Information from Pharmaceutical Companies and the Quality, Quantity, and Cost of Physicians' Prescribing: A Systematic Review

Geoff Spurling and colleagues report findings of a systematic review looking at the relationship between exposure to promotional material from pharmaceutical companies and the quality, quantity, and cost of prescribing. They fail to find evidence of improvements in prescribing after exposure, and find some evidence of an association with higher prescribing frequency, higher costs, or lower prescribing quality.


Introduction
Pharmaceutical companies in the United States spent about US$57.5 billion, or 24.4% of their revenue, on promotion in 2004 [1]. One estimate of total promotional expenditure in France for 2004 is J2,908 million (12.2% of revenue). However, another estimate is that pharmaceutical detailing cost J3,300 million and accounted for 75% of the overall cost of promotion in that year making promotion 17.3% of revenue [2]. Expenditure on promotion is aimed at maximizing returns for the corporation and shareholders [3]. The industry claims that promotion also provides scientific and educational information to healthcare professionals: ''Appropriate marketing of medicines ensures that patients have access to the products they need and that the products are used correctly for maximum patient benefit. Our relationships with healthcare professionals are critical to achieving these goals because they enable us to -inform healthcare professionals about the benefits and risks of our products to help advance appropriate patient use, provide scientific and educational information, support medical research and education'' [4].
There is a wide range of views amongst health professionals about pharmaceutical promotion. Qualitative studies suggest that many perceive pharmaceutical promotion to be a useful and convenient source of information [5][6][7]. Some doctors deny that they are influenced by pharmaceutical company promotion or claim that it influences others but not themselves [8][9][10]. Nonetheless, many of these physicians are willing to give significant amounts of time to engaging in promotional activities [11]. By contrast, several professional organisations have called for more control of promotional activities [12,13] because of evidence that promotion may be misleading [14][15][16][17].
The evidence base illuminating these conflicting views is growing. In 2000, Wazana identified eight studies linking pharmaceutical promotion to increased prescribing, ''nonrational prescribing,'' and increased prescribing costs [18]. A 2005 review concluded that promotion influences the prescribing by physicians in training [19], and a second review in the same year concluded that sales representatives influence prescribing [20].
Those previous reviews are now out of date, narrowly focused, or only partially assessed the relationship between information (promotional or otherwise) from pharmaceutical companies and prescribing costs and quality. The objective of this review is to examine the relationship between exposure to information directly provided by pharmaceutical companies and the quality, quantity, and cost of physicians' prescribing.

Criteria for Including Studies
Randomized controlled trials, time series analyses, before-after studies, cohort studies, case-control studies, ecological studies, and cross-sectional studies were eligible for inclusion. Studies were included if they had both a measure of exposure to any type of information directly provided by pharmaceutical companies and a measure of physicians' prescribing. We excluded studies that looked at the indirect provision of information, for example, through continuing medical education courses that were funded by unrestricted grants from pharmaceutical companies. Case series, case reports, abstracts, news items, and short reports were excluded.
Exposure to information directly provided by pharmaceutical companies was defined as including pharmaceutical sales representative visits, advertisements in journals or prescribing software, presentations from pharmaceutical companies to groups, meetings sponsored by pharmaceutical companies, mailed information including advertisements, and participation in sponsored clinical trials. We did not include studies of other forms of promotion such as gifts or samples or studies of indirect forms of information provision such as sponsored education.
The outcome measures were the quality, frequency, and costs of prescribing. We looked for additional articles in the references of each retrieved article including review articles in an iterative, exhaustive process. Efforts to find additional studies included placement of messages on email drug discussion groups, contacting experts in the field, and asking Australian subsidiaries of international pharmaceutical companies for information. All languages were considered.

Selection of Studies
The title and abstract, if available, of all articles detected by the database searches were reviewed by two authors. Articles that possibly met the inclusion criteria were retrieved and subjected to a formal inclusion process independently by two different authors. Differences of opinion were resolved by consensus and if necessary a third author was involved.

Quality Appraisal
Articles meeting inclusion criteria were appraised for methodological quality independently by two authors. Randomized studies were assessed for adequacy of randomization method, allocation concealment, blinding, follow-up, and use of intention to treat analyses [21]. Controlled cohort and case-control studies were assessed using the Newcastle-Ottawa scales [22]. For other nonrandomized studies, quality appraisal included assessment of sources of bias, for example presence of a control group, selection methods, control of confounding, response rate (.80%), and use of appropriate statistical tests [23]. Studies were only excluded from the review if two authors found there was insufficient information to appraise their quality. Disagreements were resolved by discussion with a third author.

Data Extraction
For included studies, two authors independently extracted data on study site, dates of data collection or publication, types of participants (primary care providers, specialists, and residents), study medication(s), exposure to information from pharmaceutical companies, and prescribing outcomes.

Reporting of Results
For quality of prescribing we accepted the original authors' definitions of what constituted more (or less) appropriate prescribing.
We divided studies into two groups on the basis of whether the information was delivered with or without conventional promotional techniques. This distinction was made because information delivered with versus without conventional promotion may produce different effects on prescribing.
Conventional promotional techniques were defined as advertisements (in journals and software), representatives' visits, attendance at pharmaceutical sponsored meetings, and mailed information from pharmaceutical companies. In addition, we included in this group studies looking at total promotional investment/summated scores of commercial information use/ general use of commercial sources. The other group of studies included warning letters, participation in company sponsored trials, and representatives' visits for nonpromotional purposes.
A narrative synthesis of results was undertaken following the MOOSE guidelines and meta-analysis performed where appropriate data were available (Text S1) [24]. The unit of analysis was defined as the combination of exposure to a type of information from a pharmaceutical company (for example pharmaceutical sales representative visits or journal advertisements) and a type of prescribing outcome (quality, frequency, and cost of prescribing). Thus studies were treated as a single unit of analysis if they measured the same type of exposure and the same type of outcome regardless of the number of drugs covered in each study. We classified each analysis as positive or negative rather than no association detected if the p value was less than 5% (p,0.05) regardless of the magnitude of the effect.
We reported standardized effect measures (Pearson correlation coefficients, odds ratios [ORs], or beta coefficients) where study reports provided them or the data needed to calculate them. For econometric studies, we also reported t statistics where they were reported or it was possible to calculate them.
Meta-analysis was not appropriate for the outcomes of quality of prescribing and cost of prescribing because in both cases the studies examined different exposures or outcome measures and/or lacked control groups. We undertook a meta-analysis for one component-studies of frequency of prescribing with identifiable control groups where the information exposure was delivered with conventional promotional techniques. We used ORs for change in prescribing frequency as the outcome measure. Where studies had suitable designs for inclusion in the meta-analysis but ORs and standard errors were not published we contacted corresponding authors. Out of ten studies [25][26][27][28][29][30][31][32][33][34], we received four replies of which three provided the information we required [29][30][31].
Heterogeneity was assessed using the tau squared test with a sensitivity analysis to investigate likely sources of heterogeneity. Factors identified a priori as possible explanations for heterogeneity were study design, study quality indicators, year of publication, type of exposure to pharmaceutical company information (active versus passive), and physician characteristics (level of experience and also primary care provider versus specialist). We defined active exposure as information presented to physicians at meetings or during pharmaceutical sales representatives' visits. We defined passive exposure as journal advertisements, mailed information, advertisements on clinical software, and participation in sponsored clinical trials. Studies reporting more than one unit of analysis were subjected to sensitivity analysis. Meta-analysis was performed using RevMan (version 5.0.24) with further analysis conducted using Stata version 10.0 (Stata Corporation).

General Characteristics of Studies
The most common study design was cross-sectional (24/58 studies, 41%). There were also two cluster randomized controlled trials, one controlled-cohort study, two case-control studies, 24 time-series analyses, and five before-after studies. Over half (55%) of the studies were conducted in the United States. Characteristics of included studies are outlined in Table 6.

Pharmaceutical Company Information and Prescribing Quality
Prescribing quality was measured by ten studies with 14 units of analysis [37,39,58,59,61,64,74,77,81,82] (Table 7). Quality was assessed in four distinct ways: quality scoring of prescribing decisions, guideline adherence, prescribing appropriateness of an individual drug class, and prescribing range. Three studies used quality scores calculated by coding physicians' drug choices in responses to clinical vignettes [74,81,82]. One of these used an expert panel to derive a quality index (1-100) judging primary care providers' prescribing in response to both their actual prescribing and clinical vignettes [81]. In the latter study learning about the drug first from pharmaceutical sales representatives was associated with lower quality of actual prescribing but the number of pharmaceutical sales representatives' visits was not. There was no significant association between primary care providers seeing more pharmaceutical sales representatives or first learning about the drug from pharmaceutical sales representatives and lower quality responses to case vignettes [81]. Another study combined scales examining indication, effectiveness, safety, dosage, duration, and polypharmacy to produce a seven-point scale measuring rationality of prescribing [74]. Primary care providers' selfreported reliance on pharmaceutical companies for information was associated with lower quality scores [74]. A third study used a quality score for a hypertension scenario where thiazides were considered very appropriate and all other drug groups were considered very inappropriate [82]. Self-reported rates of attendance at pharmaceutical company-sponsored meetings were associated with slightly lower quality scores, but self-reported rates of pharmaceutical sales representative visits had no significant association [82].
Residents attending a sponsored meeting were more likely than nonattending residents at the same hospital to prescribe the sponsoring company's medication, both when it was appropriate according to the authors and when it was not [39].
Primary care providers who saw more pharmaceutical sales representatives and those who used the pharmaceutical industry in general as a source of information prescribed a wider range of drugs [61]. The authors suggested that this was a sign of lower prescribing quality in the context of recommendations that primary care providers use a limited list of drugs they know well [61].
Two studies measured guideline adherence. One found less adherence by primary care providers who received more frequent visits from pharmaceutical sales representatives [64], while the other found no change in adherence by primary care providers participating in a clinical trial sponsored by a pharmaceutical company [37].
One study of warnings conveyed by pharmaceutical sales representatives and mailed information [58], one of mailed warnings alone [59], and one of representatives' visits and advertisements [77] found that there was no alteration in overall rates of prescriptions judged to be inappropriate.

Conventional Promotional Techniques
Pharmaceutical sales representative visits. Of the 29 studies of pharmaceutical sales representative visits, 17 found only an association with increased prescribing of the promoted drug [26,32,33,38,40,[43][44][45][46][47][48][49][50]63,67,78,79]. None found less frequent prescribing. Of the remaining 11, six studies had mixed results: finding a significant association with more frequent prescribing for some measures but no significant association for others [27,42,62,69,73,81]. Five did not detect any significant relationship [31,52,68,72,77]. One study did not use statistical tests for associations. It found that during the time that spending by pharmaceutical companies on promotion of a medication dropped to zero, there was also a significant drop in prescribing of that medication. However most of the decreases in promotion and prescribing occurred after the publication of evidence of problems with that medication [56].
Nine of these studies with either positive or mixed results provided insights into features of pharmaceutical sales representative visits that modified the impact of these visits on prescribing [40,46,49,62,67,69,73,78]. An association with more frequent prescribing was more likely when pharmaceutical sales representatives visited groups of physicians, when physicians had lower baseline prescribing of the promoted drug [62], and when physicians had larger prescribing volumes overall [67]. Longer pharmaceutical sales representative visits to physicians and residents were also more likely to be associated with increased prescribing [69,73]. More frequent pharmaceutical sales representative visits were associated with diminishing returns [46,50,69].
In addition to increasing the promoted drug's market share, pharmaceutical sales representative visits were associated with a decrease in the market share of competitor products [78]. Pharmaceutical sales representative visits were more likely to be associated with more frequent prescriptions for drugs judged more effective and also for drugs with more side effects [40]. However the authors of that study did not attempt to measure whether higher levels of use represented a change in prescribing quality. Another study found that pharmaceutical sales representative visits were associated with a greater increase in market share for new entrants into a therapeutic field than was positive scientific information [49].
Journal advertisements. Four out of the eight studies measuring the effects of journal advertisements presented data but did not include statistical tests [25,34,70,80]. One of these noted use of a medication class increased after pharmaceutical advertising commenced in a country where the medication class was previously available but was not promoted [25]. One study visually compared graphs of the monthly number of advertisements and prescriptions for a group of nine drugs and found no clear relationship between the extent of the advertising of a drug and the amount of prescribing by general practitioners [70]. One study found that the market share of a medication was higher amongst physicians who recognised the advertisement for that medication compared to those who did not [80]. The last study observed decreased prescribing of two drug classes at the same time that advertising decreased [34].
Of the four studies that included statistical tests, one found that journal advertisements have a more pronounced effect on market share for the advertised drug than does positive scientific information published in medical journals [49]. A cross-sectional study found contradictory results. Self-reported infrequent use of journal advertisements by physicians to learn about new medications was not associated with frequency of prescribing. However, infrequent use of journal advertisements was associated with less chloramphenicol prescribing [77]. One cross-sectional study found that physicians who recalled advertisements became prescribers of the advertised products in consistently larger proportions than those who did not recall advertisements [75]. Another study found that 9% of high prescribers of new drugs cited advertisements as an influence on their prescribing compared to 0% for low prescribers; however, this was not a statistically significant association [67].
Attendance at pharmaceutical company-sponsored meetings. There were eight studies of pharmaceutical companysponsored meetings. Five found positive associations with prescribing frequency [28,31,43,60,65]. Three studies did not detect a significant association [33,39,40].  Mailed information from pharmaceutical companies. One of the three studies of mailed promotional material found an association with increased prescribing [76]. The others found no association [33,67]. Advertising in clinical software. A single study examined the effect of advertising in clinical practice software and found no association with prescribing frequency for six medications and less prescribing of one medication [29]. The overall result was no association between advertising and prescribing frequency.
Total promotional investment. Eight studies combined the outcome measures for various exposures to pharmaceutical company information or measured overall promotional investment, a proxy for the amount of exposure to information from pharmaceutical companies. Three studies found that total promotional investment was positively associated with prescribing frequency [30,33,51]. Two studies found both positive results and no association [53,54]. One study did not detect an association [52].
Meta-analysis of promotional information and prescribing frequency. We pooled results from a total of seven studies using a random effects model to examine whether exposure to promotion was associated with prescribing of the promoted medication. The seven study results included in the metaanalysis showed significant heterogeneity (I 2 = 91% [95% confidence interval (CI) 84%-95%], tau 2 = 0.35), and therefore we have presented the forest plot without the pooled outcome ( Figure 2) [29,30,31,38,39,63,75]. Using sensitivity analysis we found that study design, quality factors, year of publication, and type of physician did not explain this heterogeneity. One study provided two units of analysis with outcomes amenable to metaanalysis: a significant association for attendance at sponsored meetings and a nonsignificant result for pharmaceutical sales representative (PSR) visits [31]. We included only that nonsignificant result in the forest plot ( Figure 2). When metaanalysis was conducted using the significantly positive result for attendance at a pharmaceutical company-sponsored meeting, the summary result and level of heterogeneity did not differ greatly. The largest difference detected was between exposure to active promotional information (OR 2.34, 95% CI 1.50-3.65), (I 2 = 59%, 95% CI 0%-86%, tau squared = 0.11) [31,38,39,63] and passive promotional information (OR 1.24, 95% CI 0.72-2.15) (I 2 = 89.5%, tau squared = 0.14) [29,75].

Information Delivered Without Conventional Promotion Techniques
Five studies looked for associations between information delivered without conventional promotion techniques and the frequency of physicians' prescribing [35,36,37,58,59]. One randomized controlled trial partnered a local health authority and a pharmaceutical company with the aim of promoting a less expensive drug [35], and the other randomized controlled trial aimed to promote rational prescribing through evidence-based detailing by a pharmaceutical company in partnership with an academic institution [36]. Neither found an association with physicians' prescribing. A single controlled-cohort study of a pharmaceutical company-funded randomized controlled trial found that physicians' participation in recruiting subjects was associated with an increase in the number of prescriptions of the sponsoring company's drug [37]. One time-series analysis found no change in the rate of decline in the prescribing of a medication when the main manufacturer was required by a regulatory agency to deliver an educational program warning about problems with the drug via mailed information and pharmaceutical sales representative visits [58]. A cross-sectional study found no change in prescription rates following warning letters regarding drug side effects [59].

Pharmaceutical Company Information and Prescribing Costs
Eight studies (Table 9) [35,41,50,55,66,68,69,71] measured prescribing costs as costs per physician, price elasticity, and changes in generic prescribing (ten units of analysis). In the United States, one econometric time-series analysis found that pharmaceutical sales representative visits were associated with increased price sensitivity among physicians prescribing in one therapeutic class [69], and another found the opposite effect for hypertension [50]. A third, more recent, econometric study found that promotional outlay (the total for pharmaceutical sales represen-    tative visits, journal advertisements, and direct mail) was associated with reduced price sensitivity for primary care providers and psychiatrists in 11 therapeutic classes consisting of more than 50% of the Dutch drug market [41]. Of three cross-sectional studies, two detected an association between pharmaceutical sales representative visits and higher prescribing costs [66,71], but one did not detect an association [68]. One study also found that low cost prescribers were more likely to have rarely or never read promotional mail or journal advertisements from pharmaceutical companies than high cost prescribers [66]. One time-series analysis found that reduced exposure to pharmaceutical sales representative visits and promotional material was associated with an increase in generic prescribing [55]. A randomized controlled trial of pharmaceutical sales representative visits in a noncommercial partnership between a pharmaceutical company and a local health authority measured physicians' prescribing costs for the target drug class and found no effect [35].

Overview
We found that the reported relationship between exposure to information provided directly by pharmaceutical companies and the quality, frequency, and cost of prescribing varied from case to case. However, with only one exception [39], the included studies reported that exposure to information from pharmaceutical companies was associated with either lower prescribing quality or no association was detected. Similarly, exposure to information from pharmaceutical companies was associated with either an increase in prescribing frequency or no association was detected. Three studies found that exposure was associated with increased drug sales up to a point of diminishing returns beyond which more promotion was increasingly less effective [46,50,69]. Finally, with only one exception [69], exposure to information from pharmaceutical companies was associated with an increase in prescribing costs or no association was detected.
Most of the included studies measured the frequency of prescribing. Amongst these, the studies of informational exposure where physicians are active participants, such as representatives' visits, sponsored meetings, or sponsored trials, more consistently found associations with higher prescribing frequency than studies of more passive exposures, such as journal advertisements and mailed information. Poor study quality precludes confident conclusions about journal advertising. However, one higher quality econometric analysis found that advertisements in journals were associated with a more pronounced effect on market share than positive scientific findings published in journals [49]. Also there are claims in the marketing literature that the relatively low cost of passive methods and their ability to synergistically increase the effectiveness of active methods makes them cost effective components of sales campaigns [84].

Limitations of Included Studies
All of the included studies had design limitations (Tables 1-5). We found only two randomized controlled trials [35,36]. Both lacked adequate reporting of allocation concealment and blinding. These two trials did not examine standard promotional practice but instead assessed novel partnerships of government or academia with industry aiming for less expensive, higher quality prescribing. On the basis of these two negative randomized controlled trials, it seems unlikely that similar partnerships will have beneficial effects on prescribing. No definite conclusions can be extrapolated from these studies to standard promotional practice.
All other included studies were observational and thus able to measure associations but not prove causation. There is a risk that reported associations may be false positives, and that statistically significant findings may not necessarily be clinically significant. One example is the study by Mizik     Canli [32] PSR visits were associated with increased antibiotic prescribing a p = 0.0001* Chintagunta [42] Higher levels of detailing were associated with higher market share for that brand in the three of the countries studied and no significant difference in two others

Change in Prescribing Frequency Results
Azoulay [49] PSR detailing is associated with diffusion of product information and performance on the product market with marketing activities having a more pronounced effect than scientific information in the form of clinical trial reports c

Change in Prescribing Frequency Results
Mackowiak [52] PSR visit expenditure was not associated with a change market size nor market share for benzodiazepines or diuretics No association detected Becker [77] PSR visits per month were not associated with chloramphenicol prescribing Gamma statistic 0.236; not significant

Journal advertisements
Hemminki [25] Journal advertisements were associated with a trend for increased hormone replacement therapy (HRT) prescribing in Estonia

Increased prescriptions
Charbit [34] Journal advertising was associated with increased prescriptions of ARA. When journal advertisements for ACE inhibitors and CCB decreased, their market share also decreased 10.5% decrease in mean annual advertising of ACE inhibitors associated with 19.3% decrease in market share 11% decrease in mean annual advertising for CCBs associated with 19.3% decrease in market share. 20.5% increase in mean annual advertising rate for ARAs associated with 22.9% increase in market share Prosser [67] Journal advertisements were no more likely to be cited as a prescribing influence by high prescribers than by low prescribers 9% high prescribers versus 0% of low prescribers; Fisher exact test; p = 0.18 Azoulay [49] Journal advertisements were associated with diffusion of product information and performance on the product market with marketing activities having a more pronounced effect than scientific information in the form of clinical trial reports c Kreyenbuhl [31] Attendance at pharmaceutical sponsored CME meetings more than once in the preceding month was associated with the ''add'' rather than ''switch'' strategy for antipsychotic medication prescribing a OR 2.32 (95% CI 1.29-4.18); p = 0.005* Venkataraman [40] Attendance at pharmaceutical sponsored meetings was not significantly associated with prescriptions for 7 out of 12 brands Beta coefficient 20.659 (significant with a 90% CI) Narayanan [43] Attendance at pharmaceutical company-sponsored meetings was associated with an increase in promoted medication market share A 1% increase in expenditure on ''other marketing activities'' (including meetings) was associated with increases in market shares for promoted drugs ranging from 0.02% to 0.04% (p,0.05) Huang [65] Attendance at pharmaceutical sponsored conferences was associated with more prescriptions of the corresponding sponsored antidepressant a

Change in Prescribing Frequency Results
Bowman [60] Attendance at pharmaceutical sponsored courses was associated with more prescriptions of medication made by sponsoring company Before and 6 mo after 3 sponsored course involving sponsoring company's drugs:

Information delivered without conventional promotion
Andersen [37] Participation in pharmaceutical funded research was associated with increase in the sponsoring company's share of asthma drug in practices conducting the trial compared to control practices 6.7% increase (95% CI 3.0%-11.7%) b increase in prescriptions associated with visits from pharmaceutical sales representatives [45]. Associations may also arise from confounding, bias, or chance. False negatives or inaccurate estimation of effect sizes may result from small sample sizes, measurement errors, overly complex models, or ''contamination'' when prescribers who are thought to be unexposed are actually influenced by other methods. For example in a study of promotional meetings, nonattenders may be influenced by sales representatives thus reducing the difference from attenders in their prescribing. Another possible source of contamination is indirect influence by colleagues who have been influenced directly.
To the extent that the measured associations are real, causality may be bidirectional. The influence of information from pharmaceutical companies on prescribing is a likely explanation for the associations given that the major purpose of pharmaceutical promotion is to influence prescribing [3]. However, it is also possible that physicians who prescribe larger quantities, more expensively or less appropriately may allow themselves to be exposed to, or attract, more promotional information.
Some studies found no association between exposure to information from pharmaceutical companies and prescribing outcomes or small effect sizes that seem unlikely to be clinically significant. Some of these may be false negatives or underestimations caused by study flaws, but it is likely that information from companies sometimes has little or no effect, especially when the information is not designed to increase sales, e.g., letters warning about safety problems. Most of the studies included in this review examined single components of promotional campaigns that may have little or no effect alone but have a synergistic effect in combination with other components. Promotion may be less effective if it is used beyond the point of diminishing returns or is up against similarly effective promotion for another similar product.   Given the controversial nature of this topic, there are many reasons why the studies could be biased overall in either direction. Authors may have produced results consistent with their ideological bias. Also reciprocal obligation to funders who preferred certain results may have lead to bias with or without conscious awareness. Publication and outcome reporting bias may have led to underrepresentation of negative, positive, uninteresting, or unwanted findings.

Strengths and Weaknesses
The strengths of this review include use of a comprehensive search strategy over multiple databases without any language exclusions. We consulted widely with experts in the field and we used validated instruments to assess quality of the studies. However, only one of the included studies was conducted in a low-income economy, as defined by the World Bank, so the effects of promotion there are less certain [33]. This study found a positive association between pharmaceutical promotion and prescribing frequency. Promotion may be more influential in these countries given the relative paucity of independent sources of information [85,86].
Our efforts to access data that was not in the databases we searched had mixed results. Messages on e-mail discussion groups and contact with experts yielded five additional studies subsequent to the initial search [34,43,[80][81][82] whose results were consistent with the entire review. By contrast, pharmaceutical companies did not provide us with any information that was not already in the public domain. However five studies included in this review analyzed confidential data from pharmaceutical companies and their results were also consistent with the review as a whole [33,35,37,40,46].
Given the wide range of knowledge and experience among the sources that we consulted and the expertise in our group, we are confident that we exhausted all reasonable avenues in our attempt to obtain additional literature.

Data Interpretation
Of the 58 studies included in this review, 38 studies reported a single unit of analysis with 25 (66%) finding significant associations between exposure to information from pharmaceutical companies and the quality, frequency, and cost of prescribing and eight (21%) finding no associations. The remaining five (13%) had multiple measures and found significant associations on some measures but not on others. The 20 studies with more than one unit of analysis reported 49 units of analysis of which 21 (43%) found significant associations, 24 (49%) found no associations, and four (8%) found mixed results. The difference between the results of the single versus multiple unit of analysis studies is significant (p,0.05 Freeman-Halton extension of the Fisher exact test). This difference may have been caused by publication bias against publication of single unit of analysis studies when no association was found. We believe the pattern of results suggests that there was little or no reporting bias for the multiple unit of analysis studies. Because the multiple unit of analysis studies found no association more often than the single unit of analysis studies, multiple mentions of the former studies in our narrative synthesis will not exaggerate the frequency of findings of significant associations. Table 9. Relationship between exposure to information from drug companies and prescribing costs (by year of publication and then study design/size). Interpretation of our meta-analysis requires caution because many studies included in the narrative synthesis could not be included in the meta-analysis. Where a sufficient number of studies could be combined, there was significant heterogeneity. The summary result has not been presented because it is unlikely to accurately reflect the true effect size of most promotional campaigns for two main reasons. First, effect sizes varied widely so it is likely that promotional campaigns often have effect sizes far from average. Second, single promotional techniques are likely to be less effective individually than campaigns employing multiple promotional methods.
A sensitivity analysis found the difference between passive and active promotion is one possible cause of heterogeneity. Other possible explanations for variation in the effectiveness of promotion include variation from campaign to campaign in the relative benefits of the drug being promoted, the promoter's skills and budget, and the target group's level of resistance to promotion.

Conclusions
The limitations of studies reported in the literature mentioned above mean that we are unable to reach any definitive conclusions about the degree to which information from pharmaceutical companies increases, decreases, or has no effect on the frequency, cost, or quality of prescribing. In theory, advertising may be beneficial in several ways: by distributing information and thus improving the quality of prescribing [20,78], by reducing costs through increasing price-elasticity [69], by increasing prescribing of drugs that provide better health outcomes, or by improving the cost-effective use of healthcare resources. Because of the limitations of both the included studies and this review we have not disproved those theories but we have found little evidence to support them and have found some evidence of increased costs and decreased quality of prescribing. Any conclusions about harm or benefit for patients are speculative because none of the studies that we found examined clinical outcomes. One clear conclusion from this review is that we did not find evidence of net improvements in prescribing associated with exposure to information from pharmaceutical companies. Some argue that prescribers have an ethical duty to avoid exposure to pharmaceutical promotion [13,[87][88][89]. Even ineffective promotional information may be harmful if it wastes prescribers' time or if the money spent on promotion increases the cost of medicines [90]; this is of concern given the large expenditure involved [1,2]. In the absence of evidence of net improvement in prescribing from exposure to promotional information, we recommend that practitioners follow the precautionary principle and thus avoid exposure to information from pharmaceutical companies unless evidence of net benefit emerges.

Supporting Information
Alternative Language Abstract S1 Malaysian translation of the abstract by NO. Editors' Summary Background. A prescription drug is a medication that can be supplied only with a written instruction (''prescription'') from a physician or other licensed healthcare professional. In 2009, 3.9 billion drug prescriptions were dispensed in the US alone and US pharmaceutical companies made US$300 billion in sales revenue. Every year, a large proportion of this revenue is spent on drug promotion. In 2004, for example, a quarter of US drug revenue was spent on pharmaceutical promotion. The pharmaceutical industry claims that drug promotion-visits from pharmaceutical sales representatives, advertisements in journals and prescribing software, sponsorship of meetings, mailed information-helps to inform and educate healthcare professionals about the risks and benefits of their products and thereby ensures that patients receive the best possible care. Physicians, however, hold a wide range of views about pharmaceutical promotion. Some see it as a useful and convenient source of information. Others deny that they are influenced by pharmaceutical company promotion but claim that it influences other physicians. Meanwhile, several professional organizations have called for tighter control of promotional activities because of fears that pharmaceutical promotion might encourage physicians to prescribe inappropriate or needlessly expensive drugs.
Why Was This Study Done? But is there any evidence that pharmaceutical promotion adversely influences prescribing? Reviews of the research literature undertaken in 2000 and 2005 provide some evidence that drug promotion influences prescribing behavior. However, these reviews only partly assessed the relationship between information from pharmaceutical companies and prescribing costs and quality and are now out of date. In this study, therefore, the researchers undertake a systematic review (a study that uses predefined criteria to identify all the research on a given topic) to reexamine the relationship between exposure to information from pharmaceutical companies and the quality, quantity, and cost of physicians' prescribing.
What Did the Researchers Do and Find? The researchers searched the literature for studies of licensed physicians who were exposed to promotional and other information from pharmaceutical companies. They identified 58 studies that included a measure of exposure to any type of information directly provided by pharmaceutical companies and a measure of physicians' prescribing behavior. They then undertook a ''narrative synthesis,'' a descriptive analysis of the data in these studies. Ten of the studies, they report, examined the relationship between exposure to pharmaceutical company information and prescribing quality (as judged, for example, by physician drug choices in response to clinical vignettes). All but one of these studies suggested that exposure to drug company information was associated with lower prescribing quality or no association was detected. In the 51 studies that examined the relationship between exposure to drug company information and prescribing frequency, exposure to information was associated with more frequent prescribing or no association was detected. Thus, for example, 17 out of 29 studies of the effect of pharmaceutical sales representatives' visits found an association between visits and increased prescribing; none found an association with less frequent prescribing. Finally, eight studies examined the relationship between exposure to pharmaceutical company information and prescribing costs. With one exception, these studies indicated that exposure to information was associated with a higher cost of prescribing or no association was detected. So, for example, one study found that physicians with low prescribing costs were more likely to have rarely or never read promotional mail or journal advertisements from pharmaceutical companies than physicians with high prescribing costs.
What Do These Findings Mean? With rare exceptions, these findings suggest that exposure to pharmaceutical company information is associated with either no effect on physicians' prescribing behavior or with adverse affects (reduced quality, increased frequency, or increased costs). Because most of the studies included in the review were observational studies-the physicians in the studies were not randomly selected to receive or not receive drug company information-it is not possible to conclude that exposure to information actually causes any changes in physician behavior. Furthermore, although these findings provide no evidence for any net improvement in prescribing after exposure to pharmaceutical company information, the researchers note that it would be wrong to conclude that improvements do not sometimes happen. The findings support the case for reforms to reduce negative influence to prescribing from pharmaceutical promotion.