Truace
Subscribe

Offline AI found 308 housing cases in 25,031 cancer support notes, a study reports

Study title: Housing insecurity and cancer in South Australia: artificial intelligence-enabled insights from a statewide cancer support service

A South Australian study used a locally hosted, offline large language model to read 25,031 case notes written by staff at a statewide cancer support service between December 2020 and December 2025. Of those interactions, 308 (1.23%) were confirmed as housing-related after at least two reviewers checked the flagged records.

Read the study abstract

Objective: This study demonstrates the use of a locally hosted, offline artificial intelligence approach to identify, quantify, and describe housing-related needs documented in routine case notes from a statewide cancer support service. Methods: In this retrospective, exploratory service evaluation, a locally hosted, offline large language model screened 25,031 staff-generated case notes recorded between December 2020 and December 2025 for housing-related interactions and grouped confirmed interactions into cate…

Truvace record2 min read1 sourceEvidence: Moderate · 65/100 — Observational study · n = 25,031 · 1 independent sourcePeer-reviewedTRV-2026-2425 · cite or check it
Housing insecurity and cancer in South Australia: artificial intelligence-enabled insights from a statewide cancer support service"Medical Laboratory" by ben.dracup, CC BY 2.0.

In brief

The confirmed cases showed what housing trouble looked like for people with cancer. Lack of support from health and social services appeared in 44.16% of them, financial barriers to housing stability in 37.66%, and homelessness or reliance on emergency or crisis housing in 36.04%.

In more than half of the confirmed interactions, housing was not the reason the person first got in touch and came up only during conversation. The authors say the findings fit existing evidence for linking cancer, social and housing services, and for giving housing more policy recognition in cancer care.

Main points

  1. The model screened 25,031 staff-generated case notes recorded from December 2020 to December 2025.
  2. 308 interactions (1.23%) were confirmed as housing-related.
  3. Lack of support from health and social services was the most common category, at 44.16%.
  4. Financial barriers to housing stability appeared in 37.66% of confirmed interactions.
  5. Homelessness or reliance on emergency or crisis housing appeared in 36.04%.
  6. In more than half of confirmed interactions, housing emerged only during discussion rather than as the initial reason for contact.
  7. At least two reviewers independently verified flagged records and their categorisation.

The gain

In a retrospective evaluation, a locally hosted offline AI model screened 25,031 cancer support case notes and, after reviewer checks, identified 308 housing-related interactions that were otherwise buried in routine records.

The full story

The work was a retrospective, exploratory service evaluation of routine case notes from a statewide cancer support service in South Australia. A locally hosted model running offline screened the notes for housing-related interactions, which keeps sensitive health information on local systems.

Confirmed interactions were sorted into categories using a predefined framework, and flagged records and categorisations were independently verified by at least two reviewers. Results were summarised descriptively.

Because housing often surfaced only during conversation rather than as the reason for contact, the authors say routine service data can reveal needs that people do not initially raise. They link this to coordinated pathways between oncology, social and housing services.

Sources

  1. Peer-reviewedAustralian Health Review2026-12-03

The debate