Multiple myeloma (MM) is one of the most common haematological malignancies(Reference Huang, Chan and Lok1). Although survival from MM has improved with therapeutic advances, it is costly to treat, and it remains responsible for a substantial burden of mortality, morbidity and increasing health care costs(Reference Mafra, Laversanne and Marcos-Gragera2). While our knowledge of MM aetiology has improved, there is much that remains to be learnt, and an improved understanding of the association between common modifiable factors and MM risk is necessary to better inform prevention strategies.
Diet represents an essentially universal and highly modifiable factor that could potentially influence MM risk. To date, most of the evidence examining associations between diet and MM risk comes from studies examining individual dietary components, and the findings are mostly inconclusive. While meta-analyses of vegetable(Reference Sergentanis, Psaltopoulou and Ntanasis-Stathopoulos3) and meat consumption(Reference Caini, Masala and Gnagnarella4) did not identify evidence of associations with MM risk, the results of two meta-analyses of observational studies suggest that fish consumption might be inversely associated with MM risk(Reference Caini, Masala and Gnagnarella4,Reference Wang, Wu and Zhu5) , potentially via n-3 associated pathways(Reference Shah, Parikh and Castro6). Inflammation(Reference Bataille, Boccadoro and Klein7–Reference Jourdan, Tarte and Legouffe12) and insulin-related(Reference Pollak13–Reference Castillo, Mull and Reagan15) pathways have also been linked to myeloma pathogenesis.
There are fewer published studies examining associations between MM risk and dietary patterns. Dietary pattern analysis uses scores or indices to characterise overall diet, which can have benefits over an individual food or nutrient approach(Reference Hodge and Bassett16). One prospective analysis found that a diet with higher inflammatory potential, as measured by higher empirical dietary inflammatory pattern (EDIP) score, was associated with increased risk of MM, as were diets with higher insulinaemic potential, as measured by higher scores on the empirical dietary indices for hyperinsulinaemia (EDIH) and for insulin resistance (EDIR) but only for men, and with wide CI(Reference Lee, Fung and Tabung17–Reference Tabung, Wang and Fung20). Prospective analyses of the Nurses’ Health Study and Health Professionals Follow-up Study found higher dietary quality, as measured by Alternative Healthy Eating Index (AHEI)-2010 score(Reference Chiuve, Fung and Rimm21), was associated with reduced MM-specific mortality(Reference Lee, Fung and Tabung22). Unexpectedly, there was reported a weak positive association between AHEI score and MM risk, again with wide CI(Reference Lee, Fung and Tabung17). An analysis of the healthful plant-based dietary index(Reference Satija, Bhupathiraju and Rimm23) (hPDI) score in the NIH-AARP diet and health study reported an inverse association with MM risk for the highest quartile compared with lowest quartile(Reference Castro, Parikh and Eustaquio24).
We aimed to investigate whether specific diet-related exposures, such as fish intake, the dietary indices listed above and a healthy lifestyle index, are associated with MM risk.
Methods
We used data from the Epidemiology of Multiple Myeloma in Australia (EMMA) study to examine the influence of diet on MM risk. EMMA is an Australian population-based family case–control study which was designed to investigate the epidemiology of MM; the study design, conduct and recruitment have been previously described in detail(Reference Cheah, Bassett and Bruinsma25). EMMA recruited incident cases of MM, primarily via cancer registries in Victoria (2010–2016) and New South Wales (2013–2016), Australia; some additional cases were recruited via clinic-based pathways in each state. Eligible cases were within 12 months of diagnosis, aged between 20 and 74 years, with histologically confirmed MM or plasmacytoma. Some cases recruited with the non-malignant precursor condition monoclonal gammopathy of undetermined significance (ICD-10: D47.2), although eligible for recruitment to EMMA, have been excluded from this analysis. The recruited controls were siblings or spouses of cases. Eligibility criteria for cases and controls are presented in Table 1.
Epidemiology of Multiple Myeloma in Australia eligibility criteria
* Cases with a sole diagnosis of monoclonal gammopathy of unknown significance (ICD-O-3: M9765/1), i.e. not diagnosed with malignant Multiple myeloma/plasmacytoma (ICD-O-3: M9731–9734/3), were eligible for recruitment to Epidemiology of Multiple Myeloma in Australia but are not included in this analysis.
This study was conducted according to the guidelines laid down in the Declaration of Helsinki, and all procedures involving research study participants were approved by the Cancer Council Victoria Human Research Ethics Committee and the NSW Population and Health Services Research Ethics Committee. All participants provided written informed consent to participate in the study.
Data collection
All participants completed a validated FFQ(Reference Bassett, English and Fahey26), as well as questionnaires measuring other health and lifestyle exposures, occupation, residential, medical and family history. In response to the FFQ items, participants reported their usual dietary intake for the year beginning two years prior to recruitment.
Exposure definitions
Details of dietary components used to define the following dietary scores, and the healthy lifestyle index, are included in online Supplementary Information Appendix 1.
Modified alternative healthy eating index score
Dietary quality was measured using a modified alternative healthy eating index score (mAHEI), based on the alternative healthy eating index (AHEI-2010)(Reference Chiuve, Fung and Rimm21). The original AHEI-2010 was based on eleven dietary components, each scored 0 to 10, with a higher score indicating a healthier diet. There is a positive contribution to an individual’s AHEI score from higher intakes of fruit, vegetables, nuts, legumes, wholegrains, long chain n-3, PUFA and moderate intake of alcohol. Conversely, higher intakes of sugar-sweetened beverages and fruit juices, red or processed meats, trans fats and sodium reduce an individual’s score. The modified (mAHEI)-2010 index score used in this analysis excluded trans fats (data unavailable, and negligible intake in Australia) and alcohol consumption (considered separately)(Reference Hodge and Bassett16). Although alcohol is part of the original construction of the AHEI-2010 (presented as a secondary analysis), we exclude this component in the primary analysis to better examine our hypothesis that better dietary quality is associated with reduced MM risk. Alcohol is well known to be associated with other harms, and there is a growing body of evidence that supports an inverse association between alcohol consumption and MM risk(Reference Floud, Hermon and Simpson27,Reference Psaltopoulou, Sergentanis and Sergentanis28) .
Empirical dietary inflammatory pattern score
EDIP was originally developed via a process of reduced rank regression to derive a dietary pattern associated with IL-6, C-reactive protein and tumour necrosis factor α receptor type 2, with subsequent stepwise linear regression then used to identify and weight the most important of the thirty-nine predefined component food groups, with eighteen weighted food groups (nine anti-inflammatory and nine pro-inflammatory) contributing to the final model(Reference Tabung, Smith-Warner and Chavarro18). Of the eighteen dietary components included in the original EDIP, seventeen were available in EMMA (organ meat unavailable). EDIP score was calculated for EMMA participants using the available similar items from the EMMA FFQ, with the application of the weights derived from the original study(Reference Tabung, Smith-Warner and Chavarro18).
Empirical dietary index for hyperinsulinaemia
The original empirical dietary index for hyperinsulinaemia (EDIH) was developed using a stepwise multivariable linear regression approach to identify the most important of thirty-nine predefined component food groups contributing to hyperinsulinaemia, C-peptide concentration (a marker of insulin production) was the dependent variable(Reference Tabung, Wang and Fung20). This index was developed to assess the insulinaemic potential of dietary patterns and contains eighteen food groups in the final index, thirteen of which are positively associated with insulinaemia and five that are inversely associated with insulinaemia. Of the eighteen dietary components included in the original EDIH, seventeen were available in EMMA (cream soup unavailable). To investigate the association between dietary hyperinsulaemic potential and MM risk, we calculated EDIH scores using weights derived in the original study applied to available similar items from the EMMA FFQ(Reference Tabung, Wang and Fung20).
Empirical dietary index for insulin resistance
The EDIR was developed in a similar manner as the EDIH, using the ratio of triacylglycerol to high-density lipoprotein as the dependent variable, as a marker of insulin resistance, in a stepwise multivariable linear regression approach. The final index contains eighteen food groups, ten of which are positively associated with insulin resistance, and eight of which are inversely associated. Of the eighteen dietary components included in the original EDIR, seventeen were available in EMMA (cream soup unavailable). EDIR was also calculated using the EMMA FFQ items with derived weights from the original study(Reference Tabung, Wang and Fung20).
Healthful plant-based dietary index
The hPDI was developed based on eighteen food groups which can be subcategorised as healthy plant foods (contribute positive score), unhealthy plant foods (contribute reverse score) and animal food groups (contribute reverse score). All eighteen food groups were available in the EMMA FFQ. These were ranked in quintiles and participants given positive (five for highest quintile of intake through to one for lowest) or reverse scores (one for highest intake quintile and five for lowest) depending on their quintile and the food group subcategory. For example, a participant within the highest quintile of vegetable consumption (healthy plant food) would add five to their hPDI score, if they were in the highest quintile of refined grains (less healthy) this would add one to their score. The items comprising each food group were converted to serves per day and summed to get total serves per food group, participants were then categorised into sex-specific quintiles based on cut-offs determined using controls only, after which their quintile based hPDI score was calculated.
Epidemiology of Multiple Myeloma in Australia healthy lifestyle index
To measure adherence to a healthy lifestyle, we calculated a composite healthy lifestyle index score (EMMA-HI) largely based on the WCRF/AICR 2018 recommendations for cancer prevention (see Appendix 1)(Reference Shams-White, Brockton and Mitrou29). Participants were assigned scores based on the ‘healthiness’ of their categorisation of measures of dietary composition, adiposity, smoking history and alcohol consumption. For diet, higher fruit, vegetable and fibre intake, each contribute a positive score; while higher levels of consumption of ultra-processed foods, red and processed meat, and sugar-sweetened beverages are associated with lower scores. For alcohol, scores were based on reported alcohol consumption measured in grams per day, with highest scores for abstention, followed by consumption within NHMRC recommendations(Reference Conigrave, Ali and Armstrong30). For smoking, never smokers were assigned the highest score, followed by former, and then current smokers. To avoid reverse causation bias, BMI and waist circumference measures are based on participant report for a time point 5 years prior to recruitment.
Fish consumption
We investigated the association between overall fish consumption and MM risk, and between oily fish consumption and MM risk, as a continuous measure per sd increase in consumption. One sd (28 g/d) is approximately equal to two additional 100 g servings of fish per week.
Covariates
Covariate selection for analysis models was based on a review of the literature and analysis of causal diagrams (online Supplementary Figures S1–S3). All analyses were adjusted for age at questionnaire (continuous), sex, country of birth (Australia/New Zealand, Europe/UK, or other), remoteness (Major city or outside major city), socioeconomic index (quintile) and state of residence (New South Wales or Victoria). All analyses of dietary indices (except EMMA-HI) and fish consumption exposures were additionally adjusted for total energy intake (kJ/day, continuous), adiposity (BMI in kg/m2 continuous), smoking status (binary: never/ever) and pack-year history (continuous, mean-centred). The mAHEI, hPDI and fish consumption analyses further additionally adjusted for alcohol consumption (g/d, continuous). Multiple imputations were used to handle missing data for all covariates included in analysis models.
We excluded participants who had a sex-specific total energy intake in the extreme highest percentile (> 99) or lowest percentile (< 1) were missing responses to > 50 % of FFQ items or who had an extreme < 0·5 percentile or > 99·5 percentile of sex-specific BMI (5 years pre-questionnaire measure).
Analysis
The primary analysis was conducted for a participant set comprising all eligible EMMA cases and controls.
Exposures investigated included several dietary indices, a composite healthy lifestyle index and fish consumption. Dietary indices were investigated as continuous exposures per sd increase in score and per sd increase in intake for fish intake. Linearity assumption for each exposure was assessed by Wald ratio tests of quadratic and cubic terms added to the model. We also assessed the association between family history of MM and exposures of interest.
MM risk associated with each exposure was estimated using unconditional multivariable logistic regression, with robust estimates of variance accounting for clustering on sibships. In addition to the (1) primary unconditional analyses, sensitivity analyses were conducted using subsets of cases and controls to account for the case-family design, (2) unconditional logistic regression, excluding sibling controls (spouse controls only), (3) conditional logistic regression, excluding spouse controls (matched sibships) and (4) conditional logistic regression, excluding sibling controls (matched case-spouse pairs).
In addition to the above sensitivity analyses, a secondary analysis of all dietary exposures (except the lifestyle index, which includes adiposity as part of the measure) was conducted excluding BMI as a covariate. Although BMI was considered a confounder in the primary analysis, it could also potentially mediate the effect of dietary exposures. We also conducted a secondary dietary quality analysis, using a version of the AHEI that included alcohol as a component and a sensitivity analysis assessing potential statistical interaction between dietary exposures and sex.
Multiple imputation
Multiple imputation was applied to the available observed data to estimate plausible values for the missing data, incorporating uncertainty. Under the assumption that data are missing at random, multiple imputation is a valid method and is advantageous compared with a complete case analysis approach. Given the different variable types of data to be imputed, we used the multiple imputation using chained equations approach; for example, linear regression was used to impute continuous variables, and ordinal logistic regression was used to impute ordered categorical variables.
Analyses were conducted using Stata MP4 v16.1 statistical software.
Results
Of the 2183 registry-pathway cases contacted by the study team, 969 (44 %) cases consented to participate, as did an additional sixty-six cases recruited via the clinic pathway. Most consented cases (846 = 82 %) returned completed questionnaires. 746 cases were included in the analysis, after removing one ineligible case, one duplicate record, sixty-two individuals with monoclonal gammopathy of unknown significance, twenty cases in the extreme lowest or highest percentiles of estimated energy intake, four cases in the extreme lowest and highest half percentiles of BMI, and twelve cases missing more than 50 % of FFQ items. (selection flow chart: Figure 1). The median time between diagnosis and questionnaire was 10 months. We found no evidence of a non-linear effect for any exposure of interest.
Selection flow chart.
Figure 1. Long description
The flowchart illustrates the recruitment process for cases and controls in a study on multiple myeloma. It begins with case recruitment from the Victorian Cancer Registry and the New South Wales Cancer Registry, detailing eligibility criteria, doctor approaches, and consent obtained. The process includes steps such as identifying eligible cases, obtaining study consent, and completing questionnaires. It also shows the exclusion of cases due to various reasons like refusal, non-response, and ineligibility. The flowchart further depicts the recruitment of controls, including the number of individuals approached, those who consented, and those who completed the study. It highlights the final analyzed cases and controls after exclusions.
Of the 1253 family members approached to participate in the study, 870 (69 %) consented, and most of those returned completed questionnaires (741 = 85 %). Following the removal of eight controls ineligible due to a personal history of haematological cancer, ten controls in the highest or lowest percentile energy intake, ten controls in the extreme half percentiles of BMI and seven controls missing > 50 % of FFQ items, there were 706 controls included in the analysis (417 siblings and 289 spouses). Of participants included in the analysis, 74 % had complete information on all exposures of interest.
Of the participants included in the primary analysis (Table 2), men accounted for 58 % of cases and 42 % of controls. The mean BMI was similar for cases (27·1 kg/m2) and controls (27·2 kg/m2), and cases had a similar age at questionnaire completion (63·1 years) to controls (62·3 years).
Epidemiology of Multiple Myeloma in Australia participant characteristics
Table 2. Long description
The table compares participant characteristics in a study on the epidemiology of multiple myeloma in Australia. It includes data on cases and controls, with columns for the number and percentage of participants, mean age at questionnaire completion, standard deviation, mean body mass index, and standard deviation. The table also breaks down data by state, sex, alcohol consumption, smoking status, family history, and country of birth. Additionally, it provides dietary scores with median and interquartile ranges. Notable trends include a higher percentage of men among cases compared to controls, similar mean body mass index for both groups, and a slightly higher mean age at questionnaire completion for cases. The table also shows variations in alcohol consumption, smoking status, family history, and country of birth between cases and controls.
* EMMA = the Epidemiology of Multiple Myeloma in Australia study. UK = United Kingdom. Family history (1st deg) = first-degree family history of multiple myeloma. IQR = interquartile range (25th percentile, 75th percentile). Modified AHEI = modified Alternative Healthy Eating Index-2010 score. hPDI = healthful Plant-based Dietary Index score. EDIP = empirical dietary index for inflammation. EDIH = empirical dietary index for hyperinsulinaemia. EDIR = empirical dietary index for insulin resistance. Dietary scores presented in this table-based calculated after multiple imputation.
In the primary analysis (forest plot: Figure 2), a higher mAHEI score (i.e. higher quality diet, per sd increase in score) was associated with reduced MM risk (OR = 0·88, 95 % CI = 0·78, 0·98). A higher hPDI score was also associated with reduced MM risk per sd increase (OR = 0·91, 95 % CI = 0·81, 1·02), though CI were somewhat wider. The three ‘unhealthy’ dietary indices were associated with increased MM risk in the primary analysis, per sd increase in score, EDIP (OR = 1·20, 95 % CI = 1·07, 1·35), EDIH (OR = 1·15, 95 % CI = 1·02, 1·31), EDIR (OR = 1·21, 95 % CI = 1·08, 1·37). There was not clear evidence of association per sd increase of EMMA-HI healthy lifestyle index score (OR = 0·94, 95 % CI = 0·84, 1·05) or fish consumption (total fish OR = 0·95, 95 % CI = 0·85, 1·07; oily fish OR = 0·96, 95 % CI = 0·85, 1·08); each of these exposures had an estimated OR below unity, but with CI that included 1.
Dietary exposures and MM risk, results of primary analyses and additional analyses without BMI adjustment: forest plot of OR and 95 % CI. OR and 95 % CI were estimated using multivariable logistic regression. All analyses were adjusted for age at questionnaire (continuous), sex, country of birth (Australia/New Zealand, Europe/UK or other), remoteness (major city or outside major city), socioeconomic index (quintile) and state of residence (New South Wales or Victoria). All analyses of dietary indices (except EMMA-HI lifestyle index) and fish consumption exposures were additionally adjusted for total energy intake (kJ/d, continuous), adiposity (BMI in kg/m2 continuous) and smoking status (binary: never/ever) and pack-year history (continuous, mean-centred). The mAHEI, hPDI and fish consumption analyses further additionally adjusted for alcohol consumption (g/day, continuous). Multiple imputation was used to handle missing data for all covariates included in analysis models. We excluded participants who had a sex-specific total energy intake in the extreme highest percentile (> 99) or lowest percentile (< 1), were missing responses to > 50 % of FFQ items or who had an extreme < 0·5 percentile or > 99·5 percentile of sex-specific BMI (5 years pre-questionnaire measure). *no BMI adj. = without adjustment for BMI. While additional results without adjustment for BMI are plotted for most exposures, only one result is plotted for EMMA-HI as there was no adjustment for BMI in the main analysis.
Figure 2. Long description
A horizontal dot plot displays odds ratios and 95 percent confidence intervals for various dietary exposures and their association with multiple myeloma risk. The x-axis represents the odds ratio ranging from approximately 0.8 to 1.2, with a vertical red line at 1 indicating no effect. The y-axis lists different dietary exposures, including mAHEI, hDPI, EDIP, EDIH, EDIR, EMMA-HI, all fish, and oily fish, each with two rows: one for the main analysis and one for the analysis without BMI adjustment. Each dot represents the odds ratio, with horizontal lines indicating the 95 percent confidence intervals. The plot shows that some exposures, like mAHEI and hDPI, have odds ratios below 1, suggesting a potential protective effect, while others, like EDIP and EDIH, have odds ratios above 1, indicating a possible increased risk. The confidence intervals vary in width, reflecting the precision of the estimates. All values are approximated.
Secondary analyses omitting BMI as a covariate were generally consistent with the primary analyses (see Figure 2). There was a similar association between mAHEI (OR = 0·87, 95 % CI = 0·78, 0·97) and hPDI (OR = 0·90, 95 % CI = 0·80, 1·01) with reduced MM risk, and there was an increased risk of MM associated with EDIP (OR = 1·22, 95 % CI = 1·08, 1·36), EDIH (OR = 1·17, 95 % CI = 1·03, 1·32) and EDIR (OR = 1·23, 95 % CI = 1·09, 1·38). The result from the secondary analysis of the association between dietary quality including the alcohol component in AHEI was also similar to that from the primary analysis (OR = 0·88, 95 % CI = 0·78, 0·98).
Sensitivity analyses
Sensitivity analyses conducted to assess potential sex–interaction (see Figure 3) or account for case-family design using differing subsets of cases and controls (online Supplementary Table S1) were generally consistent with the primary analysis.
Dietary exposures and MM risk, results of sex-stratified analyses: forest plot of OR and 95 % CI. OR and 95 % CI were estimated using multivariable logistic regression. All analyses were adjusted for age at questionnaire (continuous), country of birth (Australia/New Zealand, Europe/UK or other), remoteness (Major city or outside major city), socioeconomic index (quintile) and state of residence (New South Wales or Victoria). All analyses of dietary indices (except EMMA-HI lifestyle index) and fish consumption exposures were additionally adjusted for total energy intake (kJ/day, continuous), adiposity (BMI in kg/m2 continuous) and smoking status (binary: never/ever) and pack-year history (continuous, mean-centred). The mAHEI, hPDI and fish consumption analyses further additionally adjusted for alcohol consumption (g/day, continuous). Multiple imputation was used to handle missing data for all covariates included in analysis models. We excluded participants who had a sex-specific total energy intake in the extreme highest percentile (> 99) or lowest percentile (< 1), were missing responses to > 50 % of FFQ items or who had an extreme < 0·5 percentile or > 99·5 percentile of sex-specific BMI (5 years pre-questionnaire measure). *p(int) = P-value of Wald test for sex–interaction term.
Figure 3. Long description
A horizontal dot plot compares the odds ratio of various dietary exposures and multiple myeloma risk, stratified by sex. The x-axis represents the odds ratio ranging from 0.8 to 1.6, while the y-axis lists different dietary indices and fish consumption categories for women and men. Each dot represents the odds ratio for women and men, with horizontal lines indicating the 95% confidence intervals. The plot includes data for mAHEI, hDPI, EDIP, EDIH, EDIR, EMMA-HI, all fish, and oily fish. Notable patterns include higher odds ratios for EDIP, EDIH, and EDIR in women compared to men. The p-values for interaction terms are provided next to each dietary exposure, indicating the significance of sex differences. All values are approximated.
Although CI were generally wider in these sensitivity analyses, higher scores on the ‘unhealthy’ indices EDIP, EDIH and EDIR, all maintained similar magnitude and direction of association with increased MM risk, while both the hPDI and mAHEI maintained a mostly similar inverse direction of association (in the conditional case-sibling control and unconditional case-spouse control analyses) and magnitude of estimated effect, although there was an attenuation of effect size for hPDI score in the conditional case-sibling sensitivity analysis.
We found no substantial evidence of sex interaction for any of the exposures assessed in the primary analysis. Nor did we find clear evidence of association between family history of MM and any exposure of interest (online Supplementary Table S2).
Discussion
From this large case–control study, we found that a healthier diet was associated with reduced MM risk, and higher scores on dietary indices associated with biomarkers for inflammation, insulinaemia and insulin resistance were associated with increased risk of MM. There were no substantial associations observed for fish consumption or a composite lifestyle index score.
One strength of this study is the large sample of incident cases. Due to registry-based recruitment, practically all cases of MM diagnosed during the study period were able to be identified. Another strength is its family-based design, compared with other methods of control sampling, the family design may provide generally higher control participation rates, less confounding by unmeasured genetic and early-life environmental factors and reduced volunteer bias(Reference Milne, John and Knight31). The use of causal diagram analysis to guide covariate selection is another strength which allowed appropriate adjustment for known confounders(Reference Greenland, Pearl and Robins32).
There were also some inherent limitations of this study. The generalisability of our findings may be limited by the predominantly white European background of participants in the study. Due to the case–control design, with measurement of exposures at baseline depending on participant recall, there remains the possibility of measurement error and information bias due to differential recall. This potential bias should be mitigated by the use of family controls, and the similar data collection process, including FFQ, used for cases and controls.
Despite controlling for known confounders, residual confounding bias remains a possibility due to unknown confounders or incomplete accounting for confounders. Although the absence of physical activity measurement in this study is a potential limitation, there is little evidence of any substantial association between physical activity and MM risk, and thus, this is unlikely to introduce any substantial confounding(Reference Marinac, Birmann and Lee33). Although mechanically plausible, we could not assess potential effect modification by physical activity of the association between dietary scores and MM risk. Reverse causation is also potential limitation for studies of modifiable exposures, but this should have been minimised in this study by the selection of an exposure period for dietary components beginning two years prior to recruitment, during the pre-diagnosis period for cases. There is the potential for selection bias arising from differential participation according to severity of disease. Selection bias arising from missing data is also possible, but the use of multiple imputation should mitigate this so long as assumptions hold regarding the structure of missing data(Reference Sterne, White and Carlin34).
Similar to our findings, previous studies have reported that empirical dietary indices associated with biomarkers of inflammation (EDIP), insulinaemia (EDIH) and insulin resistance (EDIR) were associated with MM risk. A prospective USA cohort analysis reported an increased risk of MM associated with a higher EDIP score (HR, per sd = 1·16, 95 % CI = 1·02, 1·32); and higher EDIH and EDIR scores were associated with increased MM risk for males, within wider CI(Reference Lee, Fung and Tabung17). Each of these indices was developed utilising a similar reduced rank regression approach in the Nurses’ Health Study, and although they have been validated in other USA cohorts, they may not necessarily translate to populations in different countries. We utilised these scores in this analysis without directly validating their association with their respective biomarkers in EMMA. Though conceivably a limitation, Australia and the USA share a highly similar Western dietary pattern, and almost all relevant items in the original indices were also available in the EMMA FFQ(Reference Azzam35).
If the associations of higher MM risk with higher scores on any of these empirical indices represent an underlying causal mechanism, it is plausible that the mechanism involves one or more of the biological pathways associated with these indices. The EDIP was developed to assess the inflammatory potential of dietary pattern and has been shown to be robustly associated with plasma inflammatory markers including C-reactive protein, IL-6, tumour necrosis factor α receptor 2 and adiponectin. C-reactive protein is a well-known inflammatory marker which has been associated with MM prognosis(Reference Bataille, Boccadoro and Klein7), IL-6 is an inflammatory cytokine and a reported growth factor for MM cells which has been reported to be associated with adverse outcomes in MM patients(Reference Ludwig, Nachbaur and Fritz10). Tumour necrosis factor α receptor 2, expressed by myeloma cells, may be activated by tumour necrosis factor α and has been reported to promote growth of MM cells and cell cycle progression(Reference Jourdan, Tarte and Legouffe12).
Diet could also influence MM risk via insulin-related pathways. The EDIH was developed to measure the hyperinsulinaemic potential of diet and was shown to be robustly associated with fasting plasma C-peptide, a marker of β-cell secretion which is secreted from the pancreas along with insulin in response to elevated blood glucose levels(Reference Heding and Rasmussen36). The EDIR was developed to assess the insulin resistance potential of diet, via the ratio of triacylglycerol to high-density lipoprotein cholesterol ratio which is a common biomarker for insulin resistance, reflecting changes in lipid metabolism associated with insulin resistance(Reference Baneu, Văcărescu and Drăgan37). Both insulin and insulin like growth factor, which is associated with insulin resistance, have been reported to be associated with MM and other tumour cells in vitro, as well as with cancer risk and prognosis(Reference Sprynski, Hose and Kassambara14,Reference Pollak38) . Many observational studies(Reference Castillo, Mull and Reagan15), including a large pooled prospective analysis(Reference Ardisson Korat, Deubler and Bertrand39), have reported an association between history of type 2 diabetes mellitus and increased risk of MM. Metformin, a glucose lowering medication, has been reported to reduce the risk of progression from monoclonal gammopathy of unknown significance to MM, potentially via pathways involving reductions in hyperglycaemia and hyperinsulinaemia(Reference Boursi, Mamtani and Yang40).
We observed a reduction in MM risk associated with higher scores on a modified version of the AHEI, and there was a suggestive association between higher scores on hPDI measuring adherence to a healthful plant-based diet and lower MM risk. Similar to our findings, a prospective analysis of the hPDI(Reference Satija, Bhupathiraju and Rimm23) reported an association with reduced MM risk for the highest compared with lowest quartile (HR = 0·83, 95 % CI = 0·71, 0·96) in a USA cohort(Reference Castro, Parikh and Eustaquio24). There is also evidence from a pooled prospective analysis in the UK that adherence to a vegetarian/vegan diet may reduce MM risk, compared with a diet including meat (RR = 0·23, 95 % CI = 0·09, 0·59)(Reference Key, Appleby and Crowe41). Although other studies have not identified a substantial association between AHEI-2010 score and MM risk(Reference Lee, Fung and Tabung17), higher pre-diagnosis AHEI-2010 score has been reported to be associated with longer survival of MM cases(Reference Lee, Fung and Tabung22). The association between higher scores on the ‘healthy’ dietary indices and reduced MM risk might also be explained by similar inflammation and glucose-related pathways. However, other notable potential mechanisms include body composition, which may be influenced by diet and higher adiposity, and has been shown to be associated with greater risk of MM diagnosis(Reference Marinac, Birmann and Lee33,Reference Lauby-Secretan, Scoccianti and Loomis42) and the microbiome.
The microbiome may be substantially influenced by variations in diet, and there is accumulating evidence that microbiome alterations are associated with cancer, particularly colorectal malignancy(Reference Pleguezuelos-Manzano, Puschhof and Rosendahl Huber43), and other chronic disease outcomes(Reference Hou, Wu and Chen44). It is plausible that dietary pattern might also influence MM risk via gut microbiome-associated immune modulation. Differences have been reported in bacterial diversity and abundance between MM cases and healthy controls from metagenomic analysis, with enrichment of nitrogen-recycling species in MM cases possibly resulting from disease associated accumulation of urea-nitrogen(Reference Jian, Zhu and Ouyang45,Reference Antoine Pepeljugoski, Morgan and Braunstein46) . Higher quality diets, or more healthful plant-based diets, as measured by mAHEI and hPDI, generally have a higher proportion of less processed, fibre-rich and plant-based foods. Such diets are associated with greater abundance of fibre-fermenting microbes, and their metabolites butyrate and other short chain fatty acids (SCFA)(Reference Louis and Flint47). Butyrate can modulate immune function and has been associated with deep and durable response to therapy in MM cases(Reference Shah, Maclachlan and Derkach48). A single-arm pilot trial of a high fibre plant-based dietary intervention in participants at high risk of MM progression, reported that two patients experienced a reduction of long-term progression trajectory, and also reported that overall, there was increased microbiome α-diversity, higher proportion of butyrate producing species, and reductions of BMI and insulin resistance, relative to baseline(Reference Shah, Cogrossi and Garces49).
We did not observe any strong evidence of association with MM risk for fish consumption, EMMA-HI score or a healthy lifestyle index based on AICR/WCRF recommendations(Reference Shams-White, Brockton and Mitrou29). The lack of association of our lifestyle score may be largely due to the fact that alcohol consumption, while increasing overall risk of cancer and thus scored appropriately in an index reflecting general lifestyle recommendations for prevention of cancer, is inversely associated with MM risk(Reference Cheah, Bassett and Bruinsma25,Reference Floud, Hermon and Simpson27,Reference Shams-White, Brockton and Mitrou29) . We specifically investigated oily fish due to the higher content of n-3 fatty acids; n-3 has been reported to induce apoptosis of MM cells in vitro(Reference Abdi, Garssen and Faber50) and genetically predicted n-3 levels have been reported to be associated with reduced MM risk, potentially via higher levels of phosphatidylcholine leading to increased cell membrane stability(Reference Li, Li and Wang51). Although there was weak evidence from the sex-stratified sensitivity analysis of an association between oily fish consumption and reduced MM risk in men, the association in women was close to null (see Figure 3). Our overall findings accord with a 2018 meta-analysis of prospective studies which reported a null association between fish consumption and reduced MM risk(Reference Sergentanis, Ioannis and Ioannis-Georgios52), although an earlier meta-analysis of case–control studies reported an inverse association(Reference Wang, Wu and Zhu5).
The results from the secondary and sensitivity analyses conducted were generally consistent with those from the primary analysis. OR for EDIP, EDIH and EDIR were consistently elevated across sensitivity analyses. Though largely consistent, associations for mAHEI and hPDI were closer to the null and slightly positive in the matched-spouse sensitivity analysis, within wide CI. This could be related to a higher degree of concordance on dietary measures between cohabiting spouses, or potentially by better control of confounding due to genetic and unmeasured early-life exposures in the sibling matched analyses. Insufficient control of confounding by sex due to sex-discordant matched pairs, or other biases potentially introduced by the conditional approach might also contribute. Given that this matched-spouse analysis included smallest sample of participants, these results could also be plausibly attributable to chance(Reference Greenland, Schwartzbaum and Finkle53).
Many of the potential mechanisms discussed in the context of an individual dietary pattern score could likely also to varying degrees explain the association of other scores. Although we have investigated each of the dietary indices as individual exposures, there is certainly overlap between the healthy dietary pattern scores, and the ‘unhealthy’ patterns also share common features and could to some extent be considered an inverse reflection of the ‘healthy’ dietary indices. For example, a dietary pattern that is associated with lower inflammatory potential might also score more highly on AHEI and hPDI, while a high-quality diet as measured by AHEI could also influence health through insulin or glucose pathways(Reference Piccand, Vollenweider and Guessous54). We note that some food groups have a similar directional influence on scores across different indices; high intake of leafy green vegetables generally scores as ‘healthy’, while high intake of red and processed meat would generally score as ‘unhealthy’. While we have largely focused on the assessment of overall dietary pattern, we cannot separate the influence of dietary patterns from that of the individual foods and nutrients comprising the overall diet, and it is possible that there are other nutrient-based mechanisms that could help to explain the observed associations(Reference Tapsell, Neale and Satija55).
Examining a novel Australian study population with an advantageous population-based family case–control design, and a large sample of cases, this study extends the evidence-base for dietary pattern and MM risk. Although further research is necessary to firmly establish causality of the observed associations, our findings align with existing healthy dietary recommendations. Adhering to a dietary pattern consisting predominantly of minimally processed plant-based foods, while affording broader benefits to non-cancer chronic disease outcomes and overall mortality, may also be associated with reduced MM risk(Reference Kim, Caulfield and Rebholz56). We look forward to publication of further results from a series (NUTRIVENTION) of recently conducted or ongoing trials(Reference Shah, Cogrossi and Garces49,Reference Castro, Sweeney and Derkach57–Reference Shah and Tan60) that might clarify the potential role of nutritional interventions in reducing risk of MM in high-risk subpopulations, altering disease progression in those with pre-malignant disease or improving quality of life and response to therapy in patients diagnosed with malignant MM.
Supplementary material
For supplementary material/s referred to in this article, please visit https://doi.org/10.1017/S0007114526107156
Acknowledgements
We acknowledge the contributions of John Hopper, Doug Joshua, Marina van Leeuwen, Gianluca Severi, Melissa C Southey and Claire Vajdic to study design and funding of the work and of Robert MacInnis to the selection of appropriate statistical analyses.
The Epidemiology of Multiple Myeloma in Australia study was supported by funding from a research grant from the National Health and Medical Research Council (NHMRC: 1029885). NHMRC had no role in the design, analysis or writing of this article. Cases were ascertained through the Victorian Cancer Registry (VCR) and the New South Wales Cancer Registry (NSWCR).
G. G., S. J. H., H. M. P., W. C. and N. W. D. contributed to the design and conception of the study; F. B. and G. G. contributed to the data acquisition; J. K. B., N. A. and R. M. contributed to the statistical analysis; S. C. performed the statistical analysis; G. G., R. M., S. J. H., H. J. and W. C. contributed to data interpretation; S. C. drafted the manuscript with A. H. All authors critically reviewed the manuscript and approved the final version.
The authors declare no conflicts of interest.
Open access
