How We Validate Our Reports Against Peer-Reviewed Research

Every company in this space says their reports are science-backed. It's such a standard claim that it's become almost meaningless. What does it actually mean? What's the difference between a report grounded in solid evidence and one that's essentially dressed-up speculation?

We think the answer to that question deserves more than marketing language. Here's how the validation process works at GeneLens, including where the bar is set, who sets it, and what happens when the science changes.

The evidence threshold we use

Not all genetic associations are equal. A variant that appeared statistically significant in a single study of 5,000 people in one ancestry group is a very different beast from a finding that's been replicated across dozens of independent cohorts totaling hundreds of thousands of people. We draw a firm line between those categories.

For health predisposition reports, we require a finding to meet all of the following before it appears in a GeneLens report: replication in at least three independent cohorts; statistically significant association after correcting for multiple comparisons; published in a peer-reviewed journal with independent review; and meaningful effect size, meaning the association has enough magnitude to be informative rather than just technically detectable.

For pharmacogenomics, we use Clinical Pharmacogenetics Implementation Consortium (CPIC) Level A and Level B guidelines as our inclusion threshold. CPIC Level A means there is sufficient evidence that the gene-drug interaction should always be considered when prescribing. Level B means strong evidence exists but the clinical penetrance may be more nuanced. We don't include Level C or D associations in clinical-facing report content.

Who does the validation

Our Scientific Advisory Board includes four members with complementary expertise: molecular genetics, clinical genomics, population genetics, and pharmacogenomics. The SAB reviews any new report category before it launches, any significant algorithmic update to an existing category, and any finding flagged by our internal science team for potential retirement or modification based on new evidence.

The internal science team monitors major genomics publications - journals including Nature Genetics, NEJM, The American Journal of Human Genetics, and PLOS Genetics - on an ongoing basis. When a relevant paper is published, it's flagged and assessed against our existing methodology. If the new evidence materially changes the interpretation of a finding we already report, that triggers a formal review cycle.

SAB review is not a rubber stamp. Members have blocked report additions on scientific grounds, and they've pushed us to retire findings that were popular with customers but where the underlying evidence had weakened. The process has a real veto.

How updates reach customers

Genetic science moves. Associations that were considered robust five years ago sometimes get revised or overturned by larger, more diverse studies. When that happens, we have a responsibility to update reports rather than leave customers with outdated interpretations.

Major updates to existing findings result in a re-run of affected customers' reports. If your cardiovascular polygenic risk score changes because we've updated the algorithm to incorporate a newer, more validated variant set, your portal result is updated and you receive a notification explaining what changed and why. We don't silently overwrite old results - we archive the prior version so you can compare.

We also publish a methodology document for each report category on the Science page of our website. These documents include the primary studies used, the effect sizes and variant counts underlying polygenic scores, and the date of the last methodology update. If a number we cite in a report has a specific paper behind it, that paper is linked. This is a deliberate transparency choice. Customers who want to verify our methodology should be able to.

Where we decline to report

There are findings we could technically generate from genotype data that we choose not to include in customer reports. Alzheimer's risk based solely on APOE status without genetic counseling infrastructure is one example. The APOE4 finding is scientifically valid, but its clinical interpretation is complex, its implications for family members are significant, and delivering it without adequate support is something we're not willing to do in a web portal. We're working on building a counselor-assisted disclosure pathway before we add that capability.

Trait reports with effect sizes below a threshold we consider meaningful are another category we leave out, even if the association is technically statistically significant. Knowing that you have a variant associated with a 0.2% higher baseline cortisol level isn't useful information for a consumer. We don't include it just to make the report longer.

These exclusions represent judgment calls. We document our reasoning internally and revisit the decisions as the science and our infrastructure evolve. The goal is a report you can trust, not a report that impresses you with volume.