Determining Variant Pathogenicity: Williams v. Quest

In 2016, Amy Williams sued Quest Diagnostics/Athena alleging that Athena failed to correctly classify a variant her son carried as being linked to his illness, and that this contributed to his death.[1] The Athena genetic test revealed Williams’s son carried a variant in the SNC1A gene, a gene that was already known to be linked with seizures and epilepsy. However, at the time of the test, Athena reported that the child’s variant’s effect on the gene was unknown and could not be classified as contributing to his illness.

Follow-up by Williams years later showed that Athena had since reclassified the variant, now determining it likely to contribute to illness. Williams sued, arguing that the initial classification was incorrect in light of research available at the time of her son’s test. The case was finally decided in 2020, when District Judge Margaret Seymour granted a summary judgment in favor of Quest Diagnostics/Athena, finding no plausible allegations of medical malpractice.

The case underscores how our knowledge about the clinical significance of any of the millions of genetic variants regularly detected in genetic tests is continuing to evolve. For context, the genetic test in question was performed in 2007, only shortly after the rough draft of the human genome was completed. It remains to be seen how common cases of genomic malpractice like that of Williams v. Quest will become in the future.[2]

As genetic testing has become more common, however, best practices for conducting, reporting, and reevaluating genetic test results have been clarified. If a similar case were considered today, a judge would have a clearer starting point to evaluate the performance, validation, and reporting of a genetic test result. For example, a judge may review whether a lab used databases like ClinGen or ClinVar to evaluate the evidence of association between tested genetic variants and traits. A judge might also consider whether the lab has implemented the ACMG-AMP guidelines for assessing genetic variants associated with disease.[3]

Another thing for judges to consider is whether a lab has a defined policy and system for reevaluating genetic test results periodically and revising reports with updated data—a practice uncommon in medical care.[4] The issue of “duty to recontact” is still a murky area of policy and law, particularly whether the scope of the laboratory’s duty should extend this far.

Ensuring a useful response to new interpretations is typically less of an issue if the treating clinician sought reclassification and so was expecting the results. But it can be more complex if the lab initiates reclassification on its own, since the revised result would be unexpected by the ordering physician and the patient, either of whom may have moved or turned to other matters or may not have seen the new report. And of course the patient may have changed care providers in the interim. Deciding whether there is a legal “duty to recontact” and defining its reasonable parameters will be an enormous challenge.

Probabilistic Genotyping Systems: United States v. Gissantaner

Since the expanded use of probabilistic genotyping systems by law enforcement, several cases have considered the value of evidence derived from these systems. While the majority of admissibility hearings held to date have not raised concerns about probabilistic genotyping software, one recent review in the U.S. District Court for the Western District of Michigan provides an opportunity examine how such tools could be evaluated by the courts.[5]

Daniel Gissantaner was charged with possession of firearm by a felon, and the case rested on a small amount of DNA taken from a gun that belonged to an unrelated family member. The forensic DNA profile from the sample revealed a mixture of at least three individuals. A common probabilistic genotyping system (STRmix) was used to determine if Gissantaner could be a contributing individual to the mixture.

The STRmix analysis reported a likelihood ratio that indicated it was 49 million times more likely that the DNA mixture included DNA from Gissantaner and two unknown individuals, rather than if the mixture included DNA from three unknown and unrelated individuals. The analysis also indicated Gissantaner’s DNA could account for 7% of the DNA in the analyzed sample.

Gissantaner challenged the admission of the STRmix DNA report as evidence, and the court conducted a Daubert hearing regarding probabilistic genotyping and STRmix. At the conclusion of the hearing—in which the court heard from prosecution and defense experts as well as the court’s own appointed expert—the court determined that the STRmix DNA report did not meet the Daubert reliability standard in this case, and Gissantaner’s motion to exclude the evidence was granted.[6] The decision was later reversed by the U.S. Court of Appeals for the Sixth Circuit, which ruled that the STRmix DNA evidence was admissible.[7]

In the original Daubert hearing by the district court, the court noted that the decision was “not an indictment of probabilistic genotyping, and certainly not of STRmix software in particular,” but rather depended on the particulars of the application in the case. The Sixth Circuit, however, determined that the Daubert standard had been improperly applied to the STRmix DNA evidence.[8] The two decisions highlight the complexity of probabilistic genotyping systems and point to technical details judges may want to consider when evaluating similar evidence in future cases.

A regular theme when evaluating forensic DNA technologies has to be consideration for the quality and quantity of the DNA analyzed. In Gissantaner, the district court found that the probabilistic genotyping system had not been adequately validated by Michigan State Police’s forensics lab for the exceptionally low quantity and low quality of DNA present in the forensic sample. The proportion of the mixture attributed to Gissantaner was only 7%, far below the minimum 20% threshold outlined by a 2016 report from the President’s Council of Advisors on Science and Technology (PCAST), which examined probabilistic genotyping systems.[9]

In the reversal by the Sixth Circuit, the court determined that the Michigan State Police forensic lab had sufficiently tested mixtures of low quantity (below 7%), though they had not reported likelihood ratios or false-positive rates from these tests.[10]

When reviewing evidence from probabilistic genotyping systems, it is also valuable to consider how clear the protocols are for using the system in the testing lab. Labs should show established standards and guidance for using the system, and they should continue to update and validate the system using both internal and independent audits.

At the conclusion of the district court’s Daubert hearing, the court found that the Michigan State Police forensic lab had provided little guidance to analysts for how to deal with exceptionally low-quantity DNA samples of complex mixtures and that comparable samples had limited validation in the state’s forensic lab.[11] In fact, the lab had only been using the STRmix system for three months before the analysis was run on the genotype data from the Gissantaner case.

Other areas for consideration are the software-specific inputs specified by the operating analyst. For example, the probabilistic genotyping system results are constrained by the number of contributors to the mixture that the analyst specifies. Determining the correct number of contributors can be difficult, especially for low quality DNA samples.

In the 2016 PCAST report, the authors suggested only mixtures of three persons or fewer, where the minor contributor constituted at least 20% of the DNA mixture, could be reliably analyzed.[12] But the PCAST report also noted that the limits of reliability were likely to improve as adequate evidence from more complex mixtures was obtained and published. Subsequent studies have aimed to address these reliability limits and show improvements in analyzing complex mixtures.[13]

Probabilistic genotyping systems are also constrained by analyst-specified parameters regarding how the DNA sample was processed by the lab. In another case, New York v. Hillary, evidence from an STRmix analysis was excluded because though the genetic data were generated by the New York State Police crime lab, the STRmix analysis was performed using parameters specific to a different lab.[14]

Though probabilistic genotyping systems aim to be more rigorous than subjective interpretations by lab analysts, they are still exceptionally complex tools that continue to rely on user-generated genotype data and directed user analysis.