Lock Readiness Dashboard & Burndown

Why this matters

Database lock dates slip when teams track the wrong signals (for example, “queries closed last month”) instead of the current state of data needed for analysis. A clear, shared dashboard aligned to pre‑agreed lock gates lets teams see issues early, assign the right owner, and lock within days of LPLV (Last Patient Last Visit) when prepared.

Plain‑language glossary used on this page

• EDC: Electronic Data Capture system where sites enter CRF data.
• CRF: Case Report Form—the study’s data collection instrument.
• LPLV: Last Patient Last Visit—the last visit of the last subject in the study.
• Lock gate / KPI: A lock gate is a pre‑agreed condition to be met before database lock; a key performance indicator (KPI) measures how close each gate is.
• Expectedness: The percentage of data that should be present (based on the protocol schedule) that is actually present in EDC or external feeds.
• Presence vs value: Presence = the record exists at all; value match = the values agree between sources (e.g., lab vs EDC).
• MedDRA / WHODrug: Coding dictionaries: MedDRA for adverse events and medical history, WHODrug for medications.
• Adjudication: Independent committee decisions on endpoints.
• PV: Pharmacovigilance safety system/source.
• Owner segmentation: Showing metrics by the team responsible to fix (site, clinical research associate (CRA), vendor, data management (DM)).

Core KPIs & thresholds

Focus on analysis‑critical domains and processes: the study‑specific critical data, named up front from the protocol, the statistical analysis plan and the risk assessment (e.g., ocular history in an ophthalmology trial, which drives eligibility and baseline risk). A general figure such as total open queries does not show whether these are clean. Targets below are typical starting points—tighten or relax based on your study’s risk and endpoint timing.

Tip: Track inflow vs closure for queries and reconciliations. A backlog that shrinks every week with controlled inflow is on course for lock; a flat or growing backlog signals risk even when weekly closures look high.

The process, step‑by‑step

1. Define gates early in the Data Management Plan (DMP). List each lock gate (data present, queries, source verification and investigator signatures, reconciliations, coding, adjudication, derivations). For each, specify KPI, threshold, data source, refresh cadence, and owner.

2. Design owner‑segmented views. Every chart/table should be sliceable by owner (site/CRA, vendor, DM analyst). This makes action routing explicit. The owner is whoever can make the fix: the site corrects its EDC data in answer to queries, a vendor corrects and re‑sends its own data, PV corrects the safety case; DM raises and closes queries but never edits site data.

3. Publish a weekly burndown.

   • Start 8–12 weeks before anticipated LPLV.
   • Show starting backlog, weekly inflow, weekly closures, and weeks to target.
   • Forecast the lock date by projecting when backlogs hit thresholds given recent closure rates and inflow. Net reduction per week = closures minus inflow; weeks to target = backlog ÷ net reduction. For example, 120 open queries with 50 closed and 20 new each week fall by 30 a week: about 4 weeks to zero. Forecast per owner; the slowest owner sets the earliest realistic lock date. In words: “At your current net reduction per week, you will reach zero criticals in ~X weeks.”

4. Run an exceptions review. Each week, list the top 10 items blocking thresholds (e.g., “Site B AE pages missing for Weeks 12–16”; “Labs Week 20 panel value mismatches”). Assign dates and named owners.

5. Dry‑run the lock (rehearsal). Execute a mock data cut and confirm KPIs hold. Validate no late feeds or unplanned derivations are required.

6. Freeze and communicate. Freeze dictionary and derivation versions, and publish a no‑surprises note: what will and will not change before lock, with dates.

What good looks like

• At‑a‑glance RAG (red/amber/green) for each gate with owner initials and ETA to green.

• Trends for the last 6–8 weeks: expectedness %, median query age, %>14 days, and count of reconciliation presence gaps.

• Owner‑segmented tables showing:

  • Top sites by aging, with last activity date and CRA contact.
  • Vendor feed timeliness (files on schedule) and unresolved mismatches.
  • Coding completion and any terms pending medical review.

• Notes panel: for every red/amber, a one‑line plan (“CRA to host query‑thon with Site B on Tue; lab vendor to re‑deliver corrected Week 22 file by Fri”).

Worked examples

Self‑contained scenarios with the context, plan, and outcome.

Example 1 — Cardiovascular (CV) trial nearing LPLV

Context

• Population: adults with hypertension.
• Critical domains: Vital Signs (VS), Adverse Events (AE), Concomitant Medications (ConMeds), central Labs.
• Today is 6 weeks before expected LPLV. Figures are for data due to date; every gate is re‑checked on the final data after LPLV.

Dashboard snapshot (Week −6)

• Expectedness: VS 97%, AE 97%, ConMeds 98%.
• Queries: 0 criticals; median age 12 days; 18% older than 14 days (clustered at Site B).
• Reconciliations: 3 presence gaps in lab files (missing Week 20 panel for three subjects).
• Coding: 99% coded; 1 term pending medical review.
• Adjudication: not applicable (no adjudicated endpoints).
• Derivations: final algorithm list approved; 1 open change request to rename variables.

Plan

• Site B coaching: CRA schedules a 2‑hour virtual session; DM pre‑flags the 25 oldest queries and rewords unclear ones without suggesting the answer; the site answers from source.
• Missing pages: DM sends each site its list of expected‑but‑missing VS and AE pages; the CRA follows up and the site enters them.
• Lab vendor: request re‑delivery of Week 20 file with missing panels; confirm the delivery date in writing.
• Coding: medical reviewer to finalize the 1 pending term by Friday.
• Derivations: convert the rename request into a controlled change; decision by governance this week.

Outcome (Week −2)

• Expectedness: VS/AE now 99%; ConMeds 99%.
• Query aging: median 8 days, 4% older than 14 days.
• Reconciliations: 0 presence gaps; no value mismatches in the re‑delivered file.
• Coding: 100% coded & QC’d, dictionaries frozen.
• Derivations: versions frozen; no open changes.
• Decision: Proceed to lock rehearsal next week.

Example 2 — Device study with adjudicated endpoints

Context

• Implantable device, 12‑month primary endpoint; Endpoint Adjudication Committee reviews events.
• External sources: imaging core lab, randomization system, PV safety database.
• Today is 8 weeks before expected LPLV; as in Example 1, figures are for data due to date.

Dashboard snapshot (Week −8)

• Expectedness (critical domains): Imaging 95% (lag from two sites); AE 98%.
• Queries: 6 criticals open (eligibility and device‑related AE terms).
• Reconciliations: Randomization presence 100%; Imaging value mismatches 7 (measurement units).
• Adjudication: 12% of required decisions pending—committee backlog.
• Coding: 96% coded; dictionary version not yet frozen.

Plan

• Imaging core lab: agree on unit normalization and re‑issue values; commit date within 10 days.
• Adjudication backlog: schedule an extra adjudication meeting (quorate under the committee charter) to clear pending decisions.
• Critical queries: fast‑track eligibility queries—DM rewords the queries so they are clear and non‑leading; the site answers from source; CRA escalates non‑responsive sites.
• Coding: freeze dictionary version at the next data cut; complete coding/QC.

Outcome (Week −3)

• Imaging expectedness 98%; value mismatches 0 after unit alignment.
• Critical queries 0; aging tail within threshold.
• Adjudication 100% complete after the extra meeting.
• Coding 100% & frozen.
• Decision: Lock rehearsal passes; the team confirms a lock date in the week after LPLV, once the last visits are entered, cleaned and signed.

Pitfalls & misconceptions

• Mistaking effort for readiness. Reporting “queries resolved last month” is historic; current open and aging reveal real risk.
• Treating all data as equally critical. Focus on study‑specific critical domains (what can shift endpoints, eligibility, or safety).
• Blending owners. Mixing site and vendor issues hides accountability; segment by owner.
• Ignoring inflow. Backlog can grow even when closures are high—track new vs closed weekly.
• Late dictionary or derivation changes. Freezing late forces rework and threatens consistency across TLFs.
• Assuming presence = quality. Reconciled value matches matter as much as presence.

Practice in the wild (checklist)

• Define lock gates, thresholds, data sources, refresh cadence, and owners in the DMP.
• Build owner‑segmented dashboards (site/CRA, vendor, DM).
• Publish a weekly burndown with inflow vs closure and a plain‑English forecast to threshold.
• Run a top‑10 exceptions review every week with dates and named owners.
• Freeze dictionaries and derivation versions 2–4 weeks pre‑lock.
• Conduct a lock rehearsal and confirm KPIs hold on a mock cut.
• Document rationale for any exceptions; secure sign‑off.

Key points to remember

• Measure the current state, not past effort. Readiness = what’s missing or still open today (expectedness, open/aging queries, recon gaps), not how much was closed last month.
• Focus on study‑specific critical data. Tie all KPIs to data that can move endpoints, eligibility, or safety; don’t treat every domain equally. Critical fields (e.g., ocular history in an ophthalmology trial) are verified against source before lock.
• Expectedness drives visibility. Track the % of required pages/rows present for analysis‑critical domains (e.g., VS, AE, ConMeds, Labs); aim for lock‑ready levels (≈98–99% near lock, every remaining gap documented).
• Presence and value both matter. Reconciliation should confirm presence (rows/files delivered) and value agreement (numbers/terms align across sources).
• Queries: zero criticals and controlled aging. Reach 0 open critical queries first and close every query by lock (exceptions documented and approved); in the run‑up keep median age ≤ 14 days with only a small tail older than that; monitor inflow vs closure to spot rising risk.
• Coding is complete and frozen. Achieve 100% coded & QC’d with frozen MedDRA/WHODrug versions to avoid post‑lock rework.
• Adjudication is finished. No “pending” decisions at lock for any event the charter requires, primary endpoints above all.
• Derivations are version‑locked. Algorithms and parameters are finalized with no open change requests; freeze before the final cut.
• Segment by owner for action. Build views by site/CRA, vendor, and DM so accountability and routing are explicit.
• Use a plain‑English burndown. Show backlog, inflow vs closure, and a simple forecast to threshold using recent net reduction (closures minus inflow).
• Validate with a lock rehearsal. Run a mock cut to prove KPIs hold under real conditions.
• Aim to lock within days of LPLV. When expectedness, queries, source verification and signatures, reconciliations, coding, adjudication, and derivations meet thresholds, locking within a few days of LPLV is realistic.