Background/Purpose: The BSRBR-RA was established to compare the long term safety of anti tumour necrosis factor (TNFi) drugs with non-biologic drugs (nbDMARD) in subjects with rheumatoid arthritis (RA).  Serious adverse events (SAEs) are identified by physician and patient reporting as well as linkage to the national death register. We used myocardial infarction (MI) as an example to explore whether there is differential under-reporting of SAEs to BSRBR between treatment groups by linking BSRBR-RA to the Myocardial Ischaemia National Audit Project (MINAP). MINAP aims to collect data on all hospitalisations with MI in England and Wales.

Methods: This analysis was limited to subjects living in England and Wales. BSRBR and MINAP were linked using deterministic matching with combinations of these parameters: first and last names, birthdate, postcode, unique National Health Service number and sex. Events from both datasets were matched by subject using a 30-day window. Matched and unmatched events were verified using the American Heart Association/European Society of Cardiology criteria for MI. Deaths from MI (reported as the underlying cause of death using ICD-10 via death register linkage) were also included as verified MIs. Age, sex, treatment group, MI phenotype, whether the subjects received cardiology care at the same hospital as their rheumatology care and location of deaths due to MIs were explored as possible reasons for non-overlap between datasets using descriptive statistics. The risk of MI was compared between subjects receiving nbDMARDs and ever-exposed to TNFi using a Cox regression model, adjusted for deciles of propensity scores (PD) (Table) using i) MIs verified from BSRBR only and ii) all MIs verified from either BSRBR or MINAP. Subjects were censored at first MI, death, last physician follow-up or 04/20/2010, whichever came first.

Results: In total, 310 verified MIs were recorded during the observation period, of which 75% were captured in BSRBR-RA, 64% captured in MINAP and 39% were captured in both datasets (Table). A fifth of the BSRBR-RA-only MIs was deaths due to MI occurring outside hospital and therefore could not have been captured by MINAP. There were no differences in the age, sex, treatment group, site of MI care or MI phenotype between matched and unmatched MIs. When additional MIs from MINAP were included in the analysis, the adjusted risk estimate for MI in subjects receiving TNFi compared to nbDMARD did not differ to estimates obtained using MIs from BSRBR-RA only: BSRBR-only MIs; hazard ratio (HR) 0.56 (95%CI 0.35, 0.91); including MINAP MIs: HR 0.59 (95%CI 0.40, 0.88).

Conclusion: A degree of under-reporting of MIs exists in BSRBR but is non-differential between nbDMARD and TNFi subjects. The additional 78 (25%) MIs from MINAP did not alter the risk estimate but increased its precision. Our findings suggest linkage with other datasets is an important method of increasing event capture and enriching data for analysis.

Table