San Diego – Having already set a Guinness world record for delivering a gene-based diagnosis in 19.5 hours, the Rady Children’s Institute for Genomic Medicine now must show that such speedy results are possible for the estimated 80,000 American children in need of such workups each year.
Toward that end, a coalition of researchers led by Dr. Stephen Kingsmore, the institute’s president and chief executive, published a study in the journal Science Translational Medicine that shows similar speed for 10 patients, illustrating that advances in genetic sequencing alongside faster analysis through artificial intelligence, can deliver results in less than 24 hours.
That’s fast enough, authors noted, for a doctor to order a genetic work-up during their morning rounds and get the results the next morning.
Such speed, even at Rady, has previously been available only in one-off situations that couldn’t be replicated on a grand enough scale to serve thousands of patients per year. But, Kingsmore said, pulling together products from cutting-edge companies, including Illumina Inc., San Diego’s homegrown sequencing juggernaut, have delivered speed that scales.
“It was very much that we had built a Ferrari, but you couldn’t use it as a daily driver,” Kingsmore said. “This is the daily driver.”
From mid-2016 through September 2018, the institute ran four trials around genetic sequencing of children and infants, enrolling a total of 401 with unexplained symptoms. Most of those, Kingsmore noted, have had their genomes sequenced and analyzed at speeds that far exceed what’s available elsewhere but are still slower than necessary to deliver a one-day turnaround time.
In order to go faster and to serve more patients simultaneously, Kingsmore said, it is necessary to automate the laborious process of connecting the genetic mutations that sequencing detects with the observed symptoms and signs in each patient’s electronic medical record. The team has turned to natural language processing and machine learning — two disciplines in the broad world of AI — to accomplish that goal.
A trial with 95 children enrolled compared the results of artificial intelligence diagnosis to traditional “manual” analysis by a trained professional and found that the AI solution was as good or better than human competition at this tedious task.
After the trial verified that the AI method was accurate, 10 patients got the full-bore high-speed work-up which, according to results, was capable of making a diagnosis in a mean time of 20 hours and 10 minutes, less than half the time needed when analysis was performed manually.
It’s important to note that even without the automation steps, Rady’s workflow is faster than can be found pretty much anywhere else. Kingsmore said typical turnaround times for a whole genome sequence and diagnosis is six weeks, with some labs able to handle the work in two to four weeks.
Dr. Micheil Innes, a clinical geneticist at Alberta Children’s Hospital, said the medical community will need to see such high-speed diagnoses in a larger number of patients for the standard of care to change, but it appears that Rady’s results, if they stand up to scrutiny, could push medicine closer to quick gene-driven diagnoses.
“I don’t think we’re ready to hand all of this work off to automation yet, but if you eliminate some of the time that we currently spend validating results, then that would be significant,” Innes said. “I think that, as a clinician, my time is better spent at the bedside than staring at a computer monitor.”