This is the largest study to date evaluating the risk of newly diagnosed cancer following TNF inhibitor use in Korean patients with RA. The study data included all TNF inhibitor claims in the Korean NHIS-NHID since the first TNF inhibitor was introduced to the NHIS in 2002 till 2016.
The study results indicated that the risk of cancer development was significantly lower in the TNF inhibitor cohort than the nbDMARD cohort before and after matching. This finding was consistent with several previous studies using the claims data. Wu et al.  and Lan et al.  reported adjusted HRs of 0.63 (95% CI, 0.49–0.80) and 0.59 (95% CI, 0.36–0.98), respectively, in the Taiwanese population. Cho et al.  reported an odds ratio (OR) of 0.42 (95% CI, 0.25–0.73) in the Korean population. Only the study by Jung et al. reported an insignificant difference in the Korean population (incidence rates ratio, 0.913; P = 0.546) . On the other hand, meta-analyses of randomized controlled trials demonstrated an increased or insignificant risk of cancer among patients receiving TNF inhibitors compared to those taking only nbDMARDs , . Prospective cohort studies based on the German biologics register (Rheumatoide Arthritis: Beobachtung der Biologika-Therapie [RABBIT]) and Australian Rheumatology Association Database (ARAD) found no difference in the risk of cancer due to TNF inhibitors use , .
These conflicting findings could be primarily due to the characteristics of the different data sources. Studies using national claims data are less likely to experience loss to follow-up, and the consequent missed diagnosis, than clinical trials or prospective cohort studies. For example, a follow-up bias of up to 33% (95% bootstrap limits of ‒30 to +152%) has been reported by a study in which a birth cohort was compared with national administrative registries . When it comes to the RA cohort, subjects treated with nbDMARDs are more likely to be lost to follow-up than those treated with TNF inhibitors since they do not need to visit the study center for nbDMARDs administration. This might have led to increased risk estimates for the TNF inhibitor cohort in previous cohort studies.
Differences in study design may have also contributed to the conflicting results. In many studies, including the study by Jung et al.  that had found no difference in cancer risk due to TNF inhibitor use, patients in the two cohorts were followed up from different or random time points in their disease course. For example, subjects in the nbDMARD cohort were observed from the start date of nbDMARD treatment and those in the TNF inhibitor cohort from the start date of TNF inhibitor use. In this situation, the TNF inhibitor cohort was likely to have a longer disease duration and a longer duration of nbDMARD use since clinical guidelines and reimbursement policies allow the prescription of TNF inhibitors only to those patients refractory to nbDMARDs. Such differences could subsequently lead to a relative increase in disease activity and complications in the TNF inhibitor cohort as the disease progresses over time. The point is that not only the presence of RA has been associated with an increased risk of cancer, but also RA disease activity , nbDMARDs use , and complications such as lung disease . We could control this potential bias by matching the start year of nbDMARDs use and following up each subject in the nbDMARDs cohort from the start date of the TNF inhibitor use by the matched pair (Figure 2). This also enabled us to control the possible bias from changes in the clinical environment over time. Several new TNF inhibitors were developed during the study period, and clinical experience accumulated; therefore, biologic DMARD use has increased globally . Clinical guidelines were amended many times as well. Unless controlled for, these changes over time might affect patient selection and the outcomes.
The anti-inflammatory effects of TNF inhibitors have been suggested to play a role in reducing cancer risk since chronic inflammation has been implicated in the pathogenesis of cancer . TNF inhibitors were found to suppress tumor progression by disrupting TNF-α-related tumor-promoting inflammatory signaling in vitro and in vivo [40–42]. Another class of anti-inflammatory drugs, NSAIDs, has also been reported to be associated with a decreased risk of cancer, especially breast, colorectal, and genitourinary cancers [43–48]. Coincidentally, the risk of these three cancer types was shown to be significantly reduced for patients treated with TNF inhibitors in the secondary endpoint analysis of this study. It might be theoretically logical to expect that drugs exerting anti-inflammatory effects such as NSAIDs and TNF inhibitors would reduce the cancer risk by controlling chronic inflammation. However, a cautious interpretation is needed because the inference that drugs with anti-inflammatory effects would also be cancer-protective is a risky oversimplification. The exact mechanism and extent of association between cancer risk and anti-rheumatic drugs remains unclear and needs to be further researched.
Lastly, the ongoing issues surrounding cancer risk and TNF inhibitors may have shaped the behavior of physicians. For example, physicians might have avoided prescribing TNF inhibitors to patients clinically judged to be at high risk for cancer, thereby affecting the risk of cancer in TNF inhibitor users.
The multivariable analysis showed that cancer occurrence in the cohort before matching was positively associated with increasing age, male sex, presence of chronic liver disease, and high PDC by nbDMARDs, corticosteroids, and NSAIDs. Age, sex, and chronic liver disease were established as carcinogenetic risk factors . The high PDC by anti-rheumatic drugs might indicate a high RA disease activity, which is already known as an accelerating factor of cancer development . Among the anti-rheumatic drugs mentioned above, increased use of corticosteroids remained a significant risk factor for cancer in the matched cohort. This finding is consistent with the result of a previous study on the influence of corticosteroids on the risk of skin cancer in patients with RA (adjusted OR, 2.96; 95% CI, 1.67–5.22 for cumulative doses of corticosteroids greater than 10 g) . Patients with cardiovascular disease in our study were less likely to develop cancer. Similarly, Wu et al. reported a negative association between ischemic heart disease and cancer in patients with RA (adjusted HR, 0.70; 95% CI, 0.54–0.92) . Those authors mentioned an independent association between cancer and the use of medications for ischemic heart disease, including NSAIDs, making this link worth further investigation.
This study has several limitations. Like many other studies using claims data, variables such as family history, smoking and alcohol use, body mass index, and laboratory data, especially on disease activity, were unavailable. However, we adjusted for the use of anti-rheumatic drugs as surrogate markers of disease activity. Secondly, the sample size of the matched cohort was not large enough to estimate the risk of site-specific cancer. Lastly, we could not obtain data on the use of drugs not covered by the NHIS. Therefore, our results may have been confounded by uninsured use of TNF inhibitors by private procurement or clinical trial participation. Data linkage between claims and clinical data is needed to overcome these shortcomings of our study.
On the other hand, this study has several strengths compared to previous studies. We used nationwide administrative data for the longest period the NHIS-NHID could provide. Since Korea offers a universal health insurance service, we were able to enlarge the sample size, minimize selection bias, obtain 15 years of data, and thus enhance the statistical power to detect rare cancer events using this data source. It also permitted a solid study design. We could include only RA incident cases, account for the time of TNF inhibitor use, and match the start year of nbDMARD use, which was impossible in previous studies due to the small sample size and short study periods. Lastly, we confirmed the validity of this study by showing the increased risk of tuberculosis in the matched TNF inhibitor cohort. The lack of significance before matching is thought to be due to the relatively large number of patients enrolled in the nbDMARD cohort in the early study period, when the diagnostic method for tuberculosis was less established .