Dementia Action Week runs from 21 to 27 September 2026 and Dementia Australia has asked the country to start talking. Around 446,500 Australians are living with dementia today and more than 1.7 million people are involved in their care. Dementia is now Australia’s leading cause of death. Yet in residential aged care we fund all of it through a model that never records the word.
That is not an oversight. It is the design. Still, if we are going to talk about dementia this week, we should talk about what the design does to the people delivering the care and to the homes carrying the cost.
AN-ACC sees the consequence, not the condition
The classification study behind AN-ACC found that specific medical diagnoses, dementia among them, are not cost drivers in their own right. What drives cost is what those conditions produce, which is frailty, impaired mobility, functional decline, cognitive loss, behavioural disturbance and technical nursing need. So the model classifies the consequence and discards the label.
There is a genuine benefit in that. Funding no longer waits on a formal diagnosis, which clinicians find notoriously difficult to reach, particularly in the early and mixed presentations we see most often. Need can be recognised before a name is attached to it.
However, the problem does not disappear. It moves. Where a diagnosis once carried the evidentiary weight, that weight now sits with the registered nurses who observe and document behaviour, and with the assessors who interpret what they wrote. Whoever writes the progress note now sets the claim. This is a documentation discipline, not a clinical one, and most organisations have not yet resourced it as such.
The second consequence is that the sector struggles to count itself. Once diagnoses are stripped from the funding record, tracking system-wide outcomes for people living with dementia becomes a modelling exercise rather than a simple query.
Why almost nobody is Class 3
Under AN-ACC, cognition quietly determines mobility. A resident who can physically walk but who needs prompting, supervision or physical support, because they cannot navigate safely or recall where they were going, belongs in the assisted mobility branch rather than the independent one. The modified De Morton Mobility Index tests whether someone can comprehend and complete a mobility task, so a cognitive deficit that prevents safe independent performance pulls that resident across the branch boundary.
Our data shows what that rule does in practice. Across the 46,696 permanently classified residents in our cohort, only 249 sit in Class 3. That is 0.53 per cent, and the national figure is identical. The class that looks, on its description, like the natural home for mobile residents living with dementia is very nearly empty, because the classification tree sends almost everyone who fits that description somewhere else. The real ambulant dementia cohort lives in Classes 6, 7 and 8, which together hold 34.6 per cent of residents.
What we tested
We took a point-in-time extract as at 22 September 2026 covering 573 residential services across 134 provider organisations, holding 49,692 residents. Of those, 46,696 carry an assessed permanent class, 2,106 are in respite classes and 890 sit on a default class. The median home holds 83 residents.
The cohort is an unusually clean mirror of the sector. Every one of the thirteen permanent class shares falls within 0.1 percentage points of the national distribution, and the cohort-weighted variable subsidy of $160.45 per resident per day sits within 16 cents of the national figure. Whatever we find here should generalise.
We built an ambulant cognitive load index, which is the share of assessed permanent residents in Classes 3, 6, 7 and 8. Those are the classes where cognition or behaviour determines the classification and where the resident can still walk. That cohort numbers 16,421 residents, or 35.2 per cent. We then tested it against each home’s care minutes performance and against its claim profile.
Finding one. Funding responds to immobility roughly twice as hard as it does to dementia
Regressing each home’s average NWAU on its ambulant cognitive load share and its not-mobile share explains 81.5 per cent of the variation in claim profile. The two coefficients are not close to each other.
- Every additional 10 percentage points of not-mobile residents adds $10.26 per resident per day to the variable claim.
- Every additional 10 percentage points of ambulant cognitive load adds $5.66.
So a home can take on a substantially heavier dementia profile and watch its claim barely move. Across our cohort, ambulant cognitive load and average NWAU correlate at just 0.11, whereas the not-mobile share correlates at 0.77. AN-ACC tracks bodies that cannot move far more faithfully than it tracks minds that cannot navigate.
The current class relativities make that concrete. A resident in Class 8, who is ambulant with low cognition and who is very often the most behaviourally complex person in the building, attracts $177.38 per day. A resident in Class 11, who is not mobile with lower function and lower pressure sore risk, attracts $201.04. That gap of $23.66 per day is $8,636 a year for every such resident. Against Class 7 the gap widens to $41.39 a day, which is $15,107 a year.
Finding two. Care minutes performance barely notices cognitive load
This was the surprise. We expected high cognitive load homes to be pushing minutes well beyond what their funding implied. They are not, at least not measurably.
Total care minutes delivery is flat across the whole range. The correlation between cognitive load and total minutes performance is 0.02, which is indistinguishable from nothing. Registered nurse minutes drift upward as cognitive load rises, moving from 109.3 to 113.3 per cent, and the share of homes falling short of the total target climbs from 19.5 to 25.7 per cent. We should be careful with both of those. Neither reaches conventional statistical significance in this sample, at p equal to 0.08 and 0.39 respectively. They are worth watching rather than worth quoting.
We also matched homes on funding, comparing the highest and lowest cognitive load thirds within a narrow band of average NWAU. The high cognitive load homes delivered slightly more total minutes and noticeably more registered nurse minutes, but the differences were not significant.
The honest reading is that care minutes targets already absorb cognitive load, because the same class mix that sets the claim also sets the target. Homes staff to their target and they largely hit it, whoever is living there. That part of the system is working as intended.
What does move performance is the overall claim profile. Higher acuity homes sit closer to the line, with a correlation of minus 0.16 between average NWAU and total minutes performance, which holds after controlling for cognitive load and home size. Larger homes run materially tighter on registered nurse minutes, correlating at minus 0.25. Across the cohort, 20.6 per cent of homes are below their total target, 8.2 per cent are below their registered nurse target, and a further 29.5 per cent sit between 100 and 102 per cent, which is a buffer that one unplanned roster gap will erase.
Finding three. The October 2025 recalibration moved money the wrong way for Class 8
The Department adjusted both the AN-ACC price and the class weightings on 1 October 2025. Comparing the cash position before and after tells a pointed story about where the model now places its confidence.
Selected classes, covering the cognition branch and the not-mobile comparators. Those six classes hold 62.6 per cent of residents in our cohort, so the table is not a complete view of the thirteen permanent classes. Shares are rounded to one decimal place, which means the column will not sum exactly to the group totals quoted in the text.
Class 3 received the single largest proportional uplift in the cognition branch, at 13.2 per cent. That looks like a direct response to exactly the criticism the sector has been making. The difficulty is that the uplift landed on a class holding one resident in every two hundred, so it moved seven cents per resident per day across our cohort.
Meanwhile Class 8, which holds 5,281 residents in our cohort, or 11.3 per cent, and represents the purest ambulant dementia cohort the model recognises, went backwards in cash terms. It is the only class in the entire cognition branch that did. Class 11, sitting directly across the mobility boundary, gained 7.8 per cent.
None of this is evidence of intent, and the recalibration cut several not-mobile classes hard as well, with Class 10 down 11.8 per cent and Class 13 down 4.5 per cent. But the net effect on a home that specialises in ambulant dementia care was close to zero, while the sector average variable subsidy rose 2.1 per cent.
That 2.1 per cent deserves its own working, because it is the benchmark which makes the Class 8 result meaningful. Holding our cohort casemix constant at 22 September 2026 and applying each vintage of weightings and price in turn, the variable component moves from $157.13 to $160.45 per resident per day. The price itself rose 4.7 per cent across that period, so the class weightings came down on net. Average casemix NWAU across our cohort fell from 0.556 to 0.543, which absorbed more than half of the price increase before a single resident changed class.
The real tension sits between Classes 7, 8 and 11
Those three classes hold 43.4 per cent of all residents and they sit at the exact point where documentation quality decides the outcome. A resident who wanders, resists care and needs two staff to redirect can land in Class 7 or Class 8 depending on how their cognition was scored, which is a difference of $17.73 a day. A resident whose mobility has declined far enough to cross into the not-mobile branch moves to Class 11 and gains another $23.66 a day, even though the behavioural workload has usually fallen rather than risen.
That is the uncomfortable shape of the model. Behavioural complexity, which is what exhausts a workforce, is worth less than physical dependency, which is what a roster can plan around. Our registered nurses carry the assessment burden for the cohort the model rewards least.
What providers should do about it
- Treat behavioural and cognitive documentation as a funding control, not a clinical afterthought. In a diagnosis-blind model, the note is the evidence. Classes 6, 7 and 8 turn on cognition scoring and compounding factors that are only visible if someone wrote them down.
- Know your reassessment position on ambulant residents. Where mobility, function or behaviour has genuinely changed, an out-of-date classification is money left on the table and a care minutes target that no longer reflects who is actually in the building.
- Watch the buffer, not the average. Almost a third of homes sit within two percentage points of their total care minutes target, which is roughly one unfilled shift.
- Model the claim profile alongside the roster. Because targets and funding both derive from the same class mix, every admission decision changes two numbers at once, and they do not move in proportion.
Dementia Action Week asks us to talk about dementia. In our part of the sector, that conversation has to include the uncomfortable fact that the funding model literally cannot hear the word, and that what it hears instead undervalues the people who need the most supervision and the staff who provide it.
Method, assumptions and sources
- Cohort. Mirus client extract as at 22 September 2026 covering 573 residential services, 134 provider organisations and 49,692 residents. Analysis restricted to the 563 services holding at least 20 assessed permanent residents. All figures are reported in aggregate and no service or provider is identified.
- Ambulant cognitive load index. Share of assessed permanent residents in Classes 3, 6, 7 and 8, which are the classes where cognition or behaviour determines the classification and mobility is retained. Not-mobile share covers Classes 9 to 13.
- Funding figures. AN-ACC national efficient price of $295.64 and class NWAU values effective 1 October 2025, compared against the price of $282.44 and NWAU values effective 1 October 2024. Variable component only. Base Care Tariff, the initial entry adjustment and supplements are excluded, so quoted amounts are not total funding per resident.
- Sector average movement. The 2.1 per cent uplift in variable subsidy is calculated on our cohort casemix as at 22 September 2026, held constant, with the October 2024 and October 2025 weightings and prices applied in turn. That makes it a like-for-like comparison of model settings rather than an observed change in claims, which would also carry casemix drift.
- Rounding. Class shares are rounded to one decimal place in the tables while group totals in the text are calculated from unrounded resident counts, so the two will differ by a tenth of a point in places. Annualised funding gaps are calculated from the published daily rates so they reconcile against the Departmental schedule.
- Care minutes performance. Total and registered nurse delivery as a percentage of each service’s own casemix-adjusted target. Targets are set quarterly by the Department from the service’s AN-ACC class mix, so performance is already acuity-adjusted.
- Statistical reporting. Correlations are Pearson unless noted. Differences described as not significant were tested at the 5 per cent level using Welch t-tests and chi-squared tests of independence.
- Sources. Department of Health, Disability and Ageing schedule of subsidies and supplements for residential and transition care. Department of Health, Disability and Ageing published AN-ACC classification data for the national class distribution [Ty to confirm exact dataset]. Eagar and colleagues, The Australian National Aged Care Classification, Medical Journal of Australia, 2020, for the diagnosis-blind design principle. Dementia Australia for Dementia Action Week 2026 dates and prevalence figures.
- Verify before reuse. AN-ACC prices, NWAU values and care minutes allocations are adjusted periodically. Confirm all figures against current Departmental schedules before relying on them for budgeting or board reporting.
