Qualitative Analysis of IYCF Practices with AI

Evidano is an AI-powered qualitative data analysis platform that accelerates thematic and cross-segment analysis of IYCF data.

This post outlines a 7-step AI-enabled workflow to turn 460 mother–infant dyad transcripts, frontline-worker IDIs and weekly process data from the Palghar study (Oct 2024–Sep 2025; 24-month total study) into stakeholder-ready visuals and evidence briefs.

• Centralize audio, transcripts, diet cards and process sheets, then apply automated redaction, diarization, translation alignment and AI-assisted coding to save weeks of manual work.
• The 7-step workflow (ingest, clean/redact, translate, codebook, AI-assisted coding, cross-segment analysis, visualize/report) produces code frequencies, exemplar quotes and cross-segment comparisons across 44 AWCs.
• Security is preserved via end-to-end encryption and models not used to train third-party LLMs, enabling work with vulnerable groups while meeting compliance needs.

The protocol collects structured WHO IYCF indicators, anthropometry, and rich qualitative inputs such as IDIs, observational notes, diet cards and weekly process indicators.

The study pairs eight WHO core IYCF indicators (initiation, exclusive breastfeeding, MDD, MMF, MAD, iron intake), anthropometry and monitoring data with qualitative transcripts; qualitative data is planned for verbatim transcription, NVivo analysis and synthesis with quantitative pre/post comparisons in SPSS.

• Units of analysis: mothers (n≈460 per round), AWWs (77 AWWs across 77 AWCs), supervisory staff and community stakeholders.
• Planned qualitative outputs: codebooks, themes on barriers/enablers, contextual recipes and SBC (social behavioural change) signal events.
• Monitoring data: weekly process indicators (feeds taken at AWCs, MDD/MMF/MAD counts, supplement uptake).

The 7-step AI-enabled workflow below describes how to move from raw audio and text to evidence briefs and stakeholder visuals in weeks rather than months.

• 1) Ingest & centralize: Upload audio files, transcripts, diet cards and survey sheets. Evidano accepts mixed formats and spreadsheets, and supports custom dictionaries for local food terms.
• 2) Clean & redact: Run automatic speaker diarization, PII redaction, and normalization (for example, 'jowar' vs 'sorghum') so coding is consistent across sources.
• 3) Translate & align: Translate local languages with a custom glossary (preserve food names and cultural phrases) and align translated transcripts to original timestamps.
• 4) Build or import codebook: Import the study’s NVivo codebook (WHO/IYCF indicators) and extend with frontline-worker codes discovered in IDIs, then lock parent→subcode hierarchies.
• 5) AI-assisted coding: Auto-suggest codes at scale, then batch-validate with human reviewers and capture code frequencies and exemplar quotes with intercoder checks.
• 6) Cross-segment analysis: Compare themes by segment (AWC, village, AWW experience level, pre/post), and use co-occurrence networks to spot clusterable barriers (for example, food access and cultural restrictions).
• 7) Visualize & report: Generate word clouds, co-occurrence maps, hierarchical code trees and export readable evidence briefs for district officers, including process-indicator trend overlays.

The checklist lists 10 QA steps to confirm before running batch AI coding.

• All transcripts timestamped and linked to audio.
• PII redaction completed and consent documented.
• Custom glossary uploaded for local food and terms.
• Baseline codebook imported and agreed by two or more analysts.
• Sample intercoder reliability test (kappa) passed.
• Process indicator spreadsheet keys match qualitative IDs (AWC codes).
• Pre/post flags present on records for segment comparison.
• Diet cards digitized and normalized.
• Weekly monitoring logs ingested for time-series overlays.
• Export templates defined for stakeholders (one-pager and slide deck).

The next steps are to centralize audio, transcripts and process sheets, then run the 7-step workflow above to cut coding time, improve reproducibility and produce stakeholder-ready visuals.

If you are preparing to analyse the Palghar IYCF protocol data, see the protocol in PLoS ONE.

• Ready for a trial? See how Evidano ingests mixed inputs, preserves local terms, runs thematic and cross-segment analyses and exports policy-ready briefs at Evidano, or Try Evidano for free.

Ethics note: This guidance is research-focused and non-diagnostic; always follow approved IRB procedures and the consent and PII protections described in the protocol.