Review Reports - Associations Between Dietary Intakes of Omega-3 Fatty Acids, Blood Levels, and Pain Interference in People with Migraine: A Path Analysis of Randomized Trial Data

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This manuscript provides a thoughtful secondary analysis of a randomized dietary intervention investigating the mechanistic role of omega-3 fatty acids (EPA and DHA) in migraine-related pain interference. The topic is clinically relevant and contributes to the literature linking diet and chronic pain modulation. The study is generally well-designed, but several methodological clarifications and editorial improvements are required prior to publication.

1] The manuscript refers to this as an “secondary analysis” but does not specify whether the path analysis was pre-specified in the original trial protocol or registered separately. Please clarify how this analysis fits within the overall trial framework and what hypotheses were defined a priori versus developed post hoc.

2] The structural equation models appear to exclude important covariates (e.g., age, sex, BMI, baseline headache frequency, medication use). While you acknowledge potential unobserved confounders, it would strengthen the paper to test the robustness of the mediation model by including or at least discussing adjustment for these factors.

3] Figures 2a and 2b (final models) need clearer labeling of latent and observed variables, including standard paths and coefficients. Add legends explaining the constructs, directionality, and significance levels.

Author Response

We thank the reviewers for their thoughtful critiques and overall support for our manuscript. Please see our responses below. Changes to the manuscript also included editing to improve readability in English.

Reviewer 1

We apologize for this oversight. The analysis was based on Aim 3 of the original protocol which called for a mediation analysis examining the path from randomized diet group to omega-3 intakes to omega-3 blood biomarkers to omega-3 metabolites to pain interference. A full description of the original plan is available in the study protocol paper. To simplify the model, based on the available sample size, we combined the two groups randomized to high omega-3 diets. A description of the original plan has been added to the introduction with clarifications in the statistical analysis section.

Introduction: “An original aim of the study was to examine pathways linking diet assignment to dietary intakes to n-3 biomarkers to pain outcomes including pain interference (Headache Impact Test) and duration as documented in our published protocol [24]. For this analysis, we chose a priori to include a second measured pain interference variable (PROMIS Pain Interference 4a) in addition to the Headache Impact. We preserved the original focus on pathways of mediation from randomized diet assignment to dietary intakes of n-3 to n-3 blood levels to pain interference. Also, a priori, we decided to focus on n-3 PUFA, combining the two active intervention diets to ensure an adequate sample size to run the structural equation models. We hypothesize that the proposed model fits the data well.”

Data analysis: “The analysis plan, including the structural model, was outlined in the protocol paper. A priori modifications to the protocol-specified model involved including a second pain interference measure, collapsing the two intervention diets groups into one, and removing the n-3 metabolites, which showed very low correlations with the other blood n-3 indicators. Because they did not meaningfully contribute to the underlying construct, the metabolites were excluded from the final model.”

[We thank the reviewers for raising this important point regarding covariates adjustment in our models. Our primary aim was to evaluate whether changes in blood EPA/DHA could account for the randomized dietary interventions effects on pain interference. Because assignment to diet was randomized, the estimated total effect of diet on the pain outcome is unbiased by baseline covariates. Thus, covariates would primarily increase precision of estimates rather than address confounding of the intervention effect.

For the mediation analysis, we intentionally specified a parsimonious model given the modest sample size (n = 182) and the number of parameters already required to model diet assignment, self-reported intake, EPA/DHA in blood, and the pain interference outcome. Adding multiple covariates would substantially increase model complexity and reduce the stability and interpretability of estimates.

We agree that by not adjusting for additional baseline covariates, our mediation estimates rely on the standard but untestable assumption of no important unmeasured confounders of the mediator–outcome relationship. We have revised the Methods and Discussion sections to make this assumption explicit.]

“Methods: Because this study was a randomized controlled trial, diet assignment can be assumed to be independent of baseline participant characteristics; therefore, the SEMs did not adjust for additional baseline covariates. However, the mediation analysis still relies on the standard—but untestable—assumption that no important unmeasured confounders exist for the relationship between blood EPA/DHA and pain interference that are themselves unaffected by diet assignment. Violations of this assumption would bias the estimated direct effect, which may therefore reflect any residual influence of such confounders”

“Discussion: Our mediation models did not adjust for potential baseline covariates that might confound the relationship between blood EPA or DHA and pain interference. As a result, the indirect effects should be interpreted under the assumption of no important unmeasured mediator–outcome confounding.”

[Thank you for pointing this out. We have added footnotes to make the interpretation of the figures easier to understand.]

“Rectangles represent observed variables, and ellipses represent latent variables. Double-headed arrows represent an expected correlation between variables. Variances, disturbances, residuals, and covariances among residuals are omitted for clarity. Values shown are standardized path coefficients. †p < .10, *p < .05, **p < .01,*** p < .001. Abbreviations: 24H = Twenty-four hour; n3 = omega-3 fatty acid; EPA = eicosapentaenoic acid; DHA = docosahexaenoic acid; WBC = white blood cell; HIT-6 = Headache Impact Test. “

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

This study analyzes data from a 16-week randomized controlled dietary trial to test whether omega-3 fatty acids (EPA and DHA) influence pain interference in migraine. It uses structural equation modeling on 182 participants to link diet assignment, self-reported intake, and blood levels of EPA/DHA to pain interference measured by PROMIS and HIT-6.

The methodology is rigorous for a secondary analysis, with appropriate modeling and good statistical fit indices (EPA: CFI = 0.99, RMSEA = 0.039; DHA: CFI = 0.981, RMSEA = 0.040). Results show that higher blood EPA at 16 weeks correlated with significantly lower pain interference (B = –0.56, p < 0.001), while DHA showed a weaker trend (B = –0.43, p = 0.057). The indirect mediation effect of diet on pain through blood EPA was significant (B = –0.23, p = 0.008), suggesting blood EPA fully mediates the dietary effect.

The work is statistically sound and clinically relevant, but its interpretive strength is limited by several issues. The authors do not control for potential unobserved confounders such as medication use, hormonal fluctuations, or psychosocial variables. The high proportion of female, educated participants restricts generalizability. The self-reported intake data are subject to recall bias, and blood fatty acid levels may not fully capture tissue incorporation relevant to migraine pathophysiology. The exploratory nature of the path model also precludes causal inference. The absence of pre-specified mediation hypotheses and lack of correction for multiple testing make the results preliminary.

Despite these limitations, the paper makes a meaningful contribution by demonstrating a mechanistic link between dietary omega-3 enrichment and improved functional pain outcomes in migraine, mediated by circulating EPA. It provides a quantitative foundation for future interventional work assessing omega-3 balance in neuro-inflammatory pain syndromes.

Language and presentation are clear and well-structured, but the discussion overstates the conclusiveness of the findings. It should be explicitly framed as an exploratory secondary analysis with hypothesis-generating value rather than confirmatory evidence. Additional sensitivity analyses adjusting for baseline migraine frequency and medication use would strengthen the inference.

Author Response

Reviewer 2

1] The authors do not control for potential unobserved confounders such as medication use, hormonal fluctuations, or psychosocial variables.

2] The high proportion of female, educated participants restricts generalizability.

[The reviewer is correct to point out that the highly educated, predominantly female population limits the generalizability of the findings and this is highlighted as a limitation in the discussion. However, it should be noted that most pain studies appear to attract more women.]

“The demographics of the study sample reflect on-going challenges in pain research to broadly reflect the general population.”

3] The self-reported intake data are subject to recall bias, and blood fatty acid levels may not fully capture tissue incorporation relevant to migraine pathophysiology.

[Diet intake was measured with two 24-hour diet recalls at each time point. This procedure uses the Nutrition Data System for Research with a standardized protocol to minimize recall bias. However, because participants who had been on the diet 10 weeks or more knew that they were expected to eat the provided diet foods, they may have overstated their adherence, leading to desirability bias.]

“In addition, the association between n-3 dietary intake and n-3 blood levels was weak, suggesting that our measurement of dietary intake may have been faulty or that unmeasured factors may have impacted this association. Although, the 24-hour food recall was considered the most reliable measure of dietary intake at the time, and the least sensitive to recall bias, it may not have captured intakes adequately. The 24-hour recall measured after 10 weeks on the diet may have resulted in social desirability bias as participants reported that they were consuming provided foods and adhering to the diet. Unmeasured factors could include genetic polymorphism of enzymes associated with n-3 metabolism or differences in the gut microbiome.”

[We agree that blood fatty acids may not fully capture fatty acid incorporation into tissue relevant to migraine pathophysiology. Unfortunately, given the nature of human clinical trials we were unable to collect other tissue without compromising the primary recruitment goal of our study. We have added language in the discussion to highlight this limitation. ]

“It should be noted that in this manuscript we report circulating concentrations of fatty acids which do not necessarily reflect incorporation of fatty acids into tissue. Though some molecules in circulation may impact migraine [39,40], it is not known whether fatty acids would have similar effects. Therefore, it is possible that the effect of EPA and DHA in the models presented here would be different if different tissues were analyzed/included.”

5] The exploratory nature of the path model also precludes causal inference. The absence of pre-specified mediation hypotheses and lack of correction for multiple testing make the results preliminary.

[The original Aim 3 was exploratory, but the path model was developed a priori from the proposed Aim 3 model. The only post hoc modification was removal of the n-3 blood metabolites as they were not making a significant contribution to the model. However, we have clarified the text to specify our hypotheses. The prespecified model included a hypothesis that change in pain interference from diet assignment was mediated by both n-3 dietary intake and n-3 blood levels. The results suggest instead that the change in pain interference was mediated by n-3 blood levels alone. We recognize that this result could be associated with some bias and revised the text accordingly. ]

“However, although we hypothesized that the effect of diet assignment on pain interference would be mediated by both dietary n-3 intakes and n-3 blood levels, we did not find that dietary intakes were an important mediator nor did we find a direct effect of diet assignment. Instead, our findings indicate that the data are consistent with an indirect pathway from diet to pain through EPA or DHA in blood. Our mediation models did not adjust for potential baseline covariates that might confound the relationship between blood EPA or DHA and pain interference. As a result, the indirect effects should be interpreted under the assumption of no important unmeasured mediator–outcome confounding.”

6] Language and presentation are clear and well-structured, but the discussion overstates the conclusiveness of the findings. It should be explicitly framed as an exploratory secondary analysis with hypothesis-generating value rather than confirmatory evidence. Additional sensitivity analyses adjusting for baseline migraine frequency and medication use would strengthen the inference.

[We have clarified that our analysis was based on a prespecified conceptual model (please see our responses above. We have also addressed the concern that we were overstating the importance of the results in our discussion.

We hesitate to add variables to the analysis due to the modest sample size. Adding medications could require multiple variables associated with different medication classes.]

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

All reviewer concerns are sufficiently addressed, and the manuscript is clear, methodologically sound, and improved.

Author Response

All reviewer concerns are sufficiently addressed, and the manuscript is clear, methodologically sound, and improved.

Thank you for your careful review.

Reviewer 2 Report

Comments and Suggestions for Authors

Thank you for the revised manuscript. It is clear the authors have addressed many of the earlier concerns, and the paper is now much closer to publishable quality. The mediation assumptions are stated more openly, the Discussion adopts a more appropriate tone, and the limitations section is more transparent. A few remaining issues still need attention, but these are minor and focused.

The mediation pathway would benefit from a brief sensitivity check—nothing complex, just a small note in the supplement showing how vulnerable the indirect effect is to plausible mediator–outcome confounding. This would strengthen confidence in the interpretation without expanding the scope of the study. Because one variance had to be fixed to resolve a negative variance issue, it would help to include a short model-comparison note in the supplement showing that the main conclusions are unchanged. Readers will appreciate seeing that the constraint does not drive the findings.

It would also improve readability to translate one standardized effect into a concrete clinical value on HIT-6 or PROMIS, even as a simple example.

The manuscript should include a brief comment on the reliability of the dietary recall measure or, if reliability cannot be estimated, a clear acknowledgment that measurement error likely weakened the intake mediation path.

Author Response

Comment 1: The mediation pathway would benefit from a brief sensitivity check-nothing complex, just a small note in the supplement showing how vulnerable the indirect effect is to plausible mediator-outcome confounding. This would strengthen confidence in the interpretation without expanding the scope of the study. Because one variance had to be fixed to resolve a negative variance issue, it would help to include a short model-comparison note in the supplement showing that the main conclusions are unchanged. Readers will appreciate seeing that the constraint does not drive the findings.

Response 1: Based on your suggestion, we ran a sensitivity analysis controlling for age, BMI, any Botox use, and depression. Controlling for these variables, interpretations were largely unchanged. The coefficients associated with the indirect effects were a bit larger and the p values were smaller. Please see the additional supplemental material related to this sensitivity analysis.

In the text: A sensitivity analysis examined a model controlling for measured variables we felt would be most likely to confound the association between blood EPA/DHA and pain interference: baseline age, BMI, botulinum toxin use, and depression.

3.3.4. Sensitivity analysis results

Including the four potential confounding variables had little effect on the model variable interpetations. The indirect effect of diet assignment on reduced pain interference through EPA and DHA were slightly stronger (-0.27) with smaller p values (0.006 for EPA and 0.011 for DHA). The associations between EPA or DHA blood levels and pain interference were also stronger (Supplementary Table 2).

Comment 2: It would also improve readability to translate one standardized effect into a concrete clinical value on HIT-6 or PROMIS, even as a simple example.

Response 2: We can see how knowing the change in score could be helpful for a clinician. For example, the standardized effect (-0.23) of the diet assignment through blood EPA or DHA corresponds to roughly a 2.3-point decrease on the HIT6, and a 3.2 decrease in PROMIS pain interference t-score. We have added this language to the results:

These standardized effects (-0.23) correspond roughly to a 2.3-point decrease in the HIT-6, and a 3.2 decrease in PROMIS pain interference t-score.

Comment 3: The manuscript should include a brief comment on the reliability of the dietary recall measure or, if reliability cannot be estimated, a clear acknowledgment that measurement error likely weakened the intake mediation path.

Response 3: Yes, this was an unexpected finding. We have expanded the paragraph describing the limitations of the dietary assessment and have added several references. Although the unannounced 24-hour food recall is considered the most reliable assessment of intake, it depends on repeated assessments, ideally more than three. In our study, we were able to assess at most two per time period, thereby increasing the variability and reducing the reliability of the measure for capturing usual intakes.

The association between n-3 dietary intake and n-3 blood levels was weak, suggesting possible error in our dietary assessment or the influence of unmeasured factors. The most reliable measure of dietary intake requires multiple unannounced food recalls, ideally more than three [36,37]. Because our study included at most two recalls per participant, we may not have captured intakes adequately [25,38]. Under optimal conditions, correlations between the average of four 24-hour food recalls and plasma fatty acid levels were only about 0.4-0.6 [38]. Moreover, the recall collected after 10 weeks on the diet may have been affected by social desirability bias as participants knew that they were expected to consume provided foods and adhere to the protocol. In a previous validation study, adding measures of social desirability to regression calibration equations helped reduce bias associated with self-reported dietary intakes by accounting for participants’ tendency to over- or under-report certain foods [39]. Unmeasured factors may also explain the weak association, such as genetic polymorphisms affecting n-3 fatty acid metabolism that influences absorption [40].

Again, we appreciate the thorough reviews and would be happy to address other elements.