NASCAR has been previously described previously and is a novel linked longitudinal dataset comprising over 12,000 individuals from Northeast Scotland diagnosed with one of eight common cancers (colorectal, lung, breast, prostate, melanoma, oesophago-gastric, cervical, ovarian) from 2007 to 2014 [10]. The primary data source for NASCAR is the NHS Grampian Cancer Care Pathway (CCPd) database which comprises information about individuals’ journey to cancer diagnosis in NHS Grampian, and two island communities NHS Orkney and NHS Shetland. All three diagnosis centres lead to treatment at a single cancer treatment centre (Aberdeen Royal Infirmary). The CCPd records information about referral, diagnosis, subsequent investigations and secondary care appointments, intra-secondary care referrals, investigations, hospital admissions and discharges, operations, and treatment. The accuracy of the CCPd has been validated [10]. Using residential postcodes (geo-reference for postcode centroid), we assigned the Scottish Index of Multiple Deprivation (SIMD) and the Scottish Government Urban-Rural Classification to the whole CCPd cohort dataset [15, 18]. The SIMD data come from a variety of different sources and data providers quality assure data before providing them to the Office of the Chief Statistician and Performance (OCSP) [19]. The OCSP also carries out further checks to ensure the data are fit for purpose. This combined dataset was then linked to the Scottish Cancer Registry (SMR06), which records information on cancer type, date of diagnosis, stage at diagnosis, treatment received and date [20]. Data quality of the Scottish Cancer Registry is monitored using routine indicators, computer validations and ad hoc studies of data accuracy and completeness of ascertainment [21–23]. Further linkage was made to hospital episode data relating to all inpatient and day cases discharged from Scottish acute hospitals (Scottish Morbidity Record 01 (SMR01)). The data quality is regularly assessed and validated [24]. Using these data, a Charlson co-morbidity index (CCI) was calculated for each patient [25]. Death registry data from the General Registry Office for Scotland (GROS) provided information relating to all principal and secondary causes of death.

For each individual within the dataset, postcodes for home residence, GP practice, cancer diagnosis and cancer treatment centre were available. To model travel times to key healthcare facilities used during each individuals cancer diagnostic pathway (GP practice, cancer diagnosis centre, and cancer treatment centre), mainland road networks from place of residence were calculated using Network Analyst extension in ArcGIS V10.2 (ESRI: Environmental Systems Research Institute, Redlands, CA, USA). A travelling time was not calculated for island patients, and island was used a distinct category.

Our analyses investigated the relationship between rurality, distance and travelling times to key healthcare facilities (GP practice, hospitals of diagnosis and treatment) and outcomes (receipt of timely treatment based on Scottish Government targets, which are within 62 days of GP referral and within 31 days of diagnosis [26]) and one year mortality. Multivariable logistic regression adjusted for potential confounding variables, including: age; sex; urban/rural; deprivation; urgency/referral status; cancer type; procedure type; CCI score; treatment type; and metastatic cancer.

Data from UK Censuses on all variables which could potentially influence timely cancer treatment and cancer mortality were obtained for each individual in the NASCAR cohort. Census population results go through a rigorous quality assurance process [27]. For each individual, information was sought on: ethnic group; country of birth; religion; marital status; general health; disability; long-term illness; deprivation; economic activity; highest qualification; occupation; hours worked per week; mode of travel to work or study; family type; living arrangements; carers in house and hours providing care per week; car and van availability; home circumstances; and housing type.

The Information Services Division (ISD) is a division of National Services Scotland, part of NHS Scotland. ISD provides health information, health intelligence, statistical services and advice that support the NHS in progressing quality improvement in health and care and facilitates robust planning and decision making. The electronic Data Research and Innovation Service (eDRIS) team are part of the Information Services Division and support Administrative Data Research – Scotland (ADR-Scotland) linkage projects.

The original NASCAR cohort was approved by the Privacy Advisory Committee of ISD Scotland (Reference number 0942/14). The cohort was constructed in collaboration between eDRIS and the Data Management Team of the NHS Grampian/University of Aberdeen Data Safe Haven (DaSH). Following secure transfer eDRIS linked CCPd data using the community health index (CHI), a unique identifier for all residents in Scotland [28]. This allows all the records from multiple data sets from primary care, secondary care, and specialist disease registries to be linked [24]. Data extraction and linkage were carried out by eDRIS and DaSH. Data were pseudo-anonymised by the DaSH team (by removal of CHI and application of a unique NASCAR ID) and placed in DaSH, which provides a secure virtual research environment, before release to our research group for analysis.

Subsequently, the NASCAR Census project received approvals from the NHS Grampian Caldicott Guardian (CG/2018/31), the Public Benefit and Privacy Panel (PBPP) of Scottish Government (1718-0012), and the ADR-Scotland (PROJ-166). The DaSH team reapplied CHI numbers to the original NASCAR cohort which was subsequently transferred securely back to eDRIS. In collaboration between eDRIS and ADR-Scotland, the requested UK Census variables were linked to the original cohort using the CHI number which was then subsequently removed and replaced with a new anonymised NASCAR Census study ID. Data were stored and analysed within the National Safe Haven, a secure virtual environment maintained by Public Health Scotland where the project data were uploaded and accessed by the research team using the dedicated ADR-Scotland secure datalink [29]. The NASCAR Census cohort consisted of the individuals which had been linked to the Census (2011 or 2001).

There were 12339 patients in the original NASCAR cohort of which 401 could not be matched to data in either the 2001 or 2011 National UK Censuses (Figure 1). Participants (n = 135) resident in a communal establishment (e.g. a nursing home) were also excluded since it was not possible to determine for how long individuals had resided or would reside there. The final cohort for analysis comprised 11803 individual patients (Table 1 for patient and pathway characteristics).

The NASCAR Census cohort is described in Table 2 and was 98.1% white with 96.9% born in the UK. Adherence to a religion was stated by 67.8% and 62.4% were married. Good general health was reported by 59.5% with 1653 (14.0%) having two or more comorbidities. Only 56 (0.5%) had four dimensions of deprivation. Current employment was reported by 34.3% and 66.8% were living as a couple. Homeownership was reported by 72.2% and only 21.4% did not have a car.

Across the whole sample non-UK born white subjects were significantly more likely to receive timely treatment than Scottish born white subjects (OR 1.61; (95% CI 1.09–2.39) (Table 3). People working more than 15 hours per week were significantly less likely to receive timely treatment than those working less than 15 hours, the odds being lowest for those working more than 49 hours (OR 0.79; (95% CI 0.67–0.93). Those with heating in their homes were 2.34 times (95% CI 1.84–2.99) more likely to receive timely treatment. Compared to manager/senior officials, those working in skilled trades (OR 1.26; (95% CI 1.05–1.50)) and personal services (1.28; (95% CI 1.04–1.56)) were significantly more likely to begin treatment within the target time of 62 days.

Compared to those working fewer than 15 hours per week, those working more than 49 hours were 0.82 times (95% CI 0.70–0.96) less likely to begin treatment within 31 days of diagnosis (Table 4).

None of the added Census variables were associated with an increased risk of mortality at one year (Table 5).

Table 6 reports the results of univariable and multivariable binary logistic regression models analysing associations between travelling times to GP and cancer treatment centre and timely treatment outcomes. Travelling time to the GP surgery was not strongly associated with timely treatment, except that island-dwellers were significantly more likely to be treated within 31 days of diagnosis compared to those in all mainland categories. Those living more than 60 minutes travelling time from the cancer centre, and on an island, were significantly more likely to be treated within 62 days of GP referral and within 31 days of diagnosis. This association was observed in the unadjusted model. Making adjustments for sociodemographic factors (model 1) did not strengthen the association. In the three models (2–4) which also adjusted for patients’ route to diagnosis, the cancer type and whether it has metastasized at the point of diagnosis, the type of cancer treatment received, and patients’ underlying co-morbidities, the association was strengthened. Substituting area-based measures of SES (model 2) with individual measures (model 3 and 4) of SES did not change the strength of the observed associations.

Table 7 reports the results of univariable and multivariable binary logistic regression models analysing associations between travelling time to GP and cancer treatment centre and mortality outcomes. There were no significant associations between travelling times to the GP surgery, overall all-cause mortality, overall all-cause one-year mortality, or cancer-specific one-year mortality. In the unadjusted models travelling time to the cancer centre was not associated with any of the three mortality outcomes and this was not changed by adjusting for socioeconomic factors (model 1). When patients’ route to diagnosis, the cancer type and whether it has metastasized at the point of diagnosis, the type of cancer treatment received and patients’ underlying co-morbidities were also adjusted for (models 2–4), those living more than 30 minutes travel from a cancer centre were significantly more likely to die overall and within one year from both all and cancer-specific causes. Replacing area-based (model 2) with individual (models 3 and 4) measures of SES did not strengthen the associations. Living on an island was not associated with increase mortality in any of the adjusted models (1–4).

In univariable statistical models, subjects’ travelling time to their cancer treatment centre was not associated with receipt of timely treatment. This remained the case when adjustments were made for both area-based and individual-level measures of SES. When cancer-related factors were added to models, those living on an island or with a more than 30 minute journey to their cancer centre were significantly more likely to receive treatment within Scottish Government targets, and were more likely to receive their diagnosis and treatment on the same day if they live on an island. Substituting individual-level for area-based measures of SES did not meaningfully change the associations. Similarly, mortality outcomes showed no association with travelling times to the cancer treatment centre in univariable models, which remained the case when adjustments were made for area-based and individual-level markers of SES. Adjustments for cancer-related factors revealed a significantly increased mortality risk across all three mortality measures for mainland subjects living more than 60 minutes from their cancer treatment centre. Again, substituting individual-level for area-based measures of SES did not meaningfully change the association. Travel times to individuals’ GP surgery was not associated with outcomes in either univariable or multivariable models.

There was also a notable general trend in all adjusted models, albeit non-significant, suggesting that the best survival outcomes (overall and cancer-specific) occurred for those living within 15 minutes of the cancer centre.

The NASCAR Census cohort/dataset provides comprehensive data from a large cohort of people diagnosed with cancer in Northeast Scotland from 2007 to 2013. Extensive information about characteristics, routes to diagnosis, outcomes, travelling times from key healthcare facilities, and socio-demographics of residential area have been linked with individual socio-demographic information from the UK Censuses of 2001 and 2011. Using individual socio-demographic information, we have been able to make adjustments for the particular characteristics of individuals. The range of socio-demographic information we have been able to add, and to adjust for, provides compelling additional evidence for the fundamental importance of residential geography in an individual patient’s cancer journey.

The NASCAR Census study is one of the first in Scotland to have linked an existing clinical cohort to data from the UK Census. The linkage was remarkably complete with only 401 (3%) cases being lost likely due to incomplete Census returns rather than data errors [30]. Using high quality and validated electronic clinical datasets via established data linkage methods allowed a high degree of completeness and accuracy. Thus, for the benefit of future researchers we have demonstrated, given the appropriate permissions, the feasibility of making these linkages.

This study enables, for the first time, a consideration around concerns about residual confounding from using an area-based measure of deprivation in studies of exploring the impact of geography on processes of cancer care. There have been previous concerns that rural postcodes are sufficiently large, and their residents sufficiently diverse as to introduce major socioeconomic confounding in studies of this type. Our data provide reassurance that, at least in Scotland, individuals’ postcodes provide an acceptable proxy for their individual socioeconomic status.

The population of Northeast Scotland is relatively affluent, and the rural population here may not represent the rural population of the rest of Scotland and the wider UK. The possibility remains that the phenomena demonstrated by NASCAR are produced by conditions of local geography and health service provision and that the situation would be different elsewhere. For example, in future, increasingly sophisticated software may enable variables such as traffic volume, road conditions and local issues of healthcare supply and demand to be included in analyses. Further, we explored time-to-treatment as one of our outcomes but considering the natural course of the cancers studied other parameters such as time to diagnosis and detailed treatment received could be equally important.

A further limitation is the fact that the UK National Census occurs every ten years. It is possible, therefore, that an individual’s personal circumstances, including socioeconomic status could vary during that time in ways which would not be captured by the Census. A further limitation is the limited follow-up period enabled in the NASCAR cohort, which may be insufficient to capture differential effects of geography and SES acting on those surviving cancer for longer.

This study is important, if incremental, in that it strongly suggests that physical location is more important than individual characteristics of people of rural-dwellers and island residents in influencing their cancer journey and outcomes.

Our study suggests that area-based indicators of socio-economic status capture, within cancer outcome models, the individual status of individuals within these areas. A study based in nine counties in the Southwest of England concluded that the Index of Multiple Deprivation had a strong relation to individuals’ health. [31] A more recent study exploring the Index of Multiple Deprivation in England found greater heterogeneity in key indicators of deprivation in rural areas than is suggested by the original indices, suggesting that existing area-based measures may be less suitable for comparing rural and urban populations [32]. However, the Scottish Index of Multiple Deprivation (SIMD) is based upon smaller output areas with a smaller population and there are differences in the content of the constituent domains [33, 34]. A study based on 10,359 participants aged 40–59 in the Scottish Heart Health Study also concluded that area-based measures of deprivation showed similar degrees of association with coronary heart disease CHD as measures based on individuals’ occupation [3] Against the context of increasing regulatory and cost burdens around data-linkage, our study also supports the argument that SIMD is an acceptable proxy for individual social status in Scotland.

With this in mind our data suggest that individual social deprivation is not the main determinant of poorer rural cancer outcomes. At first glance, this would appear to contrast with a report from the National Cancer Registration and Analysis service which explored English cancer incidence and mortality in England from 2004–2006 [35]. The NCRAS analysis found variation in cancer incidence and mortality between rural and urban areas concluding that observed difference resulted from differential distribution of socioeconomic deprivation between country and city. However, a study based upon 18,568 women diagnosed with breast cancer in Queensland, Australia between 1997 and 2006 found that area-level disadvantage was associated with breast cancer mortality independently of individual characteristics [36].

The importance of patients’ location relative to specialist cancer and diagnostic facilities has been highlighted before. A study which explored the association between local populations’ mean travelling times to their cancer treatment centre across the whole of England found that longer average travel times were associated with worse mortality from breast, lung and colorectal cancer after adjustment for age, sex, year and area deprivation, in line with our own finding of a trend to better survival for those living within 15 minutes of the cancer centre [37]. In further work by the same authors increased travelling time to a GP in England was associated with increased risk of an emergency or post-mortem cancer diagnosis [13].

Two national Danish studies have recently highlighted the complex relationship between cancer patients’ location, diagnosis and mortality [38, 39]. Amongst 37,872 Danish cancer patients diagnosed during 2006–16, longer travelling distance was associated with a longer diagnostic interval, early stage at diagnosis for hard to diagnose cancers, and later stage for easier to diagnose cancers. Direct comparison with NASCAR is difficult due to the differing methods employed and intervals measured by the Danish investigators, but the overall patterns of diagnosis appear different. Both sets of studies, however, imply that specific local circumstances of geography and health service organization impact cancer diagnosis in specific local ways. In this context, it is important to consider that some European studies have not demonstrated a rural-urban cancer survival inequality. A 2014 German study compared age-standardised five-year survival using 11 population-based cancer registries and found similar survival for urban and rural cancer patients [40]. A recent data-linkage study of 3,718 patients with colorectal cancer, diagnosed between 2007 and 2013 in Northern Sweden, one of the most sparsely populated areas in Europe, found no association between travel time to nearest hospital and survival [41]. A Norwegian study also found no evidence of under-treatment or poorer survival for more remote patients amongst 288 men with metastatic prostate cancer in Nordland County [42].

Nevertheless, most studies conducted in the developed world suggest a rural cancer disadvantage which is complex and multifactorial. Further research from different perspectives is needed to unpick the causes [1]. For example, a study published in 2008 by investigators in Dumfries and Galloway did not explore pathway delays or mortality but did report reduced hospital admissions and bed-days for cancer patients with longer travelling times, suggesting less intense treatment or management of complications as a potential mechanism for our NASCAR findings [43]. A more recent US study found similar outcomes for rural and urban cancer patients who were enrolled in oncology trials [44]. Together, these studies suggest that regional and health service organization and provision could be at least as important as the characteristics of individual patients in determining how geography impacts on cancer outcomes.

Analysis of the NASCAR-Census dataset has important implications for the relationship between residential geography and cancer in Northeast Scotland.

First, the rural “paradox” remains whether adjustments are made for area-based or individual-level markers of socioeconomic status. Those living most remote from a cancer centre are more likely to receive their treatment within Scottish Government targets but, despite this, have a greater risk of mortality than those living closer by.

Second, in the NASCAR Census study the models have been adjusted for area-based and individual level markers of socioeconomic status, both together and separately, with neither appearing to be strongly associated with the outcomes under study. This strongly suggests that sociodemographic characteristics are not the cause of geographical cancer outcome inequality in Northeast Scotland.

Travelling times were not associated with outcomes in univariate models or when adjustments were made for SES. It was only when cancer-related factors were added to the models that the paradoxical associations between travelling times and time to begin cancer treatment and mortality were revealed. Together, this strongly suggests that the mortality disadvantage in the most remote patients arises after treatment has begun.

Aside from geographical considerations our results suggest that other hitherto unrecognized inequalities could be imposed by the way in which cancer services are organized. For example, it was striking that people working fewer hours were significantly more likely to receive timely treatment. This has considerable implications for how appointment systems currently operate and perhaps suggest that greater availability of weekend or evening appointments could redress this balance. It is probably also worth pointing out that the greatest barriers to timely treatment appeared to occur for those in senior management positions, and those working more than 49 hours per week. This finding perhaps highlights how people in highly pressurized senior positions may focus less on their own well-being.

In conclusion, physical geography and its impact on cancer care appears to drive a disadvantage for all rural cancer patients irrespective of their individual socio-economic circumstances. Further qualitative and quantitative research is required to compare and contrast what happens to people after they have been diagnosed with cancer and how where they live influences treatment choices and care received after the point of diagnosis. It makes sense that, where possible, such research in based upon whole-nation samples and cohorts to allow for the potential of regional variation in the impact of geography on cancer service provision and resultant cancer outcomes. The authors plan to begin a whole-Scotland analysis exploring these issues in the upcoming SCOTSCAR study. In this way the root causes of geographical cancer inequality can be determined, and the most promising interventions elucidated.