Predictive Analytics Giving The Lowdown on Clinical Trial Performance
By Deborah Borfitz
May 9, 2023 | A data-agnostic integration platform for optimizing trial planning and conduct hit the streets three years ago, bringing near-real-time insights about every participant, site, country, region, and study to companies struggling to reduce the cost and speed the pace of clinical research. The technology, powered by machine learning, is currently underpinned by data from more than 2,000 trials and over 400,000 healthcare providers and is in the process of integrating other real-world data as well, according to Rohit Nambisan, CEO and cofounder of Lokavant.
Lokavant is one of many subsidiary biotech companies (so-called “vants”) that have spun out from Roivant Sciences, where Nambisan previously served as head of digital product. A neuroscientist by training, he has a broad view of the landscape with experience in academia, healthcare IT, and the startup sphere as well as in assay development at Novartis.
With the shift in focus from blockbuster drugs to more targeted therapeutics requiring many specialized and disconnected data sources, sponsors and CROs needed a nimble innovation engine to bring all that information together in one place for easy viewing and interpretation, Nambisan says. Lokavant was built for execution inside Roivant to fill that niche but emerged from it to make the technology more broadly available.
The company’s first contract was with Parexel in January 2020 and Lokavant has since tripled its customer base year over year, he continues, a list that includes Japan’s largest CRO (CMIC Group) as well as other biotech companies. Whatever data sources and vendors sponsors opt to use for a trial, Lokavant can connect them and create a normalized view of that study.
“The whole thing is highly permissionable and configurable [based on role]... for the kind of cross-functional dynamic needed to enable smarter trials,” says Nambisan. Data is updated directly from sources up to six times daily so users can see when a study is derailing and make the necessary adjustments.
Lokavant’s clinical trial intelligence platform is unlike any other analytics platforms used across the healthcare industry, he says. It leverages a growing proprietary database of anonymized clinical research data—including metrics such as enrollment rates, protocol deviations, data management, and quality issues—to create a repository of information for developing predictive and diagnostic models for all deployments and customers. That is, it can tell study sponsors and CROs why an event, good or bad, has occurred in a trial as well as when it might happen.
The platform has been validated both retrospectively and prospectively, notes Nambisan. Its predictive analytics has been found to result in a 70-fold improvement in enrollment forecast accuracy, over $1 million in cost savings from patient retention, and time savings of six months from detecting site noncompliance issues on a per-trial basis.
The hub-and-spoke business model of Roivant Sciences, formed in 2014, is responsible for scaling up a string of biotech companies. Lokavant’s maturation speed, Nambisan says, is tied to the fact that they were iterating on challenges directly faced by study teams in an industry that is notoriously risk-averse. At launch, Lokavant’s technology had already been “battle tested” in clinical research and able to generate the evidence to prove its worth.
Several white papers have been produced where Lokavant has validated its claims of time and dollars saved based either on an actual live deployment of the predictive analytics platform in a study or a simulation where time-stamped historical data is run through the clinical data hub. The focus here has been on better forecasting participant enrollment and retention, one of the biggest issues facing trials of every size the world over.
“It’s very hard to predict when a study will complete, as well as the likelihood of achieving the target accrual of participants,” Nambisan says. The company’s predictive model looks at historical data using clustering, a machine learning technique that identifies similar studies to the one about to be deployed and co-registers them based on certain common features—same therapeutic area or trial phase, for example, or conducted in the same country—to, say, learn about the impact of specific geographies and sites on enrollment rates.
Based on that repository of data, the platform builds a model predicting the success of the new study in terms of the metrics of consequence. Many customers therefore use the software for study planning as well as keeping a trial on track once it starts, says Nambisan.
At that point, data get pulled from source systems (e.g., clinical trial management, study startup, and interactive voice response) multiple times a day and get factored into the forecast, he continues. The model adapts and learns based on the incoming information, further improving prediction accuracy as the trial progresses.
In a clinical trial for a rare disease, the platform was able to foresee early on that the target enrollment of about 200 patients would be impossible to hit even with arbitrary changes made to the enrollment window by the study team, Nambisan says, citing one use case. Two years out, the prediction came within one month of when the last patient actually enrolled.
When it comes to reaching diversity goals in clinical trials, Lokavant recognizes the importance of ingesting novel, real-world data sources since study data residing in the coffers of outsourced vendors and site networks represent historically enrolled populations where minority groups were disproportionately absent. Nambisan points to the situation where Moderna recognized late in its COVID vaccine study that only 24% of participants were from communities of color, forcing the company to temporarily pause enrollment efforts altogether to course correct.
Moving forward, demographic information will be overlaid onto enrollment forecasts made by the Lokavant platform so sponsors and CROs can at any time know which study sites are overly representing specific majority or minority groups so adjustments can be made in real time, he says. This is in addition to making sure, at the planning stage, that they are including sites and investigators that have historically enrolled diverse populations.
This unbiased approach is made possible by the fact that Lokavant is aggregating anonymized data across multiple customers, a key advantage of the platform over any systems they may have developed in-house. It is hard, if not impossible, for sponsors and CROs to acquire one another’s data, points out Nambisan.
Real-time feedback is likewise available with a few clicks to understand site performance in terms of major protocol deviations and the challenges affecting investigators, sites, and regions, says Nambisan. This can be particularly problematic for rare disease trials where the elimination of datasets due to nonadherence can throw off timelines and conclusions, if not prematurely close a study.
Not all investigators who are high enrollers also deliver high-quality data, as he witnessed firsthand at both Novartis and Roivant Sciences. “It’s not just about selecting sites and investigators based on how fast they can enroll because... you are generating evidence that a therapeutic is safe and efficacious ... [and] waiting on the other end are patients with unmet needs.”
The Lokavant platform addresses similar quality issues in terms of data management, Nambisan adds, such as blank pages in case report forms. It offers geographically comprehensive as well as investigator-level quality metrics—for oncology studies conducted in Argentina, for example, how long it takes data entry to happen and to what degree protocol deviations are happening place to place. “We do this across a number of different areas, like site activation and also for safety.”
Customization at Scale
The only way Lokavant can improve the analytical prowess of the platform is with more representative data, which comes with time and growth of the underlying dataset, he says. The 2,000-plus trials now helping to inform predictions come from strategic collaborations with CROs and sponsors as well as deployments and customer relationships.
“Because our analytics are very sophisticated, we can oftentimes go in and retrain our models based on a customer’s dataset,” Nambisan explains, which reflects that company’s unique set of standard operating procedures. Lokavant can thus take a “mass customization approach” to forecasting while adding that intelligence to the repository of data benefiting the overall clinical trial ecosystem.
“Clinical research is typically not a big data problem and therefore requires small predictive data models,” he notes. Beyond clustering, the team leverages a suite of different tools that include Bayesian models to predict future trial performance.
It was only a few months ago that Lokavant raised $21 million in series A funding to further develop its platform. For the remainder of 2023, Nambisan says, Lokavant will be heavily focused on layering new types of datasets onto its proprietary database as well as extending data coverage to the underserved Asia-Pacific region. The company will also be putting more time into educating the market about clinical trial intelligence—an industry category, he notes, which Lokavant literally helped create.