1. Introduction

Once a drug or device is approved for use by a regulatory authority the product is generally used by larger and more diverse populations than are typically studied in the clinical trials leading up to approval. As a result, the period after approval is an important phase for identifying and understanding product safety concerns associated with both acute and chronic use. The need for postapproval (also called postmarketing) safety assessment as it exists today was, for the most part, born out of well-publicized product safety issues that were initially detected by clinicians recognizing a pattern of rare serious events, such as phocomelia caused by prenatal exposure to thalidomide¹ and rare vaginal cancers that occurred in young women who had in utero exposure to diethylstilbestrol.² The detection of serious adverse drug reactions after authorization has led to much debate about the adequacy of both industry and regulatory approaches to preauthorization assessment and testing. However, the decision to authorize a medicine is a balance between wanting to know as much as possible about the safety of a product and the need to make new drugs available for patients.³ The implication is that authorization cannot mean that a medicine is completely safe; rather, it is an assessment that at the time of authorization, the known benefits for the average patient in the approved indication outweigh the known risks. But the degree to which the known risks represent the actual safety profile of a product will depend upon the size, duration, representativeness, and thoroughness of the clinical trial program, which, in turn, is related to the complexity of the patients and the state of knowledge of the disease being targeted. Trials conducted as part of clinical development are, by necessity, of limited duration and size and generally focus on a narrowly defined population that represents only a small segment of the population with the disease or product use of interest. Clinical trial populations tend to be restricted to those who have limited concurrent disease and who are on few, if any, concomitant medications. Typically, trial protocols include lengthy lists of inclusion and exclusion criteria that further restrict the trial population. Unless a drug or a product is intended for a very narrow indication or a very rare disease, it is not feasible to require clinical trials to be inclusive of all types of patients likely to ever be exposed to it. Even in the case of a narrow indication, the potential long-term and delayed effects of a product are unlikely to be established during most clinical trial development programs.

To address the acknowledged limitations of what is known about the safety profile of a product at the time of authorization, postmarketing pharmaco- and medical device vigilance is traditionally, and by regulation, performed through spontaneous adverse event (AE) reporting. The exact requirements for spontaneous reporting to the regulatory authorities vary internationally and depend on the country/region, approval type, and product type. It is widely acknowledged, however, that spontaneous reporting captures an extremely small percentage of the actual events occurring, and that, while it is useful for identifying rare and potentially significant events,^{4, 5} it has limited use in the detection of other equally important types of events, including increases in events with a high background rate. This form of postmarketing surveillance is reactive in that one waits for AEs/reactions to be spontaneously reported assesses them for causality, and estimates the importance of the information.

As well as collecting only an indeterminate fraction of adverse reactions, this method of surveillance depends upon someone reporting the events of interest. There is some evidence that clinicians who report AEs are not typical of clinicians in general, and other reporters such as patients, lawyers, and consumer groups may have unclear motivations for reporting, which introduces further bias into the equation.^6-8

The current methods available for AE reporting are seen by many as burdensome and not amenable to incorporation into a clinician's normal workflow. Waiting for reports to arrive and accumulate may also delay the detection of adverse reactions. On the other hand a massive uptake of a new drug or device, such as seen with Viagra^® (sildenafil citrate) or coronary artery stents, may lead to a sudden flood of reports of nonserious as well as serious AEs that could potentially overwhelm established systems.

To overcome some of the difficulties associated with managing large databases of spontaneous AEs, many investigators employ statistical methods to identify signals of disproportionate reporting (SDRs). These methods identify AEs that are reported more frequently with a drug or device than would be expected compared with other event/product pairs in the database, and do not imply any kind of causal relationship.⁹ It is important to be precise as to what is meant when using the term “signal” or “signal detection” since the terms are ambiguous; in the context of automated methods of detecting statistical anomalies, the term “SDR” should be used.⁹ However, these statistical methods may not be reliable in certain situations, such as when there is major confounding or when the increased risk is small compared with the background incidence of the event.⁹ All these above-mentioned limitations mean that there are situations when spontaneous reporting may not be adequate as the sole method of postmarketing surveillance.

To address problems with traditional pharmaco- or medical device vigilance when there are particular known limitations of knowledge of the safety profile of a product and/or to further address unresolved safety concerns, some products are approved subject to postmarketing commitments, which may be requested for safety purposes as well as to address other outstanding questions. In Europe, in response to concerns over pharmacovigilance, marketing authorization applicants are required to submit a European Union-risk management plan (EU-RMP) when seeking a marketing authorization for the majority of new chemical entities and biologics. This EU-RMP states what is known and not known about the safety profile of a medicinal product, how its safety profile will be monitored investigated and characterized and what risk minimization activities will be undertaken. While many products will require only routine pharmacovigilance, for others more proactive methods of pharmacovigilance will be necessary to supplement the use of spontaneous adverse reaction reporting and periodic safety update reports. Although additional clinical trials may occasionally be mandated it is more common for observational pharmacoepidemiological studies to be conducted to ascertain the safety profile of a product under real-world use.

Other observational methods of tracking and evaluating safety data have historically included active surveillance systems, such as the prescription event monitoring (PEM) systems used in the United Kingdom (Drug Safety Research Unit),¹⁰ New Zealand (NZ Intensive Monitoring Programme), Japan (J-PEM), and elsewhere, targeting new products; and the retrospective use of administrative claims data. In the United Kingdom, the requirement that access to most secondary care is through a general practitioner has led to the use of their electronic health care systems for pharmacovigilance purposes; however, this type of integrated approach is not yet widely accessible elsewhere. In May 2008, the U.S. Food and Drug Administration (FDA) launched the Sentinel Initiative, an effort to create an integrated electronic system in the United States for AE monitoring, incorporating multiple existing data sources including claims data and electronic medical record systems.¹¹

Medical devices in the United States have surveillance programs different from those for drugs. The Safe Medical Devices Act of 1990 requires that high-risk medical devices be tracked after marketing, and that product corrections and removals be reported to FDA if actions were taken to reduce health risks. Most medical device safety tracking is accomplished through reports submitted to FDA from medical facilities when devices are implanted or explanted. In addition, hospitals, nursing homes, ambulatory surgery centers, and outpatient treatment facilities are required to report to FDA whenever they believe that a device caused or contributed to the death of a patient, though this reporting is a voluntary requirement and not enforceable or audited.¹²

Whether to comply with a postmarketing requirement or out of a desire to supplement spontaneous reporting, prospective product and disease registries are also increasingly being considered as a resource for examining unresolved safety issues and/or as a tool for proactive risk assessment in the postapproval setting. The advantage of registries is that their observational and inclusive design may allow for surveillance of a diverse patient population that can include sensitive subgroups and other groups not typically included in initial clinical trials, such as pregnant women, minorities, older patients, children, or patients with multiple comorbidities, as well as those taking concomitant medications. In contrast to clinical trials, in which the inclusion criteria are generally tightly focused and restrictive by design, registry populations are generally more representative of the population actually using a product or undergoing a procedure, since the inclusion criteria are usually broad and may potentially include all patients exposed regardless of age, comorbidities, or concurrent treatments. Data collection may lead to insights about provider prescribing practices or off-label use and information regarding the potential for studying new indications within the expanded patient population. Followup duration can be long to encompass delayed risks, consequences of long-term use, and/or effects of various combinations and sequencing of treatments. Such information can be used as a source of publications, to assist the medical community with developing recommendations for monitoring patient safety and product usage, and/or to contribute to the understanding of the natural history of the disease.

There are also many challenges to the utility of registry data for providing more clarity about safety concerns and for prospective risk surveillance. These challenges relate largely to how products are used and the legal, regulatory, and ethical responsibilities of registry sponsors. Most registries that follow specific products do so through cooperation from physicians who prescribe (or implant) these products. Depending on the setup and legal constraints of the registry, sometimes only a subsection of prescribing physicians may be involved in entering patients, a situation that raises questions about the representativeness of the physicians and their patients. However, the registry approach has the potential to be very useful for studying products that are used according to their labeled indications; it also allows for effective surveillance of products that are used off label but by the same practitioners who would use it for the labeled indication. For example, a product might be approved for people with moderate to severe asthma and used off label in patients with mild asthma, yet the prescribing medical providers would already be included in the registry and could easily provide information about all their product use. Off-label use is much more difficult to study when a medical product is used by a wide variety of medical care providers; for example, drugs that promote wakefulness or are thought to increase a patient's ability to concentrate, acting as immunomodulators. The legal, regulatory, and ethical aspects of registry sponsors also affect whether they are required to report any AEs that may be observed since only those legal entities that market (or distribute) a medical product are required to report AEs. For all other parties, such reporting is ethical and desirable, but not enforceable or required.

The purpose of this chapter is to examine the role of registries as one of the available tools for enhanced understanding of product safety through AE detection and evaluation. The examination will include both the role of registries created specifically for the purposes of safety assessment and of those in which the collection of safety data is ancillary to the registry's primary objectives. The legal obligations of regulated industries are discussed by others and are only mentioned briefly here. Similarly, issues to consider in the design and analysis of registries are covered in Chapters 3 and 13, respectively. Chapter 12 discusses practical and operational issues with reporting AE data from registries. The potential ethical obligations, technical limitations, and resource constraints that face registries with multiple different purposes in considering their role in AE detection and reporting are also discussed in this chapter. Case Examples 44 and 45 offer descriptions of how some registries have provided data for product safety assessments.

2. Registries Specifically Designed for Safety Assessment

Disease and product registries that systematically collect data on all eligible patients are a tremendous resource for capturing important information on safety. Registries commonly enroll patients who are not just different from but more complicated than those included in clinical trials, in terms of the complexity of their underlying disease, their comorbidities, and their concomitant medications.

3. Registries Designed for Purposes Other Than Safety

Registries may be designed to fulfill any number of other purposes, including examining comparative effectiveness, studying the natural history of a disease, providing evidence in support or national coverage decisions, or documenting quality improvement efforts. Although these registries may gather data on AEs and report those data (to regulatory authorities, manufacturers or others), not all data may necessarily be reported through the registry. Thus, the registry may not record all events, which would result in an imprecise, and possibly inaccurate, estimation of the true risk in the exposed population(s). A strength of comparative effectiveness registries, however, lies in the systematic collection of data for both the product of interest and concomitant, internal controls.

As an example of the limitations of assessing safety events in registries not designed for safety, a registry may be sponsored by a payer to collect data on every person receiving a certain medication. The purpose of the registry may be to assess prescribing practices and determine which patients are most likely to receive this product. The registry may also contain useful data on events experienced by patients exposed to the product, but may not be considered a comprehensive collection of safety data, or may provide information regarding a known risk or outcome rather than generating data that could identify a previously unappreciated event. Alternatively, a registry may be designed to study the effectiveness of a new product among a population subset, such as the elderly. The registry may be powered to analyze certain outcomes, such as rehospitalizations for a condition or quality of life, but may not be specifically of sufficient sample size to reliably assess overall safety in this population.

It is more challenging to accurately and precisely detect AEs of interest when a registry has not been designed for a specific safety purpose. In this situation, the registry must collect a wide range of data from patients to try to catch any possible events, or be adapted later should safety become a primary objective. Some events may be missed because the registry did not anticipate them and did not solicit data to identify them. Also, much the same as for registries designed specifically to detect AEs, some events may be so rare that they do not occur in the population enrolled in the registry or do not occur during the registry followup period. In these circumstances, registries can be designed to provide useful data on some of the events that may occur in the exposed population. Such data should not be considered complete or reliable for determining event rates, but, when the data are combined with safety data from other sources, trends or signals may become apparent within the data set.

4. Ad Hoc Data Pooling

One way to capitalize on data that, because they were collected for another purpose, may be insufficient for meaningful standalone analysis and interpretation due to study size or lack of comparators, is to pool the data with other similar data. As with any pooling of disparate data, the use of appropriate statistical techniques and the creation of a core data set for analysis are critically important, and are highly dependent on consistency in coding of treatments and events and in case identification.

It is essential to have an understanding of how every data set that will be used in a pooled analysis was created. For example, what is recorded in administrative health insurance claims depends largely on what benefits are covered and how medications are dispensed. Noncovered items generally are not recorded. For example, mental health services are often contracted for under separate coverage (so-called “carve-outs”) and not covered under traditional health insurance coverage; thus, the mental health consultations are not likely to be included in administrative databases derived from billing claims data. Also, some injectable medications (e.g., certain antibiotics) may be administered in the physician's office and thus would not be recorded through commonly used pharmacy reporting systems that are based on filling and refilling prescriptions. The absence of information may lead to false conclusions about safety issues. Also, AE data coded using the same coding dictionary (e.g., Medical Dictionary for Regulatory Activities, or MedDRA) may still be plagued by inconsistency in the application of coding guidelines and standards. Recoding of verbatim event reports may be required if feasible, prior to analysis. Depending upon the purpose for which the data were collected, data on the treatments of interest are not always recorded or are not recorded with the specificity needed to understand risk (e.g., branded vs. generic, dosage, route of administration, batch).

Another consideration is differential followup, including the duration and vigor of followup in the registries to be pooled. Particular care is needed when combining data sets from different European countries, since differences in medical practice and reimbursement may mean that superficially similar data may actually represent different subgroups of an overall disease population. Similar caution is also advisable when combining information from disparate health systems within a single country, as some treatments of interest may be noncovered benefits in some systems and consequently not recorded in that health system's records. An alternative to pooling data is to conduct meta-analyses of various studies using appropriate statistical and epidemiologic methods.

While the types of registries described above may not be individually of sufficient sample size to detect safety issues, combining data from registries for other purposes could significantly enhance the ability to identify and analyze safety signals across broader populations. Core data sets for AEs have been suggested for electronic health records systems and as part of national surveillance mechanisms (e.g., through distributed research networks). In such a network, each participating registry or data source collects a standardized core data set from which results can be aggregated to address specific surveillance questions. For example, there is significant national interest in understanding the long-term outcomes of orthopedic joint implants. Currently, there are several prominent registries in the United States with varying numbers of types of patients and types of implants. Many of these registries collect data for quality improvement purposes, but have sufficient data elements to potentially report on AEs. However, only by aggregating common data sets across many of these registries can a broadly representative population be evaluated and enough data accrued to understand the safety profile of specific types of devices in particular populations.

As described above, while not every registry is designed to evaluate safety, even registries designed for other purposes might contribute to aggregate information about potential harm from health care products or services. Yet many registries, especially disease registries, are conducted by nonregulated entities such as provider associations, academic institutions, and nonprofit research groups, whose role in AE reporting is unclear. Furthermore, sample sizes needed to understand safety signals are generally much larger than those needed to achieve useful information on quality of care or the natural history of certain diseases, and the safety analyses can require a high degree of statistical sophistication. Enrolling additional patients or committing additional resources for specialized analyses in order to achieve a general societal benefit through safety reporting is not feasible for most registries when the primary purpose is not safety. However, encouraging registries to participate in aggregation of data when such participation is at minimal cost and enhances the common good may be both reasonable and appropriate.

Many efforts are underway to improve the feasibility of broader safety reporting from both registries and electronic health records that serve other purposes. These efforts include recommending standardized core data sets for safety to enhance the aggregation of information in distributed networks, and making registries interoperable with facilitated safety reporting mechanisms or other registries designed for safety.³⁰ As facilitated reporting methodologies become more common and easier for registries to implement, there will be fewer reasons for nonparticipation. In addition, linkage of population-based registries, such as the Surveillance, Epidemiology and End Results (SEER) cancer registry program, with other data sources, such as Medicare, have proven invaluable for evaluating safety and other outcomes.

5. Signal Detection in Registries and Observational Studies

Although subject to debate, according to the World Health Organization definition, a safety signal is defined as “reported information on a possible causal relationship between an AE and a drug, the relationship being unknown or incompletely documented previously”³¹ Hauben and Aronson (2009) define a signal as “information that arises from one or multiple sources (including observations and experiments), which suggests a new potentially causal association, or a new aspect of a known association, between an intervention and an event or set of related events, either adverse or beneficial, which would command regulatory, societal or clinical attention, and is judged to be of sufficient likelihood to justify verificatory and when necessary, remedial actions.”³² The authors further posit that signals, following assessment, could subsequently be categorized as indeterminate, verified or refuted.

Additional attempts at defining or describing a safety signal for purposes of guiding product sponsors, regulators, and other researchers have come from various sources, including the Council for International Organizations of Medical Sciences (CIOMS), the FDA, and the United Kingdom's Medical and Healthcare products Regulatory Agency. Nelson and colleagues recently provided FDA with a comprehensive evaluation of signal detection methods for use in postmarketing surveillance, and included a discussion of “conventional Phase IV observational safety studies,” which would encompass registries, as part of a multipronged approach to surveillance.³³ They noted that despite a focus on automated health care data sources, such as large health care claims databases, for primary surveillance and as the basis for FDA's Sentinel Network, the need for more detailed data regarding exposure and outcome measurement, as well as collection of relevant confounder data, will require that prospective observation studies be conducted to address prespecified safety-related hypotheses.

Establishing a threshold of effect size and robustness of data that would justify action, such as initiation of additional studies, FDA action, or changes in payer coverage, remains an important question and is unlikely to be uniformly applicable to all products and situations. A report was issued in 2010 from the CIOMS Working Group VIII, whose main goal is to harmonize the development, application, and interpretation of signal detection methods for use with drugs, vaccines, and biologics and to provide practical advice.³⁴

Once a signal that warrants further evaluation is identified it is typically assessed based on the strength of the association between exposure and the event; biological plausibility; any evidence provided by dechallenge and rechallenge; the existence of experimental or animal models; and the nature, consistency, and quality of the data source.³⁵ Signals may present themselves as idiosyncratic events affecting a subset of the exposed population who are somehow susceptible, events related to the pharmacological action of the drug, or increased frequency of events normally occurring in the population (such as in the example of cardiovascular events and rofecoxib). Signals may involve the identification of novel risks, or new (or more refined) information regarding previously identified risks. If an event does appear to be product related further inquiry is required to examine whether the occurrence appears to be related to a specific treatment, a combination or sequence of treatments, or a particular dosage and/or duration of use. Events with long induction periods are particularly challenging for the ascription of a causal relationship, since there are likely to be many intervening factors, or confounders, that could account for the apparent signal.

The constant challenge is to separate a potential safety signal from the “noise,” or, in other words, to detect meaningful trends and to have a basis for evaluating whether the signal is something common to people who have the underlying condition for which treatment is being administered or whether it appears to be causally related to use of a particular product. All methods currently used for signal detection have their limitations. Attempts to use quantitative, and in some cases, automated signal detection methods as part of pharmacovigilance, including data mining using Bayesian algorithms or other disproportionality analyses, are hampered by confounders and other biases inherent to spontaneously reported data.^{36, 37} Other methodologies also attempt to identify trends over time and include potential patterns associated with other patient characteristics, such as concomitant drug exposures.

These methods of automated signal detection lack clinical context and only draw attention to deviations from independence between product exposure and events. No conclusions regarding causality can be drawn without a further qualitative and quantitative assessment of extrinsic factors (e.g., an artificial spike in reporting due to media attention) and potential confounders; in some cases, even with quantitative and qualitative assessments, the data may be insufficient to establish causality. Depending on the original data source, it may be impossible to address these issues within the database itself and either abstracted medical record data or prospective data collection may be required to gather reliable data. The long-term followup and longitudinal data generated by many registries merits particular methodological considerations, including how often to perform testing, what threshold is meaningful for a given event, and whether that threshold changes over time.

While some registries can serve as sources of initial safety signaling or hypothesis generation, they may also be used for further investigation of a signal generated from surveillance and quantitative analysis. As an example, existing data from the Swedish Coronary Angiography and Angioplasty Registry (SCAAR), sponsored by the Swedish Health Authorities, was used to look at long-term outcomes related to bare-metal and drug-eluting stents, once it became clear through FDA-designed and other registries in the postmarket setting that off-label use was very common and that the risk of restenosis and other long-term outcomes in the real-world patient population was not fully understood. Due to the existence of comprehensive national population registries in Sweden, researchers were able to reliably combine SCAAR data, which captured unselected, consecutive angiography and percutaneous coronary intervention procedure data, with vital status and hospitalization data, to examine fatality rates and cardiac events on a population level.³⁸ This use of procedure and national registries provides an example of how a registry that included a well-defined population allowed for safety assessments coincident with comparative effectiveness.

6. Potential Obligations for Registry Developers in Reporting Safety Issues

In considering what actual and potential obligations there are, or may be, for registries in product safety assessment, it is useful to separate the issues into several parts. First, there are two key questions that can be asked for each registry: (1) What is the role of registries not designed for safety purposes with respect to the search for AEs? and (2) What are the obligations, especially for those registries not sponsored by regulated manufacturers, to further investigate and report these events when found? As discussed above, registries can be classified by whether or not they were designed for a safety purpose, and also by whether or not they have specified regulatory obligations for reporting. Beyond these distinctions, several factors need to be considered including the ethical obligations of the registry developer, the technical limitations of the signal detection, and resource constraints.

Registries designed for safety assessment purposes should have a clear and deliberate plan in place, not only for detecting the signal of interest, but for handling unanticipated events and reporting them to appropriate authorities. Only in the case of registries supported by the regulated industries are rules for reporting drug or device AEs explicit. Therefore, it would be helpful if other registries would also formulate plans that ensure that appropriate information will reach the right stakeholders, either through reporting to the manufacturer or directly to the regulator, in a timely manner similar to those required by the regulated industries. There should not be two different standards for reporting information intended to safeguard the health and well-being of all.

Registries that are not designed specifically for safety assessment purposes, particularly those that are not sponsored by a manufacturer, raise more complex issues. While researchers have an obligation to the patients enrolled in any research activity to alert them should information regarding potential safety issues become known, it is less clear how far this obligation extends. In the United Kingdom, the General Medical Council includes in its advice on “Good Medical Practice” the requirement to “report suspected adverse drug reactions in accordance with the relevant reporting scheme.”³⁹ It is therefore clear that in the United Kingdom contributing to the safety profile of a medicine is regarded as part of the duties of a medical practitioner. During its review of research registries, an institutional review board (IRB) in the United States or ethics committee (EC) in Canada or the European Union may specify the creation of an explicit incidental findings plan prior to approval. Such a plan is often part of studies producing or compiling nonclinical imaging and genetic data. In addition, some investigators will have an obligation to report to an IRB or EC any unanticipated problems involving risks to subjects or others under the regulations on human research protections. In turn, IRBs and ECs have an obligation to report such incidents to relevant authorities.

At a minimum, all registries should ensure that standard reporting mechanisms for AE information are described in the registry's procedural documents. These mechanisms should also be explained to investigators and, where feasible, their reporting efforts should be facilitated. For example, all registries in the United States can make available to registry participants access to the MedWatch forms⁴⁰ and train them in the appropriate use of these forms to report spontaneous events. As described in Section 4 above, in the near future it should be possible for registries that collect data electronically to actually facilitate the reporting of AEs by linking with facilitated safety reporting mechanisms. This mechanism is attractive because it reduces the work of the investigator in generating the report and ensures that the report will go to a surveillance program prepared to investigate and manage both events and potential safety signals.

Obligations beyond facilitation are less clear. Furthermore, there are both technical and resource obstacles to thoroughly investigating potential signals, and risks that inaccurate and potentially injurious information will be generated. For example, publicizing product safety issues can result in some patients discontinuing use of potentially life-saving products regardless of the strength of the scientific evidence. As described earlier, registries designed for safety assessment should ideally have both adequate sample size and signal evaluation expertise in order to assess safety issues. Registries not designed for safety purposes may not have enough patients or statistical signal detection expertise to investigate potential signals, or may not have the financial resources to devote to unplanned analyses and investigations. It would seem that, at a minimum, registries not designed for safety purposes should use facilitated reporting (via training, providing forms, etc.) of individual events through standard channels to meet their ethical obligations, and that they should check with any institutions with which they are affiliated to determine whether they are subject to additional reporting requirements. However, should a registry identify potential signals through its own analyses, obligations arise.

While registries that are approved by IRBs report safety issues to those IRBs, incidental analytic findings, which may represent true or false signals, may need more definition and should best be further investigated and reported for the public good. One approach would be to report summary information to the relevant regulatory authority for further evaluation. To avoid doubt, registry developers should consider these issues carefully during the planning phase of a registry, and should explicitly define their practices and procedures for AE detection and reporting, their planned analyses of AEs, and how incidental analytic findings will be managed. Such a plan should lay out the extent to which registry owners will analyze their data for AEs, the timing of such analyses, what types of unanticipated issues will be investigated internally, what thresholds would merit action, and when information will be provided to regulators or other defined government entities, depending on the nature of the safety issue.

7. Summary

The ongoing challenge, in the use both of existing data and of prospective data collection efforts such as registries, is to cast a wide enough net to capture not only rare events, but also more common events and events that are not anticipated (i.e., not part of a preapproval or postapproval potential risk assessment). In some cases, existing registries may add additional data collection to address questions regarding possible AEs that arise after registry initiation. In addition, it must be considered that all observational data sources are only as strong as their ability to measure and control for potential biases, including confounding and misclassification.

Large registries, linkage and distributed network schemes, and sentinel surveillance are all tools being actively developed to create an integrated approach to medical product safety and specifically, to signal detection and verification.

In contributing to the evidence hierarchy surrounding the generation of signals for detection and confirmation of potential AEs, registries are likely to make their strongest contributions through: detection of novel AEs associated with product use as reported by treating physicians, which, in turn, constitutes a signal necessitating further study; gathering information about pregnant women and other hard-to-study subpopulations of product users; linking with additional data sources such as the Medicare-SEER data linkage, thereby broadening the range of questions that can be addressed beyond the constraints of data collected for a registry; and confirming or validating signals generated in other data, such as from automated signal generation in large claims databases. Ideally, a clear and prospective understanding among stakeholders is needed regarding whether and under what circumstances signal monitoring within registries is appropriate; the timing or periodicity of any such analyses; what should be done with the information once it is identified, and what, if any, are the ethical obligations to collect, analyze, and report safety information if doing so is not a planned objective of the registry, and if the registry sponsor is not directly required by regulation to conduct such reporting.

Thoughtfully designed registries can play important roles in these newly emerging strategies to use multiple available data sources to generate and strengthen hypotheses in product safety.

However, as with all data sources, it is important to assess the effects of registry design, the type of data, reason for the data collection, how the data were collected, and the generalizability to the target population, in order to assess the strengths, weaknesses, and validity of the results provided and their contribution to the knowledge of the safety profile of the medicine or device under study.

Case Examples for Chapter 19

References for Chapter 19

1.

Lecutier MA. Phocomelia and internal defects due to thalidomide. Br Med J. 1962 Dec 1;2(5317):1447–8. [PMC free article: PMC1926787] [PubMed: 13929004]

2.

Herbst AL, Ulfelder H, Poskanzer DC. Adenocarcinoma of the vagina. Association of maternal stilbestrol therapy with tumor appearance in young women. N Engl J Med. 1971 Apr 15;284(15):878–81. [PubMed: 5549830]

3.

Hartzema AG, Tilson HH, Chan KA, editors. Pharmacoepidemiology and Therapeutic Risk Management. Cincinnati: Harvey Whitney Books; 2008.

4.

Drazen JM, Rainey J, Begg H, et al. Adverse Drug Event Reporting: The Roles of Consumers and Health-Care Professionals: Workshop Summary. Washington, DC: National Academies Press; 2007.

5.

McClellan M. Drug safety reform at the FDA— pendulum swing or systematic improvement? N Engl J Med. 2007 Apr 26;356(17):1700–2. [PubMed: 17435081]

6.

Eland IA, Belton KJ, van Grootheest AC, et al. Attitudinal survey of voluntary reporting of adverse drug reactions. Br J Clin Pharmacol. 1999 Oct;48(4):623–7. [PMC free article: PMC2014371] [PubMed: 10583035]

7.

Moore N, Hall G, Sturkenboom M, et al. Biases affecting the proportional reporting ratio (PPR) in spontaneous reports pharmacovigilance databases: the example of sertindole. Pharmacoepidemiol Drug Saf. 2003 Jun;12(4):271–81. [PubMed: 12812006]

8.

Figueiras A, Herdeiro MT, Polonia J, et al. An educational intervention to improve physician reporting of adverse drug reactions: a cluster-randomized controlled trial. JAMA. 2006 Sep 6;296(9):1086–93. [PubMed: 16954488]

9.

European Medicines Agency. Guideline on the use of statistical signal detection methods in the Eudravigilance Data Analysis System. 2008. [September 30, 2013]. http:​//eudravigilance​.emea.europa.eu/human/docs/10646406en​.pdf.

10.

Finney DJ. The Design and Logic of a Monitor of Drug Use. J Chronic Dis. 1965 Jan;18:77–98. [PubMed: 14252273]

11.

U.S. Food and Drug Administration. The Sentinel Initiative: national strategy for monitoring medical product safety. May, 2008. [August 17, 2012]. www​.fda.gov/downloads​/Safety/FDAsSentinelInitiative​/UCM124701.pdf.

12.

Samuel FE Jr. Safe Medical Devices Act of 1990. Health Aff (Millwood). 1991 Spring;10(1):192–5. [PubMed: 2045049]

13.

Schmitt-Egenolf M. Psoriasis therapy in real life: the need for registries. Dermatology. 2006;213(4):327–30. [PubMed: 17135739]

14.

Tulner LR, Frankfort SV, Gijsen GJ, et al. Drug-drug interactions in a geriatric outpatient cohort: prevalence and relevance. Drugs Aging. 2008;25(4):343–55. [PubMed: 18361544]

15.

Hanley JA, Lippman-Hand A. If nothing goes wrong, is everything all right? Interpreting zero numerators. JAMA. 1983 Apr 1;249(13):1743–5. [PubMed: 6827763]

16.

Eypasch E, Lefering R, Kum CK, et al. Probability of adverse events that have not yet occurred: a statistical reminder. BMJ. 1995 Sep 2;311(7005):619–20. [PMC free article: PMC2550668] [PubMed: 7663258]

17.

Rogers WJ, Bowlby LJ, Chandra NC, et al. Treatment of myocardial infarction in the United States (1990 to 1993). Observations from the National Registry of Myocardial Infarction. Circulation. 1994 Oct;90(4):2103–14. [PubMed: 7923698]

18.

Odvina CV, Zerwekh JE, Rao DS, et al. Severely suppressed bone turnover: a potential complication of alendronate therapy. J Clin Endocrinol Metab. 2005 Mar;90(3):1294–301. [PubMed: 15598694]

19.

Tilson HH, Doi PA, Covington DL, et al. The antiretrovirals in pregnancy registry: a fifteenth anniversary celebration. Obstet Gynecol Surv. 2007 Feb;62(2):137–48. [PubMed: 17229330]

20.

Food and Drug Administration Amendments Act of 2007, Pub. L. No. 110-85 (2007), Title IX.

21.

Peck CC. FDA's position on the clozaril patient management system. Hosp Community Psychiatry. 1990;41(8):876–7. [PubMed: 2401475]

22.

Zeldis JB, Williams BA, Thomas SD, et al. S.T.E.P.S.: a comprehensive program for controlling and monitoring access to thalidomide. Clin Ther. 1999 Feb;21(2):319–30. [PubMed: 10211535]

23.

Biogen Idec. The TYSABRI TOUCH^® Prescribing Program. [September 30, 2013]. http://www​.tysabri.com​/safety-with-tysabri.xml.

24.

Honein MA, Lindstrom JA, Kweder SL. Can we ensure the safe use of known human teratogens?: The iPLEDGE test case. Drug Saf. 2007;30(1):5–15. [PubMed: 17194167]

25.

Kleinschmidt-DeMasters BK, Tyler KL. Progressive multifocal leukoencephalopathy complicating treatment with natalizumab and interferon beta-1a for multiple sclerosis. N Engl J Med. 2005 Jul 28;353(4):369–74. [PubMed: 15947079]

26.

Humbert M, Segal ES, Kiely DG, et al. Results of European post-marketing surveillance of bosentan in pulmonary hypertension. Eur Respir J. 2007 Aug;30(2):338–44. [PubMed: 17504794]

27.

Walker AM, Funch DP, Sulsky SI, et al. Patient factors associated with strut fracture in Bjork-Shiley 60 degrees convexo-concave heart valves. Circulation. 1995 Dec 1;92(11):3235–9. [PubMed: 7586309]

28.

Walker AM, Funch DP, Sulsky SI, et al. Manufacturing characteristics associated with strut fracture in Bjork-Shiley 60 degrees Convexo-Concave heart valves. J Heart Valve Dis. 1995 Nov;4(6):640–8. [PubMed: 8611980]

29.

Curtis JP, Luebbert JJ, Wang Y, et al. Association of physician certification and outcomes among patients receiving an implantable cardioverter-defibrillator. JAMA. 2009 Apr 22;301(16):1661–70. [PMC free article: PMC2805129] [PubMed: 19383957]

30.

The * ASTER Pilot Project: Improving the Reporting of Adverse Events. *ASTER: A Collaborative Study to Improve Drug Safety. [August 15, 2012]. www​.asterstudy.com.

31.

World Health Organization's International Drug Monitoring Programme. Uppsala Monitoring Centre (UMC). [August 15, 2012]. http://www​.who-umc.org/

32.

Hauben M, Aronson JK. Defining ‘signal’ and its subtypes in pharmacovigilance based on a systematic review of previous definitions. Drug Saf. 2009;32(2):99–110. [PubMed: 19236117]

33.

Nelson J, Cook A, Yu O. Evaluation of signal detection methods for us in prospective post-licensure medical product safety surveillance. Mar, 2009. [August 15, 2012]. http://www​.fda.gov/OHRMS​/DOCKETS/98fr/FDA-2009-N-0192-rpt​.pdf.

34.

Council for Interntational Organizations of Medical Sciences (CIOMS). Practical aspects of signal detection in pharmacovigilance. Report of CIOMS Working Groups VIII. Geneva: 2010.

35.

Meyboom RH, Egberts AC, Edwards IR, et al. Principles of signal detection in pharmacovigilance. Drug Saf. 1997 Jun;16(6):355–65. [PubMed: 9241490]

36.

Hauben M, Zhou X. Quantitative methods in pharmacovigilance: focus on signal detection. Drug Saf. 2003;26(3):159–86. [PubMed: 12580646]

37.

Szarfman A, Machado SG, O'Neill RT. Use of screening algorithms and computer systems to efficiently signal higher-than-expected combinations of drugs and events in the US FDA's spontaneous reports database. Drug Saf. 2002;25(6):381–92. [PubMed: 12071774]

38.

Lagerqvist B, James SK, Stenestrand U, et al. Long-term outcomes with drug-eluting stents versus bare-metal stents in Sweden. N Engl J Med. 2007 Mar 8;356(10):1009–19. [PubMed: 17296822]

39.

General Medical Council. Good Medical Practice. 2006. [August 15, 2012]. http://www​.gmc-uk.org​/guidance/good_medical_practice.asp.

40.

U.S. Food and Drug Administration. Reporting Serious Problems to the FDA. [August 15, 2012]. http://www​.fda.gov/Safety​/MedWatch/HowToReport/default.htm.