(A39) Federated Hospital Electronic Health Record Networks for Trial Feasibility Assessments in Rare Diseases: A Pilot Study in Neurofibromatosis Type 1

Publication Number: 322

Background: Hospital Electronic Health Record [EHR] systems, underpinned by a widely adopted Common Data Model [CDM], can inform trial protocol design, to optimize recruitment. Distributed queries executed via federated networks can provide frequency distributions of patients matching protocol eligibility criteria. Such large-scale solutions can be of particular importance in rare diseases, such as Neurofibromatosis Type 1 [NF1], where recruitment is challenging.

Objectives: To show how the Innovative Medicines Initiative’s EU-PEARL (Patient- cEntric clinicAl tRial pLatform) consortium used the Observational Medical Outcomes Partnership [OMOP] CDM and the ATLAS distributed query architecture (https://atlas-demo.ohdsi.org/) to conduct NF1 protocol feasibility assessments in an EHR system of a federated hospital network.

Methods: To identify NF1 patients with an Optical Pathway Glioma [NF1-OPG] in a hospital EHR system in the Netherlands, we created four phenotype algorithms in ATLAS, based on clinical disease specification and follow-up protocol:
D1: ‘Neurofibromatosis Syndrome’ AND ‘Neoplasm of Optic Nerve’
D2: ‘Neoplasm of Optic Nerve’
F1: ‘Neurofibromatosis Syndrome’, ‘Brain Magnet Resonance Imaging [MRI]’, AND 4 ophthalmologist visits within one year
F2: ‘Neurofibromatosis Syndrome’, ‘Brain MRI’, AND 3 ophthalmologist visits within one year
The patients selected by the four algorithms were checked against a list of known OPG patients. Cases identified by the algorithms, but not listed, were evaluated through chart review.

Results: We identified 104 NF1-OPG patients: 61 came from the pre-existing list of OPG patients; 43 patients from the ATLAS algorithms were identified as true cases via chart review. ATLAS selected 48, 76, 62, and 90 patients for D1, D2, F1, and F2, respectively. All patients selected for D1 were confirmed NF1-OPG cases. The other three algorithms selected an increasing number of non-cases (5-45). With D2, F1, and F2, we identified a further 39 true cases (23, 12, and 4). Using any of the four definitions yielded 137 patients, of which 87 were cases (positive predictive value: 63.5%). The algorithms missed 15 out of the 61 listed OPG cases.

Conclusions: OMOP CDM and ATLAS allowed identifying NF1-OPG in an EHR system. The effect of variations between phenotype algorithms was examined. For NF1, being a rare condition, we recommend directing algorithm choice towards high sensitivity. Via ATLAS, algorithms can now be shared with other sites of the network, to evaluate site recruitment potential and study eligibility criteria, and to estimate the overall NF1-OPG cohort size as an indication for enrollment planning. As the study moves closer to recruitment, additional chart review should be performed as complementary sensitivity analysis.