1. How to identify and address potential methodological issues that may have influenced our interaction analysis findings?
2. What are the possible statistical approaches to refine and improve the accuracy of additive interaction estimates, and appropriate tests for the HLA-stratification analysis, in the GxE studies?
3. Why certain interaction metrics such as the SI achieve statistical significance but not others (RERI, AP) in gene-environment interaction studies?

The interaction between the environmental and genetic factors in multiple sclerosis (MS) draws attention in the field with a particular focus on the synergistic effects of low sun exposure or vitamin D levels and HLA-DRB1*15. Studies have demonstrated a significant positive additive interaction, as measured by the attributable proportion (AP), relative excess risk due to interaction (RERI), and synergy index (SI) between low sun/vitamin D and MS risk. These findings suggest that individuals carrying the HLA-DRB1*15 risk allele who also experienced low sun exposure or low vitamin D levels are at an elevated risk of developing MS beyond what would be expected from the independent effects of these factors.
In our case, utilising data from the Australian Multi-centre Study of Environment and Immune Function, we did not observe a statistically significant positive additive interaction metrics (AP, RERI, and SI) between these factors vs MS risk, however. Using the same approach, assessing interaction between diet parameters and HLA-DRB1*15 vs MS risk, we found a significant positive interaction as measured by the SI, but not the RERI and AP. Further, a molecular causal mediation analysis by the differential DNAm modules of the associations of both low vitamin D levels and low sun exposure vs MS risk did not show materially significant difference when this has been stratified by the HLA- status, i.e., HLA-DRB1*15 risk genotype vs non-HLA-DRB1*15 risk genotype.
Despite using robust statistical methods and a well-defined dataset, our estimates for AP, RERI, and SI did not align with previously reported findings in low sun/vitamin D vs MS risk. Additionally, the SI, but not the AP and RERI, was positive and significant in the diet and HLA-DRB1*15 interaction.

These discrepancies raise concerns about potential methodological issues or analytical approaches that could account for the observed divergence in results.

Objective
The primary objective of this consultation is to assess whether methodological issues, including study design, data handling,, and statistical modelling, may have contributed to the lack of significant additive interaction.

Expected Outcomes
By consulting a methodological expert, we aim to:
• Identify and address potential methodological issues that may have influenced our findings.
• Refine our statistical approach to improve the accuracy of additive interaction estimates.
• Enhance the reproducibility and validity of our study, ensuring that our conclusions align with or meaningfully contribute to existing literature.
• Provide recommendations for future studies on methodological best practices in assessing gene-environment interactions in MS research.
Methodological Implications
• A consultation with an expert in epidemiological methods/genetic epidemiology and statistical modelling will help ensure that our findings are robust, reproducible, and accurately interpreted.
• Ultimately, resolving these methodological concerns will enhance the reliability of interaction estimates and contribute to a more comprehensive understanding of MS risk factors, potentially informing targeted prevention strategies based on genetic and environmental risk profiles.
• Addressing these concerns is critical not only for our study but also for advancing the broader field of MS research.

1. What is the best way to disentangle the effects of fertility medicines from pregnancy?
2. What alternative methods could we use to investigate the relationship between fertility medicines and PAC, considering this is a rare outcome and large-scale data is needed? E.g. an emulated trial comparing IVF medicines to clomifene.

Pregnancy-associated cancer (PAC), usually defined as cancer diagnosed during pregnancy or within one-year postpartum, occurs in 1-2 of every 100 pregnancies. Rates have been increasing over time, partly due to the higher average age of women giving birth. Little is known about risk factors for pregnancy-associated cancer, whether the pregnancy itself influences the development of cancer, or whether the cancer would have occurred anyway without the pregnancy.

Our study looks at whether gynaecological conditions and treatments influence the risk of PAC. Hormonal medicines, including ovulation stimulating fertility medicines, have been associated with risk of some cancers, such as breast and endometrial cancer. It is unknown whether these medicines may also influence the risk of developing cancer during or shortly after pregnancy.

We used population-based data for Western Australian women from the IMPROVE study. We included women on the electoral roll from 2002-2013, with data-linkage to hospital morbidity, midwives and births, Pharmaceutical Benefits Scheme (PBS), cancer and death records. We used PBS data to identify users of fertility medicines. We investigated the association for clomifene separately to gonadotropins and other ovulation stimulants.

Investigating the causal relationships between fertility medicines, pregnancy and cancer is complex. This is because women who use fertility medicines are trying to get pregnant, therefore it may be that these women are more likely than age-matched non-users to fall pregnant. Also, due to the potential relationship between fertility medicines and some cancers, we may also see higher rates of cancer for women giving birth that is occurring coincidentally to the pregnancy.

Preliminary investigations (adjusting for maternal age, birth year, socio-economic index for area of residence and remoteness) showed that women who had PAC were 1.5 times more likely to use fertility medicines (excluding clomifene) than women who gave birth without PAC (OR=1.51, 95%CI: 0.84,2.70). Within women diagnosed with cancer, women with PAC had 11 times the odds of using fertility medicines (excluding clomifene) (OR=10.8, 95%CI: 4.8,24.4) and 2.7 times the odds of using clomifene (OR=2.70, 95%CI:1.45,5.01) compared to women who had cancer without an associated pregnancy.

Our preliminary results provide some indication that there may be relationship between fertility medicines and PAC, however further investigations are required to understand whether pregnancy is acting as a mediator (or moderator) in the relationship between fertility medicines and cancer. To establish this, we plan on using a nested case-control analysis including all women aged 20-50 diagnosed with cancer (cases) and randomly select 10 age-matched controls from women without a cancer diagnosis at the index date. We will identify women who gave birth from 12 months prior to cancer diagnosis up to 6 months after cancer diagnosis (or index date for the controls). We will explore the relationship between the variables of pregnancy and prior use of fertility medicines in a mediation analysis. We anticipate there being some complexities in this analysis with the timing of both pregnancy and fertility medicines potentially playing a role. Therefore, we are looking for advice on whether there is an alternative way of exploring this association.