Introduction

Rheumatoid arthritis (RA) is the most common chronic systemic inflammatory disease. Musculoskeletal pain, swelling, and stiffness are common symptoms of RA, which can lead to joint deformity and the loss of function, thereby, it can significantly reduce patients’ quality of life and long-term prognosis of RA [1]. RA is a multifactorial disease that may be closely related to genetic, environmental factors, and autoimmune responses [2]. The disease has also many complications, among which malignant tumors are one of the most difficult to control and life-threatening complications.

The risk of cancer in patients with RA has received much attention. Many studies have investigated the relationship between RA and various cancers, and identified the relevant pathogenesis. The abnormal autoimmune mechanism may be a common cause of malignant tumors in RA [3]. In addition, related drugs for treating RA, such as biological agents and disease-modifying anti-rheumatic drug (DMARD), may increase the risk of malignant tumors [4,5,6].

In recent years, many cohort studies have shown that there was an increased risk of leukemia in RA patients. However, some other studies failed to observe the correlation between leukemia and RA. To date, there was no meta-analysis evaluating the risk of leukemia in RA. Therefore, we conducted a systematic review and meta-analysis to assess the risk of leukemia in RA.

Methods

Search strategy

We searched the literature published before December 2019 through the databases of PubMed, Embase, Cochrane Library, and SinoMed. In addition, the references cited in the above original documents if they meet the requirements were also manually searched. The search keywords included “rheumatoid arthritis,” “leukemia,” “cancer,” “tumor,” and “morbidity.” The selection criteria were as follows: (1) the study design conformed to the cohort study; (2) the article had clear data, such as SIR and confidence interval; (3) the sample size needed to be greater than 100. Exclusion criteria were as follows: (1) ambiguous data or lack of necessary data; (2) reviews, case reports, conference articles, unpublished studies, etc.

Data extraction and quality assessment

Two investigators (Xiao Luo and Mao Liu) independently extracted the original literature data. They resolved differences by reassessing, consulting relevant professionals, and discussing. The third investigator (Jie Chen) also participated in resolving differences. The extracted data included the following information: author, year of data collection, country, number of cases, length of follow-up, source of cohort, standardized incidence (SIR), and 95% CI-related data. The Newcastle-Ottawa Scale (NOS) was used to assess the quality of the study, where a score of 6 or less was classified as low quality.

Statistical analysis

In this meta-analysis, we examined the relationship between rheumatoid arthritis and leukemia based on SIR and 95% CI. The heterogeneity between the studies was evaluated using the Cochrane Q test and the Cochran I2 statistic. P < 0.1 indicated heterogeneity between the studies, and the random effects model was used [7]. Sensitivity analysis was used to exclude each of the included studies as a source of significant heterogeneity. Publication bias was assessed using Begg’s test and Egger’s test. All statistical analyses were performed using Stata 12.0.

Results

Characteristics of included studies

A total of 627 articles were searched according to the search criteria. After reading the title and reviewing the abstract, 326 articles were excluded. Through a full-text evaluation, a total of 15 studies met the inclusion criteria (Fig. 1). The main research characteristics of this meta-analysis are shown in Table 1. The number of patients in the included cohort studies from different countries between 1993 and 2016 was ranged from 862 to 84,475. The mean follow-up period of these studies was 3.95 to 17.4 years. The study of Hemminki et al. [13] and Chen et al. [10] reported the results of males and females respectively. We have calculated the combined SIR values and 95% CI. The study of Hashimoto et al. [20] and Cho et al. [21] was excluded because the data is not available.

Fig. 1
figure 1

Flow chart of the article design process

Full size image
Table 1 Characteristics of studies enrolled in the meta-analysis
Full size table

Study quality

Among the 15 included articles, 7 articles were identified as high quality (NOS score > 6). In addition, three studies were overlapping cohort studies, which were all reported by Askling et al. [16]. The patients were divided into RA, early RA, and TNF antagonist RA. The subjects of the control group were the general population. These studies were eventually included in the meta-analysis and systematic review.

Meta-analysis results

Patients with rheumatoid arthritis have an increased risk of leukemia (SIR = 1.51, 95% CI 1.34–1.70) (Fig. 2). Due to the heterogeneity, the random effects model was adopted (I2 = 55.5%, P = 0.005). In order to explore the source of heterogeneity, a subgroup analysis was conducted by study quality, geographic region, publication time, type of control, and total number of rheumatoid arthritis. Studies with less than 10,000 patients showed a significantly higher risk of leukemia in RA patients (SIR = 3.18, 95% CI: 1.66–6.09, SIR = 1.47, 95% CI: 1.32–1.63) and heterogeneity was lower (I2 = 30.2%, I2 = 48.2%). The risk of non-Chinese is higher than that of Chinese (SIR = 1.61, 95% CI: 1.36–1.91, SIR = 1.34, 95% CI: 1.09–1.63, respectively). The studies published before 2005 suggested that RA patients had a higher risk of leukemia. (SIR = 1.65, 95% CI: 1.29–2.12, I2 = 45.9%, SIR = 1.44, 95% CI: 1.26–1.55, I2 = 52.9%). There was no significant change in the incidence of subgroup analysis by population and hospital. Studies with relatively higher quality according to NOS showed a higher risk of leukemia in RA (SIR = 1.75, 95% CI: 1.35–2.27, I2 = 73.6%), but the heterogeneity was obvious. The risk of leukemia was higher in patients with rheumatoid arthritis diagnosed after 1987 (SIR = 1.53, 95% CI: 1.33–1.76, SIR = 1.41, 95% CI: 1.03–1.92). 7 articles clearly indicated that the age of the subjects was at least 16 years old, excluding juvenile rheumatoid arthritis (JRA), which had a higher risk of leukemia compared with the overall rheumatoid arthritis (SIR = 1.58, 95% CI: 1.27–1.97, SIR = 1.45, 95% CI: 1.23–1.71). Subgroup analysis showed that heterogeneity might result from the sample size.

Fig. 2
figure 2

Forest plot of studies on risk of leukemia in rheumatoid arthritis (RA). CI, confidence interval; TNF, tumor necrosis factor

Full size image

Publication bias

There was no significant publication bias based on Begg’s test and Egger’s test (Begg’s P = 0.322, Egger’s P = 0.495). Meanwhile, we conducted a sensitivity analysis for each group and excluded each of the included studies respectively to observe the stability of the meta-analysis. When two studies reported by Askling et al. [16] and Abasolo et al. [12] were removed respectively, the heterogeneity would be significantly reduced, but the combined SIR did not change much.

Discussion

Our study showed that there was a significantly increased risk of leukemia in patients with RA compared with the general population. The sample size of the study was negatively correlated with the risk of leukemia in patients with RA. It might be caused due to studies with a small sample size that would lead to a lower level of evidence and the results were highly biased.

The diagnostic criteria, treatment methods, and evaluation methods have acknowledged quite a bit of transition due to the large span of research years. Three of the articles [8, 9, 12] fulfilled the diagnostic criteria of the 1987 American College of Rheumatology classification (ACR) in our meta-analysis. Other research objects are according to the International Classification of Diseases (ICD) and did not indicate which diagnostic criteria were fulfilled. We classified the articles according to the year the subjects were diagnosed, which fluctuated between 1965 and 2010. Rheumatoid arthritis was diagnosed in 2010 fulfilled by the 1987ACR diagnostic criteria [8]. Three articles were diagnosed pre-1987 [18, 19, 22]. The results presumably inferred that newer diagnostic time is slightly higher in the risk of leukemia. Therefore, diagnostic criteria have no significant impact on our study. We need to include the 2010 American College of Rheumatology/European League Against Rheumatism (ACR/EULAR) classification criteria for RA to update our conclusions in the future.

Subgroup analysis (Table 2) indicated that the risk of developing leukemia in RA was higher in non-Chinese populations. One study on the epidemiological patterns of leukemia showed that the incidence of leukemia in North America, Oceania, and certain European countries was markedly higher than in Asia [23]. Although there is probably a role for aetiological factors, including gene-environment interactions, the reduced incidence of leukemia in low and middle human development index countries such as Asia may be due to the inadequate diagnostic capacity. Chronic lymphocytic leukemia (CLL) is the most common type of leukemia in western countries [23,24,25], while the Asian population is more susceptible to acute lymphoblastic leukemia (ALL) [24], which has a high risk in childhood and cure rate of up to 90% [26]. Middle-aged and elderly people have increased access to health care or advanced diagnostic techniques, which may lead to a higher diagnostic rate of chronic blood diseases. RA is more common in middle-aged women [1], thus, the age of disease may affect the diagnostic rate of leukemia in a patient with RA. Our subgroup analysis showed that subjects who did not include JRA had a slightly higher risk of leukemia compared with the overall rheumatoid arthritis, but significantly higher than the general population, so the inclusion of the JRA or not had little impact on the final conclusion. However, there is no relevant evidence to prove this risk factor, and further research is needed. Other subgroup analyses have moderate heterogeneity, and the analysis has limitations that may be related to different sources of data.

Table 2 Meta-analysis of leukemia incidence in RA by subgroup
Full size table

Studies have shown that the mechanism of the malignant tumorigenesis may be due to that RA leads to lymphocyte activation under chronic immune stimulation, and B cell proliferation is not controlled [27]. It may also be the cause of leukemia in RA, but there has been no research to prove this yet; the above mechanism still needs to be continuously verified. Due to the long-term severity or treatment of the disease, there could also be an increased risk of immune system deterioration [6, 28]. Some drugs inhibit the function of the immune system and cause these immune-impaired patients to become more susceptible to cancer. For example, the use of steroid hormones, cytotoxic drugs, or immunosuppressive agents may inhibit immune defense against malignant cells and lead to tumorigenesis [29]. Biologics are a new type of drug that has been widely used in rheumatic diseases in recent years. Studies have shown that patients with long-term exposure to TNF inhibitors have a higher risk of developing T cell lymphoblastic leukemia (T-LGL), which is consistent with the results of Askling et al. [21]. However, studies have shown that the use of bDMARD may not affect the risk of hematological malignancies [16]. Immunosuppressive drugs such as methotrexate may be also not involved in the pathogenesis of lymphoproliferative tumors [28]. In an animal study, it has been shown that methotrexate may be even used to treat T-type leukemia/lymphoma [30]. In recent years, high-dose methotrexate (Hd-MTX) has been used in the treatment of childhood ALL [31]. Sakura et al. [32] demonstrated that Hd-MTX can effectively and safely treat adult ALL; however, other studies have shown that methotrexate and azathioprine may increase the risk of chronic myeloid tumors [33, 34]. Some steroid hormones, such as glucocorticoids, play an important role in the treatment of leukemia due to their pro-apoptotic action, especially ALL, which is given in high doses as a standard induction therapy. Steroids are also used to treat CLL, chronic myeloid leukemia (CML), and acute myeloid leukemia (AML) [35]. Therefore, the treatment of RA includes immunosuppressive and steroids, which may cause leukemia to be hidden. But these drugs do not cure leukemia, and symptoms can still appear and be diagnosed some time later. Other immunosuppressive drugs such as phenylbutanone and d-penicillamine may be involved in the promotion of hematologic malignancies [28]. Therefore, the impact of drugs on leukemia needs to be investigated further.

In addition to drug relation, another explanation is the effect of genetic mutations on malignant tumors of immune system diseases. Molecular aberrations such as PI3K [36], AKT, mTOR, and p53 [37] were reported to be involved in the development of malignant tumors. MiRNAs may also be associated with autoimmune diseases in hematological malignancies [38]. Carcinoma-associated fibroblasts (CAFs) [39] and interleukin-6 (IL-6) also play a significant role in tumorigenesis and inflammatory diseases including RA [39, 40]. Recently, Cuenca et al. [41] studied that CD84 were associated with cancer and autoimmune disorders, including CLL.

It was found that the risk of hematological tumors in RA might be the result of interactions between genetic mutations and environmental factors [42], but there was currently no research on the mechanism of leukemia in RA.

The advantage of this meta-analysis is that our study included a large number of subjects, which significantly improve the statistical effectiveness. Our analysis included only cohort studies that might minimize selection bias or recall bias. The data, analysis, and quality assessment were independently extracted by two independent reviewers, and the third review confirmed the reliability, which led to more reliable conclusions. However, the limitations of the current meta-analysis cannot be ignored. First, the moderate heterogeneity may come from the sample and the data itself. Since the data is derived from the national database, the statistical data is prone to deviations, and no controlled trials are more convincing. Second, the confounding factors inherent in the study may affect the risk estimates. Most of the included studies are adjusted for age, gender, and follow-up years. Other important confounding factors are not standardized. Third, we exclude 2 articles that are not available for data, which may affect the outcome judgment. In that study, Sweden was the largest country included, which may have an impact on the results. It might need to be further included in a multi-country cohort study to correct the results of meta-analysis [24].

In conclusion, this meta-analysis showed that the risk of leukemia in patients with rheumatoid arthritis was increased, especially in non-Asian populations. Our study provides up-to-date information and might improve the early diagnosis strategy to optimize treatment and improve survival.