Developing the atlas of cancer in Queensland: methodological issues

Susanna Cramb, Kerrie Mengersen, Peter D. BaadeInternational Journal of Health Geographics 2011 · DOI

TL;DR

This paper details the methodological decisions and rationale behind creating a comprehensive cancer atlas for Queensland, Australia. It highlights the challenges and solutions involved in mapping cancer incidence and survival across geographical areas to identify and quantify health inequalities, aiming to inform public health interventions.

What Problem Does This Paper Address?

Health equity is a major global challenge, and significant disparities in cancer outcomes are well-documented, particularly between urban and rural or socioeconomically disadvantaged populations. To address these inequalities, they must first be quantified. Geographical variations in disease data can be effectively visualized using maps, forming "atlases." However, creating such atlases involves numerous complex methodological decisions, from choosing which cancers and geographical units to analyze, to selecting appropriate statistical models and visualization techniques. This paper addresses the need for a clear, documented approach to developing these atlases, aiming to guide future efforts in chronic disease mapping and analysis.

Methods

The authors employed Bayesian hierarchical models to produce robust estimates of cancer incidence and survival for small geographical areas (Statistical Local Areas - SLAs). Specifically, they used the Besag, York and Mollié (BYM) model for incidence, which incorporates both spatial (conditional autoregressive - CAR) and unstructured random effects to borrow strength from neighboring regions. For survival, they adapted a generalized linear model based on excess mortality, also incorporating spatial and random effects.

Key methodological choices included:

  • Choice of Cancers: Focusing on Australian National Health Priority Area cancers and others with sufficient data.
  • Geographical Areas: Using Statistical Local Areas (SLAs) as defined by the Australian Standard Geographic Classification (ASGC), mapped to a consistent 2006 boundary.
  • Outcome Measures: Using Standardized Incidence Ratios (SIRs) for incidence and Relative Excess Risk (RER) for survival, derived from modelled estimates rather than raw rates to improve robustness.
  • Adjacency Matrix: Defining neighboring SLAs using the "Queen" definition (sharing a common border), with adjustments made for island SLAs to ensure all regions had neighbors for smoothing.
  • Sensitivity Analyses: Rigorously testing the impact of different prior distributions on model convergence and results, particularly for less common cancers.
  • Convergence Assessment: Using visual inspection of trace plots and the Geweke diagnostic to ensure model stability.
  • Statistical Significance Testing: Employing Tango's MEET (Maximised Excess Events Test) to identify statistically significant spatial variation.

Datasets

  • Queensland Cancer Registry (QCR) data: De-identified data on all primary invasive cancers diagnosed among Queensland residents from 1996 to 2007. This included linkage with the Australian National Death Index for survival data.
  • Australian Bureau of Statistics (ABS) data: Estimated resident population data (grouped by age, sex, year, and SLA) and de-identified unit record mortality data for all causes of death.
  • Socioeconomic and Rurality Data: Socioeconomic Indexes for Areas (SEIFA) and the Accessibility/Remoteness Index for Australia (ARIA+) were used for supplementary analysis.

Key Insights and Ideas

  • The Power of Bayesian Hierarchical Models: These models are crucial for producing reliable estimates in small areas by borrowing information from neighbors, thus smoothing out noise and making results more robust than simple standardized rates.
  • Incidence vs. Survival vs. Mortality: The paper emphasizes that mapping cancer survival is more informative for understanding the impact of healthcare access and interventions than mortality alone, as survival reflects outcomes based on where people lived when diagnosed and treated.
  • The Importance of Methodological Transparency: Documenting the rationale behind every decision (e.g., choice of priors, adjacency definitions, outcome measures) is vital for reproducibility and for guiding future atlas development.
  • Addressing Data Sparsity: For less common cancers, careful consideration of prior distributions and model convergence is paramount to avoid spurious findings.
  • Visualizing Uncertainty: While not fully solved, the paper acknowledges the need to communicate the uncertainty associated with mapped estimates, suggesting future work in this area.

Key Results

The paper details the process of developing the Queensland Cancer Atlas, which maps geographical variations in cancer incidence and survival. While specific quantitative results for each cancer are not the primary focus of this methodological paper, the process successfully generated maps and associated statistics for numerous cancers across 478 SLAs. The sensitivity analyses demonstrated that the choice of priors can significantly impact model convergence and results, leading to the selection of a specific prior (Prior 3) that offered a good balance between convergence and shrinkage. The study successfully identified areas with higher or lower than average incidence and survival rates, laying the groundwork for further investigation into the drivers of these disparities.

Limitations and Future Work

  • Interpretation of Modelled SIR/RER: The authors note that modelled SIRs and RERs compare SLA-specific estimates to the Queensland average, not directly between SLAs, which can be biased by differing age structures.
  • Communication of Uncertainty: While Bayesian models provide credible intervals, effectively communicating this uncertainty to a broader audience remains a challenge.
  • Future Work: The authors suggest several priority areas:
  • Improving the communication of spatial results, particularly uncertainty.
  • Developing and using alternative statistical models, such as classification and regression tree (CART) models.
  • Developing more detailed models to investigate the impact of rurality, socioeconomic status (area-level and individual-level), and temporal changes.

Summary updated

Abstract

Although careful consideration of many issues is required, chronic disease atlases are a useful tool for assessing and quantifying geographical inequalities. In addition they help focus research efforts to investigate why the observed inequalities exist, which in turn inform advocacy, policy, support and education programs designed to reduce these inequalities.
BibTeX
@article{cramb2011developing,
  author = {Susanna Cramb and Kerrie Mengersen and Peter D. Baade},
  title = {{Developing the atlas of cancer in Queensland: methodological issues}},
  journal = {International Journal of Health Geographics},
  year = {2011},
  doi = {10.1186/1476-072x-10-9},
}
Figure