Limitations, assumptions, and future directions

1

Decision curves are modelled, not reconstructed

Moderate impact

Each model's published AUC is mapped to a sensitivity-specificity trade-off via a binormal approximation: assume normally distributed risk scores in cases and controls with equal variance and mean separation d' = √2 × Φ⁻¹(AUC). The binormal assumption is conservative for discrimination but loses information about the shape of the risk distribution (skewness, ceiling effects, tail behaviour).

What this means
The ranking of models by net benefit is robust. The exact net-benefit values are not, especially at very low or very high thresholds. The clinically useful range's outer edges are approximations.
2

Workforce supply is measured; the skin-cancer share of GP time is not

Moderate impact

Supply for all 88 SA4s is National Health Workforce Dataset 2024 clinical hours ÷ 40 (clinical FTE) by the SA4 of each clinician's main job, for dermatologists and GPs; the Workforce Lab adds practice nurses (main job area primary care). Where an SA4 has three or fewer dermatologists the published figures are perturbed for confidentiality, so the tool uses only a band (under 1, 1–2 or 2–4 FTE). Within the band each region gets its midpoint scaled by one common factor, so that these regions together match the published national total minus the other regions (about 40 FTE in 2024; the midpoints alone would give 54.5). The share of GP time spent on skin cancer is the state share from the Medicare calibration (NHWDS clinical hours fitted to Medicare skin-cancer procedure volumes by state). By default the differences between states come from applying the Queensland method of Lindsay 2026 to every state, which is an ATOF extension: the paper covers Queensland only. The ABS defines 108 SA4 codes; the 88 in the tool cover the resident population of the six states, the ACT and the NT. The other 20 are a “No usual address” code and a migratory, offshore and shipping code for each state and territory (18), Other Territories (Christmas Island, Cocos (Keeling) Islands, Jervis Bay, Norfolk Island) and “Outside Australia”.

What this means
Headcounts and hours are measured. The GP skin-cancer share is not: Medicare data could not pin down its level, so it rests on Lindsay/AFHW priors and is the largest single source of capacity uncertainty (the Threshold Advisor shows this region by region). Clinicians are counted where their main job is, so outreach and second jobs are not located, and regions are treated as self-contained.
3

The screening demand model is calibrated, not fitted

Moderate impact

We model the demand a given threshold creates as: of the eligible adult population (48% of national ERP, spread across SA4s by population aged 40+), the fraction above threshold t comes from the same binormal risk distribution as the decision curve for the selected model (AUC and prevalence), with the regional prevalence shifted by the SA4's invasive melanoma risk per resident aged 40+ (age-standardised SIR × the SA4's 40+ age mix).

The binormal model is not derived from individual-level data in this tool, although it reproduces the share of melanomas in each tenth of predicted risk for the QSkin tool in its own cohort, and for five tools in the independent 45 and Up Study, to within a few percentage points (Methods, section 1). It assumes the model is calibrated in each region (a score cut per SA4). The published models have no area term; with one national cut, 88 rather than 86 SA4s are feasible at 3% and 84 rather than 66 at 2% (MP16, specialist capacity model); the difference is mainly high-risk Queensland and coastal NSW areas. Which SA4s are infeasible depends on this choice. In the 45 and Up Study (Reyes-Marcelino et al., JAMA Dermatology 2026), four of the five tools predicted only 31% to 57% of the melanomas that occurred; the Olsen 2018 invasive-and-in-situ model was close. ATOF's threshold is a calibrated risk, so a tool's raw score would need recalibrating to the population before it is used to decide who is offered checks.

What this means
Demand and net benefit move together when you change model. Exact pop-above-threshold counts depend on the binormal and calibration assumptions and remain approximate.
4

The workforce capacity model uses round-number constants

Moderate impact
ConstantValueSource
Annual consults per dermatologist FTE2,800ATOF assumption, not a published rate. Lindsay 2026 gives no consultations-per-FTE figure, only a hypothetical maximum of 2,880–5,760 skin-cancer procedures a year for one full-time dermatologist.
Annual consults per skin-cancer GP FTE1,200ATOF assumption, not a published rate
Screening fraction of clinical time30%Carved out from reactive workload
Screening cadence2 yearsBreastScreen; MSAC 1699 (National Lung Cancer Screening Program), a screening precedent
Risk horizon10 yearsAustralian risk tools predict 10-year
What this means
These are central anchors with real-world spread. Test ±20% on each constant to see how sensitive a given finding is. The 30% screening-fraction anchor is the largest single point of leverage.
5

Modelled-extension workforce scenarios are speculative

High impact

Two of the five workforce scenarios in the picker are modelled extensions, flagged with an amber ~m badge:

  • Melanographer / Total Body Photography (TBP). We multiply effective derm throughput by 2.0× on the rationale that TBP decouples image capture (technician, high throughput) from image reading (specialist, async, batched). Defensible from first principles but awaiting ACEMID empirical data.
  • Tiered hub-and-spoke. GPs triage ~70% of patients locally at 1.5× normal throughput; specialists confirm ~30% at 0.5× normal load. Models the Roadmap's “tiered workforce” recommendation. The multipliers are nominal assumptions.
What this means
Numbers for these two scenarios depend on multipliers that have not been empirically validated. They illustrate the shape of what these workforce models could deliver, not the magnitude.
6

Workforce projection is a trend, per person aged 40+

Moderate impact

The projection slider scales capacity by the projected clinical FTE per person aged 40+ in each state group, from the workforce projection (NHWDS 2013–2024). Dermatologist FTE grows about 3% a year, the pre-specified winner of a 2019–24 hindcast; GP FTE grows about 1.6% a year. Both are set against ABS population projections, so dermatologist supply per person rises only about 1.35-fold by 2040 and GP supply per person aged 40+ is flat. Practice nurses are not projected.

What this means
Growing capacity 3% a year while holding demand at today's population would overstate future feasibility; setting supply against projected population avoids this. Beyond about 2030 the answer depends on training intake, which the trend cannot foresee. The pre-specified cohort model (which models entries and exits) performed worse in the hindcast and is reported, not used.
7

Overdiagnosis adjustment inherits literature uncertainty

High impact

Named reference points in the OD slider: Lindsay 2024 in situ (73.5% for 2021 and 68.5% for 2017, the midpoints of the published 71–76% and 67–70%); Lindsay 2024 thin invasive (stage IA) 28% (published estimates 22–34% across 2017 and 2021); Glasziou 2020 all-melanoma women 54%, men 58%; Bjørch 2024 international range; Adamson 2024 US estimate.

Lindsay 2024 estimated overdiagnosis for melanoma in situ and for thin invasive (stage IA) melanoma only, not for stage IB or higher. The cost-utility model applies the 28% figure to all stage I detections. That is an ATOF extension, not a finding of the paper.

We deflate net benefit by sens × prevalence × OD-rate × 1/(1-t). This is the standard Vickers-Elkin extension but inherits all of the OD-estimation uncertainty in the underlying literature, which is non-trivial: estimates vary widely by definition (in-situ vs invasive, thin vs thick), country, and ascertainment method.

What this means
The choice of OD rate dominates the deflation. Read the chart at several OD anchors, not just one.
8

Incidence and population are static snapshots

Low impact

Each region's average risk is the national average multiplied by its age-standardised incidence ratio for invasive melanoma (Australian Cancer Atlas 2.0, 2010–2019, SA2 aggregated to SA4) and by its age mix within adults 40+. The crude incidence shown on the map (invasive + in situ per 100,000 residents) is for information only and does not drive demand. Eligible adults are each region's residents aged 40 and over (ABS Estimated Resident Population at 30 June 2024).

No time-trend on incidence; the eligible population does not age forward when the workforce-year slider moves.

What this means
Reasonable for short projection horizons (≤5 yr). For 10-yr projections, ageing of the eligible population and any incidence trend should be modelled explicitly.
9

The national average risk is a single estimate

Moderate impact

The default average 10-year melanoma risk for adults 40+ is 1.25%: the national risk of a first invasive melanoma, from AIHW Cancer data in Australia 2025 age-specific incidence (2025 projections) and ABS population at June 2025, less an estimated 14% of invasive melanomas that are second or later primaries (Methods, section 3). Each region's risk scales it by the region's invasive-melanoma SIR and age mix. The slider in the decision-curve section sets it between 0.5% and 5%, and every view (map, sidebar, Workforce Lab) follows it.

The estimate is uncertain: the share of second primaries is estimated rather than published, the 2025 rates are projections, and deaths from other causes are ignored, as they are when the risk tools are built (allowing for them gives 1.10%). The plausible range is 1.12–1.32%. Higher-risk groups sit well above it: in the Queensland QSkin cohort aged 40–69 it was 1.7%. See limitation 14 for what this value does to national totals and the cost per QALY.

10

Equity stratification is descriptive, not causal

Moderate impact

The Equity Reveal section stratifies the per-region minimum feasible threshold by SEIFA quintile (ABS Index of Relative Socio-Economic Disadvantage) or remoteness (Modified Monash categories MM1 to MM7). The 1.8× disparity it shows at the default settings (Q1 median min threshold 1.6% vs Q5 0.9%) is descriptive: it shows that the workforce maldistribution correlates with disadvantage.

What this means
It does not establish that ATOF's adaptive threshold is the only (or best) policy lever to address the inequity. Workforce redistribution, alternative delivery models, and demand-side interventions are all complementary policy options.
11

The tool is not a microsimulation

High impact

ATOF is a decision-analytic dashboard, not a microsimulation or other individual-level simulation model. It does not:

  • track individual patients over time
  • model uptake / adherence heterogeneity
  • follow each round of checks (the share of melanomas found by a check is a long-run average for the interval)
  • model recall or the biopsy cascade beyond a fixed number of biopsies and excisions per check and the overdiagnosis adjustment
  • carry uncertainty through every calculation (the Threshold Advisor, the national results and the Value per person page do; the capacity map and the equity views do not)
  • produce a submission-grade ICER (the Value per person ICER and the Lab's cost per QALY are illustrative)
What this means
McLoughlin et al. (PLoS One 2025) is the microsimulation arm; ATOF is complementary. The two together cover both feasibility (microsim) and operationalisation (ATOF).
12

The risk-prediction models themselves have known weaknesses

Moderate impact
  • QSkin MP16 is QLD-derived; development AUC 0.74–0.76 fell to 0.69 (0.68–0.71) in the 45 and Up external validation, which the tool uses.
  • MIA/Vuong (AUC 0.70) is the reference for 45-and-Up validation; parsimonious enough to deploy but trades discrimination for simplicity.
  • NCI MBRAT (AUC 0.70) is US-derived; ascertainment differs.
  • Cho 2005 (AUC 0.66) is the oldest and lowest-AUC model included; kept for historical comparison.
What this means
AUCs are used as point estimates in every view except the Regional advisor, which draws them from their confidence intervals.
13

Technical caveats

Low impact
  • Desktop-only. Best viewed on screens ≥ 1280px wide. A small-screen breakpoint shows a static summary instead.
  • Provenance pills only on the main map. The four equity-geography small-multiples do not show individual provenance. All SA4s share one workforce method (see limitation 2), so provenance does not vary by region.
  • Performance on lower-end machines. The 4-up small-multiples render ~350 SVG paths simultaneously and can be slow on older hardware.
  • Cookie/analytics consent is deferred until after the tour dismisses on first visit. A fresh visitor may not see the consent banner until ~1.5s in.
14

The cost-effectiveness results rest on modelling choices

High impact

The cost per QALY (Value per person page and Workforce Lab cost panel) uses the 2026 reference case of model version 2.0: 2026 prices, the immunotherapy era, health-system perspective, published PBS prices, treatment savings counted (Methods, section 6). At a 3% threshold the modelled health gain is a small difference between melanoma deaths averted and the harms of overdiagnosis and procedures: with 2-yearly checks the programme adds about 69 overdiagnosed melanomas for each melanoma death it averts, and with 5-yearly checks about 104. The result there, about $2,422,000 per QALY with 2-yearly shared primary care, is far from good value and very sensitive to the inputs below. At the settings the tool opens on (a 2% threshold and 5-yearly checks) the programme averts about 104 melanoma deaths over 10 years and adds about 10,800 overdiagnosed melanomas, and it loses more health than it gains. With shared primary care, only thresholds of 8.5% or more cost less than $50,000 per QALY (with 3-yearly checks), and only if risk assessment costs nothing; at $1 per adult no threshold is under $100,000 per QALY at any interval. The main limits and simplifications:

  • What checks do rests on observational contrasts from two Queensland studies. Three contrasts set the size of the benefit and of the harm. Adults whose skin had been examined in the previous 3 years, against adults whose skin had not, had melanomas 1.5 mm or thicker 0.74 times as often (95% confidence interval 0.61 to 0.90) and melanomas under 1.5 mm 1.29 times as often (Aitken 2010, a case-control study; the second figure combines two thickness groups), and in situ melanomas 1.45 times as often (1.09 to 1.92; Whiteman 2022, the QSkin cohort). Neither study was randomised, so people who have skin checks may differ from people who do not in ways that also affect what is diagnosed. The model takes 1.5 mm or thicker as stage II or later and under 1.5 mm as stage I. Across the interval for thick melanomas the result at 3% with 2-yearly shared primary care runs from about $304,000 per QALY to no health gain; across the interval for in situ melanomas, from about $644,000 per QALY to no health gain; across the range for thin melanomas, from about $1,269,000 per QALY to about $12,862,000 per QALY. The model lets stage IV at diagnosis fall with checking in the same proportion as stages II and III. If it did not, the same programme would avert about 75 melanoma deaths over 10 years, not 104, and its result would be no health gain.
  • The excess of thin and in situ melanomas in examined adults is all counted as overdiagnosis. Beyond the thick melanomas that checks find earlier, every extra stage I or in situ melanoma in examined adults is treated as one that would never have caused harm. If some of them would have progressed, the model overstates overdiagnosis and understates the benefit. If only 50% of that excess were overdiagnosis, and the rest melanomas diagnosed earlier within the studies' time window, the result at 3% would be about $714,000 per QALY.
  • Whether an earlier-found melanoma is found in situ or at stage I is an assumption. No data say which. The model takes 50% in situ. A melanoma found in situ cannot cause death in the model and one found at stage I still can, so the split matters: with all of them found at stage I the same result is no health gain, and with all found in situ about $1,065,000 per QALY.
  • How often those adults were checked is an estimate. The odds ratio describes adults examined at least once in 3 years. The model takes them to have had 0.55 examinations a year, from a 1998 survey of whole-body skin examinations by GPs in regional Queensland (Janda 2004). That is the lowest value the survey figures allow. If they were checked more often, each check is worth less: at 1 examination a year the same result is no health gain.
  • The detectable period and the sensitivity of a check are assumptions. A melanoma is taken to be detectable at a check for 2 years on average before it would surface on its own (range 1 to 4), and one check to find a detectable melanoma 80% of the time (60% to 95%). They set how the benefit changes with the interval between checks, so they matter when intervals are compared. Because the size of the effect is calibrated, they move the 2-yearly result less than the other inputs here: from about $2,450,000 per QALY to about $2,327,000 per QALY over the range of the detectable period. Melanomas that grow fast have a short detectable period and are less likely to be found by a check; the model represents this only through the spread of the detectable period around its mean.
  • How long an overdiagnosed melanoma stays detectable is an assumption. A melanoma that would never cause harm does not surface on its own, so it can be found at a check for much longer than one that would progress. The model takes 10 years on average (range 5 to 20); no study has measured it. It decides how the check intervals compare. With it, 5-yearly checks find about 72% of overdiagnosed melanomas but only 30% of the melanomas that would progress. At 3% with shared primary care, 5-yearly checks then lose more health than they gain, and 2-yearly checks have the lowest cost per QALY of the intervals offered. If it were 2 years, the same as for a melanoma that would progress, 5-yearly checks would have the lowest cost per QALY (about $1,362,000 per QALY). Over the range the 2-yearly result moves little, from about $2,243,000 per QALY to about $2,536,000 per QALY.
  • Adults are in just two groups. Usual care is 37% of adults with a whole-body skin check every year and the rest with none. Source: National Sun Protection Survey 2016-17: 37% of adults 45-69 had a whole-body skin check in the past 12 months (Reyes-Marcelino 2021); 43% in the 45 and Up Study 2020 (Reyes-Marcelino 2026). In practice people are checked at every frequency, some have only part of their skin or one spot checked, and some regular checkers go more than once a year. Between 30% and 45% of adults with regular checks the result runs from about $3,930,000 per QALY to about $1,706,000 per QALY.
  • Who attends. Regular checkers are assumed to be as common above the threshold as in the population, and to attend as often as other adults. Surveys find that people at higher risk of skin cancer are checked more often (Janda 2004; the 45 and Up Study in 2020), and people who already have checks may be more willing to attend. Either would lower the health gain, because a regular checker's programme checks replace their own. If regular checkers were 1.5 times as likely to attend as other adults, the reference programme would avert about 87 melanoma deaths over 10 years, not 104. The tool has no setting for this.
  • The stage mix of invasive melanoma is from 2011 diagnoses. Invasive melanomas diagnosed under usual care are split by stage with the national registry mix, the latest published (Cancer Australia NCCI / AIHW, 2011 diagnoses, stage I 78.0 / II 14.1 / III 3.0 / IV 2.1%, unknown excluded). Stage at diagnosis may have changed since then. The in situ share is more recent (Australian Cancer Atlas 2010-2019 (persons): 165,641 in situ and 134,543 invasive melanomas). What regular checkers and other adults are diagnosed with is not observed separately: the calibration splits the population figures between them.
  • The QALY loss per biopsy or excision is an assumption. Each procedure after a check is charged 0.0011 QALY (range 0.0005 to 0.003). Source: Utility 0.971 after a benign excision (Chen 2004, in Watts 2017), taken for two weeks (assumption). It matters because a programme adds many procedures for each melanoma found earlier: over that range the result runs from about $973,000 per QALY to no health gain. Time lived with advanced melanoma is charged 0.3 QALY per episode, also an assumption.
  • No age structure. Every adult aged 40 and over is given the same detectable period, stage at diagnosis, survival by stage and health gain per death averted; only the risk of melanoma differs between people. The model cannot tell a programme's effect at 45 from its effect at 80, and it does not remove people who die of other causes during the programme.
  • Photography with AI. AI triage is given the same chance of finding a melanoma as a clinician's check, which has not been established in Australian primary care. The share of flagged adults who go on to a GP visit and a procedure (50%, range 30% to 90%) is an assumption.
  • Relapse is approximated. Advanced-melanoma episodes (stage IV at diagnosis and every relapse) are counted as deaths ÷ P(death | advanced), not modelled relapse by relapse.
  • Survival basis and calibration. 10-year melanoma-specific survival stands in for lifetime survival. Published mortality by stage is divided by the share of that stage's diagnoses that are progressive, because an overdiagnosed melanoma cannot cause death, and one factor (k = 0.76) scales every stage so that melanoma deaths per invasive melanoma diagnosed under usual care equal 8.3%, the Australian ratio of melanoma deaths to invasive diagnoses (AIHW). The calibrated advanced-disease survival (45%) is above the 28% figure for all distant-stage patients. Published AJCC8 10-year survival (stage I 0.95, II 0.84, III 0.69) is an upper-bound reference only; it is not offered as a switch in the tool.
  • Published PBS prices. Medicines are costed at published prices (the MSAC convention). Confidential rebates lower the real prices, so the treatment savings are an upper bound: at a 35% average rebate the result at 3% with 2-yearly shared primary care moves from about $2,422,000 per QALY to about $2,618,000 per QALY.
  • Health gain per death averted, and the harm of an overdiagnosis. Each melanoma death averted is worth 7.3 discounted QALYs (range 6.4 to 8.6), from the age at which Australians die of melanoma (mean 75.4), not from the age of the people screened. If the deaths a programme prevents are younger than melanoma deaths in general, the gain is larger. Each extra overdiagnosed melanoma is charged the treatment of its stage and 0.05 QALY, the same for in situ and stage I (range 0.02 to 0.26); no Australian utility study of overdiagnosed or in situ melanoma was found, so this value rests on stage I and II patients. Because a programme adds many overdiagnosed melanomas for each death it averts, this is now one of the inputs the result depends on most: over its range the result at 3% runs from about $646,000 per QALY to no health gain. The probabilistic analysis draws both.
  • Attendance is 50% by default. Half of the adults above the threshold are assumed to attend their checks, close to BreastScreen Australia (about 50%). No invited risk-based skin check programme has measured it. People who do not attend are assumed to carry on as today, every adult is still risk-assessed, and drop-out between rounds is not modelled.
  • Ages. A narrower age range changes the number of eligible adults and their average risk, but not the risk tool's accuracy (lower inside a narrow age band, because age is one of its strongest predictors) or the health gained per death averted (higher when the people checked are younger). The tools were validated in adults aged 45 and over, so starting at 40 is an extrapolation.
  • Adults below the threshold. By default they keep today's skin checks. The setting changes how many checks the regular checkers among them have, and the change is valued with the same model as programme checks: melanomas found earlier or later, and overdiagnosed melanomas, biopsies and excisions added or avoided. For the reference programme, 10% fewer checks saves about $457 million over 10 years and gains about 591 QALYs, with about 17 more melanoma deaths; more checks do the opposite. Both effects rest on the inputs above: what checks do, and the QALY losses per overdiagnosed melanoma and per procedure. In QSkin, extra screening raised biopsies and in situ melanoma more than it changed invasive melanoma.
  • The national range holds the settings fixed. The 95% range on the national results draws the uncertain cost and health inputs, not the settings, so it leaves out uncertainty about attendance, the risk-assessment cost and the risk tool's accuracy.
  • Not costed. Operable regional relapses (surgery and adjuvant therapy after a stage I–II diagnosis; leaving them out understates the savings slightly). Keratinocyte cancers found at the extra checks are costed at 14% of their treatment cost by default (the cost of treating them 1 to 3 years earlier), since most would have been treated later anyway; the full cost is a setting. Only excisions in general practice within 3 months are counted, not treatment by freezing, by specialists or in hospital.
  • Assumed unit costs. Practice-nurse and photography + AI check costs, and the risk-assessment cost ($5 per eligible adult, one round every 10 years), are assumptions, as are a few treatment components (out-of-pocket share, BRAF test, brain MRI fee). So are the clinician minutes for nurse checks, photography sessions and the procedures that follow checks.
  • Illustrative price index. Older costs are brought to 2026 with the AIHW total health price index, extrapolated two years beyond the last published year; it is a general health index, not specific to these services.
  • National figures depend on the average risk. National counts (people above the threshold, checks, melanomas, deaths, advanced-melanoma episodes) and dollar totals scale with the average 10-year risk (default 1.25%, limitation 9). Per-person results (the Value per person page) and the calibration, which is defined per melanoma diagnosed, do not. The national cost per QALY depends on it strongly, because the cost of risk-assessing every adult 40+ (about $65 million over 10 years, discounted) is fixed while the number of adults above the threshold is not. For 3%, 2-yearly shared primary care, 0.85 million adults are above the threshold at 1.25% and 2.41 million at 2%, and the result is about $2,422,000 per QALY and about $1,190,000 per QALY respectively; with the risk-assessment cost set to $0 it is about $1,395,000 per QALY and about $934,000 per QALY.
What this means
Read every cost per QALY as illustrative. At a low threshold the health gain is a small difference between deaths averted and the harms of overdiagnosis and procedures, so the result is unstable. Within their ranges, seven inputs can each remove the gain at 3% or all but remove it: the odds ratio for thick melanoma, how often the examined adults were checked, the ratio for in situ melanoma, whether earlier-found melanomas are found in situ or at stage I, the QALY loss per overdiagnosed melanoma, the QALY loss per biopsy or excision, and the share of the stage shift that is real. The ratio for thin melanoma, the cost of risk assessment and the share of adults who already have regular checks also move it a lot. The two detectable periods and the sensitivity of a check move the 2-yearly result least, but the detectable period of overdiagnosed melanomas decides which check interval looks best. The Workforce Lab and the Value per person page let you vary the risk-assessment cost, the stage-shift share, the share of adults who already have regular checks and how far programme checks replace their own. The probabilistic analysis varies the rest.

Future directions

The data sources, collaborations, and modelling extensions that would advance ATOF from the current draft to a more comprehensive analytic framework.

1

Individual-level 45 and Up Study sensitivity-specificity reconstruction

Retire the binormal approximation entirely. Reconstruct each model's sensitivity-specificity curve directly from individual-level 45 and Up risk scores and outcomes behind Reyes-Marcelino et al. (JAMA Dermatology 2026).

Outcome: the decision curves become citation-ready and the “modelled from published AUCs” caveat retires.

2

Harden the national workforce replication

ATOF applies the Lindsay 2026 supply method, published for Queensland only, to every state. The next steps are official per-state accredited-GP counts from SCCA, calendar-2023 MBS skin-cancer item counts by state from Services Australia (to add Lindsay's demand-per-clinician layer), and the Cancer Atlas's official SA2 estimates with credible intervals.

Outcome: the remaining assumptions become either sourced counts or quantified uncertainty bands.

3

ACEMID empirical multipliers for TBP/melanographer scenario

The 2.0× dermatologist-throughput multiplier for the Melanographer scenario is a first-principles assumption. Empirical multipliers from the Australian Centre of Excellence in Melanoma Imaging and Diagnosis (ACEMID) would replace it.

Outcome: the “~m” modelled-extension badge can come off the TBP scenario.

4

Yearly workforce trajectories per state

The projection slider follows state-group trends in clinical FTE per person aged 40+ (the workforce projection, NHWDS 2013–2024), not a uniform national rate. It cannot foresee changes in training intake, and the pre-specified cohort model performed worse than the trend in the hindcast. Planned training intake by state, or a cohort model refitted on more years of data, would let the slider track planned workforce growth. (Lindsay 2026 is a 2023 snapshot of Queensland and has no trajectories.)

Outcome: projection numbers stop being a global anchor and become spatially specific.

5

Confidence intervals throughout the sidebar (partly built)

The Threshold Advisor propagates uncertainty in supply bands, the GP skin-cancer share, the workforce projection, area risk, AUC and the screening share of capacity by Monte Carlo, region by region, and the cost-utility explorer has a probabilistic analysis, which the national results in the planner and the Workforce Lab also run. Still to do: carry the same uncertainty to every sidebar metric, the capacity map and the equity views.

Outcome: users can read the tool's claims with statistical context, not just point estimates.

6

Cost-effectiveness: an illustrative cost-utility explorer (built)

An illustrative version is available. The cost-utility explorer prices each 10-year-risk band's care-pathway options (from usual care up to specialist 6-monthly total-body photography) in cost, QALYs and net monetary benefit (NMB = λ·QALY − Cost), the dynamic form of Wilson 2018's Table A10.1.

Benefit runs through avoided mortality: checks find thick melanomas earlier. Since version 2.0 the melanomas diagnosed under usual care come from Australian registries, and what checks do is calibrated to Australian studies of examined against unexamined adults (Aitken 2010, Whiteman 2022). An overdiagnosis penalty (Lindsay 2024, and the extra thin and in situ melanomas found in examined adults) is charged as cost + disutility with no survival gain, and each biopsy or excision after a check costs a little quality of life. Costs are 2026 Australian prices (MBS, PBS, the National Efficient Price), with older sources indexed; a shared-input-draw probabilistic analysis produces a cost-effectiveness acceptability curve, so the genuine decision uncertainty is shown, not hidden.

The 2026 reference case. The comparator is usual care, not nothing: 37% of adults already have a whole-body skin check every year, with the biopsies and benign excisions that follow, and the rest have none. Programme checks replace a regular checker's own checks. GP checks use Goldsbury 2024's per-check procedure rates. Melanomas detected during the 10-year programme carry lifetime consequences (follow-up and end-of-life costs, QALYs lost), all discounted to programme start. The stage-shift benefit is computed on progressive melanomas only (overdiagnosed lesions removed) and scaled by an explicit lead-time realisation factor (base 75%, varied 50–100% in the PSA and adjustable on the page). Still open: the health gain per death averted uses the ages at which Australians die of melanoma, not the ages of deaths a programme would prevent.

Treatment costs. The reference case also counts the melanoma treatment that earlier diagnosis avoids in the immunotherapy era: treatment at the stage found, an advanced-melanoma episode for each stage IV diagnosis or relapse, and end-of-life care, with 10-year survival calibrated to Australian melanoma deaths (Methods, section 6; limitation 14).

Two caveats. The figures are illustrative, not a submission ICER: the stage shift is observational and not yet calibrated to a natural-history model, and the lead-time factor is an assumption until it is. And ATOF is complementary to Policy1-Melanoma (McLoughlin, Watts, Cust): the definitive cost-effectiveness estimate is Policy1's microsimulation; ATOF is a fast, transparent explorer whose stage-shift is designed to be calibrated to it.

Outcome: a decision-maker can explore the threshold-and-interval trade-off in the language MSAC uses, without mistaking the explorer for an MSAC-grade estimate.

7

Uptake and adherence modelling

ATOF has one attendance setting: the share of adults above the threshold who attend their checks (50% by default, close to BreastScreen Australia, which reaches about 50% of eligible women; the National Bowel Cancer Screening Program reaches about 41% of the people it invites, AIHW). An adherence model, with attendance that differs by age, region and round, would let programme designers compare realistic with theoretical demand.

Outcome: the demand calculations become operationally meaningful, not just theoretical.

8

Microsimulation hand-off

A formal hand-off interface to the McLoughlin et al. PLoS One 2025 microsimulation. ATOF would surface programme-level scenarios; the microsim would simulate patient-level outcomes under each.

Outcome: a complete programme-design loop: pick a scenario in ATOF → simulate outcomes in microsim → feed cost back to ATOF for HTA.

9

Equity overlays beyond SEIFA and remoteness

Add Indigenous status (AIHW Indigenous Health metrics), Modified Monash (DPA mapping), language other than English (ABS), and disability prevalence. Stratify the equity reveal on each.

Outcome: the equity conversation broadens beyond socioeconomic disadvantage to include structural and cultural dimensions.

10

Peer review and external validation

Submit a methods paper documenting ATOF's decision framework, validate the outputs against an independent cohort (45-and-Up + QSkin), and publish the codebase under an open license.

Outcome: ATOF becomes a citable tool, not a private dashboard.