Article Text
PDF
Abstract
Background We implemented a novel method for providing contextual adverse event rates for a randomised controlled trial (RCT) programme through coordinated analyses of five RA registries, focusing here on cardiovascular disease (CVD) and mortality.
Methods Each participating registry (Consortium of Rheumatology Researchers of North America (CORRONA) (USA), Swedish Rheumatology Quality of Care Register (SRR) (Sweden), Norfolk Arthritis Register (NOAR) (UK), CORRONA International (East Europe, Latin America, India) and Institute of Rheumatology, Rheumatoid Arthritis (IORRA) (Japan)) defined a main cohort from January 2000 onwards. To address comparability and potential bias, we harmonised event definitions and defined several subcohorts for sensitivity analyses based on disease activity, treatment, calendar time, duration of follow-up and RCT exclusions. Rates were standardised for age, sex and, in one sensitivity analysis, also HAQ.
Results The combined registry cohorts included 57 251 patients with RA (234 089 person-years)—24.5% men, mean (SD) baseline age 58.2 (13.8) and RA duration 8.2 (11.7) years. Standardised registry mortality rates (per 100 person-years) varied from 0.42 (CORRONA) to 0.80 (NOAR), with 0.60 for RCT patients. Myocardial infarction and major adverse cardiovascular events (MACE) rates ranged from 0.09 and 0.31 (IORRA) to 0.39 and 0.77 (SRR), with RCT rates intermediate (0.18 and 0.42), respectively. Additional subcohort analyses showed small and mostly consistent changes across registries, retaining reasonable consistency in rates across the Western registries. Additional standardisation for HAQ returned higher mortality and MACE registry rates.
Conclusions This coordinated approach to contextualising RA RCT safety data demonstrated reasonable differences and consistency in rates for mortality and CVD across registries, and comparable RCT rates, and may serve as a model method to supplement clinical trial analyses for drug development programmes.
- Rheumatoid Arthritis
- Cardiovascular Disease
- Outcomes research
- Epidemiology
Statistics from Altmetric.com
Introduction
Rheumatoid arthritis (RA) is associated with increased mortality, with longitudinal studies averaging a standardised mortality ratio of 1.5 (95% CI 1.2 to 1.8) for patients with RA compared with the general population.1 In addition, patients with RA have an increased risk of cardiovascular disease (CVD); this is not entirely attributable to traditional CV risk factors and often summarised in an approximate 1.5 multiplication factor, a level similar to that of patients with diabetes.2 ,3 While CVD and mortality are associated with inflammation and disease activity of RA4 ,5 and mitigating this with effective treatment may reduce this risk,6 it is not well understood how new treatments may affect these risks, especially as initial safety results come from randomised clinical trials (RCTs).
Patients in contemporary RA trials generally do not receive placebo for >6 months and are often followed in long-term extensions, continuing on or switched to active treatment. The limited placebo data result in added uncertainty for assessing more long-term outcomes such as mortality and CVD in patients receiving active drug. Although studies from RA registries have investigated the incidence of mortality and CV adverse outcomes,5 ,7–12 such published data have limitations in their ability to provide context for rates of adverse events observed on active treatment in an RCT programme. These include differences in patient populations, calendar dates and duration of follow-up, outcome definitions and geographical matching. To address this, we sought to better contextualise mortality and incident CV event rates from a typical modern clinical trial programme by a coordinated analysis of five international RA registries for comparison rates.
Methods
To address our study aims, we (1) identified relevant existing registries with individual-level patient data on mortality and CVD and established a new registry to enhance geographical coverage, (2) harmonised the CVD outcome definitions across the cohorts, (3) identified the most important and consistent predictors for mortality and CVD incidence across these cohorts, (4) assembled a matrix of mortality and CVD incidence rates stratified by these predictors, (5) computed aggregate-level incidences standardised to those emanating from a clinical trial programme and (6) assessed their robustness across different definitions of the cohorts and follow-up times.
Patient data sources
The clinical trial cohort included data from all fostamatinib-exposed patients in the phase II–III RA programme (core and extension studies), with a data cut-off of 19 March 2013 and will here be referred to as the Fostamatinib Clinical Trial Programme (FCTP). Five RA registries (Consortium of Rheumatology Researchers of North America (CORRONA),13 Swedish Rheumatology Quality of Care Register (SRR),14 Norfolk Arthritis Register (NOAR),15 CORRONA International and Institute of Rheumatology, Rheumatoid Arthritis (IORRA)16) were selected based on considerations that focused on optimising comparability with FCTP data. See Nyberg et al17 for in-depth methodology, registry selection and registry geographical locations. For additional details, see online supplementary table S1 and a prior study examining baseline data for all five registries.18
Outcome definitions
Mortality outcomes were all-cause mortality and CV-related mortality. For CV events, we examined six outcomes: (1) major adverse cardiovascular events (MACE), that is, non-fatal myocardial infarction (MI), non-fatal stroke and CV death; (2) acute MI; (3) stroke; (4) hospitalised heart failure; (5) hospitalised deep venous thrombosis (DVT) and (6) hospitalised pulmonary embolism. Prior to analysis, event definitions were agreed upon for best validity and comparability across registries and FCTP, depending on the type and level of data available in each registry. Adjudication and validation of outcomes were included in the FCTP and where feasible in the registries. Online supplementary table S2 describes the data sources and definitions for each outcome. For the example of MI, SRR used linkage to hospital and mortality data for ICD-10 codes (I21 or I22), NOAR used a similar combination of hospitalisation and mortality data supplemented with patient-reported data, the CORRONA registries used physician-reported information with adjudication and IORRA had patient-reported data complemented by some hospital data for validation. In each analysis, the first event of each outcome for a patient during follow-up was included.
Statistical analysis
Throughout these analyses, all individual patient data remained with the registries. Study group members and staff at each registry remained blinded to FCTP data that were not already in the public domain. For each outcome, a series of Cox regression analyses were performed within each registry to gauge the strength of association between a series of predefined covariates and each outcome (ie, to assess their potential confounding properties). Covariates assessed included age, sex, HAQ, body mass index (BMI), RA treatment history, indices of RA disease activity at baseline and trial exclusion criteria. These analyses informed the selection of covariates used to standardise incidence rates from the five registries.17
Within each registry, we assembled a ‘main cohort’ of all patients with RA ≥18 years of age, observed from 1 January 2000 or the earliest time point thereafter until end of follow-up or 31 July 2013 at the latest. We also defined a series of nested subcohorts (figure 1) defined by disease activity and disease-modifying antirheumatic drugs/biologic treatment change. To evaluate the robustness of the main analysis, we performed several predefined sensitivity analyses (SA). These were variations on the main analysis based on the definitions of the nested subcohorts defined based on trial inclusion criteria and treatment start (SA1 to SA3), with selected study exclusion criteria also applied to mirror relevant exclusion criteria in the trials (SA4 and SA5), with restricted calendar years or follow-up (SA6, SA7 and SA8), and with additional standardisation applied for baseline HAQ (SA9) (see online supplementary table S3). Some subcohorts became substantially smaller than the main cohort, and for those subcohort definitions that resulted in a new start of follow-up different from the main cohort, baseline characteristics changed accordingly (see online supplementary table S4).
Definitions and relationship between the main cohort (here denoted Cohort A) and its nested subcohorts. Note that Cohorts B and C1 are intermediate steps and were not included in the final subcohort sensitivity analyses. Note also that the C1 and C2, as well as the C1a and C1b, subcohorts are partly overlapping, as some individuals may be included in one subcohort at one point in time and then at another (later) point in the other (given the longitudinal character of the data). These cohorts correspond to the following analyses in this paper: main analysis—cohort A; sensitivity analysis 1 (SA1)—subcohort C2; SA2—cohort C1a; SA3—cohort C1b; SA4—cohort C1a with exclusions; SA5—cohort C1b with exclusions. RA, rheumatoid arthritis.
For the FCTP, event rates presented are the crude incidence rates, with exact CIs based on the Poisson distribution. The registry rates were standardised to the age and sex distribution (in SA9 also the HAQ distribution) of the FCTP, using categories as in table 1, with CIs based on a gamma distribution.19 ,20 Statistical analyses were performed using SAS V.9.2; exact Poisson CIs were calculated using StatXactprocs V.9.0. The estimated incidence rates were based on individual records with non-missing observations on all standardisation variables involved. As some level of true variation by registry and geography was expected, heterogeneity across registries and SA including rate standardisation was assessed by informal comparison of point estimates and CIs rather than through for example, statistical tests for heterogeneity.
Baseline characteristics of the five RA registry main cohorts at start of follow-up (1 January 2000, or cohort entry date if later) and the Fostamatinib Clinical Trial Programme full cohort (at treatment initiation)
Results
To illustrate cohort differences, baseline characteristics of the main cohort for each registry and the FCTP are shown in table 1. The combined registry main cohorts included 57 251 patients with RA (234 089 person-years (py))—24.5% were men, mean (SD) baseline age and RA duration were 58.2 (13.8) and 8.2 (11.7) years, and mean baseline HAQ and DAS28 were 0.88 (0.67) and 4.1 (1.5), respectively. Current smoking was most common in NOAR and IORRA (both 18%), and SRR had the highest proportion with early RA (70%). The FCTP had 3240 ever fostamatinib-treated patients with RA followed for 4486 py—with mean (SD) baseline age 53 (12) years, 18% were men, 49% had RA duration <5 years, 78% had HAQ ≥1.1 and 75% had DAS28 ≥5.1. Baseline comorbidities varied, with hypertension prevalence ranging from 40% in CORRONA International and FCTP to 8% in IORRA and 10% in SRR, coronary artery disease prevalence from 8% in SRR to 1% in FCTP and IORRA, and heart failure prevalence from 3% in SRR to 0.3% in IORRA.
Mortality
The crude all-cause mortality rates in the registry main cohorts ranged 10-fold from 0.20 to 2.14 per 100 py. After age and sex standardisation, the rates ranged from 0.19 to 0.80, and the mortality rate in the FCTP full cohort was 0.60 per 100 py (table 2). Similar results were seen for CV mortality, standardised rates 0.07–0.23 in the registries versus 0.22 in FCTP. For FCTP, SRR and NOAR, cause of death information was essentially 100% complete; however, cause-specific attribution during the study period was missing for 51% of reported deaths in CORRONA, 17% in CORRONA International and 9% in IORRA.
Summary of mortality and cardiovascular adverse event outcome incidence rates standardised by age and sex in five RA registry main cohorts, and corresponding incidence rates for the Fostamatinib Clinical Trial Programme full cohort
There was moderate variability in standardised all-cause mortality rates across SA for all registries, with the highest rates (0.70 to 1.25 per 100 py) seen after standardisation for sex, age and HAQ (SA9, figure 2 and table 3). These results suggest that differences in HAQ between FCTP and registries may be confounding the incidence rates to some extent. Restricting to patients with increased RA activity (SA1) showed higher rates in the registries, while applying FCTP exclusion criteria (SA4 and SA5) resulted in lower rates. The lowest incidence rates were observed in SA7 (biologic-naïve switchers with 18 months follow-up), ranging from 0.29 to 0.55 per 100 py, while the incidence rate in the corresponding FCTP cohort was 0.59.
Incidence of all-cause mortality in five RA registry main cohorts and across all sensitivity analyses, and corresponding incidence rates for the Fostamatinib Clinical Trial Programme cohorts
Incidence of mortality (A), major adverse cardiovascular events (B), acute myocardial infarction (C) and stroke (D) across the five RA registry main cohorts and the fostamatinib clinical trial programme full cohort, with crude rates, age-standardised and sex-standardised rates (main analysis), and, where available, additional HAQ-standardised rates (sensitivity analysis 9; SA9). CORRONA, Consortium of Rheumatology Researchers of North America; FCTP, Fostamatinib Clinical Trial Programme for rheumatoid arthritis; HAQ, Health Activity Questionnaire; IORRA, Institute of Rheumatology, Rheumatoid Arthritis; NOAR, Norfolk Arthritis Register; SRR, Swedish Rheumatology Quality of Care Register; Std, Standardised.
CVD rates
In the primary analysis, the age-standardised and sex-standardised incidence rates for MACE in the registry main cohorts (range: 0.31–0.77 per 100 py) were comparable with those in the FCTP full cohort (0.42 per 100 py; table 2). Similar results were seen for acute MI (0.09–0.39 vs 0.18), while for stroke the FCTP was at the low end (0.12–0.31 vs 0.13). For hospitalised pulmonary embolism, the FCTP rate was higher than the two registries with available data, although CIs were wide and overlapping. Although an incidence rate was not formally calculated due to low number of events in the FCTP, the occurrence of hospitalised DVT did not appear to be meaningfully different from the available registries.
There was some systematic variability in the standardised incidence rates for CV outcomes across the registries (table 2): for MACE, acute MI and stroke, the highest standardised rates were reported in SRR (0.77, 0.39 and 0.31 per 100 py, respectively), and the lowest were reported in IORRA (0.31, 0.09 and 0.12 per 100 py, respectively). For hospitalised heart failure, the rates ranged from 0.13 per 100 py in NOAR to 0.40 in CORRONA International.
The MACE standardised incidence rates across all registry cohorts and SA were <1 per 100 py, and for all of the SA, the MACE incidence rate for the corresponding FCTP cohort (range 0.40–0.42 per 100 py) was comparable with the middle to lower end of the registry ranges (table 4). Overall, the SA showed limited variability within registries. The highest registry rates were reported standardised for sex, age and HAQ (SA9, figure 2 and table 4), where the registry rates ranged from 0.42 in CORRONA International to 0.94 per 100 py in SRR, as compared with the FCTP full cohort rate of 0.42.
Incidence of major adverse cardiovascular events (MACE) in five RA registry main cohorts and across all sensitivity analyses, and corresponding incidence rates for the Fostamatinib Clinical Trial Programme cohorts
Discussion
Commonly, RCTs are not directly generalisable to broader patient populations. Often, <10% of community patients with RA are eligible for phase 3 trials used for US Food and Drug Administration and European Medicines Agency approval, due to strict inclusion and exclusion criteria,21 ,22 Therefore, it is often inappropriate to compare adverse event rates directly with those published from clinical and national RA registries, as RCT participants generally have more active RA, less comorbidity and are younger. The FCTP patients were younger and included more women than three of the registries, and as age and sex are important risk factors for many events, age and sex standardisation was an important tool to more appropriately compare trial and registry rates. The importance of such standardisation for the outcomes presented here is demonstrated by the stark contrasts in crude versus adjusted rates, most notably in SRR and NOAR's older cohorts (table 3). Similarly, a recent meta-analysis of longitudinal studies found mortality rates in RA to range from 1.0 to 5.2 per 100 py,1 which is higher than our adjusted rates but more similar to our crude rates, likely reflecting the higher crude rates seen in older populations.
After careful assessment, the only factors selected for standardisation were age, sex and HAQ, which were found to be strong predictors for the CV outcomes under study across registries and helped decrease variability in these rates across cohorts. For robust comparison of CV rates, we also considered standardisation for a range of other CV risk factors,23 but they were not found to be very strong and consistent predictors for the outcomes studied (data not shown), so the standardisation used is likely to capture much of the confounding present. Some of our SA were designed to accommodate bias due to confounding via restriction (eg, to patients without CV events before start of follow-up). In addition, increased variance from finer standardisation and missing or uncollected data by registry would have been a compounding problem, had we included a broader set of standardisation variables. For example, SRR did not collect smoking or BMI, while NOAR was missing BMI on 34% and smoking status on 17%. A recent study looked at MACE events in CORRONA and calculated an unadjusted incidence rate of 0.91, which is somewhat higher than our crude rate findings of 0.73 from the same registry and considerably higher than our standardised rate, demonstrating how the actual operationalisation of seemingly straightforward cohort and outcome definitions and analysis aspects may affect estimated rates.24 In comparison with published MI rates, our crude rates (0.12–0.91 per 100 py) were again consistent with a separate US RA registry that found 0.4 per 100 py,12 and the standardised rates (0.09–0.39) were overall more consistent with the findings in SRR for patients younger than 51 years,25 highlighting the influential effect of the younger age structure in the FCTP. By utilising a prospective, coordinated and prespecified approach in our study's design, we expect our event rates to have greater relevance for contextualising trial data than what can be obtained from standard literature-based epidemiological approaches and reporting. While the focus of the current report is on the methodology and harmonisation of the register data, our results also indicate that incidence rates of mortality and CV outcomes in the FCTP were comparable with standardised incidence rates in the registry cohorts, suggesting that there was no increased risk of these outcomes associated with fostamatinib exposure.
The strengths of our study include (1) collaboration across existing and new registries to allow direct access to data sources and to obtain harmonisation of outcome definitions. This enables appropriately tailored analysis, including standardisation and stratification for relevant variables, providing better support for SA and interpretation of data than more customary reliance on published literature. (2) A main analysis that uses a relatively unselected cohort over long follow-up from each registry, maximising the precision of the estimates (narrow CIs) across registries given the available data (important for infrequent events which require large sample sizes), while the supplemented SA on smaller cohorts, implying lower precision, improve accuracy by reducing potential bias in the main analysis. (3) The selection of subcohorts to more closely resemble clinical study populations for SA (increased internal validity) and to better understand the importance of these selection criteria. (4) Better temporal matching to the trials to address potential concerns around changes in risk panorama, treatment patterns and disease risk over time; this was achieved through data cut-off dates as proximal to the analysis as possible.
There are clearly general limitations when observational data are used as context for RCTs. These include differences in patient populations, intensity and length of follow-up, data availability and structure, and definitions of outcomes. Specific limitations in this study include (1) the four existing registries that contributed the majority of event data are located in the US, Sweden, UK and Japan and, may not fully represent the regional distribution in the FCTP. While CORRONA International was created to address this regional limitation, it accounted for only a tenth of the registry patients, and at the time of analysis still had a limited follow-up time. (2) The data sources, patient population coverage, outcome availability and quality of data collection varied across registries. For example, while all registries had similar prevalence of RF positivity, NOAR and CORRONA were missing this status in a substantial proportion of their patients in various subcohorts, which increases uncertainty around RF-stratified rates for those registries due to RF's connection to mortality and CVD.26 More importantly, the low CV-mortality rate in CORRONA was likely related to the large absence of cause of death information for the study period, during which linkage to the US National Death Index was not yet available. (3) Registry-based cohorts are at risk of under-reporting if patients leave the registry before events or are not adequately followed-up. In addition, CORRONA International, being a new registry, had low mortality rates with only six deaths, which is common with clinical registries in early follow-up due to potential recruitment selection bias for healthier patients, characterised by the lower percentage of smokers. (4) A clinical trial programme may be enriched for ‘healthier’ or ‘sicker’ participants, based on inclusion and exclusion criteria. While baseline hypertension prevalence was most similar in the FCTP and CORRONA (∼40%), the overall rates of other CV comorbidities were lower in FCTP participants due to CV exclusion criteria. On the other hand, trial participants undergo more thorough testing and follow-up than would be experienced by patients in registries who are generally followed as part of routine clinical practice. Therefore, earlier and subclinical disease may be more likely to be identified in clinical trial populations leading to an apparent higher risk for some outcomes. Also, patient dropout and selection over time will occur in both trials and registries, and may affect estimated rates. In addition, the lack of data on specific CV risk factors mentioned previously is a potential limitation, as availability of such data might provide better standardised rates. While the extent of either of these biases are not fully known, we tried to address them with a variety of sensitivity and exploratory analyses; this systematic approach led to optimised comparators and deeper insights regarding rate validity and variability, and lends additional weight to our overall findings.
In conclusion, we implemented a consistent methodology and analysis with standardisation of mortality and CVD rates enabling comparison across registries. We demonstrated that in real-world RA populations from different countries and with variations in RA management and co-morbidity, rates of both overall CVD and mortality were relatively consistent with other data on geographical variation in these outcomes both across and within the cohorts, most notably when the major risk factors were accounted for by standardisation for age, sex and HAQ, and SA restricting patient selection or follow-up time. Enriching data from clinical trials programmes with observational data from clinical registries is a powerful means to improve the understanding of the safety profile of new drug compounds in the preapproval phase, support continued safety assessment and serve as a starting point for proactive postmarketing risk management. The approach addresses an important need to understand rare events that are not assessable in current trials, but it must be balanced with careful consideration of the detailed methodology and safety outcomes that may not be feasible in events with more heterogeneity in definitions or the most rare events, as discussed elsewhere.17
References
Supplementary materials
Supplementary Data
This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.
- Data supplement 1 - Online supplement
Footnotes
Handling editor Tore K Kvien
During the conduct of this study, SF and LH were employees of AstraZeneca, and CG was affiliated with the Arthritis Research UK Centre for Epidemiology.
Contributors All authors fulfilled the ICMJE recommendations for authorship and have seen and approved the final version of the manuscript.
Funding AstraZeneca funded this study (largely remuneration for analyst time at the individual registries) and collaborated with the researchers in the design, analysis and interpretation. Funding from AstraZeneca did not include editorial assistance for this manuscript; medical writers were not used. NOAR is funded by Arthritis Research UK. SRR has or has had research agreements with Abbvie, Pfizer, BMS, UCB, Merck, AstraZeneca, Sobi and Roche. IORRA is supported by various grants from a large number of pharmaceutical companies, including AstraZeneca. CORRONA, Inc. has received funding in the past 2 years from AbbVie, Amgen, AstraZeneca, Janssen, Genentech, Lilly, Novartis, Pfizer, Regeneron, Vertex and UCB, through contracted subscriptions to the database, and CORRONA International LLC has received funding from AstraZeneca.
Competing interests In summary, the included registries have, or have had, agreements with MAHs/pharma for antirheumatic therapies as outlined in the ICMJE forms. AstraZeneca provided funding for this study, but it terminated the drug development programme and so has no interest in particular safety aspects. We do not consider the above as a conflict of interest that would risk influencing the results in either direction.
Ethics approval At each of the five participating institutions, there was IRB-equivalent review and approval. Informed consents were either obtained as part of inclusion into the participating registries and trial or the requirement of informed consent was waived by an IRB.
Provenance and peer review Not commissioned; externally peer reviewed.
Data sharing statement All results are contained within the manuscript. Access to raw data will follow the principles at each of the participating institutions/registries. The general rule is that raw data (here the entire registries) reside with the principal investigators/governing boards of these registries, and that access to linked data will follow the legislative framework within each country they are housed.