Optimal Pair Matching Combined with Machine Learning Predicts a Significant Reduction in Myocardial Infarction Risk in African Americans following Omega-3 Fatty Acid Supplementation

Basic Information

• Paper ID: 2510.11756
• Title: Optimal Pair Matching Combined with Machine Learning Predicts a Significant Reduction in Myocardial Infarction Risk in African Americans following Omega-3 Fatty Acid Supplementation
• Authors: Shudong Sun, Aki Hara, Laurel Johnstone, Brian Hallmark, Joseph C. Watkins, Cynthia A. Thomson, Susan M. Schembre, Susan Sergeant, Jason Umans, Guang Yao, Hao Helen Zhang, Floyd H. Chilton
• Classification: q-bio.QM (Quantitative Methods in Biology), stat.AP (Applied Statistics)
• Institutions: Multi-institutional collaboration including University of Arizona, Georgetown University, Wake Forest School of Medicine, and others
• Paper Link: https://arxiv.org/abs/2510.11756

Abstract

This study addresses contradictory clinical trial results regarding omega-3 highly unsaturated fatty acids (n-3 HUFA) by leveraging a large sample of African Americans (AfAm) from the VITAL trial to explore the impact of racial differences on n-3 HUFA supplementation response. The research employed propensity score matching methodology to match 3,766 AfAm and 3,766 non-Hispanic White (NHW) participants from the VITAL trial, addressing limitations in controlling for confounding variables between groups. Through weighted decision tree analysis and LASSO logistic regression, the study found that n-3 HUFA supplementation significantly reduced myocardial infarction (MI) risk in AfAm (OR 0.17, 95% CI 0.048, 0.60), but showed no such effect in NHW.

Research Background and Motivation

Problem Context

1. Contradictory Clinical Trial Results: The cardiovascular protective effects of omega-3 fatty acids remain controversial. The GISSI-P trial demonstrated a 14% risk reduction, but subsequent ASCEND and VITAL trials found no significant cardiovascular benefits, whereas the REDUCE-IT trial showed a 25% reduction in composite CVD incidence and mortality.
2. Neglected Racial Differences: Existing RCTs generally lack racial diversity, implying uniform omega-3 supplementation effects across all racial groups, an assumption lacking scientific evidence.
3. Genetic Variation: African Americans exhibit higher frequencies of genetic variants in the fatty acid desaturase (FADS) gene cluster, which are associated with more efficient arachidonic acid (ARA) synthesis capacity, potentially affecting omega-3 supplementation efficacy.

Research Motivation

• Utilize the large AfAm sample (5,107 participants) from the VITAL trial to explore race-specific omega-3 supplementation effects
• Control for confounding factors through advanced statistical methods, simulating randomized controlled trial conditions
• Provide scientific evidence for precision nutrition and individualized cardiovascular disease prevention

Core Contributions

1. Methodological Innovation: First combination of propensity score matching with machine learning methods to analyze race-specific effects of omega-3 supplementation
2. Important Findings: Discovery of significant myocardial infarction protective effects of omega-3 supplementation in African Americans (83% risk reduction), but no effect in White populations
3. Statistical Rigor: Employing multiple advanced statistical methods including LASSO variable selection, bootstrap standard error estimation, and weighted decision trees
4. Clinical Significance: Provides important evidence for future RCT design of omega-3 supplementation targeted at specific racial groups

Methodology Details

Task Definition

The research task is to assess the impact of omega-3 HUFA supplementation on myocardial infarction, stroke, and cardiovascular disease mortality across different racial groups. Input comprises baseline characteristics and follow-up data from VITAL trial participants, with output consisting of risk assessments and predictive models for each racial group.

Data Preprocessing Pipeline

1. Initial Dataset: 25,871 participants
2. Racial Screening: Exclusion of other races (2,719 individuals) and incomplete records (3,833 individuals)
3. Final Dataset: 19,319 participants (15,553 NHW + 3,766 AfAm)

Propensity Score Matching

Three matching methods were compared:

• Optimal Pair Matching: Optimization of overall criterion (minimizing sum of absolute pairwise distances)
• Nearest Neighbor Matching: Greedy algorithm for sequential pairing
• Genetic Matching: Genetic algorithm optimizing scaling factors for Mahalanobis distance

Assessment through standardized mean difference (SMD) demonstrated optimal pair matching performed best, ultimately yielding 3,766 matched pairs.

Machine Learning Analysis Methods

1. Weighted Decision Tree (CART)

• Classification and Regression Tree algorithm employed for predictive modeling
• Weighted algorithm applied to address data imbalance (rare MI events)
• Weight ratio = number of MI patients / number of non-MI patients
• Gini index used to determine split points, minimum node size of 20
• 5-fold cross-validation for model optimization, 1-SE rule to prevent overfitting

2. LASSO Logistic Regression

• Incorporation of interaction terms between n-3 HUFA supplementation and all variables
• glmnet package implementation for LASSO variable selection
• 5-fold cross-validation for LASSO model tuning
• Non-parametric and parametric bootstrap methods for standard error estimation

Survival Analysis

• Kaplan-Meier survival curve analysis
• Log-rank test to evaluate statistical significance of treatment effects
• Stratified subgroup analysis by race

Experimental Setup

Dataset Characteristics

• VITAL Trial: Double-blind, placebo-controlled 2×2 factorial design RCT
• Intervention Protocol: Daily 1g omega-3 (EPA 460mg + DHA 380mg) or placebo
• Follow-up Duration: Median follow-up 5.3 years (3.8-6.1 years)
• Matched Sample: 7,532 participants (3,766 AfAm, 3,766 NHW)

Baseline Characteristic Balance

Post-matching balance achieved on key variables between groups:

• Age: AfAm 62.4±6.6 years vs NHW 63.1±5.9 years
• Smoking Rate: AfAm 14.0% vs NHW 12.0%
• Diabetes Prevalence: AfAm 23.1% vs NHW 19.8%
• Fish Consumption (≥1.5 times/week): AfAm 50.2% vs NHW 49.4%

Evaluation Metrics

• Primary Endpoint: Myocardial infarction incidence
• Secondary Endpoints: Stroke incidence, cardiovascular disease mortality
• Statistical Measures: Odds ratio (OR), 95% confidence interval, p-value

Experimental Results

Primary Results

Myocardial Infarction Risk

Kaplan-Meier Survival Analysis:

• AfAm Group: Omega-3 supplementation significantly reduced MI risk (p=0.00034)
• NHW Group: No significant difference (p=0.59)

LASSO Logistic Regression Results:

• AfAm: OR = 0.17 (95% CI: 0.048-0.60), 83% risk reduction
• NHW: OR = 1.00 (95% CI: 0.74-1.34), no protective effect

Decision Tree Analysis

AfAm Population: Omega-3 supplementation status was the most important factor for MI prediction, followed by BMI, age, and diabetes status NHW Population: Omega-3 supplementation was not an important predictor; primary predictors were BMI and sex

Secondary Endpoint Results

• Stroke: No significant protective effects of omega-3 supplementation found in either racial group
• CVD Mortality: No significant differences observed in either racial group
• Age was an important predictor for stroke and CVD mortality

Statistical Validation

• Non-parametric and parametric bootstrap methods yielded consistent results
• 5-fold cross-validation ensured model robustness
• 1-SE rule applied to prevent decision tree overfitting

Related Work

History of Omega-3 Fatty Acid Research

1. GISSI-P (1999): Landmark study demonstrating 14% relative risk reduction
2. Subsequent Studies: Mixed results from OMEGA, ASCEND, and other trials
3. REDUCE-IT (2019): High-dose EPA showed 25% risk reduction
4. Meta-analyses: 2018 review demonstrated 8% reduction in cardiac death risk

Racial Difference Research

• Genetic Basis: Significant frequency differences of FADS gene variants across races
• Metabolic Differences: AfAm exhibit more efficient ARA synthesis capacity
• Clinical Evidence: Prior studies suggest race-specific effects but lack rigorous controls

Methodological Contributions

• First application of propensity score matching to omega-3 racial difference research
• Novel application of machine learning methods in nutritional epidemiology

Conclusions and Discussion

Main Conclusions

1. Race-Specific Effects: Omega-3 supplementation demonstrates significant myocardial infarction protective effects in African Americans but not in White populations
2. Substantial Effect: 83% reduction in MI risk in African Americans (OR 0.17)
3. High Specificity: Protective effects limited to myocardial infarction, with no impact on stroke or CVD mortality

Biological Mechanism Hypotheses

FADS Gene Polymorphism Theory:

• Approximately 80% of AfAm carry "derived" alleles for efficient HUFA metabolism, compared to only 43% of NHW
• High ARA/EPA ratio leads to pro-inflammatory, pro-thrombotic state
• Omega-3 supplementation corrects this imbalance, more effective in AfAm

Lipid Mediator Balance Hypothesis:

• ARA-derived pro-inflammatory lipid mediators vs EPA/DHA-derived anti-inflammatory lipid mediators
• More severe baseline imbalance in AfAm, more pronounced supplementation benefits

Limitations

1. Post-hoc Analysis Nature: VITAL trial not specifically designed for racial comparisons
2. Causal Inference Constraints: Propensity score matching cannot fully replace randomization
3. Sample Size Limitations: Reduced sample size post-matching may affect statistical power
4. Unmeasured Confounding: Potential important confounding factors not included in the model

Future Research Directions

1. Dedicated RCT: Design randomized controlled trials of omega-3 supplementation specifically for AfAm
2. Mechanistic Studies: In-depth exploration of FADS genotype and supplementation response relationships
3. Dose Optimization: Determine optimal omega-3 supplementation dosage for AfAm
4. Personalized Medicine: Genotype-based precision supplementation strategies

In-Depth Evaluation

Strengths

1. Methodological Rigor: Multiple statistical methods cross-validate findings, enhancing result credibility
2. Significant Clinical Implications: First strong evidence of omega-3 racial differences
3. Adequate Sample Size: Utilizes large RCT data with substantial statistical power
4. Strong Innovation: Novel combination of propensity score matching with machine learning

Limitations

1. Post-hoc Analysis Constraints: Cannot completely eliminate selection bias
2. Insufficient Mechanism Explanation: Lacks biomarker data supporting mechanistic hypotheses
3. Limited Generalizability: Applicability of results to other populations requires verification
4. Clinical Translation: Transition from statistical findings to clinical application requires additional evidence

Impact Assessment

1. Academic Value: Provides important theoretical foundation for precision nutrition
2. Clinical Guidance: May alter clinical practice guidelines for omega-3 supplementation
3. Policy Significance: Provides novel methodological framework for health disparities research
4. Social Impact: Helps address cardiovascular disease health disparities in AfAm populations

Applicable Scenarios

1. Clinical Practice: Develop individualized omega-3 supplementation strategies for AfAm patients
2. Drug Development: Guide race-specific nutritional supplement development
3. Public Health: Formulate targeted cardiovascular disease prevention policies
4. Research Design: Provide reference for racial stratification and sample size calculation in future RCTs

References

1. GISSI-Prevenzione Investigators. Dietary supplementation with n-3 polyunsaturated fatty acids and vitamin E after myocardial infarction. Lancet 1999, 354, 447-455.
2. Manson, J.E.; et al. Marine n−3 fatty acids and prevention of cardiovascular disease and cancer. N. Engl. J. Med. 2019, 380, 23-32.
3. Mathias, R.A.; et al. The impact of FADS genetic variants on ω-6 polyunsaturated fatty acid metabolism in African Americans. BMC Genet. 2011, 12, 1-0.
4. Chilton, F.H.; et al. Precision nutrition and omega-3 polyunsaturated fatty acids: a case for personalized supplementation approaches. Nutrients 2017, 9, 1165.

Summary: This is a research study with significant clinical importance and methodological innovation, providing strong evidence for racial differences in omega-3 fatty acid supplementation effects. Despite limitations inherent to post-hoc analysis, its rigorous statistical methods and substantial clinical findings provide important scientific evidence for precision nutrition and individualized medical care.