I would add one observation about the idea of the FDA only evaluating "safety." It's not clear what that even means. Sure, we should screen out cases like TGN1412 that send everyone to the hospital with septic shock. But there are many drugs (cancer drugs in particular) that typically aren't "safe" per se. In many cases, they have side effects that are not safe at all. The only reason to ever take such drugs is because the potential benefits outweigh the serious risks, which requires knowing something about efficacy.
An example: adriamycin (doxorubicin) is a common chemotherapy drug. It comes in bags labeled "biohazard," and the side effects include not just hair loss, nausea and vomiting, and mouth sores, but all of these (copied from the Mayo Clinic):
* "This medicine may cause irreversible heart muscle damage, leading to heart failure."
* "This medicine may increase risk of new cancers, such as acute myelogenous leukemia (AML) or myelodysplastic syndrome (MDS)."
* "If doxorubicin accidentally seeps out of the vein into which it is injected, it may damage some tissues and cause scarring."
* "Doxorubicin can temporarily lower the number of white blood cells in your blood, increasing the chance of getting an infection. It can also lower the number of platelets, which are necessary for proper blood clotting."
* "This medicine may cause a serious type of reaction called tumor lysis syndrome."
* "This medicine may decrease the amount of sperm made and affect a man's ability to father a child. It may also cause premature menopause in women."
While less common, other side effects are:
Cough or hoarseness accompanied by fever or chills
darkening or redness of the skin (if you recently had radiation treatment)
fast or irregular heartbeat
fever or chills
joint pain
lower back or side pain accompanied by fever or chills
pain at the injection site
painful or difficult urination accompanied by fever or chills
red streaks along the injected vein
shortness of breath
stomach pain
swelling of the feet and lower legs
There's no way anyone could look at safety alone, and think, "Yep, this drug is safe."
This observation is profound: the therapeutic window concept precisely defines this vital balance between effective dose and toxicity. Notably, pharmacogenomics reveals that some patients metabolize drugs so slowly that standard doses become poisonous. In my view, decoupling safety from efficacy is a biological absurdity, as a useless drug is never "safe" if it carries any side effects.
The argument linking risk to benefit is crucial, as toxicity is often inherent to the drug's mechanism. Notably, about 30% of drugs withdrawn post-approval are due to unforeseen safety issues, proving that isolating "safety" without massive efficacy data is nearly impossible. I agree: a safety-only FDA would turn patients into involuntary test subjects, as without proven efficacy, no side effect is medically justifiable.
It’s surprising that more people aren’t talking about public trust erosion wrt opening the “right to try/pass safety” floodgates. If we think it’s bad now (especially with vaccines that work) wait until the majority of approved assets do absolutely nothing. Anyway, off my soapbox.
It is kind of funny hearing all the techno-optimists act like one more preclinical candidate is all we really need. Just one more bro. Sure there are thousands of shelved assets that have no path through trialing because it’s so expensive, time consuming, and burdensome, but all we really need is one more. As you point out, better candidates will certainly help but they aren’t the only thing!
China is very fast at getting clinical data and part of that is because of their large population and access to a ton of treatment-naive patients. If the US would play nice with allies and take the lead on trial/data standardization we could also have access to similar patient populations and so could developers in other countries.
what do you mean by this: "If the US would play nice with allies and take the lead on trial/data standardization we could also have access to similar patient populations and so could developers in other countries."?
Regulators here do work with those in allied countries and have collaborated with ICH on guidelines but there are often still different trial applications, timelines, reporting requirements, feedback, manufacturing audits/inspections, diff manufacturing requirements, etc, depending on locality. Japan also requires local bridging cohorts iirc? I just feel like if we were more respected internationally right now then our FDA could try to make a shared system (would that require a formal treaty?) vs parallel systems
International data standardization, through ICH's M4E(R2) project, is the secret weapon to rival China. Notably, AI cannot "invent" this data; it requires real human results to train without bias. In my view, without this global collaboration, public distrust toward ineffective drugs will permanently break Western biomedical innovation.
Exceptional work reframing clinical trials as the actual discovery engine rather than just validation. The clinical selection vs intelligent design framing is spot-on, semaglutide's story makes it obvious how much we learn by iterating in humans. What's interseting is how this aligns with venture's power law thinking but pharma still optimizes for preclinical predictivty. More shots on goal beats better aim when the payoff distribution is this heavy-tailed.
This perspective highlights the power of platform trials, allowing multiple treatments to be tested against one control group. Notably, this approach reduces required participants by 30–50%, maximizing "shots on goal." In my view, industry must embrace this agility, as the preclinical predictability dogma stifles innovation where human empiricism excels.
There is no replacement for clinical data. The more we expand clinical trial capacity, the more we learn what works and the more breakthroughs we find.
A very thoughtful article Ruxandra. Even your Clinical Trial Abundance project/policy reflects the same rigor & depth. I am building something adjacent at - https://signalprogress.com/
Prescient piece! As a clinician, nothing is more important than to “do no harm”. As such I appreciate your understanding where the friction really lies with Eroom’s‑Law‑style stagnation. Macroscience’s policy agenda implicitly tries to move the constraint from “we cannot ethically run more trials” to “we can ethically run more efficient trials if we attack waste, redundancy, and bureaucratic friction”; but that still requires constant vigilance so that operational reform does not quietly erode substantive protections. Any serious policy program in this space has to be explicit about that tradeoff, not hide behind engineering metaphors that are clean only because metal, unlike people, does not suffer.
The distinction between metal and biology highlights the importance of decentralized clinical trials, using wearables to reduce patient burden. Notably, these tools lower dropout rates by 15%, preserving ethics while accelerating data collection. In my view, innovation must never ignore human suffering; bureaucratic efficiency is only virtuous if it serves patient dignity.
Publicly-run platform-design clinical trials which test multiple therapies head-to-head would seem to be a) much cheaper per evidence generated b) more scientifically valuable.
Is it really too difficult to design a system where this is incentivised?
Thanks for flagging this piece, hadn't seen it- sounds a great idea, but let's do more common diseases also!
I sort of imagine something like a call for companies developing therapies for a particular condition to express interest, FDA/equivalent to work out then cost for each participating company which would presumably be much less than paying for your own trial and then away we go...
I think someone should look into clinical trial infrastructure more closely. Ideally in comparison with China who I've heard is very good at it. Unfortunately, my hands are full at the moment and *too few ppl* are working on this, as well as more systems level analyses of pharma productivity in the vein of what Jack Scannell is doing.
I was thinking of building a syllabus of important themes in clinical trials and unanswered questions so ppl have a reference.
This observation is crucial as the placebo effect is often stronger in surgical interventions or medical devices than in drugs. Notably, "innovator bias" frequently leads to overestimating these techniques before rigorous testing. In my view, applying clinical selection to devices is vital, as an "elegant" technology means nothing without proven improvements in patient quality of life.
Artificial intelligence, despite its rise, struggles to model the gut microbiome, whose individual diversity radically impacts drug efficacy. In my view, prioritizing human experimentation remains essential, as current digital simulations do not yet capture this unique biological complexity.
Great piece!
I would add one observation about the idea of the FDA only evaluating "safety." It's not clear what that even means. Sure, we should screen out cases like TGN1412 that send everyone to the hospital with septic shock. But there are many drugs (cancer drugs in particular) that typically aren't "safe" per se. In many cases, they have side effects that are not safe at all. The only reason to ever take such drugs is because the potential benefits outweigh the serious risks, which requires knowing something about efficacy.
I have not actually thought about this, but you are right. In some sense, you cannot talk about safety other than in a balance with efficacy
An example: adriamycin (doxorubicin) is a common chemotherapy drug. It comes in bags labeled "biohazard," and the side effects include not just hair loss, nausea and vomiting, and mouth sores, but all of these (copied from the Mayo Clinic):
* "This medicine may cause irreversible heart muscle damage, leading to heart failure."
* "This medicine may increase risk of new cancers, such as acute myelogenous leukemia (AML) or myelodysplastic syndrome (MDS)."
* "If doxorubicin accidentally seeps out of the vein into which it is injected, it may damage some tissues and cause scarring."
* "Doxorubicin can temporarily lower the number of white blood cells in your blood, increasing the chance of getting an infection. It can also lower the number of platelets, which are necessary for proper blood clotting."
* "This medicine may cause a serious type of reaction called tumor lysis syndrome."
* "This medicine may decrease the amount of sperm made and affect a man's ability to father a child. It may also cause premature menopause in women."
While less common, other side effects are:
Cough or hoarseness accompanied by fever or chills
darkening or redness of the skin (if you recently had radiation treatment)
fast or irregular heartbeat
fever or chills
joint pain
lower back or side pain accompanied by fever or chills
pain at the injection site
painful or difficult urination accompanied by fever or chills
red streaks along the injected vein
shortness of breath
stomach pain
swelling of the feet and lower legs
There's no way anyone could look at safety alone, and think, "Yep, this drug is safe."
This observation is profound: the therapeutic window concept precisely defines this vital balance between effective dose and toxicity. Notably, pharmacogenomics reveals that some patients metabolize drugs so slowly that standard doses become poisonous. In my view, decoupling safety from efficacy is a biological absurdity, as a useless drug is never "safe" if it carries any side effects.
The argument linking risk to benefit is crucial, as toxicity is often inherent to the drug's mechanism. Notably, about 30% of drugs withdrawn post-approval are due to unforeseen safety issues, proving that isolating "safety" without massive efficacy data is nearly impossible. I agree: a safety-only FDA would turn patients into involuntary test subjects, as without proven efficacy, no side effect is medically justifiable.
Fantastic work.
It’s surprising that more people aren’t talking about public trust erosion wrt opening the “right to try/pass safety” floodgates. If we think it’s bad now (especially with vaccines that work) wait until the majority of approved assets do absolutely nothing. Anyway, off my soapbox.
It is kind of funny hearing all the techno-optimists act like one more preclinical candidate is all we really need. Just one more bro. Sure there are thousands of shelved assets that have no path through trialing because it’s so expensive, time consuming, and burdensome, but all we really need is one more. As you point out, better candidates will certainly help but they aren’t the only thing!
China is very fast at getting clinical data and part of that is because of their large population and access to a ton of treatment-naive patients. If the US would play nice with allies and take the lead on trial/data standardization we could also have access to similar patient populations and so could developers in other countries.
Thank you.
what do you mean by this: "If the US would play nice with allies and take the lead on trial/data standardization we could also have access to similar patient populations and so could developers in other countries."?
Regulators here do work with those in allied countries and have collaborated with ICH on guidelines but there are often still different trial applications, timelines, reporting requirements, feedback, manufacturing audits/inspections, diff manufacturing requirements, etc, depending on locality. Japan also requires local bridging cohorts iirc? I just feel like if we were more respected internationally right now then our FDA could try to make a shared system (would that require a formal treaty?) vs parallel systems
International data standardization, through ICH's M4E(R2) project, is the secret weapon to rival China. Notably, AI cannot "invent" this data; it requires real human results to train without bias. In my view, without this global collaboration, public distrust toward ineffective drugs will permanently break Western biomedical innovation.
Exceptional work reframing clinical trials as the actual discovery engine rather than just validation. The clinical selection vs intelligent design framing is spot-on, semaglutide's story makes it obvious how much we learn by iterating in humans. What's interseting is how this aligns with venture's power law thinking but pharma still optimizes for preclinical predictivty. More shots on goal beats better aim when the payoff distribution is this heavy-tailed.
Thank you :)
This perspective highlights the power of platform trials, allowing multiple treatments to be tested against one control group. Notably, this approach reduces required participants by 30–50%, maximizing "shots on goal." In my view, industry must embrace this agility, as the preclinical predictability dogma stifles innovation where human empiricism excels.
There is no replacement for clinical data. The more we expand clinical trial capacity, the more we learn what works and the more breakthroughs we find.
A very thoughtful article Ruxandra. Even your Clinical Trial Abundance project/policy reflects the same rigor & depth. I am building something adjacent at - https://signalprogress.com/
thank you :)
Prescient piece! As a clinician, nothing is more important than to “do no harm”. As such I appreciate your understanding where the friction really lies with Eroom’s‑Law‑style stagnation. Macroscience’s policy agenda implicitly tries to move the constraint from “we cannot ethically run more trials” to “we can ethically run more efficient trials if we attack waste, redundancy, and bureaucratic friction”; but that still requires constant vigilance so that operational reform does not quietly erode substantive protections. Any serious policy program in this space has to be explicit about that tradeoff, not hide behind engineering metaphors that are clean only because metal, unlike people, does not suffer.
The distinction between metal and biology highlights the importance of decentralized clinical trials, using wearables to reduce patient burden. Notably, these tools lower dropout rates by 15%, preserving ethics while accelerating data collection. In my view, innovation must never ignore human suffering; bureaucratic efficiency is only virtuous if it serves patient dignity.
Related talk at https://www.linkedin.com/posts/ronnyk_abtest-successrate-pos-activity-7333357558973272064-jXDo
thanks
Really enjoyed reading this.
Publicly-run platform-design clinical trials which test multiple therapies head-to-head would seem to be a) much cheaper per evidence generated b) more scientifically valuable.
Is it really too difficult to design a system where this is incentivised?
How would this work? Like massive umbrella trial infrastructure?
we proposed smth like this for rare diseases last year (Frank David idea):
https://acrobat.adobe.com/id/urn:aaid:sc:VA6C2:c3a95fb9-b04b-49f6-93c5-0711b961dd7c
Thanks for flagging this piece, hadn't seen it- sounds a great idea, but let's do more common diseases also!
I sort of imagine something like a call for companies developing therapies for a particular condition to express interest, FDA/equivalent to work out then cost for each participating company which would presumably be much less than paying for your own trial and then away we go...
I think someone should look into clinical trial infrastructure more closely. Ideally in comparison with China who I've heard is very good at it. Unfortunately, my hands are full at the moment and *too few ppl* are working on this, as well as more systems level analyses of pharma productivity in the vein of what Jack Scannell is doing.
I was thinking of building a syllabus of important themes in clinical trials and unanswered questions so ppl have a reference.
Re why for rare diseases.
It's the area where the problem of recruitment manifests itself most acutely and would allow targeted investment.
I agree it should be expanded across the board. As I said, not enough attention paid to it.
Really great piece, as simple as possible but not simpler. It all applies to non-drug medical interventions as well.
This observation is crucial as the placebo effect is often stronger in surgical interventions or medical devices than in drugs. Notably, "innovator bias" frequently leads to overestimating these techniques before rigorous testing. In my view, applying clinical selection to devices is vital, as an "elegant" technology means nothing without proven improvements in patient quality of life.
Artificial intelligence, despite its rise, struggles to model the gut microbiome, whose individual diversity radically impacts drug efficacy. In my view, prioritizing human experimentation remains essential, as current digital simulations do not yet capture this unique biological complexity.