Withdrawing Payment for Nonscientific Drug Therapy

Little is known about the effect on clinical decision making of nonreimbursement for ineffective medical technologies. Using a time-series design, we studied the effects of cessation of government payment for 12 categories of drugs of questionable efficacy (Drug Efficacy Study Implementation drugs) in a random sample of the New Jersey Medicaid population (N = 390 465) and in four cohorts of regular users of these products. We measured changes in the overall levels of prescriptions, expenditures, and physicians' use of substitute drugs. Although withdrawn drugs accounted for 7% of prescriptions in the base year, there was no measurable reduction in overall drug use or expenditures after the regulation; prescription rates actually rose from 0.86 to 1.00 monthly prescriptions per enrollee throughout the 42-month study. Controlling for preexisting trends, an estimated drop in the use of study drugs of 21.7 prescriptions per 1000 enrollees per month was offset by an increase in the use of substitute drugs of 33.7 prescriptions. Both desirable and unimproved therapeutic substitutions were observed. Used alone, curtailment of reimbursement for marginally effective therapies results in both desirable and unintended clinical substitutions and may not reduce costs. Supplementing such restrictions with education may be necessary to promote practices that are more therapeutically and economically appropriate.


Withdrawing Payment for Nonscientific Drug Therapy
Intended and Unexpected Effects of a Large-Scale Natural Experiment Stephen B. Soumerai, ScD; Dennis Ross-Degnan, ScD; Steven Gortmaker, PhD; Jerry Avorn, MD Little is known about the effect on clinical decision making of nonreimbursement for ineffective medical technologies. Using a time-series design, we studied the effects of cessation of government payment for 12 categories of drugs of questionable efficacy (Drug Efficacy Study Implementation drugs) in a random sample of the New Jersey Medicaid population (N = 390 465) and in four cohorts of regular users of these products. We measured changes in the overall levels of prescriptions, expenditures, and physicians' use of substitute drugs. Although withdrawn drugs accounted for 7% of prescriptions in the base year, there was no measurable reduction in overall drug use or expenditures after the regulation; prescription rates actually rose from 0.86 to 1.00 monthly prescriptions per enrollee throughout the 42-month study. Controlling for preexisting trends, an estimated drop in the use of study drugs of 21.7 prescriptions per 1000 enrollees per month was offset by an increase in the use of substitute drugs of 33.7 prescriptions. Both desirable and unimproved therapeutic substitutions were observed. Used alone, curtailment of reimbursement for marginally effective therapies results in both desirable and unintended clinical substitutions and may not reduce costs. Supplementing such restrictions with education may be necessary to promote practices that are more therapeutically and economically appropriate. (JAMA. 1990;263:831-839) PRESSURES continue to mount to contain costs by restricting the thera¬ peutic options of physicians. One in¬ creasingly popular but poorly studied strategy for raising the cost-effective-ness of clinical decisions is for govern¬ ment and private payers of health care to eliminate reimbursement for "mar¬ ginally effective" or "irrational" health care technologies and services. Defin¬ ing marginal therapies is difficult both technically and politically, and the dan¬ ger exists that some restrictions on phy¬ sicians' choices may harm patients. 1,2 In the Medicaid program, one com¬ mon short-range, cost-control strategy has been to eliminate coverage of cer¬ tain services, including some drugs.3 A number of European and developing na¬ tions have also restricted public financ¬ ing for specific classes of nonessential medications.4,5 Despite the widespread occurrence of these reimbursement pol¬ icies, their consequences and the magni¬ tude of their savings and costs are not well understood, particularly in the of¬ fice-practice setting, where most clini¬ cal decisions occur without ongoing monitoring.6 This report presents the results of a natural experiment in which a large number of prescription drugs that were judged to be ineffective or irrational were suddenly eliminated from reimbursement through Medicaid and other public programs. Outcomes were measured using a patient-level prescription claims database that cov¬ ered a sample of 390 465 individual pa¬ tients in the New Jersey Medicaid pro¬ gram during a 42-month period.
Prescription-drug use provides a good model for research regarding the impact of payment-restriction policies. Drug prescribing is one ofthe most com¬ mon and important clinical decisions in medical practice; approximately 75% of all visits to the physician end with one or more drug prescriptions'; total nation¬ wide expenditures for prescription drugs were approximately $33 billion in 1987. 8 Although this represents a rela¬ tively small proportion of national health care expenditures, the overall clinical and economic impact of appro¬ priate and inappropriate drug use is substantially higher. 9 This investigation was designed to extend and improve on previous studies in several ways. We used a large sample of 390 465 Medicaid patients to achieve stable estimates of drug use and studied 12 broad categories of DESI drugs, 66 potential substitutes, and 3 comparison drug categories to achieve greater va¬ lidity and generalizability of the find¬ ings. Preexisting trends were con¬ trolled for in all analyses and all substitute categories were defined in advance to reduce bias. Identifying large preintervention cohorts of DESI drug users made it possible to achieve maximum power and control for other confounders. The analysis attempted to answer the following questions: Did the reimbursement restrictions on DESI drugs reduce overall drug use and costs in the Medicaid system? What was the nature of substitute prescribing? Did specific substitution effects represent clear improvements in therapy relative to nonreimbursed drugs? Were patients who were taking combination drugs switched to effective single-agent ther¬ apies? What were the relative costs of withdrawn products vs their identified substitutes? These findings may make it possible to understand both the oppor¬ tunities and limits of restricting physi¬ cians' choices through regulated reim¬ bursement.

Study Populations and Data Sources
The population studied included all persons who were eligible for the New Jersey Medicaid program during the pe¬ riod from July 1980 to December 1983. A 40% random sample of recipients who were enrolled during this period was selected, yielding a study population of 390 465 individuals. Many members of this study population were not enrolled in Medicaid continuously throughout the study period because of fluctuations in income, family status, or other el¬ igibility criteria. To calculate rates of pharmaceutical use correctly, the monthly eligibility status and demo¬ graphic characteristics of all Medicaid recipients were ascertained; the actual eligible population during each study month was then calculated. To inves¬ tigate whether possible secular changes in the composition of the study popu¬ lation could be responsible for some of the observed changes in pharmaceutical use, we examined trends in the distrib¬ ution of age, sex, and race in New Jersey Medicaid. To further control for changes in eligibility, we also observed four cohorts of regular users of parti¬ cular DESI drugs (see later herein).
We analyzed data regarding all pre¬ scriptions actually filled and reim¬ bursed by Medicaid in the study popula¬ tion during the 42-month study (17 months before and 25 months after the change in reimbursement policy). These data included the recipient identifier, the drug product code ofthe medication, the number of units dispensed, and the date the prescription was filled. Our own work and that of others using drug claims data from Medicaid programs have found them to be highly reliable and complete.2,18'23 Selection and Classification of Study Drugs A panel of six clinicians including in¬ ternists, geriatricians, and pharmacists was convened to select DESI drugs that were suitable for study, to review the indications for which they were used, and to define all plausible substitute therapies and their relative efficacy for specific indications. The panel members were provided with appropriate refer¬ ence material throughout this process, including the National Academy of Sciences/National Research Council re¬ port12 and other background informa-tion regarding DESI drugs and sections of the American Medical Association's Drug Evaluations2* for both DESI drugs and plausible substitute thera¬ pies. The selection and evaluation pro¬ cess was carried out first by question¬ naires completed independently, fol¬ lowed by five group sessions to build consensus about the specific drug sub¬ stitutions. To avoid bias, the entire pro¬ cess was completed before the results of data analysis were known.
The DESI drugs were initially se¬ lected for study according to the follow¬ ing criteria: (1) the drug group should be widely prescribed before the policy, (2) the drug should have relatively specific indications, (3) if possible, there should exist other therapies of varying efficacy for these indications. Using these crite¬ ria, 12 DESI drug categories were cho¬ sen, which represented a broad range of both acute and chronic health problems: peripheral or cerebral vasodilators, asthma and sedative combinations, gas¬ trointestinal antispasmodics with seda¬ tives, analgesic combinations (antimi¬ graines), combination steroid-antibiotic creams and ointments, ineffective antiemetics (trimethobenzamide), analge¬ sic and sedative combinations, phenylbutazone-antacid combinations, nitrate and meprobamate combinations, di¬ uretic and potassium combinations, ce¬ rebral stimulants, and antibiotic combinations.
Several drug compendiums2527 were used to identify all marketed products chemically equivalent to the DESI drugs chosen and all the defined substi¬ tutes for each category. The panel also evaluated each substitute relative to the DESI drug it replaced in terms ofits potential for improving therapy as (1) likely, (2) uncertain, or (3) unlikely. These ratings were based on both effica¬ cy and safety. In addition to rating ther¬ apeutic improvement, panel members also estimated the proportion of use of each DESI group by indication that each substitute would likely replace. Of course, not all substitutions for the DESI products studied will necessarily be identified by our procedures because of the use of DESI drugs for indications other than the ones defined or substitu¬ tion of other drugs. Thus, our estimates of substitution effects are conservative.

DESI-USER COHORTS
Certain DESI drugs had marketed indications for treatment of chronic con¬ ditions (eg, peripheral or cerebral va¬ sodilators and bronchodilator plus se¬ dative combinations).
Identifying cohorts of regular users ofthese medica¬ tions allowed patient-level specificity in the identification of substitutes and greater control for changes in the com¬ position of the overall study group. To minimize regression toward the mean, the "long-term users" were required to have filled at least one prescription dur¬ ing each4-month period for at least 16 months before the DESI withdrawal policy.
Time-series of drug use were con¬ structed by aggregating by month all prescriptions for each category of DESI drug group, substitute medications, and comparison drugs in the full sample population, as well as in each group of DESI drug users. Four of the 12 DESI categories were not prescribed with enough frequency to allow reliable esti¬ mates of change (nitrate and mepro¬ bamate combinations, diuretic and po¬ tassium combinations, cerebral stimulants, and antibiotic combina¬ tions), so these drugs and their substi¬ tutes were not analyzed individually, but were included in group totals. Dol¬ lar values were calculated by assigning to each prescription the allowable perunit cost from the New Jersey formu¬ lary, plus the dispensing fee. All series were converted to rates by dividing each month's total by the number of persons enrolled in that cohort for that month. The resulting time-series were analyzed by specifying a segmented lin¬ ear regression model with correction for serially autocorrelated observations.28 This class of models is described in stan¬ dard econometrics textbooks,29 and its applicability to health problems has been demonstrated.2,30,31 The basic model included terms to es¬ timate the following variables: preexist¬ ing prescribing level for each drug group in the first month of the observa¬ tion period (intercept), trend in pre¬ scribing before implementation of the DESI policy, change in level of prescrib¬ ing attributable to the policy, and change in prescribing trend after the policy. Not all patients fill a prescription each month, so after withdrawal of the DESI products there was generally a brief transition period of 2 to 3 months until patients were placed on a new regi¬ men (Figs 1 and 2). For this reason, the month in which the policy was imple¬ mented and the following 2 months were excluded from the statistical models but are included in all figures.
To derive conservative estimates of the impact of the withdrawal on substi¬ tute drug use, we interpreted only sud¬ den and significant discontinuities in the levels of the time-series as likely to be true effects ofthe intervention. The two terms for trend before and after the policy were always included as covar¬ iates in the models to ensure that more modest changes following the DESI withdrawal, which could have been caused by many other gradually chang¬ ing historical factors (eg, marketing patterns), were not included as princi¬ pal effects of the intervention.
Many drugs exhibit marked seasonal Segmented regression lines were fit to the preintervention and postIntervention data as described in the text. Dotted line represents the time of the withdrawal of the DESI drugs studied. scriptions dropped to almost zero imme¬ diately following the intervention; most of the remaining DESI drugs were withdrawn in January 1983. Despite the sharp drops in the num¬ ber of DESI prescriptions filled, overall drug use rates actually rose modestly throughout the observation period, from approximately 860 monthly pre¬ scriptions per 1000 enrollees to 998 pre¬ scriptions by the end of the 42-month period (Fig 1, center). The best esti¬ mate of the change in prescription use associated with the intervention was ac¬ tually an increase of 45.1 prescriptions per 1000 enrollees per month (SE = 32.6), indicating that there was no observable reduction in total drug exposure following the DESI drug withdrawals. Similarly, no reduction in total drug use was observed after with¬ drawal of another large group of DESI drugs in January 1983. The likelihood that a modest proportion of recipients continued to pay for DESI drugs out-of-pocket14 further supports the hypothe¬ sis that total drug use did not decline and may even have risen slightly.
The bottom graph in Fig 1 suggests  no increases in levels of use occurred for any of these products following the re¬ striction policy. Therapeutic Substitution Among DESI Drug Users We next analyzed which medications were chosen to replace specific DESI drugs and their relative therapeutic ef¬ ficacy in four DESI drug user cohorts. Significant increases occurred in 10 sub¬ stitute categories: 2 of these represent¬ ed probable improvements in therapy, 3 substitutions were indeterminate, and 5 represented unlikely improvements in therapy. Unless otherwise stated, all level-change estimates are in prescrip¬ tions per 100 patients per month. All effects presented were significant at the P=.05 level and, in many cases, P<.0001.

Patients Who Took Peripheral or
Cerebral Vasodilators.-The largest group of users of a single DESI drug group comprised patients (n = 468) who received long-term treatment with pe¬ ripheral or cerebral "vasodilators" such as cyelandelate (eg, Cyclospasmol) or nylidrin (eg, Arlidin). The large major¬ ity of this population were older than 60 years (90%) and women (76%). Overall, 69% of recipients of vasodilators did not receive any ineffective substitute drugs through Medicaid after the policy change. However, large increases in use were found for two drugs (Fig 2 (Fig 2). Prescription rates for nonsedative-containing theophylline products rose from 34.4 prescrip¬ tions to 57.5 prescriptions per 100 pa-tients per month ( + 67%, Table 1); this use rate continued to rise slightly throughout the observation period. Conversely, although it was hypothe¬ sized that physicians might prescribe single-agent barbiturates or other seda¬ tives (eg, benzodiazepines) in an at¬ tempt to reconstitute the irrational combination of therapies, this effect did not occur, except for a modest increase in the use of hydroxyzine (Atarax, Vistaril) in the treatment of these patients.
The data indicate that most, but not all, patients with asthma and chronic ob¬ structive pulmonary disease continued to receive an effective respiratory agent. Patients Who Took Antispasmodic Agents That Contained Sedatives.-This category contained primarily anticholinergic agents combined with phé¬ nobarbital (eg, Donnatal Extentabs) or with chlordiazepoxide (eg, Librax). Two-hundred forty-nine patients in the sample were long-term users, 88% were women. The largest absolute substitu¬ tion effect was for single-agent benzodiazepine tranquilizers (Fig 2 and Table  1). Use of other belladonna-type prepa¬ rations increased by 113% from the ex¬ pected level of 5.7 prescriptions per month. Some barbiturate-anticholinergic combinations remained available, but physicians did not increase their prescribing of these agents.

Patients Who Took Combination
Analgesics.-This group contained various combination analgesics with or without sedatives, often prescribed for migraine (eg, Migrai). Only 38 regular users of these drugs could be identified in the 40% sample. Although effective alternative therapies existed at the time of the study, the largest increases occurred for other combination analge¬ sic products that contained constituents that were abusable and of modest effica¬ cy, such as propoxyphene, pentazocine, and barbiturates (  Table 2). Use of prochlorperazine, probably an improvement in therapy, increased by 34%; use of metoclopramide and promethazine also increased by 162% and 30%, respectively. The analgesic and sedative combina¬ tions and phenylbutazone and antacid combinations that were withdrawn were judged to have similar substitute therapies and were, therefore, analyzed together (Table 2). Following the cessa¬ tion of reimbursement for these drugs, the largest absolute increase in pre¬ scription use occurred for the nonsteroidal anti-inflammatory drugs ( + 2.1 pre¬ scriptions [ + 5.7%]), even after con¬ trolling for a rising preintervention trend. However, significant substitu¬ tion effects occurred for less-desirable substitute therapies such as analgesic combinations that contained pentazocine or propoxyphene ( + 0.4 prescrip¬ tions [ + 5.9%]). The most common re¬ sponse to the ending of reimbursement for a combination of phenylbutazone and a small dose of antacid (Butazolidin Alka) was for physicians to prescribe phenylbutazone alone, although it was probably the least desirable nonsteroidal anti-inflammatory drug available at the time (Table 2).24,pl 067) No increases were observed for the other seven cate¬ gories of analgesics that were followed up ( Table 2).
Many of the DESI combination prod¬ ucts promoted as analgesics or antispas¬ modic agents contained abusable seda-tives such as a barbiturate or me¬ probamate. 24<p%-") As a result, it is not surprising that the use of some psychoactive drugs ( Table 2) increased follow¬ ing the intervention. The most preva¬ lent substitution effect in these patients was for the benzodiazepines ( + 3.9 pre¬ scriptions [ + 10%]). However, there was also an increase in the level of use of single-agent barbiturates ( + 0.7 pre¬ scriptions [ + 8.6%]), which occurred af¬ ter a prior downward trend in their use.24"**"*"

Changes in Medicaid Drug Expenditures
Based on constant (1986) drug prices, the estimated New Jersey Medicaid ex¬ penditure for all DESI drugs during the base year was $4.5 million. No decrease in drug expenditures was found follow¬ ing removal of DESI drugs from the Medicaid formulary. In fact, taking into account preexisting trends, the best es¬ timate of the change that occurred with the DESI intervention in total Medicaid drug expenses per 1000 enrollees was a 3.2% increase of $394 per month, al-  ttCombinations of aspirin or acetaminophen with pentazocine, propoxyphene, or butalbital. ttPentazocine or propoxyphene alone. though this was not significantly differ¬ ent from zero (90% confidence interval, -$115 to +$904). This was mainly caused by the rise in substitute pre¬ scriptions and the higher cost of some of these newer agents. For example, the estimated decrease in monthly costs for study DESI drugs per 1000 enrollees was $308 (SE =$12), while the increase in monthly costs for all defined substi¬ tute drugs was $455 (SE = $130).

Predictability of Individual Substitution Effects
To determine how well specific sub¬ stitution effects were predicted by the expert panel, we compared predicted use of substitute drugs with actual sub¬ stitution patterns observed. For each substitute drug, we measured the in¬ crease observed in prescribing of the substitute drug divided by the reduc¬ tion in prescribing ofthe DESI category to which it applied. This was then com¬ pared with the panel's prediction of the likelihood of substitution for each drug. Among the 26 drugs predicted to re¬ place 10% or more of DESI drug use, 13 actually were substituted at this rate, and the others at a lower rate. Among the 22 drugs predicted to replace less than 10% of DESI use, 19 were indeed substituted at this rate, with the re¬ maining 3 drugs substituted at a higher rate.

COMMENT
The impact of various types of formu¬ lary restrictions on the prescribing be¬ havior of physicians has been a subject of continuing controversy. Few objec¬ tive data have been reported, particu¬ larly in the office-practice setting, where most prescribing occurs. The re¬ sults presented previously support the hypothesis that such strategies used alone do not necessarily reduce overall drug use or costs. In fact, our data indi¬ cate widespread increases in the use of replacement therapies that, in the ag¬ gregate, approximately equalled the costs saved through the reduction in use of DESI drugs. There is substantial evi¬ dence for physician use ofboth desirable and unimproved substitute therapies following the reimbursement changes.
Examples of unlikely improvements in therapy included the substitution of papaverine and ergoloid mesylates for pe¬ ripheral vasodilators or the substitution of pentazocineor propoxyphene-containing agents for withdrawn analgesic combinations. On the other hand, the prescriptions of single-agent bronchodilators in place of combinations of these drugs with sedatives represents a probable improvement in the quality of care, assuming that no other negative changes in the dosing of these agents occurred. The observation that sedative prescribing did not increase substan¬ tially among patients with asthma who were previously using these combina¬ tions suggests either that physicians did not feel that the psychoactive compo¬ nent was needed or that they had not been aware of it. Some substitutions, such as the use of benzodiazepines in place of antispasmodic combinations, were difficult to evaluate without more specific diagnostic information; these effects were thus rated as uncertain changes in the quality of prescribing.
Because of the quasi-experimental nature of this study, it is essential to consider alternative explanations for the effects observed. Unlike previous single-observation, before-after stud¬ ies32 that did not control for preinterven¬ tion trends, we considered only sudden discontinuities in the levels of the ser¬ ies as strong evidence of substitu¬ tion effects since modest changes in prescribing trends could be caused by many other historical factors, including promotional campaigns and changing knowledge. The large number of signifi¬ cant increases in the level of use of antic¬ ipated substitute drugs, all beginning within 2 months following the reim¬ bursement cutoff, and the lack of effect observed for unrelated drug categories (eg, insulin, digoxin) at the same point in the 42-month series reinforces the validity of the causal inferences. Fur¬ thermore, the extremely stable charac¬ teristics of the Medicaid population, and the fact that the cohorts of long-term users did not include new patients, make it highly unlikely that the effects observed were caused by contempora¬ neous changes in the population of patients.
Regression artifacts were minimized by requiring that long-term users re¬ ceive DESI products fairly consistently (one prescription per 4-month period) for 16 months before the intervention.
No other drug reimbursement changes occurred in New Jersey Medicaid dur¬ ing the observation period. Finally, it is unlikely that the findings could be ex¬ plained by multiple significance tests. Approximately half of all substitute drugs showed significant increases; many of these effects were significant beyond the P<.0001 level.
As in most other investigations of this type, this study raises as many ques¬ tions as it was designed to answer. For example, while some out-of-pocket pur¬ chases of DESI drugs were possible," especially for inexpensive or short-term therapies, the precise level of these un¬ desirable effects remains unknown.
However, in a program such as Medic¬ aid, serving very poor patients, many physicians would be likely to select an¬ other reimbursable drug from the ap¬ proved list, rather than count on the patient to pay out-of-pocket for an unreimbursed drug.33 The extent to which certain patients may have increased their use of other nondrug health ser¬ vices following the withdrawal of medi¬ cations with perceived effectiveness, especially combination drugs that con¬ tained sedatives, is an interesting topic for further research. In general, since the withdrawn agents were judged by experts to lack evidence justifying their continued marketing, we focused on prescribing rather than patient out-comes. Nonetheless, further study of patient compliance and use of other health services following changes in their regimens is warranted. Previous reports of the effects of formulary re¬ strictions on such outcomes are uninterpretable because of methodological inadequacies. 34,35 It has been noted that most adverse drug reactions are "predictable and pre¬ ventable through logical application of existing information."36 The same may be said of the unintended effects of poli¬ cy interventions. In several meetings held before the data were analyzed, a small group of physicians and pharma¬ cists were able to predict many substi¬ tution effects. The public policy implica¬ tion of this finding is that it would be quite possible to interview office-based prescribers to learn how physicians are likely to react to planned regulatory in¬ terventions. With advance knowledge of potential unintended substitutions, proposed regulations could be modified and educational programs developed to help direct alternative practices in an optimal direction. Failure to consider these predictable substitute behaviors is a common blind spot of many regula¬ tory programs.9 This study also makes clear the need to learn more about why physicians choose "nonscientific" treatments in the first place and to develop more effective and appropriately targeted quality as¬ surance policies. If patient or family de¬ mands and placebo effects are impor¬ tant factors that drive prescribing, it may be possible to develop office-man¬ agement policies and educational mate¬ rials to help physicians reduce the use of certain treatments when they are truly not needed.3M9 If lack of knowledge of appropriate treatment choices is the main issue, physicians may simply need objective advice from credible educa¬ tional sources to improve their prescrib¬ ing. We have previously shown that face-to-face educational visits by spe¬ cially trained "academic detailers" could reduce physician prescribing of three targeted drug groups without adverse substitution effects. 19 Clinically appro¬ priate substitution effects were ob¬ served, and the overall benefit-cost ra¬ tio was high, even without considering improved quality and safety of care.21 Simple restriction policies are more sus¬ ceptible to inappropriate substitution effects because they usually do not ad¬ dress physicians' and patients' per¬ ceived needs for the drug. A promising avenue for further research is to com¬ bine such restriction policies with edu¬ cational outreach programs to encour¬ age use of appropriate replacement therapies.
The applicability of these results to programs of hospital formulary restric¬ tions is less clear. In these settings, where physicians and pharmacists are involved in pharmacy and therapeutics committees, and where programs of inservice education and administrative control often accompany restricted drug lists," potential unintended substi¬ tutions may be circumvented. 41,42 It is possible that these lessons may generalize to other kinds of physician decision making. For example, physi¬ cians have been found to "creatively" adjust to restrictions that prevent the ordering of a battery of laboratory tests with a single order by simply ordering all of the same tests individually.6 In addition, it has been hypothesized that the banning in some hospitals of singleunit blood transfusions (which often provide minimum benefits relative to risk) may have resulted in their auto¬ matic conversion in many cases to 2-U transfusions, circumventing the point of the restriction. 43 In the international health arena, our results may also be informative for gov¬ ernment programs such as those in Great Britain and Germany, which have recently implemented "negative drugs lists," ending public reimbursement of remedies such as treatments for coughs and colds.4 It is tempting to generalize these findings to more restrictive regu¬ lations as well, such as Australia's Phar¬ maceutical Benefits Scheme, which ex¬ cludes particularly costly drugs in addition to ineffective ones from public reimbursement, and the Essential Drug Programs being encouraged by the World Health Organization for develop¬ ing countries." While the need to consid¬ er physicians' substitution behaviors and motivations for practice are clearly relevant, more definitive data are need¬ ed regarding the overall effects of these very different approaches to the costs and quality of care.
The findings presented herein under¬ line both the opportunities and the lim¬ its of restricting physicians' preroga¬ tives to prescribe scientifically un¬ substantiated therapies. Although cost savings were not achieved by federal payment restrictions for a group of such drugs, the frequent substitution of effi¬ cacious therapies probably represent¬ ed, at the margin, an overall improve¬ ment in quality of care. Medications, like many other medical treatments, are not risk free; the elimination of unneces¬ sary drugs, to the extent to which this occurred, probably reduced the risk of iatrogenic illness. Yet even if all ineffec¬ tive therapies could be eliminated from medical practice, there is ample evi¬ dence to suggest that the remaining "ef-fective" therapies would be misused as well9 because of misinformation and oth¬ er factors that cause nonscientific pre¬ scribing in the first place. A broader strategy is needed. Physicians-in-train¬