Welcome to our new website! If this is the first time you are logging in on the new site, you will need to reset your password. Please contact us at raps@raps.org if you need assistance.
Your membership opens the door to free learning resources on demand. Check out the Member Knowledge Center for free webcasts, publications and online courses.
Our new book is a comprehensive look at a vital part of medicines development and regulatory affairs. Grab your copy today!
Hear from leaders around the globe as they share insights about their experiences and lessons learned throughout their certification journey.
RAPS' Latest | 06 December 2012 | By Tjeerd-Pieter van Staa
Pharmacovigilance has been defined by the World Health Organization as, "The science and activities relating to the detection, assessment, understanding and prevention of adverse effects or any other drug-related problem."1 Pharmacovigilance (or risk management) should help to inform both healthcare professionals and patients, allowing them to make educated decisions when choosing a treatment.2
The main methods for pharmacovigilance are randomized trials, observational studies and spontaneous reporting of individual case reports. Randomized trials often are considered the highest evidence for pharmacovigilance decisions, as these studies allocate patients randomly to different interventions and thus are less susceptible to bias. In contrast, observational studies compare groups based on the intervention as decided by the healthcare professional. These studies are considered to be much more susceptible to bias. Spontaneous reporting is based on case reports as submitted by healthcare professionals or patients who consider the event to be caused by the drug. This chapter assesses whether our current methods for pharmacovigilance are adequate and what could be done to improve these methods.
More than 30 years ago, Hershel Jick of the Boston University School of Medicine and colleagues provided an overview of the different research strategies in risk management.3 It was argued that individual causality assessment is particularly useful for events with a low background rate that is increased substantially due to drug exposure. The severe malformations of phocomelia in the offspring of women exposed to thalidomide are examples of events that can be identified through individual causality assessment and spontaneous reporting. Anaphylactic shock following penicillin exposure is another example of an iatrogenic event that does not require the study of larger population or trial and application of statistical techniques. On the other hand, individual causality assessment may be less appropriate for an event with a higher background rate. As an example, it may be impossible to use an individual causality assessment for attributing lung cancer to drug exposure in patients who smoke. It also is not uncommon for investigators in blinded placebo-controlled trials to attribute events to a study drug while patients were actually treated with placebo. Despite these important observations, the practice of pharmacovigilance is currently segmented, with little overlap between the different approaches.
Randomized trials are viewed by some as the highest level of evidence ("gold standard" studies): any signal of a drug safety problem can only be addressed by a "properly conducted" randomized trial. The failures of epidemiological research in detecting and addressing bias are well publicized. For example, many epidemiological studies reported beneficial effects of hormone replacement on cardiovascular disease that were not supported by the findings of a large randomized trial. On the other hand, epidemiologists do like to emphasize repeatedly that randomized trials are often conducted with idealized or unrepresentative patient groups.4 The follow-up of randomized trials is often short and they may have limited statistical power to detect effects on major clinical outcomes. Also, randomized trials commonly fail to inform decisions in everyday clinical care; they address the abstract question of an intervention's efficacy under ideal conditions, rather than its effectiveness when used in usual clinical practice.5
The challenges with our current pharmacovigilance practice are highlighted by the recent example of selective cyclooxygenase-2 (COX-2) inhibitors. Selective COX-2 inhibitors were developed to minimize the upper gastrointestinal (GI) side effects of conventional nonsteroidal anti-inflammatory drugs. A possible signal of cardiovascular toxicity of selective COX-2 inhibitors had been observed in the first large trial with rofecoxib conducted to obtain marketing registration.6 However, this finding was attributed to a possible protective effect of the comparator drug naproxen. After rofecoxib was used by millions of people worldwide, the signal of cardiovascular toxicity was finally confirmed in two randomized trials that had been initiated to expand the indication of use of rofecoxib to the prevention of colorectal adenomas.7,8 Although the registration trials for rofecoxib included large numbers of patients, the external validity of these trials was limited. An analysis of users of selective COX-2 inhibitors in actual practice found substantive differences between the trials and actual clinical practice in the characteristics of patients and exposure. The trials only included patients with osteoarthritis or rheumatoid arthritis requiring high daily doses for prolonged periods of time. In contrast, the vast majority of users in actual clinical practice used selective COX-2 inhibitors intermittently for brief periods of time at low daily doses (on average about two- to three-fold lower), and patients were younger and used selective COX-2 inhibitors for conditions other than osteoarthritis or rheumatoid arthritis.9 Most of these users in actual clinical practice may not have been eligible for inclusion in the registration trials. This suggests the need for studies that are conducted with patients who represent the full spectrum of the population to which the treatment might be applied and with interventions that have real-life (rather than ideal) compliance.
Epidemiological studies typically include the full spectrum of patients using a drug in actual clinical practice and could thus complement randomized trials. However, epidemiological studies often suffer from selection bias and confounding, as patients treated with a drug may have different underlying risks compared to non-users. Statistical techniques are used to minimize the risk of confounding. One commonly used technique is regression analysis that uses information on risk factors for the outcome, making users and non-users statistically comparable with respect to the measured risk factors. Another technique that is gaining popularity is the use of propensity scores, in which risk factors for exposure are measured and used to make groups comparable. But despite these, confounding and selection bias often cause intractable problems in epidemiological studies. If the extent of information on risk factors (for outcome or exposure) is limited, confounding may not be resolved by these statistical techniques. Epidemiological studies often apply these techniques without testing whether the statistical approach indeed minimized confounding. In a recent study of the cardiovascular risks of different diabetes medications, it was found that statistical adjustment only marginally reduced confounding due to the underlying disease. In this study, statistical adjustment only marginally reduced the difference in the rate of cardiovascular outcomes between patients with and without diabetes.10
One of the potential solutions to this vexing problem of randomized trials having great internal but limited external validity and of epidemiological studies having limited internal but great external validity is to introduce randomization into the settings that collect data from actual clinical practice. If clinicians use electronic healthcare records (EHRs) rather than paper records, the follow-up of clinical outcomes could be done unobtrusively.
In the UK, almost all inhabitants are registered with a general practice that provides their first-line healthcare. General Practitioners (GPs) play a key role in the UK healthcare system, as they are responsible for primary healthcare and specialist referrals. If a patient is treated in secondary care or by a specialist, the GP is informed about major clinical outcomes, and long-term treatments are frequently handed back to the GP. Almost all GPs in the UK use computers for maintaining health records, and communication between different health providers is increasingly sent electronically. The Clinical Practice Research Datalink (CPRD), previously known as the General Practice Research Database, collates the anonymized electronic healthcare information from more than 600 practices in the UK. The CPRD now can be linked individually and anonymously to other electronic healthcare datasets, including the national registry of hospital admissions in England (Hospital Episode Statistics), the national death certificates (with primary and secondary cause of death) and prospective disease registries, such as the cancer and cardiovascular disease registries.11 This allows major clinical outcomes to be measured long-term and without intrusion through various independent data sources. This system now has been implemented within the CPRD, but it could be applied more broadly both within the UK and possibly in other countries with well-established electronic healthcare research databases.
Risk management methods could be enhanced considerably by integrating clinical trials into electronic healthcare databases. Patients could be recruited at the point of care, randomized among routinely available interventions and then followed unobtrusively using the electronic healthcare database.12 As an example, patients could have been randomized between selective COX-2 inhibitors and traditional NSAIDs shortly after the launch of selective COX-2 inhibitors. Patients would have been followed for major clinical outcomes using the electronic healthcare database or linked datasets. In the CPRD, for example, heart attacks could then have been measured long-term using several methods: GP records, linked hospital data, linked disease registry data and/or linked death certificate data. The randomization would ensure that baseline differences and confounding between comparison groups were reduced.
Figure 15-1 shows the design of these randomized evaluations within electronic healthcare databases. The electronic healthcare database is searched to compile a list of potentially eligible patients, which is sent securely to the clinician's desktop computer.13 When a patient on this list attends the practice for events related to the trial, a flag appears on screen, notifying the clinician that the patient could be recruited. The clinician is given a link to the study website. If patient and clinician agree to participate, the patient is randomized and a prescription issued for the randomly allocated treatment. Anonymized electronic healthcare records from the clinician's office are copied to the central research database, creating a separate database for analysis. The trial database can be compared periodically to the full research database for fraud detection and generalizability of the randomized population. If necessary, the GP can be contacted to validate the data. [media:1478:embed{left}]Table 15-1 outlines the research questions, interventions and measurements in two feasibility trials of randomized trials that have been initiated in the CPRD.
Cluster trials are another opportunity for improving risk management. In cluster randomized trials, entire areas or health service organizational units are allocated to intervention or control groups, with outcomes evaluated for individuals within each cluster. Cluster randomized trials are increasingly used in public health and health services research and are especially important in the evaluation of health service and public health interventions.14 Cluster trials facilitate pragmatic evaluation of the effectiveness of interventions delivered in routine practice settings. As an example, general practices could be randomized between different interventions and the effects of these interventions can be measured by comparing the rates of outcomes between the practices. A cluster trial is currently operating in the CPRD evaluating antibiotic prescribing for respiratory tract infections. Electronic prompts have been developed based on recommended clinical practice guidelines to be activated during consultations for respiratory tract infections in the selected age range. The electronic prompts promote no antibiotic prescribing or delayed antibiotic prescribing instead of immediate prescription. The prompts specifically incorporate recommendations from the recent guidelines on antibiotic prescribing in respiratory illness.15 During consultations with patients presenting with symptoms of respiratory tract infection, primary care professionals will see the prompts, which remind them of recommended standards of care in respiratory tract infections. The prompts also will provide supporting information and links to evidence that supports the recommendations, in a format suitable for printing out for patients when appropriate. The decision on whether to follow the treatment suggestions included in the prompt will remain with the GP. The GP also will be able to terminate display of the prompt at any time. There will be no intervention at control practices. The intervention phase will continue for 12 months at each practice.16
David Torgerson, director of the York Trials Unit at the University of York, et al recently suggested another approach to risk management by randomizing practices to immediate and delayed access to new medicines. When a new drug is licensed, it would be withheld on a temporary basis from one of the groups. Depending upon the nature of the treatment, this might be a few months or one or two years. Each group would be allowed some new drugs early while each group would be required to act as a control for others. The data from electronic healthcare databases could then be used to compare the rates of major clinical outcomes between practices randomized to the new and standard medicine. Evidence would be generated in a relatively short period of time on new medicines' safety and effectiveness.17
A key aspect of risk management is to identify the predictors of side effects. Pharmacogenetics could play an important role in identifying patients at high risk of experiencing certain side effects.18 Electronic healthcare databases could be used to collect pharmacogenetic data on patients who suffered a side effect during treatment.19 These databases often include large populations, and data on exposure may be readily available through prescription records. In the CPRD, a pharmacogenetic study is well under way collecting blood or saliva in a cohort of statin-receiving patients with and without increased creatine phosphokinase levels. Practices were randomized between collecting blood or saliva, which will provide information on the best method for collecting pharmacogenetic material in the CPRD setting and the recruitment rates.
This study is now being expanded to allow patient recruitment in "real-time" by providing a flag to the GP during consultation. Patients who previously were prescribed the drug of interest (e.g., statins) will be identified in the CPRD. If the side effect of interest (e.g., Stevens Johnson Syndrome) is being entered into the database by the GP, a pop-up box then will appear providing a link to the study website (for registration and access to consent forms). After the patient's consent, the patient then will be asked to provide a blood sample, which will be sent to the laboratory. This system will be real-time, and can allow prospective monitoring of both new and older medicines.
A recent perspective on the future of pharmacovigilance highlighted the need to collect better information on whether the clinical symptoms of a side effect disappear, how long it takes and what treatment was needed.20 The patients who suffer the side effect may be able to provide this information, and pharmacovigilance also can benefit from concentrating on patients as a source of information.21 The system that collects pharmacogenetic samples in patients who suffered a side effect also could be used to collect information directly from patients. The recruitment letter could include login details for the study website, asking patients to complete a questionnaire on the symptoms of the side effect, quality of life and the resolution of the side effect and its treatment. This approach would integrate the clinical and exposure details as collected in the electronic healthcare database with the perspectives and outcomes as reported by patients.
All current methods of risk management have major limitations such as the selectivity and short follow-up of many randomized trials, intractable confounding and bias in epidemiological studies and data incompleteness and difficulty in assigning causality in spontaneous reporting. Electronic healthcare databases can be used to randomly allocate patients or practices to different interventions, with the outcomes of interest measured through routinely collected electronic healthcare data. In addition, the data could be enriched by collecting additional information, such as pharmacogenetic data. These data could help to individualize treatment and better predict the likelihood of adverse outcomes.
Electronic healthcare databases may provide a real opportunity to enhance current methods for risk management. But, not all research questions can be evaluated using them, and not all databases collect high quality data for all outcomes of interest. Furthermore, healthcare professionals may differ in how they record data in the database and in the quality of data recording. Also, researchers may differ in how they extract data from a database. A study by Professor Martin C. Gulliford of King's College London and colleagues found that clinical experts differed substantially in their selection of codes. Perhaps unsurprisingly, there now have been several instances in which different investigators reported discrepant results and reached opposite conclusions when testing similar hypotheses within the same electronic healthcare database.22 Thus, any study that uses electronic healthcare data will need careful evaluation of how to measure the outcomes of interest and whether high quality data is being collected.
In conclusion, there is a real opportunity to enhance risk management by extending the use of electronic healthcare databases. Randomized trials can be conducted and the routinely collected data can be enriched by other data such as pharmacogenetics and patient-reported outcomes.
Tags: pharmaceutical, drug, risk management, pharmacovigilance