deferred care will vary by geography and we have not attempted to capture geographic variations. Further, our analysis is based on fee-for-service (FFS) data and does not consider the effects of changes in coding accuracy programs implemented by specific organizations.
Qualifications
Rob Pipich and Deana Bell certify that we are members of the American Academy of Actuaries. We meet the Academy's qualification standards for this type of analysis.
In performing this analysis we relied on data and other information from CMS. If the underlying data or information is inaccurate or incomplete, the results of our analysis may likewise be inaccurate or incomplete. We performed a limited review of the data used directly in our analysis for reasonableness and consistency and have not found material defects in the data.
Our estimates rely on a number of key assumptions that are subject to extreme uncertainty given the limited experience available at this time and the rapid and constantly changing development of the pandemic. The assumptions supporting the conclusions outlined in this paper are based on a combination of publicly available data and Milliman’s proprietary modeling and CMS data, and represent our best estimates at the time of publication. Many of these assumptions will likely change over the coming weeks and months.
Scientific knowledge of these items is incomplete and new data on the spread of COVID-19 in the United States is constantly emerging. In addition, actions taken by governmental authorities and the healthcare system related to the COVID-19 pandemic are rapidly changing. We expect these assumptions to change as more information becomes available, and our team of consultants closely monitor the impact of COVID-19 to ensure our projections are calibrated to the most current information. Due to the limited information available on the pandemic, any analysis is subject to a substantially greater than usual level of uncertainty.
This analysis examines average impact to the nationwide 2021 MA and PD risk scores; the impact for 2022 and beyond are not included in this analysis. The impact of the pandemic on claim costs is outside the scope of this analysis. The timing of a vaccine is assumed to not be available on a scale of a significant size during calendar year 2020.
1Estimating the impact of COVID-19 on healthcare costs in 2020: Key factors of the cost trajectory. Retrieved August 18, 2020, from https://www.milliman.com/en/insight/Estimating-the-impact-of-COVID19-on-healthcare-costs-in-2020.
2CMS (April 10, 2020). Applicability of diagnoses from telehealth services for risk adjustment. Retrieved August 18, 2020, from https://www.cms.gov/files/document/applicability-diagnoses-telehealth-services-risk-adjustment-4102020.pdf.
3Consistent with the Bid Pricing Tool methodology, Part C risk score impacts exclude end-stage renal disease (ESRD) and hospice beneficiaries, while Part D impacts include all beneficiaries.
4CMS. Limited Data Set (LDS) Files. Retrieved August 18, 2020, from https://www.cms.gov/Research-Statistics-Data-and-Systems/Files-for-Order/LimitedDataSets/index.
5CMS. Risk Adjustment. Retrieved August 18, 2020, from https://www.cms.gov/Medicare/Health-Plans/MedicareAdvtgSpecRateStats/Risk-Adjustors.
6Krumholz, H.M. (May 14, 2020). Where have all the heart attacks gone? New York Times. Retrieved August 18, 2020, from https://www.nytimes.com/2020/04/06/well/live/coronavirus-doctors-hospitals-emergency-care-heart-attack-stroke.html.
7Koeher, K.E. & Macy, M.L. Emergency department patients in the early months of the coronavirus disease 2019 (COVID-19) pandemic—what have we learned? JAMA Network. Retrieved August 18, 2020, from https://jamanetwork.com/channels/health-forum/fullarticle/2767238.
8“Restrictions under the Medicare program regarding beneficiary location, provider type, and geography have limited the adoption of telehealth services provided to Medicare beneficiaries.” See https://us.milliman.com/en/insight/medicare-telehealth-coverage-expansion-during-the-covid-19-pandemic.
9See https://us.milliman.com/en/insight/have-we-reached-parity-between-medicare-advantage-raps-and-eds-risk-scoresfor more details on RAPS versus EDS risk scores.