Laboratory System Replacement Forces Choice Between Vendor Views and Analytical ODS

Direct views are cheaper initially but create long-term coupling and do not solve the semantic problem. The vendor model is normalized for transactions, not for reporting; 45 reports would each encode their own joins and patient logic. Performance may be acceptable for small reports but not for near-real-time clinical review. Vendor upgrades could break views, and the team cannot modify the vendor schema.

The ODS plus dimensional model separates operational and analytical concerns. The ODS integrates and historizes data without changing the vendor model; conformed dimensions and facts support finance and clinical reporting. It also gives one governed patient view, which direct views do not.

Recommendation: a hybrid with a bias toward the ODS.

• Use the ODS for near-real-time result review and for any report needing cross-system patient identity.
• Build a small dimensional mart for finance cost reporting, where monthly latency is acceptable.
• Allow direct views only for low-risk, low-complexity operational lookups with a documented dependency and upgrade test.

With 14 weeks, the team cannot build everything. The ODS should start with patient, order, result, and provider, and the first dimensional mart should cover finance's highest-value report. The 900-hour rebuild estimate is a signal to rationalize reports, not to copy the old flat model into views.

Do not carry the old patient identifier forward as the primary key. Reuse means the same value maps to different people, so using it as a durable key would merge unrelated patients and create clinical safety risk.

Modeling approach:

• Create an enterprise patient identifier independent of source systems. Use a master patient index or matching service to link source records.
• Maintain a crosswalk from old system ID plus effective dates to the enterprise ID. The same old ID value can map to multiple enterprise IDs in different time periods.
• Separate patient and encounter: patient is the person; encounter is the visit or lab episode. Reused identifiers often conflate the two.
• Add data quality rules that flag duplicate or reused identifiers and route them to review before clinical use.
• Preserve history: do not overwrite prior links when a correction is made; use effective dating and audit.

The CMIO's 'one patient view' should mean one governed identifier with transparent links, not a forced one-to-one mapping. Clinical reports must show when identity is provisional. This is a modeling and governance decision, not only an ETL fix.

Produce artifacts that force agreement on semantics before developers rebuild 45 reports:

• Report inventory and rationalization matrix: list each report, owner, decision use, frequency, and disposition (retain, merge, retire). This can cut the 900-hour estimate more than any technical shortcut.
• Bus matrix: business processes against conformed dimensions. It exposes overlap and missing dimensions early.
• Conformed dimension list: patient, provider, department, lab test, date, and payer, with definitions and source systems.
• Logical data model for the analytical layer, including patient-encounter relationship and identifier crosswalk.
• Source-to-target mappings for the first ODS release, with transformation rules and data quality checks.
• A three-report prototype covering the highest-risk clinical and finance use cases, reviewed by clinicians and finance before full build.

These artifacts are lightweight enough for six weeks and create the sign-off points the project lacks. Without them, report developers will rebuild the old flat model in the new system and repeat the identity problem. The modeling lead's views proposal should be evaluated against the same artifacts rather than debated abstractly.