Non-Hodgkin lymphoma and obesity: A pooled analysis from the InterLymph Consortium



Nutritional status is known to alter immune function, a suspected risk factor for non-Hodgkin lymphoma (NHL). To investigate whether long-term over, or under, nutrition is associated with NHL, self-reported anthropometric data on weight and height from over 10,000 cases of NHL and 16,000 controls were pooled across 18 case-control studies identified through the International Lymphoma Epidemiology Consortium. Study-specific odds ratios (OR) were estimated using logistic regression and combined usinga random-effects model. Severe obesity, defined as BMI of 40 kg m−2 or more, was not associated with NHL overall (pooled OR = 1.00, 95% confidence interval (CI) 0.70–1.41) or the majority of NHL subtypes. An excess was however observed for diffuse large B-cell lymphoma (pooled OR = 1.80, 95% CI 1.24–2.62), although not all study-specific ORs were raised. Among the overweight (BMI 25–29.9 kg m−2) and obese (BMI 30–39.9 kg m−2), associations were elevated in some studies and decreased in others, while no association was observed among the underweight (BMI < 18.5 kg m−2). There was little suggestion of increasing ORs for NHL or its subtypes with every 5 kg m−2 rise in BMI above 18.5 kg m−2. BMI components height and weight were also examined, and the tallest men, but not women, were at marginally increased risk (pooled OR = 1.19, 95% CI 1.06–1.34). In summary, whilst we conclude that there is no evidence to support the hypothesis that obesity is a determinant of all types of NHL combined, the association between severe obesity and diffuse large B-cell lymphoma may warrant further investigation. © 2007 Wiley-Liss, Inc.

Non-Hodgkin lymphomas (NHL) can arise following rare inherited disorders of the immune system, long-term immunosuppressive drug therapy and viral infections such as human immunodeficiency virus (HIV) and Epstein-Barr virus (EBV). In such instances, severe immunosuppression resulting from exposure usually leads to the development of specific NHL subtypes. For the majority of NHL however, the cause remains unknown but it is suspected that factors which affect the immune system are involved. In particular, it has been suggested that the degree of adiposity might be important since over (as well as under) nutrition can alter immune function.1, 2 However, while several epidemiological studies have reported associations between excess weight and NHL3–15 the evidence is far from conclusive.16–28 Here we present a pooled analysis of self-reported height and weight on over 10,000 NHL cases and 16,000 controls from 18 case-control studies identified through the International Lymphoma Epidemiology Consortium (InterLymph:

Material and methods

Through the InterLymph forum, 18 case-control studies of NHL with anthropometric data collected across 13 countries in parts of North America, Europe and Japan between 1983 and 2004 were identified. Study designs are briefly outlined in Table I, and more details are published elsewhere.3, 8, 14, 18, 24, 29–36 Cases were identified using rapid ascertainment techniques, while controls were randomly selected from population registers (8 studies), outpatient clinics (3 studies) or inpatients (7 studies) hospitalized for a variety of nonneoplastic conditions such as circulatory, digestive or respiratory problems, or with traumatic or nontraumatic orthopaedic conditions. The appropriate ethical committees' approval was granted for each study and informed consent was given by all participants.

Table I. Characteristics of Case-Control Studies Included in the Pooled Analysis
StudyLocationYear of diagnosisAge rangeCases (N = 10,453)Participation (%)Controls source (N = 16,507)NParticipation (%)Reference
NCI-SEERDetroit, Michigan; Iowa; Los Angeles, California; Seattle, Washington, USA1998–200120–7052776<65 years RDD; 65+ years random selection from Centers for Medicare and Medicaid Services, stratified by study area, age, sex and race468528
Nebraska NHL studyNebraska, USA1999–200220–7538774RDD, frequency matched by age and sex5357814
Mayo Clinic Phase 1Iowa, Wisconsin, Minnesota, USA2002–200518+49966Frequency matched by age, sex and county of residence49970n/a
UCSFSan Francisco, USA1988–199521–741,30472RDD, frequency matched by age, sex, and county of residence2,402783
British Columbia StudyVancouver and Victoria, British Columbia, Canada2000–200420–8282878Random selection from Client Registry of the Ministry of Health, frequency matched by age, sex and region8484636
UKYorkshire, Lancashire, South Lakeland and parts of Southwest England1998–200316–6983370Random selection from general practice lists, individually matched by age, sex and region of residence1,1416929
SCALEDenmark and Sweden2000–200218–743,05581Random selection from population register, frequency matched by sex and age3,1877124
EpiLymph IrelandSix hospitals on the East Coast of the Republic of Ireland2001–200318–8013590Hospital controls matched by age (±5 years), sex and study region2087530
EpiLymph FinlandFinland2001–200318–8087 Hospital controls matched by age (±5 years), sex and study region75 n/a
EpiLymph GermanyLudwigshafen/Upper Palatinate, Heidelberg/Rhine-Neckar County, Würzburg/Lower Frankonia, Hamburg, Bielefeld and Munich1999–200218–8049688Random selection from population register, individually matched by sex, age and study region7104431
EpiLymph FranceAmiens, Dijon and Montpellier2000–200318–8020691Hospital controls matched by age (±5 years), sex and study region2767430
EpiLymph Czech Republic1 centre in Czech Republic2001–200318–8019590Hospital controls individually matched by age (±5 years), sex and study region3046030
EpiLymph SpainBarcelona, Tortosa, Reus and Madrid1998–200218–8042882Hospital controls matched by age (±5 years), sex and study region6319632
EpiLymph Italy2 centres in Italy1998–200418–8022293Random selection from population census list, matched by age (±5 years), sex and study region3366630
Northern ItalyAviano & Milan1983–199217–79429>97Patients admitted for acute, nonneoplastic, nonimmunologic conditions in the hospitals where cases diagnosed1,157>9718
ItalyAviano & Naples1999–200218–8422597Hospital controls, frequency matched by age (in 5-year bands), sex and study centre to cases of lymphohematopoietic neoplasms including NHL and hepatocellular carcinoma5049133
HERPACC1Aichi Cancer Centre, Nagoya, Japan1988–200018–79416≈99Random sample of patients not diagnosed with cancer, individually matched by age and sex2,260≈9934,35
HERPACC2Aichi Cancer Centre, Nagoya, Japan2001–200418–79181≈99Random sample of patients not diagnosed with cancer, individually matched by age and sex966≈9935

NHL diagnoses were pathologically confirmed and subsequently coded to the World Health Organisation (WHO) classification37 (15 studies), the REAL classification (the 1999–2002 Italian study), or Working Formulation (North Italy and UCSF). Cases with HIV were excluded. Diagnostic codes from the different studies were combined as previously described.38 The analysis here considers specific B-cell subtypes of NHL (diffuse large B-cell lymphoma: ICDO3 codes 9679/3, 9680/3, 9684/3; follicular lymphoma: 9690/3, 9691/3, 9695/3, 9698/3; chronic lymphocytic leukaemia/small lymphocytic lymphoma: 9670/3, 9823/3; marginal zone lymphoma: 9689/3, 9699/3; mantle cell lymphoma: 9673/3; Burkitt lymphoma: 9687/3, 9826/3 and other unspecified B-cell lymphoma: 9671/3, 9728/3), and T-cell lymphomas as a whole (9700/3, 9701/3, 9702/3, 9705/3, 9708/3, 9709/3, 9714/3, 9716/3, 9717/3, 9718/3, 9719/3, 9729/3, 9827/3) as well as NHL in total (defined by the above ICDO3 codes and 9591/3, 9675/3, and 9727/3).

In all studies, information on anthropometrics, demographics, lifestyle, occupations and medical histories were collected by in-person or telephone interviews. For the purposes of the present analyses, anonymized data were provided and checked for inconsistencies before coding uniformly. Within each study, height in meters was categorized using sex-specific quintiles of the height distribution among controls, and data were then combined across studies to reflect the relative position, rather than the absolute value, of this variable. In the statistical analysis, the referent category for height was taken as the 3rd quintile, since this central group contains the median and has the narrowest range. Usual adult weight was requested in 10 studies (Nebraska, UCSF, SCALE and EpiLymph studies). Elsewhere different questions were used (weight at diagnosis/interview (HERPACC1, HERPACC2); 1 year (NCI-SEER, British Columbia, North Italy, Italy); 2 years (Mayo Phase 1); or 5 years (UK) prior to diagnosis/interview).

For the pooled analysis, body mass index (BMI) was computed by dividing weight in kilograms by the square of height in metres where each study's weight variable was considered at the closest time point prior to diagnosis/interview, or else the usual adult weight. BMI was grouped using the World Health Organisation categories of underweight (<18.5 kg m−2), normal (18.5–24.99 kg m−2), grade 1 overweight (25–29.99 kg m−2), grade 2 obese (30–39.99 kg m−2) and grade 3 obese (40 kg m−2 or more).39 For a person 1.7 m (5′7″) tall, these cut-off points relate to weights of 53 kg (118 lb), 72 kg (159 lb), 87 kg (191 lb) and 116 kg (255 lb) respectively. Socioeconomic status was defined by the level of education attained, except in British Columbia and the UK where self-reported household income and a census-based household deprivation indicator were used respectively; and no socioeconomic status information was collected in the Japanese studies (HERPACC1 and 2).

Statistical analysis followed similar methods to those employed in previous InterLymph pooling projects.40–44 Firstly, individual data were combined in an unconditional logistic regression model adjusting for study, age, sex and race. To test for between-study heterogeneity, this model was compared using the likelihood ratio test with the model that included an additional term for interaction between the anthropometric variable and a variable indexing the studies. Heterogeneity was assumed to be present when the likelihood ratio test yielded a p-value <0.05. This flexible approach utilizes all data and provides one statistic to test for heterogeneity. Where the likelihood ratio test was not statistically significant, the pooled adjusted OR and 95% CI computed from all individual data in an unconditional logistic regression model are presented.

Between-study heterogeneity was further examined among risk estimates at each category of the anthropometric variables. Study-specific odds ratios (OR) and 95% confidence intervals (CI) adjusted for sex, age and race were computed using unconditional logistic regression.45 For each category of height or BMI, the study-specific ORs were combined using a random effects meta-analysis to produce a combined OR and corresponding 95% CI. The extent of heterogeneity for each category was indicated by Cochran's Q-statistic which was considered statistically significant when p < 0.10. The I2-statistic was also reported to describe the percentage of total variation in the study-specific ORs which was due to heterogeneity.46

Since the ORs were diverse across studies, a variety of approaches were applied to explore heterogeneity.47 To assess relative obesity within study populations rather than the absolute value, BMI was grouped into quintiles based on the control distributions within each study before combining the relative quintile groupings across studies; these analyses are not presented here since their findings were similar to those reported. Sensitivity analyses using various stratifications and subsets of data were also conducted.48 Study-specific ORs were combined by continent, study design and time period (corresponding to the original lymphoma classification used) as well as by level of participation. Given that the study-specific associations with BMI were heterogeneous in all analyses, forest plots with ORs pooled by continent were judged to be the most informative. Pooled ORs stratified by study design are also presented.

Within studies, analyses were performed separately for men, women, Caucasian subjects and persons aged 18–65. The resulting study-specific ORs were combined in a random-effects meta-analysis to examine heterogeneity. Potential confounding factors, such as smoking, alcohol and socioeconomic status, were assessed by comparing study-specific regression models with and without the confounding factor using the likelihood ratio test. A factor was considered a confounder when the likelihood ratio test was significant and the adjusted OR changed by more than 10%. Continuous variables corresponding to 10 cm increases in height and 5 kg m−2 increases in BMI were created to assess trends. All analyses were conducted using Stata.49


The pooled dataset from the 18 case-control studies comprised anthropometric information from 10,453 cases of NHL and 16,507 controls. Most cases (85%) were diagnosed with a B-cell lymphoma, 5% with a T-cell lymphoma and for 11%, immunophenotype was not known. The 3 most common NHL subtypes were diffuse large B-cell lymphoma (DLBCL) (32%), follicular lymphoma (FL) (22%) and chronic lymphocytic lymphoma/small lymphocytic lymphoma (CLL/SLL) (16%). A slightly higher proportion of cases were men (57%), 90% of all cases were Caucasian and the median age was 60 years. Cases tended to be older in age, of white race and of lower socioeconomic status than controls (data not shown).

Height distributions among male and female controls varied by study; for both sexes, the median height was highest in the American studies, generally decreased from Northern to Southern Europe, and was lowest in the two Japanese studies (data not shown). Among men, compared to the third quintile the odds ratio was increased in the highest quintile (OR = 1.19, 95% CI 1.06–1.34), but was close to one in the lowest two quintiles (Supplementary Table I). When examining trend within studies, no consistent population pattern emerged; most studies showed no evidence of a trend with 10 cm increases in height, 6 a significant positive trend and 2 a significant negative trend (data not shown). Similar patterns were observed for the majority of NHL subtypes. Little association between height and NHL, or its subtypes, was observed among women (Supplementary Table I).

Figure 1 gives the distribution of BMI among controls by study. Like height, studies conducted in the US had the greatest median BMI, and Japan the lowest. When BMI was classified using WHO categories, associations between BMI and NHL were heterogeneous between studies (likelihood ratio test: χ2 = 139.1, p < 0.0001). Study-specific ORs showed that the heterogeneity was most marked in Grade 1 overweight, where ORs ranged from 0.50 (95% CI 0.34–0.74) in EpiLymph Italy to 1.70 (95% CI 1.02–2.84) in EpiLymph Ireland and Grade 2 obese (ranging from OR = 0.42, 95% CI 0.24–0.74 in EpiLymph Italy to OR = 1.78, 95% CI 1.36–2.32 in UCSF) (Figs. 2b and 2c). In the underweight and Grade 3 obese categories, where the numbers of subjects were small, ORs were also diverse (ranging from OR = 0.27, 95% CI 0.03–2.34 in EpiLymph Ireland to OR = 3.14, 95% CI 0.41–23.9 in EpiLymph Finland; and from OR = 0.19, 95% CI 0.02–1.58 in EpiLymph Germany to OR = 4.23, 95% CI 1.51–11.9 in UK, respectively) (Figs. 2a and 2d). Trends with a 5 kg m−2 increase in BMI above 18.5 kg m−2 were significantly increased in 2 studies, significantly decreased in 4 studies and showed little effect in the remaining studies (Fig. 3). ORs were pooled across North America, Northern Europe, Southern Europe and Japan. In North America, a homogeneous increased OR was suggested for Grade 1 overweight (Fig. 2b) but no effect was found among Grade 3 obese (Fig. 2d), and with the exception of the Californian study (UCSF), no significant positive trends were observed (Fig. 3). Heterogeneity was still evident when the analyses were restricted to population-based studies conducted in the period 1998–2005; to those designed to code to the WHO classification; or to those where control participation rates were 70% or more. Similarly study-specific ORs were heterogeneous among men or women; subjects aged 18–65; or Caucasian subjects (data not shown).

Figure 1.

Box-Whisker plot of body mass index among controls by study. Body mass index considered to be: Underweight if <18.5 kg m−2; Normal weight-for-height if 18.5–24.99 kg m−2; Grade 1 Overweight if 25–29.99 kg m−2; Grade 2 Obese if 30–39.99 kg m−2; and Grade 3 Obese if ≥40 kg m−2 Ref.39.

Figure 2.

(a) Meta-analysis of the risk of NHL associated with BMI <18.5 kg m−2 (Underweight) compared to BMI 18.5–24.99 kg m−2 (Normal weight). Overall test for heterogeneity: Q = 13.0, p = 0.73; Variation in odds ratios (OR) attributable to heterogeneity: I2 = 0.0%. For continents: North America: Q = 1.04, p = 0.90, I2 = 0.0%; Northern Europe: Q = 7.87, p = 0.25, I2 = 23.7%; Southern Europe: Q = 1.03, p = 0.80, I2 = 0.0%; Asia (Japan): Q = 1.38, p = 0.24, I2 = 27.5%. Test for heterogeneity between continents: Q = 1.82, p = 0.61. Pooled odds ratios by study design were: Population-based studies: OR = 0.91, 95% CI 0.68–1.21, Q = 6.75, p = 0.56, I2 = 0.0%; Clinic-based studies: OR = 0.92, 95% CI 0.65–1.31, Q = 1.47, p = 0.48, I2 = 0.0%; Hospital-based studies: OR = 0.67, 95% CI 0.39–1.17, Q = 3.79, p = 0.58, I2 = 0.0%. Test for heterogeneity between study designs: Q = 1.04, p = 0.59. (b) Meta-analysis of the risk of NHL associated with BMI 25–29.99 kg m−2 (Grade 1 overweight) compared to BMI 18.5–24.99 kg m−2 (Normal weight). Overall test for heterogeneity: Q = 60.0, p < 0.001; Variation in odds ratios (OR) attributable to heterogeneity: I2 = 70.0%. For continents: North America: Q = 2.76, p = 0.60, I2 = 0.0%; Northern Europe: Q = 25.0, p = 0.001, I2 = 72.1%; Southern Europe: Q = 8.59, p = 0.04, I2 = 65.1%; Asia (Japan): Q = 0.02, p = 0.90, I2 = 0.0%. Test for heterogeneity between continents: Q = 23.4, p < 0.001. Pooled odds ratios by study design were: Population-based studies: OR = 0.97, 95% CI 0.82–1.14, Q = 41.6, p < 0.001, I2 = 80.8%; Clinic-based studies: OR = 0.99, 95% CI 0.82–1.20, Q = 0.44, p = 0.80, I2 = 0.0%; Hospital-based studies: OR = 0.91, 95% CI 0.72–1.16, Q = 14.0, p = 0.03, I2 = 57.1%. Test for heterogeneity between study designs: Q = 3.93, p = 0.14. (c) Meta-analysis of the risk of NHL associated with BMI 30–39.99 kg m−2 (Grade 2 obese) compared to BMI 18.5–24.99 kg m−2 (Normal weight). Overall test for heterogeneity: Q = 59.7, p < 0.001; Variation in odds ratios (OR) attributable to heterogeneity: I2 = 69.8%. For continents: North America: Q = 8.18, p = 0.08, I2 = 51.1%; Northern Europe: Q = 18.1, p = 0.01, I2 = 61.2%; Southern Europe: Q = 7.88, p = 0.05, I2 = 62.0%; Asia (Japan): Q = 0.01, p = 0.93, I2 = 0.0%. Test for heterogeneity between continents: Q = 25.4, p < 0.001. Pooled odds ratios by study design were: Population-based studies: OR = 1.06, 95% CI 0.83–1.34, Q = 41.3, p < 0.001, I2 = 80.7%; Clinic-based studies: OR = 1.22, 95% CI 0.90–1.67, Q = 0.03, p = 0.99, I2 = 0.0%; Hospital-based studies: OR = 0.77, 95% CI 0.60–0.98, Q = 8.51, p = 0.20, I2 = 29.5%. Test for heterogeneity between study designs: Q = 9.81, p = 0.007. (d) Meta-analysis of the risk of NHL associated with BMI ≥40 kg m−2 (Grade 3 obese) compared to BMI 18.5–24.99 kg m−2 (Normal weight). Overall test for heterogeneity: Q = 21.9, p = 0.15; Variation in odds ratios (OR) attributable to heterogeneity: I2 = 26.8%. For continents: North America: Q = 2.89, p = 0.58, I2 = 0.0%; Northern Europe: Q = 15.3, p = 0.03, I2 = 54.4%; Southern Europe: Q = 0.69, p = 0.88, I2 = 0.0%. Test for heterogeneity between continents: Q = 2.91, p = 0.23. Pooled odds ratios by study design were: Population-based studies: OR = 1.33, 95% CI 0.88–2.00, Q = 11.4, p = 0.18, I2 = 29.7%; Clinic-based studies: OR = 0.57, 95% CI 0.26–1.22, No test for heterogeneity as only 1 study; Hospital-based studies: OR = 0.51, 95% CI 0.25–1.05, Q = 2.07, p = 0.91, I2 =0.0%. Test for heterogeneity between study designs: Q = 8.41, p = 0.015. [Color figure can be viewed in the online issue, which is available at]

Figure 3.

Meta-analysis of the risk of NHL associated with 5 kg m−2 increase in BMI above 18.5 kg m−2 (Normal weight and above). Overall test for heterogeneity: Q = 87.5, p < 0.001; Variation in odds ratios (OR) attributable to heterogeneity: I2 = 79.4%. For continents: North America: Q = 15.5, p = 0.004, I2 = 74.1%; Northern Europe: Q = 37.4, p < 0.001, I2 = 81.3%; Southern Europe: Q = 5.32; p = 0.15; I2 = 43.6%; Asia (Japan): Q = 0.12, p = 0.73, I2 = 0.0%. Test for heterogeneity between continents: Q = 29.0, p < 0.001. Pooled odds ratios by study design were: Population-based studies: OR = 1.02, 95% CI 0.92–1.13, Q = 57.7, p < 0.001, I2 = 86.1%; Clinic-based studies: OR = 1.04, 95% CI 0.94–1.14, Q = 0.34, p = 0.84, I2 = 0.0%; Hospital-based studies: OR = 0.85, 95% CI 0.79–0.92, Q = 6.09, p = 0.41, I2 = 1.4%. Test for heterogeneity between study designs: Q = 23.4, p < 0.001. [Color figure can be viewed in the online issue, which is available at]

Statistically significant between-study heterogeneity was also present for the 3 most common NHL subtypes (likelihood ratio tests for WHO BMI and DLBCL: χ2 = 104.2, p = 0.002; FL: χ2 = 82.7, p = 0.003; CLL/SLL: χ2 = 58.7, p = 0.04). For these three subtypes, as for NHL as a whole, study-specific ORs varied around one in all WHO BMI groups, with tests for heterogeneity in the two-stage random effects model being significant among Grade 1 overweight and Grade 2 obese (DLBCL: Supplementary Figures 1a–1d; FL: Supplementary Figures 3a–3d; CLL/SLL: Supplementary Figures 4a–4d. In the underweight and Grade 3 obese groups, the meta-analyses generally suggested that ORs were more homogeneous and the combined risk estimates were not significantly different from one. The pooled OR for DLBCL among Grade 3 obese was increased (OR = 1.80, 95% CI 1.24–2.62, Q = 16.7, p = 0.40, I2 = 4.4%), being elevated in North America and Northern Europe, but as with all analyses in this BMI group, study-specific risk estimates were diverse, based on small numbers of subjects, and with wide and overlapping confidence intervals (Supplementary Fig. 1d). Like NHL as a whole, a 5 kg m−2 increase in BMI did not consistently increase the risk of DLBCL (Supplementary Fig. 2) or the other subtypes (data not shown). ORs for the rarer B-cell lymphomas and T-cell lymphoma were mostly not significantly different between studies, probably due to the small number of cases, and there was little suggestion of associations between these NHL subtypes and BMI (Supplementary Table II).

Pooling data from studies with the highest WHO BMI prevalences of overweight/obese controls (EpiLymph Czech Republic, Nebraska, Mayo Phase 1, EpiLymph Italy, EpiLymph Germany, Italy-Aviano and Naples, and EpiLymph Finland) gave more homogeneous ORs (likelihood ratio test: χ2 = 32.3, p = 0.12). Within this subset of seven studies, there was still little evidence that higher than average BMI increases the risk of NHL and its subtypes (Table II). These findings were consistent when data were stratified by sex, age or race.

Table II. Number of Cases and Controls, Pooled Odds Ratios and 95% Confidence Intervals for Body Mass Index by all NHL Subtypes and the Three Most Common NHL Subtypes in Studies with the Highest Prevalence of Overweight/Obese Controls1
BMI2Controls (N = 2,963)NHL3 (N = 2,108)OR495% CIDLBCL3 (N = 659)OR495% CIFL3 (N = 457)OR495% CICLL/SLL3 (N = 381)OR495% CI
  • 1

    Studies with highest prevalence of overweight/obese controls were EpiLymph Czech Republic, Nebraska, Mayo Phase 1, EpiLymph Italy, EpiLymph Germany, Italy-Aviano and Naples, and EpiLymph Finland.

  • 2

    Body mass index grouped using WHO categories where <18.5 kg m−2 is considered Underweight; 18.5–24.99 kg m−2 Normal weight; 25–29.99 kg m−2 Grade 1 Overweight; 30–39.99 kg m−2 Grade 2 Obese; and ≥40 kg m−2 Grade 3 Obese.

  • 3

    NHL, non-Hodgkin lymphoma; DLBCL, diffuse large B-cell lymphoma; FL, follicular lymphoma; CLL/SLL, chronic lymphocytic leukaemia/small lymphocytic lymphoma.

  • 4

    Odds ratios and 95% confidence intervals adjusted for study, sex, age and race were estimated using unconditional logistic regression.

  • 5

    Test for heterogeneity was conducted by testing for evidence of interaction between BMI and studies using the likelihood ratio test.

WHO category (kg m−2)
 Missing5272  31  16  7  
Test for heterogeneity5χ2 = 32.2p = 0.12 χ2= 25.6p = 0.27 χ2 = 24.7p = 0.05 χ2 = 17.8p = 0.16


The present InterLymph analysis, which is based on 18 studies from 13 countries, found little evidence to support the hypothesis that excess weight-for-height is associated with NHL. A slightly increased OR amongst the tallest men was observed compared to those who were of mid-range height but no association was found among women. The large number of subjects included in this analysis enabled examination of risks for subtypes of NHL. While findings for most were consistent with total NHL, an increased risk for DLBCL among persons with a BMI of 40 kg m−2 or more was observed in a meta-analysis of study-specific ORs. For DLBCL, ORs were elevated with overweight/obesity in North America and amongst the most obese in Northern Europe, yet studies in either region did not show an increasing trend with a 5 kg m−2 rise in BMI. Marked heterogeneity between studies was present for all categories of BMI, which remained when studies were combined by continent, study design and time period, WHO lymphoma classification used; and when data were restricted to men or women, persons aged 18–65, Caucasians alone or studies with participation rates of 70% or more. ORs were less heterogeneous amongst studies with the greatest proportions of controls with a BMI of 25 kg m−2 or more. Of the seven studies in this subset, no effect of BMI on NHL risk was observed, and the lack of association with obesity was consistent across NHL subtypes, amongst men and women, and at age ≤45, 46–55, 56–65 and ≥66 years.

Six of the case-control studies included in this pooled analysis have previously published data on NHL and obesity3, 8, 11, 14, 23, 24, 50 and a further 12 are included here for the first time. Apart from case-control studies, adiposity has been investigated in cohorts where height and weight were measured9, 10, 12, 13, 25–27 or self-reported,5, 15, 20, 21 and among persons with a hospital discharge for obesity.4, 19, 22 Cohort studies have the advantage of prospectively collected information, although not necessarily at a relevant time point. Positive associations with obesity have been reported for some cohorts,5, 9, 10, 12, 13, 15 but not for others4, 19–22, 25–27; and a further case-control investigation nested within a cohort reported a reduced risk based on measured height and weight.16 Only one additional study of case-control design—which is not part of the InterLymph consortium—has published its findings, observing an excess risk of NHL with obesity.6

Hitherto only a few individual case-control studies and two cohort studies have considered lymphoma subtypes, proposing an association with excess adiposity for DLBCL, but less so for FL and CLL/SLL.8, 11, 14, 15, 21, 22, 24, 50 A recent meta-analysis of published risk estimates suggested a slight increased risk of NHL, particularly DLBCL based upon data from both case-control and cohort studies.51 The pooled analysis presented here has the advantage of being less susceptible to positive publication bias since it is based on all studies within the InterLymph consortium that collected anthropometric information—around 40% of the data have not been presented before. Another advantage of pooling individual records is that it permits uniform categorization of data, as well as the assessment of the effects of potentially confounding factors. In this regard, adjustment for smoking and alcohol consumption did not greatly alter the risk estimates.

With respect to potential biases, participation rates were generally lower in controls than cases, and a particular concern is whether controls are representative of the populations from which cases were drawn. It is reassuring to note that pooling data from studies with control participation rates of 70% or more gave findings similar to those reported overall. Nonetheless, it is still possible that poor control participation could have influenced our findings since we cannot rule out the possibility that those with obesity-related health problems (e.g., type 2 diabetes, cardiovascular disease, respiratory difficulties and chronic musculoskeletal problems) may have been (more or) less likely to participate. If the latter applied, the increased risks in the highest BMI category could be an artefact of differential case-control participation.

The rapidly changing prevalence of obesity is a growing public health problem, and to further investigate the issue, age-standardized data calculated from height and weight measurements were sourced from the World Health Organization Global Database on BMI ( Interestingly, the relative order of the overweight (25–29.99 kg m−2)/obesity (≥30 kg m−2) prevalence across studies among our controls and that of the corresponding country-specific WHO BMI prevalence from around the year 2000 are not strongly correlated (Spearman's ρ = 0.41, p = 0.08) (Fig. 4). WHO data place the USA, Germany and the UK at the top while our self-reported information rate the Czech Republic, USA and Italy as having the highest overweight/obesity prevalence. Whilst differences between our data and WHO are likely to be related to factors such as age, sex and time period, they serve to illustrate the rapidly changing patterns and wide variations that exist around the world.

Figure 4.

Comparison of Control and WHO overweight/obesity prevalences by study. Overweight (BMI 25–29.99 kg m−2) and Obesity (BMI ≥ 30 kg m−2) prevalence from the World Health Organisation (WHO) Global Database on Body Mass Index ( WHO prevalence was derived from the most recent published age- and sex- standardized BMI data calculated from height and weight measured in clearly defined population samples; these data were largely from around the year 2000. The relative order of control overweight/obesity prevalences across studies was not similar to that from data reported on the WHO Global database for BMI (Spearman's ρ = 0.41, p = 0.08).

Self-reports of anthropometric information is known to be inexact, with height tending to be overestimated and weight underestimated.52 The nature of individual misreporting is likely to be complex, being related not only to their actual size but also to other factors such as age and sex. In a cohort of British adults, for instance, where self-reported and measured data were compared, height was overestimated most by older people, shorter men and heavier women, while the greatest underestimation of weight was amongst heavier men and women.53 This tendency for people to report BMI closer to “normal” may have diluted our odds ratios. It is also possible that weight loss associated with lymphoma may have influenced the recall of cases differently to that of controls. Because of this, at interview subjects were either asked to recall their usual weight or their weight at a specified times before diagnosis/interview, and restricting the analyses to the 6 studies (NCI-SEER, Mayo Phase 1, British Columbia, UK, North Italy and Italy) where data were requested at 1 or 5 years prior to diagnosis yielded similar results to the findings overall.

Whilst BMI derived from height and weight acts as an easily obtained estimate of adiposity, its use as a marker of obesity has several potential weaknesses. Across different ethnic groups, for example, a given BMI may not correspond to the same proportions of body fat.54 Moreover since the index was originally devised as a means of assessing average body composition among sedentary individuals of working age it may not truly reflect the degree of adiposity across the population as a whole. For instance, among the elderly where muscle mass may have started to decline, body fat mass may be underestimated by BMI whereas amongst athletes it may be overestimated. To account for the potential variation in BMI as a marker of body fat across different populations, we grouped our data according to study-specific control distributions as well as WHO BMI categories. We also repeated the analyses restricting data to Caucasians, and to North American and Northern European studies combined. Sensitivity analyses were conducted too among persons aged 65 or less (71% of our subjects), and among those who were not regular heavy exercisers where this information was requested (NCI-SEER, British Columbia and HERPACC2). These additional investigations gave similar findings to the presented results. More specific estimates of adiposity may be derived from total body fat mass and, as a marker for abdominal fat distribution, waist-to-hip ratios, but such data were not obtained in the studies included here and have only rarely been investigated with respect to NHL elsewhere, showing little effect.15, 21, 26

In conclusion, this pooled analysis of case-control studies from 13 countries, crossing 3 continents, did not find an association between NHL and increased BMI. ORs were raised in studies from some countries, namely the US, Canada and Northern European nations, but even within this group, heterogeneity was observed, questioning the validity of a combined odds ratio. The findings presented here were based on individual data from a large number of subjects enrolled in 18 studies, pooling of which were accomplished through the InterLymph consortium. Some of the advantages of this pooled analysis include information on confounders and NHL subtypes as well as data on height and weight, the constituent components of BMI. One potential confounding factor not assessed here is diet but dietary data will be examined, in conjunction with BMI, in a future InterLymph pooled analysis. Such investigations may further elucidate whether NHL or its subtypes are associated with obesity per se.