The impact of Medicaid's preferred drug lists on physicians' prescribing behaviour

Abstract

This article examines Medicaid preferred drug lists (PDLs), a cost-containment tool that designates specific drugs for use by Medicaid beneficiaries. We develop an empirical model to measure the direct and spillover effects of Medicaid PDL across Medicaid, cash and third-party payer markets; and apply product level panel data to the cardiovascular market in Illinois and Louisiana. We find a significant decrease in post-PDL Medicaid prescription shares of drugs excluded from the PDL. Spillovers onto third parties and the cash market are also statistically significant. Moreover, a more restrictive prior authorization procedure has a greater impact on prescription shares. There is evidence of gradual adjustment in prescription shares. Lastly, the impact of PDLs is stronger among physicians with a higher share of Medicaid prescriptions.

Acknowledgements

We would like to thank Pierre Azoulay, Mitali Das, Arthur Hass, John Goldsmith, Frank Lichtenberg and seminar participants at the 2006 Western Economic Association International conference for their helpful comments and suggestions. We are also grateful to Pfizer Inc., NY for providing data and support.

Notes

¹ This list is constructed using various sources, including states’ Medicaid websites, the National Patient Advocate Foundation (www.npaf.org/pdf/PDL_Overview.pdf), and the National Conference of State Legislatures (www.ncsl.org/programs/health/medicaidrx.htm).

² Manufacturers in some states can offer a supplemental rebate for placement of their drugs on the PDL. Supplemental rebates provide a state with bargaining power to reduce its drug expenditures. They allow the state to negotiate with pharmaceutical manufacturers and wholesalers that want their product to be available to Medicaid patients without prior authorization. The larger the rebate negotiated by the state, the greater the savings the state will have for drugs it covers.

³ Based on the 1997 Medical Expenditure Panel Survey (MEPS), 36.4% of African–American respondents were on Medicaid at some point in 1997, compared to 16.1% of whites. The 2002 National Drug Code (NDC) Prescriber Data for the US also show that in antidepressant, arthritis and cardiovascular markets, physicians belonging to the lowest quartile of Medicaid prescription shares practise in areas with about 8.5% of African–Americans, while those in the corresponding top quartile practise in areas where African–Americans form over 10.5% of the population.

⁴ Source: Centers for Disease Control and Prevention (CDC)/National Center for Health Statistics (NCHS) (www.americanheart.org/downloadable/heart/1046238221277FS01AF03.pdf).

⁵ In a study of congestive heart failure, Dries et al. (1999) report that 42%(22) of African–Americans in the treatment (prevention) component of the study died compared to 36%(14) of whites. In addition, African–American patients were at increased risk for progression of the disease.

⁶ The ‘physician must document that the drug requested is necessary to prevent a life threatening situation and that items covered without prior authorization … are not effective’ (Chapter 200, Physician Handbook).

⁷ It has been shown that it is cumbersome for physicians to remember which drugs are on which formulary. Therefore, for physicians treating a greater percentage of Medicaid patients, prescribing behaviour is influenced by drugs on Medicaid formulary. However, this may not mean that the best or most suitable drug is prescribed to a patient and may not be in the interest of a patient's well-being.

⁸ At the other end of the spectrum, spillover effects among insured patients with higher co-payments or without coverage for drugs included in the Medicaid PDL may be little to none.

⁹ Difference 1: before vs. after; difference 2: Maine vs. two control states; difference 3: high vs. low Medicaid practice in Maine.

¹⁰ For instance, there are physicians who occasionally prescribe Diuretics to treat other (noncardiovascular) conditions. High-script physicians will more likely be those who make at least one ‘valid’ cardiovascular prescription.

¹¹ Wang et al. (2003) adopt a similar approach. They restrict their analysis to the top 30% of Proton Pump Inhibitor (PPI) prescribers.

¹² We test Moulton's conjecture as part of the sensitivity analysis.

¹³ Das and Mohapatra (2003) undertake a similar empirical strategy in analysing equity market liberalization.

¹⁴ A general indicator, PDL, which groups together Illinois and Louisiana, is also feasible. However, one of our main objectives is to determine if the strictness of the law affects the size of direct and spillover effects; hence we choose in favour of the specification in Model 4.

¹⁵ Regressions using finer levels of decomposition, e.g. 0, (0, 5], (5, 15], (15, 25], (25, 50] and (50, 100]%, reveal that coefficients of Medicaid shares in the (0, 50] range are not significantly different from each other.

¹⁶ This is similar to the 'threat model' (Rosen, 1969), which predicts that wage developments in the nonunion sector follow, to some extent, those in the union sector. This leads to the underestimation of spillover effects.

¹⁷ The alternative of not weighting observations does not affect the qualitative results, although the fit of the model is enhanced with weighting. Also, when we run a random effects model that allows for correlation between drug products, the estimates are almost identical to the fixed effects model.

¹⁸ Data are from NDC Prescriber Data and Pfizer Lab Details.

¹⁹ The NDC samples 70% of the total dispensed prescriptions of retail pharmacies, 98% of which are matched back to their respective practitioners. Moreover, a representative sample distribution is used to ensure coverage of patient populations in each local area. Stratified sampling is performed, whereby the population is divided along social and demographic sub-populations to allow for greater patient homogeneity within each sub-population.

²⁰ Although data for Louisiana are also available up to September 2003, a number of changes to the PDL took place in July 2003. Since a 2-month period may not long enough to fully analyse the impact of these changes, the analysis is based on data prior to this date.

²¹ In the ‘other’ Category, the top three physician specialties are General Surgery, Urology and Psychiatry.

²² The lists of drug products and corresponding analytical categories in Illinois and Louisiana are available upon request. They are also available from Department of Public Aid of Illinois (www.state.il.us/dpa/assets/applets/090503_pdl.pdf) and The Louisiana Department of Health and Hospitals (http://192.60.97.227/provweb1/Pharmacy/preferred.htm).

²³ For example, new prescriptions of CCBs in Illinois would be flagged for implementation during May and June 2003, and so on. In a similar vein, in Louisiana, new prescriptions of ACE, ARBs and CCBs between July and August will be flagged as taking place during the implementation of PDL (period = ‘IMP’). The Models 4 and 5 exclude NRx entries during this period.

²⁴ States where more than 20% of the population is African–American have all instituted Medicaid PDL, which makes Mississippi the only suitable control, despite its higher share of cardiovascular Medicaid. In addition, 1990 Census data also suggest similarities between the selected states.

²⁵ We look at the trend in the share of Medicaid NRx of off-PDL drugs for high-script and low-script physician groups in Illinois. In aggregate, both types of physicians respond similarly to the PDL, with high-script physicians being more responsive.

²⁶ When the shares of off-PDL prescriptions are plotted separately for each state, it is revealed that to a small degree, third-party prescription shares in New York also fall, albeit slightly, at the same time PDL is implemented in Illinois. This underscores the importance of using control states: here, it cautions against overstating the impact of legislation-induced spillovers onto the third-party market.

²⁷ To avoid clutter, Fig. 5 does not include the corresponding plots for New York. Since New York does not have a PDL, the share of off-PDL NRx remains relatively flat across time, for all physician types and in all Medicaid, third-party and cash-payer markets.

²⁸ Elderly Pharmaceutical Insurance Coverage (EPIC) Program is a New York State-sponsored prescription plan for seniors who do not either receive full Medicaid benefits nor have superior prescription coverage (www.health.state.ny.us/nysdoh/epic/faq.htm). In the data, EPIC NRx is 11.5 (8) percent of all third-party (total) NRx. The results remain robust to the alternative where EPIC NRx is excluded.

²⁹ The numbers are the estimated coefficients divided by the average off-PDL Medicaid share in the pre-PDL period. For instance, 66.7% comes from 9.0% divided by 13.5%.

³⁰ In results not reported here, we estimate the legislation-induced impact in Illinois relative to New York and separately, in Louisiana relative to Mississippi. The corresponding estimates turn out to be −9.8 and −6.1%.

³¹ We also test the alternative where physician type is divided into three equally sized groups. The qualitative results remain robust to this specification.

³² The qualitative results are robust to (1) restricting the analysis to ‘MED’ and ‘HIGH’ physicians; (2) reducing the time lag between SR, MR and LR from 6 to 3 months.

³³ In particular, the average number of prescriptions of ‘OFF’, ‘ON’ and ‘NIL’ is respectively 620 (604), 466 (531) and 914 (854) for Louisiana (Mississippi).

³⁴ In an alternative specification, we run a regression where the dependent variable is log(NRx) instead. Off-PDL NRx is estimated to fall by 72%, while on-PDL NRx increases by 14%, both of which are fairly close to the estimates obtained in Model 10 above. In yet another specification, we run models using the log(TRx), where TRx is the total number of prescriptions. The estimated fall in off-PDL TRx is 60%, while the on-PDL TRx increases by 21%, which reinforces the postulation that overall utilization of cardiovascular drugs fell as a result of introducing a PDL in Louisiana.

³⁵ This is computed by weighting each drug type by its pre-PDL prescription share, which is respectively 15.7, 28.3 and 56.0% for ‘OFF’, ‘ON’ and ‘NIL’ drugs.

³⁶ Behavioural models suggest that the most effective therapy prescribed will control hypertension only if a patient is motivated to adhere to the prescribed medication and maintains a healthy lifestyle.

³⁷ Source: www.fda.gov/cder/ob/default.htm. This excludes class AB, since its PDL has a list of drugs included but none are explicitly excluded. With ABs included, the age difference is 2.1 years.

³⁸ ALLHAT is a practice-based clinical trial sponsored by the National Heart, Lung and Blood Institute (source: http://allhat.sph.uth.tmc.edu).