The Bandwidth: Decoding Health + AI

If you followed AI news this week, one story crowded out everything else. For once it wasn't an agent breaking out of its sandbox to menace humanity. It was the Nvidia CEO's extramarital affairs (plural), how it turned into a national security question, and ultimately led to his ouster as CEO. I'll leave that one to the tabloids and the intelligence committees. On to the rest of the week.

1) Access, with an asterisk

One message came out of last week's US midterm election: voters want life to cost less, and health care is near the top of that list. KFF's read of the midterm electorate found health care costs among voters' most pressing affordability worries.

AI isn't helping that case yet. The prevailing view is that it pushes spending up. Last week a Stanford team priced AI-generated care plans at $314 a visit, against $97 for the treating physician. And even when efficiency gains do show up, it's unclear they reach the people paying the bills. A new Health Affairs study from Brown's Jason Buxbaum and Andrew Ryan found that the bigger hospitals and insurers get, the less it costs them to administer each patient. But the savings don't get passed along. What purchasers paid rose right in line with the number of hospital stays and people covered. So if AI does make care cheaper to deliver, would any of that reach prices?

That's the backdrop for this week's news from OpenEvidence. Most clinicians know it as the place to pull up the latest evidence mid-visit. Now it's adding what comes after the clinical decision. Its new tools tell clinicians which of more than 2,500 patient support programs a patient qualifies for, send copay savings straight to the patient, and draft prior authorization and appeal letters.

OpenEvidence isn't the only one moving the price check into the exam room. GoodRx is plugging its prescription prices into Google's new Gemini for healthcare, so a clinician can ask what a drug will cost and compare pharmacies by quantity and location without leaving the workflow. Drugmakers' own discount programs will pop up too when a provider asks about cost. Given what a Mount Sinai study found this week about drug ads swaying chatbots (more below), that last part bears watching.

These things help with access, and with affordability for the individual patient at the pharmacy counter. They do nothing about the prices that make a copay card necessary in the first place. They may even push total spending up: an approved prior authorization is a claim that gets paid, and more patients starting treatment means more spending. Good for patients. Perhaps not so good for the cost curve.