Third Bridge专家称临床试验IRB/委员会服务格局稳定但AI长期可能压缩人工审评需求

[[00:01:40]]
How transferable are committee vendor selection dynamics to central IRB [institutional review board] bidding, particularly when preferred provider pricing is more or less identical?

[[00:02:03]]
This, I guess, is maybe part of the problem of this particular service. They really operate as an oligopoly that everybody is really okay with the status quo. No one is particularly good or particularly bad, but because vendors are cognisant that we need this service, one is maybe no better than another. From the US point of view, if we use a central IRB related to a university or if we use a central IRB routed through a provider like Icon or like Advarra, do we see any difference in service, being honest, not particularly, no.

[[00:02:58]]
When Advarra, WCG and Icon submit comparable bids, which project manager behaviours most influence your award decision and later predict smoother execution?

[[00:03:15]]
Really, the first thing from my point of view is I want to see a project manager who has an understanding of the protocol, who has an understanding of the drug. Quite often, when we go through the process of bidding, what I like is, I actually quite like lots of questions about the protocol, lots of questions about the disease area. That tells me that the company is actually interested rather than just paying lip service to it all. Then they come and present. One key from my point of view is so often with external providers, comparing it to a football analogy, you often get the Premiership team, the Liverpool or the Manchester City turning up to do the bid. Then actually, quite often, you get a third division, Scunthorpe United coming in to run the project. Actually, that's the type of thing that causes me the most antibodies. If I get a project manager coming in to defend the bid who says, "I'm really looking forward to working with you on this project." That's a big buy signal from my point of view. That, coupled with real engagement in taking on board the questions about the protocol, the disease area and so on, those are the two biggest buy signals from my point of view.

[[00:04:43]]
How do you distinguish genuine team chemistry during vendor presentations from polished salesmanship that may not translate into effective trial delivery? Can you think about instances where your expectations were subverted in an adverse way vs the times when you saw an alignment between what was promised and what was delivered?

[[00:05:04]]
Yes, and you can often pick these out. The key behaviour that you often get is you get a polished salesman who drops out a lot of corporate conclusions about this company does A, B and C, and this is amazing and this is our corporate commitment vs the true project manager who's collaborative, who comes in, who can dig into the weeds, who can report from their own personal experience, stories of where things have gone wrong and where things have gone right. Those are the types of differences in the delivery that you really look forward to and look to spotting to make sure you can weed out the slimy salesman from the hard worker.

[[00:05:57]]
What's the execution risk when a favoured project manager changes mid-study? Does the replacement process usually reveal meaningful differences in provider depth?

[[00:06:12]]
Unfortunately, it can do, yes. This is the one criticism I've got of almost every major provider that I've worked with, be it Icon, Iqvia, WCG, whoever, really. You do have particularly favoured project managers you like working with, but it's equally unrealistic to say, "Well, I want Jane Smith because I know she was amazing last time," and Jane Smith could be moving on to another three or four different projects by the time you've got yours up and running. I think there is, to be honest, a general risk and it is quite a high risk of lack of continuity, really.

[[00:07:00]]
Where do set-up delays most commonly occur across steering, adjudication and safety monitoring committees? How many weeks can typically be added? Which providers do you think respond the best?

[[00:07:16]]
In terms of set-up, often there's a discussion and a lack of transparency about the price the particular committee are exacting. For example, if you're contracting with an IRB there's normally not too much price negotiation. Equally, if there's a lack of transparency between what the real cost of the IRB is and what the managing organisation, the Advarra, the Icon are charging on top, that can significantly impair contracting times and can delay things by about six or eight weeks or so. The other thing that can often lead to a delay is a lack of transparency about, for example, if you're looking at a cardiovascular adjudication committee, how many events are predicted during the trial because what happens is the charging structure is often done according to how many events are predicted at a certain time, and that then drives the scheduling of meetings and it drives the amount of hours the key experts are paid for online for doing adjudication. It's important to get that right at the beginning. To get that wrong can drive significant delays of a matter of weeks, maybe even four or six weeks while these are renegotiated and people go over numbers again and again.

[[00:08:52]]
What proportion of external review activity is generated by large phase 3 programmes vs early-stage studies?

[[00:09:05]]
It's a good question. I think it's probably 85% or 90% by large phase 3 studies. Might that change in the future? Well, it's quite interesting. We have had some experience of a potential change. I ran a couple of programmes for liver for patients with quite advanced liver disease. There, we planned to have an internal safety review board, but it was made very clear by the liver division of the agency that they would prefer us to have an external review board for the data. That was really a change. I've also seen in some rare disease indications where you only have very limited exposure by the time you move into a pivotal trial. The agency and actually often the pharma company turn around and go, "Well, there's maybe too much risk baked into this programme. Let's have an external review committee to make sure nothing goes wrong." Quite often for rare diseases where you have a very limited expertise around the world, that's actually essential because you need to understand a lot more about the life course of a particular disease. In short, 85% or more to phase 3s at the moment, might this change? Well, as we go into people with more severe subsets of chronic disease with liver cirrhosis a good example or as we go more into rare diseases, there's the potential for that balance to change a little bit.

[[00:10:43]]
Once a large phase 3 trial begins, how long do adjudication, steering and safety monitoring workloads persist despite the sponsor funding pressure?

[[00:11:03]]
Well, I don't think sponsor funding pressure, being completely honest, should play a role in how long these committees run for and what committees you have because at the end of the day, you've got a GCP commitment to do the trial right if actually you need a committee right the way through, until the last event has been adjudicated, you need a committee right the way through. You can't suddenly turn around and go, "Well, once we've had a year of exposure and it's all okay, we can discontinue the committee and fire people," because you don't know about late events, you don't know about late side effects of a particular drug. The right thing to do is always to commit for the duration and actually not to discontinue partway through.

[[00:11:58]]
How quickly would repeated phase 3 failures translate into lower committee and IRB workloads, considering the long duration of already-enrolled studies?

[[00:12:10]]
Of course, most phase 3s take three or four years to cook. For a smaller company that was backing maybe two or three phase 3s over a seven- or eight-year period, that takes a while to feed through into a reduction in cash flow. If you were, for sake of argument, like Glaxo in the 2011-13 timeframe when there were a number of massive phase 3 failures within a short number of years, that then translated into a cessation of activity in the cardiovascular space within a very short time period. I was lucky enough to move on to another company, but many others weren't. Of course, that then led to a very rapid and precipitous fall-off in this type of IRB committee type activity.

[[00:13:13]]
What pricing differences would become large enough to override a strong project chemistry? Does that threshold vary by committee type or trial complexity? What are the competitive moats and propositions of a company such as Advarra against its peers? What's the influence of pricing vs pure execution?

[[00:13:36]]
A good question again. While these committees can cost between maybe USD 4m and USD 6m on to a quote of an overall phase 3 trial. In the context of a phase 3 that cost USD 250m-300m, that's still only maybe 1% or 1.5% of the total budget. Actually, if the price was suddenly to go up by 15% or 20%, you'd argue that's within variance of study costs anyway. Actually, there's quite a leeway before price becomes the consideration. It's much more about study quality, which is the main consideration and wanting to avoid either questions about committee conduct, questions about accuracy of recording of events and so on. These are all things that you absolutely want to avoid.

[[00:14:39]]
How do sponsor relationships with Veeva, Medidata or enterprise CROs [contract research organisations] influence whether Advarra is considered early, shortlisted or excluded?

[[00:15:02]]
The first thing to say is that we're most interested in the best-in-breed type of approach. We're not necessarily interested in, Parexel says we must have A, B, C or D or Iqvia says the same thing. When they come along and suggest that that's normally a big red flag to me to say, "No, thank you. We'll consider going with somebody else." They're not a major consideration. We tend to go for the best-in-breed approach. We go for the best-in-breed approach because, again, we want to make sure the ethical decisions of an IRB are entirely defendable or the quality of data generated or the quality of decision-making of a trial committee is entirely defendable, accurate and so on and so forth.

[[00:16:03]]
Where can Icon's broader CRO relationships create an advantage over Advarra or WCG by bundling committee services into larger trial awards? How do you think that typically works in the value chain?

[[00:16:22]]
My observations to where there's a potential problem is that there can occasionally be difficulties in speed or transference of data on one portion of the clinical trial apparatus speaking to another one. That's not what I want to see. We've had an event in a trial I conducted a few years ago where there was inadvertent failure to transfer over a very large quantity of event data. This meant that there was a build-up of over three months. The adjudication committee said, "Oh, this is great. You haven't had any events over the last three months." Then the CRO was saying, "Well, actually, we've had 55 events," or whatever. "Why haven't you got them?" Then it was realised there was a computer error. What I would say is having an integrated CRO reduces the possibility of that happening, and that's quite attractive in some ways but is it also necessary when actually most companies work on unified coding and so on and so forth. It's maybe not so much of a problem these days.

[[00:17:45]]
In global trials, which regulatory data-driven governance requirements most favour established committee platform providers? Where can local alternatives still compete?

[[00:17:58]]
It's a good question. I think, to be honest, still the FDA because of unification of data quality, decision-making quality, the FDA really tend to favour these type of single point of contact approaches and so on. I think in Europe, it's probably less rigid. Certainly in Japan, it's less rigid. Bearing in mind basically 90% of revenues for most drug projects still come from the US. Of course, it tends to be the US preference that wins the day.

[[00:18:48]]
Despite academic funding pressures, which sponsor-side indicators – such as protocol approvals or committee charters – would signal that commercial review demand is beginning to weaken?

[[00:19:05]]
I think really charter sign-off is pretty much the final step. You might come along and propose protocols for review maybe six months in advance of a phase 3 start. I guess if you're looking particularly for what the biggest sign of the canary in the coal mine is, then it's people not moving to final charter sign-off because that effectively illustrates that you're not moving forward because you haven't chosen your personalities, you haven't chosen the structure of your trial committee and so on.

[[00:19:52]]
Compared to WCG and Icon, where does Advarra's committee platform perform better or worse on usability, data ingestion, reporting and auditability?

[[00:20:06]]
Because Advarra have a good heritage in biometrics, what it works well in is very seamless transfer of data from the blinded to the unblinded side of the trial conduct committee. With other sponsors, particularly WCG, we've actually had to have very separate structures and personalities and had to have a separate e-mail line for communication and so on but with Advarra, because they are experts at this type of data analysis, they are very good at separating the sides of the firewall and doing this in an automated way. You need to have less personalities and less checks to make sure that there's no risk of data crossing over to the dark side as it were.

[[00:21:06]]
What interoperability gaps create the most manual work where Advarra must exchange data with Veeva, Medidata and statistic teams or sponsor systems?

[[00:21:20]]
Like I hinted at, the biggest headache bits on all of this is where you're exchanging data on unblinded patients because you have to make sure this is absolutely secure and none of the team involved in doing ongoing study conduct see any of this data. That's really the most important thing that has to go on. That still presents the greatest headache although like I said to you, Advarra are really the best of the bunch at being able to do this well.

[[00:22:02]]
Why are sponsors reluctant to buy the broader Advarra software? Which adjacent workflows would offer the most credible cross-selling entry point for Advarra?

[[00:22:14]]
Data around generation of tables, figures and listings. Basically, the statistical software for analysing the results of the trial because clearly, if actually you were to cross-sell the same bit of software for generating TFLs from the raw data, that could be an obvious easy win for Advarra to cross-sell. Why are companies a bit reluctant to do that now? I think it's because they might already have signed into buying similar pieces of software from the providers of the clinical trial database, so the Veevas and the Medidatas or actually lots of companies now are beginning to work, including ours on Claude programming to try and automate at least some of the TFL generation to help to save time and money.

[[00:23:20]]
Which platform set-up tasks are repetitive enough for AI automation? Where would human clinical judgment remain indispensable?

[[00:23:38]]
Currently, we spend a large amount of money and time employing medical writers to do generation of new charters and to rule on event types and definitions and so on. Now this should all be generalisable to AI by using an AI agent. We've actually asked the AI agent to write a draft committee manual. This has worked spectacularly well and has maybe saved at least two weeks out of the medical writing process for an initial charter. Of course, you still have to have a human in the loop to review and look at the initial versions, but it's potentially saved a significant amount of time. The other area where AI is potentially useful is for automating production of TFLs, which are useful from the unblinded data. Now it's really very possible to build, particularly around safety events like myocardial infarction or heart failure and so on to actually build an automated method to pull those out of the data. I think, again, that's particularly vulnerable to being automated by AI really very rapidly.

[[00:25:17]]
What do you see as the biggest longer-term innovations with AI in the IRB and general clinical trial review space? What are the biggest value-add propositions? What aspects of the work value chain will see the biggest benefit? Where do you see the 2-3 biggest opportunities? Is it around revenue expansion or cost saving?

[[00:25:45]]
The first one, I'd say, would be use of clinical trial software like the IDA programme that was developed by my company in conjunction with Duke, I believe, whereby they were looking to automate the event adjudication process. Now, this is currently being trialled across a number of studies, but it looked to me like as per the strapline of changing four months to four minutes, this is eminently possible across a number of languages to use an AI agent to actually read the file and to make reasonable value judgments about whether a cardiovascular event has occurred or not. I think that represents a great opportunity to save money for drug companies, but it also would reduce the effort significantly on behalf of an agency managing an IRB and actually may even reduce or obviate the need for an adjudication committee and so on because those are no longer required because you no longer need three cardiologists to agree on a particular event or not. That I see is the massive opportunity for pharma, which could save vast amounts of money, but it would also significantly reduce work for these committee providers and also for the academic institutions who constitute the committees.

The other area of automation and AI review comes around just general in-trial safety events, looking for patterns and looking potentially for stuff that could be evaluated moving forward. I think many companies are already experimenting with what's called risk-based quality monitoring, and I see that as another potentially big advantage for AI moving forward.

[[00:28:07]]
Could you talk about the work that AI-native vendors and sellers do? Who do you think will be the biggest disruptors in the space over the next few years? Who do you consider to be most promising?

[[00:28:26]]
Yes, of course. We work across all of the generic providers for example, Claude, Microsoft, OpenAI and so on, Novari. If I was to particularly highlight one, and I've written a number of AI agents to improve clinical trial conduct, I'd probably highlight Novari as potentially a big disruptor because the language used to write Novari prompts is very straightforward. The analysis appears very powerful and very rapid and the ability to hallucinate appears to be very small. I'd highlight Novari as a mover and shaker. In the area of further process automation, I'd also highlight Faculty who I've had some contact with and I've helped them with reviewing their products and their materials. They are particularly interested in improving flow of a clinical trial. One of the things that I've discussed with them over the years is if they can improve process flow with respect to charters and adjudication and so on. They're currently working mainly on process flow to improve set-up and running of clinical trials, but it's only a very short leap for them to do the latter and move into adjudication committees as well.

Then the two that seem to be at the forefront with respect to medical writing are Writer and Yseop and to be honest, I know they sell themselves as being amazing at medical writing and clinical trial writing but actually, what I would say is they're not necessarily differentiated from any mainstream AI providers. In other words, what you can do with them, you could easily do with Microsoft Copilot with the right prompts as well.

[[00:30:46]]
From a customer's perspective, does Advarra's experienced but more traditional staffing model improve reliability or reduce adaptability during complex trial changes? How agile is the company compared to peers?

[[00:31:10]]
It's a good question and it's a little bit dependent upon personality. This is why the project chemistry and making sure you get the right people is so important. With the right personalities, they're prepared to bend over backwards and they are prepared to work hard to deliver any changes to the protocol changes to the charter and so on. Actually, maybe their slightly higher staffing levels vs a company that does more automation is useful in that respect because they do deliver on time and to a relatively high level of accuracy where you need to make changes to a project. Is one company necessarily markedly better than others? I'm not necessarily sure about that.

[[00:32:12]]
Given that AI automation would be the biggest value-add for Advarra specifically, where have been the pitfalls in your relationship with customers? Where do you see AI closing execution gaps relative to peers? Which aspects are more intrinsically problematic for the company as opposed to industry norms?

[[00:32:40]]
I hinted at this a little bit and actually, one of the problems where I've had feedback from committee chairs over the years is in the formatting and the presentation of unblinded data. Of course, they never discuss the data with me because that would be against GCP, but they actually do complain about the formatting often and they say, "Well, this hasn't been quite right. We found errors in the data," and so on and so forth. Where I think AI would add significantly is it would add significantly with respect to improving the quality of that data, which is presented to, for example, an adjudication committee or to an endpoint monitoring committee, whereby it's much easier and more intuitive for people who are basically clinicians to see. They've been expected to go through pages and pages of data tables. Unless people have a significant personality, which means they're used to looking at that type of data at length, they often miss things. The biggest add would be with AI driving better presentation of unblinded data and potentially going a stage further, as I hinted at, maybe adding in a provisional view on to adjudication.

[[00:34:21]]
In the last 2-3 years, where do you think Advarra has risen or fallen behind vs WCG and others?

[[00:34:34]]
WCG appeared to have gained market share slightly. I think they've underbid a little bit vs Advarra on a number of contracts. They've also, I've heard from colleagues, bid where there's potentially two or three studies in a therapy area gone for these bundling type deals where they've come back and said, "Well, actually, you can use the same cardiologists or the same kidney doctors on this trial as on this trial, and this could help you save X or Y amount of money." They've been a bit more competitive on price than Advarra and they've been perceived a little bit better on team chemistry. They've been perceived as caring more if that makes sense. Those are the two areas where I'd highlight maybe Advarra has fallen back vs WCG. Equally, they're both a little bit ahead of Icon, where I think Icon is a big supertanker and are just basically responding and behaving like any massive CRO would.

[[00:36:03]]
You said that WCG is undercutting Advarra. In regard to the pricing differential that you see with Advarra vs WCG, which committees or review boards are these? What gaps in pricing are normal to assume?

[[00:36:28]]
It's actually not huge, but it's certainly in the range of 10% or 12% or slightly more. Actually, what they appear to have done is bundled a lot of capacity with some of the academic units in the US and because they've bundled capacity and they've committed to running more studies, they're clearly sitting on a volume type discount, effectively.

[[00:36:58]]
What do you think is a suitable response to that on Advarra's part, where the company emphasises how it bundles its own review proposals, structures RFPs [requests for proposal] and responds to pricing? What would make customers favour Advarra over WCG?

[[00:37:22]]
Like I hinted at, to an extent, maybe a race to the bottom is not the best way forward. You can either appear as the budget airline or you can appear as the airline where you pay for quality. I would almost say that it's actually slightly better to move more into the AI tool provision, say, we can automate production of some of the unblinded TFLs. We can add in maybe indicative AI to help the cardiology professors make a decision, which might speed up the time from A to B to make a decision on a particular piece of data and so on and to reduce costs overall. I'd go on a flight for quality and a flight to add in more material rather than necessarily going for a price-cutting war because no one really wins a price-cutting war.

[[00:38:23]]
What is your 12-month outlook for Advarra, given the dynamics that we've noted? You termed the race to the bottom as hypothetical, but are margins compressing or expanding in the industry?

[[00:38:38]]
There's a push-me-pull-you type thing there. Clearly, on the one hand, there's a number of new products, GLP-1s come out of the woodwork every day. There are many more drugs for weight loss, which will require cardiovascular outcome studies, many more kidney drugs, which will require outcome studies and so on. The overall market for trials is going up and so the pressure on these committees will increase, which is only going to increase margins and allow old professors and companies to push up their prices a little bit. On the other hand, too, there's the ever-present threat of increased automation from AI. If we're having to negotiate and we know, for example, that all the programming can be automated for a particular trial or we can use an agentic AI to give an indicative diagnosis to a professor to save him 20 hours reading the notes, then actually, we're going to expect that to be fed through to us with respect to a price reduction or at least a cap on further price increases.

[[00:39:57]]
What do you think are the biggest inflection points facing the US clinical review oversight industry over the next 12-18 months?

[[00:40:10]]
Good question. I think it will be interesting to see if there are positives or negatives from some of these almost conditional licensing of drugs that are useful to Americans. The two cases in point would be the oral PCSK9 from Merck and the oral GLP-1 from Lilly. Now, both of those were licensed on less data. There's a potential need to follow up patients for longer and to maybe even look at data out in the real world. Maybe one big inflection point is what happens with that. Do they licence more products in that way and hence, there's more need for post-marketing surveillance or do they licence fewer products in that way because there's a problem. If there's a problem, will this then drive up traditional outcome trials where you maybe need to have this type of committee. That, to me, is going to be the biggest inflection point over the next few years. The other one, too, that will come along is around where trials are conducted. No one quite understands, I think, at the moment where most-favoured nation status will go. Clearly, if MFN drives that you don't do your trials or launch in certain countries, this also may impact on the workload or the way that you manage committees for outcome trials.

[[00:41:48]]
What downside scenarios would most significantly threaten Advarra's review volumes – fewer phase 3 starts, lower event rates, sponsor consolidation or technology substitution? Could you qualify and rank these threats to the company's market share and business volumes?

[[00:42:12]]
Technology substitution is the big one, as I hinted at earlier. In essence, assuming the automated adjudication via AI is confirmed in a couple of trials, then people will say there's no need to have a cardiovascular adjudication committee anymore. Actually, once that goes, that will have the biggest impact on the amount of work for these committees. The other one that will come through is if there is a change in policy of the FDA, in particular, with respect to the need to reduce follow-up for exposure. Clearly, if you can go into a computer database like Epic and use that to monitor outcomes in the real world, you're not going to need a cardiovascular outcome study and you're not going to need an adjudication committee. The next one down, I see as being a potential negative headwind is around regulatory changes, which allow earlier licensing of innovative medicines. Equally, that could be an upside where you get providers maybe branching out into using real-world electronic healthcare record data to make safety conclusions. Then with respect to more or less phase 3 starts, I think the majority of big pharma still has to keep the people in the city of London happy. There's still a need for fill and flow even if people have some failures, there's a need for fill and flow to keep things coming through the pipeline. I don't see that as a major problem.

[[00:44:23]]
If sponsors increasingly in-source trial operations, which Advarra services would remain the hardest to replicate internally vs the transactional segment revenues that face the greatest pressure?

[[00:44:41]]
Clearly, you have to be at arm's length with response to managing these committees. Actually, although you could go out of your way to make that possible with using internal resource, there is always a question or is there a bad smell about running these committees from an internal resource. I think it's still very, very difficult to replicate those internally. The committees, I think, are a very defendable area of business. When it comes to the other areas of Advarra's business like, for example, trial statistics, modelling data presentation and so on, I think those are under threat significantly from artificial intelligence over the next year to 18 months. The greater artificial intelligence can be used to build, for example, products to generate TFLs and to generate data presentation, we will be jumping on the bandwagon very quickly as sponsors.

[[00:45:59]]
What is your 6-12-month outlook for Advarra relative to the company's performance over the last two years? How do you see it trending in terms of market share gain or loss, overall business profitability and competitive dynamics?

[[00:46:23]]
Ultimately, I'd say revenues will be stable or slightly declining. I think, the reason I conclude stable or slightly declining is because of some of the technology pressures that we've talked about, not because there's maybe less turnover in the market or there's less demand for their services, but there will be some technology pressures. As long as a company like Advarra is also deploying technology to improve and to automate aspects of the committees, which has been the purpose of the discussion, then they should also be able to cut spend from their bottom line to maintain profitability.

[[00:47:11]]
Is there anything else you would like to highlight around Advarra and the broader industry?

[[00:47:29]]
I think, just to highlight again that there is the potential as computer automation and continues to hot up to actually move from the traditional way that we conduct trials to this embedding of trials within the electronic healthcare record and people go, "Well, that will still be years and years away from when that actually happens." What I would probably say is that it's probably closer than people think. Actually as computing power and AI power increases, that will probably be the most significant shake-up to the model of computer software and CROs and the agencies that run committees and so on.

[[00:48:16]]
I think that's an excellent spot for us to leave off. I want to thank you for a very pleasurable conversation, filled with great key insights. I'm sure very much so as well, our readers will appreciate it. I do want to thank those readers for tuning into another Third Bridge Interview today.

[[00:48:37]]
Thanks very much. Bye-bye.

[[00:48:39]]
Take care. Bye.

Transcription ends at 00:48:41 of the recorded material