Search protocol

Design

Targeted evidence map and thematic case synthesis, conducted as a rapid structured narrative review and focused case series. Do not describe it as systematic, exhaustive, comprehensive, epidemiological or an evidence-gap map.

Scope

• Framework period: 25 July 2013 to 22 July 2026.
• Primary focused search: 1 January 2018 to 22 July 2026. The 2018 historical backfill used a complete publication-year enumeration and PDF/OCR extraction pass, retained contextual decisions for 36 cohort rows, and a limited later-migration check. The remaining interval uses targeted discovery; a reproducible update screen covered Judiciary pages published from 15 July 2025 through 22 July 2026, but the wider interval still requires its final rerun.
• Jurisdiction: published Prevention of Future Deaths reports for England and Wales.
• Primary source: Courts and Tribunals Judiciary PFD database, report pages, attached PDFs, and official recipient responses.
• Supplementary discovery: targeted web site:judiciary.uk searches because terminology and historic metadata vary.

Inclusion criteria

Include when both apply:

1. An official PFD report is available.
2. An electronic record, information architecture, system interface, digital workflow, alert, record availability, or paper–electronic boundary is explicitly part of the circumstances, concern, or requested preventive action.

Requirement translation occurs only after inclusion. A relevant report can therefore be included even when no Defence requirement is yet derived.

Exclusion criteria

Exclude passing technology mentions; generic communication failures without an identifiable information-system component; non-clinical technology; duplicate report/response publications; and media-only claims without an official report. Retain paper or hybrid failures when an identifiable information-representation, provenance, reconciliation, workflow or boundary mechanism falls within scope.

Search concepts

Combine PFD terms with:

• Records/systems: electronic health record, electronic patient record, EHR, EPR, computer system, software, database, electronic notes, patient administration system.
• Workflow: alert, flag, task, inbox, message, referral, handover, escalation, abnormal result, laboratory, radiology, prescribing, decision support.
• Continuity: interoperability, record sharing, unavailable, registration, duplicate record, screening, recall, audit trail, retrospective entry, amendment.
• Transfer: remote consultation, ambulance record, emergency department, transfer, discharge, paper record, hybrid record.

Screening process

1. Log each query family, date, candidate URL and disposition in screening_log.csv; for the next pass also retain the literal executable query, search surface/filter and result coverage.
2. Deduplicate by official PFD reference, not page title. Where the official page shows no reference, assign a transparent UNREF-<year>-<name-slug> project report ID and never present it as an official reference.
3. Conduct title/page screening and record a provisional decision.
4. Read the full official report and apply the criteria.
5. Extract included cases into included_cases.csv.
6. Check recipient responses separately; do not count them as cases.
7. Perform a final evidence audit before releasing abstract counts.

Single-reviewer screening is a limitation. Explicit criteria, stable IDs, retained official links, decision reasons, and a final recheck provide compensating auditability. The current log is an auditable candidate/disposition record, not a fully reproducible search transcript.

The focused 2018 search audit is the exception to that final limitation for its bounded publication cohort: it records the archive boundary, extraction/OCR coverage, literal official API queries, document-level hit counts, dispositions and later-migration limitation. The 2025–2026 update audit at research/audits/judiciary-update-screen-2026-07-22.md separately records its official API boundary, literal high-specificity terms, result counts, deduplication, hashes and dispositions. Neither makes the wider 2019–2026 search systematic or exhaustive.

Current-pass disposition

The current pass logged 165 unique retained candidates, included and fully extracted 112 reports, and excluded 53. It combines the 2018 publication-cohort enumeration/extraction recorded in the search audit, 36 retained cohort decisions, targeted 2019–2026 discovery, the bounded 2025–2026 publication-date update screen and a limited check for later-migrated 2018 reports. No retained candidate remains in a pending decision state; 343 other publication-cohort rows lack a retained contextual disposition and are not treated as documented exclusions. One included PFD covers three deaths; it remains one report row. The targeted search must be rerun immediately before submission to capture newly published or newly indexed reports.

The CSV value include_interim means “included in the completed current pass”; interim refers to the release state pending the final search rerun and independent audit, not to an unresolved screening decision.

Search log initiated 14 July 2026

The CSV is the row-level disposition authority; this table is a human-readable summary. Q-008's exact executable boundary, literal terms and result hashes are retained in its audit. Earlier search_query cells preserve concise families rather than every literal query string or result set, so a final wider update must retain the stronger transcript fields.