From The Laws of Medicine, a short book by Siddhartha Mukherjee:

My medical education had taught me plenty of facts, but little about the spaces that live between facts. I could write a thesis on the physiology of vision. But I had no way to look through the fabric of confabulation spun by a man with severe lung disease who was prescribed “home oxygen,” but gave a false address out of embarrassment because he had no “home.” (The next morning, I got an irate phone call from the company that had attempted delivery of three canisters—to a Boston storefront that sold auto parts.)

I had never expected medicine to be such a lawless, uncertain world. I wondered if the compulsive naming of parts, diseases, and chemical reactions-frenulum, otitis, glycolysis—was a mechanism invented by doctors to defend themselves against a largely unknowable sphere of knowledge. The profusion of facts obscured a deeper and more significant problem: the reconciliation between knowledge (certain, fixed, perfect, concrete) and clinical wisdom (uncertain, fluid, imperfect, abstract).

Medicine is a soft science involving the application of limited knowledge amidst uncertainty. Historically, it was wholly unscientific. In the modern era, we apply the philosophy of Karl Popper:

In his 1934 book, The Logic of Scientific Discovery, the philosopher Karl Popper proposed a crucial criterion for distinguishing a scientific system from an unscientific one. The fundamental feature of a scientific system, Popper argued, is not that its propositions are verifiable, but that its propositions are falsifiable-i.e., every theory carries an inherent possibility of proving it false. A theory or proposition can only be judged “sci-entific” if it carries within it a prediction or observation that will prove it false. Theories that fail to generate such “falsifiable” conjectures are not scientific. If medicine is to become a bona fide science, then we will have to take up every opportunity to falsify its models, so that they can be replaced by new ones.

Mukherjee posits three “laws”:

LAW ONE
A strong intuition is much more powerful than a weak test.

LAW TWO
“Normals” teach us rules; “outliers” teach us laws.

LAW THREE
For every perfect medical experiment, there is a perfect human bias.

These amount to: 1) The conditional probability of Bayesian reasoning means that bad tests done indiscriminately yield more noise than information. 2) Outliers shouldn’t be ignored, but should be studied. The rare exceptions often drive our discoveries. 3) Cognitive biases are an inescapable part of the human condition.

Regarding medicine as presaged by Lewish Thomas, who in 1983 published the most popular book about the dawn of modern medicine in The Youngest Science:

Thomas would not have predicted this stickiness of uncertainties and constraints; the future of medicine that Thomas had imagined was quite different. “The mechanization of scientific medicine is here to stay,” he wrote optimistically in The Youngest Science. Thomas presaged a time when all-knowing, high-precision instruments would measure and map all the functions of the human body, leaving little uncertainty and even fewer constraints or gaps in knowledge. “The new medicine works,” he wrote. “The physician has the same obligations that he carried, overworked and often despairingly, fifty years ago-but now with any number of technological maneuvers to be undertaken quickly and with precision…. The hospitalized patient feels, for a time, like a working part of an immense, automated apparatus.

He is admitted and discharged by batteries of computers, sometimes without even learning the doctors’ names. Many patients go home speedily, in good health, cured of their diseases…. If I were a medical student or an intern, just getting ready to begin, I would be more worried about this aspect of my profession. I would be apprehensive that my real job, taking care of sick people, might soon be taken away, leaving me with the quite different occupation of looking after machines.”

In reality, things have panned out quite differently: despite the increasing accuracy of tests, studies, and equipment, the doctors of today have to contend with priors, outliers, and biases with even deeper and more thoughtful engagement than doctors of the past. This is not a paradox.

Ironically, Thomas’s passage from 1983 feels more prescient about current fears than Mukherjee’s from 2015.

Some career advice from the archives of an otherwise defunct blog from venture capital billaionaire Marc Andresseen:

In my opinion, it’s now critically important to get into the real world and really challenge yourself—expose yourself to risk—put yourself in situations where you will succeed or fail by your own decisions and actions, and where that success or failure will be highly visible.

By failure I don’t mean getting a B or even a C, but rather: having your boss yell at you in front of your peers for screwing up a project, launching a product and seeing it tank, being unable to meet a ship date, missing a critical piece of information in a financial report, or getting fired.

Why? If you’re going to be a high achiever, you’re going to be in lots of situations where you’re going to be quickly making decisions in the presence of incomplete or incorrect information, under intense time pressure, and often under intense political pressure. You’re going to screw up—frequently—and the screwups will have serious consequences, and you’ll feel incredibly stupid every time. It can’t faze you—you have to be able to just get right back up and keep on going.

That may be the most valuable skill you can ever learn. Make sure you start learning it early.

This is one of the things about a career in medicine that is very hard to do early on without going outside the school-industrial complex. In the olden days, medical students did more stuff. In many cases, contemporary students are essentially not permitted to have ownership of anything.

Cultivating the skill to deal with responsibility, decision-making admist uncertainty, and dealing with suboptimal interactions & results is probably the biggest differentiator between a happy and unhappy approach to practicing medicine.

As is the case with so many things in medicine, our selection process—for medical school, residency, and nearly every stage until a person has the requisite skills to practice—is riddled with suboptimal, if not entirely meaningless or counterproductive, proxy measures.

When we select college students for medical school, or medical students for certain residencies, the presumed goal is to select people who will become good doctors. On the first pass, we want to find those who will make good doctors generally, and on the second, those who will excel in specific fields. But if we’re honest, we’re terrible at this. We are absolutely abysmal at identifying the right people for the job.

Magical Metrics

The metrics required for field competitiveness are a reflection of supply/demand (i.e. the relative scarcity of training spots) more than the cognitive firepower, personality/disposition, or physical skills required for competency.

Our testing systems are, at best, inefficient measures of general intelligence and hard work. While those traits are important, the process we use to assess them requires students to spend countless hours learning material of dubious utility, often at the expense of other potentially valuable endeavors.

Economics teaches us that when resources are limited, the question isn’t whether something has value—most things do— but whether it’s the best use of those resources. There is no empirical evidence that our current testing paradigm or medical school curriculum broadly is the best use of applicants’ time, or that it effectively selects for the traits we actually care about.

On top of flawed testing—and accelerated by pass/fail Step 1—we have an obsession with CV-padding. Whether due to laziness, limited resources, or a lack of better alternatives, we’ve created a system that rewards applicants for amassing measurable activities that give the illusion of merit, even when those activities are empty gestures.

This leads to students wasting their free time on meaningless research, instead of pursuing hobbies, passions, or other fulfilling endeavors. To this, we’ve added a layer of well-intentioned but often questionable community service and outreach activities—nice in theory, but of dubious impact for both the person doing them and the community receiving them. None of these activities are inherently bad. Research isn’t bad. Helping people certainly isn’t bad. But when these tasks are reduced to tokenized, measurable units, we have to ask ourselves: are they being done in service of actual good, and is this the best way to accomplish that good? Is it possible for the outsider observer to distinguish the real deal from the slop?

Time & Money

Another valid criticism is that these demands favor students of means. Those with time, money, and connections are better positioned to engage in these resume-building activities, take gap years, or pursue additional schooling to improve their applications. This creates barriers for students without these resources. As if four years of undergraduate education weren’t enough preparation for a medical career, we’ve created a system where applicants need even more time, money, and effort just to qualify for medical school.

The irony is that other countries, and even parts of our own, have occasionally managed to streamline this process. Medicine faces a “good problem”: more people want to enter the field than there are spots. But we also face a deeper issue: many who do get those spots end up unhappy in practice or with no intention to practice in the first place. Some of this is due to the inherent difficulties of working in our broken healthcare system, but part of it likely reflects poor selection among applicants. Our laziness comes at a cost: we rob students of the full breadth of life when we force them into a narrow mold and make them compete in proxy contests trying to accumulate  “experience.”

What makes a good doctor?

What actually makes a good doctor? Hard work, kindness, and resilience are likely more important than test scores or resume padding. Yet, as Peter Drucker famously said, “What’s measured gets managed.” When we measure test scores and activities, we optimize students to achieve those outputs. But those outputs come at a cost. And too often, that cost includes losing good, happy, and fulfilled students, trainees, and practicing physicians.

Laws of unintended consequences combined with a crappy system, this can’t be the right way for people to spend their time collecting brownie points:

When I see the graphs showing the medical student research arms race, it seems to me that the numbers between unmatched and matched students are pretty similar in most fields.

Is anyone actually fooled by the low-quality fluff? Are students just doing this to themselves? pic.twitter.com/ei6hfSpJns

— Ben White, MD (@benwhitemd) August 25, 2025

(chart via NRMP’s Charting Outcomes)

(I apparently shared a similar but different chart a couple of years ago as well.)

The prevailing belief is that in the era of pass/fail Step 1, students need to compete on research to stand out. I think that is probably not quite the reality, as we still have a measurable Step 2. I think we’re really seeing is not the need to stand out because of P/F Step 1 but rather the combination of relatively increased available time and greater uncertainty:

Time & Pressure:

With the pressure of Step 1 removed for strong students (who are in no danger of failing), pass/fail Step 1 has enabled many students to spend more time generally polishing their applications. This has been compounded by pass/fail curricula more broadly. Learning enough to pass simply doesn’t take the same amount of time as aiming for a perfect score.

Research is typically felt to be “more important” than other extracurriculars and it’s easy to quantify, but people are also certainly also checking boxes for volunteer opportunities and clubs. Everyone seems to have been 1 of 4 co-presidents of their local Magical Interest Group.

Schools went pass/fail for a variety of good reasons, but nature abhors a vacuum. It’s been filled with measurable trash.

Uncertainty:

We already had our longstanding competition due to the scarcity of “desirable” residency spots, but other unintended consequence of all these pass/fail components is that it delays knowing how competitive you really are for your desired field.

It used to be that you received a disappointing score on Step 1, and—before clerkships even started—you adjusted your dreams of dermatology.

Now that you can’t know if your Step 2 score will be competitive until you’ve already essentially entered application season. It makes intuitive sense to do everything else in your power to polish your potential turd if you want to maximize your chances for your desired specialty + location combination.

Step 2 is the new Step 1; it’s just harder to plan a career around.

So what?

Reasonableness at the n=1 level aside, I think this is a problem.

The research slop is largely meaningless. The work itself is mostly garbage, and people are wasting time, money, and resources filling the dregs of pay-to-publish journals. We’re also incentivizing volume over quality so that students are incentivized to pretend that random surveys and opinion pieces are research instead of spending real time doing real work that could have a meaningful impact on other people or actually develop valuable skills. Most work is read by no one except AI bots, and the last thing we need are the LLMs internalizing a bunch more fake research and observational BS.

Time is zero-sum. The question can therefore never just be: is there value? Despite the mockery and dismissal above, of course there is some value. The question has to be: is this the best use of limited resources to achieve the goals of graduating good doctors?

Building a true meritocracy with holistic application selection is an incredible challenge. Matching people to a limited prospective jobs based on both their desires and their aptitudes is truly hard, and the desperation to shine is just as reasonable here as high school students buffing their resumes for college admissions.

Easy Mitigation Steps

We can’t change the overall game, but we can adjust the rules to nudge the behaviors to our desired outcomes.

The NRMP needs to at least start reporting the median and not the mean. Even better, we should split application success into quartiles. Students currently see this data and are mislead, because long tail outliers are dragging the mean up. Many who are “below average” for their field aren’t actually below the median. Break these things down by quartile and maybe then we’ll see how “required” and impactful research really is in most fields.

We should probably also limit the ERAS length and separate posters/abstracts/presentations from publications. We need to limit the double-counting that distorts the averages and change the incentives to promote diving deep to do meaningful work.

The Great Filter

After posting that chart and out of curiosity, I did a 100% unscientific poll on Twitter with 71 responses:

Residency program directors and staff, do you filter medical student residency applications by research experiences?

— Ben White, MD (@benwhitemd) August 26, 2025

So, I can’t pretend that the students are wrong to play the game. It just means even more that, as a field, we need to adjust our systems and incentives to drive our actual desired behavior and improve our actual observable outcomes.

Of course, how widespread this type of filtering really is and the actual impact for different specialities would make for a great research project.

Deskilling and automation bias will be big problems with useful AI, but what do you call it when someone never has to develop skills in the first place?

Unskilling?

(Apparently, a at least one new paper describes it as “never-skilling”)

Residency faculty, do you have an AI usage policy for your trainees? Why or why not?

— Ben White, MD (@benwhitemd) August 23, 2025

My son’s middle school has a detailed AI usage policy. It’s hard to believe medical training doesn’t require some thought on how to ensure robust, resilient skill acquisition.

In the 1960s, economist William Baumol attempted to explain why services like healthcare and education keep getting more expensive: they’re labor-intensive, and there’s a ceiling on how much productivity can improve without sacrificing quality.

This idea—known as Baumol’s cost disease—goes like this:

• In sectors like manufacturing or tech, productivity is routinely increasing. You can automate, outsource, and scale (often all three).
• In labor-intensive fields like medicine or education, that’s a lot harder. You can’t operate on two people at once or scale up human empathy. A physician visit in 2025 takes the same amount of time as one in 1995 (okay, a maybe half as long because healthcare is terrible now).

The “disease,” according to Baumol? Wages still rise across the board, even in those low-productivity-growth fields.

Why?

Because the cardiologist isn’t just competing against other doctors—she’s also competing against the broader economy. If productivity increases in other sectors boost wages, medicine has to keep up just to retain talent.

This means that costs rise even if productivity doesn’t.

So why hasn’t physician pay risen?

If you follow Baumol’s logic, we should expect physician pay to have risen steadily just to track broader wage inflation. But that’s not what Medicare has done.

• The Medicare Physician Fee Schedule uses a conversion factor to translate RVUs into actual payment.
• That conversion factor has fallen from $36.78 in 1998 to $32.74 in 2024—and that’s before inflation.
• Adjusted for inflation, that’s a real pay cut of over 40% per RVU.

The government that sets the rules is also the dominant customer setting the prices.

Physicians are providing the same service (and often more of it), with higher expectations, greater documentation, and more liability—and getting paid less to do it.

This is the exact opposite of what Baumol’s model would predict.

Rising costs, falling pay

The paradox: overall healthcare cost increases are outpacing inflation even though clinician pay is falling in real terms. If Baumol’s cost disease is supposed to explain rising prices due to rising wages in stagnant-labor sectors, then how can healthcare spending keep growing when labor (e.g. doctors and nurses) is squeezed?

Because labor isn’t the main thing driving healthcare costs.

The three dominant forces:

1. Regulation & administrative bloat

Healthcare hasn’t just added labor—it’s added layers. In the U.S., we have more administrators per capita than any other country, and the fastest growth in healthcare employment has been in non-clinical roles.

We’ve created a complex system that requires armies of coders, billers, compliance officers, prior auth specialists, and case managers just to keep the machine moving. These people may be necessary to varying degrees, and some may unlock revenue through their work, but none generate revenue through patient care: they are, on the whole, a drag that adds cost to the system.

Baumol predicted rising costs due to labor intensity, not bureaucratic overgrowth. The U.S. did both.

2. Technology that adds cost more than efficiency

In theory, technology should help reduce costs by boosting productivity. But in healthcare, it often adds capabilities rather than replaces old ones.

• MRI didn’t replace the physical exam. It just got added to the diagnostic workflow. (Okay maybe that one is a bad example.)

• Robotic surgery didn’t make operations faster or cheaper. It made them more expensive—and arguably more marketable.

• EMRs don’t actually make charting more efficient, because they allow for more elaborate and demanding documentation rules.

On the whole, new certainly doesn’t mean cheaper.

3. The disconnect & moral hazard of third-party payment

Unlike most sectors, patients don’t directly pay for services. That disconnect between consumption and payment drives demand beyond what you’d see in other labor-intensive industries. As in, when you think a physician visit actually costs a $35 copay with infinite free mychart messages after, you have no idea what you’re really asking for.

Insurers buffer the cost, employers shift premiums, and the government subsidizes the system. We’ve uncoupled the market forces of supply and demand.

So of course we want the newest, shiniest things, and when insurance “pays” for it, there’s little incentive to say no. Combine that with medicolegal defensive medicine and the customer-service/patient-satisfaction mindset, and it’s only worse.

Structural forces vs. policy levers

Baumol’s cost disease explains why costs in healthcare should rise: it’s structurally labor-bound. But instead of acknowledging that, Medicare has tried to hold the line on overall costs by cutting per-unit reimbursement. This creates a massive disconnect between how much it costs to provide care and how much physicians are paid to do it. Meanwhile, we have sabotaged ourselves with processes and guardrails without really figuring out long-term when they’re actually helping and removing the ones that don’t. We are awash in open-loop errors.

This is why independent practice is vanishing. It’s why private equity has a foothold. It’s why primary care is on life support and many doctors are shifting toward concierge and other DPC models to opt out of the system. Practices get squeezed, so they look for scale and efficiency—not for better care, but for survival.

Private insurance, everyone’s favorite bogeyman, picks up some slack by paying more than the government (and profiting handsomely as a middleman processing claims), but even that’s tied to underlying government payments. So we see further consolidation, burnout, cost-shifting, and administrative creep.

Bigger doesn’t mean better, but it does mean more negotiating clout.

The deeper tension

This is the heart of the issue:

Natural economic forces push healthcare costs up.

Political mechanisms try to push them down.

Even well-intentioned regulations are converted to pure bloat or subverted by administrative capture, resulting in painful and expensive inefficiencies as compliance becomes the dominant force in healthcare.

And in between is the physician workforce, stuck trying to deliver high-quality care under increasingly unsustainable conditions.

Another year of the NRMP match results, and Family Medicine continues to be a relentless slow-moving disaster within the house of medicine. 805 unfilled postions, only 28% filled by US MDs. Just 1,501 US MDs in the whole country matched to one of the most critical jobs in all of healthcare (1739 applied, but the discrepency is probably a reflection of FM being a back up option for several hundred people).

I think people see this and point out several obvious deficiencies:

• Pay
• Prestige/respect
• Midlevels

All true. All essentially impossible to easily fix within medicine and our training paradigms. Some people discuss the possiblity of special loan reimbursement, and that I suppose is obliquely helpful, but the reality is that PSLF already exists and there are already programs for working in underserved areas. Debt is a problem, but I don’t think tackling that head-on is going to solve the decline of primary care in the US.

Another suggested solution I often hear is to make family medicine sexier by allowing for different fellowships, creating more training options and allowing family docs to broaden their skills into things like dermatology.

There may be something to this, though I suspect in most cases, there probably isn’t. Even if such broadening were successful, it is probably counterproductive to the actual goals of primary care. A backdoor into dermatology is probably not going to solve a shortage of qualified practitioners. Nor do I think additional training is going to improve perceptions of prestige or respect.

The thing the ACGME can do to make things better are to change the training composition/requirements and especially length. Family Medicine should probably be a shorter, outpatient-focused course of training for general practitioners in the US.

In the era of massive midlevel expansion, it simply can’t be three years long. Anything else isn’t going to work to get people interested again.

In a world where many institutions struggle to attract aspiring family practitioners, I suspect the only solution is to fight fire with fire. I think we need more efficient training. We need to acknowledge that while more training is always good, it isn’t always necessary. And if we can’t get the job done in less time (though Canada is two years), then we need to seriously consider the efficiency of our process and the ability of our tools to assess competence.

We have, for too long, resorted to a proxy metric of time to tell us that somebody is skilled. This crude tool shouldn’t be the best we can hope for going forward. Nor will it help us address a possible post-AI world where physician retraining may become a more pressing concern.

So, I think the answer is just to start by shaving off a year and getting it done in two years.

(In a fantasy world, training duration would be as long as it needs to be. Strict training lengths are important to hospitals using residents for predictable labor, not because every doctor needs the exact same amount of training time to reach competency. The ebb and flow of patients in a resident clinic is probably slightly easier to accommodate than hospital service coverage.)

Given the current reality that many people in family medicine do not want to practice a significant amount of inpatient medicine, potentially refocusing a portion of that to an optional third year instead of making it a core part of the residency is likely one way to offer flexibility without fundamentally changing the field. Offering different paths for those who want to work in rural areas doing procedures and those who want to do OB are great ideas, but some serious introspection to figure out what the core of a PCP/GP should be in the US is overdue. I won’t claim to know the answer, but the match results tell us some stakeholders need to figure it out.

I also want to preempt anyone who wants to argue that doctors are already poorly trained and that shortening training will worsen that problem. The answer is, of course, all things being equal, that shorter training will be worse training. Many older physicians indeed believe that younger physicians are graduating “less well-trained” than previous generations. Part of that is a manifestation of reduced training volume. Part of it may be related to the increasing complexity of medicine. And part of it may be related to cultural shifts, such as decreased studying after work, and other such factors.

But that assumption also implies that there is no fat to trim, that all months of training are essentially equally useful, and that a shorter process should look the same as the longer process, just worse. All training is useful, but some is clearly necessary. The reality is that we cannot afford to ignore training quality. We need to provide better, more effective training.  We need better measures. We need to reward hard work and variable skill so that the most competent people can graduate when they’re ready and not just when they’re older.

And ultimately, we need to rethink our fixation on time as the defining measure of competency. It’s not. It’s a crutch.

If you have ever dabbled in personal productivity or internet software, you may have heard of—or already be using—Notion.

Notion is an incredibly powerful and sometimes complicated web software that allows you to create, essentially, a digital brain/personal database for just about anything.

Many companies use Notion for literally everything, and there are certainly many powerful features (some you can pay to unlock), but the personal version of Notion is free and incredibly powerful (and is what I use).

One of its key features is that it allows you to easily create and update databases—think flexible, easy-to-use Excel spreadsheets on the fly—including all different kinds of complicated fields like file attachments and tags that can then be searched and filtered accordingly.

I have multiple workspaces in my Notion for multiple reasons, including everything I keep track of to run Independent Radiology, but one use case I was also able to get my wife (who smartly does not share my obsession with digital tools) to implement was Notion to organize her CME.

This is the very simple way I have my CME tracking set up:

A simple table that allows me to easily enter the name, hours, and tag (opiates, ethics, etc) for every CME item I complete—and then attach or upload the corresponding certificate so the proof is never lost. Adding more columns like date done or granting organization would be just a simple click. I make a new table every year.

Because Notion is on the web and is a well-funded, stable platform, this means I can add or find any element at any time whenever I need it—and retrieve it easily if I ever get audited by the Texas Medical Board, the American Board of Radiology, the hospital, or anybody else.

I have everything I need in one place to prove my compliance—and the pics to prove it.

If you’re keeping your CME in a physical folder, or in a folder on your desktop, or just letting them accumulate in your email inbox and hoping you’ll be able to find them in the event of an audit, I would consider trying Notion as an easy way to organize this—and potentially other parts of your life. I find it to be an easier way to stay organized than a thousand Google docs & sheets.

(Alternatively, the easiest thing you could do is just use OrbitCME (reviewed here back in the day, $20 discount affiliate link thingie here). It’s a webservice with a browser plugin that will give you credit for using sites like Radiopaedia, PubMed, or UptoDate, which means you can easily get all your CME hours taken care of passively just by doing what you’re already doing. You can upload your external CME events to it and keep everything in one place. Pricier but undeniably super convenient.)