Drug development has been vexed by the following metrics for decades: 5% of eligible patients participate in a clinical trial, 25% of sites do not enroll any patients, and 80% of trials are delayed because of patient recruitment.
Despite increased access to data (EMR, claims, registries, feasibility, CTMS, etc.) and advances in analytical techniques (that is, artificial intelligence, machine learning; AI-ML) over the past several years, the aforementioned metrics haven’t moved much in the industry. A high degree of reliability in predicting site performance (that is, patient recruitment) and, in turn, trial outcomes, specifically rate of recruitment (RoR), remains elusive. Getting these predictions correct are crucial as net present value calculations and investments in new trials are often contingent on these numbers being reliable and accurate.
The challenge to reliably predict the ability of a site to recruit patients and RoR is very complex, multifactorial problem. Several factors that can positively or negatively impact these predictions, including but not limited to:
- Competitive Landscape: Number of competing ongoing trials, approved products/standard of care at a given site and country
- Protocol Complexity: Increased complexity tends to negatively impact recruitment rates
- Target Population: How restrictive (or not) the inclusion and exclusion criteria are for a given protocol
- Benchmark Trials: How similar trials in the past have performed and how sites have performed in those trials
- Incidence and Prevalence of the Disease
- Feasibility Data from Sites
- Investigator Enthusiasm: New and innovative mechanisms of action generate increased enthusiasm compared to “me too” products. Highly unmet medical needs also tend to generate high investigator engagement.
Trial planning teams, often comprised of trial physicians, trial managers, and analysts, grapple with these variables in making key decisions related to protocol design and number of countries and sites needed to complete recruitment (calculated as RoR). The common approaches and solutions often available within a pharma company, CRO, or in the market place involve manual approaches of data mining and analysis or are point solutions specifically addressing one or two of the aforementioned variables.
For example, commercially available EMR mining products assist with protocol design and even heat mapping of site and patients but are unable to take into account the other variables that influence site selection, such as past site performance (that is, number of patients enrolled, screen failure, site activation time, etc.) or current level of competition at sites (that is, number of concurrent trials).
The reverse is also true. Past site performance by itself doesn’t support site selection and RoR prediction, as restrictive inclusion/exclusion criteria (as assessed by EMR mining) and number of concurrent trials may negatively impact the number of eligible patients at a site. Thus, these data points exist in multiple silos, presenting a challenge to the trial teams of gaining quick access to these data and impeding their ability to rapidly overlay these data to perform dynamic “if-then” analyses that enable key decisions regarding protocol design and site selection.
Furthermore, just gaining access to the data is not enough, as it leaves the end user with the “last mile” problem of lacking these datasets in analyzable formats and being usable to derive rapid insights using advanced AI-ML techniques. EMRs have become more ubiquitous in the past several years, but without the ability to mine unstructured data (that is, doctor’s notes, radiology reports), the utility of these data are highly limited. Also, just using past site performance without the insights of how specific data points will impact site performance (e.g., if time-to-site activation impacts the number of patients a site will recruit) can leave the end user with an avalanche of data and metrics to sort through to make sense of the data/metric that is most meaningful. This is a time-consuming and costly process.
In summary, the contemporary approaches in the industry offer an interesting paradox of access to plethora of data coupled with advanced analytical techniques. However, those datasets live in silos, often in non-analyzable-ready formats denying the end user the ability to leverage advances in analytical techniques to get deep insights and to perform rapid “on the fly if-then” analysis to make critical decisions that have an impact on key outcomes of protocol design and site
selection. This conundrum needs a new approach to solve this vexing problem.
A unique way to disrupt this status quo is to leverage a “platform based” approach. This entails:
- Bringing the relevant variables/data (e.g., EMR, claims, past site performance data from CTMS, feasibility data, prevalence/incidence, clinicaltrials.gov, benchmarking) sitting in multiple silos into a common data layer in an analyzable-ready format.
- Leveraging sophisticated AI-ML techniques such as Natural Language Processing for unstructured data (allowing for deeper levels of target population [inclusion/exclusion] analysis), random forest, deep learning, and other sophisticated algorithms for doing analysis on past trial, country, and site performance and delivering insights and predictions on future performance.
- Giving trial planning teams a unique and intuitive user interface to interact with the data in a dynamic fashion and run a variety of simple to complex “if-then analyses” and provide unique insights into data/metrics that enhance or inhibit country and site performance.
- Smart workflow that is designed to fit into the way work is actually performed, allowing for seamless collaboration between different members of trial teams and how a clinical trial process moves from planning to protocol design, country, and site selection.
A platform that is able to deliver on all these features empowers trial teams to make more informed data-driven decisions that are likely to positively influence country and site selection, which in turn will increase the reliability and accuracy of prediction of site performance and RoR for a trial. The business impact of increased accuracy, reliability, and efficiency can be significant in terms of time and cost savings.
Amit Gulwadi is senior vice president, clinical innovations, at Saama Technologies.