Fast take: A June 18, 2026 PLoS ONE study protocol lays out a sequential mixed-methods approach; NLP web-scraping of 50 medical school sites, 25 administrator interviews, and 15–24 student focus groups, to build a standardized evaluation framework for wellbeing programs (www.journals.plos.org/plosone/article? id=10.1371%2Fjournal.pone.0351759). If your team evaluates wellbeing, curriculum, or student services, this paper shows how to combine computational breadth with qualitative depth. Read on to learn a repeatable workflow for qualitative analysis of medical student wellbeing and how to run the same pipeline faster and reproducibly in Evidano (www.evidano.com). Note: this is research-focused and non-diagnostic.
Findings snapshot
| Date | Component | Planned sample / n | Method | Key detail |
|---|---|---|---|---|
| June 18, 2026 | Study published (protocol) | , | Sequential mixed-methods | Phase I web-scrape → Phase II KIIs → Phase III FGDs → Phase IV framework |
| Phase I | Web content analysis | 50 medical schools (random sample) | Python crawler, embeddings (all-miniLM-L6-v2), UMAP, HDBSCAN | Classify text into preventative, reactive, cultural, structural |
| Phase II | Key informant interviews | ≥25 administrators | 1-hr semi-structured KIIs, thematic coding (Dedoose) | Hybrid inductive-deductive codebook; inter-coder kappa checks |
| Phase III | Student focus groups | 15–24 students (3 FGDs) | 1.5-hr FGDs, thematic analysis | Groups stratified by training stage; transcripts analyzed for saturation |
| Timeline | Recruitment & analysis | Feb–Aug 2026 (collection); results Apr 2027 | Sequential triangulation and joint displays | Public code archiving planned where allowed |
What happened, study design in plain English
This protocol operationalizes a sequential-exploratory mixed-methods pipeline: first, a reproducible web-scrape and NLP-based content analysis of public medical school webpages; second, purposive KIIs with school wellbeing leads; third, student FGDs to surface lived experience; finally, triangulation to produce a standardized evaluative framework.
- Web scraping: custom Python crawler (requests + BeautifulSoup), sentence embeddings, similarity filtering.
- Extraction: fine-tuned BERT for targeted Q&A; c-TF-IDF + MMR for cluster labeling.
- Clustering & visualization: UMAP (n_neighbors=15, min_dist=0.1, n_components=2, random_state=42) then HDBSCAN (min_cluster_size=5, min_samples=3).
- Qual phase: coded in Dedoose with inter-coder reliability (Cohen’s kappa) and iterative codebook refinement.
- Integration: joint displays and comparative matrices to find convergence/divergence across data sources.
What this means for qualitative analysis of medical student wellbeing
For researchers & evaluators
Combine breadth (web-scrape + NLP) with depth (KIIs, FGDs) to avoid response bias common in voluntary surveys. The protocol’s random sampling of 50 LCME schools improves representativeness compared with opt-in surveys.
Use reproducible seeds, archived scripts, and explicit model versions (they name all-miniLM-L6-v2 and BERT variants) to increase transparency and replicability.
For medical education leaders
Expect common gaps in structural programming (e.g., assessment policies, pass/fail adoption). Triangulated evidence will show whether public messaging matches student experience, a decision point for policy changes.
Administrator interviews reveal implementation constraints and sustainability levers not visible on webpages.
For UX/insight teams
Webpage visibility and phrasing matter: NLP similarity thresholds and sentence-ranking reveal which messages students actually see. Pair that with FGDs to validate how language maps to behavior (help-seeking, program uptake).
Do more, faster with Evidano, map the protocol to a repeatable workflow
Phase I equivalent; Website & social scraping
Problem: building a robust, maintainable crawler and handling diverse site structures.
Evidano: run scalable website and social scraping jobs, auto-segment pages, and produce sentence-level extracts ready for embedding and clustering (no manual pipeline wiring).
Phase I-II bridge; Embeddings, clustering, and codebook seeding
Problem: converting noisy text into meaningful clusters and reproducible keywords.
Evidano: one-click embedding (custom model choices), UMAP/HDBSCAN visualization, and exportable cluster keywords using c-TF-IDF + MMR to seed your qualitative codebook.
Phase II-III; Transcription, translation, and secure qualitative coding
Problem: messy audio, inconsistent transcripts, PII risks, and coder drift.
Evidano: enterprise transcription with custom dictionaries, PII redaction, parallel translation, importable codebooks, team coding with inter-coder comparison, and encrypted storage. Data is never used to train third-party models.
Synthesis & reporting
Problem: stitching web-scrape metrics and thematic codes into joint displays for stakeholders.
Evidano: generate thematic + frequency matrices, cross-segment comparisons (e.g., institution type, training stage), co-occurrence networks, and exportable visual reports for panels or LCME submissions.
Checklist: Reproduce this protocol in 7 steps
Follow these steps to implement the study architecture with AI support:
- 1) Select a reproducible sample and fix the random seed (they use a 50-school LCME sample).
- 2) Run a crawler to harvest page titles, URLs, and sentences; save metadata (publish dates, links).
- 3) Embed sentences (choose a stable model), filter by cosine similarity to question embeddings to keep relevant sentences.
- 4) Dimensionality reduction + HDBSCAN for clusters; extract c-TF-IDF labels and qualitatively validate cluster samples.
- 5) Use cluster labels to draft a codebook and recruit purposive KIIs and FGDs for contextual depth.
- 6) Transcribe audio, code with double-coding and kappa checks, and iterate the codebook until saturation.
- 7) Produce joint displays triangulating web-scrape frequencies with themes; share a standardized framework for program evaluation.
Wrapping up & next moves
Godbole et al. (published June 18, 2026) provide a rigorous, reproducible template for evaluating medical student wellbeing that pairs automated web-scale capture with deep qualitative methods (www.journals.plos.org/plosone/article? id=10.1371%2Fjournal.pone.0351759). If your team is responsible for program evaluation, accreditation evidence, or student experience research, you can reproduce this architecture faster with Evidano: ingest scraped pages, run embeddings and clustering, auto-seed codebooks, and manage secure transcription and team coding in a single platform.
Ready to turn protocol into production? Start a pilot with Evidano to run Phase I-equivalent scraping and thematic mapping in days, not months: www.evidano.com