Stage at breast cancer diagnosis and distance from diagnostic hospital in a periurban setting: A South African public hospital case series of over 1,000 women


  • Caroline Dickens,

    1. Section of Environment and Radiation, International Agency for Research on Cancer, Lyon, France
    2. Department of Internal Medicine, Faculty of Health Sciences, University of Witwatersrand, Johannesburg, South Africa
    Search for more papers by this author
  • Maureen Joffe,

    1. Wits Health Consortium, MRC/Wits Developmental Pathways to Health Research Unit, Faculty of Health Sciences, University of Witwatersrand, Johannesburg, South Africa
    Search for more papers by this author
  • Judith Jacobson,

    1. Department of Epidemiology, Mailman School of Public Health, Columbia University, New York, NY, US
    Search for more papers by this author
  • Francois Venter,

    1. Information Management Directorate, Gauteng Provincial Department of Health, Bank of Lisbon, Johannesburg, South Africa
    Search for more papers by this author
  • Joachim Schüz,

    1. Section of Environment and Radiation, International Agency for Research on Cancer, Lyon, France
    Search for more papers by this author
  • Herbert Cubasch,

    1. Department of Surgery, Faculty of Health Sciences, University of the Witwatersrand, Johannesburg, South Africa
    2. Chris Hani Baragwanath Academic Hospital Breast Clinic, Soweto, South Africa
    Search for more papers by this author
  • Valerie McCormack

    Corresponding author
    1. Section of Environment and Radiation, International Agency for Research on Cancer, Lyon, France
    • Correspondence to: Valerie McCormack, Section of Environment and Radiation, International Agency for Research on Cancer, 150 cours Albert Thomas, Lyon 69008, France, Tel.: +33-4-72–73-85-66, Fax: +33-4-72–73-83-20, E-mail:

    Search for more papers by this author

  • Conflicts of interest: Nothing to report


Advanced stage at diagnosis contributes to low breast cancer survival rates in sub-Saharan Africa. Living far from health services is known to delay presentation, but the effect of residential distance to hospital, the radius at which this effect sets in and the women most affected have not been quantified. In a periurban South African setting, we examined the effect of a geographic information system (GIS)-measured straight-line distance, from a patient's residence to diagnostic hospital, on stage at diagnosis in 1,071 public-sector breast cancer patients diagnosed during 2006–2012. Generalized linear models were used to estimate risk ratios for late stage (stage III/IV vs. stage I/II) associated with distance, adjusting for year of diagnosis, age, race and socioeconomic indicators. Mean age of patients was 55 years, 90% were black African and diagnoses were at stages I (5%), II (41%), III (46%) and IV (8%). Sixty-two percent of patients with distances >20 km (n = 338) had a late stage at diagnosis compared to 50% with distances <20 km (n = 713, p = 0.02). Risk of late stage at diagnosis was 1.25-fold higher (95% CI: 1.09, 1.42) per 30 km. Effects were pronounced in an underrepresented group of patients over age 70. This positive stage–distance association held to 40 km, and plateaued or slightly reversed in patients (9%) living beyond this distance. Studies of woman and the societal and healthcare-level influences on these delays and on the late stage at diagnosis distribution are needed to inform interventions to improve diagnostic stage and breast cancer survival in this and similar settings.

In sub-Saharan Africa (SSA), breast cancer is the most commonly diagnosed cancer in women (2012 estimates) but its survival rates are poor, despite this cancer having a good prognosis in other settings.[1] Within South Africa's public health sector, first-world cancer treatment is available at little cost to breast cancer patients, thus diagnosis of this cancer at earlier stages could prevent deaths. Drivers of advanced stage at diagnosis and of low breast cancer survival in SSA include poor existing healthcare infrastructure, lack of early-detection programs and unavailability, inaccessibility and lack of adherence to treatment.[2, 3] Studies in the West (e.g., in the United States and United Kingdom) and in Africa (South Africa, Egypt and Sudan) have found that urban versus rural residence, reduced travel burdens and geographical proximity to primary healthcare providers are associated with earlier diagnostic stage.[4-8] In SSA, distances to treatment hospitals may be up to hundreds of kilometers and personal transport is uncommon; therefore, the effect of distance-to-hospital needs to be precisely quantified as it is likely to play a major role.

In post-apartheid South Africa, the public-sector health service, which serves ∼75–80% of all patients, has a strong community level healthcare approach, complimented by a hierarchical referral system through district hospitals.[9] In this sector, a breast cancer patient typically first notices disease symptoms herself and proactively seeks help in primary healthcare clinics. She is then referred to a secondary hospital and subsequently to a tertiary hospital for diagnostic work-up and treatment. Public-sector patients usually travel by “public” transport (privately run fare-charging minibuses) or by walking.[10, 11] Transport costs represent more than 10% of monthly household expenditure for the poorest quintile of the population.[10] Transport or other barriers encountered before diagnosis may result in delays in diagnosis, especially for patients who reside far from the treating hospital, but they have not been quantified in this health system in relation to breast cancer.

In this study, we aimed to precisely quantify the association of GIS-measured distance from home to hospital with breast cancer clinical characteristics at diagnosis in South Africa, in particular with stage at diagnosis. In over 1,000 breast cancer patients diagnosed at South Africa's largest public referral hospital, the Chris Hani Baragwanath Academic Hospital (CHBAH), we investigated whether the effects of distance to hospital were present within a small radius (<60 km).

Material and Methods


CHBAH, situated in Soweto, is a tertiary public hospital with a catchment population of ∼3 million people residing in Soweto and the surrounding southern areas of Gauteng Province up to 60 km away, including Katlehong, Orange Farm and Sebokeng. Situated adjacent to the financial capital of South Africa in Johannesburg, parts of these areas have transformed significantly during the political and economic changes of the past two decades. Their populations today are heterogeneous; Soweto and Katlehong contain both poor and affluent areas, informal and formal housing.[12] Many residents, however, still live in substandard conditions and unemployment is high, e.g., formal employment in adults is 41–43% in Soweto and Katlehong.[12] Further south of Soweto and of CHBAH, areas such as Orange Farm, Evaton and Sebokeng are, on average, of lower socioeconomic status. Formal employment rates are 36%[12] and entrepreneurial informal employment is more common. Primary health clinics are located in all these communities and refer patients to CHBAH.

Study population

CHBAH's Breast Clinic, operational since 2000, receives referred and walk-in patients.13 Most breast cancer patients are symptomatic at diagnosis as there is no population-wide breast cancer screening program. Diagnostic workup, by clinical examination, imaging, cytology and histology, follows the same protocol for all patients and a small fee of R40 (≈$5) is waived if patients have no means to pay. From the date of first presentation at the Breast Clinic, this clinic aims to achieve a confirmatory diagnosis within 1–3 weeks and treatment commencement within one further week. Therapeutic options are chemotherapy (primary or neoadjuvant, in Johannesburg city center, not at CHBAH), surgery, radiotherapy and hormonal therapy. Trastuzumab is not currently available in the public sector.[14]

In 2006, a clinical data system was established at the CHBAH Breast Clinic to systematically record core clinical and histological data for patients newly diagnosed with breast cancer. Between October 2006 and January 2012, data on 1,104 consecutive female patients newly diagnosed with histologically confirmed, invasive breast cancer were entered into the database. An average of 11 breast cancer patients were diagnosed per month in 2006-2007, which rose to 20 per month in 2010-2011. Data extracted for the present investigation include age at diagnosis, residential address, race [black African, white, Indian or Asian, colored (i.e., mixed ancestry)], point of last referral before arriving at CHBAH, AJCC TNM stage[15] included as stages I, II, III and IV [n = 1,051 (98% complete)] and Scarff-Bloom Richardson grade (1 = well, 2 = moderate or 3 = poorly differentiated). As per routine clinical practice, immunohistochemistry was performed for estrogen [ER, n = 944 (88% known)], progesterone [PR, n = 941 (88% known)] and HER2 [n = 903 (84% complete)] receptors on prechemotherapy core biopsy specimens (further details provided in full elsewhere).[16] Cutoffs of >1% (score 1, 2 or 3) were considered ER-positive and PR-positive and for HER2 status, scores of 0, 1 and 2 were considered HER2-negative and score 3 as HER2-positive. A HIV test (ELISA method) was offered to all women, and is considered here as a social indicator and not a breast cancer risk factor, as we have previously shown that the proportion of HIV-positive breast cancer patients in this population is equal to that of the age-matched underlying catchment population.[13] The study was approved by the University of the Witwatersrand Human Research Ethics Committee (M130369, M130118).

GIS-based straight-line distance

The geo-coordinates of each patient's reported residential address (physical street name and house number) were obtained through a manual geo-coding process using Google street map services through the web interface. Using these coordinates, the shortest straight-line distance to CHBAH from the woman's residence was calculated using either GIS software (ESRI ARCGIS 9.3 with a Transverse Mercator projection system, central meridian of 29, Hartebeesthoek_1994 as Datum and WGS_1984 as Spheroid) or the Haversine formula below [where the patient's residential address is (latitude inline image, longitude inline image) and CHBAH is at ( inline image= 26.26144S, inline image= 27.94051E), coordinates were converted to radians and 6,371 km used as the average radius of the earth]; hereafter referred to as “distance.” inline image. The current analysis is based on the 1,071 (97%) breast cancer patients diagnosed during October 2006 to January 2012 for women both residential address and age at diagnosis were known.

Availability of correlates of distance

In addition to individual-level data, we incorporated ward-level socioeconomic indicators for women residing within the three most common residential areas: Soweto (patients originated from 44 wards), Orange Farm/Sebokeng (30 wards) and Katlehong municipalities (20 wards). In these areas, among the most densely populated of South Africa, the median number of woman over age 20 in a ward was 9,534 (range 2,929–20,615), with a median of 7,851 households per ward. Each woman was assigned to the ward whose centroid was nearest to her residential address. Ward-level socioeconomic indicators (listed in Table 2) were obtained from the 2011 South African census.[12] Finally, to gauge travel burdens to CHBAH in terms of time and cost, we interviewed 73 women attending the Breast Clinic in March 2013, of which 92% used minibuses (16-seater commuter buses) to reach the hospital, 5% private car, 1% hospital transport and 1% walked. Median minibus costs for a one-way trip were R8 for journeys <0.5 hour, R11 for 0.5–1.0 hours and R22 for >1 hour journeys (R20 ≈ $2.20 in 2013); average monthly income for women over the age of 20 ranges from R1250 in the Orange Farm/Sebokeng area to R2250 in Soweto; thus, one return trip transport costs are 1–3% of monthly income.

Statistical analyses

We used generalized linear models to assess associations with stage at diagnosis (primary outcome), treated as a binomial outcome of late (stage III/IV) versus early (I/III) stage, using a log link function to obtain risk ratios [RR, exp(β)] and an identity link function for absolute differences (100*β) in % late-stage disease. 95% confidence intervals (95% CIs) for effect estimates are reported. The explanatory variable of interest, straight-line distance to hospital, was included using indicator categorical variables and as a continuous trend restricted to <50 km, beyond which data were sparse. We first assessed crude associations with distance and thereafter examined associations mutually adjusted for age in 5-year age categories, race (black/non-black), year of diagnosis and referral source and finally we further adjusted for ward-level socioeconomic indicators. The ward-level socioeconomic indicators (% with primary education only, % informal housing and % low household income) were divided into tertiles, with 1 being the least deprived and 3 the most deprived, and included as categorical variables. The same approach was used to investigate associations with the binary outcomes for ER, PR and HER2 status.


A total of 1,071 breast cancer patients were included. Ninety percent were black African women and mean age at diagnosis was 55.4 years (SD 14.3). Overall, 54% of patients presented with late-stage (stages III or IV) disease, 67% of tumors were ER and/or PR positive and 21% triple-negative (Table 1). Residential locations of breast cancer patients in relation to CHBAH are shown in Figure 1. Forty-five percent of patients resided within 10 km of CHBAH and 95% within 50 km. Most patients (61%) resided in the Soweto area [average distance 7.5km (SD 4.0)]. Patients' residences also clustered near Katlehong [23.9 km away (SD 4.0)], Orange Farm and Sebokeng [28.5 km (6.0)] and Vanderbijlpark [47.5 km (2.7)] (Table 1 and Fig. 1). Only 46% of patients identified their last referral point, but of these patients, 49% were referred from a primary healthcare clinic, 22% from a secondary hospital (primarily patients from beyond 20 km to the hospital), 12% directly from a local doctor and 18% were self-referrals, mostly from the proximate area (Table 2).

Table 1. Characteristics of 1,071 breast cancer patients diagnosed at the Chris Hani Baragwanath Academic Hospital during October 2006 to January 2012
 NColumn %
  1. Mean tumor size for all women was 40 mm (n = 863).

  2. a

    % (n) of each characteristic that was missing were 0% for age and distance (by definition of sample set) and for residential area and year of diagnosis, 1% (n = 20) for stage, 12% (n = 128 for ER and PR and 168 for HER2) for receptor status, 22% (n = 231) for tumor size, 54% (n = 578) for referral source and 0.5% (n = 4) for race.

Demographic characteristics[1]  
Age at diagnosis (years)<4015714.7
Year of diagnosis2006–200716715.6
RaceBlack African96490.0
 Indian or Asian181.7
Straight-line distance to hospital (km)<518317.1
Residential areaNearest area: Soweto65461.1
 Farther areas: Orange Farm and Sebokeng12311.5
Referral route to hospitalVia: Primary health clinic24349.3
 Secondary hospitals10721.7
 Local doctor5711.6
Clinicopathologic characteristics  
Receptor statusER+ and/or PR+62866.6
 ER− and PR−31533.4
HIV status for 67% of womenHIV positive13218.4
 HIV negative58681.6
Table 2. Distance to hospital in relation to woman-level and tumor characteristics of breast cancer patients
 Distance to hospital (straight-line distance, km) 
  1. a

    p-Value from χ2 test for associations with categorical age and race, and linear test for trend for continuous outcomes

  2. Restricted to wards from more densely populated areas namely Soweto, Katlehong and Orange Farm/Sebokeng

  3. b

    Actual age of women used, not ward-level age.

Breast cancer cases (N)183 299 242 188 61 98  
Woman-level characteristics             
% of diagnoses in 20062008 37.2 32.4 31.4 35.1 47.5 33.70.33
Age at diagnosis (years)Mean (SD)57.5 (15.2)57.0 (14.7)54.4 (14.1)52.5 (12.9)53.9 (11.3)55.2 (14.8)0.002
Race% Black African 92.9 93.0 85.1 94.7 95.1 79.6<0.001
Referral source, via:Primary health clinic5566.38664 .26252.12325.6731.81123.9<0.001
 Secondary hospital22.453.7119.24752.21254.63065.2 
 Local doctor1012.11712.71815.188.9313.612.2 
Ward-level characteristics, mean[2] 
% Primary education or less14.6 14.2 14.0 18.7 20.9  <0.001
% Annual household income ≤R960029.4 27.4 24.7 32.9 27.0  <0.001
% Individual monthly income ≤R80050.3 51.2 49.3 61.0 58.0  <0.001
% Informal residential housing1.7 2.1 2.4 3.0 14.0  <0.001
Mean age of women in contributing wards over 3046.3 46.4 45.5 45.0 48.2  NAb
Tumor characteristics             
Stage (98% complete)Stage I105.5175.8156.321.123.477.10.02
 Stage II8245.113445.49740.96736.41627.63636.7 
 Stage III7842.911639.310745.19652.23662.14849.0 
 Stage IV126.6289.5187.61910.346.977.1 
 % Late stage (III/IV) 49.5 48.8 52.7 62.5 69.0 56.1 
Receptor status% ER+ and/or PR+ 69.1 68.9 65.6 64.1 57.9 64.80.58
 % HER2+ 23.8 28.9 21.7 25.8 33.9 27.70.37
 % Triple negative 18.8 18.8 22.2 26.5 23.6 21.70.52
Figure 1.

Residential locations of breast cancer patients diagnosed at CHBAH breast clinic during 2006–2012, overlaid with circles showing straight-line distances to the breast clinic. [Color figure can be viewed in the online issue, which is available at]

The percentage of patients presenting with advanced stage at diagnosis increased from 50% for distances <20 km to 62% for distances of 20–29 km and 69% for distances of 30–39 km (Table 2). In the 9% of patients who lived beyond 40 km, the percentage of late-stage tumors (56%) did not increase further. Stage–distance associations may have been confounded by age, race or year at diagnosis, as these factors were associated with distance and stage at diagnosis, as explained below.

Patients with over 20 km distance to hospital were younger, by an average 2.7 years (mean age 53.5 years, SD 13.2) than those who lived within 20 km (mean 56.2, SD 14.7). This difference is partially, but not entirely accounted for by an older proximate at-risk population—25% of over 30-year-old women were aged ≥50 in Soweto compared to 20% in more distal areas. After adjusting for differences in population structure, only 9 and 14% of breast cancer patients from distal areas (Katlehong and Orange Farm/Sebokeng, respectively) were over age 70 compared to 21% of patients from the proximate area of Soweto. Correspondingly much fewer (21%) over 70-year-old patients resided more than 20 km away compared to 36% and 34% of patients under age 50 and 50–69 years, respectively (Fig. 2). Distances were also associated with race, but not with year of diagnosis or HIV status (Table 2). Indian/Asian patients were concentrated 10–20 km away, a distance that includes the predominantly Indian area of Lenasia, and a greater proportion of white women resided beyond 40 km. Ward-level socioeconomic indicators also revealed large differences in the at-risk populations. Areas beyond 20 km from CHBAH had a lower socioeconomic status than proximate areas, as measured by a greater proportion of the female population whose highest educational level was primary school, lower household and individual incomes and by a greater proportion of informal housing (Table 2). In this study population, late stage at diagnosis was also more common in older patients [RR 1.03 (95% CI: 0.99, 1.07) per 10 years for late vs. early stage, not in tables] and in diagnoses before 2008 [RR 1.34 (95% CI: 1.17, 1.53) for diagnosis in 2006–2007 vs. 2008–2012], but was not associated with race.

Figure 2.

Relative distribution of travel distances within 50 km of the CHBAH Breast Clinic, for breast cancer patients under 50 years (n = 384), 50–69 years (n = 437) and 70 years and older (n = 176).

Risk ratios for late stage at diagnosis associated with distance are provided in Table 3. The crude risk of a late-stage tumor was 39% higher (95% CI: 11, 75) among women who lived 30–39 km from the hospital than for those who lived within 5 km. Across distances of up to 50 km, a statistically significant linear trend of more advanced stage with greater distance was observed (RR 1.24 per 30 km distance). These trends were present and as strong after adjustment for both demographic and ward-level socioeconomic indicators (Table 3). When adjusting for last referral point, RRs did not change greatly. Further, there was no clear evidence that the distance–stage association differed by last point of referral (interaction test p = 0.17). The distance–stage association was present in patients referred to CHBAH from clinics [RR 1.71 (1.35, 2.22), 22% of all patients], self-referrals [RR 1.48 (1.06, 2.07), 8% of all patients] or via other hospitals [RR 1.46 (0.74, 2.88), 10% of all patients]. There was weak evidence of effect modification of the distance–stage association by age (p = 0.10). The association was positive in all age groups, but was strongest in older women: RRs (95% CI) per 30 km were 1.26 (1.00, 1.59) under age 50 years, 1.08 (0.88, 1.33) at ages 50–69 years and 1.57 (1.14, 2.17) at ages 70 and over (not in tables).

Table 3. Risk ratios for the association of breast tumor stage at diagnosis with residential distance of breast cancer patients from Chris Hani Baragwanath Academic Hospital
 Risk ratio for late-stage (III/IV) breast cancer
 CrudeAdjusted for age, race, referral source and yearAdjusted for age, race, referral source and socioeconomic indicators[1]
Distance to hospital (km)RR95% CIRR95% CIRR95% CI
  1. Adjustment were carried out using continuous variables for age and year, and categories for race (black vs. non-black) and referral source (clinic, hospital, self, not known, GP).

  2. a

    Ward-level socioeconomic indicators restricted to patients residing in Soweto (n = 634), Orange Farm/Sebokeng (n = 145) and Katlehong (n = 125). 2Restricted to women residing within 50 km of CHBAH.

5–9.90.990.82, 1.190.980.82,, 1.23
10–,,, 1.34
20–, 1.531.311.08, 1.571.331.04, 1.68
30–39.91.391.11, 1.751.401.12, 1.751.330.93, 1.91
40–991.140.89, 1.461.170.92, 1.50 
100+1.060.68, 1.671.120.71, 1.76 
Trend per 30 km21.241.09, 1.411.251.09, 1.421.351.11, 1.65

Figure 3 shows the absolute differences in the probabilities of having a late-stage tumor as a function of distance to diagnosing hospital, predicted for a 50-year-old black African patient, generated from a model adjusting for race, age and referral pattern. Model predictions based on both categorical (points and 95% CI error bars) and continuous (solid line with 95% CI represented by dotted lines) distance using splines are consistent; distal patients have a 15 to 20 percentage point absolute increase in the probability of a late-stage tumor compared to proximate patients. Further than 40 km from the breast clinic, the stage–distance association plateaued somewhat, with some suggestion that it reversed slightly, but with few patients residing at those distances, confidence intervals were very wide.

Figure 3.

Probability (95% confidence interval) of a woman with breast cancer being diagnosed at late stages (stage III or IV) according to her straight-line distance from home to CHBAH using both categorical and continuous variables.

There was no association of distance to hospital with tumor hormone receptor (ER, PR or HER2) status (Table 2). This crude association did not materially alter after adjustment for potential confounders (Supporting Information Table 1), despite black patients having a greater risk of ER-negative tumors compared to non-black women [RR 1.66 (95% CI: 1.12, 2.43), not in tables].


Our study showed that public-sector breast cancer patients diagnosed at CHBAH who live farther from the hospital were diagnosed at later stages than those who lived nearer. The association was detectable beyond 20 km from the hospital and was of a clinically meaningful magnitude. The effect was stronger in older patients, and an underrepresentation of such patients at greater distances may be a continuation of this phenomenon, i.e., older women residing far from CHBAH being more vulnerable to both advanced diagnostic stage and to never reaching the tertiary hospital for diagnosis and treatment.

Distance-stage association

CHBAH is a public-sector tertiary hospital and as such is the third tier in a hierarchical referral structure. Private healthcare in South Africa is available to those who can afford medical insurance, and thus exhibits great socioeconomic and racial disparities, with 7% of black Africans having private medical insurance, compared to ∼70% of whites.[17] More advanced stage at breast cancer diagnosis in distally located patients likely arose from several patient- and health system-level delays. For the latter, primary healthcare facilities are often underutilized, sometimes being perceived as suboptimal by patients.[10] Some of the clinics are short-staffed[9] and nursing staff at these clinics may not be adequately trained in symptom recognition and clinical breast examinations, leading to unnecessary delays in referral. Some patients (one-fifth of proximate patients in our study) choose to bypass the referral system and go directly to the tertiary hospitals.[11] Alternatively, women residing further from the hospital may be less aware of the breast clinic, and less “breast aware” in general, being exposed to fewer health awareness campaigns in the media. The financial burdens associated with travelling to the tertiary hospital are an added barrier.[3] A round trip to the hospital may cost 5–14% of a woman's weekly income for a domestic worker on the minimum wage, in addition to the loss of work hours. Many women, including older women, have responsibilities as family caregivers; they may need to organize child/grandchild minding during their visit the clinic or hospital. Distance-linked diagnostic delays appeared to affect older patients to a greater extent, possibly owing to a concentration of barriers in this group. It is unlikely that competing risks in rural areas would have explained the underrepresentation of older patients, as several dominant causes of premature deaths (HIV/AIDS, tuberculosis and lower respiratory infections[18]) would disproportionately affect lower socioeconomic groups who have lower breast cancer risks. In particular, HIV, a dominant cause of premature mortality in South Africa, is not a risk factor for breast cancer and we have already demonstrated that the HIV prevalence in this breast cancer patient group matches that of the background population; HIV status of breast cancer patients was also not associated with stage.[13] There is a need to specifically address the barriers in this patient population, and to investigate whether some older patients are not being referred through the health system for cancer diagnosis and treatment.

Late stage at diagnosis

The late-stage distribution, even in patients living adjacent to CHBAH who have more immediate access to the hospital, is suggestive of further barriers to presentation. Locally residing patients included some who were diagnosed when palliation was the only care possible. The 50% stage III/IV disease observed in patients living within 20 km is a more advanced stage distribution than that seen in some, mostly non-African, unscreened populations or unscreened age groups in which less than 20% stage III/IV has been observed.[19-21] Nevertheless, worse stage distributions (60–90% stage III/IV) exist in many other African settings.[22-24] Further, we have previously documented how the stage distribution has improved from 70 to 50% stage III/IV in just 5 years.[16] Unfortunately, we do not have accurate measures of the time delays associated with differences in stage at diagnosis between proximal and distally residing patients. Instead we use breast tumor growth models to provide an approximate order of magnitude of the time delays.[19] Assuming mean tumor sizes of 10 mm for T1, 35 mm for T2 and for T3 and for T4 a range from the mean ± 10 mm tumor size obtained from SEER database (2010 breast cancers), the stage distributions seen at CHBAH translate to a potential 2- to 4-month diagnostic delay associated with distal (≥20 km) compared to proximally residing (<20 km) patients. Stage for stage, tumor sizes may have wider ranges in the South African setting compared to SEER populations; thus, this time-delay estimate is approximate.

Further factors contributing to delayed diagnosis may include lack of education concerning where to go to seek help, poor knowledge of symptoms, lack of breast awareness, fear and beliefs held on the causes of cancer and whether it is curable. Low utilization of healthcare services by women with noncommunicable diseases in general has been documented in this same Sowetan setting.[25] In some African settings where traditional healers are consulted, conventional western treatments are thought to interfere with indigenous remedies and cancer is believed to be a death sentence.[3, 26-28] In South Africa, especially within rural communities, cancer is sometimes believed to be caused by witchcraft and traditional healers are consulted to reverse this sorcery before a patient presents for treatment at a hospital.[6] In other disease areas, traditional medicine is used in conjunction with Western medicine in the hope of offsetting the side effects of treatment.[29]

Previous studies in SSA have observed more advanced disease in rural compared to urban patients.[6-8] In Ethiopia rural breast cancer patients were more impoverished than their urban counterparts and, as they reside further from treatment centers, they presented at later stages and were less compliant with treatment.[30] However, few previous studies were able to quantify associations with distance to diagnostic hospital precisely. In other settings, such as the United States, studies in low-income groups found associations between late stage at diagnosis and both distance from screening and diagnostic facilities and lack of health insurance.[31-34] Diagnosis based on symptoms, the principle means of breast cancer detection in low- and middle-income countries, rather than on screening, is also associated with delays in diagnosis.[35-39]

Strengths and weaknesses

Our study benefits from its large sample size and the fact that it was based on a consecutive case series in a public hospital setting. The sample represents women who reach this stage of the health system, but the series is not a complete population-based sample. Notably, as evidenced in the probable underrepresentation of older patients, some patients may never reach the tertiary facility and furthermore private sector cases are not included. In the absence of a population-based cancer registry covering Gauteng, it is difficult to accurately evaluate the selection biases introduced, but it is unlikely that they would give rise to the stage–distance association observed as cancer is not treated in secondary hospitals and CHBAH is the tertiary referral hospital. We used calculated straight-line distance to the hospital rather than actual individual-level travel time and cost. However, straight-line distance has been shown to approximate true travel distance and any errors introduced by using this measurement would have been nondifferential in nature.[40] We used self-reported place of residence which could not be validated. However, any patients temporarily residing within Soweto solely for the purpose of treatment at CHBAH would, if anything, have led to an attenuation of our results because of a greater misclassification of later-stage distal patients. We did not have optimal individual-level data with which to fully investigate mediating pathways but, as a proxy, we used ward-level indicators for adjustment. Furthermore, we only had data on the last referral point, not on the entire referral path. Our correlates of socioeconomic status are strengthened by the fact that we made use of both educational- and income-related variables in our analysis.[41] However, these ecological measures incur measurement error that attenuates associations to a greater extent as area size increases. Previous work in the United Kingdom has demonstrated the validity of using ecological data at the electoral ward-level, although associations were weaker than those observed using the smaller enumeration district level.[41] Average ward size in SA is larger than the UK electoral ward, so the attenuation would be greater in our study. Thus, measurement errors in the distance calculation and in socioeconomic factors are most likely random in nature and as such, the magnitude of the associations presented here would be underestimated.


As part of an increasing noncommunicable disease burden affecting aging and westernizing populations in general, the cancer burden of many middle income countries such as South Africa is expanding, and cancer control efforts need gearing up to manage the burden.[42] Understanding barriers to early presentation, diagnosis and treatment commencement are essential for cancer in particular, as diagnosis and treatment centers are usually located in a few specialized referral hospitals, and early diagnosis combined with prompt treatment can halt disease spread and metastases. Earlier diagnosis is particularly important for breast cancer as this tumor has the potential of excellent prognosis,[39, 43] especially in South Africa where diagnostic and therapeutic facilities are available, which is not the case in all SSA countries. Stage at diagnosis is a strong predictor of breast cancer survival, with 5-year survival rates reduced from 90% for stage I breast cancer to 65% at stage II, 33% at stage III and only 6% at stage IV (Indian data).[44] Equivalent recent statistics are needed for South Africa, but are likely to have a similar trend. To improve the stage-at-diagnosis distribution, research is needed in two areas: first, to identify context-specific barriers that drive later diagnostic stage in general and specifically in distally residing patients, and second, to identify interventions to achieve earlier diagnosis and treatment and ultimately improve breast cancer survival rates. Guidelines for breast cancer management in such settings are provided by the Breast Health Global Initiative.[39, 45] Appropriate early-presentation and diagnosis programs for these contexts need to consider the current state of the breast cancer burden including its relatively young age, low incidence rates, advanced symptomatic disease presentation and context-specific health infrastructure.[45-47] There are several relevant research efforts in this area. In Sudan and India, initial results from both a cluster-randomized trial of triennial clinical breast examination (India) and the training of female volunteers in the detection of breast abnormalities (Sudan) have demonstrated improved stage at diagnosis.[48, 49] Similar interventions at the primary healthcare level in South Africa may also have the desired effects of earlier presentation and faster referrals and thus need to be investigated. We have focused on late diagnosis, but the distance to the tertiary hospitals is likely to affect postdiagnosis treatment access and adherence, further impacting on survival. Cycles of chemotherapy and radiotherapy require repeated hospital visits for which travel-associated barriers may be amplified as patients become weaker.[50] Additionally, later stage at diagnosis, present at a 20 km radius of the hospital, may worsen for rural patients traveling greater distances (often up to hundreds of kilometers) to tertiary treatment centers.


Data collection was supported in part by the Columbia University – South Africa Training Program for Research on AIDS-related Malignancies through the National Cancer Institute, NIH (grant # 1D43CA153715-03). The work reported in this paper is being undertaken during the tenure of a Postdoctoral Fellowship from the International Agency for Research on Cancer, partially supported by the European Commission FP7 Marie Curie Actions – People – Co-funding of regional, national and international programmes (COFUND). The study design, analysis and manuscript writing were undertaken by the International Agency for Research on Cancer.