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

The Harvard community has made this article openly available. Please share how this access benefits you. Your story matters

Citation Ross-Degnan, Dennis. 1993. “Examining Product Risk in Context.” JAMA 270 (16) (October 27): 1937. doi:10.1001/ jama.1993.03510160055029.

Published Version doi:10.1001/jama.1993.03510160055029

Citable link http://nrs.harvard.edu/urn-3:HUL.InstRepos:32692587

Terms of Use This article was downloaded from Harvard University’s DASH repository, and is made available under the terms and conditions applicable to Other Posted Material, as set forth at http:// nrs.harvard.edu/urn-3:HUL.InstRepos:dash.current.terms-of- use#LAA Examining Product Risk in Context Market Withdrawal of Zomepirac as a Case Study Dennis Ross-Degnan, ScD; Stephen B. Soumerai, ScD; Eric E. Fortess, ScD, MPH; Jerry H. Gurwitz, MD

Objective.\p=m-\Toexamine changes in the prescribing of analgesics after the mar- as nonsteroidal anti-inflammatory drugs ket entry and subsequent withdrawal of zomepirac sodium, a nonsteroidal anti\x=req-\ (NSAIDs). The drug was first marketed inflammatory drug (NSAID), following repeated reports of zomepirac-related deaths. in the United States by McNeil Phar¬ evaluate this natural we conducted time-series maceutical, Spring House, Pa, in No¬ Design.\p=m-\To quasi experiment, vember under the analyses to compare prescribing in two cohorts of primary care physicians from July 1980, proprietary name Zomax.5 Indications for which it 1980 1983. through September was marketed included relief of moder¬ care to Setting.\p=m-\Studyphysicians provided outpatient pharmaceutical patients ate to severe as well enrolled in the New postoperative pain, Jersey Medicaid program. as acute and chronic orthopedic condi¬ Participants.\p=m-\Weidentified 260 primary care physicians who provided 10 or tions, osteoarthritis, muscle-contraction more prescriptions for zomepirac (zomepirac prescribers) and 308 who provided headache, dysmenorrhea, and the 10 or more prescriptions for NSAIDs other than zomepirac (other-NSAID prescrib- chronic pain of cancer.6 Zomepirac ers) in Medicaid during the study period. achieved rapid acceptance, accounting Main Outcome Measures.\p=m-\Monthlyrates of prescribing for zomepirac and for 11% of new analgesic prescriptions within months of its several categories of substitute analgesics among Medicaid patients seen by study 4 introduction.7,8 physicians. Main Results.\p=m-\Zomepiracaccounted for a stable 11.0% of analgesic prescrib- For editorial comment see 1976. ing among the zomepirac-prescriber cohort; label changes and manufacturer 11 months the withdrawal from market product-risk warnings before product's the The first report of an apparent ana- had no impact on use. After market entry, zomepirac prescribers reduced use of phylactic reaction to zomepirac was pub¬ other NSAIDs and propoxyphene (hydrochloride or napsylate) in comparison with lished in April 1981, about 5 months after other-NSAID prescribers (-8.1% and -2.8% of total analgesic prescribing, the product was released.9 In July 1981, respectively; P<.001). After the product's withdrawal from the market, zomepirac McNeil included a mild warning in all prescribers showed significant increases in relative prescribing of other NSAIDs product labeling stating that "reactions (+6.8%; P<.001), propoxyphene (+2.1%; P<.05), and analgesics containing bar- have been reported."1" After further case biturates reports of zomepirac-associated anaphy- (+2.7%; P<.001). laxis in the medical litera¬ Conclusions.\p=m-\Thesudden withdrawal of from the market resulted appeared zomepirac the manufacturer sent in substitutions of but also of alternative ture,11·12 warning not only other NSAIDs, analgesics that carry letters in 1982 to 200000 risks of and adverse effects. in April physi¬ habituation Apparent gains patient safety resulting cians, alerting them to the drug's poten¬ from market withdrawal of medications must be evaluated in comparison with risks tial for serious allergic reactions. How¬ of medications likely to be substituted. ever, 1 week later, the company launched (JAMA. 1993;270:1937-1942) a major 10-week sales campaign ("Op¬ eration 111") intended to increase sales ofzomepirac and tolmetin, two ofits most DURING the past two decades, an in¬ the market, as in the recent case of tria- successful analgesics. On March 3, 1983, creasing number of government, medi¬ zolam in the United Kingdom.3 One ra¬ a Syracuse, NY, television report cited cal, and lay press reports have focused tionale for product removal is an un¬ five zomepirac-associated deaths, includ¬ on the problem ofunanticipated adverse stated assumption by regulators and ing a dramatic account by a physician reactions to prescription drugs.1 Gov¬ policymakers that all clinical risks at¬ who suffered a life-threatening anaphy- ernment and industry responses have tributable to a drug are eliminated when lactic reaction.13 The following day, Mc¬ ranged from modest label warnings2 to it is withdrawn; rarely do they examine Neilvoluntarily recalled the product from withdrawal ofthe offending product from the comparative risks and benefits of the market.14 Following 2 years of law¬ alternative medications that may be sub¬ suits and hearings by both the Food and From the Drug Policy Research Group, Department stituted for the withdrawn product.4 We Drug Administration and Congress, Mc¬ of Social Medicine, and the Department of Ambulatory are not aware of any controlled studies Neil permanently withdrew zomepirac Care and Prevention, Harvard Medical School, Boston, in Mass (Drs Ross-Degnan and Soumerai); the Depart- that have examined this question. This May 1985.16 ment of Public Management/Health Administration investigation analyzes changes in the Zomepirac's rapid capture of a sizable Concentration, Suffolk University, Boston, Mass (Dr use of various alternative fol¬ share of the followed and the for of Clinical Strat- analgesics analgesic market, Fortess); Program Analysis stable use for an extended be¬ egies, Gerontology Division, Department of Medicine, lowing the market entry and withdrawal by period Brigham and Women's Hospital and Harvard Medical of zomepirac sodium. fore sudden withdrawal, provides an op¬ School, Boston, Mass (Dr Gurwitz). is a syn- to assess both and Reprint requests to Department of Ambulatory Care Zomepirac prostaglandin portunity expected and Prevention, Harvard Medical School, 126 Brookline thetase inhibitor, one of a class of anal¬ unanticipated substitution effects caused Ave, Suite 200, Boston, MA 02215 (Dr Soumerai). gesic products referred to collectively by market availability of a popular drug.

Downloaded From: http://jamanetwork.com/pdfaccess.ashx?url=/data/journals/jama/9819/ by a Harvard University User on 02/22/2017 Table 1.—Prescriptions for Study Analgesics per 100 Medicaid Recipients in Practice by Provider Specialty, July 1980 Through November 1981 Study Analgesics Physicians, Zomepirac Other Analgesic Analgesic All Study Physician Group No. Sodium NSAIDs* With Opioidst With Barbiturates Analgesics General practice 5.1 39.2 34,4 8.8 87.5 Internal medicine 73.7 12.9 147.4 Family practice 238 3.7 45.3 33.0 8.2 90.2 All Primary Care Physicians 1183 4.9 49.0 100.7 Dentistry, oral surgery 140 3.3 4.9 81.8 94.2 Pediatrics 121 0.2 2.7 4.9 0.7 8.5 General surgery 36.1 33.3 80.9 Obstetrics, gynecology 106 1.3 9.8 21.7 5.7 38.5 Other specialty 287 3.1 29.8 42.8 7.0 82.7 All Non-Primary Care Physicians 776 1.5 13.1 4.0 *NSAID indicates nonsteroidal anti-inflammatory drug. tAnalgesIc products containing propoxyphene (hydrochloride or napsylate), pentazocine, meperidine hydrochloride, or codeine.

Few studies have examined drug sub¬ dac, and tolmetin. We did not include recipients. We also obtained a complete stitution in a critical way,16 and even fewer over-the-counter products such as ac¬ Medicaid provider file, which included have studied the comparative risks as¬ etaminophen or salicylates since we the stated specialties of all physicians sociated with substituted agents.17 Herein lacked complete data on their use. participating in the New Jersey Med¬ we analyze shifts in analgesic prescrib¬ The study drugs selected by the ex¬ icaid program during the study period. ing patterns associated with zomepirac's pert panel vary greatly in safety and Using prescriber and recipient identi¬ market entry; changes in product use efficacy. For example, zomepirac has fying numbers, we calculated for every during its market life span, particularly been estimated to have 500 to 1000 times provider the total number of undupli- at the time of warnings about its safety; the risk of producing a severe anaphy- cated Medicaid recipients in our patient and the impact of its rapid withdrawal. lactic reaction as other NSAIDs.15 The sample for whom a medication of any We conclude with a discussion ofthe need opioid analgesics can be habituating, and type was prescribed, which was used as to evaluate drugs in relation to theirlikely have also been associated with the oc¬ a proxy for Medicaid practice size. substitutes and implications for health currence of other adverse effects, in¬ policy decision making. cluding hip fracture.20 Propoxyphene Selection of Physician Study Group be no more effective an may analgesic To describe changes in prescribing an¬ METHODS than and its use has been asso¬ aspirin, algesics in a stable population of pri¬ ciated with a substantial number of over¬ Study Analgesics care we first identified dose deaths.21 Other mary providers, We identified all drugs that were po¬ opioid analgesics 1964 physicians, dentists, and osteopaths such as can a risk of tential alternatives to meperidine pose with at least one filled for analgesic zomepi¬ toxic effects.22·23 Fi¬ prescription rac at the time of the as deter¬ neuropsychiatrie in 6-month study, that contain any study analgesic every mined an of nally, analgesics long-act¬ the This by expert panel physicians barbiturates period throughout study. group, and clinical familiar with re¬ ing carry potential safety which was for the pharmacists risks in and for abuse.24 responsible majority search on Medicaid pharmaceutical prac¬ dosing liability of all analgesic prescriptions to our tices. To be included in the study, a medi¬ sample of New Jersey Medicaid recipi¬ Data Sources cation had to be marketed for the treat¬ ents, averaged over 7600 analgesic pre¬ ment of mild to moderate pain and be Drug claims data from the New Jer¬ scriptions per month, or 4.4 analgesic available by prescription only. The panel sey Medicaid Management Information prescriptions per 100 patients. Control¬ used contemporary drug compendia18·19 System provided information on pre¬ ling for changes in prescribing due to as references to aid in selecting alterna¬ scriptions filled and reimbursed during the entry and withdrawal of zomepirac tive product classes and identifying all the 39-month study period (July 1980 from the market, analgesic prescribing marketed products chemically equivalent through September 1983). All claims for remained approximately constant dur¬ to the generic entities contained in each the study analgesics were extracted from ing the follow-up period. class. The identified drugs were grouped a previously selected 40% random Since we were primarily interested in into five categories: (1) zomepirac; (2) sample of patients in the New Jersey the prescribing of zomepirac in general other NSAIDs; (3) analgesics containing Medicaid program who received at least ambulatory practice, we used the medi¬ propoxyphene (hydrochloride or napsy- one prescription for any drug during cal specialty data to identify 1183 pri¬ late); (4) analgesics containing another this period, a total of 173 726 individu¬ mary care physicians, defined as gen¬ opioid (codeine, hydromorphone hydro- als. The drug claims data used included eral practitioners, family practitioners, chloride, meperidinehydrochloride, mor¬ recipient identifier, drug product code, and internists. These physicians repre¬ phine sulfate, oxycodone, or pentazocine); date the prescription was filled, and a sented 60% of consistent prescribers, and (5) analgesics containing barbiturates prescriber identification number. and wrote a total of 475 600 reimbursed (combinations with butalbital). The We used these drug claims data to analgesic prescriptions (80.2% of the to¬ "other-NSAID" category included all identify all physicians who prescribed tal) to 96989 Medicaid patients (55.8% NSAIDs other than zomepirac that were one or more analgesic prescriptions to a of total recipients of any drug) during available prior to 1984: benoxaprofen, member of this sample of New Jersey the 39-month study. As indicated in diflunisal, fenoprofen, ibuprofen, in- Medicaid drug recipients. Thus, physi¬ Table 1, these primary care physicians domethacin, meclofenamate sodium, cians were not chosen randomly, but were more likely than other providers mefanamic acid, naproxen, oxyphenbuta- were selected by virtue of their pre¬ to prescribe NSAIDs in general (53.9 zone, phenylbutazone, piroxicam, sulin- scribing to a random sample of drug prescriptions per 100 patients during

Downloaded From: http://jamanetwork.com/pdfaccess.ashx?url=/data/journals/jama/9819/ by a Harvard University User on 02/22/2017 this period vs 14.6 prescriptions per 100 preference or market share of each cat¬ patients from other physicians) and egory). This method provided both a zomepirac in particular (4.9 vs 1.5 pre¬ stable measure of physician choice of scriptions per 100 patients, respectively). analgesic, the behavior of interest, and To examine product substitution ef¬ controlled for any difference in rates of fects in primary care, we identified 260 baseline analgesic prescribing between physicians who provided 10 or more zome¬ the two study groups. The resulting pirac prescriptions during the 28-month monthly time-series data were analyzed period of product life from November using segmented linear regression mod¬ 1980 through February 1983 (called els. These models estimated, for the two ^ $r y¡P zomepirac prescribers). Because other prescriber cohorts, the underlying level secular trends in drug utilization could and trend in proportional use of each conceivably have resulted in increased analgesic category, and any discontinui¬ prescribing of some substitute drugs ties in use when zomepirac entered the independent of the withdrawal of zome¬ market and when it was withdrawn. The pirac, we constructed a comparison group same model also included terms to of 308 primary care physicians who did estimate any differences between zome¬ 4& ^ ^ ^ ^ *& s<& <& not prescribe zomepirac, but who pro¬ pirac prescribers and other-NSAID & & & & 2&¿0* & vided 10 or more prescriptions for other prescribers in changes in analgesic pref¬ y? y? y$> NSAIDs the erence when entered the during study period (called zomepirac 20 other-NSAID prescribers). market and when it was withdrawn.29 Zomepirac Ü)"D Zomepirac Appears Zomepirac Withdrawn

, on US Market From US Data RESULTS ÇJ1Û Market Analyses «'S 10 1^'-^ ~^* 5 "> We first constructed time series of an¬ Zomepirac prescribers averaged 6.2 an¬ algesic utilization by computing the algesic prescriptions per 100 recipients OD. monthly number of analgesic prescrip¬ per month (95% confidence interval [CI], tions overall and within each analgesic 5.9 to 6.6), compared with a slightly lower && ^ v& 4& 4& ^ 4& All months were 5.5 re¬ ^s category. adjusted to analgesic prescriptions per 100 &^a^ y**«*^& ^^& contain the equivalent of 30 days. We cipients (95% CI, 5.2 to 5.7) among other- then divided monthly utilization by the NSAID prescribers. There was no ob¬ Zomepirac Prescribers — number of Medicaid servable trend in total use in Other-NSAID Prescribers unduplicated drug analgesic —- recipients in the practices of physicians either group during the study period. in each (as described ex¬ the group above), However, during period zomepirac Fig 1.—Time-series analyses of prescribing zome¬ pressing the results as utilization per 100 was on the market, total analgesic use pirac sodium (bottom), other nonsteroidal anti- recipients. The resulting time-series data rose among zomepirac prescribers by an inflammatory drugs (NSAIDs) (top), and pro- for total analgesic use were analyzed by estimated 0.5 prescriptions per month, poxyphene (hydrochloride or napsylate) (middle) among primary care physicians prescribing zome¬ a linear this rise was not specifying segmented regression although significant (95% pirac (n=260) vs those prescribing NSAIDs other model with correction for serially auto- CI, -0.0 to 1.0). than zomepirac (n=308). correlated observations.2629 This model The rapid acceptance and stability of estimated the overall level and trend in zomepirac prescribing over time among from the market, prescribing of other analgesic prescribing during the study the study cohort of zomepirac prescrib¬ NSAIDs rose rapidly among former period, as well as the size and signifi¬ ers paralleled the patterns observed in zomepirac prescribers to the rate ob¬ cance of any changes in level of prescrib¬ the entire population of Medicaid phy¬ served in the comparison cohort. ing (discontinuities) after zomepirac en¬ sicians prescribing analgesics. After a We also compared monthly rates of tered the market, and immediately fol¬ rapid rise in prescribing during the first prescribing of the other study analge¬ lowing its withdrawal from the market. 2 months of its availability, zomepirac sics between zomepirac prescribers and In this and in all other time-series mod¬ accounted for 11.0% of total analgesics other NSAID prescribers. The presence els, data from November through De¬ prescribed in this group (95% CI, 10.4% of zomepirac on the market was asso¬ cember 1980 and March through April to 11.6%; Fig 1, bottom). There was no ciated with a clear divergence of trends 1981, the periods of immediate market discernible reduction in zomepirac pre¬ between the two groups in the use of adjustment following zomepirac's release scribing by these primary care physi¬ single-agent and combination pro¬ and withdrawal, respectively, were ex¬ cians associated with the April 1982 poxyphene products (Fig 1, middle). cluded to obtain more precise estimates. warning letter issued by McNeil con¬ Prior to the release of zomepirac, future For zomepirac, the time-series model cerning the potential for severe allergic prescribers of zomepirac showed a applied only to the 28-month period it reactions. slightly higher rate of prescribing pro¬ was on the market. We added a term to Other NSAIDs accounted for slightly poxyphene than the comparison group. the model to estimate changes in zome¬ less than half of all analgesic prescrip¬ After zomepirac entered the market, the pirac prescribing following the April 1982 tions in both primary care prescriber relative rates of propoxyphene use in nationwide warnings to physicians con¬ cohorts before zomepirac was released these two groups reversed, with other- cerning adverse drug reactions. (Fig 1, top). The use of other NSAIDs NSAID prescribers having a slightly but Substitution effects were estimated rose significantly in both cohorts consistently higher preference for pro¬ by contrasting the time series for zome¬ throughout the study period. However, poxyphene than zomepirac prescribers. pirac prescribers vs other-NSAID pre¬ once zomepirac became available, the When zomepirac was withdrawn, the scribers in each of the five analgesic trends in use diverged, with the rate of rates ofprescribing propoxyphene in the categories. We converted each drug cat¬ prescribing other NSAIDs increasing two groups became very similar. egory's monthly utilization data in each more slowly among zomepirac prescrib¬ The results of segmented linear re¬ category to its proportion of total anal¬ ers than among other-NSAID prescrib¬ gression models of analgesic preference gesic prescriptions (ie, the proportional ers. After zomepirac was withdrawn are presented in Table 2. As shown in

Downloaded From: http://jamanetwork.com/pdfaccess.ashx?url=/data/journals/jama/9819/ by a Harvard University User on 02/22/2017 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 Zomepirac Other Other Sodium NSAIDs Propoxyphene} Opioids Barbiturates Use at beginning of study period (June 1980) 0.000 0.475(0.0061) 0.210(0.0042) 0.206 (0.0040) 0.109(0.003) Monthly trend in use (July 1980-September 1983) <0.001§ 0.002II (0.0003) <0.001 -0.0 2 (0.0002) <0.001 Changes in average monthly use among prescribers of other NSAIDs While zomepirac on market <0.001 0.059II (0.0087) -0.027||(0.0061) -0.01311(0.0058) -0.01811(0.0045) Following withdrawal <0.001 -0.02311(0.0010) 0.008 (0.0069) 0.033II (0.0066) -0.018# (0.0051) Difference between prescribers of zomepirac and prescribers of other NSAIDs While zomepirac on market 0.110||(0.001) -0.081||(0.0047) -0.02811 (0.0033) -0.002(0.0031) 0.001 (0.0024) Following withdrawal -0.110||(0.003) 0.06811(0.0118) 0.02111(0.0082) -0.005 (0.0078) 0.02711(0.0061) *NSAID indicates nonsteroidal anti-inflammatory drug. fBased on time-series regression models. ^Propoxyphene hydrochloride or propoxyphene napsylate. §SEs not reported for proportions <0.001. l|P<.ooi. HP<.05. #P<.01.

Fig 2, during the period zomepirac was algesic market very soon after it was shifts in choice of analgesic therapies. on the market (November 1980 through introduced, and maintained this level Primary care physicians were the pre¬ February 1983), the total preference for consistently throughout its market life. dominant prescribers ofzomepirac. When NSAIDs grew significantly in both In our large Medicaid claims database, zomepirac became available, it appears groups of prescribers. However, among zomepirac accounted for 12.4% of to have been adopted by some primary our cohort of zomepirac prescribers, we NSAID use while it was on the market, care physicians not only as a preferred estimate that NSAID use (including and 6.4% of total analgesic prescribing, NSAID, but also as an apparently safer, zomepirac) was 2.9% higher than among findings that are reasonably consistent more rational alternative to analgesics other-NSAID prescribers (95% CI, 1.9% with national figures.31"33 such as propoxyphene for treating gen¬ to 3.9%; P<.001). This difference in pref¬ Zomepirac's rapid acceptance and eral acute or chronic pain.31 When zome¬ erence was balanced by a lower use of large market share suggest an effective pirac was withdrawn, the majority of its propoxyphene analgesics (-2.8%; 95% marketing effort by McNeil. Nationally use was replaced by other NSAIDs, an CI, -2.2% to -3.4%; P<.001) among disseminated mailed warnings to pre¬ apparent gain in product safety. How¬ zomepirac prescribers. The use of other scribers in April 1982 concerning zome¬ ever, significant increases in preference opioid analgesics and barbiturates de¬ pirac's potential to cause anaphylactic for propoxyphene and fixed-combination clined significantly in both groups. reactions had no detectable effect on barbiturate analgesics were also seen. After zomepirac was withdrawn from zomepirac use in our cohort of260 zome¬ These substituted products themselves the market, over two thirds ofthe 11.0% pirac prescribers, or in the larger cohort carry notable risks ofadverse clinical con¬ of analgesic prescribing previously al¬ of 1964 physicians. One explanation for sequences, including habituation and ac¬ located to zomepirac was offset by a the warning's apparent lack of impact cidental death. relative increase in the use of other was the launch of a major promotional NSAIDs (+6.8%; 95% CI, 4.4% to 9.2%; campaign targeting zomepirac prescrib¬ Limitations of Study Methods P<.001). Despite this increase, total ers 1 week after the mailed letter. It is Several limitations ofour methods and NSAID use among zomepirac prescrib¬ possible that the apparent overall lack threats to the validity of our findings ers remained about 1.4% lower than in ofeffect ofproduct warnings, rather than must be considered. The generalizabil- the comparison group. Relative increases reflecting a stable market, may mask a ity of our results may be limited by a in two other analgesic categories offset situation in which some physicians re¬ number of factors. Our sample selection the remaining reductions in zomepirac duced prescribing due to concerns about process eliminated any physician who prescribing. Preference for pro¬ the product, while others increased pre¬ did not prescribe at least one analgesic poxyphene was 2.1% higher among scribing in response to intensified mar¬ to a member of our random sample of former zomepirac users (95% CI, 0.4% keting efforts. patients. We also observed only the Med¬ to 3.8%; P<.05), and relative use of bar¬ Regardless of the promotional cam¬ icaid portion of each study physician's biturates was 2.7% higher (95% CI, 1.5% paign's effectiveness, mailed warnings practice, and we further required that to 3.9%; P<.001). The use of other opioid about pharmaceutical products have physicians have some analgesic prescrib¬ analgesics rose significantly in both been shown to be very weak behavior- ing for the entire duration of the study. groups during this period. change interventions in comparison with The relatively small sizes of the Med¬ more icaid for made COMMENT direct, person-to-person methods.1 practices many physicians We previously reported a similar lack of analysis at the physician level impos¬ and the Market Zomepirac Analgesic effect of repeated warnings by the US sible. However, the distribution of use We are not aware of any well-designed Food and Drug Administration and phar¬ of individual NSAIDs in our analyses studies that have investigated the in¬ maceutical manufacturers aimed at re¬ closely approximates the patterns ob¬ tended and unintended impacts of with¬ ducing misuse of propoxyphene analge¬ served in a representative national drawal of a drug by its manufacturer sics due to their risk of toxic effects and sample of computerized pharmacies.31 due to unexpected and serious adverse habituation.21 There is no reason to believe that the drug reactions.30 Zomepirac was an in¬ There is ample evidence that zomepi¬ prescribing habits of our study cohort teresting product to study because it rac's entry into and exit from the mar¬ differ significantly from the habits of captured a substantial share of the an- ketplace was accompanied by significant physicians who did not meet the require-

Downloaded From: http://jamanetwork.com/pdfaccess.ashx?url=/data/journals/jama/9819/ by a Harvard University User on 02/22/2017 eluding a comparison group of other- or voluntary withdrawal of pharmaceu¬ NSAID prescribers. However, we had tical products thought to be unsafe or no control over which primary care phy¬ ineffective.4 Although limited, the best sicians were classified as zomepirac pre¬ available data suggest that the quality of scribers or other-NSAID prescribers. prescribing can either increase or de¬ Our cohorts of zomepirac prescribers crease following the removal of specific and other-NSAID prescribers differed medications from a national formulary35 slightly at baseline in rate of analgesic or the withdrawal of Medicaid reimburse¬ use in their practices. Other-NSAID pre¬ ment for scientifically unsubstantiated scribers may also differ in other ways; therapies.4 One study reported no reduc¬ for example, they may be less likely to tion in risk of self-poisoning when dan¬ try new analgesic products, or they may gerous products implicated in drug abuse see a different mix ofpatients and, there¬ were removed from the national market, fore, use different medications. How¬ primarily because of increased abuse of ever, although differences between these drugs ofequal or greater toxicity.17 Thus, two groups are likely to exist, the over¬ suboptimal product substitution follow¬ all similarity ofpatterns of analgesic use ing well-intended regulatory initiatives before zomepirac was first marketed, seems to be a generalized phenomenon the stability of prescribing throughout that policymakers must consider when¬ the study period, and the visible dis¬ ever restrictions on drug availability are continuities in use of specific drugs fol¬ proposed. lowing the changes in the marketplace Risk trade-offs due to product sub¬ give credibility to the measured substi¬ stitutions can occur not only when drugs tution effects. are banned or withdrawn, but when¬ Could the observed changes in drug ever prescriber choices are constrained, use be explained by other changes in the whether by economic or administrative marketplace? Divergence of the utiliza¬ limits. Even less drastic interventions, tion trends for zomepirac and other- such as audited triplicate prescription NSAID prescribers immediately follow¬ forms for benzodiazepines, have been ing the market entry of zomepirac and associated with undesirable substitution the sudden convergence oftrends in cer¬ effects.36 Decision makers should care¬ tain substitute categories following the fully consider the relative risks of com¬ market withdrawal of make products when determining phar¬ for zomepirac peting Fig 2.—Changes in preference selected catego¬ it that fac¬ maceutical coverage in all set¬ ries of analgesics among prescribers of zomepirac very unlikely unrecognized policies sodium (n=260) and prescribers of nonsteroidal tors were responsible for these substi¬ tings that maintain formularies or anti-inflammatory drugs other than zomepirac tution effects. otherwise restrict access to specific (n=308) both following the release of zomepirac Because of the retrospective nature products, from major federal programs after it was withdrawn from the market (top) and of the and limitations on available and the De¬ (bottom). The drugs marked with an asterisk are study (eg, Medicare, Medicaid, analgesic preparations containing propoxyphene data, it is impossible to assess certain partment of Veterans Affairs) to indi¬ hydrochloride, another opioid, or a barbiturate. characteristics of the study physicians vidual health organizations. that might influence their reactions to This study examined a product to ment for prescribing activity in Medic¬ product withdrawals. For example, we which significant, dramatic, and imme¬ aid, or that their prescribing habits know little about case-mix profile, spe¬ diate health risks were ascribed. Ap¬ would vary for the larger, non-Medicaid cific motivations for choosing zomepirac proximately 40 deaths were attributed portion of their practices. vs alternative analgesic products, or the to anaphylactic reactions to zomepirac We were only able to obtain data on influence of patient requests for par¬ in the Food and Drug Administration's analgesic prescribing for 4 months prior ticular types of analgesic. To truly un¬ Sentinel Reporting System in the 28 to the release ofzomepirac, and 6 months derstand the dynamics of prescribing months it was on the market37; a revised following its withdrawal from the mar¬ behavior, one would have to character¬ Food and Drug Administration estimate ket. Consequently, longer-term trends ize prospectively these and other clini¬ totaled 14 zomepirac-related deaths.10 in the prescribing of competing analge¬ cal and nonclinical factors.34 Based on data from 1982 and early sic of as¬ 1983,31·32 about 15 million prescriptions products, independent changes Product Risks sociated with zomepirac's entry into and Understanding for zomepirac were written while it was and Benefits in Context withdrawal from the market, were dif¬ on the market. Taken together, these ficult to detect. However, because ofthe Some drugs have substantial risks that estimates indicate a total risk of 0.9 to rapid rise in the popularity of zomepirac are tolerated because oftheir established 2.7 deaths per million zomepirac pre¬ and its sudden withdrawal, short-term benefits in relation to available thera¬ scriptions. Thus, substitution of other changes in prescribing habits were peutic alternatives (eg, zidovudine). NSAIDs, which have a far lower risk of clearly discernible. However, even less toxic drugs do not anaphylaxis,15 for zomepirac may rep¬ Other methodological problems arise always prove safe or effective for every resent safer prescribing. from the quasi-experimental nature of individual. From the epidemiologie However, propoxyphene and barbi¬ the study design. Interrupted time-se¬ rather than the clinical perspective, pre¬ turate-containing analgesics, which we ries analysis, the strongest statistical scribing following a regulatory change have shown to be competing therapeu¬ technique for analysis ofnonexperimen- or product withdrawal becomes rela¬ tic substitutes for zomepirac, also carry tal data, may not detect subtle changes tively safer if the combined statistical risks. For example, it has been estimated in trend, and cannot conclusively deter¬ risk of all products prescribed within a that there are approximately 50 deaths mine the reasons for observed changes. therapeutic category decreases. per million propoxyphene prescriptions, We compensated for this deficit by in- Recent examples exist of the banning about 40% of which were attributable to

Downloaded From: http://jamanetwork.com/pdfaccess.ashx?url=/data/journals/jama/9819/ by a Harvard University User on 02/22/2017 overdose during our observation pe¬ mum, this study highlights a need for passively, to forced changes in their riod.38-40 Propoxyphene and barbiturates more effective communication to medi¬ range of therapeutic options. The de¬ are also associated with other nonfatal cal practitioners before or during prod¬ gree of improvement in quality of care adverse outcomes, including habitua- uct withdrawals or regulatory changes through drug-restriction policies de¬ tion.21 However, these drugs' risks are to alert them to both the drug removal pends on the characteristics of available less dramatic than zomepirac's, are less and preferred therapeutic alternatives.4 therapeutic alternatives, the perceptions likely to be reported, and are therefore Since mailed circulars appear to be in¬ of prescribers and patients toward those less visibly attributable to the products effective means to accomplish these alternatives, and the degree to which in question. tasks, other strategies need to be ex¬ timely education and promotional efforts We did not have access to morbidity plored, perhaps using credible opinion can influence prescribing. At the very and mortality data necessary to conduct leaders or local medical associations as least, a growing body of literature sug¬ a reliable population-based risk analy¬ communication channels. gests thatprivate and public policymak¬ sis. However, the key policy lesson to be In summary, one cannot evaluate the ers should identify likely effects of ra¬ learned goes beyond the important is¬ impact of withdrawing a given drug as¬ tional and irrational product substitu¬ sue of whether zomepirac's withdrawal sociated with serious adverse drug re¬ tion before banning specific agents in from the market was associated with an actions except in relation to the risks order to characterize, and hopefully pre¬ increase or decrease in drug-related mor¬ posed by substituted products. Although vent, unintended outcomes. tality. How could market withdrawal may assume that with¬ policymakers This study was supported in part by grant HS have been better accomplished? Edu¬ drawing a drug eliminates its entire bur¬ 05554 from the Agency for Health Care Policy and cating prescribers about appropriate den ofrisk, this is certainly not the case. Research, Rockville, Md; and grants from the Danish pain management remains a long-term This study underscores the need for poli¬ International Development Agency, Copenhagen, for to consider risk in context and Denmark, to the International Network for the Ra¬ strategy decreasing analgesic risks, cymakers tional Use of Drugs, Boston, Mass; and clinical inves¬ but such education is time-consuming to anticipate unintended drug substitu¬ tigator award K08 AG00510 from the National In¬ and difficultto operationalize.1 At amini- tions. Physicians respond actively, not stitute on Aging, Bethesda, Md (Dr Gurwitz). References 1. Soumerai SB, McLaughlin TJ, Avorn J. Improv- state pharmaceutical benefit programs in im- 31. Baum C, Kennedy DL, Forbes MB, Jones JK. ing drug prescribing in primary care: a critical analy- proving medication utilization. Health Aff Mill- Drug Use in the United States\p=m-\1982:Fourth An- sis of the experimental literature. Milbank Q. 1989; wood. 1990;9:36-54. nual Review. Washington, DC: Division of Drug 67:268-317. 17. Ekeberg 0, Jacobsen D, Flaaten B, Mack A. Experience, National Center for Drugs and Bio- 2. Soumerai SB, Ross-Degnan D, Spira J. The ef- Effect of regulatory withdrawal of drugs and pre- logics, Food and Drug Administration, US Dept of fects of professional and media warnings about the scription recommendations on the pattern of self\x=req-\ Health and Human Services; December 1983: association between aspirin use in children and poisonings in Oslo. Acta Med Scand. 1987;221:483\x=req-\ 25-45. Reye's syndrome. Milbank Q. 1993;71:217-252. 487. 32. Baum C, Kennedy DL, Forbes MB. Utilization 3. Brahams D. Triazolam suspended. Lancet. 1991; 18. American Hospital Formulary Service. Drug of nonsteroidal anti-inflammatory drugs. Arthritis 338:938. Information 86. Bethesda, Md: American Society Rheum. 1985;28:686-692. 4. Soumerai SB, Ross-Degnan D, Gortmaker S, of Hospital Pharmacists; 1986. 33. Koch H, Knapp DE. Utilization of Analgesic Avorn J. Withdrawing payment for nonscientific 19. Drug Facts and Comparisons. Philadelphia, Pa: Drugs in Office-Based Ambulatory Care: National drug therapy: intended and unexpected effects of a JB Lippincott; 1983. Ambulatory Medical Care Survey, 1980-81. Hyatts- large-scale natural experiment. JAMA. 1990;263: 20. Shorr RI, Griffin MR, Daugherty JR, Ray WA. ville, Md: National Center for Health Statistics; 831-839. Opioid analgesics and the risk of hip fracture in the 1984;96:1-12. NCHS Advance Data Series. 5. US FDA requires PMS on UT side-effects elderly: codeine and propoxyphene. J Gerontol. 1992; 34. Schwartz RK, Soumerai SB, Avorn J. Physi- with J&J's Zomax. Scrip-World Pharm News. 47:M111-M115. cian motivations for nonscientific drug prescribing. 1980;539:13. 21. Soumerai SB, Avorn J, Gortmaker S, Hawley Soc Sci Med. 1989;28:577-582. 6. American Medical Association. AMA Drug S. Effect of government and commercial warnings 35. Ferrando C, Henman MC, Corrigan OI. Impact Evaluations. 5th ed. Chicago, Ill: American Medi- on reducing prescription misuse: the case of pro- of a nationwide limited prescribing list: preliminary cal Association; 1983:98-100. poxyphene. Am J Public Health. 1987;77:1518-1523. findings. DICP. 1987;21:653-658. 7. Zomax continues to boost McNeil's US Rx drug 22. Kaiko RF, Foley KM, Grabinski PY, et al. Cen- 36. Weintraub M, Singh S, Byrne L, Maharaj K, sales. Scrip-World Pharm News. 1981;590:10. tral nervous system excitatory effects of meperi- Guttmacher L. Consequences ofthe 1989 New York 8. US broker outlines regulatory status ofJohnson dine in cancer patients. Ann Neurol. 1983;13:180\x=req-\ State triplicate benzodiazepine prescription regu- & Johnson compounds. Scrip-World Pharm News. 185. lations. JAMA. 1991;266:2392-2397. 1981;591:9. 23. Shochet RR, Murray GB. Neuropsychiatric tox- 37. Company hopes rewriting label will clear drug. 9. Samuel SA. Apparent anaphylactic reaction to icity of meperidine. J Intensive Care Med. 1988;3: Washington Post. September 27, 1983:D-12. zomepirac (Zomax). N Engl J Med. 1981;304:978. 246-252. 38. National Institute on Drug Abuse. Statistical 10. Weiser B, Walsh E. Drug firm strategy: avoid 24. Rall TW. Hypnotics and sedatives: ethanol. In: Series: Data From the Drug Abuse Warning trial, ask secrecy. Washington Post. October 25, Gilman AG, Rail TW, Nies AS, Taylor P, eds. Good- Network (DAWN) Annual Data 1981. Rockville, 1988:A-1. man and Gilman's The Pharmacological Basis of Md: Division of Epidemiology and Statistical 11. Smith VT. Anaphylactic shock, acute renal fail- Therapeutics. Elmsford, NY: Pergamon Press Inc; Analysis, National Institute on Drug Abuse. US ure, and disseminated intravascular coagulation: 1990:358-364. Dept of Health and Human Services, Public suspected complications of zomepirac. JAMA. 1982; 25. Soumerai SB, Avorn J, Ross-Degnan D, Gort- Health Service, Alcohol, Drug Abuse, and Men- 247:1172-1173. maker S. Payment restrictions for prescription drugs tal Health Administration. Series I, No. 1. 12. Ross SR, Friedman CJ, Lesnefsky EJ. Near under Medicaid: effects on therapy, cost and equity. 39. National Institute on Drug Abuse. Statistical fatal bronchospasm induced by zomepirac sodium. N Engl J Med. 1987;317:550-556. Series: Annual Data 1983: Data From the Drug Ann Allergy. 1982;48:233-234. 26. SAS Institute. Statistics. In: SAS User's Guide. Abuse Warning Network (DAWN). Rockville, 13. Zomax' immediate withdrawal by McNeil. Wkly Cary, NC: SAS Institute; 1982. Md: Division of Epidemiology and Statistical Pharm Rep. 1983;32:2-3. 27. Rao P, Miller RL, eds. Applied Econometrics. Analysis, National Institute on Drug Abuse. US 14. Strom BL, Carson JL, Morse ML, West SL, Belmont, Calif: Wadsworth Publishing Co; 1971. Dept of Health and Human Services, Public Soper KA. The effect of indication on hypersensi- 28. Gortmaker SL, Clark CJG, Graven SN, Sobol Health Service, Alcohol, Drug Abuse, and Men- tivity reactions associated with zomepirac sodium AM, Geronimus A. Reducing infant mortality in tal Health Administration. Series I, No. 3. and other nonsteroidal anti-inflammatory drugs. rural America: evaluation of the Rural Infant Care 40. National Institute on Drug Abuse. Statistical Arthritis Rheum. 1987;30:1142-1148. Program. Health Serv Res. 1987;22:91-115. 15. Corre KA, Spielberg TE. Adverse drug reac- 29. Gillings D, Makuc D, Siegel E. Analysis of in- September 1982. Rockville, Md: Division of Epid-¬July\x=req-\e tion processing in the United States and its depen- terrupted time series mortality trends: an example iology and Statistical Analysis, NationalIns dence on physician reporting: zomepirac (Zomax) to evaluate regionalized perinatal care. Am J Pub- tutei- on Drug Abuse. US Dept of Health and Human as a case in point. Ann Emerg Med. 1988;17:145\x=req-\ lic Health. 1981;71:38-48. tServices, Public Health Service, Alcohol, Drug 149. 30. Marshall E. Guilty plea puts Oraflex case to Abuse, and Mental Health Administration. Series 16. Soumerai SB, Ross-Degnan D. Experience of rest. Science. 1985:229:1071. G, No. 12.

Downloaded From: http://jamanetwork.com/pdfaccess.ashx?url=/data/journals/jama/9819/ by a Harvard University User on 02/22/2017