Skip to main content
Premium Trial:

Request an Annual Quote

Google's Schmidt Tells HIMSS That AI Will Enable Precision Medicine, Won't Replace Doctors

LAS VEGAS (GenomeWeb) – During his keynote address here this week at the annual Healthcare Information and Management Systems Society conference, former Google CEO Eric Schmidt announced a new application programming interface for Google Cloud, a launch that will likely impact genome informatics and diagnostic technology in the near future.

While Google has had a cloud API for genomics for several years, the new, broader healthcare API seeks to facilitate interoperability of clinical, administrative, research, and genomic data to support analytics and machine learning on the Google Cloud platform.

In a blog post by Gregory Moore, vice president of healthcare for Google Cloud, that Google published shortly before Schmidt took the stage in Las Vegas, the company highlighted several genomic informatics services built with the new API.

Variant Transforms is an open-source app created with the API to feed genomic variant information into Google's BigQuery. Google said that consumer genomics firm Color has adopted Variant Transforms to expand its cancer diagnostics.

Spanish open-source software developer Kanteron Systems added telegenomics on the Google Cloud platform to support precision medicine, while WuXi NextCode is getting ready to make its genomics data management system and related apps available to all Google Cloud users, according to the blog post. WuXi officially announced its partnership with Google this week.

Under the partnership, WuXi will host its core technology suite on the Google Cloud Launcher marketplace. This technology platform includes GORdb, WuXi NextCode secondary analysis, the Sequence Miner case-control research application, and the Clinical Sequence Analyzer clinical interpretation system.

The partners also will integrate Google genomics and research tools with WuXi's platform, beginning with Google's DeepVariant secondary analysis pipeline, and alongside other open-source analysis pipelines and tools available through Google Cloud. They said they would release the suite on Google Cloud Launcher in May at the Bio-IT World Conference in Boston.

At HIMSS, Schmidt took advantage of the speaking slot at the largest health IT gathering of the year — the conference drew a record 44,000 people — to announce the launch of the API, and talked extensively about how predictive analytics should, would, and must change healthcare for the better.

While Schmidt stepped down as executive chairman of Google parent company Alphabet in January, he remains a technical advisor to the internet behemoth.

"It's one thing to be able to classify. It's another to be able to predict the next step in an outcome, but what I want is prediction," Schmidt said. "Predictive analytics will change healthcare," he added, by facilitating earlier interventions. 

"A prediction means the clinician will offer you a prescription ahead of when they normally would have," Schmidt said.

"We have physicians inside of our company who believe that if [our] algorithms for prediction work, we can predict outcomes in the ER, for example, 18 to 24 hours earlier than any other observation system because [of] the deep analysis that now can be done against the historic data," he explained. "We'll see if that's true, but I think they're right," Schmidt said.

Quoting Shakespeare’s "Hamlet," he said, "We defy augury," meaning humans cannot predict their own fates. But "[m]achines can. That's the point here."

Schmidt asserted that healthcare is morphing into an information science.

He estimated that he has about 5 gigabytes of his own healthcare data in various places, most of it from various kinds of imaging. "Can you imagine when we have a combination of sensor data plus continuous behavioral data, which we're going to get from your smartphone and the various smartwatches that are coming, plus all the molecular data and so forth? This data explosion is profound," Schmidt said.

If healthcare figures out how to manage and analyze this data properly, there could be a benefit in terms of reduced mortality "probably comparable to a new drug, and far, far cheaper," according to Schmidt.

Schmidt last spoke at HIMSS in 2008. At that event in Orlando, Florida, the then-CEO gushed to a packed, buzzing room about the much-hyped Google Health personal health records platform — released a few months later — that failed spectacularly after never finding much of a market. No PHR before or since has caught on unless it has been tightly tied to a specific electronic health record.

Ten years ago, before a $37 billion federal EHR incentive program called "meaningful use," the majority of health records were kept on paper. That has changed, but so much information remains siloed in proprietary databases.

"We don't get any value for that," except for the vendor that built the app, Schmidt said. "That's precisely not how the internet was built."

He suggested that healthcare needs a new type of data store, like Google Cloud, but that alone is not sufficient. It will require a common digital health data store, normalized data, longitudinal trend analysis, machine learning, and the ability to predict next steps, supported by voice.

"We're much closer than you think we are to the vision that I outlined," Schmidt said. "This is fundamentally a search problem, and Google is very good at search problems."

Schmidt apologized for sounding somewhat "obnoxious," but added, "This stuff's happening too slowly because you've not moved to the kind of platforms that scaled my side of the industry. You've not moved to complete interconnection. You've not moved to data architectures that are open. You've not moved to the cloud. And I'm ignoring machine learning, which is relatively new."

No matter how smart computers get, though, Schmidt was adamant that machines are not going to replace medical professionals anytime soon because not only do AI systems make mistakes, the tech industry has not been able to explain some of those errors.

"For things that are life-critical, I want a human in the loop," Schmidt said. "I want [clinicians] to use computers as data sources, along with their good judgment and colleagues, and so forth, and I want them to be in charge. In my vision, the doctor and the nurse — the clinicians — are more important," Schmidt said.

The Scan

Suicidal Ideation-Linked Loci Identified Using Million Veteran Program Data

Researchers in PLOS Genetics identify risk variants within and across ancestry groups with a genome-wide association study involving veterans with or without a history of suicidal ideation.

Algorithm Teases Out Genetic Ancestry in Individuals at Biobank Scale

Researchers develop an algorithm known as Rye to tease apart ancestry fractions in admixed individuals at a biobank-scale, applying it to 488,221 UK Biobank participants in Nucleic Acids Research.

Multi-Ancestry Analysis Highlights Comparable Common Variants at Complex Trait-Linked Loci

Researchers in Nature Genetics examine common variants implicated in more than three dozen conditions, estimating genetic effect similarities across ancestry tracts in admixed individuals.

Sick Newborns Selected for WGS With Automated Pipeline

Researchers successfully prioritized infants with potential Mendelian conditions for whole-genome sequencing or rapid whole-genome sequencing, as they report in Genome Medicine.