Development and implementation of cancer clinical trial patient screening using an electronic medical record-integrated trial matching system

Abstract

Objectives

Clinical trial enrolment is critical for the development and approval of novel cancer therapeutics, but patient identification and recruitment to clinical trials remains low and multiple trials accrue slowly or fail to meet accrual goals. Informatics solutions may facilitate clinical trial screening, ideally improving patient engagement and enrolment. Our objective is to develop and implement a system to efficiently screen queried patients for available clinical trials.

Methods

At Stanford, we designed and implemented a personalised clinical trial matching system, integrating our electronic medical record, clinical trials management system and a third-party software-based solution to directly connect providers with clinical research coordinators and appropriate trials.

Results

Over 3 years of a staged rollout, significant increases in clinical trial screening requests and subsequent enrolment have been observed. The total number of screening referrals increased from 20 in the first year to 236 in the third year. Enrolment related to screening referrals, the ‘conversion rate’, ranged from 16% to 26% of referred patients.

Conclusion

Clinical trial matching systems can increase awareness of available trials and provide a mechanism to increase clinical trial accrual, especially when implemented at the point of care for easy access at treatment decision points. Here, we describe the process of creating and implementing a bespoke clinical trial matching software integrated into the electronic medical record. Having validated the utility of the platform, we will focus on further efforts to drive utilisation through software features.

Key message.

• WHAT IS ALREADY KNOWN ON THIS TOPIC

  • Currently, finding a clinical trial match for patients requires knowledge of available options by clinicians or utilisation of many different independent systems. Electronic health record (EHR)-based clinical trial matching algorithms are expected to boost accrual into clinical trials.

• WHAT THIS STUDY ADDS

  • In this study, we describe an end-to-end custom matching algorithm initiated using the EHR, populated using a machine learning system to extract information from public and private clinical trials management systems to offer clinical trial options to clinicians and facilitate screening by research teams. This required development of integrations between many commonly used services, including the clinical trial database, electronic medical records data structures and an online graphical user platform. Our approach led to an appreciable increase in referrals to research teams and subsequent enrolment in clinical trials.

• HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY

  • We believe that leveraging the power of software to enhance clinical trial matching is a key aspect of next generation medical systems and anticipate that our described experience can provide guidance for the process.

Introduction

Enrolment in clinical trials is critical for therapeutic advances in the treatment of cancer. However, fewer than 10% of patients with cancer participate in clinical trials in the USA, and those groups enrolled on trials are often not representative of the population as a whole.^{1 2} The American Society of Clinical Oncology and Association of Community Cancer Centers recently released a research statement outlining the urgent need for improving trial participation, with clear recommendations for stakeholders, including sponsors and investigators.³ Technology has the potential to streamline the screening process, leveraging the electronic medical record (EMR) to increase the identification of eligible patients and provide an avenue of communication for referring physicians and site principal investigators (PIs).⁴ As there is an increasing trend for academic cancer centres to have multiple satellite sites, with PIs and treating physicians spread across disparate locations, a central trial database may facilitate identification and screening of available, relevant trials at the point of care during treatment decision points.⁵ There have been many efforts across institutions to develop and implement algorithms to help identify and screen potential patients for clinical trials, resulting in improved efficiency.⁶,⁸ More recent attempts have even used machine learning models, including natural language processing (NLP) for unstructured data and machine learning models to refine the selection. We have recently developed and rolled out an integrated system to match patients to clinical trials open at the Stanford Cancer Institute. In this paper, we describe our technology platform, including its integration into the EMR, utilisation trends following the progressive rollout of the system, and discuss the roadblocks and lessons learnt.

Methods

We customised a cloud-based platform (XpertScreen, developed by XpertDox, Birmingham, Alabama, USA), to streamline patient matching and study team referrals to ongoing clinical trials at cancer centres. This personalised trial matching system integrates with the EMR (Epic Systems, Verona, Wisconsin, USA) by passing through patient-associated data such as age, sex and primary oncologic diagnosis, as well as user authentication information, through a Substitutable Medical Applications, Reusable Technologies (SMART) on Fast Healthcare Interoperability Resources (FHIR) connection. It is instantly accessible within the patient EMR system for oncologists and other healthcare providers to initiate a clinical trial search (figure 1).

Figure 1. Workflow diagram of clinical trial screening. EHR, electronic health record.

At its core, XpertScreen leverages machine learning to navigate the complexity of trial eligibility criteria from ClinicalTrials.gov and internal datasets, transforming dense medical language into clear, actionable insights. Instead of requiring patients and clinicians to manually interpret intricate trial requirements, XpertScreen organises them into intuitive ‘Conditions’ and ‘Filters’ using a sophisticated NLP framework.

How it works

XpertScreen employs a multistage NLP pipeline to extract, classify and structure eligibility criteria, ensuring a precise and user-friendly experience (figure 2):

Figure 2. Illustrating the process behind how the clinical trial matching software works.

• Named entity recognition: It identifies critical clinical entities such as age range, biomarkers, prior treatments and lab values using medical ontologies like Systematised Nomenclature of Medicine (a standardised clinical terminology), Unified Medical Language System (a framework for integrating biomedical terminologies) and Common Terminology Criteria for Adverse Events (a standardised classification for adverse effects in clinical trials).

• Rule-based pattern matching: It detects numerical constraints (eg, age ranges) and captures key performance scores, lab values and treatment histories.

• Ontology mapping and JSON structuring: It converts extracted data into a predefined JSON (JavaScript Object Notation, a lightweight data format for structured information) schema, standardising trial eligibility details across multiple data sources.

For example, an unstructured eligibility statement—‘Has previously failed all available and suitable therapies for AML. Disease relapse or the presence of refractory disease after ≥2 cycles of chemotherapy must be documented.’—is transformed into a structured JSON format, enabling seamless filtering and retrieval (online supplemental figure 1A).

Overcoming integration challenges

Achieving this level of automation and efficiency was not straightforward. Integrating with Epic’s electronic health record (EHR) system via SMART on FHIR (a standardised Application Programming Interface (API) framework for secure access to healthcare data) required careful mapping of structured patient data to the often unstructured criteria found in clinical trials. Handling authentication, patient consent and real-time queries—while ensuring strict Health Insurance Portability and Accountability Act (HIPAA) compliance—demanded meticulous engineering and rigorous testing.

Even though FHIR delivers EHR data in a structured JSON format, direct interpretation remains a challenge. Patient conditions, for instance, are often buried in deeply nested JSON objects with multiple identifiers (online supplemental figure 1B).

To ensure accurate filtering, we developed mappings for International Classification of Diseases (ICD) codes that unify related diagnoses, such as recognising both ‘Colon Cancer’ and ‘Colorectal Neoplasm’ as the same condition for clinical searches.

Beyond EHR integration, XpertScreen also syncs with OnCore, Stanford’s Clinical Trial Management System. OnCore plays a vital role in managing protocols, tracking participants, handling financials and ensuring regulatory compliance. To keep trial information accurate and up to date, we built a robust data pipeline that automatically ingests, standardises and updates trial details. This prevents outdated eligibility criteria from affecting the matching process, ensuring patients receive the most relevant and current options.

Streamlining access with security and oversight

To make trial access seamless for healthcare professionals, XpertScreen integrates with Stanford’s Single Sign-On, allowing clinicians, research coordinators and administrative staff to log in securely with their Stanford ID. This eliminates the need for additional credentials while maintaining high security and ease of use.

While automation plays a crucial role, human oversight remains essential. Each trial’s eligibility criteria must be carefully reviewed and validated through ongoing collaboration with PIs, research coordinators and regulatory teams. Because eligibility language varies across specialties and institutions, striking the right balance between automation and expert curation is key to maintaining accuracy and reliability.

Precision matching for better outcomes

XpertScreen categorises clinical trials based on key factors, ensuring precise patient-trial matching:

• Cancer type: It matches patients to trials based on specific diagnoses (eg, breast, lung or other cancers).

• Cancer stage: It aligns trials with disease progression, from localised to metastatic.

• Surgical operability: It distinguishes trials for resectable and non-resectable tumours.

• Therapeutic history: It considers prior treatments to determine eligibility.

• Tumour-associated mutations: It identifies trials targeting specific genetic markers.

By bridging technical, operational and clinical gaps, XpertScreen is redefining how clinical trials are matched to patients, accelerating the journey from diagnosis to treatment. This innovation brings precision oncology closer to reality, empowering patients with access to cutting-edge therapeutic opportunities tailored to their unique medical profiles.

In practice

Once activated by a provider in the EMR, the matching system automatically imports comprehensive patient data, including demographic information and cancer stage, type and mutations. The user is prompted to verify and potentially edit the patient’s information, as needed. Using smart prompts, the user can enter such information in less than 15 s. The system stores patient’s information thereafter for future search, further improving the user experience. After the patient’s information is finalised, the system creates a rank-ordered list of potential trial options at the site, starting with the best match at the top, and always including at least one trial (figure 3A). From the list provided, the provider can select one or more eligible trial options for the patient and click the ‘Request Screening’ button to initiate a private health information (PHI)-secured email communication thread connecting the referring provider with the primary clinical research coordinators (CRCs) and PI (figure 3B). From a workflow perspective, CRCs and PIs have been requested to prescreen the patient within 48 hours based on verifying all information in the patient’s records in the EHR. If the patient is found to be eligible based on prescreening, the research team proceeds with consent and formal trial screening. The matching system strictly adheres to privacy standards, ensuring PHI/HIPAA compliance, SOC2 Type 2 certification, ISO:27001 certification and robust encryption for data both at rest and in transit.

Figure 3. (A) The matching system is integrated within the right-hand pane of the EMR. (B) The matching system screening request form uses secure email communication. EMR, electronic medical record.

All information on the matching system is updated two times a week, including updating the number of trials open or closed status and CRC contact information, thus minimising erroneous matches. Additionally, the matching system features an administrative console that grants the Clinical Trials Office administration the ability to manually adjust Conditions and Filters criteria—a feedback system through the matching system allows requests to be sent and tracked. Furthermore, the matching system records user logins, screening requests and, in the future, will allow follow-up of screening completion and assessment of platform utilisation.

Results

Rollout and utilisation

The platform was unveiled to providers during the second quarter of 2021 in a soft go-live. The EMR access link and tutorial slides were presented to cancer clinical research leaders and faculty physicians to gather feedback on the platform while it was being refined and improved. Without mandatory requirements to access the platform, we observed an organic 26% yearly increase in the number of platform users (figure 4A). The number of sessions launched per user nearly tripled between 2021 and 2023 from 3.7 launches per user to 8.8 launches per user. Initially, users selected a trial in 79% of the sessions, but sent a referral in only 21%. Over subsequent years, the percentage of platform visits that resulted in screening referrals increased to 63% in year 2, then 72% in year 3 (figure 4B). The total number of screening referrals increased from 20 in the first year to 166 in the second year and to 236 in the third year (figure 4C).

Figure 4. (A) Unique users per calendar year. (B) Percentage of matched trial queries that were submitted to the clinical trial team for patient screening. (C) Screening requests sent to the study team by quarter. (D) Participants consenting to study within 60 days of screening referral. (E) Conversion rate: percentage of participants registered and treated on the matched referred study. (F) Indirect referrals within 60 days of referral. Indirect referral is defined as a patient who did not enrol on the matched study but enrolled on to another clinical trial.

Trial matching

The ‘trial matching rate’ is defined as the number of participants that signed consent for a clinical study within 60 days of a screening request, compared with the total screening submissions in the matching system. In 2022, 43 patients signed a consent form for a study, which resulted in a 26% (43/166) matching rate of the platform. 16% (37/236) of referred participants in 2023 signed a consent form by mid-February of 2024 (figure 4D).

The ‘conversion rate’—defined as a consented patient registering and treated with at least one study intervention—increased yearly from 60% in 2021 to 86% in 2023 (figure 4E). While some patients did not consent to any specifically matched studies, they signed consent to a different disease-relevant clinical trial within 60 calendar days of the initial referral (indirect referral). The percentages of these indirect referrals, 11%, were the same in the second and third years (figure 4F).

Discussion/future directions

Maximising enrolment in clinical trials is a critical step for successful cancer clinical research at academic medical centres. We integrated a matching system to increase clinical trial awareness and screening requests with the hope of increasing accrual across the enterprise. By integrating this software directly within the EMR, we offer a fast and simple approach to search for available clinical trials and send referrals for screening at the point of care where treatment decisions are being made. Integrating local clinical trial information directly into the EMR workflow has multiple advantages. First, it increases visibility into various actively accruing trials across cancer groups. For matrix cancer centres with subspecialised oncologists and network general oncologists, most of the clinical trial infrastructure is concentrated in specific cancer research groups, and visibility into those trials is not easily found across the network. Second, it streamlines the process of assessing a patient for clinical trial eligibility. In a busy oncology clinic where treatment decisions sometimes need to be made rapidly, it can save time to automatically load patient information, filter relevant trials instantly and establish communication with multiple PIs and their research teams without searching for contact information. Because of this, many providers—even the PIs of trials—find the matching system useful while in clinic.

Since the initial rollout of the matching system 3 years ago, we have seen substantial growth of providers using the platform, an increase in the number of sessions resulting in a referral and subsequently an increase in conversion rate. These results indicate that there is a substantial interest among all stakeholders to ensure that the patients at Stanford Cancer Institute (SCI) are enrolled into clinical trials with the goal of improving care for both current and future patients. Over time, there was some decrease in trial matching rate which could be attributed to the fact that initially providers were handpicking which patients they were referring to study teams. Overall, by providing an interface that was easy to use and efficient, we were able to tap into this unmet need among patients, providers and researchers to boost clinical research.

In order for these results to be generalisable, we want to highlight the active work that the clinical research team did in promoting the use of the platform. Outreach included provider champions and programme managers presenting information at town halls and meetings during the rollout. In addition, we created a new ‘Research Timeout’ in the non-research chemotherapy ordering process and are now requiring users to select a reason that a patient is not on a clinical trial, with a link to the matching system to help with the search process. We anticipate this will create a culture across the network of considering clinical trials before changing therapy, rather than as an afterthought. Real-time data tracking and visualisation dashboards may be created to discover any new trends in clinical trial practice, as well as use them to inform the expansion of our trial portfolio and choose network satellite sites for enrolment.

There have been multiple cancer centres that have also built and implemented a software-based solution to improve clinical trial matching.⁹,¹¹ One study at the Mayo Clinic was a feasibility study to evaluate whether a clinical trial matching solution could accurately determine trial eligibility as compared with manual review by trained nurses. They found that accuracy varied by disease type (breast vs lung) but in general was at least 75%. Another group at Virginia Commonwealth found that there was significantly improved efficiency and decreased research staff time burden with an automated clinical trials eligibility screening tool. The seismic innovation in the use of generative AI and large language models has also been recently applied to clinical trial matching in an algorithm called TrialGPT, which resulted in an astounding 87% accuracy and also reduced screening time by 42%.¹² In general, automated cancer screening and matching tools have proven to be accurate and effective in matching patients to potential clinical trials and can potentially improve the productivity of research staff.

While building and implementing the matching system, we encountered several issues that became important learning points. Our initial attempts at matching resulted in too many trial results that were not relevant to the patient. We realised that we needed to be strict on matching cancer type. This was challenging because some patients did not have the correct ICD code in the chart or had a history of multiple cancers, so the tool could match based on inactive cancers. In a sense, this is reflective of the messy nature of real-world data in the EMR which is not routinely verified nor audited. To attempt to fix this, we made a single selection of histology a critical first step. Also, we made it an option for providers to send a general referral message to a cancer research group (multiple PIs and CRCs all included on a single email) if a specific matching trial was not found. Second, regarding communication with clinical trial teams, we considered whether messages via EMR or secure email would be preferable. We decided that emails sent to the PI and cc’ing the clinical research staff would be the best option to facilitate open communication,¹³ but integration within the EMR, including follow-up to ‘close the loop’ is a future consideration. This was mainly because most of the PIs and clinical research staff communicate mainly by email while EMR messaging was primarily for direct patient clinical care.

Conclusion

In summary, we built and implemented a customised clinical trial matching system directly into our healthcare EMR. A significant increase was observed in clinical trial queries and patients enrolling in trials on implementation. We are continuing to optimise the platform and anticipate an increase in clinical trial accrual as well as other improvements in the quality of cancer clinical trial research over the upcoming years.

Data availability statement

Data sharing not applicable as no datasets generated and/or analysed for this study.