Algorithm Denies 85-Year-Old 15 Hours/Week Home Care for 5 Months

A production algorithm in Australia's aged care system has overruled a human assessor and left an 85-year-old woman with four hospital admissions and a brain bleed waiting five months for approved care. For AI practitioners, the Integrated Assessment Tool is a high-stakes case study in automated decision thresholds, data quality, and the absence of a meaningful appeals loop.

The most striking aspect of this story is the collision between a vulnerable older person's acute clinical needs and an automated decision system that appears unable to see them. Helen Parkinson, 85, was approved in March 2026 for a level four home care package after an in-home assessor determined she needed about 15 hours of care per week. Five months later, she is still waiting. The reason, according to the reporting by Dana Daniel in The Standard and The Examiner, is that the controversial Integrated Assessment Tool (IAT) - which has replaced human decision-making in Commonwealth-funded aged care - categorised her case as not urgent. This categorisation persists despite four hospital admissions in 2026 and a fall that left her with bleeding on the brain. The human assessor's recommendation, based on face-to-face observation, was effectively overruled by a classification algorithm. That is a profound accountability problem, not just a bureaucratic delay.

“

Helen Parkinson is a retired public servant who worked for the Department of Prime Minister and Cabinet under Bob Hawke and Paul Keating; she is articulate and has a local newspaper covering her story.

Context matters. Australia's aged care system has been in a state of reform for years. Following the Royal Commission into Aged Care Quality and Safety, the government introduced a Single Assessment System and the Integrated Assessment Tool to standardise how older people are assessed for Commonwealth-funded services. The intent was to reduce inconsistency, speed up decisions, and target resources to the most urgent cases. On paper, an algorithm can do all of that at scale. But in practice, this case suggests the tool may be prioritising according to a scoring model that does not adequately weight recent acute events such as falls with cerebral bleeding, recurrent hospital admissions, and bereavement. Helen Parkinson's husband died only a month before the article was published, after spending three years in a nursing home. That loss left her living alone and heightened her need for supervision - a factor the in-home assessor explicitly identified, but the algorithm apparently did not.

The clinical risk is immediate. Parkinson currently receives a level two package, which funds around five hours of care per week. Her approved level four package would provide roughly 15 hours - enough for daily assistance and supervision to prevent another fall in the bathroom or elsewhere. Without that support, the probability of further falls, hospital readmissions, and eventual permanent residential aged care entry rises. This is a perverse outcome for a system trying to support ageing in place. Residential aged care is more expensive than home care and often contrary to patient preferences. Parkinson herself says she has always been independent and wants to stay at home. If the algorithm's non-urgent classification forces her into residential care, the system will have saved money on home care packages while incurring higher downstream costs and doing harm to a person's autonomy.

The technology implications are equally serious. The IAT is described as controversial and as having replaced human decision-making. That means the algorithm's urgency classification is not simply advisory; it appears to determine access to funding. Yet there is no indication in the reporting of an appeal pathway that allows a clinician or assessor to override a non-urgent classification when new medical events occur after the initial approval. Parkinson was approved in March, but her four hospital admissions and brain bleed happened sometime that year. If the algorithm's input data was collected at the assessment point and not updated with subsequent hospital records, the classification could be stale. Even if it was updated, a non-urgent label for a woman with a brain bleed and recent hospitalisation is clinically questionable. Without transparency into the tool's scoring weights, thresholds, and data refresh cycle, patients and clinicians cannot understand or challenge the decision.