Evidano is an AI-powered qualitative data analysis platform that helps scale coding, quality assurance, and automated scoring. Researchers and UX/clinical teams often ask: can the way patients remember consultations change how well they retain health advice? A July 20, 2026 PLOS ONE study (n=245) shows that recounting perceived validating vs. invalidating consultations raised odds of recalling short health messages by ~19% (and ~22% in the fully randomized subset). This post translates that finding into a reproducible qualitative analysis workflow (from textual recall prompts to coding, QA, and cross-segment comparisons) using AI-enabled tools. If you analyse transcripts, patient narratives, or survey comments, you’ll learn how to replicate the study’s checkpoints (including AI-generated response detection and automated keyword scoring) and how Evidano supports tagging, thematic synthesis, and secure reporting. Note: this article focuses on research methods and is non-diagnostic and research-only.
Key Takeaways
Evidano is an AI-powered qualitative data analysis platform that helps teams reproduce the study pipeline showing perceived clinician validation improves recall of short health messages.
A July 20, 2026 PLOS ONE study of people with chronic pain (data collected Oct 8–30, 2024) found that asking participants to write about a validating consultation increased incidental recall of 20 short audio messages by ~19% overall and ~22% in a fully randomized subset (final analyzed n = 245).
- Lee et al. (Published Jul 20, 2026) reported ~19% higher odds of recalling each message in the full sample and ~22% in the fully randomized subset, with data collected Oct 8–30, 2024 and final n = 245.
- Automated keyword scoring in the study matched human raters at 97.53% agreement and the authors removed 12 likely AI-generated responses during QA.
- Analytic recommendation: treat perceived validation/invalidation as first-class qualitative codes, use item-level binary scoring, and model outcomes with binomial GLMMs controlling for covariates like pain duration and immersion.
- Evidano supports ingesting narratives, AI-assisted pre-coding, automated keyword scoring with human-in-the-loop validation, and secure exports for GLMM-ready datasets.
Fast take + source
Fast take: Lee et al. (Published July 20, 2026) found that asking people with chronic pain to write about a validating consultation increased incidental recall of 20 short audio health messages, with data collected Oct 8–30, 2024 and a final analyzed sample of n = 245.
- Key numbers: n = 245 retained; approximately 19% higher odds of recalling each message in the full sample (18.5% reported in model) and 22% in the fully randomized subset.
- Quality controls reported: 12 responses flagged as likely AI-generated were removed; automated recall scoring validated against human raters with 97.53% agreement.
- Why it matters for analysts: subjective validation in consultation narratives appears linked to downstream cognitive outcomes, a target measurable with qualitative methods.
Read the original: PLOS ONE.
Findings snapshot
Findings snapshot: the table below summarizes dates, metrics, sources, and implications from the PLOS ONE report and the study methods.
Findings snapshot
| Date / Window | Metric | Value | Source | Implication |
|---|---|---|---|---|
| Data collection | Dates | 8–30 Oct 2024 | PLOS ONE methods | Online recruitment via Prolific; audio-based recall task |
| Publication | Published | 20 Jul 2026 | PLOS ONE | Peer-reviewed open access (DOI in source) |
| Sample | Analyzed N | 245 (Validation = 124; Invalidation = 121) | Results | >80% power for small-moderate effects |
| Primary effect | Odds increase for recall | ~19% (full sample); ~22% (randomized subset) | GLMM estimates | Perceived validation linked to better recall |
| Automated QA | Scoring agreement | 97.53% | Methods (keyword matching validation) | Automated coding validated vs. human scoring |
What happened and why it matters for qualitative researchers
The study used participants' autobiographical descriptions of past consultations (validation vs. invalidation) as an affective prime, then presented 20 real-world health messages via short audio clips and measured incidental recall using a keyword-matching scorer, and the direct effect of subjective validation on recall was robust after covariate controls.
- Design takeaways: subjective perceptions (what patients say happened) are analytic variables, treat them as first-class qualitative codes and as experimental manipulations when using autobiographical reactivation.
- Measurement takeaways: short audio stimuli and item-level approaches (20 items per participant) increase sensitivity; GLMMs are appropriate for nested item data.
- QA takeaways: guard against low-quality or AI-generated text, the authors removed 12 flagged responses and used attention checks.
Implications for researchers, UX teams and clinicians
For qualitative researchers
Qualitative researchers should treat perceived validation/invalidation as thematic codes with frequency, co-occurrence and sequence analyses.
Treat perceived validation/invalidation as thematic codes with frequency, co-occurrence and sequence analyses: does validation co-occur with planning language or concrete recall phrases?
Use item-level scoring (keyword lists validated against human coders) and mixed models for inferential claims rather than collapsing to a single recall score.
Build QA into the pipeline: attention checks, immersion measures, and automated detection of AI-like responses (the paper removed 12 flagged cases).
For UX and patient-experience teams
UX and patient-experience teams should map clinician behaviors to patient narratives and quantify which behaviors predict recall or adherence.
Map specific clinician behaviors (listening, acknowledgement, explanation) to patient narratives and quantify which behaviors predict better recall or adherence.
Segment by pain duration, demographics, or setting to reveal where validation interventions would have the greatest impact.
For trialists and clinical educators
Trialists and clinical educators should consider embedding short validation training or simulated consultations and use narrative prompts to measure cognitive impacts on recall.
Consider embedding short validation training or simulated consultations and use pre/post narrative prompts to measure cognitive impacts on recall during pilot studies.
Complement observational coding with patient autobiographical reports to capture perceived validation, which may be more predictive of outcomes than externally coded behavior.
Do more, faster with Evidano
Problem: messy narrative inputs and AI noise
Evidano helps address messy narrative inputs and AI noise by ingesting texts and flagging likely AI-generated patterns for rapid review.
Manual screening for AI-like responses is slow. Evidano ingests written autobiographical narratives and flags potential AI-generated patterns and duplicate structures so you can review quickly.
Problem: inconsistent coding across raters
Evidano reduces inconsistent coding by supporting importable codebooks and AI-assisted pre-labeling with batch review of disagreements.
Solution: import a codebook into Evidano, run AI-assisted coding to pre-label validation/invalidation themes, then batch-review disagreements. Reproducible code hierarchies and inter-rater reports speed calibration.
Problem: scaling item-level recall scoring
Evidano scales item-level recall scoring through automated keyword scoring with human-in-the-loop validation mirroring the study's 97.53% agreement process.
Solution: Evidano supports automated keyword scoring with human-in-the-loop validation (mirrors the study's 97.5% agreement process) and produces item-level datasets ready for GLMM or export to R.
Problem: comparing segments (e.g., pain duration)
Evidano enables cross-segment frequency and thematic analyses and generates visualizations to show where validation effects concentrate.
Solution: run cross-segment frequency and thematic analyses, generate co-occurrence networks and hierarchical code→subcode visualizations to show where validation effects concentrate.
Security & compliance
Evidano encrypts data end-to-end and does not use your data to train third-party models, supporting GDPR and IRB-conscious projects.
Evidano encrypts data end-to-end and does not use your data to train third-party models, important for sensitive health narratives and GDPR/IRB-conscious projects.
Checklist: 7-step AI-enabled workflow to reproduce this study
This 7-step checklist converts consultation narratives into analyzable outcomes and visuals and reproduces the study pipeline.
- 1) Ingest transcripts/Open-text responses into Evidano (or upload spreadsheets of survey responses).
- 2) Run an AI-assisted quality pass: attention-check flags, likely-AI-generated text flags (review and exclude as appropriate).
- 3) Import or create a codebook for 'validation' vs 'invalidation' plus immersion/affect tags; run auto-coding and review edge cases.
- 4) Prepare item-level recall scoring: create keyword dictionaries for each target message and validate against a 10% human-coded sample.
- 5) Export item-level binary recall data and covariates (age, pain duration, immersion) for GLMMs, or run built-in cross-segment analyses.
- 6) Visualize results: thematic frequencies, co-occurrence networks, and hierarchical code maps for stakeholder reports.
- 7) Produce an executive brief and share via secure export; keep raw sensitive data redacted (PII tools available in-platform).
FAQ: qualitative analysis of clinical consultations
How do I convert subjective reports (like 'felt validated') into reliable codes?
Build a short operational definition, train a small human coding set, use AI-assisted pre-coding, then measure agreement and iterate.
Build a short operational definition, train a small human coding set, use AI-assisted pre-coding, then measure agreement and iterate. The PLOS study used guided prompts and immersion checks to improve consistency.
Can automated scoring really match human judgment?
Automated scoring can match human judgment when keyword sets are validated and refined against human-coded samples, achieving >95% agreement in practice.
With validated keyword sets and iterative refinement, automated scoring can achieve >95% agreement (the study reported 97.53%). Always validate on an initial sample.
How should I handle potential AI-generated participant text?
Use automated detection flags, manual review protocols, and document exclusions to handle potential AI-generated participant text.
Use automated detection flags, manual review protocols, and document exclusions. The study removed 12 flagged cases before analysis and logged decisions for reproducibility.
Is this appropriate for clinical use?
These methods are research-focused and explanatory, and any clinical application requires ethics approval and appropriate oversight.
These methods are research-focused and explanatory. Any clinical application should follow ethics approval and avoid diagnostic claims without proper oversight.
Wrapping up & next steps
Wrapping up: Lee et al. (Jul 20, 2026) provide proof-of-concept that perceived clinician validation links to modest but reliable gains in recall for health messages, approximately 19–22% odds increase.
- For teams analyzing transcripts, survey comments, or consultation narratives, treat subjective consultation memories as analyzable variables, build robust QA into your pipelines, and use item-level scoring with appropriate models.
- Ready to scale this approach? Try Evidano for free.
- If you want a reproducible starter: import 100 narratives, create a 2-codebook (validation/invalidation), validate automated scoring on a 10% sample, then run cross-segment frequency and co-occurrence visualizations; Evidano will generate deliverables you can share with stakeholders.