Specialist Cohort Event Monitoring Studies: A New Study Method for Risk Management in Pharmacovigilance

SCEM studies are, by definition, observational studies to evaluate use of a drug in the naturalistic setting and fulfil the criteria for PASS [1]. SCEM studies are non-interventional because they do not interfere with the prescribing decision process of the practitioner. Data are collected on patients for whom the pharmacotherapeutic treatment decision has been made that the new product is the most appropriate treatment in the secondary care setting, as in everyday ‘real-world’ clinical practice. All specialist HCPs who can prescribe the drug, and thus all patients to whom participating specialist HCPs have access, are eligible for inclusion. A summary of the key characteristics of an SCEM study is presented in Table 1.

The underlying organizational construct of SCEM is that each study is essentially an organized system that uses observational study methods to collect uniform existing data (clinical and other) to evaluate specified outcomes for a population defined by a particular disease, condition or exposure. Akin to a registry, SCEM studies serve to provide the necessary information on groups of people who possess a particular characteristic. Of major importance is that SCEM studies recognize that the study population comprises those patients who may have treatment initiated under the care of specialists and who are more complex in terms of underlying disease, co-morbidities and concomitant medications than the general disease population treated in primary care. SCEM studies also use a targeted data collection method, which allows exposure to be more effectively defined using actual treatment dates from secondary care medical records with reference to diagnosis of the clinical condition for which the new drug is indicated or another well-defined event with a clearly recognizable onset. This mitigates the risk of immortal time bias [5]. This type of bias can arise when the period between cohort entry and the date of the first exposure (e.g. to a drug), during which an event has not occurred, is either misclassified or simply excluded and not accounted for in the analysis [5]. Commonly observed in classical cohort studies from large health care databases (where a treated group is compared with some reference group) conducted in the primary care setting, immortal time can bias results in favour of the treated group because the period of immortality is misclassified with regard to treatment, and the outcome cannot occur (Fig. 2).

SCEM studies are conducted in accordance with national and international guidelines [6‐9]. Following the principles of good pharmacoepidemiology practice [10], a full protocol is written for each SCEM study in accordance with the EU PASS guidelines. Patient information security is assured through strict measures guided by Drug Safety Research Unit (DSRU) policies. A single UK-wide ethical opinion is applied for from the National Research Ethics Service. Local Research and Development (R&D) approval is also required at each participating trust. Consent is required for access to information from existing secondary care hospital charts, and also (for some SCEM studies) to enable contact with patients’ GPs to access information from existing primary care charts.

All SCEM protocols are now reviewed by the Pharmacovigilance Risk Assessment Committee (PRAC). Once the RMP is approved, SCEM studies are registered on the European Network of Centres for Pharmacoepidemiology and Pharmacovigilance (ENCEPP) e-register of studies (http://​www.​encepp.​eu). They may also be registered on clinicaltrials.gov, where appropriate.

The DSRU works collaboratively with the UK National Institute of Health Research (NIHR) Clinical Research Networks (CRNs), which are associated with undertaking research with health service providers (trusts) at a national level. SCEM studies are evaluated against specific eligibility criteria before being adopted into the NIHR portfolio [11]. These eligibility criteria include confirmation that the study is for research and a requirement for high-quality peer review of the protocol and other study documents. Accordingly, relevant subject matter CRN study officers are able to link the DSRU with specialist HCPs within hospital and/or secondary care settings. Thus, the SCEM study population is recruited through an active research network of specialist HCPs relevant to the clinical condition for which the new product may be prescribed.

Two SCEM studies—the Observational Assessment of Safety of Seroquel (OASIS) [12] and the Observational Safety Evaluation of Asenapine (OBSERVA) [13]—have used the Mental Health Research Network, whilst a third—the Rivaroxaban Observational Safety Evaluation (ROSE) [14]—has used the Non-malignant Haematology Speciality Group, the Cardiovascular Speciality Group and the Stroke Research Network. Specialist HCPs within these networks are invited to express interest prior to the start of the study after adoption into the NIHR portfolio. In particular, they are informed that they will be participating in an observational study, which will monitor the use of a new product in accordance with the requirements set out within an RMP.

Once consent has been obtained, patients are observed. The length of the patient observation period depends on the questions sought to be answered by the particular SCEM study. Information is provided by specialist HCPs, and, where necessary, further information is obtained from the patients’ GPs. The study design utilizes a multiphase approach for secondary data collection from medical records, which is intended to facilitate cooperation with data reporting, as well as spreading the workload (Fig. 3).

The first phase of an SCEM study involves collecting specialist HCP demographic and specialist expertise data. Such data is a combination of self-reported information (upon registration) and may also be supplemented from publically available information from relevant professional bodies during the course of the study. In the second phase, since the principle of event monitoring applies, and data are abstracted from existing medical records, data collection is regarded as secondary use in accordance with EU regulatory guidelines [1]. Information on baseline characteristics and exposure at the start of the observation (the index date) is obtained from the first-wave questionnaire, whilst information on outcomes and changes in exposure are collected from subsequent waves during observation. Reported data are examined for events of interest, which have been defined according to the research objective and may include known, potential or missing risks. All clinical events of medical interest (targeted/non-targeted) and serious adverse event reports [15] may undergo further evaluation and follow-up as appropriate. This process continues until the target final study milestone is reached and the final cohort of evaluable¹ patients is confirmed.

A number of SCEM studies are ongoing, and an overview is presented in Table 2 to illustrate the potential applications of SCEM in the context of PV and risk management [12‐14]. The common aim is to examine the safety and use of the study drug. They may include a comparator group of patients receiving standard care, or another counterfactual group, to be monitored concurrently. An SCEM study may also follow a single exposure cohort design, depending on the research question being addressed. These studies have been designed to address specific research questions, including characterization of real-life drug use, evaluation of particular sub-populations, adherence to prescribing recommendations or guidelines, and targeted surveillance or analysis of specific events, including those considered to require special monitoring by regulatory authorities. Each of these studies has been undertaken as part of a post-approval commitment within the RMP.

Signal detection and evaluation is one of the primary concerns of PV to try and distinguish possible ADRs from adverse events that are unrelated to use of the study drug. Several methods are undertaken using SCEM study data for signal detection, both qualitative and quantitative. The framework for how the data are analysed is based on the cohort design. After a defined point in time (starting a new treatment), patients are observed to see if they develop a particular endpoint or outcome. The longitudinal nature of the data permits examination of temporal relationships between outcomes and new exposures. Information is gathered on the numerator (the number of reports of an event) and denominators in terms of the number of patients and also the number of patient-weeks of exposure. Calculating and ranking crude incidence density (ID) rates is one of a number of standard quantitative evaluations used in event monitoring methodology for signal generation purposes as part of initial inspection of all event data for general safety surveillance. It is used as a means of identifying early potential signals as priorities for further evaluation. Medical judgment, however, is also part of this evaluation and prioritization process. Calculation of the crude incidence (risk) of incident events by week or IDs (or rates) of incident events, expressed usually in counts per 1,000 patient-weeks of treatment (or observation), provide estimates of the ‘real-world’ frequency of reported events. These are descriptive methods in which one may observe disproportionately higher counts than expected from the summary of product characteristics or clinical trial data.

Within-group comparisons of crude ID rate differences or ratios according to different time periods after the start of treatment can identify signals of events that occur significantly more frequently after the study drug is started. The null hypothesis is that the incidence rates are constant between the two time periods being compared; the alternative hypothesis is that the incidence rates are different. A 95 % confidence interval (CI) is applied to the rate difference or ratio (based on the normal approximation). Thus, if the rate of events in the first month (ID₁) is being compared with the rate in the second month and the third month combined (ID_2–3), where the confidence limits around the rate difference estimate or the rate ratio point estimate exclude the null value (0 or 1, respectively), the null hypothesis is rejected. These results can be considered to be a signal of an event associated with starting treatment with the study drug if the estimates are positive, or correspondingly a signal of a delayed-onset event if the events are negative. In comparing these two time periods, the assumption is made that, given an event, its reporting is equivalent in both periods in a fixed cohort.

As this approach may sometimes mask significant signals in specific risk groups, the subgroups defined by specific characteristics (e.g. previous history of a condition, previous/concurrent use of selected medications, off-label indication groups) will have risks and IDs calculated and compared according to strata for relevant events, where appropriate. An example of this approach is seen in the OASIS study [12]. Since it was anticipated that dosing regimens would be individualized, it was necessary to account for variation of the dose over time, since some events may be dose related. For the exposure group of interest (patients receiving Seroquel XL™) and the internal comparator group (patients receiving quetiapine immediate release [IR]), for each patient, six periods of person-time fixed at 2-week intervals were created and the modal dose was calculated according to the dose recorded for more than 50 % of each period. This approach does introduce an artificial time structure; however, group-level analysis of person-period data permits simple univariate exploration of associations between the dose, events and selected risk factors in defined periods for signal detection purposes.

For SCEM studies in particular, the relatively short timeframe can result in low counts for some events. As such, the statistical assumptions that underpin such comparisons may no longer apply. Thus, other methodological approaches need to be considered, such as the self-controlled case series approach to explore changes in the risk of oral events over time in users of asenapine in the OBSERVA study [13, 16]. Complementary quantitative analyses may also be undertaken. These include capturing information on, and ranking by frequency, the reasons for stopping the study drug and comparisons with ranked IDs. These values can be used to compare and contrast groups descriptively. Regardless of which method is used, all signals then require confirmation or refutation by further study. A qualitative assessment of cases contributing data to a signal may be undertaken to include evaluation of patient demographic characteristics, treatment details, detection and clinical features and management of those events of interest, resolution, relevant investigations prior to and during therapy, the patient’s relevant medical history and concurrent medication, and any sequelae. Data are derived from the SCEM, and follow-up questionnaires are sent to gather other relevant essential information for construction of a case-series summary descriptive table.