Timely patient recruitment is critical to meeting clinical trial timelines and budgets. And, the speed of patient recruitment is directly tied to selecting the best and most engaged sites. However, that’s not an easy task. The information available for identifying suitable investigators comes from diverse sources and is often inconsistent and of poor quality. The result? Slow site selection and underperforming sites. In fact, industry data shows that 19% of trial sites fail to enrol a single patient, and over 30% of sites fail to meet their enrolment goals. Ultimately, these failures extend study timelines – delays that carry massive daily costs of $600K to $8M.

ICON has built a proprietary site section platform, One Search, which automatically ingests, aggregates and cleans data from multiple sources including quantitative historical data amassed on investigators to propose high performing sites for selection. Through an advanced process that makes use of human-supported machine learning (ML) we’re able to more accurately forecast the performance of a site or trial. In this way, we’re able to recommend high-performing sites quickly – generally within 48 hours.

To assess our effectiveness executing study start up, we reviewed study start-up data for two groups of studies: those in which ICON was not responsible for proposing and selecting sites and those in which ICON was responsible for those activities. We examined metrics from 41 full-service, non-vaccine studies, with the actual last subject randomised (ALSR) between May 2021 and April 2022.

Let’s look at start-up speed another way. As Figure 3 illustrates, ICON was able to propose 100% of the number of sites contracted in six to seven weeks, on average, and to select 100% of those around week 21. Figure 4 shows that for non-ICON studies, the first goal was met around the 17-week mark, on average, and the second goal was never met. In other words, sponsors, when responsible for their own site selection, did not ever succeed in selecting all of the sites they actually needed.

We believe that one reason for this shortfall is that sponsors tend to underestimate the number of sites that they will actually need. In contrast, ICON knows from experience that one must consider about 25% more sites than will ultimately be contracted. (ICON’s Site Identification staff may even propose a greater overage in some cases.) Figure 4 reveals that typically, sponsors’ site selection process tended to plateau right around the time that a list of desired sites was complete.

A further analysis of the time from FSP to FSS shows that just over 10 of the 16, or 63%, of the studies managed by ICON selected the contracted number of sites within 30 weeks from first site proposed for pre-selection visit (PSV).

A further analysis shows that when sponsors managed their own start-up activities, only 48% (12 out of 25) of studies had selected the contracted number of sites within 30 weeks from first site proposed for PSV.

It is apparent from comparing the timelines in Figures 3 and 4 that it is not just site recommendations that lag in sponsor-managed studies, but site selection as well. Study managers simply do not act on the information on sites that they have on hand as promptly as they might. And, regrettably, once sponsors get close to their target number of sites with investigators that they believe are of high-quality, their selection process frequently stalls. We have also observed that many study managers prefer to review sites in “batches,” a practice that inevitably delays start up and patient enrolment.

A key takeaway from this is that deciding quickly on which sites to engage increases the probability of meeting the FPR & LPR milestone baseline on time or within 7 calendar days by almost 20% (comparison of studies where ICON vs. the sponsor is responsible for proposal and selection of sites).

Does ICON’s accelerated methodology for providing a list of recommended sites result in a selection of less-qualified sites? ICON has examined this carefully and can confidently state that it does not. Our performance statistics on the number of recommended sites that actually randomise patients bears this out.

Figure 5 shows that across non-vaccine studies in 2021, in active sites that are still enrolling, nearly 57% of sites recommended by ICON recruited a subject. This is a 20% improvement over sites that were recommended by sponsor and not by ICON, and it is a 14% improvement over when the sponsor is solely responsible for site identification.

For activated sites that have completed enrolment, 66% of sites recommended by ICON recruited a subject. This is a 16% improvement over sites that were recommended by the sponsor and not by ICON, and a 34% improvement over when the sponsor is solely responsible for site identification.

ICON’s database, AI applications, and experience allow us to recommend and select qualified study sites faster than sponsors working on their own. Across the trials we examined, this saved nearly five months in start-up time.

We are able to select 100% of the needed sites by week 30, on average, in part because we build in a certain amount of overage in our selection process to account for non-participating sites.

The quality of our selected sites does not suffer. Indeed, the sites that we identify are 34% more likely to enrol a subject than sites that sponsors identify without ICON involvement.