FOLLICULAR LYMPHOMA INCIDENCE AND MORTALITY IN RELATION TO OVERWEIGHT, OBESI TY AND PHYSICAL ACTIVITY: A META - ANALYSIS

In the last few years, there has been a growing interest in exploring the association between risk factors such as overweight, obesity and physical activity, and incidence of various cancers. Meta-analysis was performed to investigate the risk ratio of follicular lymphoma incidence and mortality in overweight and obese individuals, and in individuals with a different physical activity levels using the random-effects model. A literature search through September 2016 was performed. Case-control studies accounted for over 2.100 cases and 12.700 controls, whereas cohort studies accounted for over 2.600 cases in cohort of about 3.000.000 individuals. In overweight individuals (body mass index between 25 and 29.99 kg/m) risk ratio for the development of follicular lymphoma was 1.03 (0.95-1.11; 95% CI; p = 0.51) and in obese (body mass index ≥ 30 kg/m) it was 1.15 (1.01-1.31; 95% CI; p = 0.04) when compared to individuals with normal body mass index (< 25 kg/m). The risk ratio of specific follicular lymphoma mortality in overweight was 0.59 (0.38-0.91; 95% CI; p = 0.02), while in obese patients it was 1.08 (0.681.71; 95% CI; p = 0.75). In patients with the highest physical activity levels, the risk ratio for follicular lymphoma occurrence was 0.95 (0.75-1.21; 95% CI; p = 0.68) when compared to patients that had the lowest physical activity levels. In summary, our meta-analysis has shown statistically significant direct association between obesity and follicular lymphoma incidence.


Introduction
Follicular lymphoma (FL) is the second most common subtype of non-Hodgkin lymphoma (NHL) in Europe and North America, and accounts for about 20-30 percent of all NHL cases (1). Although it is known that the risk factors, such as alcohol (2), cytogenetic abnormality (3), immunosuppression, and some autoimmune diseases (4), may influence FL incidence, indubitable risk factor for its occurrence is not yet known.
In the last few years there has been a growing interest in exploring an association between risk factors as obesity (OB) and physical activity (PA), and incidence of various cancers (5,6). Obesity, the prevalence of which has lately increased in many countries (7), is associated with a chronic state of inflammation. Thus, as the principal source of inflammatory cytokines, adipose tissue may be linked with the development of cancer (8,9). For many malignancies over-weight (OW) and OB are noted as risk factors (10,11). However, the authors' findings regarding FL are controversial. While some claim that there is an association between these risk factors and FL incidence (12)(13)(14)(15), others deny it (16,17). Furthermore, current research on reduction of PA associated with OB is focused on their link with various cancers (6). So far, there has been no metaanalysis of the association be-tween OW and OB and a relative risk of FL mortality, while the influence of PA on FL incidence emphasized only in one published meta-analysis (18).
The aim of this study was to evaluate the association of OW and OB with FL incidence and mortality, as well as association of PA with FL incidence, using the both cohort and case-control studies. Also, we investigated the influence of OW, OB and PA on FL incidence depending on the type of study, gender and region in which examinees lived.

Study selection
Using PubMed and searching the reference list of retrieved articles, the literature search through September 2016 was performed. The key words searched were: follicular lymphoma, overweight, obesity, body mass index, meta-analysis, physical activity, mortality. No language restrictions were imposed. Also, references cited in the corresponding articles were incorporated in the search. Both prospective and case-controls studies were included in the metaanalysis. Abstracts of articles, narrative studies, letters to the authors and cross-sectional studies were not included in this meta-analysis.
A study was relevant if it contained data reporting on the association between body mass index (BMI) and incidence and mortality of FL as well as the association between PA and incidence of FL. We identified a total of 39 articles with data that were potentially eligible for inclusion in this meta-analysis, of which 36 were related to FL incidence (31 were included in the analysis of influence of OW and OB, and 9 were included in the analysis of the influence of PA on FL incidence, with 4 studies overlapping, that is containing both topics), while 3 studies, all with a prospective design, were related to FL mortality. A total of 19 studies related to FL incidence were excluded because these articles did not include data on BMI values in relation to FL incidence. From the studies of FL mortality, one was excluded because it lacked data on mortality in relation to BMI. At the end of the identification process there were 17 applicable studies left with the data on FL incidence (14 for analysis of the influence of OW and OB (12)(13)(14)(15)(16)(17)(18)(19)(20)(21)(22)(23)(24)(25)(26), and 7 for analysis of the influence of PA, from which 4 were previously included (13,15,17,25,(27)(28)(29) as well as 2 studies with data on association between OW and OB and FL mortality (30,31).

Data analysis
For both cohort and case-control studies, we extracted the following data: authors last names; publication year; study type; source (region); gender and age; case -total cohort or case-control size; time of follow-up; method of BMI determining; type and assessment of PA; the highest and the lowest PA level; adjustments (for cohort studies) or matching (for case-control studies); the risk ratio (RR) and hazard ratio (HR) for cohort studies and the odds ratio (OR) for case-control studies, with 95% confidence interval for all (95% CI).

Statistical analysis
OW was defined as BMI values between 25 and 29.99 kg/m 2 (pre-obese), OB as BMI values over 30 kg/m 2 (obese class I), while the high OB was defined as BMI values over 35 kg/m 2 (obese class II) according to the WHO criteria (32). Normal BMI (<25 kg/m 2 ) values were the reference category for all computing related to BMI. For analysis of the influence of PA, we compared the highest to the lowest PA level. The lowest level was the reference category for all comparisons. For studies with the highest PA level as the reference category, instead of the lowest level, the inverse value was used. If risk estimates for PA were reported for more than one period, the most recent data were used. For studies that reported recreational and occupational activity separately the former was preferred, and if a publication included data on total activity, these data were used. For all analysies, if study reported risk estimates for men and women separately, values for both were analyzed.
The summary RR with 95% CI was computed to assess the risk for FL incidence and mortality in overweight and obese examinees in relation to examinees with normal BMI, or the risk for the development of FL in examinees with the highest versus the lowest PA level. In the analysis of FL incidence (influence of OW and OB, and PA), we performed subgroup analysis depending on the type of study, gender and region in which the examinees lived. While analyzing the mortality depending on the BMI index, it was calculated as a specific/univariate or nonspecific/multivariate mortality.
DerSimonian-Laird method for the random-effects model was used to compute the outcome of study-specific RR (33). To assess statistical heterogeneity among studies, the Q and I2 statistics were used (34). Heterogeneity was defined as mild (I2 of 25%), moderate (I2 of 50%) and severe (I2 of 75%). Publication bias was noticed on funnel plots and was assessed by using Egger's test (35). Statistical significance was defined for p-value < 0.05. All statistical analyses and graphics were performed with Comprehensive Meta-Analysis (Version 3; Biostat, Inc, Englewood, NJ).

Results
For studies observing the influence of OW and OB on FL incidence, a total number of cohort studies were 7 and studies included 1,979 cases in cohort of 2,712,088 individuals; a total number of case-control studies were 7 and these studies included 1,959 cases and 11,725 controls.
For studies observing the influence of OW and OB on FL mortality, a total number of cohort studies were 2 and studies included 493 cases in a cohort of over 215,000 individuals.
For studies observing the influence of PA on FL incidence, a total number of cohort studies were 4 (two of which overlapped with previous included 7 cohort studies related to FL incidence) and studies included 490 cases in a cohort of 661,878 individuals; a total number of case control studies was 3 (two of which overlapped with previous included 7 case-control studies related to FL incidence) and these studies included 583 cases and 4,583 controls (Tables 1 and 2).  In overweight examinees, the statistically significant RR for specific mortality was 0.59 (0.38-0.91; 95% CI; p = 0.02), while for non-specific was 0.72 (0.46-1.12; 95% CI; p = 0.14). There was no statistically significant heterogeneity among studies. There was no statistically significant publication bias ( Figure 1 and Table 3).
In both obese and high obese examinees, an RR of FL incidence was 1.11 (0.98-1.25; 95% CI; p = 0.10). There was no statistically significant heterogeneity among overall studies (Figure 2).

Discussion
This research was set with the aim of assessing the importance of BMI and PA for the development of FL. Our meta-analysis of cohort and case-control studies pointed out that overweight individuals did not have increased risk for the development of FL in comparison to individuals with normal weight, while obese individuals had 15% higher risk compared to non-obese. Overweight individuals had a higher risk of FL specific mortality than those with normal weight. However, a note of caution is mandatory since the number of included publication with data on mortality is small. Results of cohort and case-control studies related to the association between BMI and FL incidence did not show statistically significant difference.
In previous meta-analyses the association between BMI and FL incidence was investigated and statistically non-significant positive association between BMI and the relative risk for the development of FL was found. Thus, in a meta-analysis by Larsson and Wolk (2011) relative risk of FL and 5kg/m 2 BMI increase were positively associated (1.03; 0.93 to 1.13; 95% CI), but statistically non-significant (36). In addition, the second meta-analysis from the same authors (2007) demonstrated statistically nonsignificant association between obese individuals and the relative risk for the development of FL (37). However, this meta-analysis did not include data on overweight individuals (37). Nonetheless, the findings of the current study do not support the previous research. Moreover, the results of this study indicated that OB individuals had increased risk for the development of FL.
In terms of smaller body fat percentage in men, it may be assumed that the females would be more susceptible for the development of FL. Also, considering that subcutaneous adipose tissue is predominate in women, it may be assumed that visceral adipose tissue has a much more essential role for the development of many cancers, including FL (38,39). This type of body fat is associated with obesity-related diseases (40), because it is more harmful and inflammatory active than subcutaneous fat which is generally thought to be responsible for OB. However, based on the results of our meta-analysis, there was no increased risk of FL associated with either overweight and obese men or women.
Finally, a direct association between OW and OB, and FL incidence is more pronounced among Americans as compared to Europeans. But, none of these differences were statistically significant. For both overweight and obese American individuals, increased risk of FL was formally not significant (p = 0.06). These results seem to be consistent with other research which has not found the association between BMI and NHL incidence across strata of geographic region (36). In the study of Castillo et al. (2014) for both overweight and obese American individuals the RR for the development of diffuse large B-cell lymphoma was significantly increased, but not in Europeans (41). Although different life habits and harmful effects of the environment may cause the differences between these two groups of individuals, no association was identified in this meta-analysis. In addition, more research on this topic is necessary before the association between geographic area and the risk of FL is more clearly understood.
The prevalence of OB is increasing and is associated with other risk factors such as increased intake of high-fat food and decreased PA (7,42). McTiernan (6) proposed that excessive weight and a sedentary lifestyle are linked with about 25% of cancers, and that PA may decrease risk for various cancers by several mechanisms. In this meta-analysis only, for both men and women statistically inverse association of PA with the risk for the development of FL was found. However, this subgroup analysis could have been affected by small sample size. Thus, this finding cannot be extrapolated to all the patients. Thus, the summary risk estimate derived from all studies showed no significant influence of physical activity on the risk of FL. The findings of this study are consistent with the data obtained in other meta-analyses which focused not only to FL but also generally to NHL (18) or NHL and Hodgkin lymphoma (43).
There are numerous studies that were designed to explore the association between OB and cancer (5,44), but the mechanisms of this causal link are still unclear. Basically, the focus is on three hormonal systems that include insulin and insulinlike growth factor (IGF) axis, sex steroids and adipokines. Obesity-related inflammatory cytokines, genetic background, obesity-related hypoxia, and oxidative stress were also noted as possible causes, and more other etiological factors which may have a role in the occurrence and development of cancer (45).
This meta-analysis has some limitations. First, other obesity-related factors, such as change of weight during the time, diet and caloric intake, may also influence the risk for the development of FL. Second, the sample size may be a source of erroneous conclusions. For example, statistically significant inverse association between BMI and FL mortality was found in overweight individuals, while non-significant direct association was presented in obese. Furthermore, these results are not conclusive because of the dependence on FL subtype, treatment, comorbidity and SEER stage. Third, determining of BMIs in all included studies was based on the principle of personal declaration. Any discrepancies, such as false weight loss in obese, may have an unquestionable impact on the final results. Fourth, the BMI indicates total body fat (combined subcutaneous adipose tissue and visceral adipose tissue) (45), and therefore it cannot assess what type of body fat has a greater impact -visceral adipose tissue (VAT) or subcutaneous adipose tissue. And finally, all associations either direct when the effects of excessive weight or OB were analyzed, or inverse when the influence of PA was analyzed, may be influenced by publication bias. In this metaanalysis, there was not evidence of publication bias in most of the studies.
In summary, this meta-analysis of cohort and case-control studies has identified the direct association between obesity and FL incidence. The research has also shown statistically significant association between BMI and FL mortality in overweight individuals, as well as association between PA and the risk of FL incidence in both male and female individuals. It could be argued that these last two results could have been affected by small sample size. The current data highlights the importance of obesity for the risk of FL developments. It would be interesting to assess the impact of other risk factors such as insulin and IGF, sex steroids and adipokines involved in the association between OB and malignant lymphomas. Future meta-analyses on the current topic with a larger number of included studies are recommended to confirm these results.