We read with interest the article by Somers et al in Arthritis Care & Research reporting on the incidence of systemic lupus erythematosus (SLE) in the UK (1). This study suggests a markedly higher incidence rate for SLE than we found using the same database (2) and we imagine your readers will be interested in understanding how these differences arose.
Various methodologic issues need to be considered when estimating the incidence of a chronic relapsing–remitting disease such as SLE using a clinical database (3). One such issue is the duration for which SLE patients can be in remission. To enable the distinction between incident (new) and prevalent (pre-existing) cases, the investigator needs to be confident that the first record encountered in the database is actually the first occurrence of the disease for that patient. To account for this, a minimum period of registration before diagnosis needs to be defined. Lewis et al (4) described how this minimum period of time will differ between acute, chronic, and neoplastic outcomes, and concluded that patients' “early portion of follow-up” should be excluded, with the duration of this followup portion varying by disease and being longest for chronic recurrent diseases. We chose to require ≥3 years' prior data without any record of SLE before we accepted the SLE patients identified as likely incident cases, rather than patients who were having an SLE flare or relapse. For the entire study population, therefore, the first 3 years of data did not contribute to the denominator in calculating the incidence rates. The choice of 3 years was based on the reported mean ± SD duration of remission for patients with SLE, which varies between studies but is generally reported to be 2.3 ± 1.1 years (5). Somers et al required ≥1 years' worth of data before first diagnosis (1). If 1 year is indeed enough, then the resulting incidence rates should be the same as a 3-year period, since the amount of person-time discarded from the study is accounted for in the calculation of the denominator. However, if 1 year is not enough, then this will have led to the inclusion of prevalent cases in the study by Somers et al, resulting in overestimated incidence rates.
From our understanding of the article, Somers and colleagues included patients based on the identification of relevant medical codes (1), which presents an additional problem. In our study we required supporting evidence for a diagnosis such as relevant drug treatment, a diagnosis made in a hospital (we assumed this would have been by a rheumatologist), and/or meeting ≥4 American College of Rheumatology (ACR) criteria (2, 6). In contrast to conducting automated searches only, we manually reviewed >4,500 potential cases initially identified. For the largest part of the database, free text comments were available with diagnoses, which informed our decision as to whether the potential case was incident or not and whether the patient was an SLE patient or not. For instance, free text recorded together with an SLE diagnosis could state “excluded.” At this stage, we excluded many patients we believed to have disseminated or cutaneous forms rather than systemic-type lupus, prevalent rather than incident, and rheumatoid arthritis rather than SLE.
We believe that as a result of our requirement for supporting evidence of a diagnostic code combined with our manual review of the full patients' records, we excluded more cases than we would have if we had carried out the research in the same way as Somers et al. We believe the study by Somers and colleagues was at a higher risk of including prevalent SLE cases and those that were not actually SLE patients. This will have led to overestimated rates.