MatchMiner: an open-source platform for cancer precision medicine

Widespread, comprehensive sequencing of patient tumors has facilitated the usage of precision medicine (PM) drugs to target specific genomic alterations. Therapeutic clinical trials are necessary to test new PM drugs to advance precision medicine, however, the abundance of patient sequencing data coupled with complex clinical trial eligibility has made it challenging to match patients to PM trials. To facilitate enrollment onto PM trials, we developed MatchMiner, an open-source platform to computationally match genomically profiled cancer patients to PM trials. Here, we describe MatchMiner’s capabilities, outline its deployment at Dana-Farber Cancer Institute (DFCI), and characterize its impact on PM trial enrollment. MatchMiner’s primary goals are to facilitate PM trial options for all patients and accelerate trial enrollment onto PM trials. MatchMiner can help clinicians find trial options for an individual patient or provide trial teams with candidate patients matching their trial’s eligibility criteria. From March 2016 through March 2021, we curated 354 PM trials containing a broad range of genomic and clinical eligibility criteria and MatchMiner facilitated 166 trial consents (MatchMiner consents, MMC) for 159 patients. To quantify MatchMiner’s impact on trial consent, we measured time from genomic sequencing report date to trial consent date for the 166 MMC compared to trial consents not facilitated by MatchMiner (non-MMC). We found MMC consented to trials 55 days (22%) earlier than non-MMC. MatchMiner has enabled our clinicians to match patients to PM trials and accelerated the trial enrollment process.

Structuring clinical pathways: Creating common language where there is none

303

Background: Clinical pathways (PW) consist of Decision Criteria (DC) such as patient and disease characteristics, prior therapy, and genetic tests. Historical Dana-Farber Pathways (DFP) content was unstructured and maintained in static documents. This could lead to inconsistencies across and within PW, limits to the scope of DFP analytics, and potential discrepancies in clinical content. Methods: To transform DFP content into a digitally innovative structure the DFP team created a hierarchical Data Model (DM). The team compiled all unstructured DC in historical PW, organized them into parameters and attributes, and connected them to external ontologies (e.g., ICDO3) where appropriate. The team then applied the structured DC to historical PW to test comprehensiveness; and addressed any gaps identified in PW and the DM. Results: The DPF DM contains 32 parameters (e.g., Diagnosis) and 218 attributes (e.g., Group Stage) that can be combined to represent all 600+ pathway branch points. The comprehensiveness and nuance of the DM improves DFP’s specificity and clinical flexibility: The DM ensures that DFP gathers actionable data across all PW and creates common language that allows for disease-specific nuances; The DM rectifies gaps in historical PW, such as DC that were not mutually exclusive or conflated multiple clinical parameters; The DM creates complex DC to direct specific sub-groups of patients to the correct treatment path, even in cases where treatment recommendations differ within diagnostic groups. Conclusions: Structuring and standardizing PW content is a complex, time-intensive endeavor. However, this work addresses the challenges of managing clinical content and provides significant benefits for future PW development. A structured DM ensures PW are comprehensive, logical, and built on a framework of inter-operability standardization. A DM also allows DFP to connect with the EMR for auto-navigation, which streamlines provider workflow. The ongoing work required to build and maintain a structured PW DM is worth the result: rich actionable data that can illuminate and standardize practice patterns within and across diseases and institutions. It creates an insightful solution to broadly manage the cancer population.

Everyone needs robust analytics: Integrating medical ontologies into the Dana-Farber Pathways

304

Background: The majority of clinical data is unstructured. Medical ontologies standardize the language used to communicate diagnoses, symptoms, and procedures used in healthcare. Incorporating medical ontologies into the Dana-Farber Pathways (DFP) provides structure and removes ambiguity by creating definitions for each data point. This exercise was critical for the development of the DFP into a fully automated platform that optimizes clinician workflow, reduces clicks, and expands the potential for data analysis. Methods: The DFP data model was reviewed to identify value sets contained in an existing medical ontology. DFP required employing multiple ontologies, which was difficult, as many ontologies overlap (e.g., SNOMED, NCI, ICD-O-3, ICD-10) and none were developed for this intention. For example, the majority of DFP necessitate site/histology as a criterion for decision-making. In our Electronic Data Warehouse (EDW), site/histology are stored as free text, as ICD-O-3 codes from the cancer registry, and as ICD-10 codes from Epic. To obtain site/histology in a structured format, both ICD-O-3 and ICD-10 were needed. Ongoing maintenance of the linkages between ontologies and value sets is required as ontologies are updated and DFP are enhanced. Results: The relationship between the DFP value set and the SNOMED ontology was 1:1. However, other relationships were not equal or straightforward. Some ontology values were too detailed for DFP. These were either grouped together (e.g. tumor site) or just one value was chosen (e.g. drug class). Others didn’t contain all the values required by DFP (e.g. regimens). Conclusions: DFP requires structured data related to ontologies but not identical to and not limited by the scope of ontologies. Choosing medical ontologies utilized in an EDW facilitates the implementation of auto-navigation, which ultimately streamlines provider workflow. A structured data model connected to ontologies collects valid and clearly defined data to support robust analytics.

[Table: see text]

Abstract 2288: Assessing patient trial readiness for precision cancer medicine clinical trials through computational trial matching and rule-based logic applied to genomic and clinical data

More than 1,100 phase I treatment trials are registered with clinicaltrials.cancer.gov. Patient identification and recruitment to these trials is particularly challenging due to the vast molecular data now available for patients and the increasingly complex nature of the molecular eligibility criteria. Additionally, Phase I trial staff often manage a portfolio of dozens of trials, making patient identification, recruitment, screening, and enrollment an arduous task. To address the need for better and more efficient patient identification, we have developed MatchMiner, a novel open source computational platform for matching patient-specific genomic profiles to precision cancer medicine clinical trials, and a pre-screening tool that includes options for filtering and sorting patients based on clinical data elements. MatchMiner excels at identifying patients who are eligible for clinical trials based on genomic, cancer type, age, and gender criteria, but the great majority of patients are not currently ready for a trial. The pre-screening tool allows for the integration of additional clinical data, including genomic testing date, appointments, treatment plan history, and radiology scan impression text, in order to provide an assessment of ‘trial readiness'. To integrate these tools into the clinical workflow, we have implemented a weekly MatchMiner Tumor Board Review process, where together with the DFCI Early-Phase Drug Development Center (EDDC) physicians and staff, a thorough assessment of potential patient-trial matches is performed. Currently we have scaled this initiative to include 18 genomically-driven trials. Each week, ~9000 patient-trial matches are computed across all open trials, which are then computationally filtered to a list of ~300 patient-trial matches utilizing rules-based logic applied to our clinical data sources. The MatchMiner team then performs a manual review of pre-filtered matches based on additional requirements, such as evidence of progression in the radiology scan text impressions. The final list of ~10-20 patients is reviewed with the EDDC, and patients deemed to be ‘trial ready' are flagged. E-mail notifications are then sent out to the patient's oncologist, alerting them to a potential genomically-driven clinical trial opportunity. The combination of computational patient-trial match identification and rules-based computational pre-filtering based on additional clinical data has allowed us to screen thousands of patients per week, which would not be possible otherwise. To evaluate the success of this initiative, we are tracking patient consults and trial enrollments, and capturing feedback on workflow integration and utilization. Additionally, efforts to further automate trial readiness measurements to minimize the need for manual review are ongoing.

Citation Format: Catherine Del Vecchio Fitz, Khanh Do, James Lindsay, Suzanne Hector-Barry, Zachary Zwiesler, Priti Kumari, Tali Mazor, Tamba Monrose, Adem Albayrak, John Methot, Michael Hassett, Geoffrey Shapiro, Ethan Cerami. Assessing patient trial readiness for precision cancer medicine clinical trials through computational trial matching and rule-based logic applied to genomic and clinical data [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2018; 2018 Apr 14-18; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2018;78(13 Suppl):Abstract nr 2288.