Delayed presentation and diagnosis of breast cancer in African women: a systematic review

Purpose Africa has low breast cancer incidence rates but high mortality rates from this disease due to poor survival. Delays in presentation and diagnosis are major determinants of breast cancer survival, but these have not been comprehensively investigated in Africa. Methods MEDLINE, Embase, and Global Health were searched to identify studies reporting on delays in presentation and/or diagnosis of breast cancer published between January 1, 2000 and May 31, 2016. Data were synthesized in narrative, tabular, and graphical forms. Meta-analyses were not possible due to between-study differences in the way delays were reported. Results Twenty-one studies were included in the review. Study-specific average times between symptom recognition and presentation to a health care provider ranged from less than 1 to 4 months in North Africa and from less than 3 to greater than 6 months in sub-Saharan Africa. Study-specific average times from presentation to diagnosis were less than 1 month in North Africa but ranged from less than 3 to greater than 6 months in sub-Saharan Africa. Reported reasons for these delays included patient-mediated (e.g., socioeconomic factors) and health system–mediated factors (e.g., referral pathways). Conclusions This systematic review revealed marked delays in presentation and diagnosis of breast cancer in Africa. Identification of their drivers is crucial to the development of appropriate control strategies in the continent.


INTRODUCTION
Women in Africa currently have one of the lowest incidence rates of breast cancer worldwide (1). However, the burden from this cancer is expected to increase markedly in the next decades. A growing aging population alone, i.e. assuming incidence rates will remain constant, will lead to an estimated 119,918 new cases in 2030, a near doubling in the number of incident cases over 20 years (2). The increase will be even more marked as incidence rates are likely to rise due to the adoption by African women of more westernized lifestyle profiles, particularly reproductive patterns characterised by late age at first full-term pregnancy, lower parity, reduced lifetime breastfeeding duration as well as increases in postmenopausal weight (3).
Despite breast cancer incidence rates being still relatively low in Africa, mortality rates from this disease are as high, or higher, than in high incidence countries due to poor survival (1). Furthermore, the proportion of breast cancer cases and deaths at premenopausal ages is higher in Africa than in high-income countries (HICs), where disease incidence is highest, reflecting the younger age structure of the continent's population and possibly also distinctive risk factors and/or tumour characteristics. Consequently, breast cancer in Africa disproportionately affects women in the prime of their lives and hence it has particularly marked familial, societal and economic consequences.
A recent systematic review (4) shows that a high proportion of breast cancer patients in sub-Saharan Africa (SSA) are diagnosed with late-stage disease leading to poor survival (5). Studies from HICs have shown that delays between onset of symptoms and start of treatment are main determinants of late-stage presentation and poor survival (6). Previous studies have attempted to examine delays in breast cancer presentation, diagnosis and treatment in Africa (5,7) but, to our knowledge, these have not been comprehensively investigated across the continent. Knowledge of the length of time intervals between symptom recognition, presentation, diagnosis and start of treatment -and of the factors that may influence them -is key to the development of strategies to shorten them. Therefore, we conducted a systematic review to investigate delays in presentation and diagnosis of breast cancer in Africa, and their determinants. Figure 1 depicts a patient's trajectory from the moment she first notices symptom(s) to the time when treatment starts as well as the factors that may affect her journey. In HICs with free universal access to health care the delay from a woman first noticing potential symptoms of breast cancer to her presentation to a health care provider is labelled as "patient delay" as it is essentially driven by patient-mediated factors. In contrast, the time from first medical consultation to the beginning of definitive treatment is labelled as "provider delay" as it is driven predominantly by M A N U S C R I P T A C C E P T E D ACCEPTED MANUSCRIPT 6 health system-mediated factors. However, in many African settings the picture is likely to be far more complex as delays in both presentation and diagnosis are likely to result from a complex interplay between patient-mediated and health system-mediated factors. For instance, a woman may delay presentation not only because of her lack of breast cancer awareness but also because of the unavailability of health care providers in her area of residence. Similarly, a woman who first presents with a suspicious cancer may delay diagnosis due to fear of its consequences (e.g. mastectomy, death). In this review, we will consider presentation delays as the time interval from symptom recognition to presentation to the first health care provider, diagnostic delays as the time interval between presentation and breast cancer diagnosis, and treatment delays as the time interval between diagnosis and start of cancer treatment.

Conceptual framework
These terms do not carry any judgement on whether these delays are primarily induced by patient-mediated or provider-mediated factors.

Search methodology
The PRISMA statement guidelines (8) were followed to select relevant publications on delays in breast cancer presentation and diagnosis in Africa. Papers were eligible for inclusion in the systematic review if they reported findings from primary research studies conducted in Africa; reported on delays in presentation and/or diagnosis of female breast cancer patients; and were published between the 1 st January 2000 and the 31 st May 2016. No language restrictions were imposed. Relevant publications were searched in the electronic databases MEDLINE, Embase, and Global Health. A search strategy using synonyms (including truncations) and subject headings of the search concepts "breast cancer", "late diagnosis", "Africa" and "determinants", and the Boolean operators "AND" and "OR" was used (Appendix A). All titles and abstracts were screened to identify potentially eligible papers and the full-text for these retrieved and critically reviewed to assess eligibility and, if eligible, to extract relevant data.

Data extraction
The data extraction from each eligible paper was carried out independently by two reviewers (CE and IdSS) using a specifically developed standardised data extraction form. The following information was extracted: the type of catchment population (e.g. country; urban, rural or mixed); the study design (quantitative, qualitative, mixed); the type of recruitment source (primary, secondary or tertiary hospital/clinic) and approach (eligibility criteria; recruitment period; type of sample: consecutive or convenience, i.e. opportunistic; sample size); patient (e.g. age) and tumour characteristics (e.g. stage, size, histology, symptoms); source (e.g. patient, medical records) and timing of collection (e.g. prior or after diagnosis) of data on delays and their reasons; reported times between symptom recognition, M A N U S C R I P T A C C E P T E D ACCEPTED MANUSCRIPT 7 presentation, diagnosis and start of treatment; and patient-mediated and health system-mediated factors that might have influenced them. Disagreements between the two reviewers were discussed and a consensus reached.

Quality assessment of the eligible papers
The quality of the articles included in the review was assessed independently by the same two reviewers. A standardized quality assessment form was developed which included parameters to assess the potential for selection and information bias as well as the appropriateness of the analytical methods used, including those for dealing with potential confounders (Appendix B). The overall quality score of a paper was expressed as the sum of its parameterspecific scores, which could range from 0 (lowest) to 30 (highest). The higher the score, the higher the methodological quality of the paper; the lower the score, the more likely its findings might have been affected by biases.

Data synthesis
Data were synthesised in narrative, tabular and graphical forms. Study-specific mean (SD), or median (range), presentation, diagnosis and treatment delays are presented; if only categorical data were reported in the original publication we used them to estimate the median, or a weighted mean, whenever possible. Studies differ greatly in the way they obtained information on potential reasons for delays and in the way such data were presented (Appendix C).
Most studies simply presented data in a descriptive way (e.g. percentages), but a few used logistic regression methods to estimate crude and/or adjusted odds ratios (OR) for delayed presentation, diagnosis or treatment for each variable examined, with studies using different cut-off points to define such delays (e.g. from ≥2.2 to >6 months for delay in presentation and from >2 weeks to ≥6 months for delays in diagnosis; Appendix C). One study in North Africa (9) reported on delays but only examined factors associated with late (III/IV) versus early stage at diagnosis; late stage was taken here as a proxy for delays between symptom recognition and diagnosis. Findings are shown separately for studies conducted in North Africa (i.e. in Algeria, Egypt, Libya, Morocco, Sudan, Tunisia, and Western Sahara) and sub-Saharan Africa (SSA, i.e. countries in East, Middle, South and West Africa) as defined by the United Nations (10).

RESULTS
A total of 315 papers (after removal of duplicates) were identified through electronic searches and their titles and abstracts screened for potential eligibility ( Figure 2). In all, 35 articles were retrieved for full-text review. Of these, only 21 were eligible for inclusion in the review: 16 quantitative studies, three qualitative studies and two mixed (quantitative and qualitative).  Table 1 summarizes the main characteristics of each participating study. Of the 18 quantitative and mixed design studies, eight (44%) were conducted in North Africa and ten (56%) in SSA, with their sample sizes ranging from 44 to 350. In contrast, all three qualitative studies were conducted in SSA, with sample sizes ranging from 9 to 31. All studies were hospital-based cross-sectional surveys that relied on consecutive samples of patients, except for two small qualitative studies (11,12) which relied on convenience samples. Eligibility was restricted to women with advanced breast cancer in one study in North Africa (13) and in four (three quantitative (14)(15)(16) and one qualitative (11)) in SSA.
The large majority recruited breast cancer patients diagnosed predominantly in the years 2000-2010, but two studies in North Africa (17,18) and two in SSA (19,20) included patients diagnosed after 2010 whereas one study in SSA recruited patients diagnosed prior to 2000 (21) ( Table 1). The average (mean/median) age at breast cancer diagnosis was in the 40s in the large majority of studies. Most studies involved collection of data through structured or semistructured questionnaires, usually administered by the researchers or medical staff around the time of diagnosis, but four were conducted retrospectively using medical records (14,15,17,22). Information on ethnicity was provided in only one study, which stated that its subjects were all Black (12). Information on tumour stage at diagnosis was available for seven (88%) studies in North Africa and nine (69%) in SSA. Among studies with stage information, and whose subject eligibility was not dependent on it, the proportion of patients with late stage (III/IV) was very high (range: 46%-61% in North Africa; 76%-91% in SSA; Table 1).

Delays in presentation and diagnosis
The time interval between symptom recognition by the woman to presentation, i.e. to first visit to a health care provider, varied substantially across studies but, overall, it was shorter in North Africa than in SSA (Table 2; Figure   3a). Of the five North African studies that reported on presentation delays, most yielded median estimates of <2.5 months; the only exception was a study in Libya (24) with a median presentation time of 4 months. Of the five studies in SSA that provided estimates of time from symptom recognition to presentation only one (25) reported a median time M A N U S C R I P T A C C E P T E D ACCEPTED MANUSCRIPT 9 of <2.5 months, with the remaining reporting average times ranging from 3.4 months in Mali (21) to >6 months in South Africa (15).
Fewer studies in North Africa (18,24,26) and in SSA (19,25,27) gave estimates of the time between presentation and diagnosis, or between diagnosis and start of treatment. Nevertheless, the length of these intervals tended to be shorter than the length of the corresponding intervals between symptom recognition and presentation in North Africa (all <1 month), but not in SSA (Figure 3b).
Five North African studies provided median estimates of the total delay from symptom recognition to date of breast cancer diagnosis or start of treatment ( Figure 3c). Two of these studies recruited only advanced breast cancer cases with average total delays of 8 (13) and 12 months (14). Median estimates of the total delay from symptom recognition to diagnosis for the remaining three studies ranged from 4 (18) to 8.5 months (17). Five SSA studies provided average times from presentation to diagnosis or start of treatment ( Figure 3c), with their estimates ranging from 7.9 months in Ghana (28) to 15 months in Rwanda (19); median delays were known to be >6 months for two studies (25,27) but their exact values could not be estimated. In addition, a small qualitative study (n=11) in Botswana reported a median time from first symptom(s) to presentation at the hospital where the diagnosis was finally made of 3 years (12).
The number of health care providers visited prior to the one where the diagnosis was made were reported by only one study in North Africa (26) and four in SSA (19,21,23,27), with estimates ranging from a median of 1.5 in Egypt (26) to >5 in Rwanda (19); however, these estimates are not entirely comparable because traditional and religious healers were included in two of these studies (23,27).
A few studies examined whether delays were associated with late stage (III/IV) at diagnosis. The study by Benbakhta et al. (18) in Morocco reported a 6.81-fold (95% CI: 3.65, 12.7) increase in the odds of late stage among patients who delayed presentation by >64 days relative to those who presented ≤64 days of symptom recognition.
Similarly, the odds of late stage among patients who experienced a diagnostic delay of ≥50 days was 1.84 (95% CI: 1.05, 3.23) times higher than among those diagnosed <49 days of their first presentation to a health care provider (18).

Factors associated with delays
Appendix C summarises the reasons most commonly reported by the quantitative studies in the review for late presentation to the first health care provider. They fell into the following categories: (i) socio-economic factors such as low educational level; (ii) lack of breast cancer awareness and poor knowledge of early-detection methods (e.g. breast self-examination); (iii) type of initial symptoms: painless, not taken seriously or hoping they would resolve soon; (iv) fear of the disease, its treatment (e.g. mastectomy) or death, or of being a burden to the family; (v) belief in traditional medicine or spiritual cures; (vi) financial constraints; and (vii) poor access to health care (e.g. living too far away from a health care provider; lack of transportation). Benbakhta et al. (18) found in mutually-adjusted analysis that a delay in presentation of ≥2.2 months in Morocco was positively associated with low socio-economic conditions (e.g. living in a rural area, being illiterate, being a housewife (vs. being employed) and having low socio-economic level) and lack of breast cancer awareness (e.g. negative family history of cancer, no knowledge of breast self-examination) (Appendix C). In contrast, Mousa et al. (26) found no association between delay in presentation >3 months in Egypt and a woman's socio-economic characteristics or type of symptoms before or after adjustment for potential confounders. In South Africa, Marcus et al. (15) found in mutually-adjusted analysis positive associations with late presentation (>6 vs. 3-6 months) with increasing age and a previous cancer diagnosis, but not with educational level, marital status or being employed/unemployed. A mutually-adjusted analysis of data from a study in Rwanda (19) revealed a four to five-fold increase in the odds of late presentation (≥6 months) for patients with low or no education, and for those who visited a traditional healer first, but no independent associations with other socio-economic, breast cancer awareness, symptom or health services-related variables (Appendix C). Overall, the findings from the qualitative studies supported the evidence from the quantitative studies (11,12,20,23) The reasons given by the patients for delays between presentation and diagnosis, or start of treatment, included patient-mediated factors (e.g. socio-economic factors, type of symptoms, having tried traditional treatments first, financial problems, fear of the disease and/or its treatment, and denial) as well as health care provider-mediated factors (e.g. travel time to health care provider, the number and type of health care providers contacted prior to diagnosis, delayed referrals or non-referrals, misdiagnosis, wrong advice or false reassurances, delays in obtaining diagnostic confirmation and in starting treatment) (Appendix C). The study in Morocco by Benbakhta et al. (18) found in mutually-adjusted analyses that a delay >1.7 months between presentation and start of treatment was associated with older age, illiteracy, low socio-economic level, distance to health care provider ≥100 kms and ≥3 consultations prior to the diagnostic one. Mousa et al. (26) in Egypt showed that after adjustment for potential confounders the odds of a delay >2 weeks from the first medical consultation to arrival at the diagnostic centre was not associated with the M A N U S C R I P T A C C E P T E D ACCEPTED MANUSCRIPT patient's age, socio-economic conditions or type of symptoms but was strongly associated with the type of the first health care provider visited and the navigation pathway followed by the patient (Appendix C). In Rwanda, Pace et al. (19) found in mutually-adjusted analyses a 2.69 (95% CI 1.24, 5.84) higher odds of a delay ≥6 months for patients who visited five or more health care facilities prior to diagnosis, but no associations with the patient's socioeconomic conditions, reproductive history or type of symptoms. In the qualitative studies (Appendix C), some women reported poor clinical practices (e.g. inadequate diagnosis by general doctors (11)), hospital strikes (20), or having sought alternative care after receiving the diagnosis).

DISCUSSION
To our knowledge, this is the first systematic review of studies that reported on delays in a woman's breast cancer journey in Africa. Its findings highlighted three main issues. Firstly, there is a paucity of published data on delays in the presentation and diagnosis of the most common female cancer in Africa (2) There is strong evidence that a delay from symptom recognition to diagnosis of more than three months is associated with later stage at presentation and poorer survival (6). This review revealed substantially longer delays in both North Africa and SSA, with reported average times from symptoms recognition to diagnosis between 4 and 15 months. These estimates are in line with those observed in other low and middle income countries (LMICs) (e.g. 7.6 months in Brazil (29); 5.5 months in Malaysia (30)) but much higher than those observed in HICs (e.g. 34 days in France (31); 48 days in the USA (32)). The very long time intervals from symptom recognition to diagnosis in Africa resulted from delays in both presentation and diagnosis. All studies in this review, with the exception of two (9,33), reported average presentation intervals between 2.2 months and >6 months, much longer that those observed in HICs (e.g. 9 days in the United Kingdom (34); 16 days in Germany (35)). Similarly, reported diagnostic intervals in Africa were much longer than those found in HICs (e.g. from 10 to 42 days in France (31), Germany (36) and the USA (32)), but similar to what has been described for other LMICs (e.g. median of 5 months in Brazil (29), Colombia (37) and Mexico (38)).

M A N U S C R I P T A C C E P T E D ACCEPTED MANUSCRIPT
As we had hypothesized in our conceptual model, delays in presentation in Africa were found to be associated not only with patient-mediated factors (e.g. low educational level, poor breast cancer awareness, use of alternative care medicine) but also with health services-mediated factors (e.g. distance to the nearest health care centre). These results are similar to those from previous studies -e.g. being unaware of the warning signs or tests for breast cancer (5), patients only seeking conventional care when traditional treatment has failed (39), or inability to afford the costs of treatment (40). Similarly, delays in diagnosis in Africa were influenced by both patient-mediated factors (e.g. low educational level, financial problems) and health system-mediated factors (e.g. type of first health care provider visited, number of providers visited prior to diagnosis, type of navigation pathway followed before reaching the diagnostic centre). A high number of referrals makes the patient's journey through the health system longer resulting in a more advanced tumour stage at diagnosis; however, it is also conceivable that a low number of referrals might reflect a more aggressive tumour, or a longer time interval before presentation to the first health care provider, and thus a more advanced tumour that was easily identified by the physician. Of note, however, is the fact none of the papers directly examined health system factors, e.g. through interviews with health care providers, relying instead on patients' reports.

Strengths and limitations of the review
Major strengths of this review include the systematic search strategy used to identify eligible English and non-English publications, and the use of standardised methods for data extraction and synthesis. The review also has weaknesses.
Its representativeness may have been compromised by several factors. First, publication bias cannot be excluded as grey literature was not included in this review. Second, the review included studies from only 4 of the 7 North African countries and 11 of 51 SSA countries, albeit the latter comprised studies from all four SSA regions (i.e. from Eastern, Western, Southern and Middle Africa). Third, none of the studies in the review were population-based; they were all hospital-based, predominantly from tertiary hospitals as these are the only ones in most African countries to have appropriate cancer diagnostic and treatment facilities. However, such studies excluded, by design, the large number of patients who never reach tertiary hospitals, some of whom are never diagnosed. Hence, the included patients who reached tertiary facilities are unlikely to be a representative sample of all breast cancer patients in Africa.
The methodological quality of most papers was low. In particular, measurement errors may have affected the validity of the review's findings as although most of the studies recruited women prospectively, patients were asked to remember the time from first symptom(s) to presentation, and this might have introduced recall errors, and even biases.
Little detail was provided in the original papers on the specific instruments used to collect information and the methods

M A N U S C R I P T A C C E P T E D ACCEPTED MANUSCRIPT
13 used to estimate times to presentation, diagnosis and treatment, including on the way questions to patients on time intervals were formulated and on how relevant time-related events (e.g. dates of contact with a first health care provider, breast cancer diagnosis and start of treatment) were defined. Between-study differences in these methodological issues may have affected their comparability. When questioned about the reasons for delays patients might have been reluctant to admit less orthodox behaviours such as the use of traditional medicine. Reassuringly, however, the studies that examined associations between self-reported delays and late stage at diagnosis showed, as expected, strong positive associations. Many studies had relatively small sample sizes and thus their ability to precisely quantify delays, and their power to detect associations, were limited. There were large variations across studies in the way data were analysed (e.g. only a few quantitative studies attempted to control for confounders; none of the qualitative studies conducted theoretical analyses), and summary findings presented, hampering between-study comparisons and precluding the conduct of meta-analyses.

CONCLUSIONS
Several studies in Africa have shown that early stage breast cancer is associated with better survival than late stage disease (41,42), consistent with early diagnosis and treatment being associated with reductions in mortality from this disease in the region. The long presentation and diagnostic delays identified by this review indicates that there is considerable potential to introduce interventions aimed at shrinking the time intervals between symptom recognition and diagnosis. Mammography screening is often advocated as the best intervention to improving early diagnosis of breast cancer but the findings from this review strongly argue against adopting such an approach in African settings.   Table 2 for more detailed information on study-specific estimates of delay. A dashed line indicates that the delay estimate shown in the Figure is an under-estimation of the median value (the latter could not be calculated from the data provided in the original paper). No delay estimates for Otieno et al. (16) are shown because average time from symptoms to diagnosis could not be estimated (>3 mths for 73% of patients -all with advanced BC -with no further information provided; see Tables 1  and 2).   n/a n/a n/a n/a n/a 16 n/a n/a n/a n/a n/a 20 Me: 44.5 n/a n/a n/a n/a n/a 10 n/a n/a n/a n/a n/a n/a M A N U S C R I P T A C C E P T E D ACCEPTED MANUSCRIPT 20  (9) Egypt (n=343) Md: <1 mth n/a n/a n/a n/a n/a n/a n/a Mbuka-Ongona, 2012 (12) Botswana (n=11) Time from first symptom to presentation at study hospital (PMH): Me: 3 yrs; Ra: 1 -10 yrs n/a n/a Pruitt, 2015 (20) Nigeria (n=31) n/a n/a n/a a Study recruited only patients with advanced breast cancer (see Table 1