Original practice scenario: A hypothetical medical-technology support system receives the same device issue from a distributor, a service technician and a customer. Design record linkage and review.
Read model answer +
Preserve original reports and make potential linkage a separate, reversible decision. Similar descriptions can refer to one incident or several distinct incidents; deduplication must not erase evidence.
Work through it
Store source report identity, device identifier where available, event window and structured symptoms, minimizing personal information in matching features.
Use deterministic exact matches first, then flag uncertain candidate pairs with explainable evidence and reviewer-controlled merge decisions.
Maintain an incident group with member reports and linkage history, allowing split corrections and recalculating counts without rewriting the original submissions.
Tradeoffs to discuss
Aggressive matching reduces duplicate review work but risks combining independent incidents; conservative matching preserves distinctions while increasing workload.
Common pitfalls
Do not use a shared device model or generic complaint phrase as sufficient proof of the same event.
This technical workflow must not substitute for applicable medical-device reporting or clinical assessment requirements.
Original practice scenario: A hypothetical medical-technology analytics team measures new-device adoption across hospital systems. Each customer account owns several sites and each site may have multiple devices. Define useful cohort metrics.
Read model answer +
Choose whether adoption means account, site or device activity. These grains answer different questions, and a raw join can make one hospital system appear to adopt repeatedly.
Work through it
Define an eligible population and a first qualifying activation event, excluding test units and retaining the applicable account-site relationship at event time.
Build one row per chosen cohort unit, then calculate activation and later activity with explicit observation windows and a consistent time zone.
Report counts at account, site and device grain separately, preserving unknown ownership mappings and sites too recent to complete the follow-up window.
Tradeoffs to discuss
Account-level metrics describe customer reach while device-level metrics reflect deployment depth; a single percentage cannot replace both.
Common pitfalls
Do not count administrative transfers as new adoption without an explicit rule.
Distinguish missing activity telemetry from a confirmed inactive device before labeling retention.
