Overall Survival in Patients with Rheumatoid Arthritis and Solid Malignancies Receiving Biologic Disease-Modifying Anti-Rheumatic Therapy

Abstract

Introduction/Objectives

The effects of biologic disease-modifying anti-rheumatic drugs (bDMARDs) in patients with rheumatoid arthritis (RA) and cancer is largely unknown. We examined overall survival (OS) in patients with RA and solid malignancies receiving bDMARDs.

Methods

We performed a retrospective cohort study of patients with RA and solid malignancies seen at MD Anderson Cancer Center between 2002 and 2014. Cox proportional hazard regression models, stratified by tumor type and stage, were fit evaluating use of bDMARDs as a time fixed, and time varying covariate.

Results

We identified 431 RA patients with solid malignancies: 111 (26%) received bDMARDs after their cancer diagnosis. Median OS from cancer diagnosis was 16.1 years. Of the patients receiving bDMARDs, most had localized disease, and only 14 (13%) had advanced cancer. In the stratified Cox models no statistically significant differences were observed between patients who received tumor necrosis factor inhibitors (TNFi) or patients who received non-TNFi, compared to those who did not receive bDMARDs (Hazard ratio (HR), 0.67; 95% confidence interval (CI), 0.31, 1.44; HR, 1.10; 95% CI, 0.26, 4.60 respectively). In breast cancer patients, those receiving TNFi or non-TNFi had a numerically higher but statistically non-significant HR compared to those who did not receive bDMARD: HR, 1.40 (95% CI, 0.42, 4.73), and HR, 1.37 (95% CI, 0.22, 8.42) respectively.

Conclusion

No significant differences in OS were observed between patients who received bDMARDs and those who did not. Additional data is needed to evaluate other cancer outcomes such as recurrence and progression, and patients with advanced cancer.

Background

Biologic disease-modifying anti-rheumatic drugs (bDMARDs) including tumor necrosis factor inhibitors (TNFi) and non-tumor necrosis factor inhibitors (non-TNFi) substantially improve the outcomes of RA. However, there are concerns that bDMARDs may interfere with the immune response to cancer and increase risk of recurrence or new cancers. For example, tumor necrosis factor has been shown to have pleiotropic effects on various stages of tumor genesis and may alter the natural history of cancer;^1,2 B-cell depletion with the use of rituximab may impair immunosurveillance as seen in the setting of lymphoma where treatment with rituximab may be associated with increased risk of secondary malignancies.³

Randomized clinical trials of bDMARDs for RA treatment have systematically excluded patients with cancer, thus resulting in a paucity of data on the safety of these agents in patients with cancer. Several observational studies have reported no increased risk of de novo malignancies with the use of biologic agents, other than possibly for lymphoma, for which patients with RA are at increased risk independently of therapy^4–9 Few studies have examined the use of bDMARDs in patients with RA and concomitant or history of cancer.^10–12 Although no significant harms in cancer outcomes were reported in these studies, the majority of patients were cancer survivors with no evidence of disease for many years. Given these uncertainties, and theoretical concerns of potential tumor immunity suppression, there has been reluctance to use bDMARDs in patients with RA and cancer.

In 2012 the American College of Rheumatology (ACR) guidelines,¹³ and the Canadian guidelines,¹⁴ cautioned against the use of bDMARDs in patients with recent history of cancer i.e. within 5 to 10 years of cancer treatment, but this was based on low quality of evidence. The updated ACR guidelines in 2015 recommended treating patients with RA and previously treated solid malignancies as patients without such a history.¹⁵ A recent systematic review evaluating consensus recommendations regarding the management of RA in patients with cancer revealed that the various recommendation did not agree on the safety of using bDMARDs.¹⁶ Nevertheless, with the dissemination of these guidelines the utilization of bDMARDs in patients with cancer appears to have increased. In a recent study using data from an observational registry, CORRONA, approximately one third of the patients were receiving bDMARDs or conventional synthetic DMARDs immediately following cancer diagnosis.¹⁷ Furthermore approximately 15% of patients initiated bDMARDs within the first 3 years following cancer diagnosis. We evaluated the utilization of bDMARDs in RA patients seen at MD Anderson Cancer Center with a solid malignancy diagnosis between 2002 and 2014.¹⁸ We found that a quarter of RA patients received bDMARDs following cancer diagnosis, of which about half were receiving bDMARDs prior to cancer diagnosis and continued therapy thereafter.

As the evidence on the safety of bDMARDs in patients with a history of cancer arises from small observational studies, it precludes robust conclusions. In order to add to the existing evidence, we conducted a study at MD Anderson Cancer Center, one of the largest cancer centers in the world, to examine the time dependent association of bDMARD therapy on overall survival (OS) in patients with RA and concomitant solid malignancies.

Patients and Methods

Study design.

We conducted a retrospective cohort study of patients with prevalent RA and solid malignancies initially seen at MD Anderson Cancer Center. Cohort identification with details of inclusion and exclusion criteria for the cohort can be found elsewhere.¹⁸ Briefly, patient’s ≥ 18 years of age, with a histologically confirmed diagnosis of solid malignancy identified from our institutional tumor registry, excluding non-melanoma skin cancer, seen at MD Anderson Cancer Center between 2002 and 2014, with potential RA were identified using International Classification of Diseases version 9 (ICD-9) code 714.0 (confirmed by electronic medical record review), and with at least 3 months of medical care received at our institution from the time of registration in the tumor registry were included. Potential RA cases identified by using ICD 9 code were confirmed by electronic medical record review by two independent reviewers. All identified RA also had to have documented co-management by a rheumatologist in their electronic medical record. All patients were followed up until 2016 (Supplemental figure 1). Institutional review board approval was obtained prior to data collection. The use of patient information complied with the Health Insurance Portability and Accountability Act.

Data collection.

Data was obtained from the institutional tumor registry and electronic medical records, details of which have been previously published.¹⁸ Information abstracted included date of registration, date of cancer diagnosis, age at cancer diagnosis, gender, race, type of malignancy, and stage of malignancy. We recorded the use of DMARDs, including conventional synthetic, biologic and targeted synthetic DMARDs. None of the patients in the identified cohort received targeted synthetic DMARDs. The use of bDMARDs was divided into 3 categories for data analysis: 1) no bDMARDs, 2) tumor necrosis factor inhibitors (TNFi) (etanercept, adalimumab, infliximab, certolizumab, and golimumab), and 3) non-tumor necrosis factor inhibitors (non-TNFi) (abatacept, rituximab, and tocilizumab).

Data analysis.

The primary outcome overall survival was defined as the time from cancer diagnosis to death or last follow-up. To model the effect of bDMARDs use on overall survival (OS) since cancer diagnosis, the bDMARDs status was defined as a time-varying factor which was divided into 3 categories: 1) no bDMARDs, 2) TNFi agents, and 3) non-TNFi agents. A stratified Cox proportional hazard regression model, with bDMARDs as a time-varying covariate and stratified by tumor type and stage was fitted on OS. To reduce immortal time bias resulting from cancer diagnosis to the time of first bDMARD treatment, landmark analysis was performed with 1-, 2-, and 3-year after cancer diagnosis as the landmark time points. At each selected landmark time point, a Cox model was built to compare patients that received bDMARDs to patients that did not receive bDMARDs at the respective landmark time point (1-, 2-, 3-year). Patients that received a bDMARD after the respective landmark time point were included in the no bDMARD group until that landmark point. In a sensitivity analysis, we built Cox models at 1-, 2-, and 3-year landmark points but we only included as controls patients who had not received bDMARDs throughout the observation period (as opposed to including them in the landmark period before they received treatment as for the primary analysis). Backward elimination method with staying criteria of p = 0.05 was used to build the final multivariate model. A two-sided p-value of less than 0.05 indicates significance. All statistical analyses were performed in SAS 9.4.

Results

A total of 431 RA patients with solid malignancies were included in our analysis (Table 1). Of these 320 had no documentation of receiving bDMARDs, 60 were receiving bDMARDs at the time of cancer diagnosis and continued after cancer diagnosis, and 51 started bDMARDs after cancer diagnosis. None of the patients that were receiving bDMARDs prior to cancer diagnosis stopped bDMARDs after cancer diagnosis. None of the patients received targeted synthetic DMARDs. Details on the utilization of RA therapies in this cohort have been previously published.¹⁸

Table 1.

Baseline characteristics of RA patients with cancer stratified by the time to initiation of bDMARD therapy^*

	Total (n=431) N (% by column)	bDMARDs at the time of cancer diagnosis and continued thereafter (n=60) n (% by row)	Started bDMARDs after cancer diagnosis (n=51) n (% by row)	No bDMARDs (n=320) n (% by row)	P value
Cancer diagnosis age (years), mean ± SD	59.9 ± 11.0	57.4 ± 10.3	55.3 ± 10.7	61.1 ± 10.9	<0.01
Gender, n (%)					0.30
Female	316 (73)	43 (14)	42 (13)	231 (73)
Male	115 (27)	17 (15)	9 (8)	89 (77)
Race					0.86
White	324 (75)	47 (14)	37 (11)	240 (75)
Black	47 (11)	6 (13)	5 (11)	36 (76)
Hispanic	41 (10)	6 (15)	5 (12)	30 (73)
Other	19 (4)	1 (5)	4 (21)	14 (74)
Stage					0.28
Post Treated – no evidence of disease	110 (25)	16 (14.5)	16 (14.5)	78 (70)
In situ	21 (5)	5 (24)	3 (14)	13 (62)
Localized	154 (36)	22 (14)	11 (7)	121 (79)
Regional	83 (19)	12 (14.5)	12 (14.5)	59 (71)
Distant	63 (15)	5 (8)	9 (14)	49 (78)
Tumor type					0.02
Bladder transitional cell carcinoma	8 (2)	1 (12.5)	1 (12.5)	6 (75)
Breast	175 (41)	23 (13)	21 (12)	131 (75)
Cervix adenocarcinoma	2 (0.5)	0	1 (50)	1 (50)
Endometrial carcinoma	17 (4)	5 (29)	1 (6)	11 (65)
Gastrointestinal neuroendocrine tumor	2 (0.5)	1 (50)	0	1 (50)
Head of pancreas adenocarcinoma	8 (2)	1 (12.5)	1 (12.5)	6 (75)
Kidney	9 (2)	2 (22)	1 (11)	6 (67)
Lung	64 (15)	5 (8)	5 (8)	54 (84)
Melanoma	22 (5)	6 (27)	2 (9)	14 (64)
Oral cavity squamous cell carcinoma	16 (4)	1 (6)	2 (13)	13 (81)
Ovary mixed cell adenocarcinoma	7 (1.5)	0	2 (29)	5 (71)
Prostate adenocarcinoma	67 (15)	7 (10)	8 (12)	52 (78)
Rectum adenocarcinoma	7 (1.5)	1 (14)	1 (14)	5 (72)
Thyroid papillary adenocarcinoma	22 (5)	6 (27)	3 (13)	13 (60)
Vagina/vulva squamous cell carcinoma	5 (1)	1 (20)	2 (40)	2 (40)

Abbreviations: bDMARD biologic disease-modifying anti-rheumatic drugs, RA rheumatoid arthritis, SD standard deviation

^*bDMARDs evaluated include tumor necrosis factor inhibitors (etanercept, adalimumab, infliximab, certolizumab, and golimumab), and non- tumor necrosis factor inhibitors (abatacept, rituximab, and tocilizumab)

In the overall cohort, median OS from cancer diagnosis was 16.1 (95% CI: 10.6, not reached) years. A stratified (tumor type and stage) multivariable Cox model after adjusting for age at cancer diagnosis showed that differences between patients who received TNFi and those who did not receive bDMARDs were not statistically significant (Hazard ratio (HR), 0.67, 95% confidence interval (CI), 0.31, 1.44); no significant differences were observed between patients who received non-TNFi and those who did not receive bDMARDs (HR, 1.10; 95% CI, 0.26, 4.60) compared to patients that did not receive any bDMARDs (p=0.58 for the overall difference between the 3 groups) (Table 2). The Kaplan-Meier curves of OS are shown in Figure 1.

Table 2.

Univariate and multivariate Cox proportional hazard regression models for overall survival stratified by tumor type and stage in patients with RA and solid malignances

Characteristic	HR	(95% CI)	Overall effect p-value	HR	(95% CI)	Overall effect p-value
	Univariate model			Multivariate model**
Age at cancer diagnosis	1.05	1.03–1.08	<0.01	1.05	1.03–1.07	<0.01
Gender			0.04
Female	Ref
Male	1.87	1.03–3.38
Race			0.78
White	Ref
Black	1.03	0.54–1.96
Hispanic	0.50	0.13–1.92
Other	0.93	0.44–1.96
bDMARD use *			0.49			0.58
No bDMARDs	Ref			Ref
TNFi	0.65	0.30–1.37		0.67	0.31–1.44
Non-TNFi	1.25	0.30–5.14		1.10	0.26–4.60

^*time varying covariate

^**Cox proportional hazard model stratified by tumor type and stage

Abbreviations: HR, Hazard ratio; CI, confidence interval; RA, rheumatoid arthritis; bDMARD biologic disease-modifying anti-rheumatic drugs

Figure 1.

Kaplan Meier curves stratified by bDMARD use in RA patients with solid malignancies

Abbreviations: RA, rheumatoid arthritis; bDMARD biologic disease-modifying anti-rheumatic drugs

To evaluate the effects of cancer stage on overall survival we built a multivariable model in a subset of patients with only breast cancer as it comprised the largest sub-cohort with 175 patients, 41% of the overall cohort. In a Cox model stratified by stage and after adjusting for age at cancer diagnosis, no significant differences were observed between patients who received TNFi, non-TNFi, and those who did not receive bDMARDs. However, patients receiving TNFi or non-TNFi had a numerically higher hazard of death compared to those who did not receive bDMARD: HR, 1.40 (95% CI, 0.42, 4.73), and HR, 1.37 (95% CI, 0.22, 8.42) respectively. These differences were not statistically significant, with p=0.83 for the overall effect (Table 3).

Table 3.

Univariate and multivariate Cox proportional hazard regression models for overall survival in patients with RA and breast cancer

Characteristic	HR	(95% CI)	Overall effect p-value	HR	(95% CI)	Overall effect p-value
	Univariate model			Multivariate model
Age at cancer diagnosis	1.05	1.01–1.09	<.01	1.05	1.01–1.09	<.01
Race
White	Ref		0.80
Black	1.00	0.38–2.62
Hispanic	0.60	0.11–3.36
Other	0.56	0.14–2.21
bDMARD use *
No bDMARDs	Ref		0.88	Ref		0.83
TNFi	1.20	0.36–3.98		1.40	0.42–4.73
Non-TNFi	1.49	0.25–9.00		1.37	0.22–8.42

^*time varying covariate

Abbreviations: HR, Hazard ratio; CI, confidence interval; RA, rheumatoid arthritis; bDMARD biologic disease-modifying anti-rheumatic drugs; TNFi, tumor necrosis factor inhibitors; non-TNFi, non-tumor necrosis factor inhibitors

Table 4 shows the results of the Cox proportional multivariable models at the 1-, 2- and 3-year landmark time points as described in the methods section. At the 1-year time point the hazard of death was lower in patients received bDMARDs compared to patients who did not receive any bDMARDs and at the 2- and 3-year time points the hazard of death was higher in patients who received bDMARDs compared to patients that did not receive any bDMARDs. In a sensitivity analysis, where patients who did not receive bDMARDs until the respective landmark time points were excluded we found similar results (Table 5). However, the results in the main model and sensitivity analysis were not statistically significant.

Table 4.

Vital status and multivariate models at 1-, 2-, and 3-year landmark time points including patients who did not receive bDMARDs until the respective landmark time points

Vital status at landmark time points by bDMARD use	1 year landmark		2 year landmark		3 year landmark
	Alive/censored	Death	Alive/censored	Death	Alive/censored	Death
No bDMARD	217	89	199	64	176	47
TNFi	49	10	43	9	35	6
Non-TNFi	3	0	3	0	3	0
Multivariate models
	HR	95% CI	HR	95% CI	HR	95% CI
Age at cancer diagnosis	1.04	1.01–1.07	1.05	1.01–1.08	1.08	1.03–1.13
bDMARD use
No bDMARD	Ref		Ref		Ref
TNFi or Non-TNFi	0.61	0.23–1.67	1.14	0.38–3.39	1.34	0.38–4.66

Abbreviations: HR, Hazard ratio; CI, confidence interval; bDMARD biologic disease-modifying anti-rheumatic drugs; TNFi, tumor necrosis factor inhibitors; non-TNFi, non-tumor necrosis factor inhibitors

Table 5.

Sensitivity analysis showing vital status and multivariate models at 1-, 2-, and 3-year landmark time points excluding patients who did not receive bDMARDs until the respective landmark time points

Vital status at landmark time points by bDMARD use	1 year landmark		2 year landmark		3 year landmark
	Alive/censored	Death	Alive/censored	Death	Alive/censored	Death
No bDMARD	185	80	169	56	146	40
TNFi	49	10	43	9	35	6
Non-TNFi	3	0	3	0	3	0
Multivariate models
	HR	95% CI	HR	95% CI	HR	95% CI
Age at cancer diagnosis	1.03	1.00–1.06	1.05	1.01–1.09	1.10	1.04–1.16
bDMARD use
No bDMARD	Ref		Ref		Ref
TNFi or Non-TNFi	0.57	0.21–1.54	1.02	0.35–3.02	1.24	0.35–4.33

Abbreviations: HR, Hazard ratio; CI, confidence interval; bDMARD biologic disease-modifying anti-rheumatic drugs; TNFi, tumor necrosis factor inhibitors; non-TNFi, non-tumor necrosis factor inhibitors

Discussion

Since the introduction of bDMARDs there have been concerns of their possible effects on tumorigenesis. Both TNFi and non-TNFi agents can alter the immunosurveillance of cancer and perhaps increase the development of malignancies. Despite these concerns in the general RA population, therapy with bDMARDs has not been shown to affect de novo cancer risk.^19,20 However, there is limited data on their use in patients with a history of cancer, or with concomitant active cancer. We conducted a real-world observational study at one of the largest cancer centers in the world to further address this issue. Our results showed that the use of bDMARDs in patients with RA and cancer was not significantly associated with worse overall survival. However, a clinically meaningful effect on mortality cannot be completely ruled out since patients receiving non-TNFi agents had a 10% greater hazard compared to patients who did not receive any bDMARDs. Interestingly, use of TNFi was associated with better survival, however the results were not statistically significant. In the subgroup of patients with breast cancer the hazard for death was increased by 40% and 37% for those who received TNFi or non-TNFi respectively compared to patients who did not receive bDMARDs, but the differences were not statistically significant. We also conducted landmark analysis designed to account for immortal time bias at 1-, 2-, and 3-year landmark time points, which did not show statistically significant differences in the hazard of patients who received bDMARDs versus patients who did not receive any bDMARDs. However, the landmark analysis at 2- and 3- year time points after cancer diagnosis had numerically greater hazards in patients who received a bDMARD compared to patients who did not receive a bDMARD. Sensitivity analysis showed similar numerically greater hazards and also statistically non-significant. Although no statistically significant results were observed, several of the HR estimates showed increased risk, with some including upper limits of the 95% CI greater than 2.0, suggesting that clinically meaningful effects on survival cannot be completely ruled out.

Xie et al conducted a systematic review and meta-analysis of biologic therapies on risk of new or recurrent cancer in patients with RA and prior cancer.²¹ It included 12 studies representing 13,598 patients accounting for 32,473 patient years of follow-up. They concluded that patients receiving biologics did not show an increased risk of recurrent or new cancer when compared to conventional synthetic DMARDs. Similar results were found when stratified by tumor type, time to initiation of bDMARDs after primary cancer diagnosis or exposure duration to bDMARDs. A recent nationwide population-based study conducted in Denmark showed similar results reporting that use of TNFi was not associated with an increased risk of cancer recurrence or new cancer development in patients with immune-mediated diseases (RA, inflammatory bowel disease, or psoriasis) and cancer.¹²

Few studies have evaluated the effects of biologics on survival while controlling for cancer stage, an intrinsic factor that is related to cancer mortality. A population-based cohort study from Sweden evaluating use of TNFi in patients with RA and cancer showed an overall all-cause mortality similar to biologic naïve patients.¹⁰ However only 3–8% of the population evaluated had stage 3–4 disease so the study conclusions may not be directly applicable to patients with advanced cancer. A Danish population based cohort study showed an increased hazard of death in patients that received bDMARDs before primary cancer diagnosis HR 1.53 (95% CI 1.13, 2.09), however this effect was not seen when adjusted for extent of primary cancer (HR 1.23 95% CI 0.94, 1.60), and it is unclear how many patients had regional spread or advanced disease..¹¹ Nevertheless, the adjusted models did show numerically higher hazards albeit not statistically significant, in all models irrespective of when the bDMARDs were initiated (before, after or before and after primary cancer). These results are somewhat similar to ours where numerically higher hazards were identified, and a clinically relevant risk cannot definitively be ruled out. In our study 15% of the patients had advanced disease precluding us from making firm conclusions for such patients.

Our study has several strengths. We built stratified Cox proportional multivariable models at different landmark time points to reduce the immortal time bias. Our study is also one of only a few that have evaluated survival in patients with RA and cancer receiving bDMARDs, while most others have focused on cancer recurrence and development of new cancers. Evaluating effects on survival is especially important now since many cancers have vastly improved survival rates. Patients in our database had long follow-up with a median of 6.23 years thus providing enough time for assessment of robust survival.

Our study has a few limitations. We evaluated the effects of biologics compared to biologics naïve patients, but were not able to evaluate the effects of TNFi use and non-TNFi use separately as time varying covariates in the landmark analysis due to the small sample sizes in the groups. It is possible that TNFi and non-TNFi agents may have varying effects on survival; however, we did not find statistically significant differences in the stratified Cox model of the entire cohort. We built stratified Cox proportional models stratified by tumor type and stage, however there may be residual confounding which is inherent in observational studies. Also, because of the small sample size we were unable to differentiate between the various non-TNFi received by the patients. The majority of patients in our cohort had localized cancer or no evidence of disease at the time of treatment, and our numbers are too small to infer outcomes in patients with more advanced disease. Finally, confounding by indication is a limitation encountered in observational studies. In our study it is possible that patients that had a better cancer prognosis received TNFi agents and those with a poorer prognosis received non-TNFi or no bDMARDs. We controlled for the strongest predictors of survival, tumor type and stage in our models. However, confounding by indication cannot be completely ruled out.

It is expected that approximately 40% of men and women will eventually develop cancer,²² many of whom will have concomitant RA. Therefore, it is essential to provide guidance on how to best treat these patients, optimizing their survival but also improving their quality of life. Our study provides some evidence that many patients with RA and solid malignances may receive bDMARDs without fear of detrimental effects on OS. However, most of the patients receiving these agents had localized cancer or no evidence of disease, which only allows us to generalize our findings primarily to this population. Despite these results, there was a non-significant increase in hazard of death in patients with breast cancer that needs to be further investigated in larger populations. Moreover, important knowledge gaps remain, and additional larger studies are also needed to study the potential effects of bDMARDs on cancer outcomes such as recurrence or progression, especially in patients with active or advanced cancer.

Key points.

• We found no statistically significant differences in OS between patients with RA and concomitant solid malignancies who received bDMARDs and those who did not.

• Most patients who received bDMARDs had been diagnosed with early stage cancer

• As few patients with advanced cancer received bDMARDs safety for this group cannot be established

• No significant differences were observed between TNFi and non-TNFi but the sample size was small