Alcohol and tobacco use and risk of multiple myeloma: A case‐control study

Abstract Multiple myeloma (MM) is the second most common hematological cancer and causes significant mortality and morbidity. Knowledge regarding modifiable risk factors for MM remains limited. This analysis of an Australian population‐based case–control family study investigates whether smoking or alcohol consumption is associated with risk of MM and related diseases. Incident cases (n = 789) of MM were recruited via cancer registries in Victoria and New South Wales. Controls (n = 1,113) were either family members of cases (n = 696) or controls recruited for a similarly designed study of renal cancers (n = 417). Adjusted odds ratios (OR) and 95% confidence intervals (CI) were estimated using unconditional multivariable logistic regression. Heavy intake (>20 g ethanol/day) of alcohol had a lower risk of MM compared with nondrinkers (OR = 0.68, 95% CI: 0.50–0.93), and there was an inverse dose–response relationship for average daily alcohol intake (OR per 10 g ethanol per day = 0.92, 95% CI: 0.86–0.99); there was no evidence of an interaction with sex. There was no evidence of an association with MM risk for smoking‐related exposures (p > 0.18). The associations between smoking and alcohol with MM are similar to those with non‐Hodgkin lymphoma. Further research into potential underlying mechanisms is warranted.


INTRODUCTION
Multiple myeloma (MM) is a plasma cell neoplasm arising from the malignant transformation of mature postgerminal center B cells [1].
MM is typically preceded by the asymptomatic precursor condition monoclonal gammopathy of undetermined significance (MGUS), which progresses to multiple myeloma at an average rate of 1% per year [2].
Few risk factors for multiple myeloma have been firmly established, and most known risk factors are nonmodifiable; advanced age [8], male sex [4], black African ancestry, and positive family history [9] have all been linked to increased MM risk. Apart from certain chemical and occupational exposures [10 -12], body mass index (BMI) is the only well-established modifiable risk factor for MM [13].
Although lifestyle factors such as tobacco smoking and alcohol use are responsible for a large proportion of cancers [14], research on common modifiable risk factors for MM has been limited by the inability of smaller observational studies to detect effects of public health significance [4]. Large observational studies and meta-analyses have indicated that while tobacco smoking is unlikely to be associated with MM risk [15 -17], alcohol consumption may be inversely associated [15,18 -22], and this inverse association is possibly stronger for women, and wine drinkers [20,23]. Observational studies investigating non-Hodgkin lymphoma risk have identified a similar inverse association with alcohol [24,25].
The aim of this study is to investigate whether tobacco and alcohol consumption are associated with the risk of MM, and whether the association between alcohol and MM is modified by sex.

Study population and recruitment
To investigate the effect of these common modifiable risk factors on MM risk, we conducted an analysis using the Epidemiology of Multi- During recruitment, cases were asked for consent to invite family members to participate as controls ( Table 1). The EMMA study aimed to recruit as controls at least one family member selected from living relatives unaffected by hematological malignancy; preference was for the same-sex sibling closest in age to the case, followed by any sibling of the same sex, and if available, the case's spouse or partner was also recruited. Multiple sibling controls were recruited from some families to help balance numbers for those cases without siblings.
We also included additional controls from the Consortium for the Investigation of Renal Malignancies (CONFIRM) study, a case-control family study of renal cancer with a similar design and questionnaire (see Supporting Information 1).

Data collection
Consenting cases and controls completed self-administered questionnaires on lifestyle, health and medical history, family history, residential and occupational history, and diet [27].

Statistical analysis
Alcohol-related exposures investigated include average daily ethanol intake (continuous and categorical) and beverage type. Average ethanol intake in grams per day was calculated based on reported frequency, quantity and type of alcoholic beverages consumed in the year starting 2 years prior to interview. For each type of alcoholic beverage, we converted frequencies to daily equivalents and estimated the volume consumed in grams per day. The ethanol intake in grams per day was then estimated using the average ethanol content per 100 g of each type of alcoholic beverage from the Australian Food Composition Database [28]. Grams of ethanol for each beverage type were summed to give daily average ethanol intake overall, and for each of wine, beer, and spirits separately. Participants were subsequently categorized as nondrinkers (0 g/day), moderate drinkers (1-20 g/day), or heavy drinkers (>20 g/day) based on the National Health and Medical Research Council guidelines current at the time of recruitment and data collection (one Australian standard drink contains 10 g of pure ethanol) [29].
For tobacco, we investigated smoking status, pack-years (including an ever-smoking indicator), smoking duration, and intensity [30]. Participants were categorized according to smoking status (never vs. ever -at least seven cigarettes/week for a year, and forever smokers, current vs. former smokers -ceased smoking at least 2 years prior), duration of use (total years, mean-centered), smoking intensity (average cigarettes per day, mean-centered), mean-centered pack-years ((duration × intensity)/20), age at initiation (years), and time since cessation (years).
In the primary analysis cases of MM and related diseases including MGUS were combined, as were controls from EMMA and CONFIRM studies.

Covariates
Covariates included in the primary analyses included sex, age (continuous), state (Victoria or New South Wales), and country of birth (Australia/New Zealand, Europe/UK, or other). Analyses examining alcohol exposures were additionally adjusted for smoking status, and smoking analyses were adjusted for alcohol intake (continuous).
Covariates for inclusion in our models were selected based on the literature and following causal diagram analysis (Supporting Information 3).
The risk of MM associated with various exposures was estimated using unconditional multivariable logistic regression to estimate odds ratios (OR) and 95% confidence intervals (CI). All statistical tests were two-sided and p-values < 0.05 were considered statistically significant.
Robust standard errors were used to account for clustering within sibships. To investigate the effect of beverage type while holding total alcohol intake constant, we estimated beverage-specific substitution effects: for example, the effect of substituting one additional standard drink (10 g ethanol) per day of wine for one fewer standard drink of other alcohol on MM risk, by subtracting the regression coefficient for the estimated effect of nonwine alcoholic beverages, from the regression coefficient for wine [31]. We assessed potential two-way interactions between sex and alcohol intake using the Wald-test for the interaction term in models for sex and alcohol consumption. Participants missing data for key exposures or covariates were excluded from the analysis ( Figure 2).
Sensitivity analyses were performed adjusting for BMI (continuous), since there is some evidence for bidirectionality in the associations between BMI and both alcohol and smoking [32 -35]. We also performed sensitivity analyses restricted to EMMA study participants to assess potential bias that may be introduced by control selection, matching or strong familial correlation for risk factors: (1)

Sensitivity analyses
Sensitivity analyses for alcohol and smoking exposures were generally consistent with the primary findings (see Supporting Information 2).
This was true for those sensitivity analyses using conditional logistic regression, in matched sets of cases either with spouses, or with siblings, as well as in unconditional regression analyses excluding siblings

F I G U R E 4 Alcohol consumption and multiple myeloma risk, stratified by sex
(EMMA participants only) or restricted to non-EMMA controls. There were no substantial differences in results from analyses adjusting for BMI.

DISCUSSION
In this large case-control family study, we observed an inverse association of alcohol consumption with MM risk, both for heavy drinkers rela-tive to nondrinkers and with increasing alcohol consumption. However, we found no conclusive evidence of differences in this association by sex. Nor did we find any substantial association between tobacco usage and myeloma risk.
Despite accumulating epidemiological evidence for this association, the biological mechanism by which alcohol consumption might reduce MM risk is not yet understood. It has been suggested that low alcohol intake can improve insulin sensitivity, and thus might indirectly influence the risk of MM via diabetes-or obesity-linked mechanisms [36].

F I G U R E 5 Tobacco consumption and multiple myeloma risk
Mammalian target of rapamycin (mTOR) signaling, a target relevant for MM tumorigenesis [37], was found to be inhibited in human lymphoma xenograft models by chronic low dose ethanol [38]. Others have suggested that resveratrol, found in grape skin and red wine, could reduce MM risk, despite its low oral bioavailability [39 -41]. In vitro, resveratrol has demonstrated inhibition of STAT3 and NF-κВ, suppression of MM cell proliferation and potentiation of the apoptotic effect of bortezomib [42,43]. Other polyphenols found in wine, beer, and dark spirits have also demonstrated anti-tumorigenic properties via NF-κВ and other pathways in MM cells, and differences in phenolic content and concentration could potentially contribute to the previously reported differences in association with myeloma risk by alcoholic beverage type [44,45]; however, alcoholic beverage type substitution analyses did not support these hypotheses. Another potential mechanism is via the effect of alcohol on inflammatory markers; moderate to high alcohol consumption (15-30 g/day) is inversely associated with circulating interleukin-6 (IL-6), a cytokine which can stimulate the growth of myeloma cells and has been associated with poor prognosis, and circulating C-reactive protein (CRP) a surrogate for IL-6 [46 -48].
While other studies have suggested there may be a stronger inverse association for women compared with men for alcohol consumption and MM risk [18,20,21], this study did not demonstrate statistical interaction between alcohol and sex, despite finding larger inverse effect sizes for alcohol consumption and MM risk for women. Substantially larger samples may be necessary to convincingly infer or exclude interaction.
In the primary analysis, we identified no substantial associations for smoking status, pack-year history, time since cessation, smoking duration, age at smoking initiation, or smoking intensity with risk of MM.
This general lack of association is consistent with most epidemiological literature investigating smoking and MM [4,16,17,23].
A strength of this study is its family-based design, with stronger motivation for control participation, potential reductions in volunteer and recall bias [49] and the improved cost-effectiveness of an integrated recruitment process [50]. With volunteer controls sourced from the general population, it is becoming increasingly difficult to achieve satisfactory response rates [51]. Other strengths of this study include a large incident case population, adjustment for known confounders in the analysis, and the use of sibling controls which could reduce confounding by unmeasured early life or genetic factors [52].
One limitation of this study was our inability to completely differentiate lifetime alcohol abstainers from those who may have been prompted to more recent intake reduction. Although the study examined a historical alcohol-exposure window 2 years prior to questionnaire, alcohol-intake reduction is often associated with ill health and advancing age, and as such we cannot entirely exclude reverse causation or residual confounding bias in our results [53]. Another limitation was the inability to directly examine ethnicity. Individuals of African ancestry have been found to have an elevated risk of MM. However, this could not be adequately examined due to an insufficient number of African-background participants. We found the overseas country of birth to be associated with increased MM risk, which indicates that some early life exposures or genetic factors may potentially contribute to MM risk. Due to age restriction, results may not apply to individuals aged 75 or older.
The family-based design also has some inherent limitations; exposures tend to be correlated within families, which means that family-based studies may have less power to detect certain associations than similar studies with unrelated controls. [54] We might expect this to be pronounced especially for sibling-controls with shared genetic and early-life exposures, and perhaps with shared later-life socio-environmental exposures for spouse-controls, disregarding potential assortative mating [49,50]. Given that tobacco and alcohol consumption are complex traits, for which there is evidence for both genetic and environmental influences, this could have affected precision [55 -58]. Yet the findings of this family-based study were similar to those from previous studies using population controls, even when unrelated controls were excluded from the analysis.
This suggests that the simultaneous use of multiple types of familial controls might mitigate against statistical inefficiency for certain exposures.

CONCLUSION
To our knowledge, EMMA is the first multi-center case-control study to investigate the epidemiology of MM in Australia. This study extends the evidence base for alcohol, tobacco, and MM risk by examining a novel Australian study population, and the use of family controls complements previous findings from other observational study designs [16,20].
Although this study finds an inverse association between alcohol consumption and MM risk, we do not recommend alcohol consumption as a measure for MM prevention as other studies have found that any level of alcohol consumption increases overall cancer risk and all-cause mortality [59].
Further research investigating the potential causality and mechanisms underlying the observed MM-alcohol association is recommended. While a randomized controlled trial would be inappropriate due to ethical considerations, other designs, such as Mendelian randomization, could provide additional insight into causality.

ACKNOWLEDGMENTS
The authors wish to thank Gianluca Severi and Melissa Southey for their contributions to the study design and execution.

DATA AVAILABILITY STATEMENT
The data that support the findings of this study are not publicly available due to privacy or ethical restrictions.

CONFLICT OF INTEREST
The authors declare no conflict of interest.