Sketch a Healthcare Data Pipeline and Trace How Records Flow From Source to Consumer

You are interviewing for a product manager role, and this round includes a technical question. Pick a healthcare use case, then sketch the data pipeline that would support it: draw its components as boxes and arrows, and walk the interviewer through how data travels from the systems where it originates, through each stage, to the people or applications that use it. The interviewer does not name a use case, so choose one and state it before you start drawing. ```hint Start from the consumer Decide who reads the pipeline's output and how fresh it has to be before you pick components. Then mark the points where patient-identifying data enters, changes shape and becomes visible to someone. ``` ### Clarifying Questions - Which use case should the pipeline serve: hospital records feeding a population-health dashboard, readings from home monitoring devices, or data gathered during a clinical trial? - How fresh must the output be: a report refreshed overnight, or an alert within minutes of a new reading? - Which source systems and formats are involved, for example an electronic health record exposing HL7 or FHIR interfaces, insurance claims files or device data feeds? - Which privacy rules apply (HIPAA, for a US provider), and may analysts work with identified patient data or only de-identified data? ### What a Strong Answer Covers - A concrete use case and set of consumers, with freshness and volume requirements that visibly drive the design - Distinct stages from ingestion through raw storage, validation and transformation, curated storage and serving, with the data flow between them made explicit - Healthcare-specific handling: standard formats and code sets, matching the same patient across systems, consent, de-identification, access control and audit logging - Reliability and data quality: validation, duplicates, late-arriving records, reprocessing and monitoring - Product judgment: trade-offs such as batch versus streaming or managed versus self-built components, and how to tell whether the pipeline is serving its users ### Follow-up Questions - A nurse reports that a patient's reading showed up on the dashboard a day late. How would you find where the delay happened? - How would the design change if the same data also had to train and feed a model that predicts which patients are at risk of readmission? - An analyst asks for raw identified records to debug a metric. How do you respond, and what would you build so the request is not needed next time?