|Home | About | Journals | Submit | Contact Us | Français|
To review literature on the impact of FDA drug risk communications on medication utilization, health care services use, and health outcomes.
The authors searched MEDLINE and the Web of Science for manuscripts published between January 1990 and November 2010 that included terms related to drug utilization, the FDA, and advisories or warnings. We manually searched bibliographies and works citing selected articles and consulted with experts to guide study selection.
Studies were included if they involved an empirical analysis evaluating the impact of an FDA risk communication.
We extracted the drug(s) analyzed, relevant FDA communication(s), data source, analytical method, and main outcome(s) assessed.
Of the 1432 records screened, 49 studies were included. These studies covered sixteen medicines or therapeutic classes; one-third examined communications regarding antidepressants. Most used medical or pharmacy claims and few rigorously examined patient-provider communication, decision-making or risk perceptions. Advisories recommending increased clinical or laboratory monitoring generally led to decreased drug use, but only transient and modestly increased monitoring. Communications targeting specific subpopulations often spilled over to other groups. Repeated or sequential advisories tended to have larger but delayed effects and decreased incident more than prevalent use. Drug-specific warnings were associated with particularly large decreases in utilization, though the magnitude of substitution within therapeutic classes varied across clinical contexts.
While some FDA drug risk communications had immediate, strong impacts, many had either delayed or no impact on health care utilization or health behaviors. These data demonstrate the complexity of using risk communication to improve the quality and safety of prescription drug use, and suggest the importance of continued assessments of the effect of future advisories and label changes. Identifying factors that are associated with rapid and sustained responses to risk communications will be important for informing future risk communication efforts.
The Food and Drug Administration (FDA) is responsible for protecting the public by ensuring the safety of prescription drugs. However, new safety information often emerges after drug approval, requiring the FDA to systematically evaluate and respond to new evidence about drug safety as it accumulates. This authority increased with the passage of the 2007 FDA Amendments Act (FDAAA) (1).
After identifying a new safety concern, the FDA must decide how to effectively communicate this risk information to the general public and providers (2). These communications range from minor revisions to the drug’s label to major safety advisories. Historically the FDA has communicated such risks through a combination of methods, including “Public Health Advisories” and “Safety Alerts” directed to the general public (referred to as “advisories” throughout), or “Dear Healthcare Provider” letters directed to prescribers. Additionally, the FDA may modify the drug label by including a “Black Box warning” when risks are particularly severe (3). During the past two decades the Agency has used these methods to communicate hundreds of drug risks discovered after market approval (4).
Risk communication is a complex science (5–8) and understanding the intended and unintended outcomes from prior FDA advisories and label warnings may help to guide future risk communication. We sought to summarize the current literature regarding the impact of FDA risk communications on utilization of drugs, use of other health care services, and health outcomes.
We searched MEDLINE (via PubMed) and Web of Science (Thomson Reuters) for relevant studies published between January 1, 1990 and November 10, 2010. A series of MeSH headings were specified to identify terms related to each concept: (1) drug utilization, (2) the FDA, and (3) advisories or warnings (Appendix). Each search string was executed in PubMed and the three completed searches were combined after excluding duplicated records. A similar search was executed in the Web of Science and results from both searches were combined.
Our systematic search identified 1,432 non-duplicated articles for review (Figure 1). We included manuscripts that analyzed empirical data on the impact of FDA risk communications on healthcare utilization or health behaviors. We defined risk communication as activities used to inform the public or healthcare providers of a medical product’s risk. We excluded letters to editors, commentaries, news articles, and studies of populations or advisories issued by regulatory authorities outside of the United States. We also excluded articles evaluating the impact of risk management plans, market withdrawal, studies of non-drug medical products or devices, and those whose primary focus was not an assessment of an Agency communication.
Using these criteria, one investigator (AH) and a research assistant (MY) independently evaluated articles retrieved through the search strategy, reference lists of selected articles, and articles citing included studies. There was 98% agreement between the reviewers regarding article selection (kappa=.73, 95% confidence intervals .63–.83). A second author (GCA) evaluated the 26 articles where disagreement was present; 20 were excluded. Forty-nine studies were selected for inclusion.
Our primary goal was to synthesize information about the effect of FDA advisories across the studies identified. To do so, we categorized studies into one of four non-mutually exclusive groups based on whether the study examined Agency recommendations regarding: (1)(2) avoiding co-prescribing due to drug-drug interactions; (3) avoidance of use among a subpopulation; or (4) general caution regarding a product. We reasoned that this categorization might provide a useful taxonomy for considering the intended and potential unintended consequences of FDA communications (Figure 2). Given the distinct nature of studies based on surveys or qualitative methods, we examined these investigations separately.
In addition to extracting details of the primary outcomes studied, we also identified the drug(s) analyzed, type(s) of FDA communication assessed, data source, analytic method, and any secondary outcomes assessed. We categorized outcomes into one of six groups: (1) utilization or dispensing of the targeted drug or therapeutic class, (2) use of substitutes for the targeted drug or drug class, (3) spillover effects, such as a communication targeting children impacting adults, (4) diagnoses and non-prescription utilization, (5) health behaviors and health outcomes (e.g., glycemic control), and (6) patient or physician knowledge, attitudes or beliefs regarding risk communication messages (9).
The studies utilized a variety of data sources including: medical or pharmacy claims, surveys, medical records, prescribing audits, focus groups, and vital statistics (Table 3). Sample size and composition varied widely; from a focus group of 35 health professionals (10) to a retrospective analysis of nearly 800,000 privately insured patients’ health claims (11). Statistical methods also varied widely and many studies used multiple methods. Twelve studies (24%) employed time series analyses, 17 (35%) used other types of regression, 11 (22%) used descriptive statistics alone, and 9 (18%) used other means of analysis such as Pearson’s chi-square test. The most common outcomes assessed were use of the targeted drug or drug class (40 studies), changes in diagnosis and non-prescription utilization (18 studies), and changes in substitute use (18 studies) (Table 4).
Assessments of the impact of FDA recommendations on laboratory or clinical monitoring included studies of glucose testing for patients taking atypical antipsychotics (12,13), liver enzyme monitoring for patients taking troglitazone (14,15) or pemoline (16), and increased follow-up visits for patients taking antidepressants (17–19). Of studies that assessed laboratory monitoring, there was no evidence of a large or sustained impact of the FDA recommendations. For example, rates of glucose testing among patients taking antipsychotics were unchanged following the FDA’s recommendation to monitor patients for hyperlipidemia and diabetes (12,13). Rates of liver enzyme monitoring following a black-box warning for troglitazone were low prior the first FDA advisory, increased modestly over subsequent advisories, and waned over time (14,15), ultimately contributing to market withdrawal. Recommendations to increase clinical monitoring also appeared to have little impact on clinical practice (17–19).
Studies of co-prescribing focused on drug-drug interactions related to three therapies: cisapride, terfenadine, and tramadol. Most studies suggested that although clinical practice is slow to respond to such warnings, often taking months to years, inappropriate co-prescribing decreases over time. For example, several studies suggested the largest declines in cisapride co-prescribing occurred following the FDA’s third such warning (20–23), although not all studies indicated such a change (24). Declines in terfenadine co-prescribing also occurred although the largest effects were not observed until 18 months or more following the initial warnings (25–27). Only one study concluded there was an increase in terfenadine co-prescribing between 1991 and 1992, although the authors examined absolute rather than relative co-prescribing events (28). Finally, a study of tramadol found a non-significant change in co-prescribing following an FDA warning, although the observed trend suggested that co-prescribing decreased (29).
Several studies examined communications cautioning use of a medicine or therapeutic class among a patient subpopulation, including over-the-counter [OTC] cough medicines and promethezine among children under 2 years of age, antidepressants among children and adolescents, angiotensin-converting enzyme [ACE] inhibitors among pregnant women, telithromycin among patients with myasthenia gravis, and antipsychotics among the elderly with dementia or those with or high risk of diabetes or metabolic syndrome.
Most, but not all (30, 31), investigations focused on changes in utilization among the subpopulation as well as spillover effects whereby non-targeted populations may have been impacted. For example, between January 2006 and February 2007 the FDA issued a series of communications contraindicating telithromycin use in patients with myasthenia gravis, and removing indications for bacterial sinusitis and acute bacterial exacerbations of chronic bronchitis. Claims for telithromycin from a large commercial insurer dropped by 80% between January 2006 and February 2007, while the proportion of all telithromycin prescribed for revoked indications was similar before and after the label change (32). A nationally representative physician audit identified substantial decreases in atypical antipsychotic use among elderly with dementia following an April 2005 black box warning that were accompanied by more modest decreases in use among those without dementia and for FDA-approved indications (33). ACE inhibitor exposure among pregnant women in Medicaid from 1986–2003 appeared to increase four-fold, with no decrease following the FDA’s warning (31).
Most studies of FDA communications targeting a subpopulation have focused on a series of FDA communications about antidepressant use among children and adolescents that were issued between June 2003 and October 2004. Most evidence suggests there were substantial decreases in prescribing among children (11,18,34–39) with more modest decreases among adults (34,36). The antidepressant advisories also had a variety of other effects, such as shifts in the market share accounted for by paroxetine, fluoxetine and escitalopram(40), a shift away from diagnoses from primary care physicians (11,34,41), and a lack of non-antidepressant drug substitution (11,34,38,41,42). There is mixed evidence as to whether there were changes in psychotherapy utilization rates among children (11,18,41,43). Little is known regarding the health outcomes associated with these advisories, although the single study examining the relationship between antidepressant warnings and suicides among youth in the U.S. and the Netherlands found that suicide rates increased following declines in prescribing (44).
A final series of communications have related a general safety concern regarding a drug or therapeutic class without emphasis regarding specific monitoring schedules, co-medication use, or avoidance in select subpopulations. Investigations have examined settings including topical calcineurin inhibitors, depo-provera, droperidol, long-acting beta-agonists (LABAs) and glitazones. The response to these communications has varied considerably. A study of droperidol saw a significant decrease of use following the December 2001 black box warning, accompanied by an even greater increase in use of ondansetron, a known substitute (45). Studies examining the use of LABAs among patients with asthma suggested infrequent controller use (e.g., inhaled corticosteroids) among individuals prior to their receiving a LABA, and neither investigation suggested large changes in rates of controller use following the FDA warnings (46,47). By contrast, studies of rosiglitazone and pioglitazone following a May 2007 (rosiglitazone) and August 2007 (classwide) advisory identified significant decreases in rosiglitazone use (48–51) with stable pioglitazone prescribing(48) and significant increases in use of other anti-diabetes drugs (51).
Of the 49 studies included in the systematic review, 8 studies used surveys and 1 used focus groups in order to examine parental or provider attitudes, beliefs, or experiences regarding FDA communications about a drug or therapeutic class. These analyses suggested high levels of provider awareness regarding general communications or labeling changes (10,52,53), though lower levels of awareness of specific recommended monitoring schedules in the case of antidepressants (10,52). Provider agreement with the content of the communication was high in the case of OTC cough and cold treatments among young children (54) and much lower in the case of LABAs (53), droperidol (55,56), and recommendations for monitoring patients taking antiepileptic drugs (19). In the only study examining parental attitudes and beliefs, the majority of respondents, though aware of the 2008 advisory regarding OTC cough and cold treatments, did not believe the products were potentially dangerous (54).
Many of the nine survey-based or qualitative studies examining providers’ and patients’ knowledge, attitudes, and beliefs were of small or selected populations (e.g., convenience samples), and few provided adequate information to assess efforts to minimize the effects of non-response and socially-desirable response bias (57,58). The majority of studies in this review consisted of empirical analyses of patients’ health care utilization, and most used either descriptive analysis or regression discontinuity approaches. Of studies employing regression models, several included methods to account for the outcome frequency (e.g., binomial or poisson models), to examine and correct for first-order and higher levels of correlation (e.g., seasonality) within the data, and to account for the potential of a lag between an FDA communication and its impact. Despite this, none of the quasi-experimental studies analyzed used a difference-in-difference approach with a control group, often difficult to identify in these types of studies, to account for secular trends. Similarly, few studies statistically adjusted for influential scientific reports, media coverage, litigation, or industry promotion, although several acknowledged their potential influence on the outcomes of interest. Finally, most studies looked at aggregate patterns of prescribing rather than patient-level changes in utilization, including changes in treatments among disease-specific cohorts.
Regulatory advisories are a common method of disseminating emerging safety risks from prescription drugs, and the role of the FDA in communicating will grow with the Sentinel Initiative and other regulatory efforts to improve safe prescription use (59,60). Most studies we examined were empiric analyses of drug and non-drug utilization; many fewer examined patient-provider communication and decision-making or risk perception. Since the communications had highly variable effects, the small number of contexts investigated relative to the number of settings where advisories have been issued is noteworthy.
The communications examined differed in several ways, such as whether they focused on a single drug, targeted a specific subpopulation, or addressed an idiosyncratic risk. Although the nature of the studies precludes quantitative comparisons, several suggestive conclusions can be drawn. First, advisories recommending greater monitoring do not appear to have led to large and sustained changes. For example, increases in laboratory monitoring for troglitazone, pemoline or antipsychotics and clinical monitoring for antidepressants were generally incomplete and/or temporary and often led to decreases in medication use. Second, in settings where incident and prevalent use was examined, such as in the case of cisapride, terfenadine, and antidepressants, warning information appears to have been adopted more quickly for new rather than continuing medication users. Despite this, it is not clear from these reports whether continuing users or their providers were well informed regarding the specific risks addressed. Third, principles of risk communication suggest that warnings will be most effective in cases where they are specific, where acceptable alternatives are available and where the messaging is reinforced over time (5). The case of cisapride provides one example, where the first and less specific messaging did not lead to large changes in co-prescribing with contraindicated therapies, compared with subsequent more specific communications (21,22). Interestingly, the magnitude of substitution within a class where an advisory was issued for a single agent varied across settings, from little evidence of substitution of pioglitazone for rosiglitazone, to modest increases in clarithromycin for telithromycin, to larger increases in antidepressants other than paroxetine, albeit with additional advisories regarding other specific anti-depressants. Finally, based on the selected primary data collected from surveys and focus groups, most physicians appeared aware of general safety concerns, though many provider respondents disagreed with advisory content. The single study examining parental attitudes and beliefs suggest they had limited understanding or awareness of the OTC cough medicine communication.
As with other public policies, FDA communications have the potential for unintended consequences. Antipsychotic and antidepressant advisories were associated with decreased use in non-targeted populations, and there is evidence of increases in suicides among children following the antidepressant advisories (44). Decreases in telithromycin use also appeared to occur non-selectively; the advisory did not appear to affect the proportion of all uses that were for off-label indications (32). These studies are not the first to suggest the potential for unintended consequences from regulatory warnings regarding prescription drugs (61,62), and their potential demonstrates the challenge that policy-makers face to design risk messages that are maximally effective, specific, and that don’t become risks themselves.
Our study has a number of limitations. First, the number of different therapeutic areas, communications, outcomes, and study designs used precluded anything more than a qualitative analysis of findings from the studies that we gathered (63). The heterogeneity of studies does not permit for an “average effect” of FDA communications to be estimated, nor an estimate of the relative potency or impact of one form of communication compared with another. Second, regulatory advisories exist within a sea of information that may influence health care utilization and behaviors (64,65), and thus the complexity of the phenomenon studied, and multitude of potential influences that may shape behavior, also limit conclusions drawn from this type of review. Third, our study was limited to the peer-reviewed literature regarding FDA drug risk communications. Finally, although the behavioral responses that most studies examined are relevant to stakeholders, they provide incomplete information on the efficacy of a communication; the absence of a strong response may indicate a previously informed population, or a population where the new information was insufficient to prompt different decision-making, whereas the presence of a strong response may indicate over compensation on the part of patients and providers (8).
Our review also raises several questions. None of the studies examined included rigorous assessment of how the effect of advisories may vary across groups of physicians, practices, or other communities. Given the complex determinants of health care utilization and health behavior, there is likely substantial variation in responsiveness to FDA communications among different subsets of providers and patients (66). In addition, although one-fifth of the studies examined were based on surveys or qualitative investigations, little is known regarding how these risk communications affected patients’ communication with health care providers or their risk perceptions. Also, few studies(43) examined medication switching among individual patients. Thus, although some studies provide insights regarding how a regulatory communication may lead to changes in market share within a therapeutic class, almost none provide a window for understanding how individual patients’ pharmacotherapies change as a result of an advisory or label warning. Finally, few studies examined health outcomes associated with medication changes that may have ensued following an FDA communication. Since the goals of risk communication are to minimize potential harms, understanding the impact of an FDA communication’s intended (and unintended) consequences on patient health outcomes should be a priority.
The FDA’s primary mission is to protect the public health (67), yet there are numerous challenges in designing and disseminating risk information, including a finite and ever evolving evidence base as well as social, psychological, and cognitive dimensions of risk perception and communication (5–8). The Sentinel Initiative, as well as the FDA’s greater authority to enforce post-approval Risk Evaluation and Mitigation Strategies, will increase the volume of safety information that must be conveyed to providers and the general public (68). This review indicates the varied and unpredictable impact that such outreach can have, and suggests the importance of continuing to characterize the effect of advisories and warnings on a variety of behaviors to enhance the science of risk communication regarding prescription drugs.
This work was supported by the Agency for Healthcare Research and Quality (RO1 HS0189960) and the University of Chicago Clinical and Translational Science Award. Dr. Dorsey was supported by the Robert Wood Johnson Physician Faculty Scholars Program and Dr. Huskamp was supported by a Robert Wood Johnson Foundation Investigator Award in Health Policy Research. The funding sources had no role in the design and conduct of the study; collection, management, analysis, or interpretation of the data; and preparation, review, or approval of the manuscript for publication.
The authors gratefully acknowledge Mounica Yanamandala for research assistance and Greg Simon and Paul Shekelle for helpful comments on previous manuscript drafts.