Video: Google I/O: 5 lessons
Google is betting that its deep learning systems can sort through the electronic health record morass.
At Google I/O, CEO Sundar Pichai outlined how the company was using its artificial intelligence and machine learning infrastructure to better predict healthcare outcomes. The field is emerging on multiple fronts, but much of healthcare data is unstructured and requires a lot of wrangling.
For Google, the interest in healthcare is more of a way to prove its models and algorithms in the field. There's also a natural extension for Google Cloud Platform. Google has partnered with Fitbit on data and health APIs, too. Given that backdrop, Google AI is keenly interested in health.
In a paper published during Google I/O, the company outlined how it is approaching electronic health records. The company noted that it worked with Stanford, UC San Francisco and The University of Chicago to explore how deep learning models can apply to hospital patients. The key development is that Google is looking to use data as it is. The data prep alone can sink many analysis efforts.
Google's paper outlined the data prep challenge:
We hypothesized that these techniques would translate well to healthcare; specifically, deep learning approaches could incorporate the entire EHR, including free-text notes, to produce predictions for a wide range of clinical problems and outcomes that outperform state-of-the-art traditional predictive models. Our central insight was that rather than explicitly harmonizing EHR data, mapping it into a highly curated set of structured predictors variables and then feeding those variables into a statistical model, we could instead learn to simultaneously harmonize inputs and predict medical events through direct feature learning.
According to Google, its models were able to scale and be accurate in predicting unexpected readmissions, discharges, and inpatient mortality. As ZDNet has previously reported, AI is being leveraged to better improve care as well as reduce costs.
Google noted its efforts, which were statistically important, were just a start. The predictions and deep learning models were using retrospective data only.
Indeed, there are multiple efforts to use AI and machine learning in healthcare.
- Qure.ai launches AI system to read head CT scans and find abnormalities
- Why Fitbit could make a good Google acquisition
- Conway's Law and the future of healthcare data management: Genome, Blockchain, and GDPR
- Healthcare's $3 trillion question: Should the likes of Google and Facebook control this data?
- Google takes health care to the cloud with new API, partnerships
- Microsoft rolls out more AI-infused healthcare services, software
- Death and data science: How machine learning can improve end-of-life care
- Apple can win electronic medical record game with Health Records in iOS 11.3: Here's 7 reasons why
- CVS Health and Aetna bet $69 billion merger on analytics, data, digital transformation
- Roche acquires Flatiron Health in $1.9 billion deal
- Amazon, Berkshire Hathaway and JPMorgan Chase & Co. health venture likely boon for AWS
- Project Baseline: Alphabet's five-year plan to map the entire journey of human health
- DeepMind and the NHS: What it's really like to use Google's kidney health app
- Apple's healthcare plans under the microscope: From iPhone apps to Apple Watch and what comes next
- Move over HealthKit: Why Apple's ResearchKit is proving the real hit with doctors
- Apple CareKit: Building the future of healthcare, one iOS app at a time
- Apple Health and Apple HealthKit: What you need to know