The White House is eyeing a proposal that would try to identify early warning signs of violent behavior in people with mental illness, the Washington Post reports.

• Supporters see it as a way for President Trump to address mass shootings without making changes to gun laws.

Why it matters: As we've written time and again, most mentally ill people are not violent, and the majority of mass shooters have not been diagnosed with a mental illness.

• And while the stated intention of the proposal is a good one, there are reasons to question whether it'll work.

The big picture: This proposal is part of a larger push to create a new agency called the Health Advanced Research Projects Agency, or HARPA. It would be modeled on DARPA, the research arm of the Pentagon.

• The administration first discussed creating HARPA in 2017. It's gotten new momentum following recent mass shootings, per the Post.
• Trump reportedly likes the idea of HARPA, but Congress would have to act to create a new agency.

Details: The mental-illness proposal would use artificial intelligence to try to identify changes in someone's mental state that could make that person more likely to become violent.

Yes, but: AI excels at identifying hidden patterns in huge amounts of data, and has been used before to help identify mental conditions — like finding the imperceptible vocal qualities that might suggest depression, Axios' Kaveh Waddell writes.

• But scientists warn that people often have different tells, and that applying the same test across large populations can give rise to costly mistakes.

There's a lot of skepticism about whether making prices more transparent will do anything to lower them, but there ways to maximize the odds, Gilbert Benavidez and Austin Frakt write in JAMA Forum.

Driving the news: Both the Trump administration and Congress are pushing pricing transparency measures, banking on the idea that if consumers know how much health care services cost, they'll opt for the cheaper ones.

• But this hasn't worked in the past. People who have access to price information rarely use it to shop around.
• One study found that only 2% of people with access to price transparency tools actually used them.
• This is even true for people with high deductibles.

The other side: Other incentives, when paired with transparency, actually do get people to shop for cheaper care.

• The most effective seems to be reference pricing, in which payers set a maximum amount they'll reimburse for a shoppable service. Patients who use services that cost more than this maximum amount have to pay the difference out of pocket.
• A slightly less effective tool is rewards programs, in which patients receive rebates when they use services below a set amount.
• Between these 2 approaches, the stick works better than the carrot.

The bottom line: Getting the incentives right is important, but so far transparency is being pushed as a standalone approach.

Go deeper: Washington's favorite health policy isn't a silver bullet

Buying a new car every year would be a very impractical expense. It would also be cheaper than a year's worth of health care for a family, the Kaiser Family Foundation's Drew Altman writes in today's column.

Why it matters: The cost-shifting and complexity of health insurance can hide its high cost, which crowds out families' other needs and depresses workers’ wages.

By the numbers: Health care for a family covered by a large employer cost, on average, $22,885 last year.

• That's $2,000 more than the sticker price for a brand-new Volkswagen Beetle.
• If the iconic Beetle isn't your style, $22,885 would also be more than enough to get you a Ford Focus ($17,950), a Toyota Corolla ($18,600) or a Hyundai Sonata ($22,050).

Between the lines: Roughly $15,000 of that $22,885 comes from employers' contribution to their workers’ premiums.

• Workers chip in an average of $4,706 per year premiums, and then spend an additional of $3,020 out of pocket. Combined, that's almost 4 times more than the average family spends on gas in a year.

As health care costs continue to rise, they'll be comparable to even fancier cars. They're already inching up toward the cheapest Cadillac — a familiar car metaphor.

• The Affordable Care Act's "Cadillac tax" was intended to put downward pressure on prices by taxing the most generous health plans. But it actually affects a broad range of plans, and Congress has delayed the tax until 2022. The House has voted to repeal it altogether.

HHS yesterday proposed making it easier for doctors to coordinate the care of people with substance or opioid use disorder.

• Strict confidentiality rules regarding addiction treatment have unintentionally created barriers to addressing the opioid epidemic.
• For example, a doctor treating a patient for a separate medical problem may prescribe that patient opioids without knowing they've struggled with opioid use disorder.

What they're saying: "This was all well-meaning," HHS Secretary Alex Azar said of the confidentiality rules, per AP.

• "The idea was people won’t seek treatment if they feared that information would be available to law enforcement. (But) a highly restrictive regime on the control of that information has served as a barrier to safe, coordinated care for that same patient."

The health care system loves to hate "fee-for-service," in which hospitals and doctors get paid for every test, procedure and service they provide.

Reality check: Fee-for-service still isn't remotely close to going away, Axios' Bob Herman writes.

The latest case in point: Advocate Aurora Health, the large hospital system based in Illinois and Wisconsin, released financial documents yesterday showing its "capitation" revenue dropped by $25 million in its latest quarter. 

• Capitation is when insurers pay a monthly per-person payment to providers, instead of paying for each individual service, and providers consequently are responsible for all of those patients' care.
• Why the large drop in "value-based" care? Because one of its capitation contracts was converted "to a fee-for-service arrangement," the system said.