Self-reported history of infections and the risk of non-Hodgkin lymphoma: an InterLymph pooled analysis

Abstract

We performed a pooled analysis of data on self-reported history of infections in relation to the risk of non-Hodgkin lymphoma (NHL) from 17 case-control studies that included 12,585 cases and 15,416 controls aged 16–96 years at recruitment. Pooled odds ratios (OR) and 95% confidence intervals (95% CI) were estimated in two-stage random-effect or joint fixed-effect models, adjusting for age, sex and study centre. Data from the two years prior to diagnosis (or date of interview for controls) were excluded. A self-reported history of infectious mononucleosis (IM) was associated with an excess risk of NHL (OR=1.26, 95% CI=1.01–1.57 based on data from 16 studies); study-specific results indicate significant (I²=51%, p=0.01) heterogeneity. A self-reported history of measles or whooping cough was associated with an approximate 15% reduction in risk. History of other infection was not associated with NHL. We find little clear evidence of an association between NHL risk and infection although the limitations of data based on self-reported medical history (particularly of childhood illness reported by older people) are well recognised.

Introduction

Consistent associations between immune suppression and an increased risk of NHL have been identified in many studies and have led to the suggestion of an underlying infectious cause.^{1, 2} Some rare subtypes of NHL have been associated with specific infections although for the majority, evidence is uncertain.³ Here we report results from an international pooled analysis of case-control studies that examined data on self-reported history of infections in relation to the risk of NHL among adults.

Material and methods

Study population

We performed a pooled analysis of data from 17 case-control studies identified through the InterLymph Consortium (www.epi.grants.cancer.gov/InterLymph). Participating studies (Table 1) met the following eligibility criteria: cases diagnosed with incident histologically confirmed NHL as adults (age 16–96 years); collection of personal history of infections; and electronic data set available in March 2007. The pooled analysis was approved by the University of New South Wales Human Research Ethics Committee. Each participating study obtained local institutional ethical approval and informed consent from participants and, provided a de-identified data set with individual information for their study participants.

Table 1.

Characteristics of case-control studies participating in the pooled analysis of infections and risk of NHL

Study acronym	Location (reference)	Year	Age (years)	Matching variables	Cases		Controls
					n	Rate1 (%)	n	Rate1 (%)	Source
Aviano-N (18)	Aviano; Napoli, Italy	1999– 2002	18–84	None	225	97	504	91	Patients admitted to hospital for non-neoplastic, non-immunological conditions
BC (19)	British Columbia, Canada	2000– 2003	20–80	Age, sex, region	828	79	845	46	Random selection from client registry of Ministry of Health
Connecticut (20)	Connecticut, USA	1995– 2001	21–84	Age	597	72	716	47–69	<65y: RDD2; ≥65y: random selection from CMMS3
EpiLymph (4)	Spain	1998– 2003	17–96	Age, sex, region	435	82	630	96	Patients admitted to hospital for infectious, parasitic, mental, nervous, circulatory, digestive, endocrine, metabolic or respiratory conditions
	Germany	1999– 2002	18–82	Age, sex, region	518	87	518	44	Random selection from population registries
	Ireland	1998– 2004	19–85	Age, sex, center	142	90	208	75	Patients admitted to hospital for infectious, parasitic, mental, nervous, circulatory, digestive, endocrine, metabolic or respiratory conditions
	Czech Republic	2001– 2003	19–82	Age, sex, region	199	90	199	60	Patients admitted to hospital for infectious, parasitic, mental, nervous, circulatory, digestive, endocrine, metabolic and respiratory conditions
	France	2000– 2003	18–82	Age, sex, region	217	91	272	74	Patients admitted to hospital for infectious, parasitic, mental, nervous, circulatory, digestive, endocrine, metabolic and respiratory conditions
	Italy (Sardinia)	1998– 2004	25–81	Age, sex, region	219	93	336	66	Random selection from population registries
Italy (6)	Turin; Novara; Forli; Vercelli; Varese; Verona; Florence; Siena; Latina; Ragusa; Imperia	1990– 1993	20–74	Age, sex, region	1,640	82	1,771	74	Random selection from demographic or National Health Service files
Mayo (20)	Minnesota, Iowa, and Wisconsin, USA	2002– 2005	20–87	Age, sex, region	497	65	499	69	Patients attending a pre-scheduled general medical exam
NCI-SEER (21)	Detroit; Iowa; Los Angeles; Seattle, USA	1998– 2001	20–74	Age, sex, region, race	757	76	587	52	<65y: RDD2; ≥65y: random selection from CMS3
Northern Italy (5)	Aviano; Milan, Italy	1983– 1992	17–85	None	429	>97	1,157	>97	Patients admitted for non-neoplastic, non-immunological conditions in hospitals where cases diagnosed
NSW (9)	New South Wales, Australian Capital Territory, Australia	2000– 2001	20–74	Age, sex, region	694	85	694	61	Random selection from electoral rolls
SCALE (22)	Denmark; Sweden	1999– 2002	18–74	Age, sex, country	3,055	81	3,187	71	Random selection from population registries
UCSF (8)	San Francisco, CA, USA	1988– 1993	21–74	Age, sex, region	1,305	72	2,404	78	All ages: RDD2;plus for ≥65y: random selection from CMMS3
UK (23)	Parts of north and southwest England	1998– 2003	16–69	Age, sex, region	828	75	897	71	Random selection from general practice lists

¹Participation rate

²RDD=Random digit dialing

³CMMS=Centres for Medicare and Medicaid Service

Six studies reported response rates in cases of 65–79%, five of 81–87% and six of 90–97%. Response rates in controls were: 44–61% (six studies), 66–74% (five studies) and 75–99% (six studies). Seven of the studies used hospital controls and 10 were population-based. Information on ethnicity was available only in a minority of studies (although most participants were likely to be Caucasian). Some studies have previously reported associations by NHL subtype,^4–8 or total NHL.⁹ Overall, the studies were conducted at 40 study sites or centres in 12 countries. Organ transplant recipients and individuals with HIV, as well as hospital controls admitted for infection (n=7), were excluded.

Exposure assessment

Past history of the conditions of interest was ascertained by structured questionnaires, either self-completed (two studies), by telephone (four studies) or in-person interview (11 studies); none of the studies allowed participation of proxies. All reports on occurrence of the conditions of interest within the two years prior to NHL diagnosis (cases) or interview (controls) were excluded from the pooled analyses, to avoid the possibility that onset of disease was responsible for an increased risk of infection. No attempt made to verify self-reported data with information obtained from examination of medical records.

Tumour classification

Most studies verified NHL diagnoses by histopathology review (report review in four, slide review in two, and a mixture of both in six studies). All WHO classification subtypes of NHL, except multiple myeloma, were included in the analysis, including the combined category of chronic lymphocytic leukemia/small lymphocytic lymphoma/prolymphocytic lymphoma/mantle-cell lymphoma (CLL/SLL/PLL/MCL), as recommended for epidemiological analyses.¹⁰ Cases originally classified using the WHO/ICD-O-3 scheme (seven studies) were categorized directly into the hierarchical groupings of the InterLymph nested classification. Cases classified using earlier schemes were converted where possible to WHO subtypes using the nested classification algorithms. Site of lymphoma, for cases with extranodal NHL, was recorded in six studies (7882 cases).

Statistical analysis

Study specific odds ratios (OR) and 95% confidence intervals (CI) were computed from unconditional logistic regression models, before using a two-stage random-effects model to estimate the pooled relative risk (hereafter called ‘risk’) of NHL, and a joint fixed-effects model to estimate risk by NHL subtype. All models were adjusted for the matching variables age, sex, and region/study centre. Analyses were conducted for men and women separately and showed no statistically significant differences between the sexes. The inclusion of a priori specified potential confounders - socioeconomic status, race, smoking history and sibship size - did not change the pooled estimates by more than 10%, and thus they were not retained in the final models. Furthermore, restriction to Caucasians only or stratification by birth order (only and 1^st born, 2^nd and 3^rd born, 4^th or higher order) did not reveal any systematic differences in the pooled risks across the conditions of interest.

Heterogeneity among study-specific ORs was assessed using Cochrane´s Q statistic and the I² statistic.¹¹ In the presence of significant heterogeneity (p<0.10), forest plots and influence analyses were used to identify outlying studies. Also, sensitivity analyses were conducted to examine the potential effects of type of data collection (in-person interview, self-completed questionnaire, telephone interview); source of controls (hospital or population); case ascertainment (ascertainment of slides in at least in a 10% subset of cases); response rates in cases and controls and study region/continent. As there were no systematic differences found between studies or design factors, results have been presented for all studies combined. All statistical tests were two-sided, and the level of statistical significance for risks was set to α=0.05. Analyses were performed using STATA software version 10.0 (Stata Corporation, College Station, TX).

Results

The participating studies (Table 1) collectively contributed 12,585 cases and 15,416 controls to the analysis, from 12 countries across three continents. The distribution of pooled cases by age, sex and other characteristics broadly reflect known demographic patterns (Table 2).¹² As expected, most tumours were B cell subtype.

Table 2.

Demographic and tumor characteristics of the pooled study participants

Demographics	NHL Cases (%)	Controls (%)	Tumor characteristics	NHL Cases
Pooled total	12,585	15,416	Pooled total	12,585
Sex			B-cell NHL	10,396
Men	6814 (54.1)	8338 (54.1)	Follicular lymphoma	2,582
Women	5771 (45.9)	7078 (45.9)	Diffuse large B-cell lymphoma	3,513
Age (years)*			CLL / SLL / PLL / MCL	2,142
<20	25 (0.2)	56 (0.4)	Marginal zone lymphoma	707
20 – 29	356 (2.8)	812 (5.3)	Burkitt’s lymphoma	93
30 – 39	844 (6.7)	1537 (10.0)	Lymphoplasmacytic lymphoma	285
40 – 49	1657 (13.2)	2233 (14.5)	Hairy cell leukemia	118
50 – 59	3266 (26.0)	3511 (22.8)	Precursor B-cell	24
60 – 69	4028 (32.0)	4428 (28.7)	B-cell NOS^	932
70 – 79	2262 (18.0)	2671 (17.3)	T-cell NHL	719
80+	147 (1.2)	168 (1.1)	Mycosis fungoides / Sezary syndrome	237
Median (range)	60 (17 – 89)	58 (16 – 96)	Peripheral T-cell	370
Education / SES†			Adult T-cell leukemia / lymphoma	2
Low	5007 (39.8)	5488 (35.6)	NK/T-cell§	19
Medium	4180 (33.2)	5484 (35.6)	T-cell large granular lymphocytic leukemia	7
High	3293 (26.2)	4344 (28.2)	T-cell prolymphocytic leukemia	7
Unknown / other	105 (0.8)	100 (0.6)	Precursor T-cell	29
Race‡			T-cell NOS	48
White	4788 (38.0)	5809 (37.7)	NOS NHL	1,470
Black	183 (1.5)	305 (2.0)	Precursor NOS	57
Other / mixed	497 (3.9)	525 (3.4)	NOS	1,413
Missing	7117 (56.6)	8777 (56.9)

^*The inclusion of Hodgkin lymphoma cases and matched controls and the removal of HIV-related cases in several studies led to an imbalance in the age distribution of controls compared to the non-Hodgkin lymphoma cases

^†Education / socioeconomic status (SES) groups in each study was based on the tertile distribution of years of education (10 studies) or SES levels obtained from census data (2 studies) in controls

^‡Race was generally not recorded in the European studies.

^§NK/T-cell lymphoma nasal type/aggressive NK-cell leukemia

^{^}Not otherwise specified.

Table 3 shows the number of affected cases and controls, OR and 95% CI for a self-reported history of infections. Risk of NHL was elevated with a self-reported history of infectious mononucleosis (IM, OR=1.26, 95% CI=1.01–1.57), but was slightly lower for two other infections (measles: OR=0.84, 95% CI=0.76–0.93; whooping cough/pertussis: OR=0.85, 95% CI=0.78–0.93). Study-specific results for IM (Figure 1) showed significant heterogeneity (I²=51%, p=0.01); one study (UCSF) had a significantly decreased OR and, in general, the elevated ORs were found in studies using hospital controls rather than population-based controls. The risks of NHL subtypes in relation to a report of previous IM, is shown in Table 4: IM showed significantly elevated OR specifically for the CLL group (OR=1.71, 95% CI=1.30–2.25), and for T-cell NHL (OR=1.41, 95% CI=1.01–1.97), but not for other subtypes. Study-specific and NHL sub-type specific results for measles and whooping cough showed no significant heterogeneity.

Table 3.

Self reported childhood and adult infections, in relation to risk of NHL

	Ever during lifetime
Condition	Cases	Controls	OR (95% CI) *	Heterogeneity
	Ever / never	Ever / never
Childhood infection
Measles	4470 / 1016	6446 / 1483	0.84 (0.76 – 0.93)	8 studies, p=0.48, I2=0.0%
Mumps	3124 / 1829	4514 / 2769	0.99 (0.91 – 1.07)	7 studies, p=0.75, I2=0.0%
Whooping cough / pertussis	1513 / 2618	2294 / 4090	0.85 (0.78 – 0.93)	6 studies, p=0.87, I2=0.0%
Chickenpox	4851 / 1753	6406 / 2363	0.94 (0.87 – 1.03)	10 studies, p=0.40, I2=4.8%
German measles / rubella	1695 / 2599	2410 / 4076	0.95 (0.85 – 1.07)	7 studies, p=0.14, I2=38.3%
Scarlet fever	451 / 3241	698 / 5264	0.98 (0.77 – 1.24)	6 studies, p=0.01, I2=65.4%
Diphtheria	56 / 1256	56 / 1321	0.99 (0.60 – 1.62)	2 studies, p=0.24, I2=26.8%
Adult infections
Infectious mononucleosis	521 / 11412	612 / 13484	1.26 (1.01 – 1.57)	16 studies, p=0.01, I2=51.4%
Tuberculosis	251 / 10305	244 / 12346	1.16 (0.96 – 1.39)	13 studies, p=0.78, I2=0.0%
Diarrhoea	169 / 2215	249 / 2559	0.88 (0.67 – 1.17)	7 studies, p=0.13, I2=39.0%
Herpes simplex / cold sores	2593 / 5490	2999 / 6738	1.05 (0.97 – 1.13)	13 studies, p=0.26, I2=17.8%
Genital herpes	85 / 4753	127 / 5781	0.96 (0.72 – 1.28)	5 studies, p=0.96, I2=0.0%
Shingles	552 / 5509	655 / 7876	1.05 (0.93 – 1.19)	8 studies, p=0.43, I2=0.0%

^*Odds Ratio (OR) and 95% confidence interval (CI) from a two-stage random-effects model, adjusted for age (in 5 year intervals), sex, and study centre.

Figure 1.

Study-specific results for a self-reported history of infectious mononucleosis in relation to risk of non-Hodgkin lymphoma.

Table 4.

The impact of infectious mononucleosis on the risk of NHL, stratified by NHL subtype

NHL subtype	Infectious mononucleosis
	Ever/never	OR (95% CI)*
Controls	612 / 13484	1.00 (ref)
B-cell	442 / 9618	1.12 (0.98 – 1.28)
Follicular lymphoma	141 / 2340	1.10 (0.90 – 1.34)
Diffuse large B-cell lymphoma	161 / 3238	1.06 (0.89 – 1.28)
CLL / SLL / PLL / MCL	74 / 2038	1.71 (1.30 – 2.25)
Marginal zone lymphoma	27 / 661	0.97 (0.64 – 1.47)
Burkitt lymphoma	6 / 85	1.09 (0.46 – 2.57)
Lymphoplasmacytic lymphoma	9 / 270	1.14 (0.57 – 2.27)
Hairy cell leukemia	2 / 114	0.53 (0.13 – 2.23)
Precursor B-cell	0 / 24	-
B-cell NOS	22 / 848	0.98 (0.62 – 1.53)
T-cell	43 / 660	1.41 (1.01 – 1.97)
Mycosis fungoides / Sezary syndrome	10 / 222	1.03 (0.53 – 1.97)
Peripheral T-cell	28 / 333	1.72 (1.14 – 2.59)
Precursor T-cell	3 / 26	1.87 (0.54 – 6.44)
T-cell NOS	2 / 46	1–32 (0.30 – 5.74)
NHL NOS	36 / 1134	1.00 (0.70 – 1.42)
Precursor NOS	2 / 44	1.00 (0.23 – 4.39)
NHL NOS	34 / 1090	1.00 (0.70 – 1.45)

^*Odds ratio (OR) and 95% confidence interval (CI) from a joint-fixed effects model, adjusted for age (in 5 year intervals), sex, and study centre

Discussion

In this pooled analysis of case-control study data on self-reported history of infections in relation to the risk of non-Hodgkin lymphoma (NHL) we found that a reported history of infectious mononucleosis (which is caused by infection with EBV in adolescence or later life) was associated with a moderate increase in risk of NHL, although there was significant heterogeneity across studies. However, the strength of the association varied by NHL subtype, being highest in those that have been previously associated with EBV (CLL and T-cell NHL).¹³ We cannot completely exclude the possibility of some diagnostic overlap with Hodgkin lymphoma, a proportion of which are caused by EBV and which has been linked with past history of IM in previous studies, although this is unlikely given the pathology review process.^{3, 14} Furthermore, the data on IM from all participating studies is based on self-report, rather than linkage with medical records – it is unclear if a formal laboratory-confirmed diagnosis of IM was ever made. It is possible, therefore, that the finding is unreliable, although it is also possible that a diagnosis of IM may be more reliably reported than certain other conditions. Indeed, a recent prospective study does not support the view that EBV is a major cause of NHL in apparently immunocompetent individuals.¹⁵ That said, significantly elevated odds ratios were specific to the CLL group and for T-cell NHL, lending more credence to a possible real association. Most cases of NHL occurred among people in later life and, where relevant data were available, IM occurred among children or young adults, years prior to diagnosis. No significant differences were identified, in the risk of NHL, by age at diagnosis of IM (data not shown).

In contrast, a self-reported history of measles or whooping cough was associated with a small reduction in risk of NHL. Since there is little other literature to support the view that certain infections reduce the risk of NHL, it is most likely that this finding is spurious and probably due to chance or systematic bias in reporting. These infections typically occur in childhood and have become less common with vaccination; furthermore, cases tend to be relatively elderly and recall of infections may not be accurate. No other report of infection was associated with NHL. The strengths of this pooled analysis include the large study size and the ability to stratify risk estimates by study design features and NHL subtypes. Thus the pooled data do not support a generalised effect of past infection on NHL risk, or risk of a specific NHL subtype.

In summary, therefore, we find little clear evidence of an association between risk of NHL as an adult and infection (as a child or adult) although the weakness of data based on self-report of medical history (particularly when older people are asked to report on childhood illness) is well recognised.^{16, 17} Indeed the principle limitation of this analysis is the reliability of self-reported exposure data on past medical history underpinning any associations. This problem is common to all of the individual participating studies and emphasises the need for primary source data on past infections (from medical records for example) in such investigations. Furthermore, the individual studies did not use a common pre-coded questionnaire and so data were collected in slightly different ways and for certain conditions – such as “diarrhoea”, the cause is ill-defined. Another problem common to some but not all of the participating studies is the lack of diagnostic detail recorded for the disease outcome (lymphomas) – ICD-O3 is the current gold-standard for classification of haematological malignancies, but was not universally applied across studies. Understanding of the epidemiology of non-Hodgkin lymphomas has been impaired by an inability to adequately measure relevant exposures and outcomes – it is critical that these issues are addressed in future studies.