Appendix: Guidance for Industry - Good Pharmacovigilance Practices and Pharmacoepidemiologic Assessment

APPENDIX: GUIDANCE FOR INDUSTRY – GOOD PHARMACOVIGILANCE PRACTICES AND PHARMACOEPIDEMIOLOGIC ASSESSMENT

INTRODUCTION

This chapter provides guidance to industry on good pharmacovigilance practices and pharmacoepidemi-ologic assessment of observational data regarding drugs, including biological drug products (exclud-ing blood and blood components). Specifically, this chapter provides guidance on (1) safety signal identi-fication, (2) pharmacoepidemiologic assessment and safety signal interpretation and (3) pharmacovigilance plan development.

FDA’s guidance documents, including this guid-ance, do not establish legally enforceable respon-sibilities. Instead, guidances describe the Agency’s current thinking on a topic and should be viewed only as recommendations, unless specific regula-tory or statutory requirements are cited. The use of the word should in Agency guidances means that something is suggested or recommended, but not required.

BACKGROUND

PDUFA III’s Risk Management Guidance Goal

On June 12, 2002, Congress reauthorized, for the second time, the PDUFA III. In the context of PDUFA III, FDA agreed to satisfy certain performance goals. One of those goals was to produce guidance for industry on risk management activities for drug and biological products. As an initial step towards satis-fying that goal, FDA sought public comment on risk management. Specifically, FDA issued three concept papers. Each paper focused on one aspect of risk management, including (1) conducting pre-marketing risk assessment, (2) developing and implementing risk minimization tools and (3) performing post-marketing pharmacovigilance and pharmacoepidemi-ologic assessments. In addition to receiving numerous written comments regarding the three concept papers, FDA held a public workshop on April 9–11, 2003, to discuss the concept papers. FDA considered all of the comments received in developing three draft guidance documents on risk management activities. The draft guidance documents were published on May 5, 2004, and the public was provided with an opportunity to comment on them until July 6, 2004. FDA considered all of the comments received in producing the final guidance documents.

•   Pre-marketing risk assessment (pre-marketing guidance).

•   Development and use of risk minimization action plans (RiskMAP guidance).

•   Good pharmacovigilance practices and pharma-coepidemiologic assessment (pharmacovigilance guidance).

Overview of the Risk Management Guidances

Like the concept papers and draft guidances that preceded them, each of the three final guidance docu-ments focuses on one aspect of risk management. The Premarketing Guidance and the Pharmacovig-ilance Guidance focus on pre- and post-marketing risk assessment, respectively. The RiskMAP Guid-ance focuses on risk minimization. Together, risk assessment and risk minimization form what FDA calls risk management. Specifically, risk management is an iterative process for (1) assessing a product’s benefit–risk balance, (2) developing and implement-ing tools to minimize its risks while preserving its benefits, (3) evaluating tool effectiveness and reassessing the benefit–risk balance and (4) making adjustments, as appropriate, to the risk minimization tools to further improve the benefit–risk balance. This four-part process should be continuous throughout a product’s life cycle, with the results of risk assess-ment informing the sponsor’s decisions regarding risk minimization.

When reviewing the recommendations provided in this guidance, sponsors and applicants should keep the following points in mind:

·    Many recommendations in this guidance are not intended to be generally applicable to all products.

Industry already performs risk assessment and risk minimization activities for products during devel-opment and marketing. The Federal Food, Drug, and Cosmetic Act (FDCA) and FDA implement-ing regulations establish requirements for routine risk assessment and risk minimization (see e.g. FDA requirements for professional labelling and adverse event monitoring and reporting). As a result, many of the recommendations presented here focus on situa-tions when a product may pose a clinically important and unusual type or level of risk. To the extent possi-ble, we have specified in the text whether a recom-mendation is intended for all products or only this subset of products.

·    It is of critical importance to protect patients and their privacy during the generation of safety data and the development of risk minimization action plans.

During all risk assessment and risk minimization activities, sponsors must comply with applicable regulatory requirements involving human subjects research and patient privacy.

·    To the extent possible, this guidance conforms with FDA’s commitment to harmonize interna-tional definitions and standards as appropriate.

The topics covered in this guidance are being discussed in a variety of international forums. We are participating in these discussions and believe that, to the extent possible, the recommendations in this guidance reflect current thinking on related issues.

·    When planning risk assessment and risk minimiza-tion activities, sponsors should consider input from health care participants likely to be affected by these activities (e.g. from consumers, pharmacists and pharmacies, physicians, nurses and third party payers).

·    There are points of overlap among the three guidances.

We have tried to note in the text of each guidance when areas of overlap occur and when referencing one of the other guidances might be useful.

THE ROLE OF PHARMACOVIGILANCE AND PHARMACOEPIDEMIOLOGY IN RISK MANAGEMENT

Risk assessment during product development should be conducted in a thorough and rigorous manner; however, it is impossible to identify all safety concerns during clinical trials. Once a product is marketed, there is generally a large increase in the number of patients exposed, including those with co-morbid conditions and those being treated with concomitant medical products. Therefore, post-marketing safety data collection and risk assessment based on observational data are critical for evaluat-ing and characterizing a product’s risk profile and for making informed decisions on risk minimization.

This guidance document focuses on pharmacovigi-lance activities in the post-approval period. This guid-ance uses the term pharmacovigilance to mean all scientific and data gathering activities relating to the detection, assessment and understanding of adverse events. This includes the use of pharmacoepidemio-logic studies. These activities are undertaken with the goal of identifying adverse events and understand-ing, to the extent possible, their nature, frequency and potential risk factors.

Pharmacovigilance principally involves the iden-tification and evaluation of safety signals. In this guidance document, safety signal refers to a concern about an excess of adverse events compared to what would be expected to be associated with a product’s use. Signals can arise from post-marketing data and other sources, such as pre-clinical data and events associated with other products in the same phar-macologic class. It is possible that even a single well-documented case report can be viewed as a signal, particularly if the report describes a positive rechallenge or if the event is extremely rare in the absence of drug use. Signals generally indicate the need for further investigation, which may or may not lead to the conclusion that the product caused the event. After a signal is identified, it should be further assessed to determine whether it represents a potential safety risk and whether other action should be taken.

IDENTIFYING AND DESCRIBING SAFETY SIGNALS: FROM CASE REPORTS TO CASE SERIES

Good pharmacovigilance practice is generally based on acquiring complete data from spontaneous adverse event reports, also known as case reports. The reports are used to develop case series for interpretation.

Good Reporting Practice

Spontaneous case reports of adverse events submit-ted to the sponsor and FDA, and reports from other sources, such as the medical literature or clinical stud-ies, may generate signals of adverse effects of drugs. The quality of the report is critical for appropriate evaluation of the relationship between the product and adverse events. FDA recommends that sponsors make a reasonable attempt to obtain complete infor-mation for case assessment during initial contacts and subsequent follow-up, especially for serious events⁸ and encourages sponsors to use trained health care practitioners to query reporters. Computer-assisted interview technology, targeted questionnaires or other methods developed to target specific events can help focus the line of questioning. When the report is from a consumer, it is often important to obtain permission to contact the health care practitioner familiar with the patient’s adverse event to obtain further medical information and to retrieve relevant medical records, as needed.

FDA suggests that the intensity and method of case follow-up be driven by the seriousness of the event reported, the report’s origin (e.g. health care practitioner, patient and literature) and other factors. FDA recommends that the most aggressive follow-up efforts be directed towards serious adverse event reports, especially of adverse events not known to occur with the drug.

Characteristics of a Good Case Report

Good case reports include the following elements:

•   description of the adverse events or disease experi-ence, including time to onset of signs or symptoms;

•   suspected and concomitant product therapy details (i.e. dose, lot number, schedule, dates and duration), including over-the-counter medications, dietary supplements and recently discontinued medications;

•   patient characteristics, including demographic information (e.g. age, race and sex), baseline medi-cal condition before product therapy, co-morbid conditions, use of concomitant medications, rele-vant family history of disease and presence of other risk factors;

•   documentation of the diagnosis of the events, including methods used to make the diagnosis;

•   clinical course of the event and patient outcomes (e.g. hospitalization or death);9

•   relevant therapeutic measures and laboratory data at baseline, during therapy and subsequent to therapy, including blood levels, as appropriate;

•   information about response to dechallenge and rechallenge and

•   any other relevant information (e.g. other details relating to the event or information on benefits received by the patient, if important to the assess-ment of the event).

For reports of medication errors, good case reports also include full descriptions of the following when such information is available:

•   products involved [including the trade (propri-etary) and established (proper) name, manufacturer, dosage form, strength, concentration and type and size of container];

•   sequence of events leading up to the error;

•   work environment in which the error occurred and

•   types of personnel involved with the error, type(s) of error and contributing factors.

FDA recommends that sponsors capture in the case narrative section of a medication error report all appropriate information outlined in the National Coor-dinating Council for Medication Error Reporting and Prevention (NCC MERP) Taxonomy.¹⁰ Although sponsors are not required to use the taxonomy, FDA has found the taxonomy to be a useful tool to cate-gorize and analyze reports of medication errors. It provides a standard language and structure for medi-cation error-related data collected through reports.

Developing a Case Series

FDA suggests that sponsors initially evaluate a signal generated from post-marketing spontaneous reports through a careful review of the cases and a search for additional cases. Additional cases could be identified from the sponsor’s global adverse event databases, the published literature and other available databases, such as FDA’s AERS or vaccine adverse events reporting system (VAERS), using thorough database search strategies based on updated coding terminology [e.g. the Medical Dictionary for Regulatory Activities (MedDRA)]. When available, FDA recommends that standardized case definitions (i.e. formal criteria for including or excluding a case) be used to assess poten-tial cases for inclusion in a case series.¹¹ In general, FDA suggests that case-level review occur before other investigations or analyses. FDA recommends that emphasis usually be placed on review of seri-ous, unlabelled adverse events, although other events may warrant further investigation (see ‘SAFETY SIGNALS THAT MAY WARRANT FURTHER INVESTIGATION’ for more details).

As part of the case-level review, FDA suggests that sponsors evaluate individual case reports for clinical content and completeness and follow-up with reporters, as necessary. It is important to remove any duplicate reports. In assessing case reports, FDA recommends that sponsors look for features that may suggest a causal relationship between the use of a product and the adverse event, including

•   occurrence of the adverse event in the expected time (e.g. type 1 allergic reactions occurring within days of therapy and cancers developing after years of therapy);

•   absence of symptoms related to the event before exposure;

•   evidence of positive dechallenge or positive rechal-lenge;

•   consistency of the event with the established phar-macological/toxicological effects of the product, or for vaccines, consistency with established infec-tious or immunologic mechanisms of injury;

•   consistency of the event with the known effects of other products in the class;

•   existence of other supporting evidence from pre-clinical studies, clinical trials and/or pharmacoepi-demiologic studies and

•   absence of alternative explanations for the event (e.g. no concomitant medications that could contribute to the event and no co- or pre-morbid medical conditions).

Confounded cases are common, especially among patients with complicated medical conditions. They (i.e. cases with adverse events that have possible etiologies other than the product of concern) could still represent adverse effects of the product under review. FDA recommends that sponsors carefully evaluate these cases and not routinely exclude them. Separate analyses of unconfounded cases may be useful.

For any individual case report, it is rarely possi-ble to know with a high level of certainty whether the event was caused by the product. To date, there are no internationally agreed upon standards or criteria for assessing causality in individual cases, especially for events that often occur spontaneously (e.g. stroke and pulmonary embolism). Rigorous pharmacoepi-demiologic studies, such as case–control studies and cohort studies with appropriate follow-up, are usually employed to further examine the potential association between a product and an adverse event.

FDA does not recommend any specific categoriza-tion of causality, but the categories probable, possi-ble or unlikely have been used previously.¹² If a causality assessment is undertaken, FDA suggests that the causal categories be specified and described in sufficient detail to understand the underlying logic in the classification.

If the safety signal relates to a medication error, FDA recommends that sponsors report all known contributing factors that led to the event. Many refer-ences are available to assist sponsors in capturing a complete account of the event (Cohen, 1999). FDA recommends that sponsors follow-up to the extent possible with reporters to capture a complete account of the event, focusing on the medication use systems (e.g. prescribing/order process, dispensing process and administration process). This data may be infor-mative in developing strategies to minimize future errors.

Summary Descriptive Analysis of a Case Series

In the event that one or more cases suggest a safety signal warranting additional investigation, FDA recommends that a case series be assembled and descriptive clinical information be summarized to characterize the potential safety risk and, if possi-ble, to identify risk factors. A case series commonly includes an analysis of the following:

•   the clinical and laboratory manifestations and course of the event;

•   demographic characteristics of patients with events (e.g. age, gender and race);

•   exposure duration;

•   time from initiation of product exposure to the adverse event;

•   doses used in cases, including labelled doses, greater than labelled doses and overdoses;

•   use of concomitant medications;

•   the presence of co-morbid conditions, particularly those known to cause the adverse event, such as underlying hepatic or renal impairment;

•   the route of administration (e.g. oral vs. paren-teral);

•   lot numbers, if available, for products used in patients with events and

•   changes in event reporting rate over calendar time or product life cycle.

Use of Data Mining to Identify Product–Event Combinations

At various stages of risk identification and assess-ment, systematic examination of the reported adverse events by using statistical or mathematical tools, or the so-called data mining, can provide additional infor-mation about the existence of an excess of adverse events reported for a product. By applying data mining techniques to large adverse event databases, such as FDA’s AERS or VAERS, it may be possi-ble to identify unusual or unexpected product–event combinations warranting further investigation. Data mining can be used to augment existing signal detec-tion strategies and is especially useful for assess-ing patterns, time trends and events associated with drug–drug interactions. Data mining is not a tool for establishing causal attributions between products and adverse events.

The methods of data mining currently in use usually generate a score comparing (1) the fraction of all reports for a particular event (e.g. liver failure) for a specific drug (i.e. the ‘observed reporting fraction’) with (2) the fraction of reports for the same partic-ular event for all drugs (i.e. ‘the expected reporting fraction’) (Evans, 2000). This analysis can be refined by adjusting for aspects of reporting (e.g. the report-ing year) or characteristics of the patient (e.g. age or gender) that might influence the amount of reporting. In addition, it may be possible to limit data mining to an analysis for drugs of a specific class or for drugs that are used to treat a particular disease.

The score (or statistic) generated by data mining quantifies the disproportionality between the observed and expected values for a given product–event combi-nation. This score is compared to a threshold that is chosen by the analyst. A potential excess of adverse events is operationally defined as any product-event combination with a score exceeding the spec-ified threshold. When applying data mining to large databases (such as AERS), it is not unusual for a product to have several product–event combinations with scores above a specified threshold. The lower the threshold, the greater the likelihood that more combi-nations will exceed the threshold and will warrant further investigation.

Several data mining methods have been described and may be worth considering, such as the Multi-Item Gamma Poisson Shrinker (MGPS) algorithm (DuMouchel and Pregiborn, 2001; Szarfman et al., 2002), the Proportional Reporting Ratio (PRR) method (Evans, 1998; 2000) and the Neural Network approach (Bate et al., 1998). Except when the observed number of cases with the drug event combination is small (e.g. less than 20) or the expected number of cases with the drug event combination is <1, the MGPS and PRR methods will generally identify similar drug-event combinations for further investigation.

Although all of these approaches are inherently exploratory or hypothesis generating, they may provide insights into the patterns of adverse events reported for a given product relative to other products in the same class or to all other products. FDA exercises caution when making such comparisons, because voluntary adverse event reporting systems such as AERS or VAERS are subject to a variety of reporting biases (e.g. some observations could reflect concomi-tant treatment, not the product itself, and other factors, including the disease being treated, other co-morbidities or unrecorded confounders, may cause the events to be reported). In addition, AERS or VAERS data may be affected by the submission of incomplete or duplicate reports, under-reporting or reporting stim-ulated by publicity or litigation. As reporting biases may differ by product and change over time, and could change differently for different events, it is not possible to predict their impact on data mining scores.

Use of data mining techniques is not a required part of signal identification or evaluation. If data mining results are submitted to FDA, they should be presented in the larger appropriate clinical epidemio-logical context. This should include (1) a description of the database used, (2) a description of the data mining tool used (e.g. statistical algorithm, and the drugs, events and stratifications selected for the anal-yses) or an appropriate reference and (3) a careful assessment of individual case reports and any other relevant safety information related to the particular drug-event combination of interest (e.g. results from pre-clinical, clinical, pharmacoepidemiologic or other available studies).

Safety Signals that may Warrant  Further Investigation

FDA believes that the methods described above will permit a sponsor to identify and preliminarily char-acterize a safety signal. The actual risk to patients cannot be known from these data because it is not possible to characterize all events definitively and because there is invariably underreporting of some extent and incomplete information about duration of therapy, numbers treated and so on. Safety signals that may warrant further investigation may include, but are not limited to, the following:

•   new unlabelled adverse events, especially if serious;

•   an apparent increase in the severity of a labelled event;

•   occurrence of serious events thought to be extremely rare in the general population;

•   new product–product, product–device, product– food or product–dietary supplement interactions;

•   identification of a previously unrecognized at-risk population (e.g. populations with specific racial or genetic pre-dispositions or co-morbidities);

•   confusion about a product’s name, labelling, packaging or use;

•   concerns arising from the way a product is used (e.g. adverse events seen at higher than labelled doses or in populations not recommended for treatment);

•   concerns arising from potential inadequacies of a currently implemented risk minimization action plan (e.g. reports of serious adverse events that appear to reflect failure of a RiskMAP goal)14 and

•   other concerns identified by the sponsor or FDA.

Putting the Signal into Context: Calculating Reporting Rates Versus Incidence Rates

If a sponsor determines that a concern about an excess of adverse events or safety signal warrants further investigation and analysis, it is important to put the signal into context. For this reason, calcu-lations of the rate at which new cases of adverse events occur in the product-exposed population (i.e., the incidence rate) are the hallmark of pharmacoepi-demiologic risk assessment. In pharmacoepidemio-logic studies (see ‘PHARMACOEPIDEMIOLOGIC STUDIES’), the numerator (number of new cases) and denominator (number of exposed patients and time of exposure or, if known, time at risk) may be readily ascertainable. In contrast, for spontaneously reported events, it is not possible to identify all cases because of under-reporting, and the size of the population at risk is at best an estimate. Limitations in national denominator estimates arise because accurate national estimates of the number of patients exposed to a medical product and their duration of exposure may not be available;

·   it may be difficult to exclude patients who are not at risk for an event, for example, because their exposure is too brief or their dose is too low¹⁵ and

·   a product may be used in different populations for different indications, but use estimates are not available for the specific population of interest.

Although we recognize these limitations, we recommend that sponsors calculate crude adverse event reporting rates as a valuable step in the inves-tigation and assessment of adverse events. FDA suggests that sponsors calculate reporting rates by using the total number of spontaneously reported cases in the United States in the numerator and estimates of national patient exposure to product in the denomina-tor (Rodriguez et al., 2001).¹⁶ FDA recommends that whenever possible, the number of patients or person time exposed to the product nationwide be the esti-mated denominator for a reporting rate. FDA suggests that other surrogates for exposure, such as numbers of prescriptions or kilograms of product sold, only be used when patient-level estimates are unavailable. FDA recommends that sponsors submit a detailed explanation of the rationale for selection of a denom-inator and a method of estimation.

Comparisons of reporting rates and their tempo-ral trends can be valuable, particularly across similar products or across different product classes prescribed for the same indication. However, such comparisons are subject to substantial limitations in interpretation because of the inherent uncertainties in the numera-tor and denominator used. As a result, FDA suggests that a comparison of two or more reporting rates be viewed with extreme caution and generally consid-ered exploratory or hypothesis generating. Reporting rates can by no means be considered incidence rates, for either absolute or comparative purposes.

To provide further context for incidence rates or reporting rates, it is helpful to have an estimate of the background rate of occurrence for the event being evaluated in the general population or, ideally, in a subpopulation with characteristics similar to that of the exposed population (e.g. pre-menopausal women and diabetics). These background rates can be derived from (1) national health statistics, (2) published medi-cal literature or (3) ad hoc studies, particularly of subpopulations, using large automated databases or ongoing epidemiologic investigations with primary data collection. FDA suggests that comparisons of incidence rates or reporting rates to background rate estimates take into account potential differences in the data sources, diagnostic criteria and duration of time at risk.

Although the extent of under-reporting is unknown, it is usually assumed to be substantial and may vary according to the type of product, seriousness of the event, population using the product and other factors. As a result, a reporting rate higher than the back-ground rate may, in some cases, be a strong indica-tor that the true incidence rate is sufficiently high to be of concern. However, many other factors affect the reporting of product-related adverse events (e.g. publicity and newness of product to the market), and these factors should be considered when interpreting a high reporting rate. Also, because of under-reporting, the fact that a reporting rate is less than the back-ground rate does not necessarily show that the prod-uct is not associated with an increased risk of an adverse event.

BEYOND CASE REVIEW: INVESTIGATING A SIGNAL THROUGH OBSERVATIONAL STUDIES

FDA recognizes that there are a variety of meth-ods for investigating a safety signal. Signals warrant-ing additional investigation can be further evaluated through carefully designed non-randomized observa-tional studies of the product’s use in the ‘real world’ and randomized trials. The Premarketing Guidance discusses many types of randomized trials, includ-ing the large simple safety study, which is a risk assessment method that could be used either pre- or post-approval.

This document focuses on three types of non-randomized observational studies: (1) pharmacoepi-demiologic studies, (2) registries and (3) surveys. By focusing this guidance on certain risk assessment methods, we do not intend to advocate the use of these approaches over others. FDA encourages spon-sors to consider all methods to evaluate a particular safety signal. FDA recommends that sponsors choose the method best suited to the particular signal and research question of interest. Sponsors planning to evaluate a safety signal are encouraged to communi-cate with FDA as their plans progress.

Pharmacoepidemiologic Studies

Pharmacoepidemiologic studies can be of vari-ous designs, including cohort (prospective or retrospective), case–control, nested case–control, case–crossover or other models.¹⁷ The results of such studies may be used to characterize one or more safety signals associated with a product or may examine the natural history of a disease or drug utilization patterns. Unlike a case series, a pharmacoepidemiologic study which is designed to assess the risk attributed to a drug exposure has a protocol and control group and tests pre-specified hypotheses. Pharmacoepidemi-ologic studies can allow for the estimation of the relative risk of an outcome associated with a product, and some (e.g. cohort studies) can also provide esti-mates of risk (incidence rate) for an adverse event. Sponsors can initiate pharmacoepidemiologic studies at any time. They are sometimes started at the time of initial marketing, based on questions that remain after review of the pre-marketing data. More often, however, they are initiated when a safety signal has been identified after approval. Finally, there may also be occasions when a pharmacoepidemiologic study is initiated before marketing (e.g. to study the natural history of disease or patterns of product use or to estimate background rates for adverse events).

For uncommon or delayed adverse events, phar-macoepidemiologic studies may be the only practical choice for evaluation, even though they can be limited by low statistical power. Clinical trials are impractical in almost all cases when the event rates of concern are less common than 1:2000–3000 (an exception may be larger trials conducted for some vaccines, which could move the threshold to 1:10 000). It may also be difficult to use clinical trials: (1) to evaluate a safety signal associated with chronic exposure to a product, exposure in popu-lations with co-morbid conditions or taking multiple concomitant medications or (2) to identify certain risk factors for a particular adverse event. On the other hand, for evaluation of more common events, which are seen relatively often in untreated patients, clinical trials may be preferable to observational studies.

Because pharmacoepidemiologic studies are obser-vational in nature, they may be subject to confound-ing, effect modification and other bias, which may make results of these types of studies more difficult to interpret than the results of clinical trials. Some of these problems can be surmounted when the relative risk to exposed patients is high.

Because different products pose different benefit– risk considerations (e.g. seriousness of the disease being treated, nature and frequency of the safety signal under evaluation), it is impossible to delin-eate a universal set of criteria for the point at which a pharmacoepidemiologic study should be initiated, and the decision should be made on a case-by-case basis. When an important adverse event-product asso-ciation leads to questions on the product’s benefit– risk balance, FDA recommends that sponsors consider whether the particular signal should be addressed with one or more pharmacoepidemiologic studies. If a sponsor determines that a pharmacoepidemiologic study is the best method for evaluating a particu-lar signal, the design and size of the proposed study would depend on the objectives of the study and the expected frequency of the events of interest.

When performing a pharmacoepidemiologic study, FDA suggests that investigators seek to minimize bias and to account for possible confounding. Confounding by indication is one example of an important concern in performing a pharmacoepidemiologic study (Strom, 2000). Because of the effects of bias, confounding or effect modification, pharmacoepidemiologic studies evaluating the same hypothesis may provide different or even conflicting results. It is almost always prudent to conduct more than one study, in more than one environment and even use different designs. Agree-ment of the results from more than one study helps to provide reassurance that the observed results are robust.

There are many references describing methodologies for pharmacoepidemiologic studies, discussing their strengths and limitations (Strom, 2000) and provid-ing guidelines to facilitate the conduct, interpretation and documentation of such studies.  Consequently, this guidance document does not comprehensively address these topics. However, a protocol for a pharmacoepidemiologic study generally includes

•   clearly specified study objectives;

•   a critical review of the literature and

•   a detailed description of the research methods, including

o   the population to be studied;

o   the case definitions to be used;

o   the data sources to be used (including a ratio-nale for data sources if from outside the United States);

o   the projected study size and statistical power calculations and

o   the methods for data collection, management and analysis.

Depending on the type of pharmacoepidemiologic study planned, there are a variety of data sources that may be used, ranging from the prospective collection of data to the use of existing data, such as data from previously conducted clinical trials or large databases. In recent years, many pharmacoepidemiologic studies have been conducted in automated claims databases (e.g. HMO and Medicaid) that allow retrieval of records on product exposure and patient outcomes. In addi-tion, recently, comprehensive electronic medical record databases have also been used for studying drug safety issues. Depending on study objectives, factors that may affect the choice of databases include the following:

•   demographic characteristics of patients enrolled in the health plans (e.g. age and geographic location);

•   turnover rate of patients in the health plans;

•   plan coverage of the medications of interest;

•   size and characteristics of the exposed population available for study;

•   availability of the outcomes of interest;

•   ability to identify conditions of interest using stan-dard medical coding systems [e.g. International  Classification of Diseases (ICD-9)], procedure codes or prescriptions that could be used as markers;

•   access to medical records and

•   access to patients for data not captured electroni-cally.

For most pharmacoepidemiologic studies, FDA recommends that sponsors validate diagnostic find-ings through a detailed review of at least a sample of medical records. If the validation of the specific outcome or exposure of interest using the proposed database has been previously reported, FDA recom-mends that the literature supporting the validity of the proposed study be submitted for review.

FDA encourages sponsors to communicate with the Agency when pharmacoepidemiologic studies are being developed.

Registries

The term registry as used in pharmacovigilance and pharmacoepidemiology can have varied meanings. In this guidance document, a registry is ‘an organized system for the collection, storage, retrieval, analy-sis, and dissemination of information on individual persons exposed to a specific medical intervention who have either a particular disease, a condition (e.g., a risk factor) that pre-disposes [them] to the occur-rence of a health-related event, or prior exposure to substances (or circumstances) known or suspected to cause adverse health effects’.¹⁹ Whenever possible, a control or comparison group should be included (i.e. individuals with a disease or risk factor who are not treated or are exposed to medical interventions other than the intervention of interest).

Through the creation of registries, a sponsor can evaluate safety signals identified from spontaneous case reports, literature reports or other sources and evaluate factors that affect the risk of adverse outcomes, such as dose, timing of exposure or patient characteristics.²¹ Registries can be particularly useful for

•   collecting outcome information not available in large automated databases and

•   collecting information from multiple sources (e.g. physician records, hospital summaries, pathol-ogy reports and vital statistics), particularly when patients receive care from multiple providers over time.

A sponsor can initiate a registry at any time. It may be appropriate to initiate the registry at or before initial marketing, when a new indication is approved or when there is a need to evaluate safety signals identified from spontaneous case reports. In deciding whether to establish a registry, FDA recommends that a sponsor consider the following factors:

•   the types of additional risk information desired;

•   the attainability of that information through other methods and

•   the feasibility of establishing the registry.

Sponsors electing to initiate a registry should develop written protocols that provide (1) objectives for the registry, (2) a review of the literature and (3) a summary of relevant animal and human data. FDA suggests that protocols also contain detailed descrip-tions of (1) plans for systematic patient recruit-ment and follow-up, (2) methods for data collection, management and analysis and (3) conditions under which the registry will be terminated. A registry-based monitoring system should include carefully designed data collection forms to ensure data quality, integrity and validation of registry findings against a sample of medical records or through interviews with health care providers. FDA recommends that the size of the registry and the period during which data will be collected be consistent with the safety questions under study, and we encourage sponsors to discuss their registry development plans with FDA.

Surveys

Patient or health care provider surveys can gather information to assess, for example

•   a safety signal;

•   knowledge about labelled adverse events;

•   use of a product as labelled, particularly when the indicated use is for a restricted population or numerous contraindications exist;

•   compliance with the elements of a RiskMAP (e.g. whether or not a Medication Guide was provided at the time of product dispensing) and22

•   confusion in the practicing community over sound-alike or look-alike trade (or proprietary) names.

Like a registry, a survey can be initiated by a sponsor at any time. It can be conducted at the time of initial marketing (i.e., to fulfill a post-marketing commit-ment) or when there is a desire to evaluate safety signals identified from spontaneous case reports.

FDA suggests that sponsors electing to initiate a survey develop a written protocol that provides objec-tives for the survey and a detailed description of the research methods, including (1) patient or provider recruitment and follow-up, (2) projected sample size and (3) methods for data collection, management and analysis.²³ FDA recommends that a survey-based monitoring system includes carefully designed survey instruments and validation of survey findings against a sample of medical or pharmacy records or through interviews with health care providers, whenever possi-ble. FDA recommends that survey instruments be vali-dated or piloted before implementation. FDA suggests that sponsors consider whether survey translation and cultural validation would be important.Sponsors are encouraged to discuss their survey development plans with FDA.

INTERPRETING SAFETY SIGNALS: FROM SIGNAL TO POTENTIAL SAFETY RISK

After identifying a safety signal, FDA recommends that a sponsor conduct a careful case level review and summarize the resulting case series descriptively. To help further characterize a safety signal, a spon-sor can also (1) employ data mining techniques and calculate reporting rates for comparison to back-ground rates. Based on these findings and other avail-able data (e.g. from pre-clinical or other sources), FDA suggests that a sponsor consider further study (e.g. observational studies) to establish whether or not a potential safety risk exists.

When evaluation of a safety signal suggests that it may represent a potential safety risk, FDA recommends that a sponsor submit a synthesis of all available safety information and analyses performed, ranging from pre-clinical findings to current observa-tions. This submission should include the following:

•   spontaneously reported and published case reports, with denominator or exposure information to aid interpretation;

•   background rate for the event in general and specific patient populations, if available;

•   relative risks, odds ratios or other measures of association derived from pharmacoepidemiologic studies;

•   biologic effects observed in pre-clinical studies and pharmacokinetic or pharmacodynamic effects;

•   safety findings from controlled clinical trials and

•   general marketing experience with similar products in the class.

After the available safety information is presented and interpreted, it may be possible to assess the degree of causality between use of a product and an adverse event. FDA suggests that the sponsor’s submission provides an assessment of the benefit–risk balance of the product for the population of users as a whole and for identified at-risk patient populations and, if appropriate, (1) propose steps to further investigate the signal through additional studies and (2) propose risk minimization actions.²⁴ FDA will make its own assessment of the potential safety risk posed by the signal in question, taking into account the information provided by the sponsor and any additional relevant information known to FDA (e.g. information on other products in the same class) and will communicate its conclusions to the sponsor whenever possible. Factors that are typically considered include:

•   strength of the association (e.g. relative risk of the adverse event associated with the product);

•   temporal relationship of product use and the event;

•   consistency of findings across available data sources;

•   evidence of a dose-response for the effect;

•   biologic plausibility;

•   seriousness of the event relative to the disease being treated;

•   potential to mitigate the risk in the population;

•   feasibility of further study using observational or controlled clinical study designs and

•   degree of benefit the product provides, including availability of other therapies.

As noted in ‘BACKGROUND’, risk management is an iterative process and steps to further investigate a potential safety risk, assess the product’s benefit–risk balance and implement risk minimization tools would best occur in a logical sequence, not simultaneously. Not all steps may be recommended, depending on the results of earlier steps.²⁵ FDA recommends that assessment of causality and of strategies to minimize product risk occur on an ongoing basis, taking into account the findings from newly completed studies.

BEYOND ROUTINE PHARMACOVIGILANCE: DEVELOPING A PHARMACOVIGILANCE PLAN

For most products, routine pharmacovigilance (i.e. compliance with applicable post-market requirements under the FDCA and FDA implementing regula-tions) is sufficient for post-marketing risk assessment. However, in certain limited instances, unusual safety risks may become evident before approval or after a product is marketed that could suggest that consid-eration by the sponsor of a pharmacovigilance plan may be appropriate. A pharmacovigilance plan is a plan developed by a sponsor that is focused on detect-ing new safety risks and/or evaluating already iden-tified safety risks. Specifically, a pharmacovigilance plan describes pharmacovigilance efforts above and beyond routine post-marketing spontaneous reporting and is designed to enhance and expedite the spon-sor’s acquisition of safety information. The develop-ment of pharmacovigilance plans may be useful at the time of product launch or when a safety risk is iden-tified during product marketing. FDA recommends that a sponsor’s decision to develop a pharmacovigi-lance plan be based on scientific and logistical factors, including the following:

•   the likelihood that the adverse event represents a potential safety risk;

•   the frequency with which the event occurs (e.g. incidence rate, reporting rate or other measures available);

•   the severity of the event;

•   the nature of the population(s) at risk;

•   the range of patients for which the product is indicated (broad range or selected populations only) and

•   the method by which the product is dispensed (through pharmacies or performance linked systems only).

A pharmacovigilance plan may be developed by itself or as part of a RiskMAP, as described in the RiskMAP Guidance. Sponsors may meet with representatives from the appropriate Office of New Drugs review division and the Office of Drug Safety in CDER, or the appropriate Product Office and the Division of Epidemiology, Office of Biostatistics and Epidemi-ology in CBER regarding the specifics of a given product’s pharmacovigilance plan.

FDA believes that for a product without safety risks identified pre- or post-approval and for which at-risk populations are thought to have been adequately studied, routine spontaneous reporting will be suffi-cient for post-marketing surveillance. On the other hand, pharmacovigilance plans may be appropriate for products for which (1) serious safety risks have been identified pre-or post-approval or (2) at-risk populations have not been adequately studied. Spon-sors may discuss with the Agency the nature of the safety concerns posed by such a product and the determination whether a pharmacovigilance plan is appropriate.

A pharmacovigilance plan could include one or more of the following elements:

•   submission of specific serious adverse event reports in an expedited manner beyond routine required reporting (i.e. as 15-day reports);

•   submission of adverse event report summaries at more frequent, pre-specified intervals (e.g. quar-terly rather than annually);

•   active surveillance to identify adverse events that may or may not be reported through passive surveillance. Active surveillance can be (1) drug based: identifying adverse events in patients taking certain products; (2) setting based: identifying adverse events in certain health care settings where they are likely to present for treatment (e.g. emer-gency departments etc.) or (3) event based: iden-tifying adverse events that are likely to be asso-ciated with medical products (e.g. acute liver failure);

•   additional pharmacoepidemiologic studies (e.g. in automated claims databases or other databases) using cohort, case–control or other appropriate study designs (see ‘BEYOND CASE REVIEW: INVESTIGATING A SIGNAL THROUGH OBSERVATIONAL STUDIES’);

•   creation of registries or implementation of patient or health care provider surveys (see ‘BEYOND CASE REVIEW: INVESTIGATING A SIGNAL THROUGH OBSERVATIONAL STUDIES’) and  additional controlled clinical trials.

As data emerges, FDA recommends that a sponsor re-evaluate the safety risk and the effectiveness of its pharmacovigilance plan. Such re-evaluation may result in revisions to the pharmacovigilance plan for a product. In some circumstances, FDA may decide to bring questions on potential safety risks and pharmacovigilance plans before its Drug Safety and Risk Management Advisory Committee or the FDA Advisory Committee dealing with the specific prod-uct in question. Such committees may be convened when FDA seeks (1) general advice on the design of pharmacoepidemiologic studies, (2) comment on specific pharmacoepidemiology studies developed by sponsors or FDA for a specific product and safety question or (3) advice on the interpretation of early signals from a case series and on the need for further investigation in pharmacoepidemiologic stud-ies. Although additional information is being devel-oped, sponsors working with FDA can take interim actions to communicate information about potential safety risks (e.g. through labelling) to minimize the risk to users of the product.