Examining Product Risk in Context. Market Withdrawal of Zomepirac as a Case Study

Objective.\p=m-\To examine changes in the prescribing of analgesics after the market entry and subsequent withdrawal of zomepirac sodium, a nonsteroidal anti\x=req-\ inflammatory drug (NSAID), following repeated reports of zomepirac-related deaths. Design.\p=m-\To evaluate this natural quasi experiment, we conducted time-series analyses to compare prescribing in two cohorts of primary care physicians from July 1980 through September 1983. Setting

the market, as in the recent case of tria- zolam in the United Kingdom.3One ra¬ tionale for product removal is an un¬ stated assumption by regulators and policymakers that all clinical risks at¬ tributable to a drug are eliminated when it is withdrawn; rarely do they examine the comparative risks and benefits of alternative medications that may be sub¬ stituted for the withdrawn product.4We are not aware of any controlled studies that have examined this question.This investigation analyzes changes in the use of various alternative analgesics fol¬ lowing the market entry and withdrawal of zomepirac sodium.
Zomepirac is a prostaglandin syn- thetase inhibitor, one of a class of anal¬ gesic products referred to collectively as nonsteroidal anti-inflammatory drugs (NSAIDs).The drug was first marketed in the United States by McNeil Phar¬ maceutical, Spring House, Pa, in No¬ vember 1980, under the proprietary name Zomax.5 Indications for which it was marketed included relief of moder¬ ate to severe postoperative pain, as well as acute and chronic orthopedic condi¬ tions, osteoarthritis, muscle-contraction headache, dysmenorrhea, and the chronic pain of cancer. 6Zomepirac achieved rapid acceptance, accounting for 11% of new analgesic prescriptions within 4 months of its introduction.7,8   For editorial comment see 1976.
The first report of an apparent ana- phylactic reaction to zomepirac was pub¬ lished in April 1981, about 5 months after the product was released.9In July 1981, McNeil included a mild warning in all product labeling stating that "reactions have been reported."1"After further case reports of zomepirac-associated anaphy- laxis appeared in the medical litera¬ ture,11•12 the manufacturer sent warning letters in April 1982 to 200000 physi¬ cians, alerting them to the drug's poten¬ tial for serious allergic reactions.How¬ ever, 1 week later, the company launched a major 10-week sales campaign ("Op¬ eration 111") intended to increase sales of zomepirac and tolmetin, two ofits most successful analgesics.On March 3, 1983, a Syracuse, NY, television report cited five zomepirac-associated deaths, includ¬ ing a dramatic account by a physician who suffered a life-threatening anaphylactic reaction.13 The following day, Mc¬ Neil voluntarily recalled the product from the market.14Following 2 years of law¬ suits and hearings by both the Food and Drug Administration and Congress, Mc¬ Neil permanently withdrew zomepirac in May 1985. 16omepirac's rapid capture of a sizable share of the analgesic market, followed by stable use for an extended period be¬ fore sudden withdrawal, provides an op¬ portunity to assess both expected and unanticipated substitution effects caused by market availability of a popular drug.Few studies have examined drug sub¬ stitution in a critical way,16 and even fewer have studied the comparative risks as¬ sociated with substituted agents.17Herein we analyze shifts in analgesic prescrib¬ ing patterns associated with zomepirac's market entry; changes in product use during its market life span, particularly at the time of warnings about its safety; and the impact of its rapid withdrawal.
We conclude with a discussion ofthe need to evaluate drugs in relation to their likely substitutes and implications for health policy decision making.

Study Analgesics
We identified all drugs that were po¬ tential analgesic alternatives to zomepi¬ rac at the time of the study, as deter¬ mined by an expert panel of physicians and clinical pharmacists familiar with re¬ search on Medicaid pharmaceutical prac¬ tices.To be included in the study, a medi¬ cation had to be marketed for the treat¬ ment of mild to moderate pain and be available by prescription only.The panel used contemporary drug compendia18•19 as references to aid in selecting alterna¬ tive product classes and identifying all marketed products chemically equivalent to the generic entities contained in each class.The identified drugs were grouped into five categories: (1) zomepirac; (2) other NSAIDs; (3) analgesics containing propoxyphene (hydrochloride or napsy- late); (4) analgesics containing another opioid (codeine, hydromorphone hydrochloride, meperidinehydrochloride, mor¬ phine sulfate, oxycodone, or pentazocine); and (5) analgesics containing barbiturates (combinations with butalbital).The "other-NSAID" category included all NSAIDs other than zomepirac that were available prior to 1984: benoxaprofen, diflunisal, fenoprofen, ibuprofen, in- domethacin, meclofenamate sodium, mefanamic acid, naproxen, oxyphenbutazone, phenylbutazone, piroxicam, sulin- dac, and tolmetin.We did not include over-the-counter products such as ac¬ etaminophen or salicylates since we lacked complete data on their use.
The study drugs selected by the ex¬ pert panel vary greatly in safety and efficacy.For example, zomepirac has been estimated to have 500 to 1000 times the risk of producing a severe anaphylactic reaction as other NSAIDs.15The opioid analgesics can be habituating, and have also been associated with the oc¬ currence of other adverse effects, in¬ cluding hip fracture. 20Propoxyphene may be no more effective an analgesic than aspirin, and its use has been asso¬ ciated with a substantial number of over¬ dose deaths.21Other opioid analgesics such as meperidine can pose a risk of neuropsychiatrie toxic effects.22•23Fi¬ nally, analgesics that contain long-act¬ ing barbiturates carry potential safety risks in dosing and liability for abuse. 24ta Sources Drug claims data from the New Jer¬ sey Medicaid Management Information System provided information on pre¬ scriptions filled and reimbursed during the 39-month study period (July 1980 through September 1983).All claims for the study analgesics were extracted from a previously selected 40% random sample of patients in the New Jersey Medicaid program who received at least one prescription for any drug during this period, a total of 173 726 individu¬ als.The drug claims data used included recipient identifier, drug product code, date the prescription was filled, and a prescriber identification number.
We used these drug claims data to identify all physicians who prescribed one or more analgesic prescriptions to a member of this sample of New Jersey Medicaid drug recipients.Thus, physi¬ cians were not chosen randomly, but were selected by virtue of their pre¬ scribing to a random sample of drug recipients.We also obtained a complete Medicaid provider file, which included the stated specialties of all physicians participating in the New Jersey Med¬ icaid program during the study period.Using prescriber and recipient identi¬ fying numbers, we calculated for every provider the total number of unduplicated Medicaid recipients in our patient sample for whom a medication of any type was prescribed, which was used as a proxy for Medicaid practice size.

Selection of Physician Study Group
To describe changes in prescribing an¬ algesics in a stable population of pri¬ mary care providers, we first identified 1964 physicians, dentists, and osteopaths with at least one filled prescription for any study analgesic in every 6-month period throughout the study.This group, which was responsible for the majority of all analgesic prescriptions to our sample of New Jersey Medicaid recipi¬ ents, averaged over 7600 analgesic pre¬ scriptions per month, or 4.4 analgesic prescriptions per 100 patients.Control¬ ling for changes in prescribing due to the entry and withdrawal of zomepirac from the market, analgesic prescribing remained approximately constant dur¬ ing the follow-up period.
Since we were primarily interested in the prescribing of zomepirac in general ambulatory practice, we used the medi¬ cal specialty data to identify 1183 pri¬ mary care physicians, defined as gen¬ eral practitioners, family practitioners, and internists.These physicians repre¬ sented 60% of consistent prescribers, and wrote a total of 475 600 reimbursed analgesic prescriptions (80.2% of the to¬ tal) to 96989 Medicaid patients (55.8% of total recipients of any drug) during the 39-month study.As indicated in Table 1, these primary care physicians were more likely than other providers to prescribe NSAIDs in general (53.9 prescriptions per 100 patients during this period vs 14.6 prescriptions per 100 patients from other physicians) and zomepirac in particular (4.9 vs 1.5 pre¬ scriptions per 100 patients, respectively).
To examine product substitution ef¬ fects in primary care, we identified 260 physicians who provided 10 or more zome¬ pirac prescriptions during the 28-month period of product life from November 1980 through February 1983 (called zomepirac prescribers).Because other secular trends in drug utilization could conceivably have resulted in increased prescribing of some substitute drugs independent of the withdrawal of zome¬ pirac, we constructed a comparison group of 308 primary care physicians who did not prescribe zomepirac, but who pro¬ vided 10 or more prescriptions for other NSAIDs during the study period (called other-NSAID prescribers).Data

Analyses
We first constructed time series of an¬ algesic utilization by computing the monthly number of analgesic prescrip¬ tions overall and within each analgesic category.All months were adjusted to contain the equivalent of 30 days.We then divided monthly utilization by the number of unduplicated Medicaid drug recipients in the practices of physicians in each group (as described above), ex¬ pressing the results as utilization per 100 recipients.The resulting time-series data for total analgesic use were analyzed by specifying a segmented linear regression model with correction for serially auto- correlated observations.2629This model estimated the overall level and trend in analgesic prescribing during the study period, as well as the size and signifi¬ cance of any changes in level of prescrib¬ ing (discontinuities) after zomepirac en¬ tered the market, and immediately fol¬ lowing its withdrawal from the market.
In this and in all other time-series mod¬ els, data from November through De¬ cember 1980 and March through April 1981, the periods of immediate market adjustment following zomepirac's release and withdrawal, respectively, were ex¬ cluded to obtain more precise estimates.
For zomepirac, the time-series model applied only to the 28-month period it was on the market.We added a term to the model to estimate changes in zome¬ pirac prescribing following the April 1982 nationwide warnings to physicians con¬ cerning adverse drug reactions.Substitution effects were estimated by contrasting the time series for zome¬ pirac prescribers vs other-NSAID pre¬ scribers in each of the five analgesic categories.We converted each drug cat¬ egory's monthly utilization data in each category to its proportion of total anal¬ gesic prescriptions (ie, the proportional preference or market share of each cat¬ egory).This method provided both a stable measure of physician choice of analgesic, the behavior of interest, and controlled for any difference in rates of baseline analgesic prescribing between the two study groups.The resulting monthly time-series data were analyzed using segmented linear regression mod¬ els.These models estimated, for the two prescriber cohorts, the underlying level and trend in proportional use of each analgesic category, and any discontinui¬ ties in use when zomepirac entered the market and when it was withdrawn.The same model also included terms to estimate any differences between zome¬ pirac prescribers and other-NSAID prescribers in changes in analgesic pref¬ erence when zomepirac entered the market and when it was withdrawn.29

RESULTS
Zomepirac prescribers averaged 6.2 an¬ algesic prescriptions per 100 recipients per month (95% confidence interval [CI], 5.9 to 6.6), compared with a slightly lower 5.5 analgesic prescriptions per 100 re¬ cipients (95% CI, 5.2 to 5.7) among other- NSAID prescribers.There was no ob¬ servable trend in total analgesic use in either group during the study period.However, during the period zomepirac was on the market, total analgesic use rose among zomepirac prescribers by an estimated 0.5 prescriptions per month, although this rise was not significant (95% CI, -0.0 to 1.0).
The rapid acceptance and stability of zomepirac prescribing over time among the study cohort of zomepirac prescrib¬ ers paralleled the patterns observed in the entire population of Medicaid phy¬ sicians prescribing analgesics.After a rapid rise in prescribing during the first 2 months of its availability, zomepirac accounted for 11.0% of total analgesics prescribed in this group (95% CI, 10.4% to 11.6%; Fig 1 , bottom).There was no discernible reduction in zomepirac pre¬ scribing by these primary care physi¬ cians associated with the April 1982 warning letter issued by McNeil con¬ cerning the potential for severe allergic reactions.
Other NSAIDs accounted for slightly less than half of all analgesic prescrip¬ tions in both primary care prescriber cohorts before zomepirac was released (Fig 1 , top).The use of other NSAIDs rose significantly in both cohorts throughout the study period.However, once zomepirac became available, the trends in use diverged, with the rate of prescribing other NSAIDs increasing more slowly among zomepirac prescrib¬ ers than among other-NSAID prescrib¬ ers.After zomepirac was withdrawn  -Time-series analyses of prescribing zome¬ pirac sodium (bottom), other nonsteroidal anti- inflammatory drugs (NSAIDs) (top), and pro- poxyphene (hydrochloride or napsylate) (middle) among primary care physicians prescribing zome¬ pirac (n=260) vs those prescribing NSAIDs other than zomepirac (n=308).from the market, prescribing of other NSAIDs rose rapidly among former zomepirac prescribers to the rate ob¬ served in the comparison cohort.
We also compared monthly rates of prescribing of the other study analge¬ sics between zomepirac prescribers and other NSAID prescribers.The presence of zomepirac on the market was asso¬ ciated with a clear divergence of trends between the two groups in the use of single-agent and combination pro¬ poxyphene products (Fig 1 , middle).Prior to the release of zomepirac, future prescribers of zomepirac showed a slightly higher rate of prescribing pro¬ poxyphene than the comparison group.After zomepirac entered the market, the relative rates of propoxyphene use in these two groups reversed, with other- NSAID prescribers having a slightly but consistently higher preference for pro¬ poxyphene than zomepirac prescribers.When zomepirac was withdrawn, the rates of prescribing propoxyphene in the two groups became very similar.
The results of segmented linear re¬ gression models of analgesic preference are presented in Table 2.As shown in Fig 2, during the period zomepirac was on the market (November 1980 through February 1983), the total preference for NSAIDs grew significantly in both groups of prescribers.However, among our cohort of zomepirac prescribers, we estimate that NSAID use (including zomepirac) was 2.9% higher than among other-NSAID prescribers (95% CI, 1.9% to 3.9%; P<.001).This difference in pref¬ erence was balanced by a lower use of propoxyphene analgesics (-2.8%; 95% CI, -2.2% to -3.4%; P<.001) among zomepirac prescribers.The use of other opioid analgesics and barbiturates de¬ clined significantly in both groups.
After zomepirac was withdrawn from the market, over two thirds ofthe 11.0% of analgesic prescribing previously al¬ located to zomepirac was offset by a relative increase in the use of other NSAIDs (+6.8%; 95% CI, 4.4% to 9.2%; P<.001).Despite this increase, total NSAID use among zomepirac prescrib¬ ers remained about 1.4% lower than in the comparison group.Relative increases in two other analgesic categories offset the remaining reductions in zomepirac prescribing.Preference for pro¬ poxyphene was 2.1% higher among former zomepirac users (95% CI, 0.4% to 3.8%; P<.05), and relative use of bar¬ biturates was 2.7% higher (95% CI, 1.5% to 3.9%; P<.001).The use of other opioid analgesics rose significantly in both groups during this period.

COMMENT Zomepirac and the Analgesic Market
We are not aware of any well-designed studies that have investigated the in¬ tended and unintended impacts of with¬ drawal of a drug by its manufacturer due to unexpected and serious adverse drug reactions. 30Zomepirac was an in¬ teresting product to study because it captured a substantial share of the an- algesic market very soon after it was introduced, and maintained this level consistently throughout its market life.
In our large Medicaid claims database, zomepirac accounted for 12.4% of NSAID use while it was on the market, and 6.4% of total analgesic prescribing, findings that are reasonably consistent with national figures.31"33 Zomepirac's rapid acceptance and large market share suggest an effective marketing effort by McNeil.Nationally disseminated mailed warnings to pre¬ scribers in April 1982 concerning zome¬ pirac's potential to cause anaphylactic reactions had no detectable effect on zomepirac use in our cohort of 260 zome¬ pirac prescribers, or in the larger cohort of 1964 physicians.One explanation for the warning's apparent lack of impact was the launch of a major promotional campaign targeting zomepirac prescrib¬ ers 1 week after the mailed letter.It is possible that the apparent overall lack of effect of product warnings, rather than reflecting a stable market, may mask a situation in which some physicians re¬ duced prescribing due to concerns about the product, while others increased pre¬ scribing in response to intensified mar¬ keting efforts.
Regardless of the promotional cam¬ paign's effectiveness, mailed warnings about pharmaceutical products have been shown to be very weak behavior- change interventions in comparison with more direct, person-to-person methods.1 We previously reported a similar lack of effect of repeated warnings by the US Food and Drug Administration and phar¬ maceutical manufacturers aimed at re¬ ducing misuse of propoxyphene analge¬ sics due to their risk of toxic effects and habituation.21 There is ample evidence that zomepi¬ rac's entry into and exit from the mar¬ ketplace was accompanied by significant shifts in choice of analgesic therapies.Primary care physicians were the pre¬ dominant prescribers of zomepirac.When zomepirac became available, it appears to have been adopted by some primary care physicians not only as a preferred NSAID, but also as an apparently safer, more rational alternative to analgesics such as propoxyphene for treating gen¬ eral acute or chronic pain.31When zome¬ pirac was withdrawn, the majority of its use was replaced by other NSAIDs, an apparent gain in product safety.How¬ ever, significant increases in preference for propoxyphene and fixed-combination barbiturate analgesics were also seen.These substituted products themselves carry notable risks ofadverse clinical con¬ sequences, including habituation and ac¬ cidental death.

Limitations of Study Methods
Several limitations of our methods and threats to the validity of our findings must be considered.The generalizability of our results may be limited by a number of factors.Our sample selection process eliminated any physician who did not prescribe at least one analgesic to a member of our random sample of patients.We also observed only the Med¬ icaid portion of each study physician's practice, and we further required that physicians have some analgesic prescrib¬ ing for the entire duration of the study.The relatively small sizes of the Med¬ icaid practices for many physicians made analysis at the physician level impos¬ sible.However, the distribution of use of individual NSAIDs in our analyses closely approximates the patterns ob¬ served in a representative national sample of computerized pharmacies.31 There is no reason to believe that the prescribing habits of our study cohort differ significantly from the habits of physicians who did not meet the require- -Changes in preference for selected catego¬ ries of analgesics among prescribers of zomepirac sodium (n=260) and prescribers of nonsteroidal anti-inflammatory drugs other than zomepirac (n=308) both following the release of zomepirac (top) and after it was withdrawn from the market (bottom).The drugs marked with an asterisk are analgesic preparations containing propoxyphene hydrochloride, another opioid, or a barbiturate.ment for prescribing activity in Medic¬ aid, or that their prescribing habits would vary for the larger, non-Medicaid portion of their practices.
We were only able to obtain data on analgesic prescribing for 4 months prior to the release of zomepirac, and 6 months following its withdrawal from the mar¬ ket.Consequently, longer-term trends in the prescribing of competing analge¬ sic products, independent of changes as¬ sociated with zomepirac's entry into and withdrawal from the market, were dif¬ ficult to detect.However, because of the rapid rise in the popularity of zomepirac and its sudden withdrawal, short-term changes in prescribing habits were clearly discernible.
Other methodological problems arise from the quasi-experimental nature of the study design.Interrupted time-se¬ ries analysis, the strongest statistical technique for analysis of nonexperimental data, may not detect subtle changes in trend, and cannot conclusively deter¬ mine the reasons for observed changes.We compensated for this deficit by in- eluding a comparison group of other- NSAID prescribers.However, we had no control over which primary care phy¬ sicians were classified as zomepirac pre¬ scribers or other-NSAID prescribers.Our cohorts of zomepirac prescribers and other-NSAID prescribers differed slightly at baseline in rate of analgesic use in their practices.Other-NSAID pre¬ scribers may also differ in other ways; for example, they may be less likely to try new analgesic products, or they may see a different mix of patients and, there¬ fore, use different medications.How¬ ever, although differences between these two groups are likely to exist, the over¬ all similarity of patterns of analgesic use before zomepirac was first marketed, the stability of prescribing throughout the study period, and the visible dis¬ continuities in use of specific drugs fol¬ lowing the changes in the marketplace give credibility to the measured substi¬ tution effects.
Could the observed changes in drug use be explained by other changes in the marketplace?Divergence of the utiliza¬ tion trends for zomepirac and other- NSAID prescribers immediately follow¬ ing the market entry of zomepirac and the sudden convergence of trends in cer¬ tain substitute categories following the market withdrawal of zomepirac make it very unlikely that unrecognized fac¬ tors were responsible for these substi¬ tution effects.
Because of the retrospective nature of the study and limitations on available data, it is impossible to assess certain characteristics of the study physicians that might influence their reactions to product withdrawals.For example, we know little about case-mix profile, spe¬ cific motivations for choosing zomepirac vs alternative analgesic products, or the influence of patient requests for par¬ ticular types of analgesic.To truly un¬ derstand the dynamics of prescribing behavior, one would have to character¬ ize prospectively these and other clini¬ cal and nonclinical factors.34   Understanding Product Risks and Benefits in Context Some drugs have substantial risks that are tolerated because oftheir established benefits in relation to available thera¬ peutic alternatives (eg, zidovudine).However, even less toxic drugs do not always prove safe or effective for every individual.From the epidemiologie rather than the clinical perspective, pre¬ scribing following a regulatory change or product withdrawal becomes rela¬ tively safer if the combined statistical risk of all products prescribed within a therapeutic category decreases.
Recent examples exist of the banning or voluntary withdrawal of pharmaceu¬ tical products thought to be unsafe or ineffective. 4Although limited, the best available data suggest that the quality of prescribing can either increase or de¬ crease following the removal of specific medications from a national formulary35 or the withdrawal of Medicaid reimburse¬ ment for scientifically unsubstantiated therapies.4One study reported no reduc¬ tion in risk of self-poisoning when dan¬ gerous products implicated in drug abuse were removed from the national market, primarily because of increased abuse of drugs of equal or greater toxicity.17Thus, suboptimal product substitution follow¬ ing well-intended regulatory initiatives seems to be a generalized phenomenon that policymakers must consider when¬ ever restrictions on drug availability are proposed.
Risk trade-offs due to product sub¬ stitutions can occur not only when drugs are banned or withdrawn, but when¬ ever prescriber choices are constrained, whether by economic or administrative limits.Even less drastic interventions, such as audited triplicate prescription forms for benzodiazepines, have been associated with undesirable substitution effects.36Decision makers should care¬ fully consider the relative risks of com¬ peting products when determining phar¬ maceutical coverage policies in all set¬ tings that maintain formularies or otherwise restrict access to specific products, from major federal programs (eg, Medicare, Medicaid, and the De¬ partment of Veterans Affairs) to indi¬ vidual health organizations.
This study examined a product to which significant, dramatic, and imme¬ diate health risks were ascribed.Ap¬ proximately 40 deaths were attributed to anaphylactic reactions to zomepirac in the Food and Drug Administration's Sentinel Reporting System in the 28 months it was on the market37; a revised Food and Drug Administration estimate totaled 14 zomepirac-related deaths.10Based on data from 1982 and early 1983,31•32 about 15 million prescriptions for zomepirac were written while it was on the market.Taken together, these estimates indicate a total risk of 0.9 to 2.7 deaths per million zomepirac pre¬ scriptions.Thus, substitution of other NSAIDs, which have a far lower risk of anaphylaxis,15 for zomepirac may rep¬ resent safer prescribing.
However, propoxyphene and barbi¬ turate-containing analgesics, which we have shown to be competing therapeu¬ tic substitutes for zomepirac, also carry risks.9][40] Propoxyphene and barbiturates are also associated with other nonfatal adverse outcomes, including habitua- tion.21However, these drugs' risks are less dramatic than zomepirac's, are less likely to be reported, and are therefore less visibly attributable to the products in question.
We did not have access to morbidity and mortality data necessary to conduct a reliable population-based risk analy¬ sis.However, the key policy lesson to be learned goes beyond the important is¬ sue of whether zomepirac's withdrawal from the market was associated with an increase or decrease in drug-related mor¬ tality.How could market withdrawal have been better accomplished?Edu¬ cating prescribers about appropriate pain management remains a long-term strategy for decreasing analgesic risks, but such education is time-consuming and difficult to operationalize.1At a mini- mum, this study highlights a need for more effective communication to medi¬ cal practitioners before or during prod¬ uct withdrawals or regulatory changes to alert them to both the drug removal and preferred therapeutic alternatives.4Since mailed circulars appear to be in¬ effective means to accomplish these tasks, other strategies need to be ex¬ plored, perhaps using credible opinion leaders or local medical associations as communication channels.
In summary, one cannot evaluate the impact of withdrawing a given drug as¬ sociated with serious adverse drug re¬ actions except in relation to the risks posed by substituted products.Although policymakers may assume that with¬ drawing a drug eliminates its entire bur¬ den of risk, this is certainly not the case.This study underscores the need for poli¬ cymakers to consider risk in context and to anticipate unintended drug substitu¬ tions.Physicians respond actively, not passively, to forced changes in their range of therapeutic options.The de¬ gree of improvement in quality of care through drug-restriction policies de¬ pends on the characteristics of available therapeutic alternatives, the perceptions of prescribers and patients toward those alternatives, and the degree to which timely education and promotional efforts can influence prescribing.At the very least, a growing body of literature sug¬ gests that private and public policymak¬ ers should identify likely effects of ra¬ tional and irrational product substitu¬ tion before banning specific agents in order to characterize, and hopefully pre¬ vent, unintended outcomes.
This study was supported in part by grant HS 05554 from the Agency for Health Care Policy and Research, Rockville, Md; and grants from the Danish International Development Agency, Copenhagen, Denmark, to the International Network for the Ra¬ tional Use of Drugs, Boston, Mass; and clinical inves¬ tigator award K08 AG00510 from the National In¬ stitute on Aging, Bethesda, Md (Dr Gurwitz).

Fig 2 .
Fig 2.-Changes in preference for selected catego¬ ries of analgesics among prescribers of zomepirac sodium (n=260) and prescribers of nonsteroidal anti-inflammatory drugs other than zomepirac (n=308) both following the release of zomepirac (top) and after it was withdrawn from the market (bottom).The drugs marked with an asterisk are analgesic preparations containing propoxyphene hydrochloride, another opioid, or a barbiturate.

Table
.-Prescriptions for Study Analgesics per 100 Medicaid Recipients in Practice by Provider Specialty, July 1980 ThroughNovember 1981

Table 2 .
-Changes in Analgesic Preference Following Release of Zomepirac and Its Market Withdrawal Among Prescribers of Zomepirac and Prescribers of Other NSAIDs* Study Analgesic Category, Proportional Share of Use (SE)t #P<.01.