Following the licensing of the first tumour necrosis factor (TNF)α inhibitors for the treatment of severe rheumatoid arthritis (RA) at the beginning of the current decade, the national rheumatology societies in several European countries initiated independent registries for the long-term evaluation of the safety and effectiveness of this new generation of drugs. The short-term efficacy and safety of these agents had been shown in clinical trials but their long-term effects were unknown. There was concern that cytokine inhibitors could impair the immune defence and distort immune regulation and thus lead to an increase in adverse events such as infections, malignancies and autoimmune disorders.

Spontaneous notification of adverse events to national pharmacovigilance systems leads to significant underreporting of events and lacks any opportunity to relate the events to the number of patients treated with the substance. They also lack comparators (ie, possibilities to relate numbers of observed comorbid disease events during treatment with a specific drug to numbers of disease events in patients with the same diagnosis not treated with this drug). Open label extensions have the limitation that they only observe patients previously included in randomised clinical trials (who have been shown to differ from those treated in routine practice), and seldom offer any useful comparator. The biologicals registers were established as epidemiological cohort studies or as routine registration of patients seen in rheumatology. They recruit unselected patients treated in routine care in order to overcome the drawbacks of either spontaneous reporting systems or open label extensions of clinical trials.

Since the initiation of registers for patients treated with the first TNF blocking agents, further new agents targeting TNFα, B cells, interleukin (IL)1, or costimulation molecules have been licensed for use. There is a long list of other new drugs that are being investigated for use in RA and other inflammatory rheumatic diseases. To date, all new biological drugs licensed for RA have been included in the registers.

Given the novelty and pioneering role of the rheumatology registers for other fields of medicine, their importance for the further development of pharmacoepidemiology and the richness of critical findings arising from them, it is time to balance what has been achieved so far and to outline and decide how to face the future challenges for the registers.

This paper focuses on safety aspects, the major driver for developing the registers. Further important contributions from the registers, which we do not address here, pertain to the effectiveness under real-world conditions and the cost effectiveness of biological treatments.

BACKGROUND OF THE EUROPEAN REGISTRIES

Specific features

Biologicals registers have been established in several European countries including the UK, Sweden, Germany, Spain, Norway, Denmark, The Netherlands and Switzerland. Other countries are currently setting up registers or using available data sources. Table 1 gives an overview of the larger European registers. Even though these registers differ in some aspects such as size, methods of follow-up, control groups, they do share a number of common features, which are summarised below:

• All registers were initiated by or in collaboration with the national rheumatology societies of the respective countries.

• These societies felt that the medical profession should take the lead in studying the long-term outcomes of these novel agents and initiated independent observational studies.

• The registers are not drug specific but instead include all licensed biologicals.

• With the increasing number of licensed biological agents, a considerable proportion of patients will be treated with more than one biological agent. Drug-specific registers, maintained by the respective pharmaceutical company, are not appropriate to answer questions related to the risks conferred by exposure to multiple biological agents.

• The registers are designed as epidemiological cohort studies based on recruitment of patients either at a clinical practice or population level.

• All patients ever enrolled by the register will be followed-up, irrespective of whether or not they remain on the initial drug. This is an indispensable prerequisite for the comparability of exposed and unexposed patients and for the identification of long-term effects. Events that occur after treatment has ended can only be captured if the observation period is independent of the time on treatment.

• The registers have either established comparator cohorts or use national population registers as comparison.

• In order to put the prevalence of observed adverse events into perspective, having an adequate control group is essential. These control cohorts are either embedded within the design of the register itself or rely on a parallel system of data collection at specific rheumatology sites. Some countries use historic cohorts as comparators. In the UK, Sweden and Denmark national register data on cancer and death, and in Sweden and Denmark on hospitalisation, can be linked to the biological register data.

• The registers are supported by joint grants from all companies whose products are under observation.

• After agreements with the rheumatological societies, the pharmaceutical companies agreed to provide joint grants sufficient to run high quality prospective registers. Independence of the registers in terms of design, conduct, analysis and publication of results is guaranteed. The companies usually have the right to see and comment on publications before submission but they do not have the right to determine the content and extent of publication.

• This long-term commitment is a unique example of a fully independent academic enterprise, jointly sponsored by competing drug companies.

• Some of the registers provide the companies with regular reports on adverse events that help the companies to fulfil the safety requirements of the drug regulatory authorities.

• The British, Swedish and German registries and the (currently six) pharmaceutical companies have agreed on a standardised reporting system for serious adverse events. It includes the biannual reporting of baseline clinical and demographic characteristics of the patients as well as the incidence of specified “events of interest” such as infections, cardiac diseases, haematological disorders, disorders of the central nervous system, malignancies as well as pregnancies and deaths in all treatment groups and control cohorts. The companies use the reports to fulfil the requirements of the European drug agency (EMEA). Other registers also report serious adverse events directly to the companies. Exchange of data with the safety departments of the companies aims to ensure that double reporting of serious adverse events from spontaneous reporting and from the registers is avoided.

• Since their inception, the registers have worked together closely and met regularly.

• The registers have met approximately annually since 2002 in Manchester, UK, Stockholm and Uppsala, Sweden, or Berlin, Germany. The meetings were initiated by the British, Swedish and German Registers. All other registers have been invited to participate in these workshops.1

Advantages of academia-driven drug registers

It is the major advantage over industry-driven observational postmarketing studies that all registers follow patients irrespective of treatment continuation with a specific drug. Patients are enrolled on a certain drug initially but follow-up does not end when treatment is terminated. Instead, patients are followed-up for defined periods of time (5 or 10 years) or indefinitely. Treatment with drugs other than those originally prescribed will be recorded. Monitoring follows strict rules and quality control measures to ensure complete ascertainment of exposures and events are in place. Drop-out rates are low.

With an increasing number of agents available the majority of patients will soon have been exposed to more than one biological drug. From a methodological point of view, it becomes increasingly difficult to distinguish the effects of a specific drug from those of a class of agents or from agents with different modes of action. The analysis of multidrug exposure is highly challenging and requires interdisciplinary research teams and international collaboration. New methodologies have to be developed and the results of different approaches have to be discussed within academia.2

Since the registers are run by researchers with a broad interest in disease outcomes, pharmacovigilance (mandated by the regulatory agencies or not), is not the only objective of interest. The public interest in the registers and their responsibilities towards the rheumatological community ensure that all information on long-term disease outcomes, quality of life, or costs collected in the registries will be made available, analysed and published as independent research projects.

Collaboration between the registers

Close international collaboration is a major requirement for two reasons: firstly, numbers of cases in individual registers, even though high compared to trials, may not be sufficient to gain robust knowledge on rare events such as lymphoma or pregnancy. Therefore, in order to be able to share experiences it is, secondly, mandatory to harmonise the methodological approaches. This applies to the type of patients enrolled, the time points of data collection, the items collected, the method of follow-up, or the system for coding of adverse events. New registers in rheumatology or beyond should take advantage of the experiences gained so far.

The current registers have agreed on fundamental issues for reporting adverse events to the drug companies, such as:

• a uniform reporting system for the biannual reports;

• the definition of time windows for the attribution of serious adverse events to specific drugs;

• “events of interest” which require additional and event-specific queries (eg, infections, heart failure, lymphomas, solid tumours, demyelination);

• The Medical Dictionary for Regulatory Activities (MedDRA)3 as preferred dictionary for the coding of adverse events.

METHODOLOGICAL CHALLENGES OF MULTIDRUG REGISTERS

The most obvious limitation of the biologicals registers (compared to randomised trials) involves channelling bias or confounding by indication. Guidelines effective in all European countries limit the prescription of anti-TNFα agents to patients with severe disease or bad prognosis and the failure of previous disease-modifying antirheumatic drug (DMARD) therapies. Regardless of the control group chosen, therefore, there is a high likelihood that the group of patients treated with biological agents will still have a poorer baseline disease status. Additionally, selective treatment drop-out by patients with poor prognosis may lead to a better average prognosis for those patients remaining on therapy. This influences at first hand the results on treatment effectiveness. Therefore, confounding by indication is a greater challenge in studies of effectiveness than in those on safety. However, the incidence of adverse events might also be influenced by differences in disease activity or in comorbidities associated with the event. Factors prognostic for the disease outcome such as disease activity as well as those influencing the development of adverse events (eg, chronic obstructive pulmonary disease (COPD) for lung cancer) have to be taken into account. Therefore, methods for controlling these biases and adjusting for confounding must be applied at several stages of the research process: selection biases have an influence not only at the start of biological treatment but also at clinical decisions regarding “switching” to alternative drugs.

Control groups

Since clinical features influence treatment decisions and outcomes, patients in the control group should be as similar as possible to patients starting on a biological. This similarity should include age, gender, disease duration, clinical measures of disease activity and severity, treatment history and comorbidity. Therefore, patients with previous failures on traditional DMARDs may be superior as control patients than early DMARD-naive patients, although this is eventually depending on the outcome under study. However, since there is no guarantee that the patients are comparable and key outcomes such as the risk of lymphoma may be influenced by the heterogeneity of the groups, statistical methods to minimise confounding by indication must be in place at the analysis stage.

Process of follow-up

It is essential to follow the control group with the same rigidity and protocol as the biologicals group. Differences in ascertainment of adverse events, for example, inevitably lead to non-comparable results. Linkage to national registers, when available, can minimise differences in ascertainment between groups.

The unblinded nature of the study bears in itself the risk of differential awareness and reporting. If a doctor is aware of an increased risk of certain infections in patients treated with biologicals, he might be more careful in examining for this in biologicals patients than in controls or lower his treatment threshold. Additionally, the doctor might consider it more important to report it if a biologicals patient is involved.

Therefore, any form of spontaneous reporting necessarily leads to biased results. Even within the framework of an intensively monitored observational study, this problem cannot be completely ruled out.

Data analysis

The greatest challenge lies in the adequate statistical analysis of the observational data. Several methods controlling for confounding by indication have been applied during the past decade. Propensity score risk adjustment and propensity-based matching are the most widely used methods. The propensity score reflects the probability of receiving a specific treatment based on prognostically relevant baseline data. Patients treated with biologicals and controls are stratified according to their propensity score value. Within propensity score strata, the covariates in biologicals and control patients are equally distributed.4

Propensity-based matching means that the propensity score is used to select patients from the control group who are similar with respect to their propensity scores to the biologicals patients. The problem here is that atypical patients are chosen from the biologicals and the control group due to the unequal distributions among cases and controls. These patients may therefore no longer represent the patients in each respective group.

Analysis of rare events

Even though the registers include thousands of patients and tens of thousands of patient years of follow-up, for very rare adverse events with an incidence of 1 in 1000 patient years or less the individual national registers may be underpowered. It will thus be necessary to evaluate data across registers and one approach to achieving this, which takes full advantage of the numbers of patients collected throughout Europe and taking at the same time national differences into account, is to perform nested case-control studies in each of the registers and to undertake a meta-analysis of the results. This ensures that from each register adequate controls matched for many possible background factors can be chosen.

Pooling of raw data, however, is not recommended since this might obscure differences in outcomes resulting from different patient populations, health care systems or methods of data collection.

MAJOR LESSONS ON THE SAFETY OF NEW THERAPIES LEARNED SO FAR FROM THE REGISTERS

The major concerns at start of the registries were whether the agents would trigger the development of lymphoproliferative disorders, solid tumours, or autoimmune diseases or lead to an increase in infections. Today, we can provide some intermediate answers to the majority of these questions.

Infections

Three registers have published data describing the risk of infections in patients treated with TNFα inhibitors compared to that in control patients. The absolute rates per 100 patient years were quite similar.5^–7 The relative risks compared to the control group, however, varied from 1.0 to 2.1. This led to intensive methodological discussions among the registers and showed that there will never be one “truth” since the results depend on the case mix of the patients, the method of case ascertainment, the length of follow-up or the structure of the control group. A trend to decreasing risk of infections with ongoing therapy was found7 and explained by an increased risk in the first months of therapy.2 This example does not only show the pitfalls of observational designs but also illustrates why it is imperative to collect, analyse and compare data from different independent registers. Irrespective of these concerns, the data produced from the different registers are of value, since they indicate that there may be an increased real risk for several types of infections, not only tuberculosis,8 but also that this eventual risk is within a moderate range and thus should be possible to handle in clinical practice with appropriate guidelines on awareness of the potential risk.

Cardiovascular events

Patients with RA have a twofold greater risk of experiencing a myocardial infarction.9 It is well established that atherosclerosis is an inflammatory condition associated with elevated acute phase reactants. Reducing the inflammatory burden could therefore theoretically lead to a decrease in morbidity and mortality from atherosclerosis. Data from the registers support this view.

A decrease in mortality from cardiovascular events compared to that in a clinical cohort of conventionally-treated patients and a mortality similar to that of the general population of the same age and gender was found.10 Comparing data from a register with those from a population cohort and from a national inpatient care registry, the age-adjusted and sex-adjusted incidence was significantly lower in patients with RA treated with TNFα blockers.11 The rates of myocardial infarction in 8670 patients treated with TNFα inhibitors were equal to those of conventionally-treated patients after adjustment for baseline risk factors. However, responders to anti-TNFα treatment had a significantly lower rate of MI than non-responders.12 TNFα blocking agents neither increased the risk of new heart failure nor that of worsening of a pre-existing heart failure.13

Lymphomas and solid malignancies

After an early report of an 11.5-fold increase in the occurrence of lymphoma among patients treated with TNFα inhibitors14 which disappeared when the data were included in a larger registry,15 there have been no further reports of an increased risk of lymphoma. Currently available data do not suggest that this treatment confers an additional risk for lymphomas. However, further observation is needed. By linkage with Cancer and Census Registries, no difference in the overall risk of solid cancers between patients treated with anti-TNFα agents and other patients with RA was found.16

AUTOIMMUNE DISEASES

There are case reports of autoimmune-like syndromes such as antiphospholipid syndrome, drug-induced lupus, neuroinflammatory diseases and vasculitis in patients receiving TNF blocking agents in clinical trials. New comorbidities such as psoriasis17 or lupus are reported to the registers. As the product information leaflets do not recommend routine screening for autoantibodies, this is not undertaken in the real world setting of the registers.

PREGNANCY OUTCOMES

Results from two registers on women exposed to anti-TNFα agents before or during pregnancy conferred no hint of an increased risk of adverse birth outcomes.18 19 However, the available data on pregnancies remain too limited to allow firm conclusions to be drawn.

Deaths

Two registers have reported on mortality risk. Decreased age-adjusted and sex-adjusted cardiovascular disease and cancer mortality rates compared to the population were found, whereas the mortality rate from infections was elevated.10 A lower mortality risk for patients with RA treated with anti-TNFα agents was found after controlling for age, gender and disease severity markers. This decrease was specifically seen in women.20 Both studies support the assumption that successful suppression of inflammatory activity increases survival in patients with RA.

OPEN QUESTIONS AND FUTURE DEVELOPMENTS

Increasingly, new biological agents are licensed and new registers will be established. Regulatory agencies, as well as the pharmaceutical industry have identified the registers as very useful postmarketing drug surveillance tools. The agreement on a standardised reporting system for serious adverse events and the ongoing discussions on methodological issues among the registers have increased the quality of data reported to the European Medicines Agency (EMEA) by the companies. Regulatory authorities require inclusion of patients on new drugs into the existing registers, and have thereby transformed what started as an academic enterprise with voluntary support from different companies into official pharmacovigilance tools. This is not without complication. The requirements imposed on the companies in their postmarketing drug safety monitoring plans are of ever-increasing complexity, and only in part represent research issues. The requirements also pose new demands on statistical precision and on the structure and capture of the underlying raw data. There is thus a need for communication between the different agencies, professional organisations and agencies which today are working towards a better safety surveillance system, but which do not have a defined common forum for communication. Further, it is not yet clear what the role of the registers will be within the new plans of EMEA to collaborate with national or European centres of excellence in pharmacovigilance. These centres will not be disease specific. However, it would be wise to include the experiences gained by the rheumatological drug registers into the new structures.

Major analytic challenges to the registers will result from multiple exposures of patients to an ever-increasing number of new drugs that make the identification of individual and combined effects increasingly complex. The groups of patients with identical exposure will become small despite large numbers of patients enrolled. This underlines the necessity for international collaboration in the development of methodology and in the sharing of data and results.

CONCLUSIONS

The European biologicals registers in rheumatology represent a novel generation of pharmacoepidemiological cohort studies. They were all initiated by epidemiologists and rheumatologists, and all are conducted under the auspices of the respective national rheumatology societies. They have already tremendously extended our knowledge beyond the populations and timeframes typical of randomised clinical trials. In terms of signal detection, no risks that had not already been observed in clinical trials have been identified. In terms of signal evaluation, the intermediate results with regards to prespecified outcomes are reassuring. In addition to what has been reported here, the registers have provided important data regarding what to expect in terms of effectiveness and response when new treatments are used in patients who would never be included in clinical trials due to comorbidity, severe functional disability, or currently low disease activity.

The registers have also further developed research methodology. As seen from the infections example, conflicting results led to further analyses and insights that would probably not have been gained if the results had all been similar. Therefore, from a methodological as well as a scientific point of view it is extremely helpful that we have several well conducted registers today whose results can be compared. Careful analysis within the context of each single register and comparison or even pooling of results is superior to primary pooling of data. The heterogeneity of results and their explanation leads to insights that were otherwise not achievable.

In order to further improve the collaboration and to ensure a high quality of current and future biologicals registers, a task force has recently been established under the auspices of the European League Against Rheumatism (EULAR). This task force will establish more formal guidelines and recommendations for the conduct of biologicals registers and will also further explore the opportunities and limitations of data sharing between the registers.

Drug regulatory agencies should take advantage of these new tools. They should, however, also bear in mind that these long-term enterprises have a fixed methodology that ensures all patients are observed in an identical manner over many years. The opportunities to change methodology with the inclusion of a new drug are therefore limited.

Rheumatologists prescribing the new biological agents now have better information on the balance of their benefits and risks, including aspects of costs to society. When new treatments arise, it is necessary to evaluate them in the same manner as those currently on the market and compare the effectiveness and safety of the different agents. With longer exposure to various biological treatments and more frequent switches, however, it will be increasingly difficult to establish the influence of individual substances. We are only at the outset of these methodological challenges.