Clinical Data × Systems × Automation

Proof of Work.

Selected projects exploring how standards, programming and thoughtful workflow design can make clinical-data review more structured, traceable and useful. Confidential study details are excluded or represented with synthetic examples.

Case study 01 · Standards review

CDASH Reviewer

A Python-based review pilot that compares study data dictionaries with CDASH Model v1.3, CDASHIG v2.3 and supplied terminology, then presents differences and reviewer questions with traceable evidence.

PYTHONCDASHSDTM READINESSMEDRIOREDCAP
Repository status: private. Portfolio describes the design without exposing sponsor-confidential dictionaries or licensed CDISC source files.
EDC DICTIONARY→FORMAT ADAPTER→CANONICAL MODEL→CDASH / SDTM RULES→REVIEW QUESTIONS→DECISION + EXPORT
Approach

Separate vendor shape from review logic.

Medrio XLSX and REDCap CSV inputs are normalized through adapters so downstream review logic operates on a canonical representation rather than vendor-specific columns.

Traceability

Show the evidence.

Findings retain the relevant standards and study context. Checks that cannot run are reported rather than silently treated as passes.

Human oversight

Do not automate judgment away.

Domain mappings are proposed rather than assumed, unmatched concepts require human review, and fixes are proposed only where the standard determines the answer.

Case study 02 · Study oversight

Clinical Compliance Dashboard

A Python-driven workflow concept for consolidating repeated clinical-data events into subject-level compliance, reconciliation and issue views.

PYTHONSTREAMLITEXCELRECONCILIATION
Portfolio version uses synthetic labels and examples. No sponsor, participant or confidential study data are published.
EDC EXPORT+DEVICE / ePRO→NORMALIZE→MATCH EVENTS→ISSUES LOG→DASHBOARD
Workflow

Make repeated events reviewable.

Transform event-level records into participant timelines and summaries while preserving the detail needed to investigate discrepancies.

Quality

Separate compliance from reconciliation.

Different data streams can serve different purposes; the design keeps compliance signals distinct from supporting reconciliation evidence.

Demonstration

Explore a synthetic version.

The Data Lab demonstrates the same philosophy with synthetic study metrics and issue review.

Open Data Lab →
Case study 03 · External data

External Data Reconciliation Pipeline

A structured approach to compare EDC and external records across participant, visit, date/time and expected-data context while preserving discrepancies for review.

PYTHONEXCELQCLAB / DEVICE DATA
Architecture is generalized from clinical-data workflow experience. Examples shown publicly are synthetic.
EDC↔VENDOR FILE→KEY MATCH→FIELD COMPARE→EXCEPTIONS→RECONCILED
Matching

Start with identity and context.

Participant, visit and record context are resolved before individual values are compared.

Exceptions

Make discrepancies visible.

Timing or value differences become review items with source context rather than automatic corrections.

Try it

See a reconciliation example.

The Data Lab includes a synthetic EDC-versus-lab time discrepancy.

Run example →
Case study 04 · Practical automation

Clinical Data Automation Toolkit

Small, targeted automations for recurring clinical-data work—Excel transformations, visit mapping, expected-date logic, lookup workflows, QC checks and review-ready outputs.

VBAPYTHONEXCELSASPOWER BI
The portfolio focuses on reusable patterns and synthetic examples rather than internal study files.
RAW EXPORT→VALIDATE INPUT→TRANSFORM→QC RULES→EXCEPTIONS→REVIEW OUTPUT
Excel / VBA

Meet teams where the data lives.

Use controlled macros and formulas for repeatable transformations when Excel remains the operational review surface.

Python

Scale repeatable QC.

Use configuration-driven scripts where reconciliation, validation or repeated datasets outgrow manual review.

Governance

Automation still needs evidence.

Inputs, rules, outputs, exceptions and review decisions should remain understandable to someone other than the developer.

How I think about automation

Useful before impressive.

01 · Traceable

A reviewer should be able to understand why a record was flagged and what evidence produced the finding.

02 · Human-in-the-loop

Clinical and standards judgments remain review decisions; automation should surface evidence rather than manufacture certainty.

03 · Confidentiality-safe

Public demonstrations use synthetic or generalized data structures instead of sponsor, site or participant information.