Moral hazard in physician prescription behavior

Abstract

I examine whether the choice made by physicians concerning what drug version — trade-name or generic — to prescribe is subject to moral hazard. I use a data set containing information on exactly what drug and what version was prescribed at a particular patient visit to the physician.

The results indicate that physicians' habits and the tastes acquired by patients are important. However, costs also matter. Patients having to pay large sums out-of-pocket are less likely to have trade-name versions prescribed than patients getting most of their costs reimbursed. This indicates moral hazard.

Introduction

Most trade-name version drugs manage to maintain a large market share after patent expiration although much lower-priced generic drugs with identical therapeutic effects become available. A possible explanation is that neither the final consumer of the drug, nor the one who prescribes the drug has any real reason to prefer the lower-priced generic products. Because of the way health care is financed, patients are often insulated from the extra cost associated with having the expensive trade-name version rather than the generic version prescribed.¹ And if physicians act more in the interest of the patient than the third-party payer, physicians do not have to worry about costs either. In the absence of incentives to keep costs down, prescription decisions are likely to be made on grounds other than price, e.g., perceived quality differences between different versions or physician loyalty to the manufacturer of the trade-name drug (so-called brand loyalty).

Although considerable effort has been made to examine the role of prices for the demand of post-patent drugs, no firm conclusions have yet been reached. Some studies indicate that price does not matter (e.g., Schwartzman, 1976, Hurwitz and Caves, 1988) while others indicate that they do (e.g., Grabowski and Vernon, 1992).

A potential problem contained in all of the above mentioned studies, however, is that they use aggregate sales data. If high reimbursement rates make physicians inclined to prescribe expensive trade-name versions, then we would expect the probability of having the generic version prescribed to increase with the amount that patients pay out-of-pocket. In other words, it is possible that it is the cost difference to the patient that is important for determining which version is prescribed, rather than the price difference per se. However, so long as we only possess information on total sales, this cannot be tested. In order to do that, we have to know what fraction of the cost difference falls on patients and what fraction falls on the third-party payer.

In this paper, a prescription microdata set is used to analyze the determining factors for physician prescription decisions. The data set is constructed from the records of all dispensed drugs from two (the only two) pharmacies in a small Swedish municipality. The available information includes, among other things: the identities of the prescribing physicians and patients; the drug prescribed and the price paid; and patient insurance status. I have selected seven drugs where the trade-name version has lost its patent protection and faces generic competition, and estimate a binary choice model where the dependent variable is drug version: trade-name or generic. In particular, I want to answer the following questions: Are decisions made on a patient-by-patient basis, or are physicians creatures of habit prescribing only one version at a time? How, and to what extent, do physicians take prices into account when deciding what drug to prescribe; are they more responsive to costs incurred by patients than costs incurred by third-party payers?

The two papers which this study is most closely related to are Coscelli (1998) and Hellerstein (1998). Using a US data set on physicians, their patients, and the multi-source drugs prescribed, Hellerstein examines the role of the physician in the choice of drug version. The overall conclusion is that physicians are indeed important agents. She also finds that very little can be explained by observable characteristics of individual patients. In particular, there is no indication that patients who do not have insurance coverage for prescription pharmaceuticals are more likely to have generic versions prescribed. However, the data set has two shortcomings. First, it contains no prices. Therefore, no relative prices can be calculated, and one cannot account for the magnitude of the price difference between versions. Second, the physicians surveyed were asked to record information over a 2-week period only. A particular patient will therefore only show up once, and the tastes which patients might acquire for a certain version cannot be accounted for.

In Coscelli's data set (taken from the Italian drug market), on the other hand, physicians as well as patients can be followed over the course of a 3-year period. This allows him to examine what role habits play for physicians and what role acquired tastes have for patients. He finds that habits and tastes are very important. However, since prices are always the same for different versions of the same drug by regulatory fiat in the Italian drug market, there are no economic incentives for either physicians or patients to prefer one version over the other.

As opposed to the data sets used by Hellerstein and Coscelli, in the data set used here explicit costs for both the patient and the third-party payer are available. In addition, both physicians and patients can be followed over a 3-year period. These features of the data set allow me to examine exactly how habits interact with prices: In what situations are price-differences allowed to override the habits developed by physicians and patients?

It is found that physicians prescribe heavily across versions: they do not specialize in particular versions over time or at a particular point in time. Habits are important — both patients' and physicians' — for the choice of version, but they cannot explain everything. It also matters who bears the costs. Patients having to pay a large sum out-of-pocket are less likely to have the trade-name versions prescribed than patients getting most of their costs reimbursed.

This is an indication of moral hazard, as defined by Pauly (1968): the existence of insurance leads patients to overconsume medical care because they do not bear the full marginal cost of provision. This is sometimes referred to as ex post moral hazard (Zweifel and Breyer, 1997), and it has nothing to do with the fact that insurance might change peoples' behavior so that the likelihood of falling ill increases. Instead, it arises once an illness has occurred because there is usually a choice between different treatments, and insurance might lead patients (or physicians on behalf of their patients) to choose a more expensive treatment than they would have received if they had not been insured.

The result is an interesting illustration of the problem recognized in a theory by Blomqvist (1991) explaining physician behavior. Because of the combination of both information asymmetry and third-party financing in medical care, physicians have to act as “double agents” to achieve efficiency in provision; they have to act in the interest of both the patient and the insurance provider. But the problem is that there are usually not enough instruments available to give physicians the incentives to fulfill their double agency role. Some systems induce them to act more in the interest of the patient (fee-for-service), while others induce them to act more in the interest of the insurer (HMO contracts).

What is found here is that public provision of medical care in the Swedish style — where there are neither any financial incentive to act in the interest of patients, nor any restriction as to what drugs may be prescribed (as with HMO contracts) — does not solve the problem. Physicians are found to act more in the interest of their patients than in the interest of the insurance provider.

The paper is organized as follows. In Section 2, the Swedish reimbursement system is described, with an emphasis on recent changes. In Section 3, the data set is described and some summary statistics are given. Section 4 specifies the model and Section 5 contains the results. Section 6 is the conclusion.