Boost Recall: Qualitative Analysis of Patient Narratives

Evidano is an AI-powered qualitative data analysis platform that automates transcription, thematic coding, and cross-segment analysis.

Lee et al. (published July 20, 2026) found that recalling a validating clinical consultation increased the odds of recalling brief audio health messages by about 19% per item in a final sample of n=245.

This post shows how to reproduce the study’s item-level recall workflow and how to scale transcript, audio, and narrative analysis into reproducible thematic and modeling pipelines.

Lee et al. (published July 20, 2026) found that people with chronic pain who wrote about a validating clinical consultation were more likely to recall short audio health messages than those who wrote about invalidation.

The effect was about 19% higher odds per item in the final sample (n = 245). See the paper: PLoS ONE.

• Design: quasi-random assignment to a validating vs. invalidating autobiographical recall task; incidental recall of 20 audio health messages.
• Key outcome: validation → ~19% higher odds of recalling each message (robust in randomized subset n = 186).
• Why researchers care: clinician validation appears to free cognitive resources, improving retention of advice.

Participants living with chronic pain wrote for 3+ minutes about either a validating or an invalidating healthcare consultation, then listened once to 20 brief audio health messages and later completed an incidental recall task.

Researchers scored recall with a rule-based keyword algorithm validated against manual coding (97.5% agreement). Analyses used binomial GLMMs with random intercepts for participants and items to model per-item recall.

• Primary predictor: perceived validation vs. invalidation (autobiographical recall).
• Covariates considered: pain intensity, pain-related fear (PASS-20), immersion, age, sex, pain duration.
• Robustness: results held when restricted to fully randomised subset; mediation tests found no indirect effect via pain-related fear.

This seven-step workflow reproduces the study pipeline.

Step 1 Collect: recruit and capture autobiographical responses and audio health messages; ensure attention checks and consent.

Step 2 Transcribe: auto-transcribe audio with custom dictionary, review and redact PII.

Step 3 Preprocess: normalise text, remove duplicates, flag AI-like responses for manual review.

Step 4 Code: build a codebook for validation/invalidation markers; run AI-assisted coding then refine with human QC.

Step 5 Link: align item-level recall responses to the 20 message codes (rule-based or model-assisted scoring).

Step 6 Analyse: run per-item binomial GLMMs (retain random intercepts for participants and items); explore mediators.

Step 7 Visualise & share: generate co-occurrence maps, segment comparisons, and export reproducible reports for stakeholders.

This study is research-focused and not diagnostic.

When working with patient narratives include consent, de-identification, and ethics oversight, especially if you plan clinician-facing interventions.

Small communication changes can yield measurable gains in patient recall.

Lee et al. (July 20, 2026) report approximately 19% higher odds after recalling a validating consultation (n = 245). If your team analyzes transcripts, survey text, or audio and needs reproducible thematic and cross-segment workflows, use automated tools to cut the busywork, surface the validation markers that matter, and link narratives to behavioral outcomes.

• See the original paper: PLoS ONE
• Try Evidano for free