I talk a lot about the data we leave behind. A lot of this time, this is in the context of clinical trial design. But more recently, it’s also about the data loss that drives bias in AI. And actually…that same bias exists in humans. And seeking to fix one may also fix the other!
To clarify, it’s not that no one cares (as a reason for why data loss or missingness happens). It is an artifact of our systems, funding priorities, and a whole bunch of historical decision making. Clinical trials cost money and are usually designed to answer a question around safety or efficacy, often in pursuit of regulatory approval and eventual commercial distribution. We need that (I’m not saying we don’t!), but we also need funding sources to answer additional questions related to titration, individual variance, medication impacts, more diseases, etc.
A lot of trials exclude people with type 1 diabetes, for example, so it’s always a question when new medications or treatments roll out whether they are appropriate or useful for people with T1D or if there’s a reason why they wouldn’t work. This is because sometimes T1D might be seen as a muddy variable in the study, but a lot of times it’s just a copy-paste decision from a previous study where it did matter that is propagated forward into the next study where it doesn’t matter. There is no resulting distinction between the two: there is simply a lack of representation of people with T1D in the study population and this means it’s hard to interpret and make decisions about this data in the future.
A lot of decisions around what data is collected, or who is enrolled in a population for a study, is an artifact of this copy-paste! Sometimes literal copy-paste, and sometimes a mental copy-paste for “we do things this way” because previously there was a reason to do so. Often that’s cost (financial or time or hardship or lack of expertise). But, in the era of increasing technological capabilities and lower cost (everything from the cost of tokens to LLM to the decreasing cost of storage for data long-term, plus the reduced cost on data collection from passive wearables and phone sensors or reduced hardship for participants to contribute data or reduced hardship for researchers to clean and study the data) this means that we should be revisiting A LOT of these decisions about ‘the way things are done’ because the WHY those things were done that way has likely changed.
The TLDR is that I want – and think we all need – more data collected in clinical trials. And, I don’t think we need to be as narrow-minded any more about having a single data protocol that all participants consent to. Hold up: before you get the torches and pitchforks out, I’m not saying not to have consent. I am saying we could have MULTIPLE consents. One consent for the study protocol. And a second consent where participants can evaluate and decide what additional permissions they want to provide around data re-use from the study, once the study is completed. (And possibly opt-in to additional data collection that they may want to passively provide in parallel).
The reason I think we should shift in this direction is because that, out of extreme respect to patient preferences around privacy and protecting patients (and participants in studies – I’ll use participants/patients interchangeably because it happens in clinical care as well as research), researchers sometimes err on the side of saying “data not available” after studies because they did not design the protocol to store the data and distribute it in the future. A lot of times this is because they believe patients did not consent (because researchers did not ask!) for data re-use. Other times it’s honestly – and I say this as a researcher who has collaborated with a bunch of people at a bunch of institutions globally – inadvertent laziness / copy-and-paste phenomenon where it’s hard to do the very first time so people don’t do it, and then it keeps getting passed on to the next project the same way, because admittedly figuring it out and doing it the first time takes work. (And, researchers may have not included in their budgets to cover data storage, etc.). Thus this propagates forward. Whereas you see researchers who do this on one project tend to continue doing it, because they’ve figured it out – and written it into their research budgets to do so for subsequent projects. And other times, the lack of data re-use is a protective instinct because researchers may have concern that the data may be used in a way that is harmful or negative. Depending on the type of data (genetic data versus glucose data, as two examples), that may be a very legitimate concern of study participants. Other times, it may be at odds with the wishes of the broad community perspective. And sometimes, individuals have different preferences!
We could – and should – be able to satisfy BOTH preferences, by having an explicit consent (either opt-in or opt-out, depending on the project) that is an ADDITIONAL consent specifically indicating preferences around data being re-used for additional studies.
To emphasize why this matters, I’ll circle back to AI/LLMs. A criticism of AI is that they are trained on data that is not representative; has holes; or is biased. This is legitimate and an artifact of the past design of trials. I have been thinking through the opportunities this provides now, in the current era of trial design, to be proactive and strategic about generating and collecting data that will fill these holes for the future! We can’t fix the way past trials were done; but we can adjust how we do trials moving forward, fill the holes, and update the knowledge base. This actually then fixes the HUMAN PROBLEM as well as the AI problem: because these criticisms of AI are the same criticism we should be making about humans (researchers and clinicians etc), who are also trained on the same missing, messy, possibly biased data!
More data; more opt-in/out; more strategery in thinking about plugging the holes in the knowledge base moving forward in part by not making the same mistakes in future research that we’ve made in the past. We should be designing trials not only for narrow endpoints (which we can also do) for safety and efficacy and regulatory approval and commercial availability, but also for answering the questions patients need answered when evaluating these treatments in the future: everything from titration to co-conditions to co-medications to the arbitrary research protocol decisions around the timing of treatment or the timing of the course of treatment. A lot of this could be better answered by data re-use beyond the additional trial as well as additional data collection in conjunction to the primary and secondary endpoints.
All of this is to say…you may not agree with me. I support that! I’d love to know (constructively; as I said, no pitchforks!) what you disagree about, or what you are violently nodding your head in agreement about, or what nuances or big picture things are being overlooked. And, I have an invitation for you: CTTI – the Clinical Trials Transformation Institute – is hosting a 2-hour summit on April 28 (2026) for patients, caregivers, and patient advocates designed to facilitate discussion around the range of perspectives on these topics as they apply to clinical trials. Specifically 1) the use of AI in clinical trials and 2) perspectives on data re-use beyond the initial trial that data is collected for.
If you have strong opinions, this is a great place for you to share them and hear from others who may have varying or similar perspectives. I am expecting to learn a lot! This event is not designed around creating consensus, because we know there are varying perspectives on these topics, and thus especially if you disagree or have a nuanced take your voice is needed! (I may be in the minority, for example, with my thoughts above.) 
Participation is focused (because these topics are scoped to focus on clinical trials) on anyone who has ever been part of a clinical trial, is currently navigating one, made the decision to leave a trial early, or even those who looked into participating but ultimately decided it wasn’t the right fit .
You can register for the summit here (April 28, 2026 10am-noon PT / 1-3pm ET), and if you can’t attend, feel free to drop a comment here with your thoughts and I’ll make sure it is shared and included in the notes for the event that will also be shared out afterward aggregating the perspectives we hear from.