← ATOF interactiveMethods · v0.1

Methods & Citations

The Adaptive Threshold Optimisation Framework (ATOF) brings together three inputs to the choice of a risk threshold for Australia's national targeted skin cancer screening programme: clinical net benefit, overdiagnosis harm and regional workforce capacity. A cost-utility model adds the cost per quality-adjusted life year (QALY). This page sets out the methods, parameters and sources behind every number in the tool.

Author: Dr Yagiz Aksoy · The Daffodil Centre · University of Sydney & Cancer Council NSW · LinkedIn. Dr Aksoy is a recipient of the Viertel Clinical Investigator Award. A manuscript describing ATOF is in preparation.

1. Decision curve analysis

Net benefit follows Vickers and Elkin (2006):

NB(t) = TP/N − (FP/N) × t / (1 − t)

where t is the risk threshold, TP/N and FP/N are the true and false positives per person assessed, and t / (1 − t) is the weight that converts a false positive into the same units as a true positive. Checking everyone gives NB = π − (1 − π) × t / (1 − t), and checking no one gives 0, where π is the average 10-year risk (section 3).

Each model's published area under the curve (AUC) is turned into sensitivity and specificity with a binormal model: risk scores of future cases and non-cases are normal with equal variance and means ±d/2, where d = √2 × Φ⁻¹(AUC). A person's predicted risk is the probability of being a case given their score, so the score cut-point for threshold t is s = [logit(t) − logit(π)] / d. Sensitivity is 1 − Φ(s − d/2) and the false-positive rate is 1 − Φ(s + d/2). The same model gives the share of adults above any threshold, which drives demand (section 5) and the national programme estimates (section 6).

Check against published data. Whiteman 2025 reports, for the QSkin tool in its own cohort (41,919 adults, 706 invasive melanomas over 10 years), how many melanomas fell in each tenth of predicted risk. At the cohort's AUC (0.74) and average risk, the binormal model gives a cumulative share of melanomas within 3.5 percentage points of the published one at every tenth:

Highest-risk tenths includedtop 10%top 20%top 30%top 40%top 50%top 60%top 70%top 80%top 90%top 100%
Share of melanomas, QSkin (published)33%48%64%74%82%90%94%97%99%100%
Share of melanomas, binormal model35%52%64%74%81%88%92%96%99%100%

The QSkin figures come from the cohort the tool was built in, so its accuracy there is higher than in external validation (AUC 0.69 in the 45 and Up Study, which the tool uses). The check tests the shape of the binormal assumption, not the accuracy of the tool. Source: Whiteman 2025, Supplement 1 eTable 7 (QSkin cohort, 10 years, invasive melanoma).

Second check: an independent cohort. The 45 and Up Study validation (Reyes-Marcelino et al., JAMA Dermatology 2026), which supplies the AUCs the tool uses, plots for five tools the share of participants who developed melanoma in each tenth of predicted risk. With tenths of equal size, that gives the share of the cohort's melanomas in each tenth. At each tool's AUC and the cohort's average risk, the binormal model is within 2.2 percentage points of it at every tenth, for every tool:

Tool (AUC in 45 and Up)Top 10%: observed / modelTop 30%: observed / modelLargest gapMelanomas predicted per one observed
QSkin MP16 (Whiteman 2025) (0.69)26% / 27%56% / 56%1.5 points0.42
Olsen 2018, invasive (0.69)26% / 27%55% / 56%1.8 points0.31
Olsen 2018, invasive and in situ (0.70)28% / 28%57% / 57%1.0 points1.15
Vuong 2016 (0.70)28% / 29%57% / 58%2.2 points0.57
Cho 2005 (0.66)22% / 24%54% / 52%1.8 points0.45

The proportions were read from the study's calibration figure. The same figure shows that most tools under-predict absolute risk (last column): four of the five predicted 31% to 57% of the melanomas that occurred, and only the Olsen 2018 invasive-and-in-situ model came close. ATOF's threshold is therefore a calibrated risk, not a tool's raw score (Limitations).

Each region has its own average risk: the national average multiplied by the region's risk multiplier (section 5). The binormal model is applied within each region, and national true and false positives are the regional ones added up, weighted by adults aged 40 and over. The national decision curve, the number of adults above a threshold and the cost model therefore all use the same risk distribution.

⚠ The curves are modelled from published AUCs. A planned analysis will reconstruct sensitivity and specificity directly from individual-level 45 and Up Study data (Reyes-Marcelino et al., JAMA Dermatology 2026).

2. Risk prediction models

Five models are available. Four have external-validation AUCs from the 45 and Up Study (Reyes-Marcelino et al., JAMA Dermatology 2026); NCI MBRAT, which was not part of that validation, keeps its development-study AUC. QSkin MP16 (AUC 0.69) is the default.

ModelAUCCohortVariablesCitation
QSkin MP16
Development AUC 0.74-0.76; external validation in 45 and Up 0.69 (0.68-0.71). MP16 = 16-term QSkin model.
0.69QSkin Sun and Health Study (Queensland)14 variables · Age (and age squared), Sex (and age-by-sex interaction), Ancestral risk of melanoma …Whiteman DC 2025
QSkin MP7
Parsimonious QSkin model. External AUC is for the Olsen 2018 invasive model (invasive + in situ model: 0.70, 0.68-0.71).
0.69QSkin Sun and Health Study (Queensland)7 variables · Age, Sex, Tanning ability …Olsen CM 2018
MIA/Vuong
Used as reference model in 45 and Up Study external validation.
0.70Australian Melanoma Family Study (development), 45 and Up (validation)7 variables · Age, Sex, Hair colour …Vuong K 2016
NCI MBRAT
US-derived; not part of the 45 and Up external validation.
0.70US case-control (Philadelphia, San Francisco)6 variables · Age, Sex, Sunburn history …Fears TR 2006
Cho
0.66Nurses' Health Study + Health Professionals (US)6 variables · Age, Sex, Family history of melanoma …Cho E 2005

3. Average 10-year melanoma risk

The decision curves, the demand model and the national cost panel all need π, the average 10-year melanoma risk among eligible adults. The risk models predict a first invasive melanoma in people who have not had melanoma, with deaths from other causes censored, so π is defined the same way. The default is 1.25%, derived as follows:

  1. Age-specific incidence of invasive melanoma (ICD-10 C43), persons, from AIHW Cancer data in Australia 2025 (Book 1a, Table S1a.1). The 2025 projected rates are used because AIHW reports that 2020–21 incidence was depressed by COVID-19.
  2. For each single year of age a from 40 (ABS estimated resident population, 30 June 2025), the 10-year cumulative hazard is H(a) = Σ rate(a + k) for k = 0 to 9, using the 5-year age-group rate at each attained age, and the risk is R(a) = 1 − e−H(a). The population-weighted mean for ages 40+ is 1.45%.
  3. About 14% of invasive melanomas (range 10–18%) are second or later primaries in people who have already had melanoma. This share is estimated, not published: it combines the prevalence of previous melanoma in the QSkin cohort, scaled to national incidence, with a relative risk of about 5 for a subsequent primary (Youlden 2014), and is cross-checked against national survivor counts. Removing them gives 1.45% × 0.86 = 1.25%.

With 2021 actual rates the result is 1.18%, and with 2019 rates 1.26%; the plausible range is 1.12–1.32%. Allowing for deaths from other causes (ABS 2024 death rates) gives 1.10%, but this is not used because the risk models do not allow for them. As a check, the same method gives about 1.0–1.1% for ages 40–69 nationally, against 1.7% observed in the Queensland QSkin cohort (Queensland's age-standardised incidence is about 1.3 times the national rate, and the cohort is mostly of European ancestry).

Each region's average risk is πr= π × the region's invasive-melanoma risk per resident aged 40+ (Cancer Atlas 2.0 age-standardised SIR, 2010–19, × the region's age mix within 40+), scaled so that the eligible-weighted mean of πr is π. The slider in the decision-curve section changes π for every view in the tool.

4. Overdiagnosis adjustment

Net benefit can be reduced by an overdiagnosis rate, so that melanomas found by screening that would never have caused harm count against the programme instead of for it:

NB'(t) = sens × π × (1 − OD) − [FPR × (1 − π) + OD × sens × π] × t / (1 − t)

The same adjustment is applied to checking everyone (sens = FPR = 1), so the comparison stays like for like. The risk tools and the average risk count invasive melanoma, so the default rate when the adjustment is switched on is 19%, the midpoint of Glasziou 2020's estimates for invasive melanoma (15% in women, 22% in men). Estimates for all melanoma include melanoma in situ, which the tools do not predict. This switch applies to the decision curves and the sidebar; the cost model has its own stage-specific overdiagnosis shares (section 6). Reference values:

  • 74%: Lindsay 2024, in situ (2021). in situ · AU (range 71–76%)
  • 69%: Lindsay 2024, in situ (2017). in situ · AU (range 67–70%)
  • 28%: Lindsay 2024, thin invasive (stage IA). thin invasive · AU (range 22–34%)
  • 15%: Glasziou 2020, invasive melanoma (women). invasive · AU
  • 22%: Glasziou 2020, invasive melanoma (men). invasive · AU
  • 54%: Glasziou 2020, all melanoma (women). all melanoma · AU
  • 58%: Glasziou 2020, all melanoma (men). all melanoma · AU
  • 29%: Bjørch 2024, international low. all melanoma · INTL
  • 60%: Bjørch 2024, international high. all melanoma · INTL
  • 50%: Adamson 2024, US (men). all melanoma · US
  • 65%: Adamson 2024, US (women). all melanoma · US

5. Workforce capacity

Capacity map

Each region's annual capacity for screening checks comes from its dermatologist and skin-cancer GP clinical FTE:

capacity = (D × 2800 + G × 1200) × 0.3

D is dermatologist clinical FTE in the region (NHWDS 2024 clinical hours ÷ 40; regions with 0–3 dermatologists are published only as a band, and get the band midpoint scaled by one common factor so that together they match the published national total minus the other regions). G is skin-cancer GP FTE: GP clinical FTE × the state's skin-cancer share of GP time (section 7). 2800 and 1200 are annual consultations per FTE (ATOF assumptions, not published rates), and 30% is the share of consultations available for screening, an assumption that the Threshold Advisor varies from 15% to 45%.

Demand is the number of eligible adults above the threshold, checked every 2 years:

demand = eligibler × [πr × sens(t) + (1 − πr) × FPR(t)] / 2

Sensitivity and FPR come from the same binormal model as the decision curves, at the selected model's AUC. Eligible adults are each region's resident population aged 40 and over (ABS estimated resident population, 30 June of the workforce year). A region is feasible when capacity is at least demand. Annual melanomas found (sidebar) are sens × π × eligible adults ÷ 10.

ScenarioStatusMultiplierRationale
Dermatologists onlyempirical-NHWDS 2024 dermatologist clinical FTE (hours/40; small cells banded) × 2800 consults/yr × 30% of capacity for screening.
Skin cancer GPs onlyempirical-NHWDS 2024 GP clinical FTE × state skin-cancer share (Medicare calibration) × 1200 consults/yr × 30% screening fraction.
Combined (Derm + GP)empirical-Sum of Derm + GP capacity; represents the practical Australian baseline.
Melanographer / TBPmodelledderm × 2Total Body Photography decouples image capture (technician, high throughput) from image reading (specialist, async, batched). Effective derm throughput is doubled (×2.0), because the bottleneck shifts from physical consult time to image-reading time. This is an assumption pending empirical data from ACEMID.
Tiered hub-and-spokemodelledGP × 1.5, derm × 0.5Tiered model where GPs triage ~70% of patients locally and route ~30% to specialist confirmation. GP throughput ×1.5 (focused on dermoscopy), specialist load ×0.5 (only complex cases).

Workforce projection

The workforce projection slider scales dermatologist and GP capacity by the projected clinical FTE per person aged 40+ in each state group, because demand grows with the population. Dermatologist FTE grows 3% a year (the pre-specified trend model that best predicted recent years from earlier ones), GP FTE 1.6% a year, and the population aged 40+ follows ABS Series B projections. National median growth per person aged 40+: dermatologists 1.2% a year, GPs -0.3% a year.

Workforce Lab (delivery models)

The Workforce Lab asks what share of each local workforce's clinical time the programme would take under a chosen delivery model. For region r and workforce k:

checks_r = attendance × N_r(t) / interval          N_r(t): people above threshold t (same demand model as the map)
check hours_rk = checks_r × m_k × minutes_k / 60    m_k: share of first-line checks done by workforce k (the delivery mix)
    minutes_k: the minutes setting for a GP or dermatologist (15); × 25/15 for a nurse check; × 30/15 for a photography session
procedure hours = checks_r × m_k × f_k × 5.7 / 60    minutes of biopsies, excisions and results consultations per GP check
    f_k: procedures relative to a GP check (dermatologist 0.5, skin-cancer GP 0.7, other GP 1, nurse 1); after a nurse check a GP does them
flagged photographs: checks_r × photo share × AI flag × 0.5 × (minutes + one procedure with its results consultation) / 60, GP time
derm review hours = checks_r × [referral × (GP + nurse share) × minutes + AI flag × photo share × review minutes] / 60
share_rk = hours_rk / (FTE_rk × 40 h × 46 wk)   region feasible if the largest share_rk ≤ the time cap
(38 h a week for practice nurses, whose NHWDS FTE is on a 38-hour week)

Supply is dermatologist and GP clinical FTE from NHWDS 2024 (hours ÷ 40, main-job SA4), skin-cancer GPs as above, and practice nurses from NHWDS 2024 nurses whose main job area is primary care (not projected). Dermatologist review can be local or pooled across the state or nation (teledermatology), in which case the review load is shared across the pool. Minutes per check, referral and AI-flag rates, tele-review minutes and the time cap are scenario settings, not estimates. The lowest feasible threshold is found on a 0.25% grid from 0.5% to 15%.

Time for procedures, nurse checks and photography. The biopsies, excisions and results consultations that follow checks take clinician time: a biopsy 20 minutes, a benign excision 30 and a results consultation after each 10 (assumptions), which is about 5.7 minutes per GP check at the Goldsbury 2024 rates. After a dermatologist's or GP's check that time is theirs. After a nurse check, or a photograph the AI flags, it is a GP's: shared between the GPs in the delivery mix, or taken from other GPs when the mix has none. Of the adults whose photographs are flagged, 50% see a GP in person and have a procedure. A nurse check takes 25 minutes and a photography session 30 when a GP check takes 15, the same minutes that set their prices. Attendance scales the checks and every hour that follows from them.

Ages offered. By default every adult aged 40 and over is offered risk assessment. A start age (40, 45 or 50) and an age at which it stops (70, 75, 80 or no limit) change, in each region, the number of eligible adults (the ABS population in those 5-year age groups, 2024) and their average risk: the region's risk multiplier times the mean national 10-year risk of its eligible age groups divided by the mean for all its adults aged 40+ (risk by age from the same AIHW calculation as section 3). With the default ages both factors are 1. The risk tool's accuracy and the health gained per death averted are not changed by the age range, although both depend on age.

Checks below the threshold. Adults below the threshold keep today's skin checks by default: 37% of them have a whole-body check every year and the rest have none. The setting changes how many checks the regular checkers among them have, by −25%, −10%, +15%, +35%. Personal risk information does not change screening attendance on average (meta-analysis of 19 trials, relative risk 1.00, 95% CI 0.97 to 1.03; Usher-Smith 2018), women told their breast cancer risk was low returned for screening as often as other women (Evans 2016), and in an Australian trial of melanoma genomic risk information skin examinations rose by a third in women but not in men (Smit 2021). So the default is no change, −10% and +15% are the range those studies support, and −25% and +35% are stress tests. The change is costed at today's GP check and follow-up prices and counted in GP time. Its health effect comes from the same model as programme checks (section 6): the share of their melanomas found by a check changes, thick melanomas move one way and overdiagnosed melanomas the other, and the number of biopsies and excisions that follow their checks changes too. With the QALY losses from overdiagnosis and procedures counted, fewer checks at low risk save money and gain QALYs while a few more people die of melanoma, and more checks do the opposite. At a 3% threshold, 10% fewer checks among them is about 447,000 fewer checks a year: about $457 million less cost over 10 years and 591 more QALYs, with 17 more melanoma deaths. 15% more is about 671,000 more checks a year: about $688 million more cost and 905 fewer QALYs, with 22 fewer melanoma deaths.

6. Health economics and national programme impact

A partitioned cost-utility model evaluates each 10-year risk band under each care option (usual care, a one-off GP check, GP checks every 5, 2 or 1 years, or 6-monthly specialist surveillance), over a 10-year programme with the lifetime consequences of melanomas detected in it. Costs and QALYs are discounted to programme start at 5%. The reference case uses 2026 prices and counts the melanoma treatment that earlier diagnosis avoids. This section describes version 2.0 of the model.

Version 2.0 (November 2026): what changed

  • Usual care is two groups. 37% of adults have a whole-body skin check every year and the rest have none. Version 1.0 used one average rate from Medicare data, 0.065 GP skin checks per adult a year, which left out skin-cancer clinics and dermatologists.
  • Diagnoses under usual care come from Australian registries. 55% of melanomas diagnosed are in situ (Australian Cancer Atlas 2010-2019), and invasive melanomas follow the registry stage mix. With the overdiagnosed shares of Lindsay 2024, about 50% of all melanomas diagnosed are overdiagnosed. Version 1.0 used two stage mixes from a study of people at high risk (Watts 2017).
  • What checks do is calibrated to Australian studies of examined against unexamined adults. The share of progressive melanomas found by a check before they would surface on their own is 30% with 5-yearly checks, 55% with 2-yearly checks, 72% with yearly checks and 84% with 6-monthly checks. In proportion to that share, thick melanomas fall. Overdiagnosed in situ and stage I melanomas rise with the share of overdiagnosed melanomas found, which is higher because they stay detectable for longer (72% with 5-yearly checks, 87% with 2-yearly checks and 93% with yearly checks). The sizes are set so that adults examined about 0.55 times a year have stage II or later melanomas 0.74 times as often and stage I melanomas 1.29 times as often (Aitken 2010), and in situ melanomas 1.45 times as often (Whiteman 2022). This replaces an assumed share of the stage shift for each interval.
  • A melanoma found earlier is found in situ or at stage I. No data say which. The model takes 50% in situ, and the probabilistic analysis draws the share anywhere from none to all.
  • Overdiagnosed melanomas are costed at the price of their stage, in situ or stage I. Version 1.0 costed all of them at the in situ price.
  • Programme checks replace a regular checker's own checks, one for one. A regular checker who attends is found no earlier unless programme checks are more frequent than yearly. The health gain comes from adults who are not checked today.
  • Risk assessment is one round in a 10-year programme by default. Version 1.0 costed a round every 5 years, although a repeat round has no modelled effect.
  • Keratinocyte cancers. By default only the cost of treating them earlier is counted (14% of the full cost). The full cost is a setting.
  • QALYs count two more harms. A small loss for each biopsy or excision that follows a check (0.0011 QALY) and time lived with advanced melanoma (0.3 QALY per episode).
  • Deaths are calibrated to Australian deaths per invasive melanoma diagnosed (8.3%, AIHW). Mortality by stage is per progressive melanoma: published mortality per diagnosed melanoma is divided by the share of that stage's diagnoses that are progressive.
  • Workforce. A nurse check takes 25 minutes and a photography session 30 when a GP check takes 15, and the biopsies, excisions and results consultations that follow checks take clinician time (about 5.7 minutes per GP check; section 5).
  • Photography with AI. Procedures follow the share of image sets the AI flags. Every flagged set is tele-reviewed by a dermatologist, and 50% of flagged adults then see a GP in person and have a procedure.
  • Checks below the threshold. The setting changes the checks of the regular checkers below the threshold, and counts their overdiagnosed melanomas, biopsies and excisions as well as melanomas found earlier or later (section 5).
  • Probabilistic analysis. It also draws who has regular checks, the three ratios for examined adults, how often those adults were checked, where earlier-found melanomas are found, the two detectable periods, the sensitivity of a check, procedures per check and the two new QALY losses. Every draw keeps melanoma deaths per invasive melanoma diagnosed at the observed value. Prices set by MBS fees and prices built bottom-up move separately.

The model

Usual care: regular checkers (share c, v_own checks a year) and other adults (no checks)
P(v) = s·(τ/T)·(1 − e^(−T/τ)) / [1 − (1 − s)·e^(−T/τ)],  T = 1/v
                                                 share of progressive melanomas found by a check at v checks a year, P(0) = 0
                                                 (τ = mean detectable period, s = chance that one check finds a detectable melanoma)
P_o(v): the same formula with τ_o in place of τ   share of overdiagnosed melanomas found by a check (τ_o = their mean detectable period)
Per progressive melanoma, for a group whose shares found by a check are P and P_o:
m_s(P) = m_s(0)·(1 − θ·P),  s = II, III, IV      progressive stage mix: thick melanomas fall in proportion to P
x(P) = θ·P·[m_II(0) + m_III(0) + m_IV(0)]        melanomas found earlier
m_insitu(P) = m_insitu(0) + λ·x(P);  m_I(P) = m_I(0) + (1 − λ)·x(P)      found in situ (share λ) or at stage I
o_s(P_o) = b_s + h_s·P_o,  s = in situ, I        overdiagnosed melanomas per progressive melanoma
Calibration (P̄ = c·P(v_own), P̄_o = c·P_o(v_own): the population averages; P_A = P(v_A), P_oA = P_o(v_A): the examined adults of the studies):
m_s(P̄) = R_s·(1 − α_s) / Σ_j R_j·(1 − α_j)       R: usual-care diagnoses by stage (registries); α: overdiagnosed shares (Lindsay 2024)
o_s(P̄_o) = R_s·α_s / Σ_j R_j·(1 − α_j)
θ = (1 − OR_thick) / P_A                         stage II or later OR_thick times as common in examined adults (Aitken 2010)
[m_I(P_A) + o_I(P_oA)] / [m_I(0) + o_I(0)] = OR_thin                       stage I diagnoses in examined adults (Aitken 2010)
[m_insitu(P_A) + o_insitu(P_oA)] / [m_insitu(0) + o_insitu(0)] = R_insitu  in situ diagnoses in examined adults (Whiteman 2022)
v_new = v_own − δ·min(v_own, v_prog) + v_prog    checks a year in the programme (δ = 1: programme checks replace own checks)
P_own = P(v_own),  P_new = P(v_new)              the programme changes each group's shares found by a check (P_o likewise)
p_mel = g·[1 − (1 − r)^(H/10)]                   progressive melanomas over the H-year programme (r = predicted 10-yr risk)
g = 1 / [1 − m_insitu(P̄) + o_I(P̄_o)]            progressive melanomas per invasive melanoma diagnosed under usual care
F_s = k·(1 − φ)·(1 − S10_s) / (1 − α_s)          10-yr melanoma mortality per progressive melanoma (lifetime proxy)
k:  g·Σ_s m_s(P̄)·F_s = deaths per invasive melanoma diagnosed under usual care (AIHW)
f(P) = Σ_s m_s(P)·F_s;  deaths = p_mel·[f(P_own) − ρ·(f(P_own) − f(P_new))]    ρ = share of the stage-shift benefit that is real
episodes = deaths / F_IV                         advanced-melanoma episodes: p_mel·m_IV(P_new) at diagnosis, the rest as later relapses
OD_s = p_mel·o_s(P_o,new)                        overdiagnosed melanomas at stage s (cost and disutility, no survival benefit)
Cost = p_mel·Σ_{s≤III} m_s(P_new)·C_s·DF_e + episodes·E·DF_adv + deaths·EOL·DF_d + (OD_insitu·C_insitu + OD_I·C_I)·DF_e
       + v_prog·Σ_t (1 + d)^−t·(c + follow-up procedures) + own checks not replaced·(c_GP + follow-up procedures)
QALY = Σ_t (1 + d)^−t − deaths·Q_death·DF_d − (OD_insitu + OD_I)·q_OD·DF_e − episodes·q_adv·DF_adv
       − q_proc·(procedures after programme checks and own checks, discounted)
Each result is the average of the two groups, weighted by c and 1 − c, against the same group with no programme.

Usual care. The comparator is today's skin checks. Adults are in two groups: regular checkers, a share c = 37% with 1 whole-body skin check a year, and adults with no regular checks. 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). The range used for c is 30% to 45%. Each usual-care check is priced as a GP check and is followed by a biopsy in 8.3% and an excision in 8.0% of checks (Goldsbury 2024).

What a check adds. A progressive melanoma can be found at a check for some time before it would surface on its own. That detectable period is exponential with mean τ = 2 years (range 1 to 4), and one check finds a detectable melanoma with probability s = 0.8 (0.6 to 0.95). Both are assumptions. With a check every T years, a melanoma that becomes detectable at a random time between two checks is still detectable at the next one with probability (τ/T)(1 − e^(−T/τ)), and a check that misses it leaves it to the following check with probability e^(−T/τ). Summing over checks gives P(v), the share of progressive melanomas found by a check: 30% with 5-yearly checks, 55% with 2-yearly checks, 72% with yearly checks and 84% with 6-monthly checks. The spread of the detectable period reflects how widely melanoma growth rates vary (Liu 2006, Arch Dermatol 142:1551-8). τ and s set how the benefit changes with the interval between checks; they do not set its size. An overdiagnosed melanoma would never surface on its own, so it stays detectable for much longer: τ_o = 10 years on average (range 5 to 20), an assumption. The same formula with τ_o gives P_o(v), the share of overdiagnosed melanomas found by a check: 72% with 5-yearly checks, 87% with 2-yearly checks and 93% with yearly checks. So 5-yearly checks find most of the overdiagnosed melanomas but under a third of the progressive ones, and the balance of harm and benefit depends on the interval.

Diagnoses under usual care. What is diagnosed today comes from Australian registries. Australian Cancer Atlas 2010-2019 (persons): 165,641 in situ and 134,543 invasive melanomas, so 55% of melanomas diagnosed are in situ. Invasive melanomas are split by the registry stage mix (Cancer Australia NCCI / AIHW, 2011 diagnoses, stage I 78.0 / II 14.1 / III 3.0 / IV 2.1%, unknown excluded). The overdiagnosed shares α are 73% of in situ and 28% of stage I diagnoses (Lindsay 2024). So about 50% of all melanomas diagnosed, and 22% of invasive ones, are overdiagnosed: 1.01 overdiagnosed melanomas for each melanoma that would progress. Of the melanomas that would progress, 30% are found in situ, 52% at stage I and 18% at stage II or later.

What checks do, and its calibration. For a group of adults whose share found by a check is P, stage II or later melanomas are 1 − θ·P times as common as with no checks, and the melanomas found earlier are found in situ or at stage I. Overdiagnosed in situ and stage I melanomas rise with P_o, the share of overdiagnosed melanomas found by a check. Three Australian contrasts of examined against unexamined adults set the sizes. The examined adults are taken to have had v_A = 0.55 checks a year, so P_A = 57% and P_oA = 88%. Stage II or later melanomas were 0.74 times as common in them, which gives θ = 0.45. Source: Aitken 2010 (Int J Cancer 126:450-8): whole-body clinical skin examination in the 3 years before diagnosis; melanomas 1.5 mm or thicker (two thickness groups pooled), odds ratio 0.74 (0.61-0.90). Examined adults taken to have 0.55 checks a year (Janda 2004: 11% examined in 12 months, 20% in 3 years). Stage I melanomas were 1.29 times as common. Source: Aitken 2010: odds ratio 1.38 (1.22-1.56) for melanomas 0.75 mm or thinner and 0.93 (0.79-1.10) for 0.76-1.49 mm; combined 1.29 for melanomas under 1.5 mm (those 0.75 mm or thinner taken as 80%). In situ melanomas were 1.45 times as common. Source: Whiteman 2022 (Br J Dermatol 187:515-22), QSkin: in situ melanoma 1.45 (1.09-1.92) times as common in adults whose skin was examined by a doctor in the previous 3 years. In Aitken 2010, melanomas 1.5 mm or thicker are taken to be stage II or later, and those under 1.5 mm to be stage I. The excess of stage I and in situ melanomas in examined adults, beyond the thick melanomas found earlier, is counted as overdiagnosis. With no checks there are 0.66 overdiagnosed in situ and 0.14 overdiagnosed stage I melanomas for each progressive melanoma; checks that found every overdiagnosed melanoma (P_o = 1) would add 0.45 and 0.18. No data say whether a melanoma found earlier is found in situ or at stage I: λ = 0.5 is an assumption, and the probabilistic analysis draws it anywhere between 0 and 1. Because every quantity is linear in P or P_o, the two usual-care groups together reproduce the registry diagnoses and the overdiagnosed shares exactly. v_A = 0.55 is the lowest value the survey figures allow (everyone examined in 12 months is among those examined in 3 years), so its range runs upwards only, to 1: more checks among the examined adults make each check worth less. The range used for the odds ratio is 0.61 to 0.90, for the stage I ratio 1.12 to 1.47 and for the in situ ratio 1.09 to 1.92.

Replacement. Programme checks replace a regular checker's own checks one for one (δ = 1, an assumption). A regular checker who attends then has the more frequent of the two schedules, so programme checks every 1 to 5 years leave their stage mix and the number of checks they have unchanged, and only 6-monthly checks add to what they already have. The health gain comes from adults who are not checked today. Every result is the average over the two groups, weighted by the share of regular checkers.

The risk tools predict a diagnosed invasive melanoma. Some of those diagnoses are overdiagnosed (22%), and some melanomas that would progress are found in situ (30%), which the tools do not count. So the predicted risk is converted to progressive melanomas with g = 1.11. The conversion uses the share of regular checkers among adults above the threshold, taken to be the population share, and is the same whoever attends. Q_death is the QALYs lost per melanoma death, discounted to the time of death: 7.3 (range 6.4 to 8.6; 8.5 discounted life-years before quality weighting). It comes from melanoma deaths at ages 45 and over by age and sex (AIHW, 2023; mean age at death 75.4) and the ABS life tables for 2022–2024, with each remaining year weighted by the Australian population norm for quality of life at that age (McCaffrey 2016). q_OD is the QALY loss per overdiagnosed melanoma, the same for in situ and stage I: 0.05 (range 0.02 to 0.26), from the utility of stage I and II patients at surgery and over the following year (Tran 2018 meta-analysis) against the population norm, with no loss after the first year. The low end counts the month of treatment only and the high end two years of reduced quality of life (Tromme 2014). q_proc is the QALY loss for each biopsy or excision that follows a check: 0.0011 (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 applies to the procedures after programme checks and after usual-care checks alike, so a programme check that replaces a regular checker's own check adds none. q_adv is the QALY loss while living with advanced melanoma, before death or lasting remission: 0.3 per episode (range 0.15 to 0.6). Source: Assumption 0.3 QALY (0.15-0.6): stage IV on treatment 0.58 (Tromme 2014 / Watts 2017) against the population norm, about one year. Life lost after a death is counted separately, in Q_death.

d is the discount rate. DF_e is the mean discount factor for a detection spread evenly over the programme; DF_d adds a mean of 5 years from detection to death; DF_adv starts stage IV episodes at detection and relapse episodes 2 years later; E is the discounted cost of an episode over years 1–3. Programme checks are spread evenly over the programme, on the same clock as the health events they lead to; a one-off check is at year 0. Usual-care checks are spread evenly too. ρ = 0.75 is an assumption until the stage shift is calibrated to Policy1-Melanoma. The overdiagnosed share α is 73% for melanoma in situ and 28% for stage I, and zero for stages II–IV. Lindsay 2024 estimated overdiagnosis for melanoma in situ and for thin invasive (stage IA) melanoma only; applying its stage IA figure to all stage I is an ATOF extension, not a finding of the paper.

Two treatment-cost channels. Primary treatment C_s follows the stage actually found and gets full credit: excision, sentinel node biopsy, staging and imaging, and for resected stage III oncology consultations and adjuvant immunotherapy (eligible share 78% × uptake 65% × a real-world course of 231 days). Advanced-melanoma episodes (stage IV at diagnosis and every relapse) are counted from realised deaths: episodes = deaths ÷ P(death | advanced). They therefore share the lead-time factor ρ with the mortality benefit, so the part of a stage shift that is not real comes back as relapse cost. Episodes at diagnosis are the melanomas still found at stage IV; a melanoma found earlier whose outcome does not change is treated at its new stage and counted as a relapse 2 years later. An episode costs $264,802 at published prices, built from its parts: first-line systemic therapy over the 2026 regimen mix at PBAC mean treatment durations, infusion admissions, a subsequent line, imaging and consultations, and toxicity care and radiotherapy. End-of-life care per melanoma death is the Goldsbury 2026 terminal-phase cost excluding PBS medicines (already in the episode): $48,751 in 2019 AUD ($61,865 in 2026 AUD), the excess over matched controls among melanoma patients who died of any cause. Advanced episodes avoided are the same in both treatment eras, because the era and calibration factors cancel in deaths ÷ P(death | advanced): salvage therapy changes survival, not relapse. Each overdiagnosed melanoma is costed at the primary treatment price of its stage: $883 in situ and $3,383 at stage I.

Survival and calibration. Lifetime melanoma mortality uses 10-year melanoma-specific survival by stage at detection. The pre-immunotherapy base is about I 93%, II 70%, III 55%, IV 12% (approximately AJCC7, Balch 2009). The immunotherapy era applies one salvage factor (φ = 0.18) to every stage's mortality, which takes stage IV 10-year survival from 12% to 28%. Published survival is per diagnosed melanoma. An overdiagnosed melanoma cannot cause death, so each stage's mortality is divided by the share of that stage's diagnoses that are progressive (1 − α), which gives mortality per progressive melanoma. One calibration factor, k = 0.76, then scales every stage's 10-year mortality so that melanoma deaths per invasive melanoma diagnosed under usual care equal 8.3%, the ratio of Australian melanoma deaths to invasive diagnoses (AIHW). That is 10.7% per progressive invasive melanoma, once the overdiagnosed ones are left out. The same k applies in both eras, so the difference between them is preserved. Calibrated 10-year survival per progressive melanoma in the immunotherapy era is I 94%, II 81%, III 72%, and 45% for advanced disease (P(death | advanced) = 0.55). This is higher than the 28% figure for all distant-stage patients, but it is implied jointly by AIHW deaths and PBAC first-line patient numbers, and it is consistent with CheckMate 067 10-year melanoma-specific survival (44–52%) in treated patients. PBAC check: 2,139 first-line starts a year (2023) ÷ 0.85 receiving systemic therapy ≈ 2,516 advanced episodes a year, or 0.145 for each of the 17,400 invasive melanomas diagnosed a year (0.164 if only 75% receive systemic therapy), against the model's 0.151 usual-care advanced episodes per invasive melanoma diagnosed. Adjuvant therapy gets no survival credit (CheckMate 238 showed no significant overall-survival difference). Sources: 10-yr MSS: AJCC7/Balch 2009 approx (pre-immunotherapy); AJCC8 10-yr (published upper-bound reference, not used as a model option); AIHW ACD 2017-21 (stage II proxy); AIHW melanoma deaths (~1,450) / incidence (~17,400); PBAC ~2,139 first-line advanced patients/yr.

Era, perspective and prices. The immunotherapy era (2026, the default) uses PBS 1 Oct 2026, MBS 1 Jul 2026 and IHACPA NEP 2026-27 prices, with older sources indexed by the AIHW total health price index (Health expenditure Australia 2023–24, Table 39a; 2025–26 extrapolated). The pre-immunotherapy era uses published-study treatment costs (Elliott 2017, Watts 2017) indexed to 2026, with pre-immunotherapy survival; the prices of checks, referrals and follow-up procedures are the 2026 prices in both eras, so the switch isolates the effect of treatment. The health-system perspective (the default) is government plus average patient contributions: the MBS benefit plus the average out-of-pocket cost of office care, and 12.5% for procedures (an assumption). The government perspective is Commonwealth plus state. Admitted surgery and infusions are costed at the public-hospital National Efficient Price in both. Medicines are at published PBS prices (the MSAC convention), an upper bound because confidential rebates lower the real price; the Workforce Lab's drug-discount setting applies to the manufacturer's price only.

Programme costs. Itemised MBS costs: a GP or skin-cancer GP check (MBS 23/36 + 75870, $86 health system, $74 government), a dermatologist check (105) or referral consultation (104), and biopsies and excisions with pathology and a results consultation, at Goldsbury 2024 per-check rates scaled by provider (skin-cancer GP ×0.7, dermatologist ×0.5, other GPs and practice nurses ×1). Practice-nurse and photography with AI checks have no MBS item and are built bottom-up (loaded nurse time and practice overhead; the AI licence and device cost per scan are assumptions). The 6-monthly specialist option uses the Watts 2017 high-risk-clinic visit cost and the same per-check follow-up cost as the other options. It is valued as two checks a year, with no extra effect for the clinic's photography and dermoscopy.

Photography with AI. Procedures follow the share of image sets the AI flags, not a rate per check. Every flagged set is tele-reviewed by a dermatologist, and 50% of the flagged adults (an assumption, range 30% to 90%) then see a GP in person and have a biopsy or excision. At the default flag share (10%) that is 5 procedures for every 100 photography checks, against 16.3 for every 100 GP checks. AI triage is given the same chance of finding a melanoma as a clinician's check, which has not been established.

Risk assessment. Finding out who is above the threshold costs $5 per eligible adult on average (invitation, questionnaire, follow-up), whether or not the adult responds. It is an assumption, range $0 to $15, and can be edited in the settings. Every adult 40+ is assessed whatever the threshold, so it is a fixed cost of the programme. By default there is one round every 10 years, which is one round in a 10-year programme, counted at the start. A shorter interval adds rounds and their cost (years 0 and 5 at a 5-year interval), and the model gives a repeat assessment no extra effect.

Keratinocyte cancers and patients' costs. Whole-body checks also find basal and squamous cell carcinomas. An excision of one follows 3.4 of every 100 GP skin checks (Goldsbury 2024, reported separately from its biopsy and benign excision rates, so nothing is counted twice), costed as an excision with pathology and a results consultation at 2026 MBS fees. Only the extra checks count: programme checks by clinicians less the regular checkers' own checks they replace (a keratinocyte cancer seen on a photograph is among the flagged procedures). Most of these cancers would have been treated later anyway, so by default only the cost of treating them 1 to 3 years earlier is counted (14% of the full cost at 5% discounting). The setting can count the full cost, an upper bound, or leave them out. Their excisions are charged the same QALY loss as other procedures, in the share that is costed. Patients' out-of-pocket costs are the average patient contribution on each item (about $12 per GP skin check, most being bulk-billed; 12.5% of the government benefit on procedures), shown as a change from today. They are inside the health-system totals and left out of the government view.

Attendance. The attendance setting is the share of adults above the threshold who attend their checks (50% by default). Checks, clinician hours, programme costs and health effects are all multiplied by it. Adults who do not attend stay in usual care, with today's skin checks, so they add no cost and no benefit. Regular checkers are taken to attend as often as other adults, and to be as common above the threshold as in the population, so 37% of the adults who attend are regular checkers. The cost of risk assessment does not change, because every adult is still assessed.

Uncertainty. The probabilistic analysis on the Value per person page draws shared inputs: the stage mix of melanomas diagnosed under usual care (Dirichlet, tight around the registry mix); logit-normal overdiagnosis fractions α (in situ and stage I); ρ uniform within 0.25 either side of its setting and no higher than 1 (0.5–1.0 at the default); QALYs lost per melanoma death (log-normal, CV 0.2); one salvage factor φ (Beta, mean 0.18), from which every stage's 10-year mortality per progressive melanoma is recomputed together with the drawn α, each draw then being rescaled so that melanoma deaths per invasive melanoma diagnosed under usual care stay at the observed 8.3% (the draws change which stages the deaths come from, not how many there are); P(death | advanced), which sets the number of advanced episodes for each death, drawn on its own around its calibrated value (0.40 to 0.69); a treatment cost multiplier (log-normal, CV 0.25); adjuvant uptake (Beta, mean 0.65); the advanced-episode cost (a further log-normal multiplier, CV 0.3, so about 0.4 combined with the treatment multiplier); and the QALY loss per overdiagnosed melanoma (log-normal, CV 0.75). Version 2.0 adds the following, each with the range that holds 95% of draws: the share of adults who already have regular checks (Beta, 26% to 50%); the odds ratio for stage II or later melanoma (log-normal; its spread combines the standard error of the published interval with an allowance for bias in an observational study, 0.56 to 0.98); the ratios for stage I and in situ melanoma in examined adults (log-normal, 1.10 to 1.51 and 1.10 to 1.93); the share of earlier-found melanomas that are found in situ (uniform between 0 and 1); the checks a year among the examined adults (drawn upwards only from 0.55, the lowest value the survey figures allow; 0.55 to 0.86); the mean detectable period of a progressive melanoma (log-normal, 1.0 to 3.6 years) and of an overdiagnosed one (log-normal, 4.4 to 19.9 years); the sensitivity of a check (Beta, 0.60 to 0.94); biopsies and excisions per GP check (log-normal, 11 to 24 per 100 checks); the QALY loss per biopsy or excision (log-normal, 0.0004 to 0.0025); and the QALY loss per advanced-melanoma episode (log-normal, 0.15 to 0.56). Prices move with two separate log-normal multipliers: a narrow one for prices set by MBS fees (GP and dermatologist checks, referrals, procedures; 90% to 110% of the price) and a wide one for prices built bottom-up (nurse checks, photography; 61% to 155%). 600 draws are summarised as a cost-effectiveness acceptability curve, with each risk band weighted by the number of adults in it; with 600 draws a percentage is accurate to about 3 points either way. Treatment era, perspective, drug discount and the treatment-savings setting are scenarios, not drawn. The national results are run through the same draws (300 simulations) to give a 95% range for the net cost a year and the cost per QALY, and the share of simulations in which the programme is good value at the chosen value of a QALY. The threshold, delivery model, interval, attendance, ages and risk-assessment cost are held at their settings.

National programme impact (Workforce Lab cost panel). In each region, eligible adults above the threshold are split into log-spaced risk bands using the binormal risk distribution of section 1 at the region's average risk. The adults in each band who attend get programme checks at the band's interval. The model above gives their incremental cost, QALYs, deaths, advanced episodes and overdiagnoses against usual care, for regular checkers and for other adults, and these are summed nationally. The cost per check is weighted by the delivery mix, plus dermatologist referrals and tele-review. The panel itemises the cost (10-year discounted, and average per year undiscounted): risk assessment, programme checks, referrals and tele-review, follow-up procedures, regular checkers' own checks replaced, treatment avoided by earlier diagnosis (primary treatment, advanced episodes, end-of-life care), treatment of extra overdiagnoses and of keratinocyte cancers. With the treatment-savings setting off it shows the programme budget only: the treatment lines are shown but not counted, overdiagnosis treatment stays in, and the cost per QALY is flagged as not a standard cost-effectiveness result. A check is assumed to find a melanoma equally often whoever performs it, so deaths averted do not depend on the delivery model. QALYs do, because the delivery models lead to different numbers of biopsies and excisions, and the QALY loss from those procedures is large beside the net gain. National counts and totals scale with the average risk π (section 3, default 1.25%); the per-person table on the Value per person page does not depend on it. When the national gain is under 0.5 QALYs the result is reported as no health gain, not as a cost per QALY. A programme can avert melanoma deaths and still have no health gain, when overdiagnosis, biopsies and excisions take away more health than the deaths averted add.

Parameter table

Every value the results depend on, where it comes from and where it is used. The values are read from the tool itself, so this table always matches what the tool calculates. Values marked as settings can be changed in the tool.

ParameterValueSourceUsed in
Risk and demand
Average 10-year risk of a first invasive melanoma, adults 40+1.25%AIHW Cancer data in Australia 2025 and ABS population, June 2025 (section 3)Every view
Accuracy of the default risk tool (AUC, QSkin MP16)0.69External validation in the 45 and Up Study (section 2)Every view
Risk in each region relative to the national averageby regionAustralian Cancer Atlas 2.0, invasive melanoma 2010–2019, and each region's age mix within adults 40+Demand, cost model
Eligible adults (aged 40 and over)12,930,995ABS estimated resident population, 30 June 2024, 88 regionsEvery view
Attendance50% (setting, 20% to 100%)Close to BreastScreen Australia (about 50%) and above the National Bowel Cancer Screening Program (about 41%), AIHW; no risk-based skin check programme has measured itWorkforce Lab, cost model
Ages offered40 and over (setting)Start 40, 45 or 50; stop 70, 75, 80 or no limit. ABS population by age and region, 2024; AIHW risk by ageEvery view
Checks below the threshold (regular checkers' own checks)as today (setting)−10% and +15% supported by Usher-Smith 2018, Evans 2016 and Smit 2021; −25% and +35% are stress testsWorkforce Lab, cost model
Workforce
Dermatologists, clinical FTE535National Health Workforce Dataset 2024, clinical hours ÷ 40Workforce Lab, capacity map
GPs, clinical FTE27,457National Health Workforce Dataset 2024, clinical hours ÷ 40Workforce Lab
GP time spent on skin cancer, FTE3,764GP clinical FTE × each state's share of GP time on skin cancer (section 5)Workforce Lab, capacity map
Practice nurses, clinical FTE10,616National Health Workforce Dataset 2024, nurses whose main job area is primary careWorkforce Lab
Clinical hours in an FTE year1,840 (nurses 1,748)40 hours × 46 weeks; 38 hours for nurses, as the dataset defines a nurse FTEWorkforce Lab
Minutes per skin check by a GP or dermatologist15Assumption (a standard consultation); setting, 10 to 20Workforce Lab
Minutes per practice-nurse check and per photography session25 and 30Assumptions, the same minutes that set their prices; they scale with the minutes settingWorkforce Lab
Clinician minutes for the procedures that follow a checkbiopsy 20, excision 30, results consultation 10 (5.7 per GP check)Assumptions; biopsies and excisions per check from Goldsbury 2024Workforce Lab
Most of any workforce's time the programme may use10%Policy setting, not a measurementWorkforce Lab
GP and nurse checks referred to a dermatologist5%Assumption; settingWorkforce Lab, cost model
Consultations a year per FTE (dermatologist, skin-cancer GP)2,800, 1,200; 30% available for screeningAssumptions (section 5)Capacity map, equity views
Costs (2026 Australian dollars, health-system perspective)
GP skin check$86MBS 1 Jul 2026: items 23/36 and 75870, benefit plus average patient contributionCost model
Practice-nurse check$47Built from nurse time and practice overhead (assumption; no MBS item)Cost model
Nurse photography with AI triage, per check$101Built from nurse time, device and AI licence (assumptions; no MBS item)Cost model
Dermatologist referral consultation$186MBS 1 Jul 2026: item 104Cost model
Follow-up biopsies and excisions per GP check$40 (biopsy after 8.3%, excision after 8.0%)Rates: Goldsbury 2024. Prices: MBS items with pathology and a results consultationCost model
Flagged photographs that lead to a GP visit and a procedure50% (30% to 90%)Assumption. Every flagged image set is tele-reviewed by a dermatologist firstCost model, Workforce Lab
Risk assessment$5 per eligible adult, one round every 10 yearsAssumption $5 (0-15) per eligible adult, one round per 10-year programme; the average cost of finding out who is above the threshold. Both are settingsCost model
Keratinocyte cancers treated after extra checks3.4 per 100 checks (2.2 to 4.8), $426 eachGoldsbury 2024 (keratinocyte cancer excised within 3 months after 3.4% of GP skin checks); 2026 MBS fees. By default only the cost of treating them earlier is counted (14% of the full cost); settingCost model
Patients' share of procedure costs12.5% of the government benefitSkin cancer excisions in QSkin (Lindsay 2026) and Thompson 2019; Medicare statistics 2025-26Cost model (health-system view)
Price of a usual-care skin check$86 plus $40 follow-upPriced as a GP check with its follow-up proceduresCost model (the comparator)
Advanced-melanoma episode (stage IV or relapse)$264,802PBS 1 Oct 2026 prices at PBAC mean treatment durations; IHACPA NEP 2026-27Cost model
End-of-life care per melanoma death$61,865Goldsbury 2026, terminal phase, excluding medicinesCost model
Discount rate, costs and health5%Australian convention for health technology assessmentCost model
Programme length10 years, lifetime consequencesModelling choiceCost model
Value of a QALY for the verdict$50,000No official Australian figure; settingCost model
Health effects
Adults who already have a whole-body skin check every year37% (30% to 45%); the rest have noneNational 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)Cost model (the comparator)
Programme checks that replace a regular checker's own checks100%Assumption; 0% to 100% on the Value per person pageCost model
Detectable period before a progressive melanoma would surface (mean); chance that one check finds a detectable melanoma2 years (1 to 4); 0.8 (0.6 to 0.95)Assumptions: mean detectable period 2 years (1-4) for a progressive melanoma and 10 years (5-20) for an overdiagnosed one; sensitivity of one whole-body check 0.8 (0.6-0.95)Cost model
Share of progressive melanomas found by a check5-yearly 30%, 2-yearly 55%, yearly 72%, 6-monthly 84%Derived from the row above (section 6)Cost model
Detectable period of an overdiagnosed melanoma (mean)10 years (5 to 20)Assumption: an overdiagnosed melanoma does not surface on its own, so it stays detectable for longer than a progressive oneCost model
Share of overdiagnosed melanomas found by a check5-yearly 72%, 2-yearly 87%, yearly 93%, 6-monthly 96%Derived from the row above and the chance that one check finds a detectable melanoma (section 6)Cost model
Stage II or later melanomas in adults who have skin checks (odds ratio)0.74 (0.61 to 0.90)Aitken 2010 (Int J Cancer 126:450-8): whole-body clinical skin examination in the 3 years before diagnosis; melanomas 1.5 mm or thicker (two thickness groups pooled), odds ratio 0.74 (0.61-0.90). Examined adults taken to have 0.55 checks a year (Janda 2004: 11% examined in 12 months, 20% in 3 years)Cost model
Checks a year among the adults that odds ratio describes0.55 (0.55 to 1)Janda 2004, as in the row above. It is the lowest value the survey figures allow; more checks among them make each check worth lessCost model
Stage I melanomas in adults who have skin checks (ratio)1.29 (1.12 to 1.47)Aitken 2010: odds ratio 1.38 (1.22-1.56) for melanomas 0.75 mm or thinner and 0.93 (0.79-1.10) for 0.76-1.49 mm; combined 1.29 for melanomas under 1.5 mm (those 0.75 mm or thinner taken as 80%)Cost model
In situ melanomas in adults who have skin checks (ratio)1.45 (1.09 to 1.92)Whiteman 2022 (Br J Dermatol 187:515-22), QSkin: in situ melanoma 1.45 (1.09-1.92) times as common in adults whose skin was examined by a doctor in the previous 3 yearsCost model
Melanomas found earlier that are found in situ; the rest are found at stage I50% (anywhere from none to all)Assumption: no data say whichCost model
Melanomas diagnosed under usual care, by stagein situ 55%, I 36%, II 6%, III 1%, IV 1%Australian Cancer Atlas 2010-2019 (persons): 165,641 in situ and 134,543 invasive melanomas. Invasive melanomas by stage: Cancer Australia NCCI / AIHW, 2011 diagnoses, stage I 78.0 / II 14.1 / III 3.0 / IV 2.1%, unknown excludedCost model
Share of the stage shift that is real, not lead time0.75Assumption (0.5 to 1.0) until calibrated to Policy1-MelanomaCost model
Overdiagnosed share of melanomas diagnosed under usual carein situ 73%, stage I 28%; 50% of all melanomas diagnosedLindsay 2024 (the stage IA estimate applied to all stage I), applied to the diagnoses in the row aboveCost model
Progressive melanomas per invasive melanoma diagnosed1.11Derived: 30% of the melanomas that would progress are found in situ, and 22% of invasive melanomas diagnosed are overdiagnosedCost model
Melanoma deaths per invasive melanoma diagnosed under usual care (calibration target)8.3%AIHW melanoma deaths (~1,450) / incidence (~17,400); PBAC ~2,139 first-line advanced patients/yrCost model
10-year melanoma survival by stage, per progressive melanoma (calibrated)I 94%, II 81%, III 72%, IV 45%Published stage-specific survival per diagnosed melanoma, with immunotherapy-era survival for advanced disease. Mortality is divided by the share of each stage's diagnoses that are progressive, then scaled so that usual-care deaths per invasive melanoma diagnosed match the row above (section 6)Cost model
Years from detection to death5Assumption (3 to 7)Cost model
QALYs lost per melanoma death (discounted)7.3 (6.4 to 8.6)AIHW melanoma deaths by age and sex (2023) × ABS life tables 2022–2024, weighted by Australian EQ-5D-5L population norms (McCaffrey 2016)Cost model
QALY loss per overdiagnosed melanoma, in situ or stage I0.05 (0.02 to 0.26)Stage I/II utility at surgery and over the following year (Tran 2018 meta-analysis) against the Australian population norm; no loss after the first yearCost model
QALY loss per biopsy or excision after a check0.0011 (0.0005 to 0.003)Utility 0.971 after a benign excision (Chen 2004, in Watts 2017), taken for two weeks (assumption)Cost model
QALY loss per advanced-melanoma episode, before death or lasting remission0.3 (0.15 to 0.6)Assumption 0.3 QALY (0.15-0.6): stage IV on treatment 0.58 (Tromme 2014 / Watts 2017) against the population norm, about one yearCost model

7. Data sources

  • Risk models. External-validation AUCs from the 45 and Up Study (Reyes-Marcelino et al., JAMA Dermatology 2026): QSkin MP16 0.69, QSkin MP7 (Olsen 2018 invasive model) 0.69, MIA/Vuong 0.70, Cho 0.66. NCI MBRAT 0.70 (development study; not in that validation).
  • Average 10-year risk. AIHW Cancer data in Australia 2025, Book 1a, Table S1a.1 (invasive melanoma, age-specific rates, 2025 projections); ABS estimated resident population by single year of age, 30 June 2025 (National, state and territory population, March 2026); second primaries estimated from QSkin and Youlden 2014 (section 3).
  • Workforce supply (all 88 SA4s). National Health Workforce Dataset 2024, clinical hours ÷ 40 by main-job SA4. In 2024 that is 657 dermatologists averaging 33 clinical hours a week, or 535 clinical FTE (634 and 521 in 2023). Dermatologist values resting on published cells of 0–3 people or hours are shown only as a band (disclosure control); in calculations these regions share the national total minus the published regions, in proportion to their band midpoints. Skin-cancer GP FTE = GP clinical FTE × the state skin-cancer share of GP time. The differences between states come from applying the Queensland method of Lindsay 2026 to every state (an ATOF extension; the paper covers Queensland only), and the national level is informed by Medicare skin-procedure data (the data constrain the level only weakly).
  • Accredited skin-cancer GPs. Skin Cancer College Australasia public directory, counted by state on 27 Sep 2026 (641 doctors), scaled so that Queensland = 216 (the figure SCCA gave Lindsay).
  • Incidence by region. Australian Cancer Atlas 2.0 (Cancer Council Queensland, QUT): age-standardised invasive melanoma SIR 2010–2019, SA2 aggregated to SA4 as modelled counts ÷ expected counts. In situ melanoma is not used as a measure of need, because its diagnosis partly reflects access to skin checks.
  • Population. ABS estimated resident population by SA4, 30 June 2024 (Regional population 2023–24).
  • SEIFA and remoteness. ABS Socio-Economic Indexes for Areas (SEIFA) quintiles and Modified Monash Model remoteness categories (MM1 to MM7).

8. Methodological caveats

  • Decision curves are modelled. They come from published AUCs through a binormal approximation, not from individual-level data.
  • The average risk is an estimate. The second-primary share is estimated rather than published, and the 2025 incidence rates are projections (section 3 and limitation 9).
  • Workforce shares are assumptions. Every region uses measured NHWDS 2024 clinical FTE, but the share of GP time spent on skin cancer (the Queensland method of Lindsay 2026, applied by ATOF to every state) and the share of clinician time available for screening (30%) are assumptions. NHWDS perturbs cells of 0–3, so those regions are shown only as bands.
  • Melanographer and tiered scenarios are modelled. Their throughput multipliers (×2.0; ×1.5 and ×0.5) are first-principles assumptions pending empirical data from ACEMID.
  • Value of a QALY. MSAC 1699 (National Lung Cancer Screening Program), a screening precedent, not a melanoma assessment: ICER stated as AUD $62,754/QALY; ATOF uses $65,000 as one reference value of a QALY. Annual overdiagnosis cost ≈ AUD $21.6M (Lindsay 2024).

9. Citations

37 references, in order of first author. Treatment-cost and price sources are named in section 6.

  1. Adamson AS, Naik G, Jones MA, Bell KJL (2024). Ecological study estimating melanoma overdiagnosis in the USA using the lifetime risk method. BMJ Evidence-Based Medicine, 29(3): 156–161. doi:10.1136/bmjebm-2023-112460 · PMID:38242569
  2. Aneja S, Aneja S, Bordeaux JS (2012). Association of increased dermatologist density with lower melanoma mortality. Archives of Dermatology, 148(2): 174–178. doi:10.1001/archdermatol.2011.345 · PMID:22351816
  3. Australasian College of Dermatologists (2023). Special Commission of Inquiry into Healthcare Funding - NSW submission. ACD. link
  4. Australian Bureau of Statistics (2026). National, state and territory population, March 2026 (estimated resident population by single year of age, 30 June 2025). ABS. link
  5. Australian Bureau of Statistics (2025). Life expectancy, 2022–2024 (life tables by age and sex, Australia). ABS. link
  6. Australian Bureau of Statistics (2025). Regional population by age and sex: estimated resident population by SA4 and 5-year age group, 30 June 2024 (ERP_ASGS2021). ABS Data API. link
  7. Australian Government Department of Health, Disability and Ageing (2026). National Health Workforce Dataset 2013-2024: medical practitioners by SA4 and primary specialty (Health Workforce Data Tool, accessed 27 September 2026). Health Workforce Data. link
  8. Australian Institute of Health and Welfare (2024). Cancer data in Australia. AIHW Cancer Series. link
  9. Australian Institute of Health and Welfare (2025). Cancer data in Australia 2025, data tables: Book 1a, cancer incidence (age-standardised rates and 5-year age groups), Table S1a.1. AIHW, cat. no. CAN 122. link
  10. Australian Institute of Health and Welfare (2025). Cancer data in Australia 2025, data tables: Book 2a, cancer mortality (age-standardised rates and 5-year age groups), Table S2a.1; Book 5, mean and median age at death, Table S5.2. AIHW, cat. no. CAN 122. link
  11. Australian Institute of Health and Welfare (2023). Cancer screening programs: quarterly data (participation: BreastScreen Australia 2021–2022, National Bowel Cancer Screening Program 2020–2021). AIHW. link
  12. Bjørch MF, Gram EG, Brodersen JB (2024). Overdiagnosis in malignant melanoma: a scoping review. BMJ Evidence-Based Medicine, 29(1): 17–28. doi:10.1136/bmjebm-2023-112341 · PMID:37793786
  13. Blake SC, Guitera P, Cust AE, et al. (2023). The association of dermatologist demographic density with melanoma survival in New South Wales, Australia. Australasian Journal of Dermatology, 64(3): 425–429. doi:10.1111/ajd.14113 · PMID:37353974
  14. Brentnall AR, Atakpa EC, Hill H, Santeramo R, Damiani C, Cuzick J, Montana G, Duffy SW (2023). An optimization framework to guide the choice of thresholds for risk-based cancer screening. npj Digital Medicine, 6(1): 223. doi:10.1038/s41746-023-00967-9 · PMID:38017184
  15. Cancer Council Queensland, Queensland University of Technology, Australian Cancer Atlas (2024). Australian Cancer Atlas 2.0. Version 05-2024. link
  16. Cho E, Rosner BA, Feskanich D, Colditz GA (2005). Risk factors and individual probabilities of melanoma for whites. Journal of Clinical Oncology, 23(12): 2669–2675. doi:10.1200/JCO.2005.11.108 · PMID:15837981
  17. Evans DGR, Donnelly LS, Harkness EF, et al. (2016). Breast cancer risk feedback to women in the UK NHS breast screening population. British Journal of Cancer, 114(9): 1045–1052. doi:10.1038/bjc.2016.56 · PMID:27022688
  18. Fears TR, Guerry D 4th, Pfeiffer RM, et al. (2006). Identifying individuals at high risk of melanoma: a practical predictor of absolute risk. Journal of Clinical Oncology, 24(22): 3590–3596. doi:10.1200/JCO.2005.04.1277 · PMID:16728488
  19. Glasziou PP, Jones MA, Pathirana T, Barratt AL, Bell KJL (2020). Estimating the magnitude of cancer overdiagnosis in Australia. Medical Journal of Australia, 212(4): 163–168. doi:10.5694/mja2.50455 · PMID:31858624
  20. Lindsay D, Bell KJL, Olsen CM, Whiteman DC, Pathirana T, Collins LG (2024). Estimating the magnitude and healthcare costs of melanoma in situ and thin invasive melanoma overdiagnosis in Australia. British Journal of Dermatology, 191(6): 906–913. doi:10.1093/bjd/ljae296 · PMID:39026390
  21. Lindsay D, Stangl S, Ronagh K, Cameron J, Johns R, Muir J, Collins LG (2026). Mapping the current demand and supply of dermatologists and general practitioners managing skin cancer in Queensland, Australia. Australasian Journal of Dermatology, 67(4): 214–227. doi:10.1111/ajd.70063 · PMID:41674191
  22. McCaffrey N, Kaambwa B, Currow DC, Ratcliffe J (2016). Health-related quality of life measured using the EQ-5D-5L: South Australian population norms. Health and Quality of Life Outcomes, 14(1): 133. doi:10.1186/s12955-016-0537-0 · PMID:27644755
  23. McLoughlin KG, Watts CG, Wade S, et al. (2025). A protocol for development of a microsimulation model platform to evaluate the potential benefits, harms, and cost-effectiveness of risk-tailored melanoma screening. PLoS One, 20(12): e0339177. doi:10.1371/journal.pone.0339177 · PMID:41452872
  24. Medical Services Advisory Committee (2022). MSAC Application 1699: National Lung Cancer Screening Program - Public Summary Document. Australian Department of Health. link
  25. Olsen CM, Pandeya N, Thompson BS, et al. (2018). Risk stratification for melanoma: models derived and validated in a purpose-designed prospective cohort. Journal of the National Cancer Institute, 110(10): 1075–1083. doi:10.1093/jnci/djy023 · PMID:29538697
  26. Pashayan N, Morris S, Gilbert FJ, Pharoah PDP (2018). Cost-effectiveness and benefit-to-harm ratio of risk-stratified screening for breast cancer: a life-table model. JAMA Oncology, 4(11): 1504–1510. doi:10.1001/jamaoncol.2018.1901 · PMID:29978189
  27. Reyes-Marcelino G, Espinoza D, Smit AK, Watts CG, Kang YJ, Goldsbury DE, Steinberg J, Cust AE (2026). Evaluating the performance of melanoma risk prediction models. JAMA Dermatology. doi:10.1001/jamadermatol.2026.4097 · PMID:42905683
  28. Skin Cancer College Australasia (2026). Locate a Doctor - Accredited Skin Cancer Doctors (Australia), counted 27 Sep 2026. SCCA. link
  29. Smit AK, Allen M, Beswick B, et al. (2021). Impact of personal genomic risk information on melanoma prevention behaviors and psychological outcomes: a randomized controlled trial. Genetics in Medicine, 23(12): 2394–2403. doi:10.1038/s41436-021-01292-w · PMID:34385669
  30. Tran AD, Fogarty G, Nowak AK, Espinoza D, Rowbotham N, Stockler MR, Morton RL (2018). A systematic review and meta-analysis of utility estimates in melanoma. British Journal of Dermatology, 178(2): 384–393. doi:10.1111/bjd.16098 · PMID:29077983
  31. Tromme I, Devleesschauwer B, Beutels P, Richez P, Leroy A, Baurain JF, Cornelis F, Bertrand C, Legrand N, Degueldre J, Thomas L, Legrand C, Lambert J, Haagsma J, Speybroeck N (2014). Health-related quality of life in patients with melanoma expressed as utilities and disability weights. British Journal of Dermatology, 171(6): 1443–1450. doi:10.1111/bjd.13262 · PMID:25039853
  32. Usher-Smith JA, Silarova B, Sharp SJ, Mills K, Griffin SJ (2018). Effect of interventions incorporating personalised cancer risk information on intentions and behaviour: a systematic review and meta-analysis of randomised controlled trials. BMJ Open, 8(1): e017717. doi:10.1136/bmjopen-2017-017717 · PMID:29362249
  33. Vickers AJ, Elkin EB (2006). Decision curve analysis: a novel method for evaluating prediction models. Medical Decision Making, 26(6): 565–574. doi:10.1177/0272989X06295361 · PMID:17099194
  34. Vuong K, Armstrong BK, Weiderpass E, et al. (2016). Development and external validation of a melanoma risk prediction model based on self-assessed risk factors. JAMA Dermatology, 152(8): 889–896. doi:10.1001/jamadermatol.2016.0939 · PMID:27276088
  35. Whiteman DC, Olsen CM, MacGregor S, et al. (2022). The effect of screening on melanoma incidence and biopsy rates. British Journal of Dermatology, 187(4): 515–522. doi:10.1111/bjd.21649 · PMID:35531668
  36. Whiteman DC, Olsen CM, Wang H, Law MH, Neale RE, Pandeya N (2025). A risk prediction tool for invasive melanoma. JAMA Dermatology, 161(11): 1123–1131. doi:10.1001/jamadermatol.2025.3028 · PMID:40928797
  37. Youlden DR, Youl PH, Soyer HP, Aitken JF, Baade PD (2014). Distribution of subsequent primary invasive melanomas following a first primary invasive or in situ melanoma, Queensland, Australia, 1982-2010. JAMA Dermatology, 150(5): 526–534. doi:10.1001/jamadermatol.2013.9852 · PMID:25093216