Autoimmune conditions and primary central nervous system lymphoma risk among older adults

Summary:

Primary central nervous system lymphoma (PCNSL) risk is highly increased in immunosuppressed individuals, such as those with human immunodeficiency virus infection and solid organ transplant recipients, but rates are increasing among immunocompetent older adults (age ≥65 years). We utilized data from a large, nationally-representative cohort of older adults in the United States and found that PCNSL is significantly associated with systemic lupus erythematosus, polyarteritis nodusa, autoimmune hepatitis, myasthenia gravis and uveitis. Immunosuppressive drugs given to treat these conditions may increase PCNSL risk, but these associations cannot explain the observed temporal increase in PCNSL rates, given the low prevalence of these conditions.

Primary central nervous system lymphoma (PCNSL) is a rare, but highly aggressive, extranodal non-Hodgkin lymphoma (NHL) that is confined to the brain, spinal cord and leptomeninges (Grommes and DeAngelis 2017). PCNSL rates are highly increased among immunosuppressed individuals, e.g. human immunodeficiency virus (HIV)-infected people (Gibson, et al 2014), and solid organ transplant recipients (Mahale, et al 2018). Epstein–Barr virus (EBV) infection is involved in the pathogenesis of PCNSLs among immunosuppressed individuals, while PCNSL cases linked to EBV are uncommon among immunocompetent people (MacMahon, et al 1991).

A recent analysis of data from the cancer registries participating in the United States (US) Surveillance, Epidemiology, and End Results (SEER) program showed that although incidence rate of PCNSL was low in the general population during 1992–2011 (0.6 cases per 100,000 person-years), rates increased significantly for both men (1.7% per year) and women (1.6% per year) aged ≥65 years, even after excluding HIV-infected individuals and transplant recipients (Shiels, et al 2016). The reason for this trend is unclear. Given the strong association of PCNSL with immunosuppression, it has been hypothesized that increasing use of immunosuppressive medications to treat autoimmune conditions in older adults may be contributing to this trend (Shiels, et al 2016, Villano, et al 2011). Indeed, cases of PCNSL occurring among individuals with autoimmune disorders receiving immunosuppressive medications have been reported (Balci, et al 2017, Finelli 2005). Herein, we utilized data from a large, nationally-representative cohort of older adults (aged ≥65 years) to estimate associations between autoimmune conditions and PCNSL.

SEER is a cancer surveillance program that includes 18 cancer registries covering ~28% of the US population. Medicare is a federally-funded program that provides health insurance to US adults ≥65 years. The SEER-Medicare database is an electronic linkage of SEER and Medicare that links >94% of SEER cancer cases ≥65 years of age with their Medicare claims data (1991–2015) (Warren, et al 2002). All beneficiaries are entitled to Part-A coverage (hospital inpatient care) and approximately 96% subscribe to Part-B coverage (outpatient and physician services). Some beneficiaries chose to enrol in health maintenance organization (HMO) and are not required to submit claims to Medicare. Claims data for an additional 5% random sample of Medicare beneficiaries residing in SEER geographic areas are available.

We used a case-control study design (Engels, et al 2011). PCNSL cases were first malignant cancers classified as NHL according to SEER site recode (https://seer.cancer.gov/siterecode), had the International Classification of Diseases (ICD) for Oncology (version 3) topography codes for CNS (C70.0-C72.9), and were diagnosed between 1992–2015. We excluded cases diagnosed at autopsy or by death certificate only. To ensure adequate ascertainment of medical risk factors, we required that cases had ≥13 months of Part-A, Part-B, non-HMO coverage before PCNSL diagnosis. We randomly selected 200,000 controls from the 5% random sample of Medicare beneficiaries who were alive and cancer-free as of 1 July of the calendar year of their selection. Like cases, controls were required to have ≥13 months of Part-A, Part-B, non-HMO coverage prior to their selection. Controls were frequency-matched to cases on age, calendar year, sex, and race. We excluded subjects who had any ICD version 9 (ICD-9) diagnosis code for HIV infection or solid organ transplantation at any time during the study period (codes provided in Tables S1A, S1B).

We ascertained 31 autoimmune conditions using ICD-9 diagnosis codes at least 12 months before PCNSL diagnosis/control selection (Table S1A) to minimize differential assessment of exposures in cases closer to cancer diagnosis. A diagnosis of any specific condition required at least one inpatient, outpatient or physician claim.

We compared the characteristics of cases and controls using the chi-squared or Fisher’s exact test. To avoid biases in selecting autoimmune conditions for evaluation, we calculated the expected number of PCNSL cases with an autoimmune condition under the null hypothesis (by multiplying the number of PCNSL cases by the prevalence of autoimmune condition in controls) and included those where there were ≥11 expected exposed PCNSL cases or when the prevalence of an autoimmune condition was significantly (p<0.05) different in cases and controls. We evaluated associations between selected autoimmune conditions and PCNSL using logistic regression (Figure 1 legend). We adjusted the variance of odds ratios (ORs) for repeated selection of some controls across calendar years and inclusion of some controls who later became cases (Engels, et al 2011). The p value was adjusted by Bonferroni’s method (0.05/31 evaluated conditions=0.0016) for multiple testing (Dunnett 1955). Sensitivity analyses were conducted by 1) restricting cases to diffuse large B-cell lymphoma (DLBCL), and 2) by using a stricter definition of autoimmune conditions where they had at least one inpatient, or two outpatient or physician claims at least 30 days apart.

Figure 1:

Associations between autoimmune conditions and PCNSL

The associations with PCNSL are presented for each autoimmune condition as an adjusted odds ratio (black dots) and corresponding 95% confidence interval (horizontal axis, logarithmic scale). Separate models were fitted where the outcome was occurrence of PCNSL, and the main exposure was a specific autoimmune condition.

Abbreviations: aOR, adjusted odds ratio; CI, confidence intervals; PCNSL, primary central nervous system lymphoma.

^a Autoimmune conditions were considered present if there was at least 1 inpatient, outpatient, or physician Medicare claim.

^b Odds ratios are adjusted for age categories (65–69, 70–74, 75–79, 80–84, ≥ 85 years), sex, race/ethnicity (white, black, others/unknown), calendar year of cancer diagnosis/control selection (1992–2001, 2002–2007, 2008–2011, 2012–2015), duration of Medicare coverage (1–29, 30–54, 55–67, ≥ 68 months), and average annual number of physician visits (0–2.17, 2.18–5.16, 5.17–9.63, ≥ 9.64).

^c Associations that were significant after correction of p value for multiple comparisons by Bonferroni method.

Most of the 1,727 PCNSL cases included were DLBCL (69.3%) and were in the brain (82.5%) (Table S2). By design, cases and controls were perfectly frequency-matched on age categories, sex, race/ethnicity and calendar year of diagnosis/control selection (Table I). Cases had a slightly lower duration of Medicare coverage and total number of physician claims per year than controls.

Table I:

Comparison of the characteristics between primary CNS lymphoma cases and controls

Characteristics	PCNSL casesa (N = 1,727)	Controlsa (N = 200,000)	Chi2 test p value
	Number (%)	Number (%)
Age, years
Median (IQR)	75 (70 – 79)	75 (70 – 79)
65–69	329 (19.1)	38,123 (19.1)	_
70–74	506 (29.3)	58,592 (29.3)
75–79	462 (26.7)	53,486 (26.7)
80–84	292 (16.9)	33,812 (16.9)
≥ 85	138 (8.0)	15,987 (8.0)
Sex
Males	772 (44.7)	89,400 (44.7)	_
Females	955 (55.3)	110,600 (55.3)
Race/ethnicity
White	1,467 (84.9)	169,850 (84.9)	_
Black	63 (3.7)	7,308 (3.7)
Others/Unknown	197 (11.4)	22,842 (11.4)
Year of cancer diagnosis/control selection
1992 – 2001	335 (22.5)	45,053 (22.5)	_
2002 – 2007	376 (25.3)	50,567 (25.3)
2008 – 2011	385 (25.9)	51,794 (25.9)
2012 – 2015	391 (26.3)	52,586 (26.3)
Total Part-A, Part-B, non-HMO Medicare coverage (months)b
Median (IQR)	54 (26 – 71)	55 (30 – 67)
1 to 29	467 (27.0)	50,153 (25.1)	< 0.0001
30 to 54	373 (21.6)	46,190 (23.1)
55 to 67	330 (19.1)	58,807 (29.4)
≥ 68	557 (32.3)	44,850 (22.4)
Total number of physician claims per yearb
Median (IQR)	5.42 (2.50 – 10.00)	5.33 (2.35 – 9.84)
0 to 2.17	378 (21.9)	49,417 (24.7)	0.05
2.18 to 5.16	448 (25.9)	50,573 (25.3)
5.17 to 9.63	440 (25.5)	49,989 (25.0)
≥ 9.64	461 (26.7)	50,021 (25.0)
Any autoimmune conditionc	364 (21.1)	34,598 (17.3)	< 0.0001
Systemic/Connective tissuec
Rheumatoid arthritis	109 (6.3)	11,250 (5.6)	0.22
Sjögren syndrome	14 (0.8)	1,288 (0.6)	0.39
Systemic lupus erythematosus	25 (1.5)	1,422 (0.7)	0.0003
Sarcoidosis	< 11 (< 0.6)	414 (0.2)	0.07
Polymyalgia rheumatica	33 (1.9)	3,136 (1.6)	0.25
Ankylosing spondylitis	< 11 (< 0.6)	462 (0.2)	0.32
Polymyositis/dermatomyositis	< 11 (< 0.6)	422 (0.2)	0.22
Reactive arthritis	< 11 (< 0.6)	46 (< 0.1)	1.00
Bloodc
Autoimmune haemolytic anaemia	< 11 (< 0.6)	136 (0.1)	0.33
Immune thrombocytopenic purpura	< 11 (< 0.6)	184 (0.1)	0.22
Cardiovascular systemc
Giant cell arteritis	14 (0.8)	1,407 (0.7)	0.60
Polyarteritis nodosa	< 11 (< 0.6)	141 (0.1)	0.0008
Behçet disease	< 11 (< 0.6)	11 (< 0.1)	0.09
Wegener granulomatosis	< 11 (< 0.6)	86 (< 0.1)	0.17
Endocrine glandsc
Addison disease	< 11 (< 0.6)	279 (0.1)	0.74
Graves disease	16 (0.9)	1,629 (0.8)	0.61
Hashimoto thyroiditis	16 (0.9)	1,382 (0.7)	0.24
Skinc
Psoriasis	58 (3.4)	5,020 (2.5)	0.03
Scleroderma	13 (0.8)	1,268 (0.6)	0.54
Discoid lupus erythematosus	< 11 (< 0.6)	507 (0.3)	0.77
Vitiligo	< 11 (< 0.6)	407 (0.2)	0.78
Gastrointestinal systemc
Pernicious anaemia	56 (3.2)	6,220 (3.1)	0.75
Crohn disease	13 (0.8)	1,250 (0.6)	0.50
Ulcerative colitis	15 (0.9)	2,264 (1.1)	0.30
Autoimmune hepatitis	< 11 (< 0.6)	35 (< 0.1)	0.04
Primary biliary cirrhosis	< 11 (< 0.6)	175 (0.1)	0.67
Coeliac disease	< 11 (< 0.6)	462 (0.2)	1.00
Nervous systemc
Multiple sclerosis	< 11 (< 0.6)	613 (0.3)	0.76
Guillain-Barré syndrome	< 11 (< 0.6)	265 (0.1)	0.30
Myasthenia gravis	12 (0.7)	391 (0.2)	< 0.0001
Eyesc
Uveitis	29 (1.7)	849 (0.4)	< 0.0001
Scleritis	12 (0.7)	948 (0.5)	0.18

Abbreviations: HMO, health maintenance organization; IQR, interquartile range; PCNSL, primary central nervous system lymphoma.

Note: Counts less than 11 are suppressed in accordance with the SEER-Medicare data use agreement.

^aCases and controls were frequency-matched on age categories (65–69, 70–74, 75–79, 80–84, ≥ 85 years), sex (male, female), race/ethnicity (white, black, others/unknown), and calendar year of cancer diagnosis/control selection.

^bCalculated at least 12 months before case diagnosis/control selection.

^cAutoimmune conditions were considered present if there was at least 1 inpatient, outpatient, or physician Medicare claim.

Overall, the prevalence of any autoimmune condition was higher among cases than controls (21.1% vs. 17.3%; adjusted OR [aOR]=1.24; 95% confidence interval [CI]=1.09–1.40; p=0.0007) (Table I, Figure 1). Significant associations were observed between PCNSL and systemic lupus erythematosus (SLE; aOR=1.96; 95%CI=1.31–2.92), polyarteritis nodusa (PAN; aOR=3.99; 95%CI=1.62–9.81), autoimmune hepatitis (aOR=6.31; 95%CI=1.50–26.57), myasthenia gravis (aOR=3.40; 95%CI=1.90–6.07), and uveitis (aOR=3.86; 95%CI=2.64–5.64) (Figure 1). The association with psoriasis was borderline significant (aOR=1.29; 95%CI=0.99–1.68). Associations between PCNSL and SLE, myasthenia gravis and uveitis were significant after adjusting the p-value by the Bonferroni method. These associations were consistent when the cases were restricted to DLBCLs only or when a stricter definition of autoimmune conditions was used (Table S3).

Immunosuppression is a strong risk factor for PCNSL, and HIV-infected people and transplant recipients have very high risk of PCNSL compared to the general population (Gibson, et al 2014, Mahale, et al 2018). Transplant recipients are immunosuppressed because of the use of medications to prevent rejection. Similarly, immunosuppressants given to treat autoimmune conditions may also increase PCNSL risk. In this large population-based cohort of older adults, SLE, PAN, autoimmune hepatitis, myasthenia gravis, and uveitis were significantly associated with increased risk of PCNSL, though these conditions are rare in the general US population.

Population-based studies have shown that several autoimmune conditions are associated with increased risk of NHL due to either chronic inflammatory state, dysregulation and hyper-reactivity of B-cells along with impaired T-cell control, or immunosuppressants that are used for their treatment (Anderson, et al 2009, Fallah, et al 2014). PCNSLs have been previously reported in people with SLE (Balci, et al 2017) and myasthenia gravis (Finelli 2005) and were attributed to receipt of long-term immunosuppressants, such as mycophenolate mofetil (for SLE) or azathioprine (for SLE, myasthenia gravis). In contrast, the association between uveitis and PCNSL may represent ocular manifestations of primary intraocular lymphoma (Chan, et al 2011). Associations with PAN or autoimmune hepatitis have not been previously reported but could be attributed to high doses of immunosuppressive drugs, such as azathioprine.

Increasing rates of PCNSL among immunocompetent older adults in the US are not reflective of the patterns observed for other NHLs, where rates among HIV-uninfected people aged 60+ years plateaued starting in 2004 (Shiels, et al 2013). It had been hypothesized that increasing use of immunosuppressants to treat autoimmune disorders in older adults may be driving increasing rates of PCNSL. Indeed, we found that several autoimmune conditions are associated with increased risk of PCNSL. However, as <1% of the general population have the autoimmune disorders that we found to be associated with PCNSL, it is not plausible that treatment for these conditions could account for the observed increasing trend of PCNSL among the older adults in the US.

The strengths of our study include our use of a large sample of the US older adult (age ≥65 years) population, which included more than 1,700 PCNSLs, a very rare cancer. Furthermore, SEER registries have strict quality control measures for cancer ascertainment which contribute to the reliability of PCNSL diagnoses. Our study is limited by use of Medicare claims to assess autoimmune conditions, which may have resulted in exposure misclassification. However, our findings were consistent when using a stricter definition for autoimmune conditions. In addition, misclassification of exposure is likely to be non-differential between cases and controls, driving associations towards the null. Our study population may not be entirely representative of the general older US population as older people residing in SEER areas include a relatively large fraction of racial/ethnic minorities and are more likely to reside in urban and affluent areas than an average older person in the US.(Warren, et al 2002). Most PCNLs (>90%) are DLBCLs (Grommes and DeAngelis 2017), but ~69% of PCNSLs in our study were classified as DLBCLs. It is possible that most of the cases classified as “NHL, not otherwise specified” (~25%) are DLBCLs. A sensitivity analysis, restricting our cases to those classified as DLBCLs, confirmed our findings. We did not include information on treatment for autoimmune conditions in our analysis. Although prescription data are available from Medicare Part-D claims, these data are only available starting in 2007, and would have dramatically reduced our sample size. As this study is exploratory, we highlighted all associations that were p<0.05; however, some of these could be due to chance. We adjusted the p-values for multiple comparisons, and associations for SLE, myasthenia gravis and uveitis remained statistically significant.

As PCNSL is a rare cancer, our findings need to be replicated using other large databases. Epidemiological case-control studies which focus on individual autoimmune conditions can be conducted to validate these results. Information on immunosuppressive medications need to be obtained to evaluate whether a higher cumulative doses or duration of treatment increases the risk of PCNSL.

PCNSL among older adults is a rare malignancy with poor survival. In this study, we found novel associations between PCNSL and SLE, PAN autoimmune hepatitis, myasthenia gravis and uveitis. However, immunosuppressive treatment for these conditions cannot plausibly explain the increasing PCNSL rates over time among immunocompetent older individuals, given their very low prevalence. Further studies of the aetiology of PCNSL in older adults are needed to explain this trend.

Supplementary Material

Supp TableS1-3

Table S1A: Codes for autoimmune conditions and human immunodeficiency virus infection.

Table S1B: Codes for solid organ transplantation.

Table S2: Characteristics of PCNSL cases.

Table S3: Sensitivity analyses for assessing the associations between PCNSL and autoimmune conditions.