Target journal: Lancet Oncology
 
--
Abstract (semi-structured summary; 5 paragraphs; 249/300 words)
Background: Canada is a high-income country with universal healthcare. However, Canada’s large geographic area and small population density creates challenges in accessibility to oncologic treatment, such as radiotherapy.
Methods: We conducted spatial autocorrelation using the global Moran’s I statistic to detect non-random spatial patterns in age standardized all-cancer mortality-to-incidence ratios (MIRs) across health regions in Canada, from 2010 to 2012. Global ordinary least squares (OLS) regression and geographically-weighted regression (GWR) were then applied to examine relationships between distance to nearest radiotherapy facility, sociodemographic factors, and the observed spatial patterns. 
Findings: All-cancer MIRs by health region across Canada exhibited positive statistically significant global Moran’s I index values, with a tendency towards clustering (Moran’s I = .346, p = .001). Mapping of clusters showed one high-MIR cluster (range .45–.88) involving two of three Canadian territories (Nunavut and Northwest Territories) and the north of certain provinces (Manitoba, Ontario and Quebec). A second cluster of low-MIRs (range .40–.41) was observed in the southern region of British Columbia. In both regression models, health regions with longer Euclidean distance to nearest radiotherapy centre, higher rates of smoking and lower rates of food security were significantly associated with higher MIR (r2 = .70 with OLS and r2 = .74 with GWR). 
 Interpretation: Disparities in accessibility to radiotherapy exist within Canada, which, along with other sociodemographic factors, may lead to poorer oncologic outcomes. Further work is required to better understand how best to improve access to radiotherapy in Canada for regions with poorer accessibility.
Funding: None.
--
/ 3500 words 
/ 30 references
INTRODUCTION
Approximately half of all Canadians will develop cancer in their lifetime, with cancer being the leading cause of death and premature mortality in Canada (\cite{statistics2017}). Significant progress has been made in the past few decades with regards to cancer control strategies, which has resulted in an overall decrease in age-standardised cancer incidence and mortality for the most common cancers \cite{2017}. However, despite being a high-income country with universal health care, regional disparities in cancer outcomes and accessibility to cancer services persist within Canada, particularly for low-income populations, those residing in rural and/or remote areas, and new immigrants \cite{2014}. Various factors are thought to contribute to this disparity; for example, income has been reported to influence screening uptake, whereas age and geography have been identified to have a more important influence on treatment uptake and end-of-life care \cite{Maddison_2011}
Radiotherapy is an important component of cancer treatment, and the impact of geography on its uptake and accessibility has been increasingly explored. Canada's large geographic area and small population density creates challenges in providing equal accessibility to this specialised oncologic treatment, with most facilities being located in larger tertiary centres in areas of higher population density. Longer distance, or travel time, to radiotherapy centre is a significant barrier to accessing radiation therapy \cite{Gillan_2012}, and has been associated with lower odds of radiotherapy use for patients across multiple disease sites (\cite{Liu_2015},\cite{Jones_2008}), greater chance of mastectomy for breast cancer patients versus radiotherapy \cite{Schroen_2005} and decreased utilisation of palliative radiotherapy \cite{French_2008}.
The importance of considering geographic, or spatial, variability as a separate variable when exploring the factors that influence cancer outcomes has been previously reported, with spatial auto-correlation being one method used to identify geographical clusters and patterns in cancer outcomes (\cite{Mandal_2009},\cite{Al_Ahmadi_2013}). Al-Ahmadi et al. (\cite{Al-Ahmadi2013}) examined spatial trends in Saudi Arabia, and found significant clusters of lung and prostate cancer, in addition to Hodgkin's disease among males in the Eastern region, and significant clusters of thyroid cancers in females were found in the Eastern and Riyadh regionn. This suggests that the distribution of cancer outcomes within a geographic region may not be spatially random, and that the underlying associated risk factors are also spatially related. Geographically weighted regression (GWR) models can then not only identify risk factors that may be associated with cancer outcomes, but to demonstrate that their impact is not constant across regions within the geographic space of interest; as such, both spatial auto-correlation and GWR can help identify local patterns and areas of disparity \cite{Goovaerts_2015}
Given the significant contribution of geography on the reported disparities in cancer outcomes and cancer service accessibility in Canada, we sought to explore the regional variations in cancer outcomes across Canada and the potential associated variables, including distance to nearest radiotherapy centre and sociodemographic factors, and the impact of any spatial relationships. 
METHODS
Study design
Each province in Canada is divided into health regions, which are administrative areas defined by provincial ministries of health (XX). Statistics Canada collects information on cancer incidence and mortality by health region, and we therefore conducted a cross-sectional study using this administrative level to compare oncologic outcomes and assess the potential contributing factors across Canada. For the province of Ontario which is divided into either public health units or local health integration networks, we chose to use public health units as there were a greater number of divisions within the province.