Roundup: In an Era of AI, Should We Train More Doctors?

Happy Halloween! It’s pure joy to see children in costumes, jamming their bags with sweets. It reminds me that we essentially have two kinds of people in the world: Those that will eat their sweets immediately in the few weeks after Halloween, and those that carefully preserve their treats for the months to come. It’s like an annual marshmallow test, except the results probably mean very little.

Now to healthcare: It has been another week of important news, as the industry continues to find ways to embrace technology. NVIDIA is working with Big Pharma on drug discovery; health plans are partnering with makers of wearables; and startups are continuing to raise capital for healthcare AI plays. Every company now in healthcare seems to be under pressure to have some kind of AI story, whether or not it makes sense for the underlying business.

We are clearly in a bubble, and yet there are clearly use-cases here. I continue to be very intrigued by opportunities to use AI to enhance image analysis, and assist providers in making diagnoses. And yet, it’s not that simple even in cases where the technology is proven to work. What matters - and I’m glad we’re increasingly talking about it - is how the human specialist or PCP interfaces with AI. We’re moving into an era where personalization is necessary. One study, for instance, found that AI can interfere with some radiologists’ ability to accurately assess an image, while that same technology is helpful to others. We need to study how these tools are being used in the field, and not just how effective they are in controlled settings.

What’s also been on my mind this week: While I do think we need AI, I’m also very concerned that we don’t have enough human physicians and other allied health professionals. With the Trump administration’s visa fee potentially worsening rural healthcare shortages, this is a topic we need to be addressing now. We don’t only need more technology, we also need more humans. One of my favorite data-sets to explore is the ratio of physicians per 1,000 people in the population globally. The U.S. is at 3.6 per 1,000, while many European countries are at 4 or 5 per 1,000. Cuba is an outlier at 9.4.

Other countries are doing it. So why doesn’t the U.S. train more doctors?  

So with that, an interview with one of my all-time favorite thinkers in healthcare.

1. The U.S. is at 3.6 physicians per 1,000 people, but our population is getting older and sicker. Should we be training more doctors?

The problem is actually worse than that. We need to look at the overall numbers, but then go a few levels deeper. What fields are these physicians going into? Don’t forget that in much of the rest of the world, there’s twice as many primary-care physicians as there are specialists. And we have the flip. Twice as many specialists. That’s because of the way we do our payment models and the fee schedule.

The second consideration is where are the doctors geographically located? They are very non-evenly distributed. The places people want to live and work are much higher in terms of that ratio you mentioned, and in places where people don’t want to live - oftentimes, the rural areas - it’s lower. So the problem is worse than it seems. And yet, we need more doctors as the population gets older and sicker.

But if you look at the young people going into medicine, they tend to be more women than men nowadays. My daughter’s medical school class is two-thirds women. That’s not uncommon. As a society, we aren’t set up to support that transition. Studies show that female physicians are less likely to work full-time, or discontinue completely. A big reason for that is structural and societal, with the lack of support systems for families in the U.S. that disproportionately impacts women. Gen Z is also increasingly saying we’re out of our minds in the older generations for working 60 -70 hour weeks. There’s a huge problem here. We have to figure something out. But the shift is very hard to pull off.

There’s resistance. The (medical) profession has decided that if we add more doctors, then prices will go down. That is what they’re trying to prevent. In South Korea, they dramatically increased residency spots, and existing doctors went ballistic to try to prevent it because they were afraid they would drop the prices. That’s what I fear could happen here.

2. Is AI a better solution than training more humans to go into the field?

I think it’s ‘both-and.’ Clearly there’s a whole set of things that doctors are being asked to do that the machine can do better. So should we use AI in medicine? Absolutely. And yet, I think the Silicon Valley community that thinks it will replace doctors don’t understand medical care. Humans want other humans to help them out and be a partner to them. Which is not to say that we ought not use AI, but that it won’t replace doctors. Look at radiology, where there’s clearly an argument for using AI for things like mammography to improve the accuracy and efficiency of breast cancer screening. Instead of seeing radiologists go away in light of this, we have seen the number of radiologists go up and increasing demand for the profession. You still need humans to integrate all of these things. We’ll use AI to improve outcomes but you’ll need a lot of doctors too.

3. I am seeing more physicians do part-time or full-time roles working for AI companies and startups doing things like labeling and reviewing. Is this making the problem worse?

More and more doctors in my network do not want to work full-time clinically, or even part-time and instead are taking on these side gigs. They didn’t exist when I was in medical school. That’s why these physician numbers per capita are misleading - it’s not taking into account that many physicians these days are not seeing patients. Instead, they’re labeling data sets, working for an investment bank or for a big consulting firm. I am empathetic. The current model of private care and medicine in general sucks and doctors have to do a lot of crap they don’t want to do. So now, they want to hedge their bets a little, and they want to be a part of creating a better future. Likewise, physicians are wired to be entrepreneurial. 30, 40 years ago, all these doctors ran their own businesses. Only 20% worked for big institutions. And now it’s flipped. Very few are independently practising, maybe 20%. And 80% work for CVS, Optum, a big system, or a big health plan. I think that’s a failed experiment. Patients don’t like it and the doctors don’t like it. That is also driving people to find alternatives.

4. There’s been a big rise in physicians moving to concierge medicine and taking on smaller patient panels - how is that impacting the ratio? I assume it’s making it worse.

We have got a very unequal system. The trend is exactly right and not surprising because the current insurance-based, middleman-based system is awful. I think people are realizing that all these middlemen sucking money out of the system aren’t in their interest - either the patient or the provider. We started one of the first direct primary care practices - and it’s incredibly appealing - you don’t worry about billing or coding. You just take care of the patients. What’s happening now is the market is very different. We used to charge $40 a month - or even pay what you can. Now my primary care doctor just quit and is moving to a $10,000 a year concierge practice with a vastly reduced panel size.

This ratio of 3.6 relies on physicians taking care of larger patient panels, roughly 1,600. Most of these concierge doctors are seeing 400 or less. Some are doing 100. I’ve seen examples where it’s 50. All the problems you started with related to shortages now get worse. We barely have enough doctors with normal panel sizes. Consider that for every doctor that is now only treating a panel of 200 patients, 1400 are essentially kicked to the curb and have nowhere to go. This problem is getting worse and worse. I can understand the personal motivation to make the leap here, but you will only be serving rich people. Very few poor people can afford concierge medicine. The poor people are worse and worse off with no good options.

So that brings me back to my earlier point, we need more doctors. In other parts of the world, medical education is thought of as a public good. It’s largely funded by the government. Doctors in the U.S. have a debt load of about $260,000 on average - and that completely skews things. So we need a new compact. We also need to ensure we aren’t providing publicly funded education for a physician who immediately graduates and works for a concierge practice.

The military might be a good example here. If you get your medical school funded, you agree to serve, and then you don’t have debt. Something like “primary care service” could potentially fix a lot of our shortage problems.

With Annalisa Merelli

Lilly partnered with NVIDIA

The news: Eli Lilly and NVIDIA announced a partnership to develop a supercomputer that would help with drug discovery and shorten development cycles to get medicines to people faster.

How it will work: Through the supercomputer, Lilly scientists will train AI models on millions of experiments, and will make a number of the models available on its Lilly TuneLab.

What Lilly said: "Lilly is shifting from using AI as a tool to embracing it as a scientific collaborator," Thomas Fuchs, Senior Vice-President and Chief AI Officer, told Reuters.

Why it matters: There’s no doubt in my mind that the U.S. is really good at R&D. Access to R&D is a completely different story, and we need to be considering ways to get these new medications to patients – and not just the cool technology that will help us create these therapeutics. If in theory, AI can bring the price down of drug development, how will that impact the pricing for drugs?

Novo Nordisk made an unsolicited offer to Metsera

The news: Novo Nordisk launched a bidding war against Pfizer over Metsera, a obesity -focused biotech firm. The Danish company offered $8.5 billion for the company, including $6 billion upfront, after Pfizer had offered to acquire it for $7.3 billion. Pfizer has four days to make a counteroffer.

What Metsera said: The company called Novo’s offer “superior,” Reuters reports. 

What Pfizer said: The drugmaker called Novo’s offer “reckless,” and said it would pursue legal avenues.

Why it matters: There’s a big battle for dominance in the booming weight loss market, and Metsera has a promising pipeline inclusive of a weight loss pill, a monthly injection, and two drugs that promote the feeling of fullness.

Ōura partnered with Cigna

The news: The biometric insight company is launching its first national health plan partnership with Cigna, following its partnership with Essence, a Medicare advantage plan.

The deal: Employees enrolled in Cigna health coverage will receive the wearable at a 45% discount, with the goal of improving commitment to wellness programs.

The evolution: “Ōura has evolved from a consumer wearable to a payer‑aligned, prevention‑first tool,” said the company in a press release.

Why it matters: Wearables are extremely hot right now as companies like Whoop and Oura seem to be sparking different populations of patients to engage. The key though is in making the health data relevant, and even integrated into the insurance-based system of care. The true unlock is if the core, summarized insights are even available to the provider and the patient via the EHR.

Claude for Life Sciences

The news: Anthropic released a biology package available to Claude’s premium subscribers, which features a set of tools that could be helpful in a lab.

What Anthropic said: “We are interested in Claude being a superhuman research assistant that can partner with you and collaborate with you throughout the entire scientific process,” who leads Anthropic’s life science work, told Brittany Trang.

Why it matters: Both OpenAI and Anthropic are chasing opportunities in the healthcare industry but with vastly different approaches. Anthropic seems to be more enterprise-focused, while OpenAI is seeing mass adoption amongst patients and providers.

Insight - Call the midwife

A new report published by Morgan Health surveyed 2,500 people who gave birth in the previous two years about midwife care. It found that:

In our view, the most important finding though is that maternity outcomes aren’t better for people who get insurance through an employer, versus those on Medicaid. People with employer-sponsored healthcare, for instance, are more likely to get unnecessary C-sections. There is a misconception in our industry that people on Medicaid necessarily and in all cases get worse care because the government doesn’t spend as much as commercial payers. As we have seen time and time again, there is no correlation between cost and quality in the American healthcare system. And even people with employers (and insurance via their employer) are struggling – many are barely above the poverty line.

Deals

But first, a quick correction: Marble Health raised $15.5 million, not $155 as mistakenly typed in the last roundup issue.

ThoroughCare raised $3 million: The Series A3 venture funding came from undisclosed investors, and put the company's pre-money valuation at $50 million.

OpenEvidence raised $200 million: The funding is a combination of Series C-1 and Series C-2 venture funding, in a deal led by GV which is the company's pre-money valuation at $5.8 billion

Thermo Fisher to buy Clario for $9.4 billion: In its third major acquisition this year, Thermo Fisher will take over the privately held data management company Clario.

Allswell raised $1.3 million: The LGBTQ+ mental health startup closed a pre-seed funding round led by Amboy Street Ventures.

Datavant and athenahealth partner: The two companies are joining forces to automate medical record requests and releases.

NVIDIA takes all: The AI computing giant announced deals with Verily, Innovaccer and Diligent Robotics to build dedicated AI infrastructure and capabilities.

Earnings season

by Meredith Nolan, our new HBS intern (welcome Meredith!)

One more thing

CB Insights released its State of Digital Health Q3 2025 report

Key insights include: