2.1. Literature search strategy

Searches were conducted in PubMed, Web of Science and Scopus using the following Medical Subject Headings (MESH) terms in the title and the abstract. The search was limited to English language papers that had been published through to April 15, 2019. The following keywords were used for this search:

• ("breast cancer" or "breast neoplasm" or "breast neoplasms" or "breast carcinoma" or "breast tumor" or "breast tumors" or "cancer of the breast")
• AND ("accessibility" or "geographic access" or "spatial access" or "residence characteristics" or "residence characteristic" or "neighborhood characteristic" or "neighborhood characteristics")
• AND ("SES" or "low-income" or "low income" or "low SES" or "low socioeconomic" or "socioeconomic status" or "low socioeconomic status" or "poor" or "poverty" or "disparity" or "disparities" or "deprived" or "disadvantaged" or "low resources" or "poverty area" or "deprivation" or "social class" or "socioeconomic factors" or "insecurity" or "precariousness").

2.2. Inclusion criteria

After excluding duplicate papers, 215 papers were identified by the searches in the three databases. The titles and summaries of these papers were all examined by three reviewers (B.C., A.B. and H.C.). Fig 1 presents the flowchart of the systematic literature search based on PRISMA statement guidelines [14]. The protocol for this literature search was registered in the Prospero database, registration number CRD42020193325 (this can be found at https://www.crd.york.ac.uk/prospero/display_record.php?RecordID=193325).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Flowchart of selection process.

https://doi.org/10.1371/journal.pone.0271319.g001

The selected papers had to meet three major criteria:

1. include at least one of the following measures of breast cancer outcomes: mammography use (Yes/No), stage at diagnosis, type of treatment received such as mastectomy or breast conserving surgery (BCS), and, last, breast cancer survival or mortality;
2. include socio-economic characteristics at the individual or contextual levels;
3. include one or more measures of geographic access to healthcare facilities (distance, travel time and geographic capacity).

Based on these criteria, 55 papers were selected.

In the next stage, 30 papers were excluded. We have excluded six papers in which the SES characteristics were demographic such as age or ethnicity [15–19] or used as covariates [20]. Studies were also excluded if they provided only descriptive analyses of level of access to healthcare facilities or neighborhood SES characteristics (10 papers). We excluded one paper that focused on the likelihood of not using the closest facilities [21]. Furthermore, we excluded 13 papers that used a proxy measure of geographic access such as density (for example as a measure of healthcare availability), car ownership or urban contexts (urban or rural).

After this elimination process, 25 papers were included in the review. Any papers for which inclusion was open to question were discussed by all the authors until consensus was reached.

2.3. Data extraction

For the 25 selected papers, several items of data were extracted by the primary reviewer (B.C.) and presented in an Excel spreadsheet. These were: authors, year of publication, geographical area (country, state or city), breast cancer-related outcomes, geographic access measures and relationships, SES characteristics and relationships.

2.4. Quality assessment of included studies

For quality assessment, we adapted the Effective Public Health Practice Project (EPHPP) quality assessment tool [22]. This tool is widely used to evaluate any quantitative study design. We have kept five in the eight key domains for assessment of study quality (study design, selection bias, confounders, data collection and data analysis) according to the study design of studies included. An overall rating for each study was determined based on the component ratings, ranging from 1 (low risk-of bias; high methodological quality) to 3 (high risk-of-bias; low methodological quality). Strong was attributed to those with no weak ratings and at least two strong ratings, moderate was given to those with one weak rating or fewer than two strong ratings and weak was attributed to those with two or more weak ratings. The methodological quality assessment of each of the included studies was independently assessed by three authors (BC, AB and HC). The ratings for each of the five domains, as well as the total rating, were compared between the three authors. Consensus was reached on a final rating for each included article.

3.1. What measures of breast cancer outcomes, geographic access, and SES characteristics?

The most explored outcome was the stage of cancer at diagnosis (n = 15), followed by the probability of the women receiving different types of treatment such as mastectomy, breast conserving surgery and/or radiotherapy (n = 6). Screening mammography was assessed in three papers and breast cancer mortality in two (Table 1). One paper explored two breast cancer-related outcomes: cancer stage at diagnosis and survival [26].

Using Geographical Information Systems (GIS), geographic access between the residential address of the women and the closest healthcare facility was evaluated by travel time (12 papers) or/and by travel distance (10 papers) or/and by the two-step floating catchment area method (2SFCA) (6 papers). This method is based on the results of spatial capacity modelling including population demand and healthcare provision. Three studies combined two measures such as travel time and capacity [35, 36] or travel distance and travel time [37]. Geographic measures were assessed by Euclidean distance (n = 2) or/and by car (n = 23). In the large majority of the studies, the travel time or distance was estimated between the closest healthcare facility and the women’s residential addresses (n = 11) or the centroid of their residential neighborhood (n = 18). Only two papers calculated the distance between the healthcare facilities used by the women and their residential addresses [34, 38]. In addition, the cut-off used to categorize travel distances and travel times varied (Table 2). Celaya et al. [37] proposed three classes of travel time in which the least accessible class was ">15min"; whereas in Sauerzapf et al. [27] proposed a three-classes in which the most accessible class was "<30min" and the least accessible class was ">60min".

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Cut-off of travel distance and travel time of included articles.

https://doi.org/10.1371/journal.pone.0271319.t002

Socioeconomic characteristics were assessed at individual and residential levels: fifteen studies combine data from both levels, five papers present only individual level data and five papers use data at residential level only. At individual level, the most common characteristics used were age (n = 20), ethnicity (n = 13), partner status (n = 8), health insurance (n = 7) and education level (n = 2). At residential level, the main data used were residential disadvantage or deprivation (n = 10): six papers adopted existing deprivation indicators (e.g., Index of relative socioeconomic Advantage and Disadvantage [IRSAD], index of multiple deprivation [IMD]), and four papers created their own deprivation index [29, 31, 32, 39]. Other variables used were poverty (n = 5), educational level (n = 4) and per capita income (n = 2).

3.2. What are the relationships between geographic access to health-care facilities and breast cancer outcomes?

The twenty-nine relationships between geographic access and at least one breast cancer outcome (mammography use, stage at diagnosis, treatment and mortality) explored in the 25 papers in our review are presented in Table 3. In overall, the statistical significance of the relationships was heterogeneous with 15 significant relationships and 14 as non-significant.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Relations between breast cancer outcomes and geographic access to health-care facilities.

https://doi.org/10.1371/journal.pone.0271319.t003

Women with high level of geographic access to healthcare facilities had a statistically significant higher cancer screening rate in one study (to 4), an earlier stage of cancer at diagnosis (n = 8/17), and fewer mastectomies (n = 4/6) than women with lower level of geographic access. Two studies with survival rates as outcomes observed that women with higher access to healthcare facilities have poorer survival rates than women with lower access to healthcare [26, 28]. The authors of the UK study put forward that this result may be “an artefact of imperfect control of the effects of deprivation, since inner city populations tend to be more deprived and closer to hospitals than suburban or rural populations” [26]. In the state of Parana in Brazil, the authors suggested that the municipalities close to the services of specialized treatment in oncology were also areas with high population concentration which makes access to treatment difficult [28].

As reported in Table 3, relationships varied considerably depending on the breast cancer outcomes: geographic access seems to more frequently influence the type of treatment (4/6) than whether women undergo screening (1/4). It therefore seems necessary to explore whether the type of measure of geographic access (travel time, travel distance or capacity in Table 3) influences the results of the relationships and is responsible for the differences observed between the included studies.

As presented in Table 3, several measures of geographic access were used, which raises the question of their impact on the relationships (significance and direction) with breast cancer outcomes. When geographic access measures were based on travel times, the results varied: 7 studies found negative relationships and 4 found no relationship [27, 35–37, 40]. When geographic access was measured by travel distance, a non-significant relationship was observed in 6 studies [30, 34, 37, 41–43] and a significant one in 4 studies [29, 38, 44, 45]. Finally, when modelling was applied to combine demand and travel time to healthcare (the 2SFCA method), 4 papers observed a significant relationship with breast cancer outcomes [28, 31, 32, 39] and 3 observed a non-significant relationship [35, 36, 46].

Regardless of the type of measures used to calculate geographic access to health facilities (distance, time or geographical capacity), the heterogeneity of the results is very similar and does not allow to define which proxy is the most appropriate to assess geographic access. This finding is confirmed by the fact that studies that compared two different measures of geographic access obtained similar results [35–37]. For example, the paper by Celaya et al. [37] compared travel distance and travel time by car between patients’ addresses and a mammography service in the state of New Hampshire, USA. For these two measures, the authors showed an absence of a relationship between geographic access and stage at diagnosis.

Another interesting finding is that the heterogeneity of the results regarding a relationship between geographic access and breast cancer outcomes cannot be explained by the geographical context and, in particular, the country of study: for example, among 16 studies conducted in the USA, 8 found significant relationships and the other 8 found non-significant relationships. In addition, neither the nature of the area studied (e.g. metropolitan, urban area, rural area) nor the geographic level (e.g. federal state, city, hospital) appear to affect the ability of the studies to explain differences in breast cancer outcomes.

3.3. What are the relationships between SES characteristics and breast cancer outcomes?

SES characteristics were explored at the individual level in 5 papers, at the residential level in 5 papers, and at both the individual and residential levels in 15 papers. At the individual level, SES characteristics were mostly assessed by partner status and health insurance status. Based on Table 1, age and tumor features could be assessed as major confounding variables in the relationships between SES, geographic access and breast cancer outcomes. At the residential level, SES was defined by the poverty rate or the income level (8 papers) or by composite scores such as the deprivation index (10 papers). The composite scores were based on different variables, statistical methods and geographical scales. For instance, Lian et al. [32] used 9 variables in a multivariate approach in order to define a deprivation index, whereas St-Jacques et al. [29] defined two indices (material deprivation and social deprivation), using a factor analysis based on 3 variables in each case.

At the individual level, as shown on the forest plot (Fig 2), the relationships between breast cancer outcomes and age (a), ethnic status (b) and socioeconomic status were mixed (c and d). Considering the studies as a whole, the relationships with age did not follow the same pattern. Greater age was related with: (i) more mammography use (except for 1 study [36]); (ii) lower odds of late stage at diagnosis (except for 3 studies [26, 30, 38]); (iii) receiving Breast-Conserving Surgery (BCS) (except for 1 study [34]). The relationships with ethnic origins were also not systematic and were based on many different definitions of ethnic origins which made it difficult to compare the studies. The majority of the relationships between marital status and breast cancer outcomes were not significant [36, 38, 44]. In two studies, married women or with partner had lower risks of late-stage diagnosis of breast cancer than other women [25, 37]. In two papers, married women or with partner had higher odds of receiving BCS than single women [23, 44]. According to the authors of these studies, these findings reflect the underlying issue of social support networks and social incentives which may affect women’s motivation or ability to screen and/or to receive BCS. In the six papers, the majority of the relationships between marital status and breast cancer outcomes were not significant (6 relationships). Women with no health insurance had less mammography screening and more advanced cancer stage at diagnosis than women with health insurance. The relationship between health insurance and receiving Breast-Conserving Surgery (BCS) seems to be less significant.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Forest plot showing the relationship between SES characteristics (at the individual and contextual levels) and breast cancer outcomes.

a) Age, b) Ethnicity, c) Insurance status, d) Marital status, e) Poverty rate at residential level, f) Deprivation index at residential level.

https://doi.org/10.1371/journal.pone.0271319.g002

At the residential level, as reported in six papers (Fig 2E), women residing in residential environments characterized by high levels of poverty were more prone to late-stage diagnosis than the others (except for two studies [38, 41]). The relationship with the type of treatment received was not significant except in one relationships [47], and in this case only for women residing in an area with a very high level of poverty (> = 15%). In contrast, the relationship between deprivation index (Fig 2F) at residential level and type of treatment was identified as significant in 3 studies: women who resided in the most disadvantaged areas seemed to undergo less BCS [23] and more mastectomies [24, 27] than the others. The deprivation index exhibited no consistent relationship with stage at diagnosis. Only two studies investigated the relationships with the use of mammography or survival rates: living in an area with a high level of material or social deprivation was related with lower mammography use [29], later stage presentation and higher mortality risk [26].

3.4. What are the combined effects of geographic access and SES characteristics on breast cancer outcomes?

Of the 25 papers included in the review, the large majority (n = 24) tested the independent effect of geographic access. In these studies, SES characteristics were used as predictors and/or covariates. Only one study explored the combined effect of geographic access to breast cancer facilities and SES characteristics on the probability of different breast cancer outcomes. Lian et al. [32] developed stratified models of the effects of geographic access to mammography services and neighborhood socio-economic deprivation on late-stage breast cancer diagnosis. The models show that lower geographic access to mammography services was related with greater odds of late-stage breast cancer diagnosis in less deprived neighborhoods, but not in more deprived neighborhoods.

Taking account of the fact that people do not necessarily use the closest facility, and that they do not necessarily start from home

The reviewed studies assumed that women had access to and used the closest facility to their home and that the starting point was always their home address or the polygon centroid of the residential area (when the home address was unavailable). The calculated travel time or distance was therefore the minimum possible time or distance. This may differ from the actual travel time or distance based when women do not use the health facilities that are nearest to their home. Alford-Teaster et al. [48] have shown that only 35% of women participating in the US-based Breast Cancer Surveillance Consortium in the years 2005–2012 (n = 646,553 women) used their closest mammography facility. In this sample, nearly three-quarters of women not using their closest facility used a facility within 5 minutes of it. A previous study has compared self-reported and calculated measures (by women and GIS respectively) of travel time to the maternity unit for childbirth [49]. The reported travel times were similar to the calculated travel times in peri-urban and rural areas, but agreement between the two was poor in urban areas. To overcome this limitation of theoretical accessibility, future studies will ensure that women’s actual care pathways are taken into account including information about their reported travel time and their reasons for choosing (or not choosing) certain types of healthcare facilities. As stated by Khan-Gates et al. [13], we need to further explore the “the actual geographic patterns of seeking care rather than access to the nearest facilities”. To this purpose, information on where women come from when they go to hospital and why they choose (or not choose) their hospital should be included both in questionnaire and interview. Another methodological limitation observed in the reviewed studies concerns the absence of public transport for the calculation of geographic access based only on the travel distance or time by car.

Taking account of the use of other modes of transportation

As reported by Celaya et al. [37], the car is the main mode of transport in the context of assessments of geographic access in breast cancer studies. Measures based on public transportation are rarely used (n = 3/25), and only in terms of supply density [26, 27, 30]: in these three papers, it is the availability of public transport (buses or trains) in the residential area that is measured (line density or the presence of a stop nearby). None of this research measures geographic access, either in time or distance. The lack of analyses based on travel by public transport in breast cancer issues tends to mask the use of other modes of transportation. We may assume that women who are on low incomes and/or who live in deprived areas are more limited in their choice of travel mode, in particular ownership of a private car. For instance, in London, customers who live in the most deprived areas are less likely to use a private car to travel to convenience stores [50]. In contrast, in inner city areas, where the public transport network is dense and effective, public transport can even be faster than the private car. Thus, the availability of effective and convenient public transport (metro, tram, suburban rail, and bus) may be deemed to be a leading driver of women’s mobility. Thus, advances in GIS methods and the availability of transport data sets will allow studies to make a more accurate assessment of the geographic access to healthcare facilities by public transport [51].

In addition to the efforts that must be made to improve geographic access measures (not only the closest facility and include public transportation), the quality of the findings also depends on the intersection of geographic access with socio-economic variables.

Taking account of interactions between geographic access and deprivation at the individual level

Unfortunately, the combined effects of geographic access to breast cancer facilities and SES characteristics on the probability of different breast cancer outcomes have been less frequently explored. For instance, using stratification analyses, Lian et al. [32] showed that the significance of the relationship between geographic access to mammography services and stage at diagnosis varied according to the level of deprivation: women who are more deprived and who live in more accessible areas have less access to mammography screening than non-deprived women who have poorer geographic access. In this way, stratification analysis has been used to divide the study population into several strata according to characteristics that may influence health outcomes. This would help answer the questions we posed, but which we were unable to answer, at the beginning of this systematic review: in a context of equal geographic access to health facilities, do socially and economically disadvantaged women have worse breast cancer outcomes than more advantaged women? With equivalent socio-economic status, do women with poor geographic access to healthcare facilities have worse breast cancer outcomes than women with good geographic access? In addition, the measurement of geographic access should also be considered in relation to the local context in which women live. For example, women who live in suburban or rural areas are more willing to travel longer distances than women living in urban centers [29]. To this purpose, information on urban density level of residential area of women should be collected either from self-reported questionnaires or from spatial databases (e.g., urban, suburban, rural areas).

Taking account of the variability of urban forms and local contexts

The great variability in the results may also be partly explained by differences between study areas in terms of social organization and urban forms, which result from a complex system of interactions between social, political, economic and cultural dimensions [52]. Although increasing evidence suggests that urban form affects public health [53], the urban and social morphologies of cities are rarely used to explain inequalities in healthcare access. For instance, certain suburbs of cities may exhibit a lower density of health services and transportation provision as well as a higher level of deprivation, while the opposite may apply in others [28]. Inner city areas may either be characterized by distressed housing, abandoned buildings and vacant lots or, in contrast, the highest housing prices in the city: the geographic distribution of the population and services are highly variable. As geographic and social access to healthcare is highly embedded in local contexts, it will always be difficult to draw general conclusions from the evidence. Healthcare access issues need further study in the case of urban environments in which differing public health and planning policy responses are required to meet the varied challenges [54].

Taking account of changes in the geographic access score and the deprivation level over time: The need for longitudinal analysis

Twenty of our selected papers were cross-sectional and provided a snapshot at a given time (in general over a four-year period) of the relationship between geographic access, SES and breast cancer outcomes. Only five papers considered a longer period of over 10 years. One of the papers [42], is original in that it covers a very long recruitment period of 17 years—women diagnosed during the period 1985–2002. Three of the other papers took the year of diagnosis as an explanatory variable for differences in cancer outcomes [23–25]. These three papers arrived at the same conclusion: the difference between the likelihood of having better treatment [23, 24] or less advanced cancer at diagnosis [25] between women with good access to healthcare facilities and those with poor access, has decreased over time. It would appear that, over time, the level of geographic access (measured in these three papers by the travel time) has become increasingly less significant: at the beginning of the study period, breast cancer outcomes were very different for women with poor geographic access and the others, while at the end of the period the difference between the two groups was smaller. A contrasting view is presented in the paper by McLafferty et al. [39]. This is the only study that compares two cohorts of women ten years apart (1988–1992 vs. 1998–2002). The results also show that a change has occurred over time: the impact of geographic access was statistically significant in the recent period but less so in the early 1990s. How can we explain this finding? Have inequalities in geographic access increased? Have screening techniques improved? There are many hypotheses, and a longitudinal analysis that provides a comparison at the individual level of geographic access and deprivation over time would provide a better understanding of the changes that are occurring.

Our systematic literature review has a number of limitations. First, using a quality assessment tool introduced some challenges. There is no consensus as to whether one should judge the representativeness of these characteristics of the study and the quality of the reviewed studies is based on what the authors reported in the paper. The quality assessment may not reflect a low quality of the study but might merely have been a lack of reported detail in the paper. Second, the heterogeneity of sample size, characteristics of the sample and measurement tools (both access and SES measures) limited the inter-study comparisons. Third, as in any systematic review, it is possible that some eligible studies may have been missed in our search strategy.