Congress blasts telehealth companies for data leaks

A bipartisan group of senators blasted several prominent direct-to-consumer telehealth companies for leaking users' sensitive data with companies like Google and Facebook, citing a recent investigation by STAT and The Markup that revealed dozens of companies had sent information as detailed as whether a user had added medication to their online carts.

"This data is extremely personal, and it can be used to target advertisements for services that may be unnecessary or potentially harmful physically, psychologically, or emotionally," Sens. Amy Klobuchar (D-MN), Susan Collins (R-ME), Maria Cantwell (D-WA) and Cynthia Lummis (R-WY) wrote in letters to telehealth companies Monument, Workit Health, and Cerebral this month, my colleague John Wilkerson reports. The lawmakers demanded more information on those companies' data sharing policies. Read more here.

The FTC's GoodRx action is only the beginning

The Congressional letter follows a historic action by the Federal Trade Commission, which last week took action against health data sharing by prescription drug coupon site GoodRx. But it'll be far from the regulatory agency's last target, experts say.

Three years had passed since the allegations against GoodRx first came to light following a Consumer Reports investigation revealing that the company — which also runs a telehealth marketplace — was sharing sensitive user data with Google and Facebook. And in the time it took for FTC to act, the market for online health services has exploded, forcing companies to compete fiercely for users' attention and leading to the kind of rampant data leaks detailed in STAT and the Markup's investigation.

"Combined with things like the Dobbs decision and the overwhelming focus on the sensitivity of health data, especially when it's not protected by HIPAA, I'd expect to see more of these investigations," Ben Rossen, formerly a senior attorney within FTC's privacy and identity protection division, said. Read my full story.

Artificial intelligence

Mm-hm, uh-huh: Why AI scribes can't understand us

Nuance's DAX AI tool is popular in hospitals and clinics because it promises to save burnt-out doctors time on documentation. But can AI fully understand a provider–patient visit? According to a recent study, automatic speech recognition tools have problems with small sounds — "mm-hm," "uh-uh," etc. — that can have a big impact in clinical settings. If the answer to "Are you allergic to aspirin?" doesn't get transcribed or is misunderstood, the downstream effects are…not good, my colleague Brittany Trang writes.

In the study, researchers ran high-quality re-recordings of patient–provider visits through Google and Amazon's clinical speech transcription models. (The authors were unable to get access to Nuance's DAX engine.) The rates of substituted and deleted non-lexical conversational sounds like "uh-huh" and "mm" were alarmingly high. Combined with other factors — such as non-native English speakers, high background noise, and too many speakers — the authors caution that clinical models will need more tuning before they're ready to be trusted.

Mental health

Mindstrong's mental health care plans fall apart

Mindstrong is the latest mental health tech startup buckling under economic shocks across big tech. The Menlo Park-based company, once a top name in Silicon Valley, announced it will permanently close its patient services department and cut 128 jobs across its leadership and clinician teams.

The layoffs include Mindstrong's CEO, chief financial officer, chief technology officer, and other senior roles, along with dozens of therapists and nurse practitioners. The cuts will start taking effect in late March, the company's Head of People AJ Ruiz said in a memo to the California Employment Development Department.

Mindstrong had reinvented itself multiple times in an effort to keep its head above water: first pitched as a high-tech biomarker startup using AI and sensors to track mental health symptoms, Mindstrong more recently has provided app-based mental health care.

Health records

Oracle's plans to tackle clinical trials, records search

It's been almost a year since cloud giant Oracle's blockbuster $28 billion merger with EHR company Cerner, and the company appears to be advancing toward the sweeping vision for seamless data flow and rich analytics laid out by chief technology officer Larry Ellison over the summer. Though he didn't elaborate directly on Ellison's concept for a national health record data base, chairman David Feinberg — formerly Cerner's chief executive — said combining the data repositories sets Oracle up to generate real world evidence for pharma companies' clinical trials, and to simplify clinical tasks like finding patients' most relevant health data.

Joined with Oracle's medical claims information, Cerner's health record data could lay the groundwork for automated claims adjudication, Feinberg said at a recent event in Los Angeles co-hosted by the Harvard T.H. Chan School of Public Health and the Ellison Institute. Oracle is also looking into making HR suggestions directly in the health record.

"When the nurse is giving chemo for the very first time we should know that, and [suggest] just in time training," Feinberg said, noting later that "none of this will work if we don't build our systems that are actually user friendly."

Standardizing health data for easier sharing, or interoperability, is only a step toward that goal, he said. Today's health information exchanges might let clinicians call up PDFs on individual patients, but finding the right information is still a major challenge, he said. "We're saying to clinicians, here's all the information from an interoperability standpoint, but it's not what they're looking for," he said.

Cerner is also launching a new product compiling patients' information in a more searchable system while eliminating duplicate records, he said.

Study of the week

How far AI imaging has to go

AI in medical imaging has exploded — and a new review of more than 100 studies over the past few years takes stock of what the field has accomplished so far, Ambar writes. Here are the highlights:

The promise: Some of the most alluring potential benefits, so far, are AI's ability to sift through troves of images to identify patterns or new insights and its potential to more accurately and objectively flag signs of disease. The authors also noted another frequently touted benefit: cutting down the time spent by staff reviewing images, freeing up clinicians to focus more directly on patient care.
The challenges: For all that promise, doubts abound in the research. Can AI account for low-quality pictures, insufficient data, or poorly annotated images? And as STAT has reported, there are huge issues with the oversight of algorithms, including a lack of standards on how and when to audit their performance. Other commonly raised concerns in research: trust, integration with workflows, and exacerbated biases.
The future: Ideally, AI models can be integrated not just with health records systems, but can also harness data streams from sensors, mobile devices, and wearables. "The future is expected to bring machine learning models that will be able to analyze different types of images and other clinical data at once, in ways that are more transparent and understandable," the authors wrote.

More around STAT

Check out more exclusive coverage with a STAT+ subscription

Read premium in-depth biotech, pharma, policy, and life science coverage and analysis with all of our STAT+ articles.

What we're reading

Communities are looking to Reddit and wastewater analyses to curb overdose deaths, The Washington Post
Apple Watches are making false emergency 911 calls, The New York Times
UnitedHealthcare tried to deny coverage to a patient who fought back, ProPublica

Thanks for reading! More on Thursday — Mohana

Mohana Ravindranath is a Bay Area correspondent covering health tech at STAT and has made it her mission to separate out hype from reality in health care.