Answers are grounded in this paper and cite their sources.
Background: Clinical research and biobanking workflows require retrieval and redaction of protected health information (PHI) from clinical documents before sharing with sponsors and investigators, often alongside biospecimens. Manual redaction is time-intensive, error-prone, and may increase organizational risk through privacy exposure. Automated redaction tools could reduce human burden, improve completeness, and minimize exposure. Objective: This exploratory pilot study aimed to measure the human time burden of manual document retrieval and PHI redaction across common clinical document types, quantify accuracy based on missed PHI (false-negative) patterns, and benchmark results against a fully automated redaction system applied to the same source documents. Methods: A prospective time-motion and error characterization study with six experienced research staff were performed. Each participant redacted nine documents four discharge reports, four pathology reports, and one lab-results report. Automated time was defined as zero human time, as machine runtime does not represent human effort; missed PHI was measured using the same criteria in both conditions. Results: Six participants each completed nine documents (54 total). Mean manual session time was 058:54. Mean document-level time was ∼0:06:05, excluding Research Electronic Data Capture search (0:00:51) and miscellaneous time (0:00:48). Manual redaction produced 55 missed PHI instances (∼1.02/document); the automated system missed 3 instances across 18 unique documents (∼0.17/document), all medical staff names. Conclusion: Automated redaction was associated with fewer missed PHI instances with no measurable human time burden. These exploratory results suggest automation may reduce privacy risk and eliminate the nonredaction workflow steps comprising nearly 40% of manual session time, supporting further evaluation in biospecimen research and biobanking operations.
No takes yet. Share an insight, caveat, or question.
Familusi et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: