Evidano is an AI-powered qualitative data analysis platform that ingests mixed documents and supports thematic and cross-segment analyses. The NPR/KFF reporting (July 19, 2026) documents repeated, sometimes fatal, resident-to-resident violence in memory care that is often preceded by missed warnings and ineffective interventions. This post shows researchers and operations teams how to run a reproducible qualitative analysis of dementia violence, extract early-warning themes from inspection reports, transcripts, and video notes, and turn them into targeted interventions using AI-enabled workflows and Evidano. Source: NPR (KFF reporting).
Key Takeaways
Evidano is an AI-powered qualitative data analysis platform that ingests mixed documents and supports thematic and cross-segment analyses.
The NPR/KFF July 19, 2026 investigation shows predictable warning signs in records and staff reports precede many resident-to-resident assaults, and facilities often fail to act.
- Warning signals are discoverable in unstructured sources including inspection reports, family emails, staff notes, video transcripts, and court records.
- Run a focused 7-day pilot to ingest 50–200 artifacts, draft a 12-code warning-signal codebook, run AI-assisted coding with manual validation, and deliver action-ready evidence packets.
- Evidano ingests heterogeneous sources, runs AI-assisted thematic and cross-segment analyses, and secures data through encryption without using customer data to train third-party models.
Fast Take: Why this matters to researchers & ops
The NPR/KFF July 19, 2026 investigation shows predictable warning signs in records and staff reports that precede many resident-to-resident assaults, yet facilities repeatedly fail to act.
For qualitative teams, those warnings live in unstructured sources (inspection reports, family emails, staff notes, camera transcripts, and court records) and are prime for AI-enabled synthesis.
- Quick payoff: tag, quantify, and visualize warning signals (roommate mismatch, repeated complaints, medication changes, delayed response times) across hundreds of documents.
- Use Evidano to ingest heterogeneous sources, run thematic and frequency analyses, and produce cross-segment reports that operations, compliance, and clinical teams can act on, securely (data encrypted; not used to train third-party models).
Snapshot: Key numbers from the reporting
| Date / Metric | Value | Source / Note |
|---|---|---|
| Article published | July 19, 2026 | NPR (KFF reporting) |
| CMS citations since Jan 2024 (resident-to-resident failures) | At least 700 | CMS inspection reports cited in the investigation |
| Assisted living aggression (Cornell study) | ≈ 1 in 7 residents/month | Cornell University (study referenced in article) |
| Nursing home altercations (Cornell) | ≈ 1 in 5 residents/month | Separate Cornell study referenced |
| Residents with dementia in long-term care | ≈ 900, 000 of 2.2M | CDC figures cited in article |
What the reports show (nuts & bolts)
The reports show proximate failures that repeat across cases: poor roommate and placement decisions, missed escalation windows, insufficient staffing or supervision, and delayed responses to call pendants or incident alerts.
The reporting includes concrete timestamps (for example, aides arriving 13 minutes after an alert in one case) and documentation of prior complaints that were ignored or inadequately acted on.
- Common triggers: noise intolerance, infection or pain, medication changes, wandering, unmet toileting needs.
- Common failure modes: inconsistent door locks, room changes without risk assessment, lack of one-to-one supervision for known agitators.
- Evidence sources to mine: state inspection reports, staff shift notes, family emails, incident logs, court filings, and closed-circuit video transcripts.
So what for qualitative researchers and ops teams?
For qualitative researchers
Qualitative researchers should treat the corpus as mixed-methods, pairing thematic coding of unstructured text with frequency and cross-segment counts (for example, incidents by roommate pair, shift, medication change).
Qualitative researchers should prioritize reproducibility by documenting codebook versions, including audit trails for automated coding, and exporting quote-level evidence for audits or legal review.
For clinical and facility operations
Clinical and facility operations should focus interventions on repeat signals, such as multiple complaints, prior aggression, and sudden behavioral change, rather than isolated incidents.
Clinical and facility operations should use deliverables that drive action: short, visual dashboards showing hotzones by room assignment, peak times for incidents, and staff-response latency.
For policy & compliance teams
Policy and compliance teams should aggregate evidence across facilities to detect systemic issues such as staffing, admission screening, and training gaps that individual inspections miss.
Policy and compliance teams should produce shareable, time-stamped evidence packages to support investigations, corrective action plans, and regulatory submissions.
How to run a qualitative analysis of dementia violence (AI-enabled)
Ingest & normalize
Collect inspection reports, incident logs, emails, court records, caregiver notes, and transcripted video audio, and standardize metadata fields such as date, location, and involved parties.
Use OCR to convert scanned documents and normalize fields so documents can be compared; Evidano can ingest documents and spreadsheets and supports custom dictionaries for domain terms.
Create a warning-signal codebook
Create a warning-signal codebook with 12 to 20 codes up front, for example 'noise intolerance', 'roommate conflict', 'med change', 'delayed response', or 'weaponized object'.
Import or version a codebook so AI-assisted coding is reproducible and auditable.
Automate coding & validate
Run AI-assisted coding at scale, then manually review a stratified sample to estimate precision and recall.
Use hierarchical codes and subcodes to capture nuance, for example 'weapon' broken into 'walker' and 'belt buckle', and document manual review findings to refine the model.
Cross-segment & frequency analysis
Compare incidents by roommate pairing, shift, medication change, and facility revenue model to quantify which signals predict escalation.
Use thematic, content, frequency, and cross-segment analyses to measure signal prevalence and co-occurrence.
Visualize & export
Produce stakeholder-ready visuals such as word clouds, co-occurrence networks, and hierarchical code maps to highlight patterns.
Export evidence packets that include quotes and original documents for legal or compliance review.
Iterate and collect follow-ups
Collect follow-up information using targeted surveys or AI avatars and feed results back into the corpus for continuous monitoring.
Iterate the codebook and analysis cadence as new signals or edge cases appear during follow-up.
Security & ethics note
Handle these sensitive data under strict consent and privacy rules and treat work as research or quality-improvement, not clinical diagnosis.
Evidano encrypts data and does not use customer data to train third-party models, and teams should maintain consent, retention, and redaction policies.
This week’s 7-step workflow (practical checklist)
A focused 7-day pilot can surface high-priority interventions quickly.
- Day 1: Ingest 50–200 artifacts (inspection reports, incident logs, family emails, audio/video transcripts) into Evidano and normalize metadata.
- Day 2: Draft a 12-code warning-signal codebook and import it into the platform.
- Day 3: Run AI-assisted coding; review a 10% sample for accuracy and adjust the codebook.
- Day 4: Produce frequency and cross-segment reports (roommate pairs, time-of-day, medication changes).
- Day 5: Create two visual dashboards (hotrooms map; response-latency trend) and circulate to operations and compliance.
- Day 6: Run targeted follow-up (two to three AI avatar interviews with staff or short family surveys) to validate hypotheses.
- Day 7: Deliver an evidence packet and three prioritized actions (reassign rooming, dedicated supervision for 24–72 hours, policy changes) and set a monitoring cadence.
Wrapping up: What to do next
The NPR/KFF series (July 19, 2026) shows that many violent incidents leave a paper trail of missed warnings, and the key challenge is extracting them reliably and quickly.
- Start small: run a two-week pilot on a single facility to prove signal-to-action workflows.
- Want to try this workflow on your corpus? Learn how Evidano ingests mixed documents, runs thematic and cross-segment analyses, and builds visual reports, or request a demo to see a pilot tailored to inspection reports and incident logs.
- Try Evidano for free
FAQ: Qualitative analysis of dementia violence
How can qualitative teams extract early-warning signals from records?
Qualitative teams can extract early-warning signals by mining unstructured sources such as inspection reports, family emails, staff notes, video transcripts, and court records.
Qualitative teams should pair thematic coding with frequency and cross-segment counts, document codebook versions, and export quote-level evidence for audits or legal review.
What sources of evidence should be collected and normalized?
Collect inspection reports, incident logs, emails, court records, caregiver notes, and transcripted video audio, and normalize fields such as date, location, and involved parties.
Use OCR for scanned artifacts and standardize metadata so that incidents can be compared across facility, shift, and roommate pairings.
How quickly can a pilot surface actionable interventions?
A focused 7-day pilot can surface high-priority interventions.
A 7-day pilot ingests 50–200 artifacts, drafts a 12-code codebook, runs AI-assisted coding with a 10% manual review sample, produces cross-segment reports, and delivers dashboards and an evidence packet.
How does Evidano protect sensitive data used in these analyses?
Evidano encrypts data and does not use customer data to train third-party models.
Teams should handle sensitive records under strict consent and privacy rules and treat this work as research or quality-improvement rather than clinical diagnosis.