Estimating the incidence of breast cancer in Africa: a systematic review and meta-analysis

Background Breast cancer is estimated to be the most common cancer worldwide. We sought to assemble publicly available data from Africa to provide estimates of the incidence of breast cancer on the continent. Methods A systematic search of Medline, EMBASE, Global Health and African Journals Online (AJOL) was conducted. We included population- or hospital-based registry studies on breast cancer conducted in Africa, and providing estimates of the crude incidence of breast cancer among women. A random effects meta-analysis was employed to determine the pooled incidence of breast cancer across studies. Results The literature search returned 4648 records, with 41 studies conducted across 54 study sites in 22 African countries selected. We observed important variations in reported cancer incidence between population- and hospital-based cancer registries. The overall pooled crude incidence of breast cancer from population-based registries was 24.5 per 100 000 person years (95% confidence interval (CI) 20.1-28.9). The incidence in North Africa was higher at 29.3 per 100 000 (95% CI 20.0-38.7) than Sub-Saharan Africa (SSA) at 22.4 per 100 000 (95% CI 17.2-28.0). In hospital-based registries, the overall pooled crude incidence rate was estimated at 23.6 per 100 000 (95% CI 18.5-28.7). SSA and Northern Africa had relatively comparable rates at 24.0 per 100 000 (95% CI 17.5-30.4) and 23.2 per 100 000 (95% CI 6.6-39.7), respectively. Across both registries, incidence rates increased considerably between 2000 and 2015. Conclusions The available evidence suggests a growing incidence of breast cancer in Africa. The representativeness of these estimates is uncertain due to the paucity of data in several countries and calendar years, as well as inconsistency in data collation and quality across existing cancer registries.


What is already known?
The International Association of Cancer Registries (IACR) has been the main sources of cancer data in Africa, particularly by aiding African countries towards establishment of cancer registries. North Africa appears to have more robust data on cancer, with estimated incidence rates of breast cancer across North African states comparable to some high-income settings. In SSA, current reports reveal an increasing number of breast cancer cases diagnosed in the ages 35-49 years, with many presenting at advanced late-stage disease. Largely, data availability remains an ongoing limitation in understanding and estimating the incidence of breast cancer on the African continent.

What this study adds
This study provides the first systematic review and meta-analysis of publicly available evidence on the incidence of female breast cancer in Africa. There is limited data on cancer in Africa, particularly in Central Africa; only 22 African countries were included in this study. Population-based cancer registries remain the main sources of data on breast cancer in Africa as these provided 67% of all data points. Our estimated incidence of breast cancer increased between 2000 and 2015 across both registries, suggestive of a rising breast cancer incidence in Africa. The mean age of populations covered ranged from 30.6 to 60.8 years, with over 33% and 81% of population in ages 30-49 years, and 30-59 years, respectively.

Search strategy and data sources
We conducted an initial scoping literature search to identify Medical Subject Headings (MESH) and relevant keywords, following which a final search strategy was developed. We conducted a systematic search of Medline, EMBASE, Global Health and African Journals Online (AJOL), with search dates set from January 1980 to December 2016. We also conducted additional searches of Google Scholar, and websites of the International Association of Cancer Registries (IACR) and WHO African Region (AFRO). We equally reviewed the "GLOBOCAN studies" [6,9,16], "Cancer Incidence in Five Continents (CI5) series" [17], and "Cancer in Africa: Epidemiology and Prevention" [18] for more studies or additional data for studies already selected. Reference lists of initially selected studies were further hand-searched. The list of African countries was based on the World Bank list of economies [19] (Table 1).

Selection criteria
We included population-based or hospital-based cancer registry studies on breast cancer conducted primarily on African population groups, and providing crude estimates of the cases or incidence of breast cancer among women in the population over a specified period. Studies were included if they were completed on or after the year 2000, to give a fair representation of the current burden. Hospital-based studies that only reported cases of breast cancer were also included if they provided sufficient information on the reference population to allow an estimation of the population or person-years at risk. We excluded studies on non-human subjects, and those that were mainly reviews, case reports, opinions or editorials. No language restrictions were applied.

Confirmation of breast cancer diagnosis
We included studies that identified breast cancer based on histological diagnosis with or without i) clinical evaluation by a physician, ii) radiological investigations (mammography, breast ultrasound, computed tomography scan or magnetic resonance imaging), iii) laboratory tests (estrogen or progesterone receptor (ER or PR) status, the human epidermal growth factor receptor 2 (HER2) status, and breast cancer antigen (BRCA1 or BRCA2) or iv) community reported cases.

Quality criteria
Each study was assessed for quality according to a set of five (5) predefined criteria. The first criterion was based on the cancer registration process. This considered how cancer registries collated data and the approach employed for data ascertainment. The second considered the coding criteria employed across studies to determine if the reported cancer types were classified according to the primary anatomic site Table 1. Search terms for studies on breast cancer in Africa Number SearcheS 1 africa/ or africa, northern/ or algeria/ or egypt/ or libya/ or morocco/ or africa, central/ or cameroon/ or central african republic/ or chad/ or congo/ or "democratic republic of the congo"/ or equatorial guinea/ or gabon/ or africa, eastern/ or burundi/ or djibouti/ or eritrea/ or ethiopia/ or kenya/ or rwanda/ or somalia/ or sudan/ or tanzania/ or uganda/ or africa, southern/ or angola/ or botswana/ or lesotho/ or malawi/ or mozambique/ or namibia/ or south africa/ or swaziland/ or zambia/ or zimbabwe/ or africa, western/ or benin/ or burkina faso/ or cape verde/ or cote d'ivoire/ or gambia/ or ghana/ or guinea/ or guinea-bissau/ or liberia/ or mali/ or mauritania/ or niger/ or nigeria/ or senegal/ or sierra leone/ or togo/ 2 exp vital statistics/ or exp incidence/ 3 (incidence* or prevalence* or morbidity or mortality).tw. 4 (disease adj3 burden).tw. 5 exp "cost of illness"/ 6 exp quality-adjusted life years/ 7 QALY.tw. 8 Disability adjusted life years.mp. 9 (initial adj2 burden).tw. 10 exp risk factors/ 11 2 or 3 or 4 or 5 or 6 or 7 or 8 or 9 or 10 12 exp breast cancer/ 12 1 and 11 and 12 (topography) or cellular characteristics (morphology-histology, behaviour, and grade) using the international classification of diseases (ICD) and oncology (ICD-O) guidelines [20][21][22]. The third assessed how population or person-years at risk were generated in each study. The fourth and fifth criteria checked whether the population covered in each study was representative of the target (subnational) and national populations, respectively. Each criterion was scored one (1), and studies were finally graded as high (4)(5), moderate (2)(3) or low (0-1) quality based on the number of criteria they met (Table S1 in Online Supplementary Document). All low-quality studies were excluded from the review.

Data extraction
Two reviewers (OYS and AA) independently screened studies against the inclusion and exclusion criteria and performed the data extraction. Any disagreement over article inclusion, exclusion or data extraction between the two initial reviewers was resolved through a final assessment by a third reviewer (DA). As we already employed a two-stage search and extraction process based on a combination of independent review and reassessment, we did not calculate Kappa statistics to determine agreement between the reviewers. We extracted data and relevant information systematically from each study. This included location, period, design, cancer registry, confirmation of diagnosis, data collation methods, coding criteria, data ascertainment and modality with which population or person years at risk were generated. The basic numerical data extracted were mean age (or age range), population or person years at risk, cancer cases, and crude incidence. For studies conducted on the same cancer registry over the same period, we selected the first chronologically published study, and supplementary data from other studies were added to the selected paper. All extracted data were separated into hospital-based and population-based studies, further sorted based on different regions in Africa (central, east, north, south, and west), and stored in Excel 2013 (Microsoft Inc, Redmond, WA, USA).

Data analysis
From extracted crude incidence rates of breast cancer, we conducted a random effects meta-analysis (Der-Simonian and Laird method) [23]. For studies not reporting population or person years at risk, we checked the UN population projections of the country for the study period and multiplied by the years covered. For sub-national populations, we applied the subnational population ratio to standardize the United Nations (UN) projections of the country for the study period. Standard errors were estimated from the crude incidence rates and person years assuming a Poisson distribution. Crude meta-estimates and confidence intervals (expressed per 100 000 person-years) were pooled from individual crude incidence rates and reported by registry-type for the African continent, the African sub-regions, age groups, and the years 2000 and 2015. I-squared (I 2 ) statistics and subgroup (sensitivity) analysis were conducted to assess heterogeneity between studies. We conducted a meta-regression analysis (with age and year as confounders) to further check consistency of the data extracted and estimates reported for hospital-based and population-based registries, respectively. All statistical analyses were conducted on Stata 13.1 (Stata Corp LP, College Station, TX, USA).

Systematic review
The literature search returned a total of 4648 studies. There were 4633 studies identified from the databases-Medline (1035), EMBASE (2811), Global Health (733) and AJOL (54). 15 additional studies were identified from Google Scholar, IARC and WHO AFRO websites, and hand-searching of reference lists. A total of 2858 records remained after duplicates were excluded. After reviewing titles for relevance (ie, studies providing incidence of breast cancer in any African location), 2650 studies were excluded, giving a total of 208 full texts that were assessed. After applying the selection criteria, 167 studies were excluded. A total of 41 studies were finally retained for the review   (Figure 1).

Study characteristics
The 41 studies retained were conducted across 54 study sites (locations) in 22 African countries ( Table  2). West Africa had the highest number of study sites (16) with seven of these sites located in Nigeria. South Africa closely followed with 6 sites while others were Uganda and Tunisia (5 sites each), Morocco (4 sites), Algeria, Ghana, and Libya (3 sites each), and Cameroon, Egypt, Mozambique and Niger (2 sites each). In total, SSA had 36 sites (70%) while North Africa had 18 sites (30%) (see Table 2 for details). Population-based registries accounted for 67% of all study sites. These population-based registries reported mainly active or passive case findings across referral hospitals, pathology laboratories, and communities within the geographical remit of the cancer registry. Data were abstracted mostly from pre-designed pathology/laboratory forms confirming breast cancer diagnosis, with all sites reporting morphological/histological verification of the breast cancer cases. Mean age of study population ranged from 30.6 to 60.8 years (median 50.2 years), with over 33% in the age group 30-49 years, and over 81% in 30-59 years ( Table 2).
Following the quality assessment, 23 (56%) studies were graded high, of which (20) 87% were conducted in population-based registries, which suggests the internal consistency and validity of these registries. However, most hospital-based studies were of moderate quality, as 15 (83%) of the total 18 studies graded moderate were conducted in hospital-based registries (Figure 2, Table 2, Table S1 in Online Supplementary Document). The basic limitations of many hospital-based registries were related to poorly defined data collation and ascertainment process. Moreover, catchment areas were also not well-defined, with this affecting the estimation of person-years across these registries.
Pooled crude incidence rate of breast cancer in Africa

Population-based registries
From population-based registries (36 study sites), the overall pooled crude incidence of breast cancer in Africa was 24.5 per 100 000 person years (95% Confidence Interval [CI]: 20.1-28.9) (Figure 3, plate A  Figure 2). Besides, a sensitivity analysis based on quality criteria showed studies graded as high quality had almost similar pooled incidence as the overall pooled incidence at 24.3 per 100 000 person years (95% CI 19.9-28.8), compared to moderate quality studies at 26.1 per 100 000 (95% CI 5.5-46.7) (Figure 3, plate B).
Moreover, in the population-based registries, we observed an increasing incidence with increasing age of subjects and over the study period covered ( Table 3). The pooled incidence rate was highest among persons older than 60 years at 36.6 per 100 000 (95% CI 30.7-42.4) compared to an incidence rate of 3.3 per 100 000 (95% CI 2.7-3.8) among persons aged 30-39 years ( Table 3). Between 2000 and 2015, the      Table 3).
The meta-regression analysis showed a significant association of the crude incidence rates with age and year (P = 0.0012) ( Table 4), which possibly demonstrates a relative consistency and representativeness of the data captured by population-based cancer registries. Assuming our pooled population-based registry estimates accounted for population growth, ageing and other demographic and epidemiological changes in Africa, these rates would account for 93 890 (95% CI 76 007-111,774) breast cancer cases among women in 2000. This increased to 150 526 breast cancer cases by 2015 (95% CI 107 600-192,880), based on the UN demographic projections for Africa.

DISCUSSION
As the leading cause of cancer globally among women, breast cancer has continued to attract interest among experts and clinicians, particularly towards appropriately quantifying its burden and identifying risks across countries. In Africa, experts have continued to work towards addressing several challenges in the response to a growing cancer burden [14]. However, with relatively scant, incomplete, or poorly representative data on cancer in many settings, it remains difficult to effectively address this burden [15]. This study provides the first systematic review and meta-analysis of publicly available evidence on the incidence of female breast cancer in Africa with a view to improving the understanding of the epidemiology of the disease in the region. This endeavor yielded some findings.
Our study showed that population-based cancer registries remain the main sources of data on breast cancer in Africa as these provided 67% of data in this study. Current data from most hospital-based cancer registries are rather too incomplete and inconsistent to be deployed in the understanding of cancer epidemiology in Africa [12], as evidenced from the quality assessment. In the population-based registries, we observed that the estimated regional variation in breast cancer incidence in Africa was consistent with the GLOBOCAN 2008 and 2012 studies, with the incidence in North Africa higher than observed in SSA [6,9]. The estimated incidence of breast cancer in North Africa (29.3 per 100 000) was almost same as values reported by the global burden of diseases (GBD) collaborators for the region (29.7 per 100 000) [2]. However, the GBD collaborators noted a relatively higher incidence rate in SSA in 2015 at 34.1 per 100 000 compared to our estimate of 22.4 per 100 000 (95% CI 17.2-28.0) for SSA [2]. The GBD estimates are thus suggestive of a higher incidence of breast cancer in SSA compared to North Africa, which are in contrast to our findings of the opposite, and possibly raises potential questions regarding the rising trend of breast cancer in SSA compared to North Africa. The differences may be partly attributable to sources and completeness of data, as well as modeling approaches employed in the GBD estimates compared to the present study. Although some previous reports have shown that the estimated incidence rates of breast cancer across North African states are higher than in SSA, and comparable to rates in some high-income settings [65][66][67]. The regional differences in Africa may be indicative of racial and ethnic disparities in invasive breast cancer epidemiology in the region, as already observed in the variations in incidence and mortality patterns between black and white women globally [68,69]. Dickens and colleagues noted that the clinical spectrum of breast cancer in Africa is highly heterogenous across population groups [70], with this often linked to variations in age at diagnosis, staging, time trends, and genet- ic and environmental risks [69,71]. Moreover, the different stages of fertility transition currently observed in Africa as evidenced by declining births and widening birth intervals across population groups may also be an important driver of the regional differences [71].
Across both population-and hospital-based registries, we reported lower incidence rates in Central Africa. However, these estimates have wide uncertainty intervals, possibly reflecting the limited data in the region, and the substantial variation and imprecision in reported figures already noted across African countries [10,72]. Some authors have reported the lack of data in Central Africa as a major factor mitigating against the estimation of disease burden and ensuring evidence-based interventions in the region [12,73,74]. Balekouzou and colleagues specifically noted that the epidemiology of breast cancer in Central Africa remain poorly understood, owing to sparse data on cancer in the region [75]. Ongoing civil unrest in some parts of Central Africa influencing the capacity of the health systems to adequately capture cancer cases may have resulted in a relatively low cancer ascertainment.
The mean age of populations covered in both the population-and hospital-based registries ranged from 30.6 to 60.8 years (median 50.2 years), with over 33% and 81% of population in ages 30-49 years, and 30-59 years, respectively. This, albeit subject to further validations, may be suggestive of a high incidence of breast cancer among younger age groups in Africa. For example, Jedy-Agba and colleagues reported that many cases of breast cancer in SSA were diagnosed in the ages 35-49 years, with many presenting at advanced late-stage cancer [7]. Some authors have argued that the higher proportion of younger women in Africa, due to a relatively lower life expectancy on the continent, may have been responsible for the increasing breast cancer incidence reported among younger age groups [14,76]. However, epidemiological transitions, rapid urbanization with increased adoption of unhealthy lifestyles, increasing prevalence of obesity in younger populations, changing reproductive behaviours, including early menarche, low parity, advanced age at first pregnancy, and low self-breast examination for breast cancer among adolescents and young women have been identified as possible risk factors [14,76,77].
The poor state of many health systems in Africa and their declining capacity to lead cancer preventive initiatives and response to overall health needs of the population, as compared to developed countries, are also major concerns. Abdulrahman and Rahman reported that there are significant differences in age and stage at presentation between women in Africa and Europe, with more than half of breast cancer patients in Nigeria and Libya presenting at relatively younger age, mostly with advanced stage III or IV disease [78]. Parkin et al. equally noted that the lifetime risk of dying from breast cancer in young African women is about twice the risk in high-income countries [56].
Meanwhile, we observed some important variations in the estimates reported from population-and hospital-based registries in this study. Although our estimated incidence of breast cancer increased between 2000 and 2015 in both registries, suggestive of a rising breast cancer incidence in Africa, there was however a more significant rise in breast cancer incidence with advancing age in the population-based registries. This was clearly different from our finding from the hospital-based studies. Moreover, from population-based registries, the breast cancer incidence in North Africa (29.7 per 100 000) was higher than the incidence in SSA (22.4 per 100 000), while hospital-based registries provided relatively similar estimates (23.2 per 100 000 vs 24.0 per 100000). These variations, notwithstanding our estimation approach, further raise concerns on the completeness and representativeness of data in African registries. Population-based cancer registries have been widely regarded as being the gold standard for accurate data collation on cancer [79], as they cover wide geographical areas including communities, and employ standard coding methods and estimation of person-years [12]. However, challenges have been identified in the establishment and effective functioning of population-based registries across Africa due to poor funding, lack of skilled personnel, and low priorities from many governments, with this affecting the estimates provided [12,79]. Some authors have reported that hospital-based registries can indeed complement the few population-based registries, particularly by providing data on health service provision and access, which is useful for appropriate planning and public health response [79]. To address challenges with data consistency and representativeness, efforts may also be directed at improving geographical mapping of hospitals to appropriately determine the catchment area or population covered, and generate standardized methods to calculate cancer incidence that are representative of the population [12].
The application of precision medicine across world regions is still evolving [3]. However, translating this into prevention and management of breast cancer in Africa also requires additional efforts with cancer registration [3]. It is important for future research to decipher if the advanced stage of breast cancer presentation in Africa could be linked to unique aggressive biological characteristics of the malignancy, or The limitations of this study are also related to our key findings. First, though we performed a compressive data retrieval process, the empirical estimates we provided are limited by the absence of population-based cancer registries in many African countries (only 22 African countries represented), which has implications for the precision and representativeness of our estimates. Although, about 67% of studies were conducted in population-based cancer registries, many of these registries only covered subnational populations. As highlighted in the results, Nigeria and South Africa had the highest data points in this study. However, of all entries, South Africa particularly provided a complete national registry report of histologically-diagnosed breast cancer cases covering a period of 18 years, although only recent reports suggested an upgrade to population-based cancer registration system [80]. As noted, across data from hospital-based reports, we relied on the UN demographics for the country to estimate reference populations and person-years when they were not provided. This could have led to a relative under-estimation of cancer incidence due to resultant large denominators derived from the UN demographics. Besides, estimates were not age-standardized which limits direct comparisons with other world regions. Moreover, the heterogeneity across selected studies was high (I 2 >90%), which further reflects the need to address varying study designs, case ascertainments, referral approaches, and standards for collation of cancer data across different population groups, as these are important sources of heterogeneity. Despite these limitations, it is still important to present to the public the challenges arising from data availability on breast cancer research in Africa, the varying estimates available, and the implication of this on the response to a rising burden of breast cancer on the continent.

CONCLUSIONS
The efforts of various researchers and organizations, particularly the IARC, United States National Cancer Institute (NCI) and African Organization for Research and Training in Cancer (AORTIC), must be acknowledged in the response to cancer in Africa. Data availability remains an ongoing limitation in understanding and estimating the incidence of breast cancer on the continent. Further issues from available data include inconsistency in data collation and quality across existing cancer registries. However, available evidence is suggestive of a growing incidence of breast cancer in Africa. With population growth and ageing, exposures to known risks, and relative weak health systems in Africa, it is unlikely this trend will change in the short-term in the absence of effective interventions, with this further straining the health system in many African countries. As various funders and institutions develop their focuses, we assert that strategies for controlling and treating breast cancer patients must include activities that support active and comprehensive data collation and cancer registration (population-and hospital-based) on the African continent, and more emphasis on the biological and histological characteristics of breast cancer cells across different population groups.