Atul Gawande on Priorities, Big and Small
Atul Gawande — surgeon, staff writer at The New Yorker, and author of The Checklist Manifesto and Being Mortal — joins Tyler Cowen for Ep. 26 to range across AI and diagnosis, the operating room checklist, the economics of the FDA, end-of-life priorities, and what makes a person productive.
Key ideas
- Medicine is no longer a craft practised by individuals. The volume of knowledge and skill required now exceeds what any one person can hold. The most urgent gap in medical education, Gawande argues, is teaching doctors how to function as an effective group — not how to be the smartest person in the room.
- Checklists work by forcing conversation, not by enforcing compliance. A checklist that becomes a tick-box exercise, or swells from 19 items to 81 under administrative capture, loses its entire point. The mechanism is a structured pause that turns attention on: introductions, shared goals, critical concerns, before the incision is made.
- Healthcare does improve outcomes, but only when reliability matches spending. High blood pressure kills more Americans than anything else; sixty per cent of those with the diagnosis receive incomplete or inappropriate care. The problem is not coverage — it is a failure of organised, reliable delivery. Kaiser and Minnesota both flipped cardiovascular disease off the top-killer list simply by getting adherence above 80 per cent.
- Doctors need to become counsellors, not informants. The dominant clinical mode — here are the options, here are the risks, what would you like? — leaves patients with 5 per cent of the conversation and treatment routinely misaligned with their actual priorities. Eliciting goals first, then recommending what best matches them, improves outcomes for both quantity and quality of life.
- Clarity about your own priorities is the rarest and most valuable professional asset. Asked repeatedly — about his production function, about hiring, about Ariadne Labs’ work — Gawande returns to the same answer: most people cannot say what their goal is, and most organisations are not actually optimising for patient outcomes. The ability to name the goal and spend the highest proportion of your time on it is, he suggests, ‘all it is.‘
Content
AI and the limits of diagnosis by machine
Tyler opens on artificial intelligence — specifically whether a system like IBM Watson could replace the diagnostic consultation. Gawande is sceptical, not as a reflex but for structural reasons: diagnosis is a narrative process, not a data-matching one. Patients arrive unable to locate their own symptoms. The presenting story evolves over the course of the conversation and depends on the clinician’s read of how likely this particular person is to flag a symptom as serious. Tools like Isabel — which generates a ranked differential diagnosis from a list of observations — already add value as augmentation, not replacement: the clinician does the eliciting; the software catches what the human might have anchored past.
He is similarly measured on CRISPR. The fears about designer babies strike him as misplaced — the traits that matter most are multigenic, not addressable by point mutations — but the capacity for gene drives in mosquitoes and other organisms represents a genuinely unthought-through capability. The deeper worry is cultural: society is already narrowing the neurotypical range through hiring norms, medication, and social structure, and CRISPR is one more instrument in that narrowing. George Church’s narcolepsy — and what the world would lose if a parent screened against it — is the worked example.
Checklists, sponges, and operating room culture
Gawande describes the sponge problem — surgical swabs left inside patients — as his entry into intervention design. A case-control study at his institution found a steady rate of roughly one in 1,500 to 3,000 operations over thirty years. The counting system was not broken; it was fallible in the way counting fifty-two cards is fallible — you get the wrong number at a predictable rate. The solution was a barcode system that scanned sponges in and out, eliminating the error almost entirely; the only remaining cases have been when staff ignored the machine.
The surgical checklist — modelled on aviation pre-flight procedure — cut death rates by 47 per cent across eight cities in an initial trial. It works because it requires the team to stop, introduce themselves to one another, and review the goals and key risks of the operation before the incision. What it does not work as is a compliance document. Gawande describes receiving 150 versions from hospitals that had ‘adapted’ the checklist: an 81-item version had been assembled by administrators and was universally ignored; shorter versions that had been modified by the clinical team were actually used. The failure mode is not adoption but capture — administration using the checklist to impose agendas rather than the clinical team using it to think together.
The increasing prevalence of awake patients during surgery adds another layer. Neurosurgery has long required it — the patient’s speech is real-time feedback on whether a probe is in the wrong place — but even in abdominal surgery, awake patients can flag concerns, preferences, and comfort, making the room safer when the team includes rather than works around them.
The healthcare–outcomes paradox
Cowen presents the literature on low marginal returns to healthcare — randomised trials, Amish life expectancy, doctors’ strike studies — and asks why the connection between coverage and outcomes is so weak. Gawande’s response has two parts. First, the time horizon matters: the Oregon Medicaid study lasted under eighteen months and measured a non-significant 16 per cent reduction in deaths; the threshold for significance would have required greater than 50 per cent, an impossible bar. Massachusetts data show that the value of coverage compounds — steady improvements accumulating over years as chronic disease management takes effect.
Second, coverage is a floor, not a ceiling. The real failure is within the system: high blood pressure is the leading killer, a third of adults have it, and 60 per cent are receiving care that is incomplete or inappropriate. This is not a resources question. Kaiser Permanente took that same condition from 40 per cent adequate management to above 80 per cent; cardiovascular disease fell from their number one killer to second. Minnesota achieved similar results. The gap between what good organised care produces and what typical care produces is the largest addressable opportunity in American health, and the system is not organised to close it.
Medical education and the doctor as counsellor
Gawande’s answer to ‘what is most missing from medical education’ does not concern clinical knowledge. It concerns teamwork: the profession has outgrown what any individual can manage, and training has not caught up. Knowing how to operate as an effective group, how to diagnose when the group is failing, and how to restore its function — none of this is taught or researched, and it is, he argues, the biggest lever available for improving human health.
A complementary failure is in the doctor–patient conversation. The standard model — Dr Informative, as one palliative care colleague labelled it — gives patients a menu of options and asks them to choose. The patient does 5 per cent of the talking; the doctor does 95 per cent. What Being Mortal led Gawande to recognise is that patients have goals — for quality of life, not merely quantity — that clinicians rarely surface. Eliciting those goals first and then recommending what most aligns with them produces better outcomes by both measures, and reduces the suffering that comes from treatment pursued without understanding what the patient actually wanted.
Kenneth Arrow’s observation about asymmetric information — that healthcare is the paradigm case where the seller controls both the decision and the option set — runs beneath all of this. The clinician’s pen is the most powerful tool in the room. Payment incentives shape whether it tilts toward overtreatment or undertreatment. The profession’s self-awareness about that power, Gawande suggests, has a long way to grow.
The FDA, surgical regulation, and surveillance
On drug regulation, Gawande broadly defends the FDA’s role, framing it as a trade-off about which risks a society is genuinely willing to accept — not a technical question with a correct answer. The HIV community’s engagement with the FDA in the 1980s and 1990s is his model: a public willing to accept faster approval in exchange for greater risk, working with regulators to build that trade-off into the process rather than litigating it from outside.
He is more critical of the post-approval surveillance infrastructure. Accelerating approval while gutting the monitoring that would catch harms after market launch — which he identifies as the trajectory under political pressure from drug makers — is the worst possible combination.
For surgical procedures, the FDA’s current almost complete non-regulation is a mismatch: devices and drugs are highly regulated but the operation itself is not, leaving outcome variance of 250 per cent across institutions for the same procedure with no transparency. His preferred intervention is institutional accountability — organisations required to track and report outcomes — rather than procedure-level FDA oversight.
Ariadne Labs and the Gawande production function
Ariadne Labs, co-founded with the Harvard Chan School of Public Health and Brigham and Women’s Hospital, is Gawande’s institutional attempt to apply scientific methods to healthcare delivery rather than to biological science. Its three focus areas — surgery, childbirth, and end-of-life care — were chosen because they represent the highest-risk, highest-cost, most failure-prone moments in a life. About half of its twenty-odd active projects are abroad.
The South Carolina safe surgery programme illustrates the method: five years of work in a state without a mandate (red-state politics) and without pay-for-performance incentives, achieving 40 per cent adoption of the checklist and a 22 per cent reduction in deaths. The conclusion — that getting the remaining 60 per cent would require a mandate or financial incentive — is itself a research finding.
Asked about his production function, Gawande’s answer is relentlessly about priority. Most people in healthcare organisations cannot state their actual goal; observed behaviour reveals priorities far removed from ‘the best possible care at the lowest possible cost.’ His own advantage, as he names it, is spending a very high fraction of his time on his stated priority. The same principle appears in his hiring method at Ariadne: before anyone interviews, specify what success looks like in two years — five concrete accomplishments — and use reference calls to verify that the candidate has a record of actually delivering on stated intentions.
Related
- Atul Gawande — guest
- Tyler Cowen — host