Distinguishing true from false positives in genomic studies: p values

Area of Science:

• Human genetic epidemiology
• Statistical genetics
• Biostatistics

Background:

• Differentiating true positive findings from false positives is a significant challenge in human genetic epidemiology.
• Strategies like positive predictive value (PPV) and Venice criteria are used to assess the credibility of findings.
• Journals commonly employ significance thresholds (p-values) to identify true positive results.

Purpose of the Study:

• To investigate the impact of p-value thresholds on the positive predictive value (PPV).
• To establish usable statistical significance thresholds using PPV and Venice criteria.
• To provide guidance on appropriate p-value thresholds based on prior probabilities of association.

Main Methods:

• Theoretical and empirical analyses of published data from AlzGene.
• Evaluation of the effect of various p-value thresholds on PPV.
• Application of PPV and Venice criteria to define statistical significance thresholds.

Main Results:

• At a p-value threshold of 0.05, most findings can be false positives if the prior probability of association is below 0.10, even with high statistical power.
• In underpowered studies with low prior probability (1x10^-3), a p-value of 1x10^-5 yields a high PPV (>96%).
• A p-value threshold of 1x10^-5 generally indicates strong evidence of association.
• For genome-wide association studies (GWAS) with very low prior probability (1x10^-4), a threshold of 1x10^-7 is recommended.

Conclusions:

• The choice of p-value threshold significantly impacts the reliability of findings in genetic epidemiology.
• A p-value of 1x10^-5 is a robust threshold for declaring significance in most genetic association studies.
• Context-specific thresholds, considering prior probabilities, are essential for accurate interpretation of genetic findings.