Two-thirds of newly approved cancer drugs are cleared on surrogate endpoints — tumor shrinkage or delayed progression on scans — not on whether patients live longer or feel better; the 30% tumor-shrinkage cutoff traces back to 16 oncologists measuring marbles through foam rubber in 1976.
2
The average survival benefit of 71 consecutively approved solid-tumor drugs was 2.1 months, often at a cost of $100,000–$200,000 per year of treatment, yet clinical trial populations are so selected that those marginal benefits may evaporate entirely in real-world patients.
3
Prasad's six hallmarks of sound cancer policy — independence, evidence, relevance, affordability, possibility (NIH funding), and agenda (deduplication of trials) — form a coherent framework for understanding why the system produces so many costly, marginally effective drugs.
4
Patients and physicians can reclaim agency by demanding overall survival or quality-of-life data before consent, understanding what progression-free survival actually measures, and insisting on tumor genomic testing only in the defined settings where approved targeted therapies exist.
Protocols
Concrete recipes — what, when, how much, and why
7 items
Questions to ask before consenting to a new cancer drug: OS vs. surrogate endpoint
WhatBefore accepting a new oncology treatment, ask the oncologist explicitly: (1) Does this drug improve overall survival (OS) or quality of life in a randomized trial? (2) If approved on a surrogate endpoint — PFS, response rate, tumor shrinkage — is there any OS data, even immature? (3) What was the patient population in the trial and how do I compare on age, comorbidity, and performance status?
WhenAt the treatment-decision conversation before initiating any new cancer drug, especially in the metastatic or relapsed setting.
DoseOne conversation before consent. Ask to see the original FDA approval document or trial abstract.
For whomAny cancer patient being offered a drug approved in the last 10–15 years, particularly in solid tumors in the metastatic setting.
WhyTwo-thirds of approvals rest on surrogates that do not track living longer or feeling better. Patients who consent to these drugs believing they extend life are being misinformed by a system that does not require that data.
CaveatsSome surrogate-approved drugs do subsequently demonstrate OS benefit in confirmatory trials. The question is whether that trial exists at the time of the decision.
Prasad frames the oncologist's role as empowering the patient: 'It's the doctor's role to empower the patient with what I know about the drug, what's been shown, what hasn't been shown, what the benefit might be, what the uncertainties are, what the known toxicities and risks are — and then to walk them through how they would decide if it's worth it to them.' The patient who asks for OS data is exercising exactly the autonomy Prasad wants the system to support. NCCN-recommended drugs without OS evidence are still widely covered by Medicare; the patient's informed consent is the only point in the process where personal preference about risk/benefit gets applied.
It's the doctor's role in my mind to empower the patient with what I know about the drug, what's been shown, what hasn't been shown, what the benefit might be, what are the uncertainties, what are the known toxicities and risks — and then to walk them through how they would decide: is it worth it to them to take those risks.
Tumor genomic testing: tiered decision framework for when to sequence
WhatApply a tiered approach: Tier 1 (always test) — lung cancer (EGFR, ALK, ROS1, BRAF, KRAS, MET, RET, NTRK), melanoma (BRAF), colorectal (MSI-high, RAS, BRAF), any tumor for MSI-high if high clinical suspicion of Lynch syndrome. Tier 2 (sequence to match a trial) — use broad NGS panels primarily to identify available clinical trials for your mutation profile. Tier 3 (high caution) — do not use an NGS finding to make off-label drug choices in the absence of trial data, especially when a standard-of-care regimen with known benefit is available.
WhenAt diagnosis for Tier 1 indications. Before starting next-line therapy for Tier 2. Avoid Tier 3 off-label choices as sole basis for abandoning established regimens.
DoseSingle sequencing test at tissue diagnosis; liquid biopsy for serial monitoring or when tissue is inaccessible.
For whomAll cancer patients at diagnosis for Tier 1 tumors; patients with rare tumors or young onset regardless of type; patients for whom a clinical trial might be preferable to standard therapy.
WhyGenomic instability in advanced tumors produces many mutations. Only a minority are driver events. Pursuing every seductive 'key-in-lock' mutation with an off-label targeted agent risks eroding outcomes against regimens with proven track records.
CaveatsIntra-tumor heterogeneity means a biopsy from one site may miss driver mutations in another. The same tumor sent to different sequencing companies has produced discordant mutation calls — adding clinical uncertainty.
Prasad's validation of Attia's Lynch-syndrome pancreas success: using NGS to enter a pembrolizumab trial (before tumor-agnostic approval) is the ideal use of sequencing — it matched a patient to an evidence-based trial. The contrast case is a patient with advanced disease, many prior treatment lines, and a genome that looks 'like a dinner plate dropped on the floor.' Finding a single mutation in that setting and using it to choose an off-label drug over a standard second-line regimen has led to worse outcomes in empirical analyses. Prasad's rule: if there is a standard regimen with known benefit, the bar for departing from it toward an off-label NGS-driven choice is high and requires at least basket-trial evidence.
Mechanism
Driver mutations fuel tumor proliferation; their targeted inhibition has mechanistic impact. Passenger mutations arise from genomic instability but do not drive growth — targeting them with a kinase inhibitor affects the passenger, not the engine.
Next generation sequencing tumor genomics — right off the bat there are definitely some people who definitely need to be tested for some mutations. Lung cancer — there are at least six or seven mutations that we now have FDA approved therapy for. Definitely there are different ways you could test for the same mutations but you need to know about EGFR, you need to know about ALK.
Also said
“The part that I kind of have the most friction with some of my colleagues is — after having done all these things — you do NGS on a patient and there is a mutation that is seductive. It looks like you have a drug for it. There's no trial available. And in these situations it is so seductive to believe that because we found the mutation and the target it's better to use the targeted drug than that older drug that may have a longer track record. And I think that's where people get into a bit of trouble.”— The cautionary framework — seductiveness of NGS findings versus track record of established regimens.
Applying the RCT hierarchy: insist on randomized evidence before accepting mechanistic plausibility
WhatWhen evaluating any new cancer drug, screening test, or procedure, require randomized controlled trial evidence showing patient-relevant outcomes (OS or QoL) before adopting or endorsing. Treat retrospective observational data and mechanistic arguments as hypothesis-generating, not practice-changing.
WhenAny time a treatment, screening, or diagnostic decision is being made — in the clinic, in guideline development, in coverage policy.
For whomOncologists, policy makers, patients evaluating treatment options, and clinicians in any field where mechanistic plausibility drives adoption of new interventions.
WhyPrasad's medical reversal research shows hundreds of practices adopted on observational or mechanistic grounds that were later reversed by RCTs. The hormone therapy reversal (WHI) and PVC-suppression reversal (CAST) are two canonical cases where compelling observational data plus strong mechanistic plausibility led to widespread adoption of ultimately harmful practices.
CaveatsThere is a cost to excessive skepticism — the dexamethasone example in COVID shows that waiting for a formal publication when an RCT protocol is publicly available and the prior probability is high can cause harm. Balance evidence standards against urgency.
Prasad explicitly uses the dexamethasone-COVID case to calibrate: when the RECOVERY trial protocol was publicly available, when the prior probability was high, and when a pandemic was active, acting on a press release plus protocol — 'a driver's license and a social security card' — was justified even before the journal paper. This is the boundary condition where his otherwise strict RCT-first rule should yield. The principle: it's not about waiting forever, it's about not acting on observational data plus mechanism alone when an RCT is feasible and the stakes of reversal are high.
There's such a thing as too much skepticism. The way to keep that in balance between skepticism that's so bad that it's paralyzing and blind acceptance that's so bad that you are a cheerleader for everything — the way to keep that in balance is I find going into the clinic. Because no matter how skeptical you are about drugs, you have to have those conversations with real people.
Interpreting cancer drug trial benefits: correct for patient-population mismatch
WhatWhen a trial reports a survival benefit, mentally apply a discount for population mismatch: trial patients are typically ~10 years younger, free of comorbidities (no diabetes, no renal dysfunction, no cardiovascular disease), with better performance status. The sorafenib Medicare analysis showed that real-world patients gained zero months where the trial showed 3 months — a 100% evaporation. Treat any reported median survival benefit as a ceiling, not a floor, for the average clinic patient.
WhenWhen reviewing trial results to make a treatment decision, either as a patient or clinician.
DoseApply this mental model at the treatment decision meeting; revisit whenever a new trial result is cited as justification for therapy.
For whomCancer patients considering systemic therapy, oncologists counseling patients, and policy makers setting coverage thresholds.
WhyRegistration trials systematically enroll the most robust patients to maximize tolerability and response signal. The real-world oncology patient is older, frailer, and on more concurrent medications — all of which blunt drug effect and amplify toxicity.
CaveatsNot all drugs evaporate in real-world populations — drugs with large absolute effects (imatinib in CML, pembrolizumab in MSI-high tumors) maintain benefit even in less-selected patients. The evaporation risk is highest for drugs with 1–3 month median benefit in already-selected populations.
Prasad quotes colleagues describing clinical trial patients as 'someone who could run a marathon who also happens to have cancer.' A patient receiving a dose reduction due to toxicity is receiving less drug — yet the trial was powered at full dose. The compounding effect: older, frailer patients need dose reductions more often, get less drug, and also have less physiologic reserve to translate whatever benefit exists into actual survival. The empirical evidence for this framework: the sorafenib-Medicare paper, and broader analyses showing drugs with modest RCT benefits often show no benefit in observational registry analyses of routine-practice patients.
So now imagine what happens when you give it to an older person who's frail who can't handle the full dose because they need a dose reduction because that side effect is massively more severe in that person. What's the benefit of that person? A number of empirical studies have looked at these cancer drugs and find that the benefits in the average American are maybe even absent.
No-regret moves in oncology: surgery, radiotherapy, and the narrow curative drug settings
WhatRecognize the small number of oncology scenarios where the evidence is unambiguous and the benefit transformative: (1) Surgery for localized solid tumors in fit patients. (2) Radiotherapy for locally advanced disease. (3) ABVD chemotherapy for Hodgkin lymphoma — curative in the majority. (4) BEP chemotherapy for metastatic testicular cancer — 95%+ cure rate. (5) Imatinib for CML — median survival from 3–4 years to near-normal life expectancy in Swedish registry data.
WhenThese should be offered without hesitation when indicated. Reserve the skeptical cost-benefit analysis for the vast zone of metastatic solid tumors where benefits are modest.
For whomPatients with early-stage solid tumors, Hodgkin lymphoma, testicular cancer, or CML; clinicians and patients navigating the overall oncology landscape.
WhyMost cancer policy debate focuses on the problematic marginal zone. Prasad explicitly delineates the no-brainer wins to avoid conflating warranted skepticism about marginal drugs with opposition to genuinely transformative interventions.
Prasad notes that surgery has improved even if the modality has not changed — radical mastectomy has been almost entirely replaced by breast-conserving surgery with equivalent outcomes over the past 20 years, and the Whipple procedure now carries far lower perioperative mortality than 50 years ago. Imatinib is his canonical example of genuine molecular precision: CML is 'genomically dumb,' driven by a single BCR-ABL fusion, and a drug that inhibits that fusion converts it from a disease with 3–4 year median survival to near-normal life expectancy. Swedish registry data confirms this durability. The contrast with avastin is explicit: avastin, earning $100 billion globally, cures nobody and extends median survival by 1–2 months in most indications.
You've got Hodgkin's lymphoma — well good news, we've got a four-drug combination that's curative. You've got testicular cancer and it's spread to your lung — good news, we can cure 95 plus of you. CML, this is the gleevec story — if you give a drug that inhibits that fusion event you can turn median life expectancy from three to four years to nearly normal life expectancy.
Evaluating end-of-life treatment choices: account for opportunity cost and time in clinic
WhatWhen evaluating a cancer treatment with modest expected benefit, factor in the opportunity cost: time spent in infusion suites and clinics is subtracted from whatever survival extension the drug provides. A drug that extends median survival by 2 months while requiring twice-weekly infusions for 4 weeks may consume 40% of that additional time in medical settings.
WhenBefore starting any systemic therapy with an expected benefit of less than 3–4 months, particularly in the metastatic setting near end of life.
DoseEvaluate the treatment burden (infusion schedule, monitoring visits, side-effect management) as a fraction of the expected incremental survival benefit before consenting.
For whomPatients in the metastatic or relapsed setting weighing systemic therapy, particularly when expected survival is less than 6–12 months.
WhyOncology 'try it and see' culture implicitly assumes every drug has only upside for a patient who wants to try something. Prasad argues this is false: the opportunity cost of treatment time may consume the benefit entirely, and some drugs with no OS benefit may have net-negative quality-adjusted survival.
CaveatsFor patients with young dependents, strong family reasons to maximize time, or personal values around doing everything possible, a different calculus may apply. This is a tool for shared decision-making, not a universal prescription to decline treatment.
Prasad: 'There may be an opportunity cost that instead of being somebody who in their last few months of life is going to Tahiti or going to visit a friend, I'm somebody who's tethered to the infusion suite and I got to keep coming back twice weekly for four weeks. You're taking it back from me.' This framing reframes the end-of-life treatment decision from 'any chance of benefit means try it' to a genuine risk-benefit calculation where the numerator (expected survival extension) and denominator (time consumed by treatment) are both quantified as explicitly as possible before consent.
They may actually have a net downside. They may have an opportunity cost that instead of being somebody who in their last few months of life is going to Tahiti or going to visit a friend, I'm somebody who's tethered to the infusion suite. You're taking it back from me. And so it makes people make choices about what they prioritize and how they view the end of life.
Healthcare cost rationality: apply the skin-in-the-game diagnostic
WhatWhen evaluating any high-cost medical intervention — for yourself, your patient, or as a policy matter — ask: would this person make the same choice if paying out-of-pocket? The Saudi Arabia air-conditioning analogy illustrates that rational decision-making collapses when the payer is abstracted from the cost. Use this as a diagnostic for identifying where the system is producing choices that reflect insulation from cost rather than net clinical benefit.
WhenWhen evaluating personal treatment decisions, guideline coverage recommendations, or policy proposals for cancer drug coverage.
For whomPatients evaluating whether to pursue aggressive end-of-life treatment, oncologists counseling patients, and policymakers setting drug coverage thresholds.
WhyPrasad cites Nassim Taleb's skin-in-the-game framework: when physicians and patients bear none of the financial risk, the rational response is to 'try it' for any positive signal — which is exactly what drives the overuse of marginal drugs.
CaveatsThe skin-in-the-game principle cuts in both directions: when costs are prohibitive, patients can be denied beneficial treatments they could not afford on their own. Prasad acknowledges this tension explicitly and does not offer a clean resolution.
Prasad extends the analogy: 'You pay five percent for the price of the car but you never actually get the car and you don't get to drive the car — you just get told how good the car is.' This is the consent dynamic for many cancer drugs: patients buy into the drug's claimed benefits based on a trial population they don't resemble, using other people's money, which further insulates the decision from the kind of cost-benefit scrutiny that would otherwise apply.
When we don't have skin in the game we are incapable of making rational cost decisions. You know it's like if I said to you can have any car you want but you only have to pay five percent of the cost of the car — I think you're going to make a very different decision than if I say you can have any car you want but you actually have to pay for the whole car.
What's new
Personal practice updates, fresh positions, predictions
7 items
Average cancer drug approval buys 2.1 months — and often zero months in real patients
~40 min
Fojo and colleagues analyzed 71 consecutively approved solid-tumor drugs and found an average survival benefit of 2.1 months. A Medicare dataset analysis of sorafenib (approved for liver cancer after a 3-month trial benefit) found real-world patients lived a median of 4 months whether they took sorafenib or not — the entire trial benefit evaporated.
Why this matters: This is the hardest quantitative statement in the episode: the nameplate number from an RCT is a best-case ceiling achieved in marathon-running patients, not a floor applicable to average Americans.
Background
Clinical trial populations are typically 10 years younger than the average cancer patient, free of comorbidities like diabetes and renal dysfunction — Prasad's colleagues describe them as 'someone who could run a marathon who also happens to have cancer.'
The sorafenib Medicare analysis illustrates the chasm: in the registration trial (HCC patients who could not have surgery), sorafenib yielded 11 months median survival vs. 8 months on placebo — a celebrated 3-month improvement. In Medicare recipients aged 65+, median survival with sorafenib was 4 months; survival of similar patients not on sorafenib was also 4 months. Prasad calls this 'the grand canyon of difference between real-world patients and clinical trial patients.' The implication for the practitioner: when a patient's oncologist says this drug showed 3 months benefit in a trial, the honest answer to 'what will it do for me?' is 'we don't know — you likely do not resemble the trial population.'
The average improvement was 2.1 months. So that's the average of 71 drugs. Two months can mean a lot to somebody. But two months should also come with another asterisk which is what is the cost of these medications. They now routinely run one hundred thousand to two hundred thousand dollars per year of treatment.
Also said
“In the real world they compared those people taking sorafenib who live four months to similar people who didn't take sorafenib and they also live four months. So I think that some of the benefits of these drugs do evaporate when you give them in broad populations.”— The Medicare sorafenib analysis — the clearest empirical example of trial-to-clinic benefit evaporation.
Progression-free survival is an arbitrary surrogate endpoint, not a measure of living longer
~65 min
Two-thirds of cancer drug approvals rest on PFS — the time from enrollment until the tumor grows 20% from its nadir, a new lesion appears, or the patient dies. The 20% growth threshold traces to 16 oncologists at a 1976 Mayo Clinic meeting who measured marbles through foam rubber with calipers to establish inter-rater agreement. It was an operational convenience, not a biologically meaningful threshold.
Why this matters: Knowing PFS is an operational artifact — not a validated surrogate for survival — fundamentally changes how a patient should interpret 'this drug delayed tumor progression by five months.'
Background
The FDA has increasingly accepted PFS as a primary endpoint for accelerated approval. The PFS cutoff does not track patient symptoms — a tumor 19% larger is 'stable' while 21% larger is 'progression,' yet patients feel the same.
Prasad walks through all four PFS events: death (uncommon first event in most trials), new lesion on scan, 20% growth from the smallest measured size, and — importantly — all the measurement variability in defining 'where the tumor ends.' He quotes his own analogy: 'measuring the width of a cloud between your fingers looking up at the sky.' For the POLO trial of olaparib in BRCA-mutated pancreatic cancer (approved on PFS benefit, no overall survival improvement), he argues the correct control arm should have been continuation of FOLFIRINOX — not a sugar pill — because stopping effective chemotherapy to randomize to placebo artificially inflated the PFS gap.
I spent a lot of time trying to get to the bottom of why it is this 30 percent and I found out it goes back to a 1976 paper where this Mayo Clinic doctor got a bunch of marbles and he put him on a dining table and he rolled out foam rubber and he got 16 oncologists to come to his house with calipers and measure the marbles.
Also said
“Progression-free survival is the time from when a patient enrolls on a study to one of four things. It's an arbitrary cut point. Very arbitrary. And that's why it doesn't always track with how people feel.”— Explains why PFS can be statistically positive while patients report no improvement in wellbeing or function.
Industry conflicts of interest are self-reinforcing through guideline authorship
~85 min
By law, US Medicare must cover any cancer drug recommended at NCCN guideline level 2A or higher — and the oncologists who write those guidelines frequently receive six-figure payments from the very companies whose drugs they are evaluating. Prasad describes this as a courtroom where both the prosecutor and defense are paid by the defendant.
Why this matters: The connection between personal pharmaceutical payments, guideline authorship, and mandatory Medicare coverage is the structural mechanism that makes drug-cost reform so resistant — and it identifies exactly which incentive to change.
Background
Open Payments (Sunshine Act) data shows many senior oncologists earn more from industry consulting than the median US household income. Prasad cites José Baselga's forced exit from Memorial Sloan Kettering — followed almost immediately by appointment as VP at AstraZeneca — as evidence that the existing disclosure framework does not function as a deterrent.
Prasad's proposed solution is not disclosure (he cites psychology evidence that disclosure may paradoxically increase patient trust without correcting bias) but structural incentive redesign: make guideline committee membership available only to oncologists who have divested from industry relationships, and make those seats the most prestigious career opportunities in the field. The current system inverts this — all the most visible career opportunities require industry relationships. He explicitly says he does not blame individual oncologists because 'the tiger is doing what tigers do' when the incentive system is this powerful.
The people who write the guideline that mandates Medicare must pay for this drug with no price negotiation — that person is being paid by the industry even for the exact same drug. And so those relationships I think are an order of magnitude more concerning than the resident hungry resident taking a meal.
Also said
“I want to create a different set of incentives that get people to maybe do something differently. If we had rules and policies that favored faculty members who didn't take money from the industry to be on the guidelines, suddenly career incentives look very different.”— Prasad's structural solution — change incentives, not rules — for the conflict-of-interest hallmark.
Sham-controlled trials reveal that many mechanical interventions work via placebo, not mechanism
~20 min
Arthroscopic knee debridement, vertebroplasty, and stenting for chronic stable angina (ORBITA trial) each performed no better than sham procedures — where surgeons cannulated the artery, performed a diagnostic angiogram, and made the patient wear headphones so they did not know if a stent was placed. The difference in exercise tolerance on a modified Bruce treadmill protocol: 16 seconds, non-significant.
Why this matters: The sham-controlled trial design is a methodological gold standard that the field almost never applies, which means we routinely confuse the psychological benefit of 'having something done' with the specific mechanical benefit of that procedure.
Background
In ORBITA, patients wore headphones, were told they 'may or may not' have received a stent, and exercise tolerance was the primary endpoint. Prior open-label stenting trials had shown ~1–2 minute improvement; ORBITA showed 16 seconds — within measurement noise.
The stent-for-stable-angina story is the episode's clearest through-line from the COURAGE trial (published during Prasad's residency; showed stents do not lower MI risk or mortality in chronic stable angina) to ORBITA (published later; showed even the subjective symptom benefit is largely placebo). Prasad describes why practitioners resist these findings: the patient comes back grateful, believing the stent saved their life and resolved their chest pain. Pair that gratitude with financial remuneration and Prasad calls it 'the methamphetamine of being a doctor — a highly addictive substance.' The same sham-vs-real design has been applied to vertebroplasty for osteoporotic fractures and to shoulder procedures with similar results.
When somebody comes at you and they tell you that actually it's just a placebo effect, it doesn't actually improve symptoms — it's psychologically unacceptable. How can that be? It doesn't fit with my experience and it doesn't fit with the way I've been rewarded. And I think that's in part why many of these medical practices that have evidence that goes the other way are very difficult to dispel.
Tumor genomic sequencing: a narrow set of definitive indications surrounded by a large zone of misuse
~130 min
Prasad supports NGS in lung cancer (at least 6–7 druggable mutations with FDA-approved therapies: EGFR, ALK, ROS1, BRAF, others), melanoma (BRAF), colon cancer (MSI-high), and for MSI-high across tumor types (the tumor-agnostic pembrolizumab indication). He warns against the seductive off-label use of NGS findings — a mutation that 'looks like a key' for a targeted drug in the absence of trial evidence may erode outcomes versus the older drug with a longer track record.
Why this matters: Gives patients and clinicians a clear tiered decision framework for when sequencing adds value vs. when it creates false precision and potentially harmful off-label treatment choices.
Background
Prasad describes the problem of intra-tumor heterogeneity: biopsies from different metastatic sites of the same patient have been sequenced and found to harbor different driver mutations — meaning a drug selected based on a liver biopsy might be the wrong drug for the lung met.
Attia shares a success story: a friend with Lynch syndrome-associated pancreatic cancer got NGS, Attia recognized the checkpoint-inhibitor susceptibility signal before pembrolizumab's tumor-agnostic approval, got the patient into a trial, and the patient achieved a complete response — still cancer-free a decade later. Prasad uses this as the ideal use case: NGS used to identify a trial, not to make an off-label prescribing decision. The cautionary counterpart: highly unstable genomes produce many mutations that look like targets, some are just genomic shards from widespread instability and have no functional significance. The same tumor sent to two different commercial sequencing labs has not always produced the same mutation calls.
Sometimes you actually end up eroding outcomes not enhancing outcomes. You actually make worse choices in those cases. Because the truth is that some genomic mutations are mutations that are fueling the tumor, and that if you fix those mutations you would improve the outcomes. But some mutations are the product of a genome that is undergoing massive instability and damage.
Medical reversal: practices widely adopted on observational evidence that randomized trials later debunked
~35 min
Prasad and Adam Cifu coined 'medical reversal' for practices that were not merely replaced by something better but proven to be worse than doing less. Examples: hormone replacement therapy for cardiovascular protection (HERS and WHI showed thrombotic harm); anti-arrhythmic drugs to suppress PVCs post-MI (CAST trial showed increased mortality); arthroscopic knee debridement (sham-controlled trials showed no superiority).
Why this matters: Medical reversal is the empirical basis for requiring RCTs before adopting mechanistically plausible interventions — and its frequency (Prasad's lab has catalogued hundreds of reversals) argues against deferring to expert consensus.
Background
The WHI hormone therapy result reversed a decade of prescribing backed by compelling observational data from the Harvard Nurses' Health Study, basic science showing estrogen's favorable vascular effects, and billions in Wyeth marketing spend. CAST reversed the PVC-suppression strategy after drugs that reliably suppressed PVCs on EKG were found to increase sudden cardiac death.
Prasad's key phrase: 'a lot of smart, well-intentioned people who have plausible pathophysiology, who have a compelling retrospective observational story — they can be wrong.' The pattern repeats: observational study finds association, basic science provides mechanism, widespread adoption, eventual RCT overturns practice. Prasad's lab documented that more than one-third of standard medical practices they evaluated were eventually reversed or found to be of no benefit when rigorously tested. The policy implication is that accelerated drug approvals based on surrogate endpoints replicate the pre-RCT era: we are adopting practices based on mechanistically plausible surrogates the way the PVC-suppression era adopted drugs based on a plausible surrogate (PVC counts on EKG).
A lot of smart well-intentioned people who really have plausible pathophysiology, who have a compelling retrospective observational story — they can be wrong. And has this happened elsewhere? We started investigating. Now we have lists of hundreds of items. They span everything from ways in which we screen patients, ways in which we test patients, drugs we give patients, procedures we do on patients, surgeries we do on patients.
Medicare's inability to negotiate drug prices creates a perverse ratchet
~100 min
US law prevents CMS from negotiating cancer drug prices and requires coverage of any drug recommended in NCCN guidelines — creating an incentive to get approval (not demonstrate value), influence guideline authors, and raise prices in synchronous lockstep with competitors to avoid becoming a 60 Minutes story. Avastin (bevacizumab) has earned nearly $100 billion globally while not curing a single patient in any indication.
Why this matters: Identifies the two structural levers — FDA approval bar and CMS coverage mandate — whose combination produces the observed drug pricing landscape, and explains why disclosure alone cannot fix the guidelines problem.
Prasad's proposed mechanism for correction is value-based pricing: tie reimbursement to the magnitude of the survival or quality-of-life benefit, which would create incentives for the first time to develop drugs that cure or substantially extend life rather than drugs that change scan findings. He contrasts gleevec (transformative, appropriate return) vs. avastin (non-curative, modest median benefit, ~$100B lifetime earnings). The MLR (medical loss ratio) provision of the ACA — capping insurer profit at 20% of revenue — paradoxically incentivizes insurers to want high costs because their absolute profit is a fixed percentage of a larger base, explaining why private payers have not formed a countervailing negotiating bloc.
There are different conceptions of ethics. And I think some people might put a value on caring for people at the end of life irrespective of societal implications for others. But we're not saying don't treat a person. There's still lots of treatments you'd give — you give the same treatment except leave out the avastin — and the difference is probably very slight.
Recommendations
Products, supplements, and tools mentioned in the episode
2 items
Forgive and Remember by Charles Bosk
Book
Sociologist Charles Bosk's 18-month ethnography of a surgical residency program (believed by Attia to be University of Chicago, late 1970s–early 1980s). Attia cites it as a window into the toxicity of traditional surgical training culture.
Attia uses the book to contextualize Prasad's medical school experience — both trained in environments where yelling, throwing things, and public humiliation were common. Bosk's analysis of how surgical culture shapes error-handling parallels Prasad's later analysis of how financial culture shapes guideline-writing.
There's a book that I've spoken about before in the podcast called Forgive and Remember... Charles Bosk spent 18 months with a group of surgical residents to understand the culture of surgical training.
Federal database tracking pharmaceutical and device industry payments to US physicians. Any patient can look up their oncologist's financial relationships with companies whose drugs they are being offered.
Prasad cites evidence that personal payments from industry are associated with greater pro-industry publication output even after controlling for research funding and seniority — suggesting a positive feedback loop. Patients can use the database to ask informed questions about their oncologist's relationships with manufacturers of recommended drugs.
Personal payments from the industry are associated with greater publication in the future. It's likely a positive feedback loop — working with the industry helps your career which helps you work with the industry which helps your career.
Malignant: How Bad Policy and Bad Evidence Harm People with Cancer by Vinay Prasad
Book Sponsored · disclosed
Prasad's systematic critique of cancer drug approval, conflict of interest in guideline writing, surrogate endpoint misuse, and his six hallmarks of sound cancer policy. Attia describes it as the synthesis of Prasad's decade of meta-research across oncology.
DisclosurePrasad is the author and the guest in this episode — he discusses the book throughout.
The book is structured around case studies and culminates in the six-hallmarks framework discussed in this episode's final segment: independence (conflict of interest), evidence (measure what matters), relevance (study average patients), affordability (value-based pricing), possibility (NIH funding), and agenda (deduplication of clinical trials). Prasad notes the hallmarks were an epiphany in the final stages of writing — he realized he could distill a decade of work into six principles analogous to Hanahan and Weinberg's hallmarks of cancer biology.
I start by in this chapter by saying here's a way I should have said things all along. I could have said things better. But at least for my case it's better late than never.
Ending Medical Reversal by Vinay Prasad and Adam Cifu
Book Sponsored · disclosed
The foundational book on medical practices that were widely adopted and subsequently reversed by rigorous RCTs — hormone therapy for cardiovascular protection, PVC-suppression drugs post-MI, arthroscopic knee debridement, and hundreds of others.
DisclosurePrasad is a co-author; discussed as background to current work.
Prasad describes the book as arising from his residency experience watching practices that seemed evidentially inconsistent. He began cataloguing 'medical reversals' with mentor Adam Cifu at University of Chicago. The key concept: reversal means the practice was not merely replaced by something better — it was found to be worse than doing nothing or reverting to the prior standard.
We started to ask a bunch of questions about how many medical practices are adopted based on low levels of evidence. And what happens when years later people come along and do really carefully done rigorous studies and find that some of them do not work as intended. We found that not doing it was better or whatever you did before was better.
Lines worth pulling out — contrarian, specific, or perfectly phrased
6 items
The average improvement was 2.1 months. So that's the average of 71 drugs. Two months can mean a lot to somebody. But two months should also come with another asterisk which is what is the cost of these medications. They now routinely run one hundred thousand to two hundred thousand dollars per year of treatment.
The single most-cited quantitative indictment in Prasad's body of work — the 2.1-month average reduction distills the entire cancer drug approval landscape into one uncomfortable number.
It was the methamphetamine of being a doctor — a highly addictive substance. You do this procedure, the patient comes to your office, they say thank you so much doctor, you saved my life. You pair that incredibly powerful feeling of gratitude with a little bit of financial remuneration — that's highly addictive.
Prasad's most memorable explanation for why sham-trial evidence fails to change clinical practice — the procedural gratitude loop is more powerful than a negative RCT.
Can you imagine a courtroom where the prosecutor and the defendant are both being paid by the defendant? I think many of us would say that's not really a balanced courtroom. And yet in medicine we do have many imbalanced courtrooms where almost everybody in the courtroom is getting money from the company.
The clearest one-line structural diagnosis of the conflict-of-interest problem in oncology guideline-writing.
I spent a lot of time trying to get to the bottom of why it is this 30 percent and I found out it goes back to a 1976 paper where this Mayo Clinic doctor got a bunch of marbles and he put him on a dining table and he rolled out foam rubber and he got 16 oncologists to come to his house with calipers and measure the marbles.
The marble-measuring origin story for the PFS response-rate threshold — a vivid illustration of how operational convenience calcifies into regulatory standard.
The real role for liquid biopsy might be the serial nature of it. You can track something over time and that might be the real boon. But they're already in our clinics — we use them.
Prasad's most actionable take on liquid biopsies: longitudinal monitoring is likely more valuable than single-point-in-time mutation detection.
Patients who are dying need us more than patients who are living.
Steve Rosenberg's teaching at the NCI, relayed by Attia — the bedside ethos that inverts the avoidance instinct when a patient has no more therapeutic options.
Sign in to share feedback
Tell us if this brief hit the mark or missed it — feedback feeds back into the next iteration of the prompt.
Reading is free for everyone. A free account adds the personal layer: save protocols, follow experts, and see how the other experts weigh in on this same topic.
Educational summary of the cited expert source — not medical advice. Open the source recording linked above and consult a qualified physician before acting on any protocol.
