Response to Comments on Changes in Methodology Since 1999 Rates: Risk Adjustment

We received 8 comments from managed care associations, Medicare+Choice organizations, and other parties. Several of the comments were very lengthy, and contained detailed analyses and recommendations. As we note at various junctures below, detailed analysis and information which will respond to many points raised by the commenters are contained in HCFA's March 1, 1999, Report to Congress: Proposed Method of Incorporating Health Status Risk Adjustors into Medicare+Choice Payments. The report of Health Economics Research, Inc. (HER), entitled Principal Inpatient Diagnosis Cost Group Models for Medicare Risk Adjustment, is attached to the Report to Congress as an appendix.

Comment: Several commenters recommend a delay in implementation of risk adjustment. These commenters variously cited the need for improved data collection and tracking, potential disruption in payments to M+C organizations, absence of an institutional adjustment, and the reliance of the PIP-DCG model on hospital inpatient data alone as reasons for delaying risk adjustment.

Response: We do not have the authority to delay implementation of risk adjustment, which is required by statute on January 1, 2000. In addition, we believe that a delay in implementing risk adjustment is unwarranted. We have analyzed the PIP-DCG system sufficiently to be confident that it represents an improvement over the current system of demographic-only adjustment, that it provides an appropriate interim step toward a comprehensive risk adjustment model, and that it provides appropriate levels of payment for different classes of beneficiaries. We believe that the blend transition methodology should relieve concerns about disruption of payments, especially since the initial blend percentage for the risk-adjusted portion is 10 percent. We respond to other issues, such as the absence of an institutional adjustment and the reliance of the model on hospital data, below.

Comment: One commenter objected to the decision to implement risk adjustment in a manner that produces savings in Medicare payments. The commenter requested that HCFA consider implementing risk adjustment in a budget neutral manner.

Response: While budget neutrality has been mandated by law for other payment system changes, there is no provision in the law requiring that risk adjustment be budget neutral. The purpose of implementing risk adjustment is to correct for historical payment errors caused by biased selection. The current payment system uses only demographic factors and has long been criticized as inadequate. Risk adjustment is a mechanism to pay managed care more accurately based on a model which predicts future expenditures for an enrollee based on health status and demographic factors. Based on the PIP-DCG model, if a managed care enrollee is predicted to have a higher level of expenses, then payments will be higher. Conversely, if predicted expenses are lower, payments will be lower. There is considerable empirical evidence that the Medicare program has been overpaying managed care organizations due to "selection bias" (i.e., the enrollment of healthier beneficiaries in managed care). To the degree that there has been such selection bias, risk adjustment can and should yield savings for the program. We discuss the empirical evidence for selection bias and overpayment of managed care plans by the program in the March 1, 1999, Report to Congress.

Comment: Commenters generally supported a phase-in of risk adjustment. However, some commenters recommended a longer phase-in period, such as 10 years. Some commenters expressed a preference for corridors over the blend methodology, or for a combination of corridors with the blend.

Response: We believe that a 10 year transition would be excessively long. Among other considerations, a transition of that length would long delay full implementation of a comprehensive risk adjustment system, which will take place in 2004 under our transition schedule. We also carefully considered adoption of a corridor methodology, either alone or in combination with the blend, in our transition strategy. We concluded that the blend methodology offered the best combination of appropriate incentives, simplicity, and feasibility. The blend-only methodology is both familiar from several previous transitions (e.g., both operating and capital PPS) and easily comprehensible. It also provides the most straightforward manner of proceeding from payment based fully on demographic adjustments to full risk-adjusted payment. We also believe that a blend-only methodology more effectively promotes the goals of risk adjustment during the transition period. We believe that the blend method will provide adequate safeguards against abrupt changes, in particular by providing initially for a low blend percentage of the risk-adjusted payment rate. While the corridors method might also have contributed to providing stability against abrupt payment changes, our analysis showed that implementing this methodology would have created serious systems challenges which would have been difficult to resolve.

Comment: One commenter expressed concern about the potential for large fluctuations in payments to small Medicare+Choice organizations and the possibility that some organization-level scores will be outliers that could adversely affect the stability of the organization. The commenter recommended that HCFA identify mechanisms to deal with these potential problems.

Response: Our preliminary estimate, based on encounter data submitted by November 5, 1998 by 195 organizations that had submitted sufficient data, was that aggregate payments would decrease 0.76 percent, assuming current enrollment mixes. This estimate reflected the blend of 90% demographic adjusted amount, 10% risk adjusted amount to be implemented in 2000. We have recently estimated the impact on the 285 plans that were active in September, 1998 and that did not terminate their contracts with Medicare in 1999. (Included in this group are 10 plans that merged into other active plans as of January 1, 1999.) The estimated impact of risk adjustment for 2000 on this group of plans is -0.7 percent, taking into account the blend percentages in effect for 2000. While the impact on specific organizations will vary, our analysis suggests that, except for highly unusual circumstances (e.g., a high proportion of working aged enrollees), the maximum decrease in payment to any organization from risk adjustment alone will be less than 2 percent. The analysis did not suggest that smaller organizations, or any other specific category, would experience a disproportionate impact. We will, however, continue to monitor the impacts on organizations throughout the transition period.

Comment: One commenter expressed concern about the data submission process and its adequacy to support initialimplementation of the PIP-DCG model on January 1, 2000. The commenter objected that the process has been cumbersome and unnecessarily resource intensive.

Response: Hospital encounter data were collected from managed care organizations for discharges between July 1, 1997 and June 30, 1998. Approximately 1.5 million encounters were submitted to HCFA for over 5.7 million beneficiaries. The volume of data received is sufficient to generate an estimate of the impact of risk adjustment, and to conduct other analysis in order to prepare for implementation of risk adjustment. Based on this experience, we are confident that sufficient data will be generated to calculate beneficiary risk scores and other information necessary for implementation of the PIP-DCG model on January 1, 2000.

A range of problems in the submission of encounter data have arisen. These problems have included: not following the required UB-92 format, difficulties in accurately tracking counts of discharges, failure to arrange hospital submission of encounter data, difficulties in understanding Fiscal Intermediary reports, and HCFA/FI and FSS processing problems. Plans themselves may have problematic data processing systems in-house. We have worked with Medicare+Choice organizations, managed care associations, and other parties to address many specific issues that have arisen concerning data transmission and processing, and we will continue to do so. HCFA has taken a number of specific steps to facilitate and improve the encounter data submission process. These activities have included the following:

• Encounter Data Reconciliation Analyses. HCFA has shared with plans analyses of their individual plan level data which have been successfully received at HCFA. We have further conducted analyses upon request at the provider level and by the different methods of submission to help explain discrepancies. We are in the process of sharing these analyses with the plans. The detailed provider level analyses are requiring additional time to conduct and the results of these analyses will be shared with plans over the coming weeks.
• Onsite Consultations. HCFA's contractor is planning a series of onsite consultation visits to 20 plans in order to learn more about the process of data submission. The majority of the 20 plans selected for the visits are plans that have experienced problems with encounter data submission. The information gained during these visits will assist plans to identify and resolve problems.
• HCFA Data System Fixes. Recently, processing problems have been identified that relate to beneficiaries who change from one plan to another. The estimated number of affected encounters from all plans is less than 3,000. These problems will be fixed over the next 2 months and they are not expected to impact the March 1 rate estimates, which in any case will not be used to make direct enrollee payments.
• Communication with the FIs. HCFA has shared data problems raised by the plans with the FIs. Furthermore, discussions between HCFA, FIs, and plans have been encouraged in order to address problems.

Comment: Several commenters objected to the exclusion of 1 day stays from the final PIP-DCG groups. Commenters noted that Medicare+Choice organizations may have a higher proportion of these stays than in Medicare fee-for-service. One of the commenters requested impact analysis of the exclusion and a list of any diagnoses that are disproportionately affected. One commenter challenged the assumption that short stays, defined as one day or less, are indicative of less costly illness. There are many examples of the effectiveness of multiple 1 day stays in treating serious chronic illness.

Response: We discuss the issue of 1 day stays, and their exclusion from the final PIP-DCG groups in the March 1, 1999, Report to Congress. In particular, we describe our analysis of the payment impact of excluding 1 day stays. The majority of 1 day stays are for diagnoses already excluded on the grounds that they are "vague, minor, or transitory." The effect on payment of excluding 1 day stays is therefore small: a maximum payment impact of 0.7 percent under full risk adjustment, or 0.07 percent considering the first year blend percentage. The HER Report, which is appended to the Report to Congress, contains data on costs associated with 1 day stays. It is important to reiterate that these modifications do not mean that these expenditures have been excluded from the model. Rather, the payments associated with these diseases are captured in increased payments for the base payment category. We continue to believe that excluding 1 day stays is appropriate to prevent potential gaming of the risk adjustment system to obtain higher Medicare payments by admitting enrollees for 1 day stays.

Comment: One commenter objected to the inclusion of "discretionary diagnoses" in the PIP-DCG model, and recommended returning the costs for as many of these diagnoses as possible to the base payment category.

Response: The commenter is mistaken. We placed in the base payment category all "discretionary diagnoses," which we defined as vague, non-predictive, and/or marginal diagnoses, as well as diagnoses resulting from 1 day stays. As a result, only a subgroup of seriously ill beneficiaries is identified for increased payments. As described in our January 15, 1999, notice, the diagnoses to be excluded were determined by an outside panel of clinical experts.

Comment: Several commenters continued to recommend inclusion of an adjustment for institutional status in the risk adjustment methodology. These commenters contended that, in the absence of an adjustment, organizations would be underpaid for institutionalized beneficiaries and the system would provide incentives for hospitalization over more appropriate treatment in institutional settings.

Response: We have carefully considered inclusion of an adjustment for institutional status in the model. Because of the level of interest in this issue, we present a detailed analysis in the March 1, 1999, Report to Congress. Briefly, our analysis showed that the PIP-DCG model accurately predicts the average costs across the entire group of institutionalized beneficiaries. Our analysis also showed that mean actual Medicare payments for those in post-acute care facilities, such as SNFs, are far greater than those for long-term care facilities. However, since Medicare requires that a hospital stay precede a SNF stay, those costs are already reflected in the respective PIP-DCG groups. The model is designed to make adjustments that are correct on the average for groups of enrollees. We do not believe that it would be appropriate to introduce a distinction based on patterns of treatment among beneficiaries with the same inpatient diagnosis. While those in long term care facilities incur more cost than average Medicare beneficiaries, they incur less cost than predicted by the PIP-DCG model. An institutional factor for this population would therefore actually be negative: the PIP-DCG model is actually overpaying for persons in long term care facilities. The incentives for identifying the long term institutionalized and reporting on this group are low when the result is a payment reduction. We have therefore decided not to pay based on this site of service. There are relatively few enrollees in this group and the overpayments will be small. For these reasons, we have decided not to include an institutional status factor in the payment model.

Comment: Several commenters supported our decision to delay implementation of risk adjustment for several current long-term care demonstrations, and recommended extending this delay to other programs for the frail elderly, including capitated sub-contractors to M+C organizations.

Response: We believe that it is important to restrict the delay in implementing risk adjustment to the four demonstrations that we identified. Demonstrations generally have a special status in the Medicare program because they operate on the basis of waivers from normal payment rules. These four demonstrations in particular are well-established projects focusing on providing services to special populations. During the period prior to implementing risk adjustment for these projects, we will work with these demonstrations to collect data and to analyze whether inclusion of an adjustment for functional status in the risk adjustment system is feasible and appropriate.

Comment: While reiterating their support for the use of a "time-shifted" data model (in which data from the year ending 6 months before the payment year are used in the model to determine final risk scores), some commenters recommended that the model be recalibrated to take this "time shift" into account. Another commenter requested that HCFA enter discussions to identify other approaches, such as a "continuous update" model, that could mitigate the effect of using older data.

Response: In the September 8, 1998 Federal Register notice, we presented a clear choice between the use of lagged data with the current model, and use of more recent data with interim risk scores during the first part of the payment year. The overwhelming response of commenters was in favor of using the lagged data. We have conducted general analysis, and consulted with outside experts, on whether this approach would create any systematic bias, and we are confident that it does not. While some "high cost" individuals may see a delay in higher payments made on their behalf, others for whom payments should have decreased will receive higher payments for an extended period due to the lagging of data. We believe that these cases will balance out at the plan level. The American Academy of Actuaries, in its review of the risk adjustment methodology, notes that this schedule "will result in capitation rates that lag behind the theoretical prediction period." However, the Academy goes on to observe that this "lag between reporting and application, while lengthy, is significantly shorter than the lag in other systems currently used." (The American Academy of Actuaries review is appended to the March 1, 1999, Report to Congress.) We have discussed the "continuous update" alternative with representatives of the managed care industry. This model would require plans to provide data to HCFA on a continuous flow. It would also require HCFA to conduct frequent recalculation of risk scores. We therefore do not believe that this model is operationally feasible for plans or for HCFA. In addition, the notion of continuously changing payment rates for enrollees seems to contradict the clearly stated preference of commenters on the September 8, 1998 Federal Register notice, for knowing final enrollee rates at the start of the payment year, and their concern over stability of payments.

Comment: One commenter asked for clarification of the rescaling factor that is applied to the county rate books and for a description of the interaction of the various changes in payment rules made by the Balanced Budget Act of 1997 (e.g., regional and national blending, the 2 percent minimum increase, etc.) with risk adjustment of payments, including how and when a county moves from the floor amount to a blended rate or from a blended rate to the 2 percent update.

Response: We provide a detailed discussion of the rescaling factor, as well as a description of the interaction between risk adjustment and the rate changes made in the Balanced Budget Act, in the March 1, 1999, Report to Congress.

Comment: Several commenters presented extensive lists of data requests. For the most part, these data requests related to the computation of county rates and the rescaling factor, elements of the risk adjustment model and the calculation of risk scores (including the HER report), information on the encounter data submitted for the start-up period, and information specific to each Medicare+Choice organization.

Response: Much of the information requested on the rescaling factor and the computation of the county rates is included with this announcement. Other information is posted on the HCFA Web site, or will be sent separately. As described in the January 15, 1999, notice, the rescaling factor is the ratio of risk county rate to the demographic county rate. The tables provided with this notice include the demographic county rates and rescaling factors, from which risk county rates can be determined. County worksheet data are posted on the HCFA Web site (http://www.hcfa.gov/stats/homorates/aapccpg.htm). County demographic tables will be sent under separate cover. The computation of the restandardized, risk county rates is described in the January 15, 1999, notice. Basically, these rates are computed by replacing the average county demographic factors found in the AAPCC rate book with average county risk factors. The CY 2000 risk county rates are derived from restandardized 1997 rates, which in turn were based on 1997 average risk scores calculated from 1994, 1995, and 1996 data. We are considering whether it is possible to make available county information on which the 1997 average risk scores were based, given issues of privacy and data confidentiality, as well as the resources required to prepare these data for public release.

Additional information requested concerning the risk adjustment model and the calculation of risk scores is included in the the March 1, 1999, Report to Congress and the HER Report, which is appended to it. The HER Report in particular provides detailed information on all aspects of the design and operation of the risk adjustment model, including regression formulae, the number and types of diagnoses eliminated by the discretionary diagnosis filter, detailed information on data and file construction, and other information requested by the commenters.

We sent each section 1876 risk plan a letter on December 11, 1998, with information concerning the amount of encounter data submitted by month and other information to help plans determine whether the discharge information submitted is complete. We asked plans to review the information carefully and to bring significant discrepancies to our attention. We have discussed many specific concerns and problems concerning submission and processing of encounter data with plans, and we will continue to work to resolve particular problems. However, in consideration of workload and other priorities in implementing risk adjustment, we do not now think it will be possible or appropriate to provide much of the specific information requested by the commenter, such as detailed reports on the performance of fiscal intermediaries. We will, however, continue to consider these particular requests.

In conjunction with the release of this announcement, each Medicare+Choice organization will be sent a letter estimating the percentage difference between the organization's payment under risk adjustment and payment under the current system. The estimate will be based on the organization's enrollment in September, 1998. It will show the difference between the current demographically adjusted payment and the blended payment at 90 percent of the current demographic amount and 10 percent of the risk adjusted payment amount. The letter will also include the distribution of enrollees for an average month by PIP-DCG category and for other demographic factors (e.g., age, gender, Medicaid status, previously disabled, and working aged), the distribution of PIP-DCG scores for that organization, and the counts of encounters that were used to determine the estimated payment. Organizations will be instructed to use these estimates as they prepare their adjusted community rate proposals for the year 2000. Beginning January, 2000, on a monthly basis, each organization will be provided with information on each enrollee, including the county of residence, age, gender, Medicaid status and previously disabled status, PIP-DCG score, and payment amount.

We will consider providing other data requested by the commenters as resources, time, data confidentiality, and other priorities permit.

Comment: One commenter noted that the 45-day notice in past years had included tables showing the forecast of national per capita growth rates by major line item, e.g., inpatient hospital, SNF, etc. The commenter recognized that these tables were not included in this year's notice due to constraints on the release of details concerning the President's budget, and requested that these tables be released in the future as soon as the budget is released. The commenter also requested that in the future forecasts for 3 years be presented.

Response: The tables to which the commenter referred are included in this announcement. We will consider the commenter's recommendations for future notices.

Comment: One commenter raised the possibility of computer failure anywhere in the process (transfer of data, processing of data, etc.), and asked what will happen if the Y2K computer problem causes disruption or delay in the transfer or processing of data required for risk adjustment.

Response: HCFA has established the necessary mechanisms to implement risk adjusted payments in 2000. Over the past 15 months, we have gained experience in obtaining encounter data for risk adjustment. Since December 1998, we have furthered our experience in implementing the payment methodology change by determining the risk adjustment factor for all Medicare+Choice enrollees and identifying and calculating changes to the county based rate book necessary to implement risk adjustment. We have also initiated changes to our payment system that are required in order to make payments to M+C plans in January, 2000. Finally, we have identified a Y2K contingency plan for M+C payments. These plans ensure that HCFA will be able to implement risk adjusted payments in 2000.

Comment: One commenter noted that the PIP-DCG model is based upon fee-for-service data that do not accurately reflect the experience and health status of managed care enrollees. In order to make the risk adjustment model moreaccurate, HCFA should recalibrate the factors using data from managed care plans rather than fee-for-service data. The PIP-DCG model uses 1995 data which are not appropriate for application to 2000 payments. In particular, the 1995 data reflect discretionary admissions which are no longer the norm.

Response: We have had no option but to employ fee-for-service data in order to meet the statutory deadline for implementing risk adjustment. Under any model, of course, there would necessarily be a lag between the data used for calibrating the model and the payment year, so that some change in practice patterns will always be possible during the lag period. Finally, while 1995 data were used to calibrate the PIP-DCG model, the 1996 predicted payments that result are used only in the form of a relative index.

Comment: The PIP-DCG methodology could impose penalties on managed care companies that appropriately provide care in outpatient settings.

Response: The PIP-DCG model represents a substantial improvement over the current system. Since we do not yet have full encounter data from Medicare+Choice organizations, we do not know the extent to which plans are treating chronically ill patients in an outpatient setting. Evidence from the Medicare Current Beneficiary Survey (MCBS) suggests that even if managed care provides outpatient care as a substitute for inpatient care in some cases, chronically ill beneficiaries disproportionately require inpatient care for their chronic and/or other conditions. The PIP-DCG model increases payments to the organization for many of these conditions. Nevertheless, we agree that a comprehensive model, which would include encounter data from outpatient settings, is preferable, and we plan to move toward implementing such a model as expeditiously as possible. However, implementation of the comprehensive risk adjustment model is not operationally feasible for 3 to 4 years, because of data constraints on both plans and HCFA.

Comment: One commenter contended that the refinements to the PIP-DCG groupings tend to understate the severity of illness. They argue that the sorting algorithm that excludes a diagnostic grouping with fewer than 50 beneficiaries assigned to it allows the potential that beneficiaries with rare and costly illnesses may be assigned to the base group. Another commenter questioned the sample size limitation of 1,000 beneficiaries for the PIP-DCG groups, noting that while a minimum threshold stabilizes payments in the model, admissions with very high costs may thereby be excluded.

Response: Each original PIP-DCG group retained its identity in the final payment model only if it contained at least 1,000 beneficiaries in the original sample; this minimum sample size was defined to assure stability of estimated payments in the model. If sample sizes were smaller than 1,000, the potential PIP-DCG was expanded to include PIP-DxGroups with average expenditures in the next lower range until the sample size was satisfied. However, it is incorrect to conclude, as the commenter seems to, that these costs were excluded from the model and assigned to the base payment group. If at any time during the sorting algorithm a PIP-DxGroup had fewer than 50 beneficiaries assigned to it, it and the associated costs were assigned to the base payment category. This is necessary because smaller groupings cannot provide a statistically valid basis for predicting costs.

Comment: One commenter stated that, while the working aged adjustment provided in the PIP-DCG model is logically sound, there have been problems with out-of-date and incomplete information on working aged status. A longer phase-in would be appropriate to provide time to accurately determine which beneficiaries belong in this category. Another commenter requested clarification regarding whether the working aged adjustment will be based on the enrollee's prior year status or payment year status.

Response: We recognize that there have been problems concerning out-of-date data for working aged beneficiaries under the current payment system, and we will continue to work on these problems as we implement risk adjustment. The adjustment in the risk adjustment system is based on current year status. There is detailed discussion of the working aged adjustment in the March 1, 1999, Report to Congress and the appended HER Report.

Comment: One commented that the recognition of Medicaid recipients has also been a challenge in the Medicare managed care program. Alternatively, commenters recommended a longer phase-in to provide time to improve the tracking of Medicaid status or more frequent updates to Medicaid status to account for changes.

Response: Again, we recognize that obtaining timely and accurate data on Medicaid status has been a challenge under the current payment system. We have been working on these problems, and will continue to do so as risk adjustment is implemented. Given these data issues, we believe that the approach for Medicaid status under the risk adjustment system will be an improvement over the current month-to-month concurrent adjustment. There is detailed discussion of the Medicaid adjustment in the March 1, 1999, Report to Congress and the appended HER Report.

Comment: Several commenters raised issues concerning the time allowed for submission of data and the proposed reconciliation period to account for late encounter data. One commenter stated that the deadline for submission of encounter data allows insufficient time, and expressed concern that data for beneficiaries moving from fee-for-service to Medicare+Choice organizations may be incomplete because of the time frame for submitting fee-for-service claims. Another commenter expressed concern that the proposed reconciliation may become a labor-intensive and difficult effort. More information on the reconciliation process is needed before the implementation of the PIP-DCG system, especially in the light of the limits on systems resources due to the Year 2000 issue.

Response: The deadline for submission of the encounter data (for the period July 1, 1998 through June 30, 1999) that will be used to compute risk scores for the first payment year is September 10, 1999. We are unable to extend this deadline. However, we do intend to institute a reconciliation process that will take into account late data submissions. Plans should attempt to have all data in by the deadline of September 10, 1999. However, if plans receive UB-92s from hospitals after this date, they may submit the encounter to their fiscal intermediary and the data will be processed. Plans should note that a deadline for submission of all data from a payment year will be established: this deadline will probably be June 30, 2000 for the period of July 1, 1998 to June 30, 1999. After that date, the fiscal intermediary will no longer accept these data. After the payment year is completed, HCFA will recalculate risk factors for individuals who have late encounters submitted. Then, we will determine any payment adjustments that are required. This reconciliation will be undertaken after the close of a payment year and will be a one-time only reconciliation for each payment year. Additional information on the reconciliationapproach will be provided to plans over the next several months. However, we anticipate that the major burden of this reconciliation will be on HCFA to recalculate risk scores on the basis of the late encounter data, rather than on the plans who will merely have the opportunity to continue submitting data using established mechanisms after the intitial deadline for submission of data to calculate the CY 2000 risk scores.

Comment: One commenter requested a detailed draft time line for implementation of the comprehensive risk adjustment model that will be used beginning January 1, 2004, and requested discussion with HCFA to identify issues related to the operations and capabilities of Medicare+Choice organizations that will be relevant to shaping an implementation strategy.

Response: We present information on our preliminary plans for implementing comprehensive risk adjustment in the March 1, 1999, Report to Congress. We will continue to apprise Medicare+Choice organizations in a timely fashion as our implementation plans are finalized.