Longitudinal patient records
A complete record matters only if it helps someone decide what to do next. I've built the patient's chart across providers, and the products that put it to work.
- Where
- Susan G. Komen (2019 to 2023); iCAD (2023 to 2025)
- My role
- Product strategy, requirements, architecture; IQVIA partnership
- Built
- Komen Health Cloud; ShareForCures registry
- Headline result
- One consented chart per patient; $150M+ a year in data value
- ConsentThe patient authorizes retrieval
- RetrieveRecords from every provider visited, plus claims
- StructureOne model aligned to USCDI, with NLP on notes
- ServePatient portal, navigator console, researcher portal
Outcomes
- Each patient's chart assembled, with consent, from every provider they had visited, plus claims.
- One record serving three audiences: patients, care navigators, and researchers.
- ShareForCures, a national breast cancer research registry registered on ClinicalTrials.gov, estimated by outside consultants at $150M+ a year in data value.
Komen Health Cloud
The problem
A person with breast cancer often sees many providers. No one, including the patient, has the full picture of diagnoses, treatments, medications, and results in one place.
What I built
- Consent first. We designed a patient-authorized process for retrieving records from the providers each participant had visited, and consolidated them into one longitudinal view.
- Every provider, not a fixed list. Records came from whichever providers each patient had visited, digital and paper, plus claims and genomic data, retrieved through FHIR and a data broker.
- Structure from unstructured data. Records were organized into one structured model aligned to the national standard (USCDI), with natural language processing pulling diagnoses, medications, and lab results from clinical notes and pathology reports.
- Three views of one record. A patient portal; a navigator console showing medications, encounters, appointments, barriers, preferences, and recommended resources; and a researcher portal.
- Privacy by design. Patient health data stayed separate from donor and engagement data.
- Built with a partner. I owned the product strategy, requirements, and architecture decisions, and partnered with IQVIA, which built the EHR integration, the registry, and ShareForCures.
ShareForCures
The same foundation powered ShareForCures, a national breast cancer research registry: participant consent, medical records, saliva samples for genomic sequencing, de-identified data for researchers, and clinical trial matching.
iCAD: the patient's history in imaging
At iCAD, the data pipeline linked each patient's prior mammograms so findings could be compared over time: new, stable, growing, or resolved. That history is what made AI findings actionable for radiologists (see next-best-action). We grew the training data from about 7M to 20M+ mammography images from 100+ clinical sites, with full provenance for FDA review.
What I've learned
- Provenance builds trust: every fact should trace back to its source document.
- Structure what matters for the decision at hand, not everything at once.
- The record is the foundation, not the front door. Patients and clinicians come for answers and actions.
References
- Komen personalizes patient care with Health Cloud (Salesforce customer story on the patient care center platform)
- How Susan G. Komen uses technology at the intersection of nonprofits and healthcare (November 2020)