Guidelines for good pharmacoepidemiology practice (GPP)

Pharmacoepidemiologic studies provide valuable information about the health effects of healthcare products. The ISPE Guidelines for Good Pharmacoepidemiology Practice (GPP) are intended to assist investigators with issues pertaining to the planning, conduct, and interpretation of pharmacoepidemiologic research. This paper represents the fourth version and supersedes previous versions. While the overall structure and nature of the GPP has been preserved in the current revision, new sections have been added, and the text has been updated to reflect current practice. Pharmacoepidemiology is being used increasingly to evaluate health care systems, interventions, and health-related behaviors. Pharmacoepidemiology is the scientific backbone of therapeutic risk management—the process of assessing a product's benefits and risks, and developing, implementing, and evaluating strategies to enhance the overall balance of such benefits and risks. Pharmacoepidemiology is also the scientific backbone of comparative effectiveness research (CER). These guidelines are intended to address these activities and other pharmacoepidemiologic studies. The GPP propose essential practices and procedures that should be considered to help ensure the quality and integrity of pharmacoepidemiologic research and to provide adequate documentation of research methods and results. The GPP do not prescribe specific research methods nor will adherence to guidelines guarantee valid research. The GPP are intended to apply broadly to all types of pharmacoepidemiologic research, including feasibility assessments, validation studies, descriptive studies, and etiologic investigations, and all of their related activities from design through publication. Therapeutic risk management activities provide a formal framework in which medicine, pharmacoepidemiology, and public health are integrated in the development and life-cycle management of healthcare products. Pharmacoepidemiology is the core science of risk assessment and the evaluation of the effectiveness of risk minimization interventions. Therefore, the GPP also support risk management activities. In a similar fashion, pharmacoepidemiology is the core discipline of CER. Such activities are based on problems from clinical medicine, by using rigorous methods to compare the outcomes of two or more therapeutic alternatives. Thereby, comparative effectiveness activities are defined by their research questions rather than by their methods. a statement of research objectives, specific aims, and rationale; Research objectives describe the knowledge or information to be gained from the study. Specific aims list key exposures and outcomes of interest, and any hypotheses to be evaluated. The protocol should distinguish between a limited number of a priori research hypotheses and hypotheses that are generated based on knowledge of the source data. The rationale explains how achievement of the specific aims will further the research objectives. The research question may be phrased by using the PICOT template; population, intervention, comparator, outcome, and timing. a critical review of the literature to evaluate pertinent information and gaps in knowledge; The literature review should describe specific gaps in knowledge that the study is intended to fill. The literature review might encompass relevant animal and human experiments, clinical studies, vital statistics, and previous epidemiologic studies. The literature review should also cite the findings of similar studies, and the expected contribution of the current study. the overall research design and reasons for choosing the proposed study design; Research designs include, for example, case–control, cohort, cross-sectional, nested case–control, self-controlled, randomized trials or hybrid designs. Any feasibility or pilot work that informed the choice of design should be described here. the population or sample to be studied; The population is defined in terms of persons, place, time period, and selection criteria. The rationale for the inclusion and exclusion criteria and their impact on the number of subjects available for analysis should be described, if known. If any sampling from a defined population is undertaken, description of the population and details of sampling methods should be provided. Some justification should be given to support that the necessary study size is actually attainable from the given data source or design. This could be data from informal queries or pilot studies. Considerations of generalizability from the study population to those actually receiving the drug may be voiced here. the strategies and data sources for determining exposures, health outcomes, and all other variables relevant to the study objectives, such as potential confounding variables and effect measure modifiers; Data sources might include, for example, questionnaires, hospital discharge files, abstracts of primary clinical records, clinical databases, electronic medical records, ad hoc data collection, administrative records such as eligibility files, prescription drug files, biological measurements, exposure/work history record reviews, or exposure/disease registries. If the study uses secondary data, the name of the data source should be included (e.g., Medicare, CPRD, and MarketScan). Use validated instruments and measures whenever such exist and describe the validation method and summarize what is known about the completeness and validity of those instruments and measures. If data collection methods or instruments will be tested in a pilot study, plans for the pilot study should be described. Any procedures to be used to validate diagnosis should be described. clear operational definitions of disease state, exposures, health outcomes, and other measured risk factors for outcome; An operational definition is one that can be implemented independently using the data available in the proposed study. For example “pneumocystis carinii pneumonia, episode” is not an operational definition; a better description would be “hospitalization with a principal discharge diagnosis of ICD-9-CM code 136.3.” If complex algorithms are used to establish exposure or health status, they should be described or referenced. projected study size, statistical precision, and the basis for their determination; Describe the relation between the specific aims of the study and the projected study size in relation to each outcome. In most circumstances, it is desirable to express study goals in terms of precision sought for study estimates rather than statistical power. For safety studies, it may be useful to specify the sample size that can minimally detect a pre-specified risk with a pre-specified power, for example, “the study has an 80% power to detect a relative risk of 3 or greater for drug x compared with treatment with other drugs commonly used in this condition.” For descriptive studies, for example, studies on drug utilization, where the result is a measure of frequency rather than of association, the protocol might specify the expected precision of these frequency estimates, if possible. methods used in assembling the study data; This should include a description of, or reference to, any pre-testing procedures for research instruments and any manuals and formal training to be provided to interviewers, abstractors, coders, or data entry personnel. This should also include procedures for linkage and data mining of administrative databases. If data are validated, methods should mentioned, for example, review of ICD codes. procedures for data management; Describe data management and statistical software programs and hardware to be used in the study. Describe data preparation as well as the methods for data retrieval and collection. Data management also includes methods used to correct inconsistencies or errors, impute values, or modify raw data. methods for data analysis; Data analysis comprises comparisons and methods for analyzing and presenting results, categorizations, and procedures to control sources of bias and their influence on results, for example, possible impact of biases due to selection bias, misclassification, confounding, and missing data. For instance, the statistical procedures to be applied to the data to obtain point estimates and confidence intervals of measures of occurrence or association should be presented. Any sensitivity analyses should be described.Details of the statistical analysis may be specified later, but before analysis begins, as part of a protocol amendment to the study protocol, or more typically as a separate document, usually referred to as a Statistical Analysis Plan. See Analysis under Study Conduct section. a description of quality assurance and quality control procedures for all phases of the study; Mechanisms to ensure data quality and integrity should be described or referenced if the mechanisms are published elsewhere (e.g., when publications describing procedures for existing data collection systems are published or in the public domain), including abstraction of original documents, extent of source data verification, validation of endpoints, and revision and oversight of programming. As appropriate, include certification and/or qualifications of any supporting laboratory or research groups. limitations or strengths of the study design, data sources, and analytic methods; At a minimum, issues relating to confounding, misclassification, selection bias, generalizability, and random error should be considered. The likely success of efforts taken to reduce errors should be discussed. a description of plans for protecting human subjects; This section should include information about whether study subjects will be placed at risk as a result of the study, provisions for maintaining confidentiality of information on study subjects, and potential circumstances and safeguards under which identifiable personal information may be provided to entities outside the study. Conditions under which a clinical trial would be terminated for ethical reasons (stopping rules) should be described. Procedures for monitoring results should be described, and the use of a Data Safety Monitoring Board for clinical trials should be considered for this purpose. The need for submitting the protocol to an Institutional Review Board (IRB)/Independent Ethics Committee and the requirement of informed consent should be considered in accordance with local law. Research using de-identified data from commercially or publically available secondary data sources represents the lowest risk to potential subjects because it involves the collection of anonymous or publically available data. These studies therefore may not require IRB review in all jurisdictions if the study meets other legal and regulatory requirements for the protection of human subjects. If the study is categorized as exempt from IRB review, this should be stated and duly referenced in the protocol. Further, if the investigator's sponsoring institution (e.g., university) or owner of the data source requires the registration and/or ethical review of all research, including “exempt from human subjects research,” this should also be stated. See section on Protection of human subjects. a description of plans for disseminating and communicating study results, including the presence or absence of any restrictions on the extent and timing of publication; There is an ethical obligation to disseminate findings of potential scientific or public health importance (e.g., results pertaining to the safety of a marketed medication). Authorship should follow guidelines established by the International Committee of Medical Journal Editors (http://www.icmje.org/). See also Communication section. dated amendments to the protocol; Significant deviations from the protocol, such as any changes in the population or sample that were implemented after the beginning of the study, along with the rationale, should be documented in writing. Any changes made after data analysis has begun should be documented as such and the rationale provided. archiving or registration of protocol; ISPE endorses the opportunity to register and publicly disclose hypothesis-driven pharmacoepidemiology research protocols in a suitable public site, such as the ENCePP registry (also currently functioning as the EU PAS Register) or ClinicalTrials.gov. If it is chosen, registration and disclosure should be permanent, without the option to retract the protocol. If possible, the researcher may choose to register a protocol with a published timetable for its later disclosure, rather than immediately disclose the protocol at the time of registration. This delay may be scientifically appropriate to avoid the potential for recall or other bias resulting from study participants' knowledge of the study hypothesis, such as in a case control study collecting data on past exposures. The procedures for protocol registration should also include a mechanism for amendment of previously posted protocols. In the registration statement, the investigators should declare the extent to which they were aware, through advance exploratory analyses, of the likely ultimate findings of the study at the time that the protocol is submitted. exceptions from overall protocol requirement; Active surveillance of medical products seeks to proactively identify suspected or unanticipated medical product-associated adverse events. The primary goal of active surveillance is to identify potential safety signals quickly. While these active surveillance activities generally employ sound design and analytic approaches, they are often not fully specified, protocol-driven analyses. As part of surveillance, identified safety alerts undergo further investigations using methods that are tailored to specific product–outcome pairs. If a protocol-driven assessment is deemed necessary to confirm or refute the safety signal, the analysis should follow the GPP to the extent possible. The organization(s) and individual(s) conducting and sponsoring the research shall be fully responsible for the research. The relationship, roles, and responsibilities of the organizations and/or individuals conducting and sponsoring the study should be described. The individuals responsible for pharmacoepidemiologic research, along with the type of expertise and autonomy in conducting the research, should be stated clearly. For projects sponsored by one organization (such as a pharmaceutical company or government agency) but implemented by another (e.g., an academic institution or a contract research organization CRO), responsibility for scientific integrity is shared by the collaborating institutions (e.g., sponsor, the principal investigator conducting the study, the senior qualified epidemiology staff within the CRO, and the organization that employs the principal investigator). In such situations of shared responsibility, contractual arrangements should include a timeline for study completion and potential actions to be taken if the timeline cannot be met. In particular, the contract should delineate the roles and responsibilities to be assumed by the study sponsor and the contractor(s) in communicating various aspects of the study as well as data access, ownership, and archiving. Personnel engaged in epidemiologic research and related activities should have the education, training, or experience necessary to perform the assigned functions competently. The organization should maintain a current summary of training and experience of these personnel. A list of individuals engaged in or supervising activities should be maintained and updated periodically with current job titles. Adequate physical facilities shall be provided to all those engaged in epidemiologic research and related activities. Suitable storage facilities shall be available to maintain technical records in a secure and confidential environment in compliance with local regulations. For the purposes of ensuring and documenting the contractor's conformance with the GPP, it is recommended that the study sponsor have the right during the course of the study, and for a reasonable period following completion of the study, to inspect the contractor's facilities, including equipment, technical record, programming, and records relating to the work conducted under the sponsor's contract. The nature of the audit, including procedures that ensure patient confidentiality, should be agreed upon at the outset of any contract. The principal investigator shall be responsible for the overall content of the research project, including the day-to-day conduct of the study, interpretation of the study data, and preparation and publication of the final report. These responsibilities extend to all aspects of the study, including periodic reporting of study progress as well as quality assurance. The unusual decision to terminate a study prematurely should be taken with great caution and should be based on good scientific and ethical reasons and documented in writing. There may be rare instances in which administrative reasons require study termination. Investigators and sponsors should specify and agree in advance about the circumstances under which the study could be terminated early. Included should be a mechanism for resolution of any disagreement. Approval by an IRB, Independent Ethics Committee, or other appropriate body, should be obtained for all research involving human subjects. Studies using commercially or publically available de-identified secondary data sources, or which meet certain other criteria, are not considered research involving human subjects in some countries and may be exempt from IRB review. Informed consent beyond that already obtained for participants in a research database may be needed when the research imposes a risk for patients or data containing personal identifiers is required by the investigator (e.g., medical record). The legal definition of a personal identifier varies across countries; therefore, national and local laws should be consulted when proposing to obtain this type of data. In some circumstances, and in some countries, disclosure of relevant personal medical information (e.g., medical record) without consent is permissible under public health laws, for example, when government organizations conduct infectious disease surveillance or monitoring and reporting adverse drug reactions through secondary data sources (see section on Protocol Development, N on Active Surveillance). Investigators shall ensure that personal identifiers will be removed from any study files that are accessible to non-study personnel in accordance with applicable laws and regulations. Whenever feasible, study files should be coded and stripped of personal identifiers, and code keys should be stored separate from study files. All personnel with access to data containing personal identifiers should sign a pledge to maintain the confidentiality of study subjects. For additional information, please consult the ISPE guidelines on Data Privacy, Medical Record Confidentiality, and Research in the Interest of Public Health (http://www.pharmacoepi.org/resources/privacy.cfm). Blood and serum sample collections stored after completion of clinical studies are a valuable resource. However, protecting confidentiality in such data requires special consideration, and investigators are encouraged to consult guidelines developed by the NHLBI.1 All data collected for the study should be recorded accurately, promptly, and legibly. The individual(s) responsible for the integrity of the data, computerized or hard copy, shall be identified and shall have the education, training, and experience needed to perform the assigned tasks. All procedures used to obtain, verify, and promote the quality and integrity of the data should be recorded in sufficient detail so that others can replicate them. A historical file of these procedures shall be maintained, including all revisions and the dates of such revisions. Any changes in data entries shall be documented. For any endpoint or covariate status ascertainment (in a cohort study or trial) or exposure ascertainment (in a case–control study) that requires adjudication, all measures taken to assure blinding of the adjudicators to the exposure (cohort) or outcome (case–control) status of the subject should be outlined in the protocol. Documentation of these measures needs to be maintained with the data files throughout the course of the study and archiving period. Security of the data should be maintained at all times. Access should be limited to authorized individuals. Control systems, such as document encryption, should be used to ensure the authenticity, integrity, and confidentiality of electronic records when transmitted over open networks (e.g., the Internet). Adequate back up of the data should be maintained throughout the course of the study. The analysis should be directed toward the unbiased estimation of the epidemiologic parameters of interest (e.g., risk or rate differences and risk or rate ratios). The precision of effect estimates should be quantified using confidence intervals. Comparability of populations for pooled estimates should be assured, and missing of important variables should be addressed. Interpretation of statistical measures, including confidence intervals, should be tempered with appropriate judgment and acknowledgements of potential sources of error and limitations of the analysis and should never be taken as the sole or basis for that is or is not a relation between an exposure and outcome. analyses should be conducted to the effect of critical of the Describe the need and of an or analysis when If the of such be pre-specified in the study protocol. studies shall be in a final that the study objectives, results, strengths and limitations of the study, and interpretation of the organization and its if is shall procedures under which of the conduct, results, and interpretation of an epidemiologic study will including what individuals with the research These individuals should include the principal study and/or the This may be documented in the of a company in the study protocol, or through contractual ISPE communicating estimates of epidemiologic measures, generally by using point estimates and confidence intervals, or is useful in reporting results of safety studies to include the relative and risk Investigators should not about based on the outcome of a of (e.g., a or a statement about the confidence including or not including the There is an ethical obligation to disseminate findings of potential scientific or public health For findings that could have a impact on public may be as well as ethical to the results immediately to the appropriate regulatory shall be informed of study results in a by publication in the scientific literature and at scientific or at should not be considered as a for publication in the Authorship of study should follow the guidelines established by the International Committee of Medical Journal Editors (http://www.icmje.org/). All should meet the criteria for and all meet the criteria should be of interest, and should be to to these guidelines should be described in the protocol. research sponsors shall be informed of study results in a that with local regulatory of research including and of should be whether results are or in writing. Pharmacoepidemiologic studies are usually to the relation between certain exposures and health based on comparisons of the frequency in with and without the exposure of interest statistical analyses analytic The primary goal of these analytic studies is to whether a drug exposure the risk of an adverse or protection Therefore, assessment and reporting of if scientific to this For primary data collection studies, reporting of adverse is required by in most information on suspected adverse may be identified during the course of a study, but not as a formal part of the study objectives, procedures for and reporting of safety information in the study should be defined by the sponsor and research at the time of protocol For specific relevant regulatory should be consulted analysis of electronic healthcare may also identify an in risk with a Such studies may be as study but do not require reporting of adverse or adverse data source studies that include medical record or text review by a human for example, studies in which endpoint validation is using medical record review, are an These studies are generally required to adverse drug reactions documented in the medical text If reporting requirements the protocol should specify what data are being collected and how they will be of whether reporting requirements sponsors should findings of their analytic studies in as study be maintained for the storage and retrieval of all study related An shall be to identify the to identify their and to identify by name and any that by their nature are not in the study Access to the shall be and limited to authorized personnel procedures may be necessary to ensure that access to confidential information is limited and that the confidentiality of information about study subjects is pertaining to the study, informed consent of all relevant of IRB and other and of all quality assurance and Communication of study results to the sponsor, and scientific should be for example, questionnaires, and of relevant information and other important guidelines for pharmacoepidemiology research. This list is not to be but an of important guidelines available at the time of the GPP