myTomorrows has shared new details on the methodology behind its AI-driven patient pre-screening and referral platform after releasing figures on accuracy and time savings. The company said its internal evaluation covered 490 unique patient–study matches across nine conditions, while additional performance benchmarking on Harvard’s n2c2 datasets showed state-of-the-art results. myTomorrows also clarified the publication status of its research, site-level outcomes from recent recruitment support and current adoption of its AI tools. Further information was provided on integration with hospital systems, data-protection practices, model explainability and steps to mitigate bias. The company outlined how sponsors fund the platform and described how it measures the contribution of its Patient Navigator team to recruitment efficiency.

You say 98% eligibility accuracy and “90% of time saved” — can you share the methodology and dataset behind these numbers? Who validated them?

These figures are based on an internal evaluation of 490 unique patient-study matches across nine conditions: ALS, Type 1 Diabetes, Duchenne Muscular Dystrophy, Urarchal carinoma, Oligodendroglioma, Anaplastic Astroycytoma, Pancreatic Ductal Adenocarcinoma and Pancreatic Cancer. The ‘90% of time saved’ statistic was measured across thousands of patient-study matches performed by myTomorrows, including by our Patient Navigators and using our AI pre-screening.

In addition, our AI pre-screening tool achieved state-of-the-art performance levels on Harvard’s n2c2 natural language processing (NLP) benchmark. On Harvard’s n2c2 Partners HealthCare (i2b2) NLP datasets, myTomorrows’ AI tool achieved a Macro F1 of 0.91, an improvement over Stanford’s previous best of 0.81. It also delivered a Micro F1 of 0.91, nearly matching Stanford’s record of 0.93, achieving overall state-of-the-art performance on the clinical trial matching benchmark.

Have results been published in a peer-reviewed journal or presented at a major scientific meeting?

In 2023, we published our findings on ‘arXiv’ in a paper titled ‘Improving Patient Pre-screening for Clinical Trials: Assisting Physicians with Large Language Models’. The study has been cited multiple times and is indexed on Google Scholar. Our most recent tools have not yet been published a peer reviewed journal or presented at a major scientific meeting.

Can you provide a case study or site-level example of improved recruitment timelines or reduced screen failures?

A site we supported for a Duchenne Muscular Dystrophy clinical trial in the US observed a significant improvement in recruitment efficiency. By leveraging our platform and support from our Patient Navigators, the site saw higher-quality referrals which lowered the number of screen failures. This not only helped the site achieve a 35% increase in enrollment rate, but also accelerated recruitment timelines, allowing the site to meet study milestones more quickly and operate with greater efficiency.

How many of the 320 sites onboarded are actively using the AI referral management tool, versus other myTomorrows services?

To date, around 50% of our onboarded users are using the AI pre-screening tool. As the tool is still new, we’re actively working to broaden its adoption across our user base.

The platform integrates with EHRs and CTMSs — which vendors or systems are currently live, and what limitations exist?

The platform integrates with EHRs and CTMs by accepting FHIR formatted data through myTomorrows Matching API. As the exact FHIR implementation differs per hospital, some manual optimisation per connection is required.

Data protection is emphasized (GDPR, HIPAA, ISO, SOC 2). Have you had any regulatory audits or certifications specific to clinical trial use?

At myTomorrows, we focus on data protection and information security across the board rather than for a specific workstream. While not certified specifically for clinical trial use, our existing certifications cover personal data and information security on our platform. myTomorrows is not involved in running clinical trials but supports the patient recruitment process outside of the study setting.

AI transparency is a growing issue — how do you ensure explainability for sites and investigators making referral decisions?

During AI pre-screening, all potentially eligible trials are surfaced, including those the AI deemed as ineligible, ensuring that medical professional have complete oversight. Eligibility is also shown at the criterion level, including the AI’s reasoning for its eligibility assessment. When a patient is referred, both the AI pre-screened criteria and the original medical profile are shared, allowing sites to independently review each eligibility decisions.

What safeguards are in place to prevent algorithmic bias in patient matching, especially for underrepresented groups?

Eligibility assessments are based only on the medical information provided by the medical professional and publicly listed eligibility criteria. We adopt the latest LLM models from frontier model providers, which undergo continuous improvements. Before we implement a model, we carefully review the model system card which outlines, amongst other things, model bias.

Are sponsors paying for site access, or is the platform cost carried by myTomorrows? How is pricing structured?

There is no cost for individual patients, caregivers, or healthcare professionals. We provide different models for sites, which can include an additional premium service of dedicated patient navigation. The platform is designed to support sites by delivering high-quality patient referrals, AI-driven pre-screening, and secure, compliant communication tools. Platform costs are covered by myTomorrows through paid collaborations with sponsors such as BioPharma companies seeking patient recruitment services, expanded access solutions and the collection of Real-World Data. Pricing is tailored to each sponsor’s needs.

How do you measure the impact of the Patient Navigator team on recruitment and retention separately from the technology?

We measure Patient Navigation impact by analyzing initial call-referral-to-enrollment conversion rates and funnel performance. Site feedback helps assess the quality and suitability of patients referred by myTomorrows, while patient satisfaction surveys provide insight into the value of human support. These