Exploration of the Impact of Preferred Drug Lists on Hospital and Physician Visits and the Costs to Medicaid

January 15, 2005
Matthew M. Murawski, RPh, PhD

,
Tamer Abdelgawad, MA

Volume 11, Issue 1 SP

Objective: To conduct an exploratory investigation of the possible effects of the implementation of a state Medicaid preferred drug list (PDL) on the average number of visits by Medicaid patients to hospitals and physicians, and to provide preliminary estimates of the Medicaid reimbursement costs of these additional visits.

Study Design: A regression-based, difference-in-differences retrospective analysis using anonymized patient-level data on cardiovascular-related inpatient and outpatient hospital visits and procedures, and physician visits and procedures.

Methods: The impact of the implementation of a state Medicaid PDL on a test group of Medicaid cardiovascular patients was examined. A contemporaneous group of non-Medicaid cardiovascular patients from the same state were used as controls.

Results: There was a statistically significant increase in the number of outpatient hospital visits and physician visits for the test group compared with the control group in the first 6 months after PDL implementation. There was a positive but statistically insignificant increase in the number of inpatient hospital visits. All increases in visits for the test group compared with the control group in the second 6 months after PDL implementation were positive but statistically insignificant. As a result, estimated average Medicaid reimbursement costs for cardiovascular patients in the state increased during that year.

Conclusion: The observed range of increases in hospital and physician visits is evidence for the possible existence of an unintended consequence of PDL implementation by state Medicaid programs. Precautionary research in this area is clearly called for.

(Am J Manag Care. 2005;11:SP35-SP42)

State Medicaid programs, facing substantial budgetary shortfalls, are aggressively implementing various cost-containment measures. One popular approach is the preferred drug list (PDL), which functions by encouraging the use of less expensive drugs considered equal in efficacy to more expensive alternatives in any given therapeutic class, and by soliciting additional manufacturer rebates for products included on the PDL beyond those rebates mandated by federal law. A prior-authorization mechanism is included to permit off-list prescribing, but in practice, the more onerous prior-authorization requirements become, the more the PDL resembles a closed formulary.

Adoption of PDL programs is widespread, with 29 states having instituted some form of PDL by 2003.1 Although these programs have been adopted with the expectation that they will result in significant reductions in drug acquisition costs, there is some evidence that similar restrictive-access mechanisms have been associated with increased utilization (and costs) in other components of the Medicaid budget.2-6 This cost-shifting may largely offset any achieved prescription savings, or even result in a net increase in costs. Because of the high rate of PDL adoption, the overall impact of these possible unintended effects on Medicaid costs could prove substantial. Over time, several models have been offered to explain potential increases in nonprescription healthcare utilization and costs associated with restrictive-access mechanisms. The literature suggests that substitution of less expensive generics in some therapeutic areas could lead to problems.7-9 Forced shifts from newer, more effective products to older-brand medications have been reported to result in higher rates of side effects and adverse reactions.10,11 Also, the use of front-line medications has been shown to result in significant cost avoidance.12,13 These explanations, which parallel early, anecdotal reports evaluating PDL programs,14 suggest at least 3 plausible mechanisms through which PDL-induced cost-shifting could occur: "administrative" physician visits to make or clarify the switch to PDL-approved drugs; additional physician (and sometimes hospital) visits to deal with side effects and adverse reactions to new therapies; and physician or hospital visits to address outright treatment failure.

A recent literature survey by Soumerai concluded, in part, that additional research is required to address the question: "What are the economic and clinical consequences of ...policies that promote use of lower-cost preferred drugs for major medical illnesses (such as cardiovascular disease, chronic mental illnesses and neurologic disorders), especially in low-income Medicaid populations?"15 (p142) To our knowledge, extensive research has not yet been done, possibly because of the formidable methodologic challenges involved in investigating this question. In this paper, we attempt to provide a preliminary quantitative exploration of this question by isolating the impact of PDL implementation on inpatient and outpatient hospital services and physician visits by cardiovascular patients. We also provide exploratory estimates of the costs to Medicaid of these PDL-related visits.

DATA AND METHODS

Data Description

We obtained data for prescriptions, hospital visits, and physician visits from Verispan, LLC (Yardley, Pa). The prescription dataset covers anonymous patient-level cardiovascular prescriptions, and includes the patient's age group and sex, zip code of the pharmacy where the prescription was filled, the prescriber's medical specialty, and payer information for each prescription (Medicaid, cash, or third party). The data come from electronic pharmacy records and benefit claims, and cover 50% to 55% of all prescriptions filled in the United States during the study's time frame. We used the pharmacy zip code in conjunction with zip-3 level US census 2000 data to assign an urban/rural designation to each patient as a demographic control.

International

Classification of Diseases, Ninth Revision,

Clinical Modification [ICD-9-CM]

The hospital and physician datasets cover physician and hospital visits where the primary diagnosis was hypertensive or heart disease (the following diagnosis code subheadings were covered: 394-398, 402, 404, 410-416, and 420-429). We will refer to these as cardiovascular-related visits. In addition to the diagnosis codes, the data also contain all the codes for procedures performed during the visit. The data come from electronic billing systems installed in hospitals and physicians' offices, and cover about 20% and 10%, respectively, of all US hospital and physician transactions. Table 1 contains a listing of all variables used in this study and their definitions.

The state under study implemented a Medicaid PDL in June 2002. The 3-month period immediately after the PDL was implemented was the index period (June 2002 to August 2002), and we used it for initial sample classification (described below). The prescription data span a 2-year period from June 2001 to May 2003, while the data for hospital and physician visits span a shorter, 18-month period from December 2001 to May 2003. We used the data for the 6-month period before the PDL implementation (December 2001 to May 2002) to establish baselines for our variables of interest. Then we examined changes in these variables in the two 6-month periods after June 2002.

Sample Selection and Classification

Although none of the datasets covers 100% of its target population, we are not aware of any sample selection problems or systematic omissions that would bias the direction of our results. In addition, since we conducted a difference-in-differences analysis, we needed only to make the weaker assumption that even if bias or systematic omissions were present, they affected the test and control patient groups in the same way.

patient_ ID

Each patient in the data was uniquely identified by an anonymous variable that linked all of the patient's records over time and across datasets. The initial patient sample was selected from the prescription data based on a pre-PDL history of antihypertensive prescriptions, 1 or more prescriptions filled after May 2002 in a drug class restricted by the PDL (angiotensin-converting enzyme inhibitor, angiotensin II receptor blocker, and calcium channel blocker classes were restricted), and at least 1 antihypertensive prescription filled during the index period. These criteria were intended to select patients who potentially experienced the PDL process at its inception (the index period), and whose cardiovascular-prescription claims history implied they would be most affected should the state's PDL restrictions be applied to them. We will refer to these patients as CV-PDL patients.

A CV-PDL patient was classified as a Medicaid patient if Medicaid paid for 75% or more of his/her antihypertensive prescriptions in the 1-year post-PDL period, and classified as a non-Medicaid patient otherwise. A total of 3250 CV-PDL patients satisfied the criterion for classification as Medicaid patients and served as the basis for the selection of our (retrospective) experimental test group. Of the much larger number of non- Medicaid CV-PDL patients, we randomly selected 3788 patients (1.5%) to serve as a basis for the selection of our control group.

To select our hospital test and control samples, CVPDL patients were matched against the hospital dataset to assign inpatient and outpatient hospital visits, diagnoses, and procedure data to each patient. During the period from December 2001 to May 2003, 935 of the 3250 Medicaid CV-PDL patients appeared in the hospital dataset with 1 or more hospital visits, but only 269 had 1 or more hospital visits that were actually cardiovascular related. We took the 935 Medicaid CV-PDL patients as our test group, and those without cardiovascular-related visits were assumed to have zero cardiovascular-related visits in our quantitative analysis. As a control, we selected the 1114 out of 3788 non-Medicaid CV-PDL patients who appeared in the hospital dataset with 1 or more hospital visits (only 274 had cardiovascular-related hospital visits, the rest were assumed to have zero cardiovascular-related visits).

Selection of the physician sample was done similarly. Here, 1421 Medicaid CV-PDL patients (364 with cardiovascular-related visits) and 1668 non-Medicaid CV-PDL patients (457 with cardiovascular-related visits) had physician visits during our study's time horizon, and were selected as test and control groups, respectively.

Model

delta1_visits,

delta2_visits, delta1_cost, and delta2_cost

medicaid

visits

i

Our main goal was to examine the change in the number and overall cost of hospital and physician visits per patient when the PDL was implemented. Specifically, we estimated the differences in visits and cost changes between Medicaid and non-Medicaid CV-PDL patients by regressing the change variables ( ) on the dummy, while using the demographic and behavioral variables available to us as controls. The following is a typical regression equation from our model for patient :

medicaid

In this setting, the estimate of ß1 (the dummy coefficient) represents the portion of the change variable that is solely attributable to being on Medicaid. Since a PDL directly affects only Medicaid patients, it is reasonable to interpret this coefficient as the net direct consequence of PDL implementation. The model does not capture any indirect effects of a PDL, such as possible spillovers to non-Medicaid patients resulting from induced changes in physician prescribing behavior.

RESULTS

Visits

Table 2 presents means and standard deviations for the visits data of our test and control groups by time period and type (hospital inpatient, hospital outpatient, hospital total, and physician). A brief examination of the table shows some of the trends that form the basic finding of our work: namely, that the increase in the average number of visits in the post-PDL periods for our test (Medicaid) group is generally larger than that for our control group. This observation was confirmed by the regression analysis presented below.

visits

medicaid

medicaid

Table 3 presents the results of regressions for hospital inpatient data. Although almost all of the coefficient estimates were not significant, the coefficients were positive for all model specifications and time periods, indicating an increase in the number of inpatient hospital visits for Medicaid patients compared with non-Medicaid patients. By adding the coefficients for the 2 time periods and using the pre-PDL Medicaid average from Table 2 as a baseline, we calculated an increase of 37% to 42% (depending on whether the model specification includes the demographic controls) in the average number of inpatient hospital visits in the year after PDL implementation for Medicaid patients compared with non-Medicaid patients.

visits

medicaid

Table 4 contains the results of the regressions for hospital outpatient data. Again, the coefficients were always positive and were statistically significant for the first 6-month period after PDL implementation. The combined 1 year post-PDL numbers implied an increase in outpatient hospital visits by 35% to 41% for Medicaid patients compared with non- Medicaid patients.

visits

Medicaid

Table 5 presents the physician regressions. coefficients again were positive in all model specifications and time periods, and were significant for the first 6-month period after PDL implementation. The combined coefficients for the post-PDL year implied a 66% to 78% increase (depending on model specification) in physician visits for Medicaid compared with non-Medicaid patients.

Costs

To determine the net impact of PDL implementation on Medicaid budgets, it was necessary to calculate the costs associated with the estimated increases in hospital and physician visits. Since our dataset did not include any cost-per-claim information, either at the patient or the aggregate level, we estimated costs based on published payment schedules and mean estimates of prices for similar procedures.

The state in our study uses both per-diem and diagnosis-related group (DRG) methods for inpatient Medicaid reimbursement. The majority of hospitals in which cardiovascular-related procedures take place are reimbursed under the DRG method. Since our data do not contain the actual DRGs used for reimbursement, we computed an (unweighted) average reimbursement rate16 of $7385 for all DRGs corresponding to the cardiovascular-related diagnosis and procedure codes in our data. The positive (though not significant) parameter estimates shown in Table 3 implied that the average Medicaid cardiovascular patient incurred a $162 to $185 increase in reimbursements for inpatient procedures compared with non-Medicaid patients in the year following the implementation of the PDL. This, like other cost estimates in this study, must be considered a preliminary, ballpark approximation of the cost magnitudes involved.

For Medicaid hospital outpatient visits, reimbursement was based on a simplified state-constructed listing for ambulatory procedures. If the procedure was not in the ambulatory listing, reimbursement was based on the fee-for-service system used for physician reimbursement. Since our data do not readily map into the state's ambulatory-procedure listing, we used the state's posted reimbursement rate schedule for physician office visits (the posted 2004 rates were adjusted for "medical care services" inflation to equivalent 2002 rates) to calculate all outpatient as well as physician procedure fees.

cost

Several procedures in the state's posted reimbursement schedule were listed as "hand-priced" and were not assigned a dollar figure. We computed the (unweighted) average price of all the procedures that appeared in our outpatient data ($91.66) and physician data ($198.74), and set all hand-priced procedures to these values in their respective regressions. We then computed our variable by adding up all procedure costs for a given patient in each 6-month period.

cost

medicaid

Table 6 contains the parameter estimates from the outpatient and physician regressions. Again, although the coefficient estimates were not statistically significant, they were positive across the board. The combined 1-year estimates implied $20 (61%) and $37 (151%) increases in outpatient and physician reimbursement costs, respectively, for Medicaid compared with non-Medicaid CV-PDL patients. Finally, the net increases in visits and cost for Medicaid patients compared with non-Medicaid patients discussed in this section are summarized in Table 7.

LIMITATIONS

The primary limitation of our work lies in the nature of our data sources. As already noted, these databases include only fractions of the entire salient patient groups, and the restricted sampling frame of the data relative to the population may explain why some of our tests were not statistically significant. Further, the small sample size implies that extrapolation of our results to the entire Medicaid population, or even to only the Medicaid population in our study state, is somewhat problematic; although we are not aware of any systematic bias in our sample, we cannot preclude the possibility of such bias existing. Given the additional simplifying assumptions needed to complete the cost analysis, we consider our cost estimates mostly as a demonstration of the crucial importance of further research in this area.

Data availability also imposed limits on our experimental design. Data for hospital and physician visits were only reliably available starting 6 months before PDL implementation. This ruled out replicating our entire analysis in an earlier time period when no PDL existed to serve as experimental control.

medicaid

R2

medicaid

Data limitations also contributed to restricting our research goals. We did not attempt to present full explanatory models of the change variables under study. Rather, we tried to isolate the effects of PDL implementation (as proxied by the dummy) on these variables. In that light, the low values for our regressions are not surprising, and the focus should remain on the dummy coefficient estimates and their statistical significance. The data also did not permit precise examination of case-mix or disease-intensity differences between Medicaid and non-Medicaid patients in our sample, although our use of a difference-in-differences approach–comparing slopes rather than levels–reduced the likelihood of any biases this might have caused. Finally, this study only examined visits and costs for patients on antihypertensive medications. The extent to which the PDL may have impacted visits and costs in other therapeutic categories is unknown, and merits further study.

DISCUSSION

The visit and cost-increase estimates for Medicaid compared with non-Medicaid patients may be conservative for 2 reasons. First, we only examined visits coded as cardiovascular related, although it is possible that patients sought care for problems that were not coded as cardiovascular related, but were ultimately caused by PDL-driven treatment changes. Second, the fact that our data only captured a small percentage of all physician and hospital visits for a given patient implies that estimates of PDL-driven increases in visits and costs are biased downward (the estimates of percent increases in visits and costs would still be unbiased). For example, if the data captured only 20% of the visits for the patient sample, then visit and cost-increase estimates would be biased downward by a factor of 5. These 2 reasons also may explain why some of our estimates were not statistically significant.

Visits

All the changes we found in Medicaid hospital visits (inpatient and outpatient) were positive, although only those for outpatient visits in the first 6-month period were statistically significant. The observed increases, although small in absolute terms, occurred in a population with a low average rate of hospital visits to begin with. Our estimates of percent increase (roughly 35% to 42% across inpatient and outpatient cohorts) represent a substantial increase in a highly expensive procedure.

This trend also was found for the changes we observed with physician visits: the trend is consistently upward and in some cases statistically significant. The magnitude of change (as a percentage) is larger, at 66% to 78%. This pattern is congruent with our hypothesized explanations for PDL-induced increases in healthcare resource utilization. One would expect physicians to increase their involvement with their patients in the face of higher rates of treatment complications and outright treatment failure (with hospital visits also increasing, but at a slower rate, as it is unlikely that all PDL-induced treatment complications would result in hospital inpatient or outpatient treatment).

Data limitations and low power limit our conclusions. Still, the consistent upward trend in hospital and physician visits strongly suggest the need for further investigation of changes in nonprescription resource utilization in Medicaid patients as a potential unintended consequence of PDL implementation.

Costs

Despite the exploratory nature of the cost estimates, they do offer a glimpse of the magnitude of the PDL effect on Medicaid budgets, and may provide researchers and policymakers with additional information to weigh when considering the need for future research in this area. We found that Medicaid patients with antihypertensive prescription therapy subject to PDL restrictions incurred additional costs in nonprescription medical services in the range of $219 to $242 a year. Below, we attempt to place these numbers in the context of a PDL program cost structure.

One problem with estimating the net benefit of a PDL program is the relative difficulty of obtaining accurate estimates of the costs savings accrued by that program. Changes in the drug mix and listed prices can be used to develop a rough estimate of the change in published costs. But the exact magnitude of cost savings attributable to "extra" nonfederally mandated rebates from those manufacturers whose products are listed on the PDL is not currently available.

It is possible, however, to obtain fairly robust estimates of the per-recipient per-year (PRPY) drug expenditures for Medicaid programs across the country for fiscal year 2000.17 Adjusted to 2002 dollars, the PRPY estimate for our study state is $939. Since PRPY cost is not disease specific, we also obtained, as a check, an estimate of the retail cost for a typical treatment regimen for the drug classes restricted by the PDL under analysis. The Lewin Group recently estimated the retail cost for a typical hypertension regimen for Medicare patients to be approximately $850 (adjusted to 2002 dollars using the medical care services rate of inflation).18 Medicaid outlays per patient may be smaller due to rebates (at least 15.1%), cost controls at the retail pharmacy level, and nonadherence by patients to a full therapy regimen.

Our finding of a $219 to $242 increase in cardiovascular patient costs and a baseline drug-expenditure range of $850 to $939 equate to a requirement for the PDL program to realize prescription cost reductions of more than 25% over nonimplementation before savings begin to accrue. This requirement does not include the cost of administering the program itself.

CONCLUSIONS

Our empirical results indicate that, compared with non-Medicaid patients, hospital outpatient and physician visits increased for Medicaid patients receiving antihypertensive therapy subject to PDL restrictions in the 6-month period after those restrictions were implemented within our sample. The estimates for hospital inpatient visits and for the second 6-month period were not statistically significant, but uniformly indicated a net relative increase in visits for Medicaid patients as well. The associated cost computations were based on many simplifying assumptions to compensate for data limitations. Therefore, we consider the cost estimates exploratory and merely suggestive of the possible magnitudes involved. We believe that further research designed to overcome the limitations of our work is called for, is timely, and should be performed as soon as possible, given the adoption of PDL programs by many states.

Finally, the suggestion of additional hospital and physician costs associated with the implementation of a PDL may have implications beyond Medicaid. It has been established that Medicaid programs have a distinct spillover effect, modifying the behavior of physicians treating patients who are not Medicaid enrollees, especially when Medicaid patients make up a substantial fraction of the physicians' practice.19 This means our admittedly exploratory results suggest the possible existence of a substantial externality to PDL programs, an externality that, if it does exist, is an especially pernicious one. Specifically, if PDLs lead to additional consumption of healthcare resources, and physicians subject to PDL restrictions tend to prescribe in a similar manner to their non-Medicaid patients due to spillover effects, then those non-Medicaid patients subject to such effects will have to bear additional costs that they would otherwise not face. Because physicians whose practice includes a substantial fraction of Medicaid recipients may be expected to provide medical services to patients of lower socioeconomic class, this externality would be imposed on those members of society least able to sustain the additional expense.

We are concerned about the possibility of such a burden being imposed on those members of society (largely the working poor) whose ability to absorb the additional cost is minimal. This possibility offers, in our opinion, an additional and compelling argument for the need to perform further research in this area.

Acknowledgments

We thank Kirsten Axelsen, MS, for important suggestions in selecting our initial patient sample from the prescription database. We also thank Neal Masia, PhD, and an anonymous referee for valuable comments and suggested improvements.