Geographic differences in population rates of interventional treatment for prostate cancer in Australia

Jessica K. Cameron, Upeksha Chandrasiri, Jeremy Millar, Joanne F. Aitken, Susanna Cramb, Jeff Dunn, Mark Frydenberg, Prem Rashid, Kerrie Mengersen, Suzanne K. Chambers, Peter D. Baade, David P. SmithmedRxiv 2023 · DOI

TL;DR

Where a man lives in Australia significantly influences the type of prostate cancer treatment he receives. This spatial analysis reveals that residents of remote and socioeconomically disadvantaged areas face a "triple threat": lower diagnosis rates, lower rates of interventional treatments (like surgery), and higher rates of excess mortality.

What Problem Does This Paper Address?

Prostate cancer is the most common cancer among Australian men, but treatment isn't "one size fits all." Options range from radical surgery and radiation to active surveillance. While clinical factors (like tumor grade) should drive these decisions, international studies suggest that geography and wealth often play an outsized role.

In Australia, men in remote areas are known to have lower cancer incidence but poorer survival rates. This paper investigates whether this is due to a "postcode lottery" in treatment access. By analyzing fine-grained geographic data (Statistical Area Level 2), the authors aim to identify clusters of under-treatment and determine how remoteness and socioeconomic status (SES) correlate with specific medical interventions.

Methods

The researchers employed a sophisticated suite of spatial and statistical tools to map treatment patterns across Australia from 2007 to 2016:

  • Bayesian Spatial Modeling: Used to calculate spatially smoothed Standardised Separation Rate Ratios (SSRs). This technique "borrows strength" from neighboring areas to provide more stable estimates in regions with small populations.
  • Generalized Linear Mixed Models (GLMMs): Specifically, negative binomial and Poisson models were used to identify associations between treatment rates and covariates like age, state, remoteness, and SES.
  • K-means Clustering: A machine learning approach used to group geographic areas into three distinct profiles based on their treatment rates, incidence (SIR), and excess death ratios (EHR).
  • Tango’s Maximised Excess Events Test: A statistical test used to confirm the presence of non-random spatial "hotspots" or heterogeneity across the country.

Datasets

The study integrated several high-quality Australian datasets:

  • National Hospital Morbidity Database: Provided data on "hospital separations" (treatments) for radical prostatectomy, Low Dose Rate (LDR) brachytherapy, and High Dose Rate (HDR) brachytherapy.
  • Australian Cancer Atlas: Provided modeled Standardised Incidence Ratios (SIR) and Excess Hazard Ratios (EHR) for prostate cancer.
  • Australian Bureau of Statistics (ABS): Provided population estimates, the Australian Statistical Geography Standard (for remoteness), and SEIFA indexes (for socioeconomic advantage/disadvantage).

Key Insights and Ideas

  • The Disadvantage Gradient: There is a clear linear relationship between wealth and intervention. Men in the most advantaged areas had a radical prostatectomy rate of 19.1 per 10,000 person-years, compared to just 12.9 in the most disadvantaged areas.
  • The "High EHR" Cluster: Perhaps the most concerning finding was a cluster representing 34% of the studied areas (mostly remote or disadvantaged). These areas showed low treatment rates and low diagnosis rates, but high excess deaths. This suggests these men aren't "avoiding" cancer; they are likely being diagnosed too late or lacking the means to access life-saving interventions.
  • Surgery vs. Brachytherapy: Radical prostatectomy (surgery) and HDR brachytherapy showed similar geographic distributions, often concentrated in major cities. However, LDR brachytherapy had a unique spatial footprint, suggesting its use is driven more by local clinical expertise and the availability of specific technology rather than broad demographic trends.
  • Financial Barriers: The authors point to the high out-of-pocket costs (median ~$5,000 AUD) and the prevalence of private hospital treatment for surgery as major drivers of the socioeconomic gap.

Key Results

  • Total Volume: Between 2007–2016, 94,777 interventional treatments were recorded. Surgery was the dominant choice (88%), followed by LDR brachytherapy (11%) and HDR brachytherapy (2%).
  • Remoteness Impact: Surgery rates were significantly lower in remote areas (12.2 per 10k) compared to regional (16.4 per 10k) or major city areas (15.2 per 10k).
  • State Variation: Treatment rates varied wildly by state. For example, the radical prostatectomy rate in the Australian Capital Territory (18.7 per 10k) was nearly triple that of the Northern Territory (6.8 per 10k).
  • Cluster Profiles:
    • Group 1 ("Higher RP"): 41% of areas; higher surgery rates and higher incidence.
  • Group 2 ("High LDRB"): 25% of areas; characterized by higher use of low-dose brachytherapy.
  • Group 3 ("High EHR"): 34% of areas; low treatment rates but high excess mortality.

Limitations and Future Work

  • Missing Treatment Modalities: The study lacked data on External Beam Radiation Therapy (EBRT)—the second most common active treatment—as well as active surveillance and chemotherapy. This means the "low treatment" areas might be using EBRT, though the high excess death rates suggest a genuine care gap.
  • Ecological Fallacy: As an ecological study, the results describe areas, not individuals. We cannot say for certain that a specific man's treatment was dictated by his postcode, only that the probability shifts based on location.
  • Temporal Shifts: The data spans a decade where clinical guidelines evolved. The authors suggest that future spatiotemporal modeling is needed to see if these geographic gaps are closing or widening over time.

Summary updated

Abstract

Abstract Background Treatment decisions for men diagnosed with prostate cancer depend on a range of clinical and patient characteristics such as disease stage, age, general health, risk of side effects and access. Associations between treatment patterns and area-level factors such as remoteness and socioeconomic disadvantage have been observed in many countries but have not been investigated in Australia. Methods Hospital separations data for interventional treatments for prostate cancer (radical prostatectomy, low dose rate and high dose rate brachytherapy) were modelled using spatial models, generalised linear mixed models, maximised excess events tests and k -means statistical clustering. Results Geographic differences in population rates of interventional treatments were found (p<0.001). Separation rates for radical prostatectomy were lower in remote areas (12.2 per 10 000 person-years compared with 15.0-15.9 in regional and major city areas). Rates for all treatments decreased with increasing socioeconomic disadvantage (radical prostatectomy 19.1 /10 000 person-years in the most advantaged areas compared with 12.9 in the most disadvantaged areas). Three groups of similar areas were identified: those with higher rates of radical prostatectomy, those with higher rates of low dose brachytherapy, and those with low interventional treatment rates but higher rates of excess deaths. The most disadvantaged areas and remote areas tended to be in the latter group. Conclusions The geographic differences in treatment rates may partly reflect differences in patients’ physical and financial access to treatments. Treatment rates also depend on diagnosis rates and thus reflect variation in investigation rates for prostate cancer and presentation of disease. Spatial variation in interventional treatments may aid identification of areas of under-treatment or over-treatment.
BibTeX
@misc{cameron2023geographic,
  author = {Jessica K. Cameron and Upeksha Chandrasiri and Jeremy Millar and Joanne F. Aitken and Susanna Cramb and Jeff Dunn and Mark Frydenberg and Prem Rashid and Kerrie Mengersen and Suzanne K. Chambers and Peter D. Baade and David P. Smith},
  title = {{Geographic differences in population rates of interventional treatment for prostate cancer in Australia}},
  year = {2023},
  doi = {10.1101/2023.06.12.23291310},
  howpublished = {medRxiv},
}
Figure