Fast, reproducible qualitative analysis matters when you are judging whether a training and mentorship model shortened diagnostic delays. A June 29, 2026 PLOS Medicine report on a multilevel onsite training and 6‑month mentorship program in Northwest Ethiopia found large provider gains and a 54% drop in diagnostic interval, and the original study appears in PLOS Medicine. This post shows researchers and implementation teams how to reproduce and extend those findings using AI qualitative analysis: what to extract from interviews, how to cross-segment by facility level, and how Evidano speeds transcription, thematic coding, and stakeholder-ready visualizations while keeping data private (see Evidano).
Key Takeaways
Evidano is an AI-powered qualitative data analysis platform that speeds transcription, thematic coding, and stakeholder-ready visualizations while keeping data private.
A June 29, 2026 PLOS Medicine study of a Jan–Sep 2024 multilevel onsite training and 6‑month mentorship in Northwest Ethiopia found large provider gains and a ≈54% reduction in diagnostic interval, and AI-enabled qualitative synthesis can reproduce those timelines and CFIR themes rapidly.
- The multilevel training ran Jan–Sep 2024 with 10-day hospital modules, 7-day primary-care modules, pictorial HEW modules, and mentorship, reaching 1, 020 HEWs.
- Median knowledge and practice scores rose to above 90 in many cadres, with measurable system changes: diagnostic interval fell from 56.5 to 25.8 days (≈54% decrease) and HEW referrals increased from 0.4 to 1.2 referrals per HEW per month.
- The Evidano workflow can reproduce timelines, CFIR-coded themes, and stakeholder-ready briefs from a pilot corpus in a 10–14 day checklist.
Findings snapshot
| Metric | Value (reported) | Why it matters |
|---|---|---|
| Study period | Jan–Sep 2024 (mentorship Apr–Sep 2024) | Implementation timeline for pre/post analysis |
| Participants trained | 18 hospital clinicians, 29 primary clinicians, 1, 020 HEWs | Three-tier reach: tertiary → primary → community |
| Knowledge (median) | Level I: 54.6 → 90.9; Level II: 36.4 → 90.9 | Substantial cognitive gain after training |
| Practice proficiency (median) | Level I: 87.5 →100; Level II: 68.8 →93.8 | Improved clinical behavior measures |
| Diagnostic interval | 56.5 days → 25.8 days (≈54% ↓) | Primary outcome: faster confirmation |
| Patient delay | 27.0 days → 24.5 days (≈9.3% ↓) | Smaller community-level change |
| Budget | USD 52, 762 | Feasibility signal for low-cost scale-up |
| Published | June 29, 2026 (PLOS Medicine) | Peer-reviewed source |
What happened (concise methods & outcomes)
The study evaluated a Jan–Sep 2024 multilevel onsite training and a subsequent 6‑month mentorship and measured knowledge, practice, referral, and interval outcomes.
Between January and September 2024 the University of Gondar ran a multilevel onsite training: 10 days for hospital clinicians, 7 days for primary-care staff, and pictorial modules for 1, 020 health extension workers (HEWs), followed by a 6‑month mentorship program with monthly on‑site and remote supervision.
- High fidelity of delivery: approximately 90% of planned training days were delivered and surveys had a 100% participant response rate.
- Large quantitative gains: median knowledge and practice scores rose to above 90 in most cadres.
- System outcomes: diagnostic interval decreased from 56.5 to 25.8 days (≈54% decrease) and HEW referrals increased from 0.4 to 1.2 referrals per HEW per month.
- Constraints: treatment initiation lagged, with treatment interval rising 11.9%, and civil unrest plus seasonal rains required cascade adaptations.
The study used a quasi‑experimental pre–post design without a contemporaneous control, so coded qualitative data were essential to explain causal mechanisms and contextual moderators.
Why qualitative analysis mattered here
Qualitative analysis explained how and why the intervention produced the observed quantitative changes.
Quantitative endpoints like median days show effect size, while interviews and FGDs reveal facilitators, barriers, and process adaptations that explain those numbers.
- CFIR-based interviews uncovered facilitators (supportive leadership, trusted HEWs, EMR/DHIS-2 feedback loops) and barriers (supply shortages, cultural beliefs, staff rotations).
- Qualitative timelines explained deviations: civil unrest forced cascade trainings and increased tele-support, which are important process data for replication.
- Attitudes did not improve proportionally; qualitative themes pointed to motivational and cultural gaps that require non-technical content in future curricula.
AI qualitative analysis: a reproducible workflow
Inputs to collect
Collect training materials, surveys, mentorship logs, chart-review CSVs, interview and FGD audio, and referral-form samples as the primary inputs.
Also collect metadata such as dates, facility tier, mentee role, and district to allow cross-segmentation by conflict or transport disruption status.
Step-by-step (what to run in Evidano)
Run these steps in Evidano to reproduce the study's qualitative synthesis and link it to chart-based outcomes.
1) Ingest all documents and audio into Evidano; apply custom dictionaries for local terms (for example local words for lymphadenopathy) and enable PII redaction for chart and interview transcripts.
2) Auto-transcribe interviews with domain-tuned models and run translation if interviews are in local languages, preserving glossary mappings.
3) Run an automated thematic analysis to extract themes, subthemes, and code frequencies and map codes to CFIR constructs.
4) Run cross-segment analyses comparing themes and code frequencies by facility tier, by district, and by time window (pre vs post).
5) Generate visualizations such as co-occurrence networks of barriers and facilitators, hierarchical code trees, and word clouds for rapid stakeholder briefings.
6) Use Evidano AI chat over the uploaded corpus to ask targeted questions (for example, “Which barriers were repeatedly linked to treatment delays? ”) and to produce a one-page executive brief.
7) Export reproducible reports and share clickable quotes with timestamps for presentations to ministry or funder audiences.
Why each step matters
Each step preserves context, links process to outcomes, and speeds synthesis compared with manual review.
Custom dictionaries and translation preserve local terminology, which was critical in this study where HEWs used terms like 'nififit'.
Cross-segment frequency and co-occurrence analyses surface actionable links, for example reagent shortages co-occurring with referral errors.
AI chat and exportable visuals shrink synthesis time from weeks to hours, enabling rapid decision cycles.
Implications for researchers and implementers
For evaluators
Evaluators should pair chart-based intervals with coded qualitative themes to strengthen causal plausibility in pre–post designs.
Use cross-segmentation by facility tier, district conflict status, and HEW cascade exposure to detect heterogeneity of effect.
For program teams
Program teams should embed mentorship logs and case reviews into the same analysis corpus as interviews to link process fidelity to outcomes.
Track unintended consequences such as workload displacement and attitude changes with targeted codes so adaptations can be timely.
For policy/stakeholders
Policy teams should use qualitative evidence to clarify why diagnostic intervals fell (for example referral-form improvements and tele-support) and why treatment initiation lagged (infrastructure limits).
Use themed qualitative evidence to prioritize investments and operational fixes.
Do more, faster with Evidano
Transcription & translation
Evidano accelerates transcription and translation with glossary support and built-in PII redaction to secure identifiers.
Auto-transcribe FGDs and IDIs with custom glossaries to save hours per interview while securing patient identifiers.
Thematic & cross-segment analysis
Evidano auto-generates themes, subcodes, and frequency tables and runs cross-tabulations to detect where the intervention worked best.
Run cross-segment analyses such as Level I versus Level II or unrest-affected versus stable areas to surface heterogeneity.
Visualizations for impact
Evidano creates one-click co-occurrence networks and hierarchical code trees to turn qualitative nuance into concise slides for ministries and funders.
These visuals are ideal for showing links between staffing, supplies, and delays.
AI chat & reproducible reports
Evidano AI chat lets users ask natural-language questions over their corpus and get evidence-backed summaries with cited interview excerpts and timestamps.
Exportable executive briefs and reproducible reports speed stakeholder communication.
Security & compliance
Evidano encrypts data and does not use uploaded data to train third-party models, supporting secure handling of patient charts and sensitive interview data.
Follow local IRB guidance and consent practices when handling clinical and interview data.
Checklist: run this analysis in 10–14 days
A 10–14 day checklist guides reproducible setup, transcription, coding, cross-segmentation, and brief production.
- Day 1–2: Gather materials (audio, charts, KAP CSVs) and build a local-term glossary.
- Day 3–4: Ingest data to Evidano; run transcription and translation; apply PII redaction.
- Day 5–7: Auto-code and review top themes; refine codebook and re-run coding.
- Day 8–10: Run cross-segment frequency analyses and co-occurrence networks.
- Day 11–12: Draft executive brief with AI chat; validate quotes and timelines with domain experts.
- Day 13–14: Finalize visuals and share with stakeholders; archive reproducible analysis for scale-up.
Ethics note
This analysis approach is research-focused and non-diagnostic, and requires standard ethical safeguards.
Always obtain informed consent for interview and audio use, de-identify patient records, and follow local IRB guidance when handling clinical data.
Wrapping up: your next moves
Combining chart metrics with AI-enabled qualitative synthesis reveals mechanisms, heterogeneity, and operational fixes faster than manual review.
Start by uploading a pilot set (10 interviews and 20 charts) into Evidano to reproduce timelines and CFIR themes from the PLOS study, then scale; to get started, Try Evidano for free.
FAQ: AI qualitative analysis
What were the main quantitative outcomes of the Gondar training and mentorship?
The main quantitative outcomes were large provider gains, a ≈54% decrease in diagnostic interval, and increased HEW referrals.
Specifically, diagnostic interval decreased from 56.5 to 25.8 days, patient delay decreased from 27.0 to 24.5 days (≈9.3% decrease), median knowledge rose to above 90 in many cadres, practice proficiency increased (for example Level I median practice 87.5 → 100), HEW referrals rose from 0.4 to 1.2 referrals per HEW per month, and the reported budget was USD 52, 762.
How did qualitative analysis contribute to understanding the results?
Qualitative analysis explained how and why quantitative changes occurred by identifying facilitators, barriers, and process adaptations.
CFIR-based interviews revealed supportive leadership and EMR/DHIS-2 feedback loops as facilitators, and supply shortages, cultural beliefs, and staff rotations as barriers, while timelines documented adaptations to civil unrest and seasonal constraints.
What does the reproducible Evidano workflow include?
The reproducible Evidano workflow includes ingestion of training materials and audio, glossary-based transcription and translation, automated thematic coding mapped to CFIR, cross-segmentation, visualization, AI chat, and exportable reports.
These steps preserve local terminology, enable co-occurrence and frequency analyses, and produce stakeholder-ready briefs with cited quotes and timestamps.
How long does an Evidano-enabled analysis take?
A pilot Evidano-enabled analysis can be completed in about 10–14 days following the provided checklist.
The checklist covers data gathering, ingestion and transcription, auto-coding and review, cross-segment analyses, drafting the executive brief with AI chat, and finalizing visuals for stakeholders.