Likelihood ratios for genome medicine

Alexander A. Morgan, Rong Chen, Atul J. ButteGenome 2010 · DOI · OpenAlex

Abstract

As personal genomics moves into the clinical sphere, healthcare providers face the daunting task of interpreting thousands of genetic variants. This paper proposes a framework that treats individual genetic variants as standard diagnostic tests. By applying Likelihood Ratios (LRs) from Evidence-Based Medicine (EBM), clinicians can mathematically combine a patient’s baseline risk with their genetic data to calculate a concrete "post-test probability" of disease, bridging the gap between complex research and actionable bedside medicine.

The Challenge of the "Genomic Deluge"

Direct-to-consumer genotyping and whole-genome sequencing have created a "genomic deluge"—thousands of single nucleotide polymorphisms (SNPs) linked to various diseases. For clinicians, this presents a significant challenge: the literature is often contradictory, and research typically reports "Odds Ratios," which are difficult to apply to an individual without knowing their baseline risk. Furthermore, testing healthy individuals for thousands of rare diseases risks the "incidentalome" problem, where "high-risk" markers may simply be false positives in a clinical context.

Methodology: Genomics as a Lab Panel

The proposed solution involves adopting the well-established Bayesian framework of Evidence-Based Medicine. By shifting the view of the genome from a mysterious blueprint to a massive panel of individual lab tests, the data becomes manageable for doctors trained in classical diagnostics.

Likelihood Ratios and the Fagan Nomogram

Instead of general risk, the authors define the Likelihood Ratio (LR) for a specific genotype as the probability of having that genotype if the disease is present, divided by the probability of having it if it is not. Each SNP is treated as an independent laboratory test. Because these are statistically independent, they can be "chained" by multiplying their LRs together:

$$\text{Pre-test Odds} \times LR_1 \times LR_2 \dots \times LR_n = \text{Post-test Odds}$$

To simplify this for clinical use, the authors advocate for the Fagan Nomogram, a graphical tool that allows clinicians to visually map how a genetic test result shifts a patient's probability of disease.

The Fagan Nomogram allows clinicians to visually calculate post-test probability by connecting pre-test probability and the likelihood ratio.

Quantifying Clinical Impact

To demonstrate feasibility, the authors curated a dataset of 2,092 disease-variant associations from 2,174 articles. Their analysis highlights the critical role of pre-test probability. For example, if a disease is extremely rare (e.g., 0.01% prevalence), even a "strong" genetic marker with an LR of 5 only raises the patient's risk to 0.05%, remaining clinically irrelevant.

The following table illustrates how different strengths of genetic associations and disease types affect the final risk assessment:

Comparison of pre-test and post-test probabilities based on disease type and variant strength.

Limitations and the Path Forward

While this framework provides a robust approximation for clinical use, several challenges remain. "Missing heritability" means common SNPs only explain a fraction of disease risk. Additionally, Linkage Disequilibrium (LD)—where SNPs are inherited together—requires more complex models to avoid "double-counting" risk.

The authors also highlight a significant reporting gap: only 46% of surveyed publications provided enough data to calculate an LR. They call for improved reporting standards, such as STREGA, to ensure genomic research becomes "clinically readable" and useful for evidence-based practice.

Summary updated

Abstract

Patients are beginning to present to healthcare providers with the results of high-throughput individualized genotyping, and interpreting these results in the context of the explosive growth of literature linking individual variants with disease may seem daunting. However, we suggest that results of a personal genomic analysis may be viewed as a panel of many tests for multiple diseases. By using well-established methods of evidence based medicine, these very many parallel tests may be combined using likelihood ratios to report a post-test probability of disease for use in patient assessment.
BibTeX
@inproceedings{morgan2010likelihood,
  author = {Alexander A. Morgan and Rong Chen and Atul J. Butte and Atul Butte},
  title = {{Likelihood ratios for genome medicine}},
  booktitle = {Genome},
  year = {2010},
  doi = {10.1186/gm151},
}
Figure