Volume responses to exogenous changes in Medicare's payment policies.
Costs for physicians' services under Medicare continue to increase rapidly, despite imposition of various constraints on Medicare's payment rates. Although the rate of growth in costs slows somewhat during periods of fee constraint, realized savings are typically smaller than the reduction of projected fee increases would simply. This occurs because the volume of services provided during periods of fee constraint tends to accelerate in growth, eliminating part of the savings that would otherwise accrue to Medicare. This is called the "volume offset" in this study, which presents estimates of this response.
The study examines not only thee savings-reducing offset that appears to occur during periods of fee constraint, but also the cost-reducing offset that might occur in the face of increases in real fees. Both responses may be relevant when a new physician fee schedule is implemented in 1992 under Medicare, because some physicians will see their payments from Medicare increase substantially even though the overall impact of the fee schedule is to be budget-neutral. In addition, it is necessary to consider the effects of limits on billed amounts on volume responses. Under current law increasingly stringent limits are being set on billed amounts for "nonparticipating" physicians -- that is, for physicians who retain the right to bill patients directly and to collect billed amounts rather than Medicare's allowed amounts.
The first section of this study explains the factors that might generate a volume response to changes in real fees. The second section describes the data used here to estimate behavioral responses to payment changes. The third section presents the regression estimates and translates them into an estimated volume offset. The fourth and final section uses the estimated volume offset in payment simulations to illustrate how the conversion factor for a budget-neutral fee schedule might be adjusted to account for behavioral responses.
Factors Contributing to the Volume Response
During periods of fee constraint, a volume response (observed as a temporary acceleration in growth of the number of services per enrollee or in service complexity) could arise from either of two sources:
* Greater demand for care patients in response to lower out-of-pocket
costs relative to what they would otherwise face; or
* Physician-induced demand resulting from physicians' efforts
to offset, at least partially, the fall in their practice income
compared with income that would otherwise result.
Both of these effects could be at work simultaneously. For purposes of cost estimation, the particular effect responsible for the volume response, or offset, makes no difference. The implications for reduced savings from fee constraints are the same. There might be analogous responses to increases in real fees, in that the costs of a payment rate increase might be partly offset by a decrease in volume (observed as a temporary slowing in the normal rate of growth).
This study examines the volume offset both to cuts and to increases in real fees under Medicare, allowing as well for the effects of any limits imposed on physicians' billed amounts. It assumes that the volume offset is a response to the initial change in practice receipts that would result -- before behavioral responses -- from a change in the effective rates faced by physicians (where effective rates are defined as payment rates on assigned claims, and actual charges on unassigned claims).(1)
Methods and Data
The key behavioral response estimated is the elasticity of volume with respect to receipts per service for Medicare-covered services [E(vol:rec)].(2) Medicare receipts are a function of allowed amounts on assigned claims, billed amounts on unassigned claims, and the assignment rate -- which might itself change in response to changes in Medicare's payment rates. To test for the latter response during the period studied, an estimate of the elasticity of assignment with respect to Medicare's allowed amounts [E(a:aa)] is also obtained.(3)
The data used for the estimates presented here are physician-level records containing information from Medicare claims filed in Colorado for 1976 and 1978, for services provided by general practitioners and internists.(4) Other specialty groups were excluded because they were largely unaffected by the payment rate changes that were implemented in Colorado during this period.(5)
In early 1977, Colorado combined ten payment localities into a statewide locality, substantially changing the structure of Medicare's prevailing charges in each of the original localities.(6) Consequently, physicians' receipts per service for Medicare-covered services also changed, with receipts increasing for some practices and falling for others in each of the ten original localities?
This natural experiment permits empirical assessment of physicians' responses to exogenous changes in Medicare's payment provisions without the appearance of potential problems inherent in single-year cross-section analyses, in which differences among physicians not taken into account statistically could distort the estimates. Because the change in payment rates took place at the beginning of 1977, any behavioral changes were observed after a full year's experience with the new payments. As a result, it may be reasonable to view the observed behavioral responses as equilibrium values.
One shortcoming of the data is that there is no information about physicians' receipts from non-Medicare patients. In some instances, changes in the non-Medicare sector could affect physician's practice patterns for both non-Medicare and Medicare patients. If so, failure to control for those changes in the non-Medicare sector could result in distorted estimates of the effects of the variables that are included in the analysis on Medicare. This is a common problem in analyses of this sort, and one that cannot be remedied without obtaining the cooperation of all insurers in the region selected for study.
Another problem with the data is that the period of time covered for the 19767 fiscal year is longer than the time covered for 1978, so that the measure of volume used as a dependent variable in the regression analysis is measured with error. As discussed in the Appendix, the estimated intercept in the volume equation is biased downward as a result of this measurement error, but the estimated slope coefficients -- from which the key elasticity estimates are derived -- appear to be unbiased.
The Colorado data used here were initially analyzed by Thomas Rice (1983, 1984) and Rice and McCall (1982). Reanalysis was necessary in order to obtain estimates in a form useful for estimating the costs of specific payment changes.
Two equations are estimates and reported here: one for the change in physicians' assignment rates in response to a change in Medicare's payment rates [allowed amounts per relative value unit (RVU)]; and one for the change in the volume of services provided in response to a change in physicians' effective rates (receipts per RVU).(8) Volume is measured by total RVUs per patient for each physician.(9) The observed changes in both payment rates and effective rates are deflated by the change in the Medicare Economic Index (MEI) between 1976 and 1978, in order to adjust for changes in physicians' costs over the period.(10) The change between 1976 and 1978 is used in the specification of the regression equations as a way to control for unobserved differences among the physicians in the sample that might otherwise distort the estimates.
In addition to the payment measures that are the explanatory variables of primary interest, each estimated equation includes a number of other variables to control for factors other than reimbursement that could have affected physicians' behavior over the period studied. These are measures of demand for the physicians' services including experience, board certification, specialty, sex, medical school, and physician density in the locality. They also include measures of direct practice costs, for example, whether or not the physician is part of a group, wage costs in the locality, and the extent of local urbanization. Definitions for all variables used are shown in Table 1.
Table 1: Definition of Variables Used in Regressions Variables Definition Dependent Variables Assignment Change in ratio of assigned relative value units (RVUs) t o total RVUs between 1976 and 1978 Volume Change in ratio of total RVUs to total patient count betw een 1976 and 1978 Independent Variables Payment rate Change in ratio of total allowed amounts to total RVUs be tween 1976 and 1978, adjusted by Medicare Economic Index (MEI) Effective rate Change in ratio of total receipts to total RVUs between 1 976 and 1978, where receipts are sum of allowed amounts on assigned claimed and billed amounts on unassigned claims, adjusted by MEI and holding 1976 assignment rate constant Experience Number of years between 1977 and the year the physician graduated from medical school Board certified Dummy variable indicating whether the physician has one o r more specialty certifications Female Dummy variable indicating whether the physician is female Foreign Dummy variable indicating whether the physician graduated from a medical school outside the United States or Canada Osteopath Dummy variable indicating whether the physician is an osteopath Group practice Dummy variable indicating whether the physician is in a g roup Large metropolitan Dummy variable indicating that the physician practices in a statistical area metropolitan area of more than 1 million (MSA) Small MSA Dummy variable indicating that the physician practices in or adjacent to a metropolitan area of less than 1 million Non-MSA Control group for urbanization, including physicians prac ticing in rural or semirural areas Wages Change in average wage for health sector workers in the a rea between 1976 and 1978 Physician density Change in number of nonfederal physicians per 1,000 population in the area between 1975 and 1977
Separate equations are estimated for general practitioners, for internists, and for the two groups combined. The data are weighted by each practice's total allowed amounts for 1976, so that the estimated behavioral responses will appropriately reflect the impact on Medicare's costs.
The dependent variable in the assignment equation is the change between 1976 and 1978 in the proportion of services that were assigned. Services are measured as RVUs to make the measure of quantity independent of payment rates.(11) The independent variable of most interest is the change in real allowed amounts per RVU (the payment rate).
The results indicate the changes in payment rates had some effect on assignment rates in Colorado during this period, with assignment rates increasing in response to higher payment rates and falling in response to lower payment rates (Table 2). The response is larger for internists than it is for general practitioners, and it is statistically significant only for internists.(12) Hence, this endogenous element in the definition of the effective rate used for the volume equation estimated in the next section must be eliminated to avoid estimation bias. [Tabular Data 2 Omitted]
One should note that the estimated equations explain very little of the observed variation in assignment rates in the sample. Because of the 1984 implementation of Medicare's participating physician program, under which physicians were encouraged to accept assignment on all claims, it is likely that assignment rates now are even less predictably related to payment rates than they were during the period studied. The evidence for this assertion is that in the year before implementation of the participating program, Medicare assignment rates were 53 percent; in the first full year subsequent to implementation, rates were nearly 70 percent. Assignment rates continued to climb thereafter, and currently are more than 80 percent. This dramatic increase in assignment rates has occurred despite a growing disparity between physicians' billed amounts and Medicare's payment rates. For this season, no behavioral responses for assignment rates are included in the payment simulations discussed in the final section of this article
The regression equation for volume uses as the dependent variable the change in total RVUs per Medicare patient seen by a given physician over the period; the change in real Medicare receipts per RVU (the effective rate) is the independent variable of primary interest. Rather than using reported receipts for 1978 (which would reflect any payment-induced changes in assignment), a value was calculated to estimate receipts in 1978 if assignment rates had been unchanged from 1976. Hence, the equation provides an estimate of the volume response to the change in real receipts per RVU that would have occurred due to Medicare's payment rate change had no endogenous change in assignment behavior taken place.
When responses for gainers and losers are constrained to be the same (a symmetric response), the estimated volume response is negative and significant, with a value of approximately -0.5 for both general practitioners and internists (Table 3). The estimate responses for the two specialties were not significantly different, and an estimate for the two combined was obtained. The results indicate that changes in volume would offset about half of the initial change in practice receipts that would result from changes in Medicare's payment policies. [Tabular Data 3 Omitted]
However, the volume response might differ depending on whether receipts increased or decreased as the initial result of policy changes. When different responses are permitted for gainers and losers (an asymmetric response), the size (in absolute value) of the response is larger for losers than for gainers (Table 4). For both specialties combined, the volume elasticity is -0.375 for gainers, and -0.555 for losers. This means that decreases in volume would offset a little over a third of the initial gain in receipts for gainers, while increases in volume would offset a little over half of the initial loss in receipts for losers. [Tabular Data 4 Omitted]
Imposing symmetry on the responses of gainers and losers is a strong constraint. Because the asymmetric estimates do not impose this constraint, they are probably better estimates of the volume response even though the estimated differences are not statistically significant. In the simulation results discussed in the next section, both asymmetric and symmetric estimates are used for comparison.
Adjustments to the Fee Schedule Conversion Factor
One of the issues to be resolved when a Medicare fee schedule is put in place in 1992 is how to calculate the conversion factor that will transform the relative value scale developed by the Health Care Financing Administration into a set of payment rates in a way that will achieve budget neutrality. As a first approximation, Medicare claims data may be used to estimate the conversion factor that would achieve the desired effect on spending in the absence of behavioral responses. For example, a first-order budget-neutral conversion factor could be calculated that would equate total Medicare payments under the new fee schedule (assuming no behavioral changes) to projected payments under the old payment system. This factor might result in higher (or lower) costs once the expected volume responses occurred, however. A second-order budget-neutral conversion factor would differ from the first-order factor by just enough to offset the effects of the expected volume responses.
This section provides estimates of the adjustment that would have to be made to any given first-order conversion factor in order to achieve budget neutrality once behavioral responses had occurred. The estimates are made from the fee schedule recommended by the Physician Payment Review Commission (PPRC) in their April 1989 annual report, using a microsimulation model developed from the 1986 Part B Medicare Annual Data Files and aged to 1990.(13) Estimates (which are obtained through iterative adjustments to the first-order conversion factor) are presented for two variants of that fee schedule -- one with no limits on billed amounts, and one billed amounts limited to no more than 115 percent of the fee schedule amounts (as specified under current law for 1993 and subsequent years).
Illustrative adjustment values are shown under alternative assumptions (Table 5). First, the asymmetric elasticity estimates obtained in this study are used. If no ceiling on billed amounts is imposed, the adjustment to the first-order conversion factor necessary to offset behavioral responses is very small, at .998.(14) Failure to make this adjustment would have increased costs by about $50 million if the fee schedule had been fully implemented for 1990.(15) [Tabular Data 5 Omitted]
If billed amounts were capped at 115 percent of the fee schedule amounts, behavioral responses to implementation of the new fee schedule would be larger. This is because practice receipts would fall for some unassigned claims due to the cap. (By contrast, receipts on unassigned claims would be unchanged if there were no cap.) In this case, the necessary adjustment would be .941. If the adjustment had no been made, the fee schedule would have increased costs by about $1.5 billion for 1990, rather than being budget-neutral.
Some believe, however, that no offsetting volume response would occur for practices that would gain under a payment change, although there would be a large response among losers. If the volume elasticity was set to 0 for gainers in the fee schedule simulation to reflect this belief, the necessary adjustment to the first-order conversion factor would be increased. With no cap on billed amounts the adjustment would be .964; with a 115 percent cap, the adjustment would be .921.
Symmetric volume elasticity estimates require the smallest adjustment to the first-order conversion factor. In fact, with no cap on billed amounts, the first-order conversion factor would actually have to be adjusted upward to maintain budget neutrality. Only one instance -- if the volume offset was symmetric and if all claims were assigned -- would not adjustment to the first-order conversion factor be necessarily to maintain budget neutrality subsequent to behavioral responses. In this instance, changes in receipts would equal changes in allowed amounts for each practice. Because of the symmetric offset, volume-increasing responses by losers would be just sufficient to balance volume-reducing responses by gainers.
One problem with the Rice data extract in that the period of time covered for the 1976 fiscal year is longer than the time covered for 1978 -- 456 days in 1976 compared to 313 days in 1978. This difference is unimportant for the assignment and reimbursement variables used in the study, because both numerator and denominator are affected by the length-of-year difference in the same way, with the result that the variable values are unaffected. For example, the Medicare payment rate variable is the sum of allowed amounts over the period, divided by RVUs provided over the period. Allowed amounts cumulate at the same rate as RVUs during a period with fixed payment rates. The assignment rate variable is the sum of assigned RVUs over total RVUs, and both of these cumulate at the same rate.
The difference in length of fiscal years matters for the measure of volume, however, introducing a measurement error in the dependent variable. Volume is defined as the number of RVUs provided during the year, divided by the number of patients seen during the year. The number of different patients seen by a physician over a given period of time is probably higher for longer periods of time, as new patients are accepted and established patients either die or move away, but it is unlikely that patient count increases exactly in proportion to the count of RVUs provided by the physician over the period. Because no adjustment was made to the volume measure for the estimates presented in the text, the implicit assumption was that patient count would cumulate at the same rate as RVUs. Consequently, the expected value of the measurement error for the dependent variable is negative, biasing downward the estimated intercept in the volume equation.
The only adjustment that can be made -- without retrieving the original claims data from Colorado and selecting a subset of the 1976 records by date of service -- is to adjust the 1978 RVU measure by 1.46 ( = 457/313). In this case, the implicit (and unrealistic) assumption would be that patient count would not increase at all over an additional 134 days in the fiscal year. This measure probably also has an error component, however, with a positive expected value. Estimates were made using this adjusted measure of volume, for comparison with those made using no adjustment. As expected, the primary effect of the adjustment was to increase the estimated intercept in the volume equation (changing it from negative to positive), with much smaller effects on the estimated slope coefficients (indicating lack of correlation between the measurement error and the independent variables). Because the volume offset depends only on the slope coefficient, it is not much different from the offset based on the unadjusted data.
Contributions made by others are much appreciated. Thomas Rice generously provided a tape and documentation for the data used here (data used earlier for his own work). Paul Ginsburg and Stephen Long reviewed earlier drafts and made useful suggestions for revision. Susan Hilton provided programming support.
(1)Responses in volume would probably occur because of changes in total practice receipts from all payers, not only from Medicare payers. Because the data used for simulation results contain only Medicare information, however, an estimate of behavioral responses to changes in Medicare receipts is necessary. (2)Technically, E(vol:rec) is the percent change in the volume of services in response to a 1 percent in receipts per service. (3)E(a:aa) is the percent change in physicians' assignment rates in response to a 1 percent change in Medicare's allowed amounts per service. (4)The Medicare physician fee freeze period (from 1984 through 1986) might appear to provide more recent evidence that could be used to estimate behavioral responses. However, during that same period Medicare implemented a new prospective payment system for hospitals and a new peer review oversight system, with significant effects on physicians' practice patterns. It is doubtful that the effects of the fee freeze can be successfully isolated from that experience. Even if isolation were possible, the period could provide estimates only for losers, not for gainers. (5)With the exception of general surgeons, little or no change in prevailing charges took place for the excluded specialty groups, because prevailing charges were already calculated statewide for most of their services. This was because the excluded specialists were too few in number to permit separate charge profiles to be defined for their services in each of the original ten localities. General surgeons, however, were sufficiently affected by the payment rate changes to be included in the original data set, but estimates of their volume responses to the payment changes were not credible. Whether or not surgical practices saw payments rise or fall during the sample period, volume fell. This indicates that something else -- important but uncontrolled in the regression equation -- happened along with the payment rate changes. Representatives for the Medicare carrier in Colorado can recollect no changes in the Medicare sector that would have affected use of surgical services, and they were unable to comment on possible changes in the non-Medicare sector that might have made Medicare patients relatively less attractive to surgeons. (6)The prevailing charge was one of three values used to determine Medicare's allowed amount for a given service. The allowed amount was the lowest of (1) the physician's billed amount; (2) the physician's customary charge for that service (defined as his median billed amount during the preceding year); or (3) the prevailing charge in the locality (defined as the 75th percentile of all physicians' customary charges, or a lower amount determined by the Medicare Economic Index). (7)The coefficient of variation (CV) on the primary independent variable used in regressions for this study is very large, over all regions combined and for each of the ten original pay localities. The CV -- a widely used measure of variation -- is defined as the ratio of the standard deviation over the mean. (8)Relative value units (RVUs) are used to indicate the complexity or intensity of any given service, with a higher number of units associated with more complex services. (9)A value for total RVUs was obtained by adding together the RVUs for each of the four service groups, after converting their separate RVU scales to a common scale based on payment rates. In particular, conversion factors were calculated from the 1978 data (at which time Medicare's rates were uniform throughout Colorado) by dividing the statewide sum of allowed amounts by the statewide sum of RVUs, separately for each of the four service categories. The implicit conversion factors are 2.34 for medical services; 5.22 for surgical services; 5.06 for laboratory services; and 6.05 for radiology services. (10)The Medicare Economic Index is an index of physician's practice costs developed and updated by the Health Care Financing Administration. (11)Ordinary least--squares estimation was used. Changes in Medicare's payment rates were largely exogenous because of the nature of the change in payment methodology implemented by Colorado in 1977. (12)No significant difference in assignment responses occurred between practices that gained or lost allowed amounts. (13)For a description of this model, see Appendix A in Congressional Budget Office (1990). (14)A value of 1 would indicate that no adjustment to the first-order factor was needed. (15)Projected Medicare costs for physicians' services for 1990 are about $26 billion.
Congressional Budget Office. Physician Payment Reform under Medicare. Washington, DC: U.S. Congress, April 1990. Rice, T. "The Impact of Changing Medicare Reimbursement Rates on Physician-induced Demand." Medical Care 21, no. 8 (August 1983): 803-15. --."Physician-induced Demand for Medical Care: New Evidence from the Medicare Program." In Advances in Health Economics and Health Services Research. Vol. 5. Greenwich, CT: JAI Press, 1984. Rice, T., and N. McCall. "Changes in Medicare Reimbursement in Colorado: Impact on Physicians' Economic Behavior." Health Care Financing Review 3, no. 4 (June 1982): 67-85. Sandra Christensen, Ph.D., Principal Analyst/Economist, Congressional Budget Office, Ford H.O.B., U.S. Congress, Washington, DC 20515. This article, submitted to Health Services Research on October 17, 1989, was revised and accepted for publication on March 27, 1991.
|Printer friendly Cite/link Email Feedback|
|Publication:||Health Services Research|
|Date:||Apr 1, 1992|
|Previous Article:||An empirical comparison between the board's strategic role in nonprofit hospitals and in for-profit industrial firms.|
|Next Article:||Identification of factors associated with hospital readamission and development of a predictive model.|