UBC Faculty Research and Publications

Small-area spatio-temporal analyses of bladder and kidney cancer risk in Nova Scotia, Canada Saint-Jacques, Nathalie; Lee, Jonathan S W; Brown, Patrick; Stafford, Jamie; Parker, Louise; Dummer, Trevor J. B. Feb 19, 2016

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata


52383-12889_2016_Article_2767.pdf [ 1.83MB ]
JSON: 52383-1.0307463.json
JSON-LD: 52383-1.0307463-ld.json
RDF/XML (Pretty): 52383-1.0307463-rdf.xml
RDF/JSON: 52383-1.0307463-rdf.json
Turtle: 52383-1.0307463-turtle.txt
N-Triples: 52383-1.0307463-rdf-ntriples.txt
Original Record: 52383-1.0307463-source.json
Full Text

Full Text

RESEARCH ARTICLE Open AccessSmall-area spatio-temporal analyses ofbladder and kidney cancer risk in NovaScotia, CanadaNathalie Saint-Jacques1,2*, Jonathan S. W. Lee3,4, Patrick Brown3,4, Jamie Stafford3, Louise Parker5and Trevor J. B. Dummer6AbstractBackground: Bladder and kidney cancers are the ninth and twelfth most common type of cancer worldwide,respectively. Internationally, rates vary ten-fold, with several countries showing rising incidence. This study describesthe spatial and spatio-temporal variations in the incidence risk of these diseases for Nova Scotia, a province locatedin Atlantic Canada, where rates for bladder and kidney cancer exceed those of the national average by about 25 %and 35 %, respectively.Methods: Cancer incidence in the 311 Communities of Nova-Scotia was analyzed with a spatial autoregressivemodel for the case counts of bladder and kidney cancers (3,232 and 2,143 total cases, respectively), accounting foreach Community's population and including variables known to influence risk. A spatially-continuous analysis, usinga geostatistical Local Expectation-Maximization smoothing algorithm, modeled finer-scale spatial variation in risk forsouth-western Nova Scotia (1,810 bladder and 957 kidney cases) and Cape Breton (1,101 bladder, 703 kidney).Results: Evidence of spatial variations in the risk of bladder and kidney cancer was demonstrated using bothaggregated Community-level mapping and continuous-grid based localized mapping; and these were generallystable over time. The Community-level analysis suggested that much of this heterogeneity was not accounted forby known explanatory variables. There appears to be a north-east to south-west increasing gradient with a numberof south-western Communities have risk of bladder or kidney cancer more than 10 % above the provincial average.Kidney cancer risk was also elevated in various northeastern communities. Over a 12 year period this exceedancetranslated in an excess of 200 cases. Patterns of variations in risk obtained from the spatially continuous smoothinganalysis generally mirrored those from the Community-level autoregressive model, although these more localizedrisk estimates resulted in a larger spatial extent for which risk is likely to be elevated.Conclusions: Modelling the spatio-temporal distribution of disease risk enabled the quantification of risk relative toexpected background levels and the identification of high risk areas. It also permitted the determination of therelative stability of the observed patterns over time and in this study, pointed to excess risk potentially driven byexposure to risk factors that act in a sustained manner over time.Keywords: Small-area disease mapping, BYM model, Local-EM algorithm, Bladder and kidney cancer risk,Geostatistical analysis, Spatial autoregressive analyses* Correspondence: nathalie.st-jacques@ccns.nshealth.ca1Cancer Care Nova Scotia, Surveillance and Epidemiology Unit, Room 560Bethune Building, 1276 South Street, Halifax B3H 2Y9NS, Canada2Interdisciplinary PhD program, Dalhousie University, 6299 South Street,Room 314, PO Box 15000, Halifax B3H 4R2NS, CanadaFull list of author information is available at the end of the article© 2016 Saint-Jacques et al. Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, andreproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link tothe Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver(http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.Saint-Jacques et al. BMC Public Health  (2016) 16:175 DOI 10.1186/s12889-016-2767-9BackgroundUrinary tract cancers comprise primarily cancers of theurinary bladder and kidney, the former accounting forapproximately two-thirds of all cases diagnosed. Bladdercancer is the ninth most common type of cancer world-wide (~360,000 cases per year) and the 13th most commoncause of death from cancer (~145,000 deaths per yearworldwide) [1, 2]. Kidney cancer is comparatively lesscommon, ranking twelfth and accounting for an approxi-mate 150,000 new cases and 78,000 deaths annually [3, 4].Internationally, the incidence rates for bladder andkidney cancer have been reported to vary by as much asten-fold between countries. Incidence tends to be higherin Southwestern Europe, North Africa (Egypt) and NorthAmerica; and lower in South America and Asia [1, 4, 5].Parkin [2] reports the highest estimated mortality ratesto be in Egypt, where the world-standardized rate of 34per 100,000 (in men) is more than three times higherthan the highest rates in Europe (Denmark 10.4, Spain9.7) and eight times that in the United States (US) (3.4).Several countries show increasing incidence for bothbladder and kidney cancers, although with evidence ofsome stabilization or even decreases during the 1990s [2,4]. Recent trends in stage-specific incidence rates for blad-der cancer in some US populations, suggest however, thatrates may be stabilizing in late stage disease but continueto increase in noninvasive predominantly low grade dis-ease [6]. Regardless of space, time or stage at diagnosis,rates are consistently higher for males than females [4, 5,7–9]. In fact, in most developed countries, men are atleast, a three to five time greater risk than women.Past variations in the prevalence of known etiologicalfactors, whether genetic, environmental, occupational orbehavioural, may to some extent, contribute to the re-ported temporal and geographical variations of urinarytract cancers among populations worldwide. In addition,differences in the scope of case ascertainment betweennational cancer registries may result in some countriesreporting solely invasive diagnoses while others may in-clude non-invasive or in situ diseases. Some countriescount only one primary cancer in subjects with multiplecancers in the urinary tract. In the Netherlands, suchpractice is thought to reduce the reported incidenceof bladder cancer by up to 10 % [2]. Finally, varia-tions in rates within and/or between countries can bepartly driven by the introduction of new imagingtechniques enabling the detection of pre-symptomatictumours.In Canada, bladder cancer incidence rates increasedfrom 1970 to 1981 and have since gradually declined orstabilized [10–12]. Kidney cancer incidence rates havealso stabilised in recent years among females, but con-tinue to increase at a rate of about 1.3 % amongmales [10, 11, 13, 14]. Rates of both bladder and kidneycancer are particularly high in Nova Scotia (NS), a prov-ince of 900,000 people, in Atlantic Canada. NS consist-ently has some of the highest rates of cancer in Canadafor both males and females and continues to show in-creases in the age-standardized incidence rates of bothbladder and kidney cancers. For bladder cancer, age-adjusted incidence rates estimated for 2015 exceed thoseof the national average by about 25 and 30 % amongmales and females, respectively [11]. Similarly, forkidney cancer, excesses of 30 and 45 % have been re-ported among males and females, respectively. Thisnoted excess burden of urinary tract malignancies in NS isunlikely to result from health system related factors(e.g. scope of case registration, imaging technology)given the relative uniformity of health care deliverywithin the country.This study thus, describes spatial and spatio-temporalvariations in the risk of bladder and kidney cancer forNS in order to identify those areas where rates arehigher than what would be expected given the preva-lence of known risk factors. This is an important step toguide both etiological research and public health inter-ventions in the province. We use two geospatial methodsfor modelling disease risk, both of which are appropriatefor low-density populations such as NS. The first ap-proach is a Community-level analysis using a spatialautogregression (or Besag, York and Mollie model), aBayesian method that models diseases risk for spatiallyaggregated case counts [15, 16]. The second approachestimates spatially continuous variation in risk using aLocal Expectation Maximization (local-EM) smoothingalgorithm, an emerging geostatistical method developedby Fan, Stafford and Brown [17], which models spatialand temporal variation in risk when cases are aggregatedto time-varying spatial boundaries. To our knowledge,this is the first attempt to model the risk of bladder andkidney cancer in NS and one of the first epidemiologicalapplications of the Local-EM algorithm for cancermapping in Canada.MethodsData sourcesCancer incidence data were obtained from the NSCancer Registry and were divided into two cohorts:Cohort 1 included all NS residents diagnosed with blad-der or kidney cancer between 1998 and 2010 and aged20 years and older; Cohort 2 included cases diagnosedbetween 1980 and 2010 and aged 20 years and older.Cases were coded according to the International Classifi-cation of Diseases (ICD-O) as following: bladder (ICDO:188.0-188.9; ICD-O-2/3: C67.0-C67.9); kidney (ICDO:189.0; ICD-O-2/3: C64.9). Because of a change indisease-coding over time, bladder cases included both, insitu (36 %, period 1998–2010; 21 %, period 1980–2010;Saint-Jacques et al. BMC Public Health  (2016) 16:175 Page 2 of 17Table 1) and invasive diagnoses; kidney cases includedinvasive diagnoses only.The Community-level (BYM) analysis was restricted toCohort 1. This is because the proportion of cases withincomplete residential addresses (i.e. civic street address)was fairly large prior to 1998. During those early years,most cases were assigned to a town or a six-digit postalcode, which vary greatly in size, especially betweenurban and rural settings. Depending on the spatial scaleof analysis, one postal code may belong to severalgeographic units or one unit of geography may containseveral postal codes, resulting in the potential misclassifi-cation of the spatially aggregated data. The spatiallycontinuous-grid based (local-EM) analysis was able toaccommodate data from the entire 30 year period(Cohort 2) because the method allows for bothchanges in the spatial distribution of risk over time,and accounts for uncertainties in location of caseswhere civic street addresses are missing but postalcodes or administrative regions are known.Table 1 Cases characteristics for the two periods under study, Nova Scotia, CanadaBladder KidneyTotal Females Males Total Females MalesPeriod 1998 - 2010Nova ScotiaCases diagnosed 3,292 834 2,458 2,199 863 1,336Cases analyzed* 3,232 820 2,412 2,143 848 1,295In situ 1,164 298 866 0 0 0Invasive 2,068 522 1,546 2,143 848 1,295Mean age at diagnosis (years) 71 71.2 70.5 65 66 63.7Spatial referencing (%)Civic address 86.6 85.5 86.9 85.9 86.4 85.5Postal code 2.29 2.07 2.36 2.10 1.65 2.39Town name 11.1 12.4 10.7 12.0 11.9 12.1Period 1980 - 2010Nova ScotiaCases diagnosed† 6,473 1,642 4,831 3,762 1,493 2,269Mean age at diagnosis (years) 70 70.5 69.9 65 65.9 63.8South-western Nova ScotiaCases analyzed 1,810 423 1,387 957 358 599In situ 386 86 300 0 0 0Invasive 1,424 337 1,087 957 358 599Spatial referencing (%)Civic address 43.6 40.4 44.6 47.2 50.6 45.2Postal code 52.9 56.3 51.9 49.8 46.1 52.1Town name 3.4 3.3 3.5 2.9 3.3 2.7Cape Breton IslandCases analyzed 1101 283 818 763 306 457In situ 172 41 131 0 0 0Invasive 929 242 687 763 306 457Spatial referencing (%)Civic address 43.7 45.9 42.9 53.7 54.2 53.4Postal code 47.0 41.3 48.9 39.2 36.6 40.9Town name 9.4 12.7 8.2 7.1 9.2 5.7*Excludes 116 cases (2.1 %) diagnosed in a Community for which population data was not available†Excludes 21 bladder cases (0.32%) and 10 kidney cases (0.27%) due to unavailable spatial informationSaint-Jacques et al. BMC Public Health  (2016) 16:175 Page 3 of 17The Nova Scotia Civic Address File (NSCAF) wasused to assign spatial locations (i.e. longitude-latitudecoordinates) to all cases for which a civic street addresswas available. When civic address was unavailable, theDesktop Mapping Technologies Inc (DMTI) conversionfile was used to geo-reference postal codes. For theCommunity-level model, where postal code was unavail-able or located in rural areas, a gazetteer of place nameswas used to georeference the centroid of the town. For thespatially-continuous local-EM, where postal code wasavailable, cases locations were treated as spatially censoredsomewhere within one of the census regions containing atleast one address with the postal code in question. Wherepostal code was unavailable, the local-EM analysis used theCensus Division boundaries as a second type of spatial cen-soring. Proportions of case by spatial data type, includingthe numbers of cases excluded from each analysis due touncertainty in their spatial location, are shown in Table 1.Population data from seven census years (1981, 1986,1991, 1996, 2001, 2006, and 2011) were used for thisstudy. Each census provided counts of people aged20 years and older by age and sex group, and were usedas the denominator for cases diagnosed within two yearsof a given census period.For the modelling of risk using the spatial autoregres-sive model, population estimates were aggregated at theCommunity level, a set of geographic administrativeunits, which represent groupings of neighbourhoodswith a degree of shared identity and social processes[18]. This level of spatial aggregation represents the fin-est unit of geography for which boundaries are stableover time. There were 311 Communities in NS over thestudy period with population counts up to 30,900 per-sons. In total, 36 Communities (30 First Nations Com-munities and 6 wilderness and park Communities) wereexcluded due to unavailable population information.The spatially-continuous (local EM) analysis usedpopulation counts by age and sex group at the finestlevel of geography for which digitized spatial boundarydata were available. These were census subdivision level(CSD) for the 1981 and 1986 census years; enumerationareas (EA) for the 1991 and 1996 census years; and dis-semination areas (DA) for census 2001 onward. Therewere 113 CSD in 1981 and 118 CSD in 1986. The num-ber of EA/DA ranged from 1379 to 1645 between the1991 and 2011 census periods; their size varied to targeta population of 400 to 700 individuals.It was assumed that populations were uniformly dis-tributed within these finest levels of census regions, anot unreasonable assumption if one accepts that thesecensus regions generally follow physical boundaries, suchas major streets and waterways, and are designed to befairly homogeneous. An exception is regions which areindicated by Statistics Canada to be partially uninhabited,or lying outside the population ecumene, in which casethe population is assumed to be homogeneously distrib-uted within the inhabited portion.Covariates included in the Community-level spatialautoregressive model were indicators of socioeconomicdeprivation and well water usage. The latter obtainedfrom NS Environment, aimed to account for spatial varia-tions in risk which may relate to exposure to environmen-tal sources of heavy metals such as arsenic in drinkingwater, a known risk factor for the development of bladderand kidney cancer [19]. Socioeconomic deprivationindicators were derived from socio-economic data ob-tained from Statistics Canada. They were constructedas Community-level area-based composite indices ofsocial and material deprivation intended to be used asa proxy for unavailable individual-level measures suchas smoking, a key factor in the development of urin-ary tract malignancies. Material and social depriva-tions indices were also used to capture the contextualsetting of a place of residence, which has been shownto independently predict smoking habit in both menand women and other health outcomes [20-24]. Eachindex summarized information relating to six socio-economic indicators from the 2006 Canadian Census;all of which having known links to health outcomesand known application as geographic proxies of socio-economic conditions [21, 25-28]. For people age15 years and over, these variables were: the propor-tion of people with no high school diploma, the indi-vidual average income, the employment rate, theproportion of separated, divorced or widowed, theproportion of single-parent families, and the proportion ofpersons living alone. The first three indicators reflect thematerial dimension of deprivation; the others reflect itssocial aspect. Variables were combined using a PrincipalComponent Analysis (PCA), a standard factorial approachthat recognizes the interlinked nature of variables byaccounting for their correlation and co-variation [29].Methodological details appear in Saint-Jacques et al. [30].Covariates were not included in the spatially-continuousanalysis as the local-EM method does not currentlyaccommodate covariates.Data analysesCommunity-level analysisThe Besag York and Mollié (BYM) model (see [15, 16]),a popular and convenient spatial autoregressive modelfor count data referenced to discrete spatial regions, wasused to perform Community-level analysis. The ap-proach treats the case counts by Community as responsevariables, rather than Standardized Incidence Ratios(SIR), because the latter is unstable when computedSaint-Jacques et al. BMC Public Health  (2016) 16:175 Page 4 of 17from low counts. This is particularly important in thisstudy due to the low population density of NS and therarity of the health outcomes measured. Possible spatialdependence in the data, with pairs of nearby Com-munities tending to be more similar than Communi-ties situated far apart, is accounted for with theinclusion of a spatially autocorrelated random effectterm. The BYM models the case counts as Poissondistributed and supports Baysesian inference formodel fitting, which in this study, was performedseparately for each data set (bladder male, bladder fe-male; kidney male, kidney female) using IntegratedNested Laplace Approximations [31]. Further detailspertaining to this analytical approach are describedin Additional file 1.Spatially-continuous analysisThe local-EM kernel smoothing was used to performthe spatially-continuous analysis. The method developedby Fan, Stafford and Brown [17] was extended by Lee etal. (Lee J, Nguyen P, Brown P, Stafford J, Saint-JacquesN: Local-EM Algorithm for Spatio-Temporal Analysiswith application in Southwestern Nova Scotia. Submit-ted in Ann Appl Stat; [32]) to accommodate the require-ments of modelling the cancer incidence data presentedhere. Collected between 1980 and 2010, the data weresubject to aggregation boundaries changing over timeand were geocoded with varying degrees of precision.Exact spatial locations were derived from full residentialcivic street addresses for most of the recent cancer cases,though the proportion of cases spatially referenced withpartial street address (i.e. postal codes) or with censusregions, increased with the age of the data. Where exactlocation is unavailable, the local-EM kernel smoothingalgorithm produces an optimal risk surface which aver-ages out all the possible locations at which each casecould be located. The bandwidth of the smoothingkernel is chosen by cross-validation (see Additionalfiles 2 and 3) and determines the degree of smoothingin the risk surfaces. A detailed description of themethodology is contained in Lee et al. (Lee J, NguyenP, Brown P, Stafford J, Saint-Jacques N: Local-EM Al-gorithm for Spatio-Temporal Analysis with applicationin Southwestern Nova Scotia. Submitted in Ann ApplStat) and in Nguyen et al. [32], and summarized inAdditional file 1.In this study, local-EM analyses focused on two re-gions of the province which the BYM models suggestedrisk was particularly high, as to describe localized pat-terns in risk. Two models were applied: (1) a spatialmodel testing for significant variation in risk over space,and where a spatial effect was detected; (2) a spatio-temporal model was applied to determine whether riskalso varied significantly over time. Maps were producedwhere statistically significant spatial or spatio-temporaleffects were detected. Estimated risk surfaces based onlocal-EM are not presented to minimize risk of disclos-ure of personal health information. Rather, a p-value fortesting for relative risk being lower than 1.1 (risk lessthan 10 % above the population average) at each locationand time is presented. These p-values were computedwith a parametric bootstrap, with 100 synthetic data-sets simulated with a constant relative risk of λ(s,t) =1.1 and for each s and t the p-value is the proportionof these datasets where the local-EM algorithm yieldsrisk estimates exceeding the estimate produced by thedata. Shown are exceedance probabilities, or oneminus the p-values, which are large when risk is be-lieved to exceed 1.1.The software used was R version 3.1.1 (http://www.r-project.org) in combination with the diseasemapping package [33] and the INLA software [34].This study received ethics approval from CapitalHealth Research Ethics Board. The study was a secondaryanalysis of anonymised cancer registry data obtained fromthe NS Provincial Cancer Registry and a waiver of consentwas approved.ResultsCohort characteristics summaryA total of 6,473 bladder cancers and 3,762 kidney cancerswere diagnosed in NS between 1980 and 2010 (Table 1),95 % of which included spatial information on residenceat time of diagnosis and were successfully geo-referenced.In total, 3,232 bladder and 2,143 kidney cancers were in-cluded in the analyses focusing on the 1998–2010 timeperiod, and; 2,911 bladder and 1,720 kidney cancers wereincluded in the analyses covering the 1980–2010 timeperiod, which focused specifically on cases diagnosed insouth-western (SW) NS (2,767 cases) and Cape Breton(CB; 1,864 cases) — two regions where risk was mappedat a finer spatial resolution. Geo-referencing based onexact residential location at diagnosis was more commonfor cases diagnosed in the most recent time period,between 1998 and 2010 (bladder 86.6 %; kidney85.9 %) than for cases diagnosed between 1980 and2010 (SW: bladder 43.6 %; kidney 47.2 %; CB: bladder43.7 %; kidney 53.7 %). On average, kidney malignan-cies were diagnosed at a slightly younger age thanbladder cancers (65 vs 70 years). Overall, the male tofemale ratio was about 2.9 and 1.5 for bladder andkidney cancer diagnoses, respectively.Spatial patterns of bladder cancerCommunity-level analysisEstimates and credible intervals for regression andvariance parameters obtained from the BYM modelsare shown in Table 2. These coefficients represent theSaint-Jacques et al. BMC Public Health  (2016) 16:175 Page 5 of 17log relative risk in bladder cancer incidence over theentire province and study period. None of the covari-ates – well water usage or material and socialdeprivation – significantly affected the estimated riskfor bladder cancer among males and females (Table 2).Thus, much of the observed spatial heterogeneity inrisk relates to unmeasured risk factors which ap-peared to have a similar effect on the distribution ofdisease in both males and females. Both the spatiallycorrelated and the independent random errors havestandard deviations in the range of 0.1 to 0.4, reason-ably large values considering that they apply to riskon the log scale (Table 2).Figure 1 maps the residual spatial variation in bladdercancer risk, more specifically the posterior meansE[exp(Ui)|data] of the exponentiated random effects,among males (Fig. 1a) and females (Fig. 1b). Thesevalues are equivalently the ratio between the predictedrisk λi for each community and the risk exp(μ + Xiβ)which is typical given the region's covariates Xi. Regionsof elevated risk are common in the south-western sec-tion of the province where several communities exhibitrisk well above what is typical (i.e. > 1.2). Looking atthese Community-level variations for the province, oneidentifies a clear southwest to northeast gradient amongfemales, additional pockets of high risk being observedin Cumberland county (north central region).Uncertainties associated with these maps can be visu-alized with exceedance probabilities, which are the prob-abilities that the risk in a Community or locationexceeds a given threshold, defined here as 10 % abovethe risk that would be typical given the region'sdeprivation and well water usage. We denote these prob-abilities as Pi(10 %) = Pr{λi > [1.1 exp(μ + Xiβ)] | data}, orequivalently Pr[exp(Ui) >1.1|data]. Figure 2a shows ex-ceedance probabilities for bladder cancer amongst males,with 28 communities in SW NS having a probabilityPi(10 %) in excess of 80 % and four communities havingPi(10 %) >95 %, again supporting a southwest to north-east gradient. Estimated risk in these communitiesranged between 1.24 –1.56, and between 1.39 – 1.56,respectively. The exceedance probabilities for females inSW NS are for the most part in the range of 0.2 – 0.8(Fig. 2b), as the smaller number of cases for female can-cers makes it more difficult to assess with any certaintywhether a region has risk above or below a given thresh-old. In total of 9 Communities show exceedance prob-abilities for female risk above 80 % and 2 haveprobabilities above 95 %, the latter located in south centralNS (Fig. 2b). Risk in those areas was higher than that esti-mated for males, with risk ranging between 1.38 – 1.69and between 1.58 –1.69, respectively. Over the 12 year-period, high risk areas (Pr[exp(Ui) >1.1|data] > 80 %) had33 and 52 % more cases of male and female bladder can-cer being diagnosed, respectively.Spatially-continuous analysisTable 3 a shows optimal spatial and spatio-temporal band-widths obtained from cross-validation scores (Additionalfiles 2 and 3) and p-values of Scores-Test that assess thestatistical significance for spatial and spatio-temporaleffects in bladder cancer risk in SW NS and CB. Spatialand spatio-temporal bandwidths determine the extent ofthe smoothing kernel used in risk estimation, and in thisstudy, they ranged between 3 km and 22 km in space and5 to 13 years over time. Based on these bandwidths, weobserved significant localized variations in the spatialdistribution of bladder cancer risk for males from bothSW NS and CB regions (Table 3). For SW NS, the resultssuggested that these spatial patterns also varied over time(Table 3; p = 0.07). Statistically significant spatial variationsin bladder cancer risk were not observed in females fromeither SW NS or CB regions (Table 3). These resultspossibly reflect a combination of small case counts andlocation misclassification. For example, there were only247 cases of female bladder diagnosed between 1980and 2010 in Cape Breton, and 76 % of those weregeocoded to a single location. During cross validation,half the cases would be excluded from model fittingand optimal spatial bandwidths would be determinedbased on too few events to produce stable and statis-tically significant results.Table 2 Posterior summaries for regression and variance parameters – Bladder cancer, Nova Scotia 1998-2010Bladder cancer Males FemalesParameter Mean 2.5 % 97.5 % Mean 2.5 % 97.5 %Intercept −0.105 −0.297 0.086 0.007 −0.301 0.309% using well water 0.001 −0.002 0.003 −0.001 −0.005 0.003Material deprivation −0.297 −0.109 0.048 0.055 −0.067 0.178Social deprivation 0.046 −0.023 0.116 −0.018 −0.130 0.094Spatial standard deviation 0.228 0.157 0.352 0.199 0.086 0.439Unstructured standard deviation 0.124 0.072 0.193 0.240 0.126 0.421Saint-Jacques et al. BMC Public Health  (2016) 16:175 Page 6 of 17Exceedance probabilities obtained from fitting aspatially continuous risk surface with the local-EM algo-rithm are shown in Fig. 3 for male bladder cancer in SWNS and CB. These exceedance probabilities can be inter-preted in a similar manner to the quantities from theBYM model shown in Fig. 2, with one difference beingthey refer to a threshold of 10 % above the average riskfor NS without adjustment for deprivation and wellwater usage. Another difference is these probabilitiesvary over a continuous spatial surface as opposed tobetween Communities with set boundaries and, hence,provide insights on finer resolution patterns in risk. Thus,we write, P(s;10 %) as one minus a p-value for testingλ(s) < 1.1 with probabilities being computed usingFig. 1 Posterior means relative risks for male (a) and female (b) bladder cancer, Nova Scotia 1998–2010Saint-Jacques et al. BMC Public Health  (2016) 16:175 Page 7 of 17parametric bootstrapping (see details in Nguyen et al.[32] and Lee et al. (Lee J, Nguyen P, Brown P, StaffordJ, Saint-Jacques N: Local-EM Algorithm for Spatio-Temporal Analysis with application in SouthwesternNova Scotia. Submitted in Ann Appl Stat). As ob-served using Bayesian inference, results from thesefiner-scale analyses also show probabilities of above-average risk in excess of 80 % along the Fundy shoreand near Cape Sable Island and Shelburne, areas lo-cated on the south shore of NS (Fig. 3a). In Cape Bre-ton, patterns of exceedance probabilities in excess of80 % (Fig. 3b) pointed to areas of elevated risk whereaggregated analysis based on BYM modeling hadshown Pi(10 %) to be less than 20 % (Fig. 2a).Fig. 2 Exceedance probabilities (Pi(10 %)) for male (a) and female (b) bladder cancer, Nova Scotia 1998–2010Saint-Jacques et al. BMC Public Health  (2016) 16:175 Page 8 of 17Figure 4 shows the exceedance probabilities obtainedfrom fitting a spatio-temporal risk surface to male blad-der cancer for SW NS, a region where risk varied overtime (Table 3). In this latter model, where risk varies intime as well as in space, we write P(s,t;10 %) as oneminus a p-value for testing λ(s,t) < 1.1. Here, P(s,t;10 %)is shown for four specific years, 1980, 1990, 2000 and2010. Exceedance probabilities for the intervening yearscan be found in the supplementary materials and athttp://pbrown.ca/jlee/spatio_temporal/. Note that whilepatterns of exceedance probabilities for year 2000 (i.e.Fig. 4c) includes data from 1980–2010, the 13 years clos-est to this index year will have the greatest influenceupon parameter estimates. This is because the relativeinfluence is determined by a weighting function that fol-lows a Gaussian distribution with a standard deviationof 13 years (i.e. optimal temporal bandwidth for malebladder cancer). Simultaneously, the spatial weightingfunction associated with a point estimate also followsfrom a Gaussian distribution with a standard deviationof 11 km (i.e. optimal spatial bandwidth for male bladdercancer). Overall, the results are similar to those obtainedwith the spatial model, highlighting large areas withP(s,t;10 %) above 80 % along the Fundy Shore and southportion of the region. However, when adding a temporalcomponent and thus further zooming into a finer scaleof analyses, several locations show P(s,t; 10 %) surpassing95 %, pointing to broad areas of significantly elevatedrisk where the estimated relative risk varied between1.27 – 2.84 (not shown).Spatial patterns of kidney cancerCommunity-level analysisAs observed for bladder cancer, posterior summaries forregression and variance parameters show that the mea-sured covariates had no significant influence on the esti-mated risk of kidney cancer (Table 4). Random effectsfor both spatially and unstructured random errors weresignificant, although showing greater unstructured het-erogeneity for males than previously observed with malebladder cancer risk (i.e. ranging between 0.17 – 0.27 vs0.07 – 0.19, respectively; Tables 2, 4). Maps of posteriormeans displayed strong spatial heterogeneity in male andfemale kidney cancer risk (Fig. 5a-b). Regions of elevatedrisk for male kidney cancer were common in the south-western region of the province as well as in several com-munities of CB Island, correlating with the elevated riskobserved amongst females which is uniformly high in thatregion (Fig. 5a-b). Female kidney cancer rates wereelevated in some Communities along the southern shoreof SW NS and around the south shore of central NS(Fig. 5b). Figure 6a-b shows Pi(10 %) for kidney cancerand a risk threshold that would be typical given theregion's deprivation and well water usage. In total, 11Table 3 Optimal spatial and temporal bandwidth (BW) fromcross-validation scores, bladder and kidney cancer, NovaScotia 1980-2010Spatial Spatio-temporalRegion Sex BW (Km) P-value BW (Km) BW (years) P-valueBladderSW M 11 <0.001 11 13 0.07F 3 0.41 - - -CB M 4 0.01 4 13 >0.2F 22 0.79 - - -KidneySW M 3 0.03 3 17 >0.2F 7 0.05 7 13 >0.2CB M 6 0.01 6 5 >0.2F 10 0.38 - - -Fig. 3 Bootstrapped exceedance probabilities (P(s; 10 %)) for risksurface of male bladder cancer in south-western Nova Scotia (a)and Cape Breton (b) regionsSaint-Jacques et al. BMC Public Health  (2016) 16:175 Page 9 of 17Fig. 4 Bootstrapped exceedance probabilities (P(s, t; 10 %)) for risk surface of male bladder cancer for 1980, 1990, 2000, 2010, in south-westernNova ScotiaTable 4 Posterior summaries for regression and variance parameters – Kidney cancer, Nova Scotia 1998-2010Kidney cancer Males FemalesParameter Mean 2.5 % 97.5 % Mean 2.5 % 97.5 %Intercept 0.032 −0.231 0.290 0.038 −0.259 0.326% using well water −0.001 −0.004 0.002 −0.001 −0.004 0.003Material deprivation −0.006 −0.112 0.097 0.052 −0.064 0.167Social deprivation 0.008 −0.087 0.103 0.0004 −0.107 0.109Spatial standard deviation 0.138 0.048 0.298 0.156 0.052 0.366Unstructured standard deviation 0.265 0.174 0.390 0.251 0.137 0.440Saint-Jacques et al. BMC Public Health  (2016) 16:175 Page 10 of 17Communities showed Pi(10 %) in excess of 80 % amongstmales (estimated risk: 1.36 – 2.52); 2 of these beingstatistically significant (i.e. Pr[exp(Ui) >1.1|data) >0.95;estimated risk: 1.73 – 2.52). The majority of theseCommunities are located along the south shore ofSW NS (Fig. 6a). Exceedance probabilities above 80 %for females risk were observed in 8 Communities(estimated risk: 1.35 – 1.86); 4 located along thesouth shore of SW NS and 4 along the north shoreof CB (Fig. 6b). Of these, 1 had a statistically signifi-cant probability (estimated risk: 1.87). Over the12 year-period, high risk areas (Pr[exp(Ui) >1.1|data] >80 %) had 52 and 57 % more cases of male and femalekidney cancer being diagnosed, respectively.Fig. 5 Posterior means relative risks for male (a) and female (b) kidney cancer, Nova Scotia 1998–2010Saint-Jacques et al. BMC Public Health  (2016) 16:175 Page 11 of 17Spatially continuous analysisOptimal spatial and spatio-temporal bandwidths fromcross-validation scores (Additional files 2 and 3) and as-sociated p-values testing for spatial and spatio-temporaleffects in kidney cancer risk, are shown in Table 3. Basedon these bandwidths, we observed significant variation inthe spatial distribution of kidney cancer risk in males andfemales from SW NS and in males from CB. Statisticallysignificant spatio-temporal effects were not observed(Table 3; p > 0.2) and therefore maps of exceedance prob-abilities were derived from the spatial models with 30 yearsof pooled data (1980–2010). In comparison to the resultsobtained with BYM modeling, probabilities in excess of 80and 95 % had a larger spatial extent. This pattern wasFig. 6 Exceedance probabilities (Pi(10 %)) for male (a) and female (b) kidney cancer, Nova Scotia 1998–2010Saint-Jacques et al. BMC Public Health  (2016) 16:175 Page 12 of 17generally observed across regions and genders. Inaddition, the probabilities produced by local-EM were lessspatially smooth, allowing the detection of more localizedrisk. Again, P(s;10 %) for males in SW NS showed a highprobability of excess risk along the southern shore, butalso toward the centre of the region. Significant probabil-ities of exceedance in risk of male kidney cancer were alsodetected in several areas of CB; an occurrence thatwas not observed with BYM models (Fig. 6a, 7b).Correspondingly, exceedance probabilities for femaleswere high along the southern shore of SW NS (Fig. 8).Overall, estimated relative risk for female kidney can-cer ranged between 1.34 – 1.98 and 1.45 –1.98, forP(s;10 %)|data) > 0.80 and P(s;10 %)|data) > 0.95, re-spectively. For males, these values ranged between1.53 – 2.54 and 2.01 –2.54.DiscussionSummary of findingsThis study showed evidence of spatial variation in therisk of bladder and kidney cancer in Nova Scotia.Posterior summaries for regression and variance parame-ters suggested that much of the heterogeneity in risk re-lated to unmeasured risk factors. High risk areas forbladder cancer were predominantly distributed along asouthwest to northeast gradient. Kidney cancer riskfollowed a similar distribution, although areas of elevatedrisk were also detected in various northeast Communitiesof Cape Breton, for both genders. Focusing on aggregatedspatial units (Communities), the study showed that areasidentified to have high probability of exceedance (BYM:Pr[exp(Ui) >1.1|data] > 80 %) in the risk of male (28 Com-munities) or female (9 Communities) bladder cancer had33 % (males) and 52 % (females) more cases diagnosedover the 12 year period, compared to the number of casesexpected. Similarly, high risk areas for male (11 Commu-nities) or female (8 Communities) kidney cancer had 52 %(males) and 57 % (females) more cases diagnosed than ex-pected. From a public health perspective, this translates inan excess of nearly 200 urinary tract cancer (UTC) cases(150 bladder; 45 kidney) being diagnosed in those high riskareas where the estimated risk was observed to be at least10 % above the NS average rate. Over a 12 year period, thiscorresponds to an additional 16 UTC cases annually, aconservative figure given that exceedance probabilities inexcess of both 80 % and 95 % had much larger spatial ex-tent when derived from the spatially-continuous analysisthan with the Community-level model. This was true forrisk measured in either sex or cancer site. Focusing onlocalized spatial patterns, this study also highlightedsignificant spatial and spatio-temporal variations in theFig. 7 Bootstrapped exceedance probabilities (P(s; 10 %)) for risksurface of male kidney cancer in south-western Nova Scotia (a) andCape Breton (b) regionsFig. 8 Bootstrapped exceedance probabilities (P(s; 10 %)) for risksurface of female kidney cancer in south-western Nova ScotiaSaint-Jacques et al. BMC Public Health  (2016) 16:175 Page 13 of 17risk of male bladder cancer within SW NS, with areasof elevated risk along the Fundy shore and south shoreof the region. Elevated risk of both, male and femalekidney cancer were also observed along the south shoreof SW NS. In addition, risk for both male bladder andkidney cancer varied significantly in CB, although areasof elevated risk did not always overlap. Overall, spatialpatterns were generally stable over time.Interpretation of spatial patternsPatterns of spatiotemporal heterogeneity in risk provideclues to the occurrence and influence of extrinsic factorsinvolved in the rise or fall of a disease. In this study, pat-terns of spatial variations in bladder and kidney cancersrisk were stable over time, suggesting persistent risk ex-posure. The exception being male bladder, for which theresults pointed to a temporal effect. However, the pat-tern of spatial variations in risk remained stable over a13 year period, possibly also reflecting persistent effects.Similarly, a study of space-time patterns of bladder can-cer incidence in Utah, US, detected high risk areas thatwere persistent over time [35]. These high relative riskareas were subsequently found to be associated with thepresence of Toxic Release Inventory sites, where the riskwas observed to range between 1.14 and 1.82 for bothgenders combined and between 1.12 to 1.47 for malesonly. While the processes generating the elevated risk inNS are unknown, the magnitude of the estimated risk inhigh risk areas for NS was similar to that reported in Utah,ranging between 1.24 – 1.56 and 1.38 – 1.69 among malesand females, respectively based on BYM and between1.48 – 1.99 and 1.48 – 1.95 among male from SW NS andCB, respectively, when based on local-EM. The latter tigh-ter lower bounds of the estimates are attributable tothe more conservative rule of exceedance probabilityapplied in NS (NS: Pi(10 %) > 0.8 and P(s;10 %) > 0.8;Utah: P(exp(si) >1.0|data) > 0.8) for the determinationof high risk areas. Both studies suggest an increasedeffect in females.Several factors affect the incidence of urinary tractcancers worldwide. Exposure to tobacco smoke, occupa-tional toxins and environmental source of heavy metalssuch as arsenic in drinking water, are amongst wellestablished risk factors for bladder cancer, in particular,transitional cell carcinoma which account for 90 % ofthe bladder cancer cases diagnosed in developed coun-tries [5, 7, 19]. Tobacco smoking [5, 9, 36–41] and long-term exposure to high levels of arsenic in drinking wateralso increase kidney cancer risk [19, 42] along with obes-ity [38, 43, 44], hypertension [38], the use of phenacitin-containing analgesics and exposure to trichloroethyleneand polycyclic aromatic hydrocarbons [38, 45–47].Whether measured independently or synergistically, themagnitude of influence of these risk factors for thedevelopment of UTC varies. However, meta-analyses ofover 30 years of epidemiological studies suggest, for in-stance, that tobacco smoking could increase the risk ofbladder and kidney cancer by at least 270 and 50 %, re-spectively, in current smokers compared to non-smokers[37, 48]. Exposure to arsenic in drinking water shows ef-fects of similar magnitude, increasing the risk of bladdercancer by about 40 %, 230 and 310 % at levels exposure of10, 50 and 150 μg/L, respectively [19]. Obesity has beenreported to account for 30–40 % of kidney cancer cases inEurope and the United States; and is known to increasethe risk of renal cell carcinoma in a dose–response fashion[12, 49]In this study, residual spatial variation and resultingprobabilities of exceedance for bladder and kidney cancerrisk suggest that smoking is not the only factor contributingto the observed spatial patterns. This is because the proxymeasures of smoking included in the analyses (i.e. socialand material deprivation indices) did not change the spatialvariations in risk or its magnitude. As well, the heterogen-eity in bladder and kidney cancer risk observed in high riskareas was greater than what could be accounted by knownspatial variations in smoking prevalence in Nova Scotia.Nonetheless, synergistic relationships between smoking andother un-measured risk factors cannot and should not beruled out. This is especially important in Nova Scotia, aprovince known for its high prevalence of tobacco smoking[50], obesity [51] and where inorganic arsenic in drinkingwater was observed to be a major contributor to arsenicbody burden in a study population [52]. Overall, the twospatial approaches used to model disease risk provided con-sistent and complementary results. Inclusion of a time-varying component in the spatially-continuous models per-mitted the determination of whether high average risk in agiven location was sustained over time or changed overtime; two different situations that could be derived fromthe same number of accumulated cases in an area over aset time period. As described by Abellan et al. [53], theepidemiologic interpretations of these two situationsare important. In one scenario, spatial patterns aremore likely to occur in a constant manner over timeand hence could be induced by environmental orsocio-demographic risk factors that act in a sustainedmanner. In the second scenario, the rate of case accu-mulation may be more temporally clustered with dis-tinct variability, possibly reflecting emerging short-latency risk factors that would generate high excesscases in shorter time intervals or, alternatively, due toartificial or sudden variations associated with changesin disease coding or screening practices (see details inAbellan et al. [53]). Hence, it would not be unreasonableto suggest that the observed heterogeneity in thespatial distribution of high-risk areas for bladder andkidney cancer in both SW NS and CB, support aSaint-Jacques et al. BMC Public Health  (2016) 16:175 Page 14 of 17scenario in which risk factors act in a relatively sus-tained manner over time.Strengths and limitationsThis study has important strengths. First, it is based on30 years of cancer incidence data obtained from apopulation-based cancer registry adhering to registrationstandards of both the Canadian Cancer Registry and theNorth American Association of Central Cancer Regis-tries. Those standards allow for consistency in diseasecoding over time and; ensure case ascertainment andcompleteness through a network of activities includingautomated and manual edit processes, record linkagesand data audits. In addition, the systematic collection ofspatial information at time of diagnosis enabled 100 % ofcases in Cohort 1 and 95 % of cases in Cohort 2 to besuccessfully geo-referenced with a high degree of cer-tainty, thus minimizing location misclassification (Cohort1, ~ 85 % exact location; Cohort 2, ~ 50 %). Second, thetwo statistical methods used in this study accounted forspatial dependence (random effects) in risk estimateswhich reduce the likelihood of Type I error – declaring anarea as having elevated risk when in fact its underlyingtrue rate equals the background level [54]. Third, theexceedance probability rules, Pi(10 %) > 0.8, P(s;10 %) > 0.8and P(s,t;10 %) > 0.8, used here to classify spatial risk hashigh specificity even when data are sparse, further redu-cing the risk of false alarms, although perhaps increasingthe likelihood of Type II error – declaring an area as hav-ing average risk when in fact its underlying true rate is ele-vated relative to background levels [54]. Fourth, theapplication of the local-EM algorithm treated risk as a con-tinuously varying process in space and time and so was notconstrained to be within arbitrary administrative boundar-ies which often change between census periods [52]. Thisallows for the integration and use of irregularly aggregatedor point-location data within a single framework and min-imizes loss of information. It presents a real advantage forthe estimation of disease risk in small-area analyses or forrare diseases that requires the monitoring and accumula-tion of cases collected over a long time period as it maxi-mizes statistical power and results in more meaningfulinference [55]. As such, it is reasonable to suggest that ap-plying the Local-EM framework improved the sensitivityof the study, offering a balance to the Community-level autoregressive model, a more conservative approachwith generally lower sensitivity (see [54, 55]. Finally, mod-elling the spatio-temporal variation in risk with local-EMalgorithm provided useful insights about the stabilityof the estimated spatial patterns of disease. It alsoproduced predictions that were generally less spatiallysmooth, and as such, is a more sensitive tool for thedetection of localized areas of elevated risk, whichultimately better informs health service planning,public health interventions and resource allocation.Nonetheless, this study has limitations. First, location attime of diagnosis was used as a surrogate for the locationwhere a person was thought to be exposed to factorswhich increased their risk of cancer. This is a commonapproach in the geographic analyses of many diseaseoutcomes given the difficulty of obtaining a full history ofresidence and building estimates of lifetime exposure. Theconsequent exposure misclassification can result in lessinformative maps that impedes hypothesis generation oridentification of environmentally or sociologically drivenprocesses occurring over long time periods. Second,individual-level information on important risk factorssuch as smoking frequency and duration was not availableas cancer registries do not routinely collect informationunrelated to patient care. This study used neighbourhoodsocial and material deprivation as a proxy for smokingprevalence. As a result, it is possible that maps of posteriormeans relative risks include some residual confoundingdue to smoking. Third, current algorithms for local-EM estimation do not allow for the inclusion of co-variates. Fourth, the method is computationally inten-sive. Finally, although the local-EM analyses benefitedfrom the inclusion of cases diagnosed over a longertime period, when reporting for the Cape Breton region,the number of cases was still quite low, which resultedin unstable results. This was particularly evident whendetermining optimal spatial and temporal bandwidthsin females risk for which incidence counts was about1.5 to 3 times lower than for males.ConclusionModeling the geographical distribution of disease withina population is essential to public health surveillance. Itpermits the quantification of the risk of disease relativeto expected background levels, and the identification ofunusually high and low risk areas which can guide healthservice planning, public health intervention and resourceallocation. The current approach further permits theestimation of residual spatial dependence resulting fromexposure to unmeasured risk variables, and as such,helps identify areas where other etiological factors maybe at play. In this study, spatial analyses demonstratedevidence of spatial heterogeneity in the risk of both blad-der and kidney cancers in Nova Scotia. The temporalcomponent of the spatially-continuous approach permit-ted the determination of the relative time scales of highaverage risk in a given area and hence provided an un-derstanding of the stability of the spatial patterns of theestimated risk; and the generation of hypotheses aboutthe nature of possible exposure. Based on this infor-mation, we suggest that the excess bladder and kidneycancer risk for both male and potentially, female inSaint-Jacques et al. BMC Public Health  (2016) 16:175 Page 15 of 17south-western NS may be driven by exposure to unknownrisk factors that act in a sustained manner over time. Fur-ther research may uncover the nature of these factors andlead to future opportunities for disease prevention.The findings from this study warrant further investiga-tion in three main areas. First, further work is requiredin the area of exposure modeling in order to elucidatethe potential factors driving the observed patterns ofvariations in the risk of UTC in NS. Second, they high-light the need for the development of local-EM methodsthat incorporate individual- and neighborhood-level co-variates. Finally, they reaffirm the need for the establish-ment of a public health platform that would enable thecollection of individual- and/or neighborhood level in-formation relating to disease causing-risk factors, suchas behavioural, occupational and environmental factors.Such information permits more accurate quantificationand understanding of disease risk.Additional filesAdditional file 1: Analytical details [27, 55, 56]. (DOC 42 kb)Additional file 2: Spatial cross-validation scores for the selection ofoptimal bandwidths. (PNG 152 kb)Additional file 3: Temporal cross-validation scores for the selectionof optimal bandwidths. (PNG 109 kb)AbbreviationsBYM: Besag York and Mollie; CB: Cape Breton; Local-EM: local expectationmaximization; NS: Nova Scotia; SW: South-Western; UTC: urinary tract cancer.Competing interestsThe authors declare that they have no competing interests.Authors’ contributionsNSJ extracted the cases files; georeferenced cases; conducted all analyses relatingto BYM application, constructed tables and figures, drafted and revised themanuscript; JL modified existing Local-EM methods to incorporate temporality,carried-on all work relating to local-EM based analysis; PB devised the study,drafted section describing Local-EM methods, reviewed the article critically forimportant statistical content, provided assistance in the interpretation of theresults, supervised NSJ and JL for statistical work; JS assisted JL in developing thelocal-EM methodology, supervised JL, reviewed the article critically for importantstatistical content; LP devised the study, reviewed the article critically for importantintellectual content and provided assistance in the interpretation. TD devisedthe study, supervised the overall work, reviewed the article critically forimportant intellectual content and provided assistance in the interpretation.All authors read and approved the final manuscript.AcknowledgementThis work was supported by the Canadian Cancer Society [grant number19889]; the Nova Scotia Health Research Foundation [MED SRA 009 5524 toN.S.J.]; and the Canadian Institute for Health Research [201010GSD-249658-164753 to N.S.J.]. We thank Ron Dewar from Cancer Care Nova Scotia for hisinvaluable guidance, and Cancer Care Nova Scotia for its continued support.Author details1Cancer Care Nova Scotia, Surveillance and Epidemiology Unit, Room 560Bethune Building, 1276 South Street, Halifax B3H 2Y9NS, Canada.2Interdisciplinary PhD program, Dalhousie University, 6299 South Street,Room 314, PO Box 15000, Halifax B3H 4R2NS, Canada. 3Department ofStatistical Sciences, University of Toronto, 100 St. George St., Toronto M5S3G3ON, Canada. 4Cancer Care Ontario, 620 University Ave, Toronto M5G2 L7ON, Canada. 5Department of Pediatrics and Population Cancer ResearchProgram, Dalhousie University, 1494 Carlton Street, PO Box 15000, HalifaxB3H 4R2NS, Canada. 6The University of British Columbia, School ofPopulation and Public Health, 2206 East Mall, Vancouver V6T 1Z3BC, Canada.Received: 29 May 2015 Accepted: 22 January 2016References1. Murta-Nascimento C, Schmitz-Dräger B, Zeegers M, Steineck G, KogevinasM, Real F, et al. Epidemiology of urinary bladder cancer: from tumordevelopment to patient’s death. World J Urol. 2007;25:285–95.2. Parkin DM. The global burden of urinary bladder cancer. Scand J UrolNephrol. 2008;42(4 supp 218):12.3. Kidney cancer statistics | World Cancer Research Fund International[http://www.wcrf.org/int/cancer-facts-figures/data-specific-cancers/kidney-cancer-statistics]4. Mathew A, Devesa SS, Fraumeni JFJ, Chow W-H. Global increases in kidneycancer incidence, 1973–1992. J Cancer Prev. 2002;11:171–8. April 2002.5. Janković S, Radosavljević V. Risk factors for bladder cancer. Tumori.2007;93:4–12.6. Nielsen ME, Smith AB, Meyer A-M, Kuo T-M, Tyree S, Kim WY, et al. Trends instage-specific incidence rates for urothelial carcinoma of the bladder in theUnited States: 1988 to 2006. Cancer. 2014;120:86–95.7. Leppert JT, Shvarts O, Kawaoka K, Lieberman R, Belldegrun AS, Pantuck AJ.Prevention of Bladder Cancer: A Review. Eur Urol. 2006;49:226–34.8. Shariat SF, Sfakianos JP, Droller MJ, Karakiewicz PI, Meryn S, Bochner BH.The effect of age and gender on bladder cancer: a critical review of theliterature. BJU Int. 2010;105:300–8.9. Burger M, Catto JWF, Dalbagni G, Grossman HB, Herr H, Karakiewicz P, et al.Epidemiology and Risk Factors of Urothelial Bladder Cancer. Eur Urol. 2012.10. Kachuri L, De P, Ellison LF, Semenciw R. Cancer incidence, mortality andsurvival trends in Canada, 1970–2007. Chronic Dis Inj Can. 2013;33:69–80.11. Canadian Cancer Society’s Advisory Committee on Cancer Statistics.Canadian Cancer Statistics 2015. Toronto, ON: Canadian Cancer Society;2015. ISSN 0835-2976.12. De P, Otterstatter MC, Semenciw R, Ellison LF, Marrett LD, Dryer D. Trends inincidence, mortality, and survival for kidney cancer in Canada, 1986–2007.Cancer Causes Control CCC. 2014;25:1271–81.13. Liu S, Semenciw R, Morrison H, Schanzer D, Mao Y. Kidney cancer inCanada: the rapidly increasing incidence of adenocarcinoma in adults andseniors. Can J Public Health Rev Can Santé Publique 1997;88:99–104.14. CANSIM - 103-0553 - New cases and age-standardized rate for primarycancer (based on the May 2015 CCR tabulation file), by cancer type and sex,Canada, provinces and territories [http://www5.statcan.gc.ca/cansim/a26?lang=eng&retrLang=eng&id=1030553&paSer=&pattern=&stByVal=1&p1=1&p2=31&tabMode=dataTable&csid]15. Besag J, York J, Mollié A. Bayesian image restoration, with two applicationsin spatial statistics. Ann Inst Stat Math. 1991;43:1–20.16. Waller L, Carlin B. In: Diggle P, Fuentes M, Gelfand AE, Guttorp P, editors.Handbook Of Spatial Statistics. Boca Raton, FL: Taylor and Franci. USA:Chapman and Hall CRC Press; 2010.17. Fan C-PS, Stafford J, Brown PE. Local-EM and the EMS Algorithm. J ComputGraph Stat. 2011;20:750–66.18. Terashima M, Rainham DGC, Levy AR. A small-area analysis of inequalities inchronic disease prevalence across urban and non-urban communities in theProvince of Nova Scotia, Canada, 2007–2011. BMJ Open. 2014;4:e004459.19. Saint-Jacques N, Parker L, Brown P, Dummer TJ. Arsenic in drinking waterand urinary tract cancers: a systematic review of 30 years of epidemiologicalevidence. Environ Health. 2014;13:44.20. Shohaimi S, Luben R, Wareham N, Day N, Bingham S, Welch A, et al.Residential area deprivation predicts smoking habit independently ofindividual educational level and occupational social class. A cross sectionalstudy in the Norfolk cohort of the European Investigation into Cancer(EPIC-Norfolk). J Epidemiol Community Health. 2003;57:270–6.21. Marmot M, Wilkinson RG. Social Determinants of Health. Second. GreatBritain: Oxford University Press; 2008.22. Cohen SS, Sonderman JS, Mumma MT, Signorello LB, Blot WJ. Individual andneighborhood-level socioeconomic characteristics in relation to smokingprevalence among black and white adults in the Southeastern UnitedStates: a cross-sectional study. BMC Public Health. 2011;11:877.Saint-Jacques et al. BMC Public Health  (2016) 16:175 Page 16 of 1723. Turrell G, Hewitt BA, Miller SA. The influence of neighbourhooddisadvantage on smoking cessation and its contribution to inequalities insmoking status. Drug Alcohol Rev. 2012;31:645–52.24. Hossain MP, Palmer D, Goyder E, El Nahas AM. Association of deprivation withworse outcomes in chronic kidney disease: findings from a hospital-basedcohort in the United Kingdom. Nephron Clin Pract. 2012;120:c59–70.25. Pampalon R, Raymond G. A deprivation index for health and welfareplanning in Quebec. Chronic Dis Can. 2000;21:104–13.26. Pampalon R, Hamel D, Gamache P, Philibert MD, Raymond G, Simpson A. Anarea-based material and social deprivation index for public health in Québec andCanada. Can J Public Health Rev Can Santé Publique. 2012;103(8 Suppl 2):S17–22.27. Blackweel DL. Family structure and children’s health in the United States:findings from the National Health Interview Survey, 2001–2007. Vital HealthStat. 2010;246:1–166. Data from the National Health Survey Interview Survey.28. Remes H, Martikainen P, Valkonen T. The effects of family type on childmortality. Eur J Public Health. 2011;21:688–93.29. Legendre P, Legendre L. Numerical Ecology. Amsterdam, Netherlands:Elsevier; 1998.30. Saint-Jacques N, Cui Y, Parker L, Dummer T. Premature mortality due tosocioeconomic inequalities in Nova Scotia, Canada. Health Place. 2013.31. Rue H, Martino S, Chopin N. Approximate Bayesian inference for latentGaussian models by using integrated nested Laplace approximations.J R Stat Soc Ser B Stat Methodol. 2009;71:319–92.32. Nguyen P, Brown PE, Stafford J: Mapping cancer risk in southwesternontario with changing census boundaries. Biometrics 2012;68:1228–1237.33. Brown P. Model-based geostatistics the easy way. J Stat Softw. 2015;63(12):1–24.34. Martins TG, Simpson D, Lindgren F, Rue H. Bayesian computing with INLA:New features. Comput Stat Data Anal. 2013;67:68–83.35. Fortunato L, Abellan JJ, Beale L, LeFevre S, Richardson S. Spatio-temporalpatterns of bladder cancer incidence in Utah (1973–2004) and theirassociation with the presence of toxic release inventory sites. Int J HealthGeogr. 2011;10:16.36. Sasco AJ, Secretan MB, Straif K. Tobacco smoking and cancer: a brief reviewof recent epidemiological evidence. Lung Cancer. 2004;45(Supplement 2):S3–9. Trends in Diagnosis and Therapy of Lung Cancer.37. Gandini S, Botteri E, Iodice S, Boniol M, Lowenfels AB, Maisonneuve P, et al.Tobacco smoking and cancer: A meta-analysis. Int J Cancer. 2008;122:155–64.38. Chow W-H, Dong LM, Devesa SS. Epidemiology and risk factors for kidneycancer. Nat Rev Urol. 2010;7:245–57.39. Freedman ND, Silverman DT, Hollenbeck AR, Schatzkin A, Abnet CC.Association between smoking and risk of bladder cancer among men andwomen. JAMA. 2011;306:737–45.40. Pou SA, Osella AR, Diaz MDP. Bladder cancer mortality trends andpatterns in Córdoba, Argentina (1986–2006). Cancer Causes Control CCC.2011;22:407–15.41. Ferrís J, Berbel O, Alonso-López J, Garcia J, Ortega JA. Environmentalnon-occupational risk factors associated with bladder cancer. Actas UrolEsp. 2013;37:579–86.42. IARC. A Review of Human Carcinogens. C. Metals, Arsenic, Fibres and Dusts.Monograph. Lyon: International Agency for Research on Cancer; 2012.p. 41–93. Arsenic and Arsenic Counpounds.43. Lipworth L, Tarone RE, McLaughlin JK. The Epidemiology of Renal CellCarcinoma. J Urol. 2006;176:2353–8.44. Wang Y, Chen X, Song Y, Caballero B, Cheskin LJ. Association betweenobesity and kidney disease: A systematic review and meta-analysis. KidneyInt. 2007;73:19–33.45. Haalboom B, Elliott SJ, Eyles J, Muggah H. The risk society at work in theSydney “Tar Ponds”. Can Geogr. 2006;50(2):227–241.46. Lambert TW, Guyn L, Lane SE. Development of local knowledge ofenvironmental contamination in Sydney, Nova Scotia: Environmental healthpractice from an environmental justice perspective. Sci Total Environ. 2006;368:471–84.47. Kiriluk KJ, Prasad SM, Patel AR, Steinberg GD, Smith ND. Bladder cancer riskfrom occupational and environmental exposures. Urol Oncol Semin OrigInvestig. 2012;30:199–211. Environmental Exposures and GenitourinaryMalignancies.48. Zeegers MPA, Tan FES, Dorant E, van den Brandt PA. The impact ofcharacteristics of cigarette smoking on urinary tract cancer risk. Cancer.2000;89:630–9.49. Calle EE, Kaaks R. Overweight, obesity and cancer: epidemiological evidenceand proposed mechanisms. Nat Rev Cancer. 2004;4:579–91.50. Reid J, Hammond D, Rynard V, Burkhalter R. Tobacco Se in Canada: Patternsand Trends. Waterloo, ON: Propel Centre for Population Health Impact,University of Waterloo; 2015.51. Gotay CC, Katzmarzyk PT, Janssen I, Dawson MY, Aminoltejari K, Bartley NL.Updating the Canadian Obesity Maps: An Epidemic in Progress. Can JPublic Health. 2012;104:64–8.52. Dummer TJB, Yu ZM, Nauta L, Murimboh JD, Parker L. Geostatistical modellingof arsenic in drinking water wells and related toenail arsenic concentrationsacross Nova Scotia, Canada. Sci Total Environ. 2015;505:1248–58.53. Abellan JJ, Richardson S, Best N. Use of space-time models to investigate thestability of patterns of disease. Environ Health Perspect. 2008;116:1111–9.54. Richardson S, Thomson A, Best N, Elliott P. Interpreting Posterior RelativeRisk Estimates in Disease-Mapping Studies. Environ Health Perspect. 2004;112:1016–25.55. Holowaty EJ, Norwood TA, Wanigaratne S, Abellan JJ, Beale L. Feasibility andutility of mapping disease risk at the neighbourhood level within a Canadianpublic health unit: an ecological study. Int J Health Geogr. 2010;9:21.•  We accept pre-submission inquiries •  Our selector tool helps you to find the most relevant journal•  We provide round the clock customer support •  Convenient online submission•  Thorough peer review•  Inclusion in PubMed and all major indexing services •  Maximum visibility for your researchSubmit your manuscript atwww.biomedcentral.com/submitSubmit your next manuscript to BioMed Central and we will help you at every step:Saint-Jacques et al. BMC Public Health  (2016) 16:175 Page 17 of 17


Citation Scheme:


Citations by CSL (citeproc-js)

Usage Statistics



Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            async >
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:


Related Items