Cancer Causes & Control https://doi.org/10.1007/s10552-021-01515-0 ORIGINAL PAPER Overall and central obesity and prostate cancer risk in African men Ilir Agalliu1  · Wei‑Kaung Jerry Lin2 · Janice S. Zhang1,2 · Judith S. Jacobson3 · Thomas E. Rohan1 · Ben Adusei4 · Nana Yaa F. Snyper4 · Caroline Andrews5 · Elkhansa Sidahmed5,6 · James E. Mensah7 · Richard Biritwum7 · Andrew A. Adjei8 · Victoria Okyne7 · Joana Ainuson‑Quampah9 · Pedro Fernandez10 · Hayley Irusen10 · Emeka Odiaka11 · Oluyemisi Folake Folasire11 · Makinde Gabriel Ifeoluwa11 · Oseremen I. Aisuodionoe‑Shadrach12 · Maxwell Madueke Nwegbu12 · Audrey Pentz13 · Wenlong Carl Chen13,14,15 · Maureen Joffe13,16 · Alfred I. Neugut17 · Thierno Amadou Diallo18 · Mohamed Jalloh18 · Timothy R. Rebbeck5,6 · Akindele Olupelumi Adebiyi11 · Ann W. Hsing2,19,20,21 Received: 17 May 2021 / Accepted: 29 October 2021 © The Author(s), under exclusive licence to Springer Nature Switzerland AG 2021 Abstract Purpose African men are disproportionately affected by prostate cancer (PCa). Given the increasing prevalence of obesity in Africa, and its association with aggressive PCa in other populations, we examined the relationship of overall and central obesity with risks of total and aggressive PCa among African men. Methods Between 2016 and 2020, we recruited 2,200 PCa cases and 1,985 age-matched controls into a multi-center, hospital-based case–control study in Senegal, Ghana, Nigeria, and South Africa. Participants completed an epidemiologic questionnaire, and anthropometric factors were measured at clinic visit. Multivariable logistic regression was used to examine associations of overall and central obesity with PCa risk, measured by body mass index (BMI), waist circumference (WC), waist-to-hip ratio (WHR), and waist-to-height ratio (WHtR), respectively. Results Among controls 16.4% were obese (BMI ≥ 30 kg/m2), 26% and 90% had WC > 97 cm and WHR > 0.9, respectively. Cases with aggressive PCa had lower BMI/obesity in comparison to both controls and cases with less aggressive PCa, suggesting weight loss related to cancer. Overall obesity (odds ratio: OR = 1.38, 95% CI 0.99–1.93), and central obesity (WC > 97 cm: OR = 1.60, 95% CI 1.10–2.33; and WHtR > 0.59: OR = 1.68, 95% CI 1.24–2.29) were positively associated with D’Amico intermediate-risk PCa, but not with risks of total or high-risk PCa. Associations were more pronounced in West versus South Africa, but these differences were not statistically significant. Discussion The high prevalence of overall and central obesity in African men and their association with intermediate-risk PCa represent an emerging public health concern in Africa. Large cohort studies are needed to better clarify the role of obesity and PCa in various African populations. Keywords Prostate cancer · African men · Body mass index · Obesity · Central adiposity · Sub-Saharan Africa Introduction Prostate cancer (PCa) is the second most commonly diag- nosed solid tumor and the sixth leading cause of cancer Akindele Olupelumi Adebiyi and Ann W. Hsing have contributed equally as senior authors. deaths among men worldwide [1, 2]. In 2018, about 1.3 million men were diagnosed with PCa, and 360,000 men * Ilir Agalliu died from it [1]. Incidence and mortality rates of PCa ilir.agalliu@einsteinmed.org vary significantly by race/ethnicity and by geographic * Akindele Olupelumi Adebiyi region [1–3]. African American and Afro-Caribbean men adebiyi20012002@yahoo.com have the highest PCa incidence and mortality rates in the * Ann W. Hsing world [2, 4]. Prostate cancer risk in African men is less annhsing@stanford.edu clear. Despite potential underreporting of PCa in Africa Extended author information available on the last page of the article Vol.:(012 3456789) Cancer Causes & Control [3, 5], the most recent estimates from the International Materials and methods Agency for Research on Cancer indicate an age-adjusted PCa incidence rate of 84.5 per 100,000 person-year for Study population and recruitment African men [2, 4, 6]. In comparison, current age-adjusted PCa incidence rates among US black men are 175.2 per The MADCaP is an international consortium established to 100,000 men [4]. Recent data suggest that PCa mortal- investigate genetic and epidemiological risk factors of PCa ity rates among African men are among the highest in among men of African ancestry [25]. For this study, the the world [1, 2, 4], suggesting that PCa in this popula- MADCaP team included researchers in seven tertiary-care tion is either diagnosed at advanced stage due to limited hospitals and their affiliated universities in West and South access to health care and PCa screening, or has an unusu- Africa and four twinning centers at US universities. ally aggressive pattern. The World Health Organization Men aged 30 years or older, who resided in the catch- (WHO) estimated that the annual number of deaths from ment areas defined by the seven tertiary-care hospital centers PCa in Africa is expected to increase from 42,298 in 2018 between 2016 and 2020 and reported no European, Middle to 94,909 in 2040, a 124.4% increase in the next two dec- Eastern, or Asian grandparents or parents were eligible for ades [7]. This increase is higher than those estimated for recruitment. We excluded men who had any prior cancer North America (+ 101.2%), Europe (+ 58.3%), and Asia diagnoses, except for non-melanoma skin cancer. All centers (+ 105.6%) [7]. used common standardized protocols for subjects’ recruit- Despite the high incidence of PCa worldwide, other ment, interviews, and data collection and processing [25, than age, family history of PCa, and race/ethnicity [8, 9], 26]. few etiological factors have been established. Obesity, usu- ally defined as a body mass index (BMI) ≥ 30 kg/m2, has been linked to PCa, but it is more consistently associated Prostate cancer cases with PCa mortality and aggressiveness than with overall PCa incidence [10–12]. For example, in a large meta-anal- Men diagnosed with histologically confirmed PCa within ysis of 17 cohort studies including 76,978 cases, obesity the 6-month period prior to the date of study enrollment was not associated with total PCa risk, but was associated at each center, were eligible and recruited through Depart- with statistically significant 14% and 24% increased risks ments of Urology and Oncology at participating hospitals. In of aggressive cancer and PCa-specific mortality, respec- fact, this eligibility criterion applied mostly to PCa patients tively [10]. The Pooling Project of Prospective Studies of recruited during the first year (i.e., 2016) of the study period; Diet and Cancer recently reported positive associations all subsequently enrolled cases were newly diagnosed PCa between baseline BMI and risks of advanced PCa [haz- patients. The median time between PCa diagnosis and ard ratio (HR)  = 1.30, 95% CI 0.95–1.78] and  PCa- recruitment into the study for all PCa cases was 29 days specific mortality (HR = 1.52, 95% CI 1.12–2.07) when (0.98 months), and the interquartile range (IQR) was from comparing BMI ≥ 35.0 versus 21–22.9 kg/m2 [13]. In this 13 days (0.43 months) to 47 days (1.57 months). However, study, waist circumference (WC) and waist-to-hip ratio for cases enrolled in the first year of the study period, the (WHR), were also associated with 14% and 16% increased median interval between PCa diagnosis and recruitment was risks of high-grade PCa, respectively [13]. Unlike BMI, 69 days (2.3 months) and the IQR was 34 days (1.13 months) these measures reflect adipose tissue accumulation in the to 144 days (4.8 months). Physicians in the participating abdominal region [14]. Several studies, although not all, departments reviewed medical charts to confirm PCa diag- have suggested that central obesity, measured by WC, noses and pathological tumor characteristics. WHR, or waist–height ratio (WHtR), is more consistently associated with risks of overall PCa or more aggressive Controls cancer compared to BMI [13, 15–22]. Noting the increasing prevalence of obesity in sub-Saha- Men with no history of PCa or other cancers, who were ran Africa (SSA) [23, 24], as well as the rising incidence of seen for other conditions or diseases in the departments not PCa in this region [1, 2], we investigated the relationships affiliated with urology or oncology at participating tertiary- of overall obesity/BMI and central obesity measurements care hospitals, and who resided in the same catchment area (e.g., WC, WHR and WHtR) with risks of total PCa and as cases were recruited as controls [25]. The main hospi- more aggressive cancer in a large, multi-center, hospital- tal departments for recruitment of controls were Internal based case–control study of patients recruited in Senegal, Medicine (including Cardiology), Family Medicine, Gen- Ghana, Nigeria, and South Africa through the Men of Afri- eral Surgery (not including Urology), Ophthalmology, and can Descent and Carcinoma of the Prostate (MADCaP) Orthopedics. Controls were frequency matched to PCa cases consortium. within each hospital by 5-year age group, and participating 1 3 Cancer Causes & Control center. The study protocol and procedures were approved measurements and PCa risk [27]. We used normal prob- by the Institutional Ethical Review Boards (IRBs) of all ability plots and Q–Q plots to visualize data, and Kol- participating institutions. All cases and controls provided mogorov–Smirnov and Shapiro–Francia tests to evaluate written informed consent to participate in the study. The normality assumptions. The BMI was normally distributed; participation rates/proportions ranged from 89% to 100% in however, WC, WHR and WHtR were not normally distrib- PCa cases, and 85%–99% in controls across seven participat- uted, and therefore were analyzed as categorical variables, ing hospitals. However, overall, 95% of both eligible cases using either center-specific quartiles based on the distribu- and controls agreed to participate and completed the study tions among controls, or, when available, standard/clinically protocol and procedures. meaningful cutoff points. We used the WHO definitions of general obesity as BMI ≥ 30 kg/m2, and central obesity as Data collection WC > 95 cm, WHR > 0.9 and WHtR > 0.59, although the cutoff points for WHR were further modified based on Interview empirical distribution of data and literature [28–32]. Each anthropometric variable was fitted separately in each logistic All study participants completed an epidemiological ques- regression model. Models were adjusted for the following tionnaire through an  in-person interview that collected categorical variables: participating hospital center in SSA, detailed information on demographics (e.g., age, ethnicity/ age at enrollment (5-year categories), occupation, cigarette tribe), lifestyle and social factors (e.g., cigarette smoking, smoking status (i.e., never, former, and current smoker), alcohol consumption, physical activity, education, occupa- presence of hypertension and diabetes (see Table 1 for cat- tion, income), family history of PCa, personal medical his- egories). Each of these variables satisfied the criteria for tory of chronic diseases and history of PCa screening. PCa confounding because they changed the ORs between body cases were also queried about signs and symptoms of PCa size measurements and PCa by 10% or more. Education, and any cancer treatment that they had received. Both cases alcohol drinking habits, and moderate physical activity did and controls provided consent for the team to access their not change the ORs by 10% or more; therefore, they were medical records to abstract clinical information relevant to not included as confounders in our final models. To evalu- PCa diagnosis and pathological features, comorbid condi- ate the linear trend of PCa risk with increasing values or tions, and hospitalizations. quartiles of body size measurements, we performed both the Cochran–Armitage and Wald tests (we present the Wald test Body size measurements p values from multivariate logistic regression models) [33]. We excluded 82 (3.7%) cases and 62 (3.1%) controls At the clinic visit immediately following recruitment, trained with missing anthropometric measures from the analysis. In study personnel took anthropometric measurements from addition, we excluded 17 (2.3%) controls with PSA ≥ 20 ng/ each participant. Height (cm), weight (kg), waist (cm) and ml because of the possibility that their higher PSA levels hip (cm) circumferences were measured using a stadiom- might have been due to undiagnosed PCa, and not from eter, beam scale, and non-stretching measuring tape, respec- other comorbidities such as benign prostatic hyperplasia tively, using standardized protocols. Body mass index (BMI) (BPH) or prostatic inflammation. We also carried out a sen- was calculated as weight (kg) divided by height in meters sitivity analysis where we excluded controls with a serum squared (kg/m2). Waist-to-hip ratio (WHR) was calculated PSA > 4 ng/ml (to be consistent with the recommended as waist circumference (cm) divided by hip circumference guidelines for PSA screening cut-point used in the US). (cm). Waist-to-height ratio (WHtR) was calculated as waist The proportion of missing data was < 10% (median 4–5%) circumference (cm) divided by height (cm). The data coordi- for most variables; cases had slightly more missing data on nating center at DFCI implemented quality control measures anthropometric measures and some other variables (e.g., and data harmonization across centers [25]. cigarette smoking, occupation) compared to controls (see Table 1). Statistical Analysis Clinical variables and PCa risk categories We grouped PCa We compared characteristics between PCa cases and con- cases based on biopsy Gleason score (GS): ≤ 6, 7(3 + 4), trols using Student t-tests or Wilcoxon rank-sum tests for 7 (4 + 3), 8–10, corresponding to the Grade Groups 1, 2, normally or not-normally distributed continuous variables, 3, ≥ 4. We also grouped PCa cases based on D'Amico risk and Chi-square tests for categorical variables; all tests were classification scheme [34] using the following three cat- 2-sided using significance level α = 0.05. We used logistic egories: low-risk (i.e., T1–T2a, GS ≤ 6, and PSA ≤ 10  ng/ regression to estimate odds ratios (ORs) and 95% confidence ml), intermediate-risk (i.e., T2b, GS = 7, or PSA 10–20), intervals (CIs) for the associations between anthropometric and high-risk group (i.e., ≥ T2c, GS = 8–10, or PSA > 20). 1 3 Cancer Causes & Control Table 1 Selected characteristics Characteristics Cases Controls of prostate cancer cases and controls from the Men of n = 2,200 n = 1,985 African Descent and Carcinoma of the Prostate (MADCaP) n % n % Consortium Participating centers/hospitals  Hôpital Général de Grand Yoff, Dakar, Senegal 232 10.5 226 11.4  37 Military Hospital, Accra, Ghana 184 8.4 187 9.4  Korle-Bu Teaching Hospital, Accra, Ghana 403 18.3 386 19.4  University College Hospital, Ibadan, Nigeria 201 9.1 123 6.2  University of Abuja Teaching Hospital, Abuja, Nigeria 98 4.5 104 5.2  Stellenbosch University, Cape Town, South Africa 170 7.7 139 7.0  WITS Health Consortium, Johannesburg, South Africa 912 41.5 820 41.3 African r egiona  West Africa 1,118 50.8 1,026 51.7  South Africa 1,082 49.2 959 48.3 Age at enrollment (years)   < 60 330 15.0 491 24.7  60–69 943 42.9 866 43.6  70–79 764 34.7 506 25.5   ≥ 80 163 7.4 122 6.1 Marital status  Single/never married 76 3.5 139 7.0  Married 1,632 74.2 1,494 75.3  Divorced or separated 152 6.9 139 7.0  Widowed 195 8.9 151 7.6  Missing 145 6.6 62 3.1 Education  No formal education or < 4 years of schooling 349 15.9 329 16.6  5–12 years of schooling 696 31.6 499 25.1  Some secondary or Senior secondary schooling 486 22.1 722 36.4  Post-high school training 98 4.5 88 4.4  Some college 105 4.8 89 4.5  College graduate or postgraduate 297 13.5 159 8.0  Other 23 1.0 38 1.9  Missing 146 6.6 61 3.1 Smoking status  Non-smokers 1,024 46.5 973 49.0  Former smokers 737 33.5 640 32.2  Current smokers 253 11.5 297 15.0  Missing 186 8.5 75 3.8 Alcohol drinking status  Non-drinker 714 32.5 682 34.4  Former drinker 799 36.3 663 33.4  Current drinker 498 22.6 564 28.4  Missing 189 8.6 76 3.8 Occupation  Professional 252 11.5 153 7.7  Managerial 154 7.0 106 5.3  Technical/sales/administrative/office worker 227 10.3 328 16.5  Service 190 8.6 240 12.1  Operators, fabricators and laborers 560 25.5 520 26.2  Farmer 69 3.1 78 3.9 1 3 Cancer Causes & Control Table 1 (continued) Characteristics Cases Controls n = 2,200 n = 1,985 n % n %  Artisan 158 7.2 128 6.4  Other o ccupationsb 439 20.0 364 18.3  Missing 151 6.9 68 3.4 Diabetesc  No 1,794 81.5 1,615 81.4  Yes 287 13.0 318 16.0  Missing 119 5.4 52 2.6 Hypertensionc  No 1,059 48.1 1,112 56.0  Yes 1,022 46.5 821 41.4  Missing 119 5.4 52 2.6 Hypercholesterolemia (high blood cholesterol)c  No 1,934 87.9 1,851 93.3  Yes 147 6.7 82 4.1  Missing 119 5.4 52 2.6 Number of comorbid c onditionsc,d  0 834 37.9 790 39.8  1–2 1,080 49.1 997 50.2   ≥ 3 286 13.0 198 10.0 First-degree relatives with prostate cancere  0 1,795 89.4 1,755 95.0  1 160 8.0 54 2.9   ≥ 2 32 1.6 8 0.4  Missing 21 1 31 1.7 Moderate-intensity physical a ctivityf  No 1,819 82.7 1,606 80.9  Yes 194 8.8 301 15.2  Missing 187 8.5 78 3.9 Serum PSA at PCa diagnosis (cases) or at time of recruitment n = 704 (controls) (ng/ml)g  0–3.9 11 0.5 599 85.1  4–9.9 252 11.5 68 9.7  10–19.9 341 15.5 20 2.8  20–49.9 455 20.7 11 1.6   ≥ 50 993 45.1 6 0.9  Missing 148 6.7 Clinical tumor s tageh  cT1 653 29.7  cT2 809 36.8  cT3 212 9.6  cT4 157 7.1  Missing/unknown 369 16.8 Gleason score/grade group (GG)   ≤ 6/GG 1 363 16.5  7(3 + 4)/GG 2 506 23.0  7(4 + 3)/GG 3 385 17.5  8–10/GG 4 or 5 789 35.9  Missing 157 7.1 1 3 Cancer Causes & Control Table 1 (continued) Characteristics Cases Controls n = 2,200 n = 1,985 n % n % D'Amico risk classification g roupi  Low risk (T1–T2a and GS ≤ 6 and PSA ≤ 10) 74 3.4  Intermediate risk (T2b, or GS = 7, or PSA 10–20) 393 17.9  High risk (≥ T2c, or GS ≥ 8, or PSA > 20) 1,608 73.1  Missing 125 5.6 a West Africa: Hôpital Général de Grand Yoff, Senegal; 37 Military Hospital, Ghana; Korle-Bu Hospi- tal, Ghana; University College Hospital, Nigeria; University of Abuja Teaching Hospital, Nigeria, South Africa: Stellenbosch University and Wits Health Consortium b The distribution of other occupations among 439 cases and 364 controls is as follows: 10.3% and 11.3% were businesses, sales or trade; 9.6% and 8.5% were drivers, 4.1% and 4.7% were security, and 1.8% and 1.9% were pastors, respectively c Based on information extracted from medical records of cases and controls d Comorbidity score is calculated from the sum of 19 diseases for each participant. These diseases include: high blood pressure, malaria, diabetes, high blood cholesterol, rheumatoid arthritis, HIV/AIDS, ulcers, asthma, heart attack, chronic back pain, urinary tract infection, chronic bronchitis, hepatitis, thyroid dis- ease, depression/anxiety, cirrhosis, syphilis, gonorrhea, and herpes. This information was extracted from medical records of participants e First-degree relatives include blood-related father, brothers and/or sons of men who were not adopted: 2,008 cases and 1,848 controls f Defined as moderate-intensity sports, fitness or recreational (leisure) activities that can cause small to moderate increase in breathing or heart rate such as brisk walking, cycling. This information was collected via self-reported questionnaire g PSA in controls is reported only among 704 men who had laboratory serum PSA levels recorded in their medical records h Clinical Stages: cT1 (T1, T1a, T1b, or T1c), cT2 (T2, T2a, T2b, or T2c), cT3 (T3, T3a, T3b, or T3c), and cT4 (T4) i D'Amico Risk Category: low-risk (T1–T2a and GS ≤ 6 and PSA ≤ 10 ng/ml), intermediate-risk (T2b, or GS = 7, or PSA 10–20 ng/ml), and high-risk (≥ T2c, or GS ≥ 8, or PSA > 20 ng/ml) About 7% of all PCa patients did not have clinical data on anthropometric measure and stratification covariates in sep- Gleason score or diagnostic PSA; however, a larger propor- arate logistic regression models containing the main effects; tion (17%) had missing or unknown clinical tumor stage. we used likelihood ratio tests to evaluate the statistical sig- We excluded 125 cases (5.7%) from D’Amico risk classifi- nificance of the interaction terms [33]. All analyses were cation analyses because they had missing data on all three performed using R (v3.6.0) and Stata (StataCorp version clinical parameters. We used polytomous logistic regression 16). [35] to examine associations between anthropometric meas- ures of overall and central obesity and PCa risk stratified by Gleason score/Grade group or D’Amico PCa risk levels as Results described above. These models were adjusted for the same confounders as the main analysis of total PCa risk. A total of 2,200 PCa cases and 1,985 controls were included in our analyses (Table 1). About 51% of PCa cases and con- Stratified analyses We also carried out several stratified trols were recruited in West Africa and 49% in South Africa. analyses by African region (i.e., West vs. South Africa), Most PCa cases (95%) were recruited from urology clinics. and by presence or absence of diabetes, hypertension, heart The majority of controls were recruited from departments attack or hypercholesterolemia or by the number of comor- of ophthalmology (40%), internal medicine (25%) or fam- bid conditions (e.g., 0, 1–2, and 3+), which were abstracted ily medicine (7%), and general/orthopedic surgery (16%). from medical records of all PCa cases and controls to exam- Relative to controls, cases were slightly older (although the ine whether associations between body size measurements distributions of their 5-year age category were similar within and PCa risk varied across different strata. To test for effect each participating hospital, in concordance with the match- modification, we included interaction terms between each ing algorithm), were slightly more educated, and were more 1 3 Cancer Causes & Control likely to have worked in professional or managerial occupa- medicine and ophthalmology, whereas WHR was higher in tions. In comparison to controls, PCa cases were less likely controls from family medicine. to be current smokers (11.5% vs. 15%) or current alcohol Cases with aggressive PCa had lower body weight/BMI drinkers (22.6% vs. 28.4%). Relative to controls, PCa cases compared to controls or to cases with early-stage/low-grade were three times more likely to have a first-degree family PCa, suggesting weight loss related to cancer. For instance, history of PCa (9.6% vs. 3.3%), slightly more likely to have the prevalence of general and central obesity decreased from hypertension (46.1% vs. 41.4%) or three or more comorbidi- 25% to 13.5%, and from 41.6% to 24.1%, respectively, with ties (13% vs 10%), but less likely to have diabetes (13% vs increasing D’Amico PCa risk classification from intermedi- 16%). Among PCa cases the distribution of Gleason score ate to high risk. In addition, 22.6% of PCa cases and 17.6% (GS) was as follows: 16.5% had GS ≤ 6, 23% GS: 7 (3 + 4), of controls reported having lost 5 kg or more in the past five 17.5% GS: 7 (4 + 3), and 35.9% had GS of 8–10 (Table 1). years before recruitment (p < 0.001). A higher proportion The majority (81%) of cases had very high serum PSA lev- of cases with more aggressive PCa pathological features: els at PCa diagnosis: 15.5%, 20.7% and 45.1% had PSA of i.e. GS 8–10 (25.4%), stage CT4 (29%) or those with high- 10–19.9, 20–49.9, and ≥ 50 ng/ml, respectively. Based on the risk PCa (24%) reported the highest weight loss in compari- D’Amico risk classification algorithm, 3.4% and 17.9% of son to patients with low-risk (11.8%) or intermediate-risk PCa cases were classified as low- or intermediate risk, while PCa (15.5%; p < 0.001). Most body size measurements were the majority (73.1%) of patients was high risk. moderately correlated with age and one another; however, Among controls, 30.4% were overweight and 16.4% the correlations with WHtR were generally weaker com- were obese (Fig. 1). The prevalence of central obesity was pared to correlations with other anthropometric factors (see very high; 90% and 70% of controls had WHR > 0.90, and Supplemental Table S2). WHtR > 0.50, respectively. Overall obesity and central obe- Table 2 shows the associations of overall and central sity measures were more common in South than in West obesity measures with risk of total PCa. Overall obesity Africa (Fig. 1). The prevalence of overall and central obe- (BMI ≥ 30), and WC or WHtR were not associated with sity measures varied among controls recruited in different total PCa risk. However, men in the intermediate and high- hospital departments, although their age distribution was est category of WHR had ORs of 0.77 (95% CI 0.66–0.90) similar (see Supplemental Table S1). The prevalence of and 0.68 (95% CI 0.56–0.83), respectively, for total PCa overall obesity (BMI ≥ 30) was the highest among controls compared to men in the lowest category (p trend < 0.001). from ophthalmology (20%), followed by those from internal Although the prevalence of general and central obesity was (17%) and family medicine (15%), but was the lowest among higher in South Africa, the patterns of associations between controls from general surgery (8%). However, the patterns of body size measurements and total PCa risk were similar central obesity were not clearly associated with the depart- between West and South Africa (Table 2). To address the ment of control recruitment; some measures (e.g., WC and issue of undiagnosed PCa among controls, we carried out a WHtR) were higher in controls recruited from internal sensitivity analysis in which we excluded 105 controls with Fig. 1 Prevalence of overall obesity (BMI ≥ 30  kg/m2) and central obesity measures among all controls, as well as stratified by geographic region. Abbreviation: BMI body mass index, WHR waist-to-hip ratio, WHtR waits-to-height ratio 1 3 Cancer Causes & Control 1 3 Table 2 Body size measurements and risk of total prostate cancer among all participating centers and stratified by geographic region (West vs. South Africa) Body size All C entersa,b West Africab,c,d South Africab,e,f measurements Controls Cases OR 95% CI Controls Cases OR 95% CI Controls Cases OR 95% CI n % n % n % n % n % n % BMI (kg/m2)   < 18.5 109 5.9 121 6.5 1.03 0.77–1.36 57 5.7 85 8.5 1.30 0.90–1.88 52 6.0 36 4.2 0.78 0.49–1.25  18.5–24.9 879 47.3 920 49.3 1.00 Ref 529 53.1 572 57.1 1.00 Ref 350 40.6 348 40.2 1.00 Ref  25–29.9 563 30.3 532 28.5 0.88 0.75–1.03 301 30.2 253 25.2 0.80 0.65–1.00 262 30.4 279 32.3 1.01 0.79–1.28   ≥ 30 308 16.6 294 15.7 0.88 0.72–1.08 109 10.9 92 9.2 0.83 0.60–1.15 199 23.1 202 23.4 0.94 0.72–1.23  p for trendg 0.11 0.07 0.73 Waist circumference (WC, cm)h   ≤ 82.5 450 24.2 409 21.8 1.00 Ref 284 28.4 258 25.5 1.00 Ref 166 19.3 151 17.5 1.00 Ref  82.6–90.0 481 25.8 539 28.7 1.19 0.98–1.43 296 29.6 355 35.1 1.29 1.02–1.64 185 21.5 184 21.3 0.94 0.69–1.3  90.1–97.0 456 24.5 422 22.5 0.95 0.78–1.16 269 26.9 233 23.1 0.92 0.71–1.19 187 21.7 189 21.8 0.92 0.67–1.27  97.1–158 475 25.5 505 26.9 1.10 0.90–1.34 151 15.1 164 16.2 1.18 0.88–1.58 324 37.6 341 39.4 0.96 0.71–1.3  p for trend 0.90 0.92 0.86 Waist-to-hip ratio (WHR)   ≤ 0.95 695 37.4 798 43.2 1.00 Ref 363 36.3 410 40.8 1.00 Ref 332 38.6 388 46.1 1.00 Ref  0.96–0.99 774 41.6 716 38.8 0.77 0.66–0.90 478 47.8 443 44.0 0.79 0.64–0.97 296 34.4 273 32.5 0.75 0.60–0.95   ≥ 1 391 21.0 333 18.0 0.68 0.56–0.83 158 15.8 153 15.2 0.80 0.59–1.09 233 27.1 180 21.4 0.60 0.46–0.77  p for trend < 0.001 0.06 < 0.001 Waist-to-height ratio (WHtR)   ≤ 0.54 1032 55.6 1003 53.7 1.00 Ref 658 66.1 656 65.4 1.00 Ref 374 43.4 347 40.2 1.00 Ref  0.55–0.59 433 23.3 435 23.3 0.98 0.83–1.16 218 21.9 212 21.1 0.94 0.75–1.18 215 25.0 223 25.8 1.00 0.77–1.28   > 0.59 392 21.1 429 23.0 1.07 0.89–1.28 120 12.0 135 13.5 1.11 0.84–1.48 272 31.6 294 34.0 1.04 0.82–1.33  p for trend 0.50 0.60 0.74 a Adjusted for seven hospital centers in West and South Africa, age at enrollment (5-year age category), occupational categories, smoking status (i.e., never, former, current), hypertension (yes vs no), and diabetes (yes vs. no) b Controls with PSA ≥ 20 ng/ml and participants with missing values on anthropometric factors or covariates were not included c Adjusted for five centers in West Africa, age at enrollment, occupation, smoking status, hypertension, and diabetes d West Africa: Hôpital Général de Grand Yoff, Senegal; 37 Military Hospital, Ghana; Korle-Bu Hospital, Ghana; University College Hospital, Nigeria; University of Abuja Teaching Hospital, Nigeria e Adjusted for two centers in South Africa, age at enrollment, occupation, smoking status, hypertension, and diabetes f South Africa: Stellenbosch University and Wits Health Consortium g Body mass index (BMI) < 18.5 kg/m2 was excluded from the p for trend analysis h Cutoff points for WC were based on the quartile distribution among controls Cancer Causes & Control serum PSA > 4 ng/ml. Results of this sensitivity analysis Nevertheless, the 95% CI overlapped for most of the strati- were similar to those of the main analysis (see Supplemental fied analyses, and therefore results were not statistically sig- Table S3). nificantly different across various strata. Given that overall and central obesity are associated with Table 3 shows the associations of body size measurements other cardio-metabolic factors, we also conducted several with risk of PCa stratified by Gleason score (GS) among all stratified analyses by presence or absence of diabetes, either participating centers. Although, general obesity/BMI was hypertension, heart attack or hypercholesterolemia, or by not associated with low-grade PCa (GS ≤ 6), the highest the number of comorbid conditions (e.g., 0, 1–2, 3+; see categories of WC and WHtR were positive associated with Supplemental Table S4). Although, in general, patterns of modest increased risk of GS ≤ 6. With regard to GS 7 = 3 + 4 associations of body size measures with total PCa risk were PCa, both general obesity (OR= 1.23; 95% CI: 0.91–1.68), similar, overall obesity (BMI ≥ 30) was inversely associated and the highest categories of WC (OR= 1.48; 95% CI: with PCa risk only among men without hypertension, heart 1.06–2.05), and WHtR (OR = 1.44; 95% CI: 1.09–1.90) attack or high cholesterol (OR = 0.62; 95% CI 0.44–0.88), were positively associated with risk of cancer (Table 3). or among those without any comorbidities (OR = 0.65; 95% By contrast, the associations of general and central obe- CI 0.45–0.94), but not among men with these comorbidities sity measures, except for WHR, with GS 7 = 4 + 3 prostate (Supplemental Table S4). Interestingly, among men with tumors were inverse or null. Overall obesity (OR= 0.70, 95% three or more comorbidities, overall obesity was associ- CI 0.53–0.92), and the highest categories of WHR (OR= ated with a twofold higher risk of PCa (OR = 2.18; 95% 0.66, 95% CI 0.51–0.86), and WHtR (OR = 0.78, 95% CI CI 1.18–4.03, p trend = 0.02), as was the highest category 0.61–1.00) were also inversely associated with high-grade of WHtR (OR= 2.10; 95% CI 1.19–3.70, p trend = 0.01). PCa (i.e., GS 8–10). There was consistent inverse association Table 3 Body size measurements and risk of prostate cancer stratified by Gleason score/grade group (GG) among all participating centers Body size Controls Gleason score < 6/GG 1 Gleason score 7 (3 + 4)/ Gleason score 7 (4 + 3)/ Gleason score 8–10/GG 4 measurements GG 2 GG 3 or 5 Cases ORa,b 95% CI Cases ORa,b 95% CI Cases ORa,b 95% CI Cases ORa,b 95% CI BMI (kg/m2)   < 18.5 109 21 1.03 0.61–1.73 27 1.14 0.72–1.82 15 0.68 0.38–1.21 55 1.11 0.78–1.60  18.5–24.9 879 159 1.00 Ref 191 1.00 Ref 174 1.00 Ref 377 1.00 Ref  25–29.9 563 103 1.02 0.77–1.36 141 1.12 0.87–1.45 100 0.87 0.65–1.15 176 0.71 0.57–0.89   ≥ 30 308 51 1.07 0.74–1.56 93 1.23 0.91–1.68 52 0.77 0.53–1.11 96 0.70 0.53–0.92  p for trend 0.75 0.23 0.16 < 0.001 Waist circumference (WC, cm)c   ≤ 82.5 450 65 1.00 Ref 81 1.00 Ref 66 1.00 Ref 187 1.00 Ref  82.6–90.0 481 98 1.37 0.96–1.95 109 1.13 0.82–1.57 102 1.31 0.93–1.86 217 1.07 0.84–1.37  90.1–97.0 456 87 1.26 0.87–1.81 107 1.16 0.83–1.61 76 0.99 0.69–1.44 144 0.72 0.55–0.94  97.1–158 475 84 1.54 1.05–2.25 156 1.48 1.06–2.05 97 1.14 0.78–1.65 163 0.78 0.59–1.02  p for trend 0.06 0.02 0.96 0.01 Waist-to-hip ratio (WHR)   ≤ 0.95 695 137 1.00 Ref 181 1.00 Ref 147 1.00 Ref 317 1.00 Ref  0.96–0.99 774 143 0.88 0.67–1.17 167 0.81 0.63–1.04 127 0.72 0.55–0.95 264 0.74 0.60–0.91   ≥ 1 391 51 0.69 0.47–1.01 94 0.81 0.60–1.10 61 0.57 0.40–0.81 123 0.66 0.51–0.86  p for trend 0.07 0.12 0.001 0.001 Waist-to-height ratio (WHtR)   ≤ 0.54 1,032 175 1.00 Ref 209 1.00 Ref 178 1.00 Ref 416 1.00 Ref  0.54–0.59 433 86 1.28 0.95–1.73 108 1.11 0.85–1.47 81 0.99 0.73–1.34 153 0.80 0.64–1.01   > 0.59 392 72 1.47 1.05–2.05 135 1.44 1.09–1.90 82 1.04 0.75–1.43 136 0.78 0.61–1.00  p for trend 0.02 0.01 0.84 0.03 a Adjusted for seven hospital centers in West and South Africa, age at enrollment (5-year groups), occupation, smoking status, hypertension and diabetes b Controls with PSA ≥ 20 ng/ml and all participants with missing values were  excluded; c Cutoff points were based on the quartile distribution among controls 1 3 Cancer Causes & Control Table 4 Body size measurements and risk of prostate cancer stratified by Gleason score/grade group (GG) in West Africa Body size Controls Gleason score < 6/GG 1 Gleason score 7 (3 + 4)/ Gleason score 7 (4 + 3)/ Gleason score 8–10/GG measure- GG 2 GG 3 4 or 5 ments Cases ORa,b 95% CI Cases ORa,b 95% CI Cases ORa,b 95% CI Cases ORa,b 95% CI BMI (kg/m2)   < 18.5 57 18 1.22 0.67–2.22 18 1.44 0.79–2.62 12 1.11 0.56–2.2 35 1.29 0.81–2.07  18.5–24.9 529 122 1.00 Ref 111 1.00 Ref 100 1.00 Ref 228 1.00 Ref  25–29.9 301 79 1.15 0.82–1.61 55 0.99 0.68–1.44 36 0.74 0.48–1.14 82 0.65 0.48–0.89   ≥ 30 109 26 0.98 0.59–1.62 19 0.94 0.53–1.64 9 0.55 0.26–1.17 37 0.81 0.52–1.24  p for trend 0.78 0.76 0.08 0.06 Waist circumference (WC, cm)c    ≤ 82.5 284 51 1.00 Ref 48 1.00 Ref 38 1.00 Ref 114 1.00 Ref  82.6–90.0 296 77 1.39 0.92–2.09 65 1.20 0.78–1.84 61 1.48 0.94–2.35 146 1.27 0.93–1.73  90.1–97.0 269 67 1.26 0.83–1.92 51 1.10 0.70–1.73 36 0.97 0.58–1.62 77 0.73 0.51–1.03  97.1–158 151 51 1.76 1.11–2.79 40 1.60 0.97–2.62 22 1.20 0.66–2.17 51 0.84 0.56–1.27  p for trend 0.04 0.12 0.98 0.07 Waist-to-hip ratio (WHR)   ≤ 0.95 363 99 1.00 Ref 73 1.00 Ref 66 1.00 Ref 165 1.00 Ref  0.96–0.99 478 110 0.88 0.62–1.23 99 1.03 0.71–1.48 64 0.60 0.39–0.91 164 0.77 0.58–1.02   ≥ 1 158 36 0.73 0.44–1.20 32 0.99 0.57–1.72 27 0.64 0.34–1.18 56 0.76 0.50–1.14  p for trend 0.21 0.98 0.07 0.09 Waist-to-height ratio (WHtR)   ≤ 0.54 658 142 1.00 Ref 129 1.00 Ref 106 1.00 Ref 266 1.00 Ref  0.54–0.59 218 64 1.31 0.92–1.87 42 0.96 0.64–1.44 31 0.90 0.57–1.42 74 0.80 0.59–1.10   > 0.59 120 39 1.47 0.95–2.28 32 1.47 0.92–2.36 20 1.24 0.71–2.16 43 0.81 0.54–1.21  p for trend 0.05 0.19 0.65 0.16 West Africa: Hôpital Général de Grand Yoff, Senegal; 37 Military Hospital, Ghana; Korle-Bu Hospital, Ghana; University College Hospital,Nigeria; University of Abuja Teaching Hospital, Nigeria a Adjusted for five hospital centers in West Africa, age at enrollment (5-year category), occupation, smoking status, hypertension and diabetes b Controls with PSA ≥ 20 ng/ml and participants with missing values were  excluded c Cutoff points were based on the quartile distribution among controls of WHR with risk of PCa across all Gleason score, with 95% CI 0.60–0.92) were also inversely associated high-risk the highest category (WHR ≥ 1) showing statistically sig- PCa, but not with other measures of central obesity (WC or nificant ORs of 0.57–0.81 across GS categories. Patterns of WHtR; Table 6). Although some associations of overall and associations between anthropometric factors and PCa risk central obesity with intermediate-risk PCa were stronger in stratified by Gleason score were generally similar between West Africa compared to South Africa (Table 7), results West Africa (Table 4) and South Africa (Table 5). were not statistically significantly different. Overall and central obesity were also associated with D’Amico PCa risk groups (Tables 6, 7). Since only 74 cases (3.4%) were classified as low risk, associations for this group Discussion are not presented. The associations of overall obesity (OR = 1.38, 95% CI 0.99–1.93) and several central obesity meas- In this large multi-center, hospital-based case–control study ures: e.g., WC > 97 cm (OR = 1.60, 95% CI 1.10–2.33), of urban African men, we found that half of the study sub- or WHtR > 0.59 (OR  = 1.68, 95% CI 1.24–2.29) with jects were overweight or obese, and 90% of them had cen- intermediate-risk PCa were consistently positive (Table 6). tral obesity as defined by WHR > 0.90. Despite evidence However, intermediate-risk PCa was inversely associated of higher weight loss (≥ 5 kg before recruitment) among with WHR (OR = 0.56; 95% CI 0.39–0.80 when comparing PCa cases with more aggressive PCa (i.e., those with a GS WHR ≥ 1 vs. ≤ 0.95). Overall obesity (OR = 0.77, 95% CI 8–10 or D’Amico high-risk group), we found that several 0.61–0.95), and the highest category of WHR (OR = 0.74, parameters of overall and central obesity were statistically 1 3 Cancer Causes & Control Table 5 Body size measurements and risk of prostate cancer stratified by Gleason score/grade group (GG) in South Africa Body size measurements Controls Gleason score < 6/GG 1 Gleason score 7 (3 + 4)/ Gleason score 7 (4 + 3)/ Gleason score 8–10/GG GG 2 GG 3 4 or 5 Cases ORa,b 95% CI Cases ORa,b 95% CI Cases ORa,b 95% CI Cases ORa,b 95% CI BMI (kg/m2)   < 18.5 52 3 0.58 0.17–1.97 9 0.83 0.38–1.81 3 0.29 0.09–0.97 20 1.04 0.58–1.88  18.5–24.9 350 37 1.00 Ref 80 1.00 Ref 74 1.00 Ref 149 1.00 Ref  25–29.9 262 24 0.80 0.46–1.40 86 1.29 0.90–1.86 64 1.09 0.74–1.62 94 0.81 0.58–1.12   ≥ 30 199 25 1.11 0.62–1.97 74 1.41 0.96–2.08 43 0.98 0.63–1.53 59 0.64 0.44–0.94  p for trend 0.81 0.07 0.99 0.02 Waist circumference (WC, cm)c    ≤ 82.5 166 14 1.00 Ref 33 1.00 Ref 28 1.00 Ref 73 1.00 Ref  82.6–90.0 185 21 1.33 0.64–2.76 44 1.02 0.61–1.73 41 1.08 0.62–1.87 71 0.69 0.46–1.06  90.1–97.0 187 20 1.24 0.59–2.61 56 1.28 0.77–2.13 40 0.99 0.57–1.74 67 0.62 0.41–0.95  97.1–158 324 33 1.15 0.56–2.36 116 1.38 0.86–2.23 75 1.12 0.66–1.88 112 0.63 0.43–0.94  p for trend 0.89 0.10 0.73 0.04 Waist-to-hip ratio (WHR)   ≤ 0.95 332 38 1.00 Ref 108 1.00 Ref 81 1.00 Ref 152 1.00 Ref  0.96–0.99 296 33 0.91 0.55–1.50 68 0.66 0.46–0.95 63 0.82 0.56–1.21 100 0.71 0.52–0.97   ≥ 1 233 15 0.54 0.29–1.03 62 0.69 0.48–1.01 34 0.51 0.32–0.81 67 0.58 0.41–0.82  p for trend 0.08 0.04 0.01 < 0.001 Waist-to-height ratio (WHtR)   ≤ 0.54 374 33 1.00 Ref 80 1.00 Ref 72 1.00 Ref 150 1.00 Ref  0.54–0.59 215 22 1.17 0.65–2.12 66 1.29 0.87–1.9 50 1.08 0.71–1.65 79 0.79 0.56–1.11   > 0.59 272 33 1.35 0.78–2.35 103 1.45 1.01–2.08 62 1.05 0.70–1.58 93 0.76 0.54–1.06  Wald test for trend 0.28 0.05 0.81 0.09 South African centers included: Stellenbosch University and the Wits Health Consortium a Adjusted for hospital centers in South Africa, age at enrollment (5-year group), occupation, smoking status, hypertension and diabetes b Controls with PSA ≥ 20 ng/ml and all participants with missing values were excluded c Cutoff points were based on the quartile distribution among controls significantly associated with 23%–68% higher odds of GS Africans [37]. Although the prevalence of obesity in African 7 = 3 + 4 PCa or D’Amico intermediate-risk category, but American men is reported to be over 41% [38], the average there was no association with total PCa risk. Although the BMI among men in all African regions has increased stead- prevalence of general and central obesity were higher in ily in the past 25 years [24]. South vs. West Africa in our data, which was similar to other Central obesity measurements were highly prevalent reports [24], the associations between body size measures among controls (ranging from 44% to 90%), and were and risks of overall PCa and by Gleason score or D’Amico consistently high across all seven participating centers in risk score did not differ much by African region. West and South Africa. In recent years, the reported preva- The high prevalence of general obesity and several meas- lence of central obesity has been alarmingly high in Afri- ures of central obesity in African men in our study (pre- can countries. For example, in a study of Ghanaian adults sented in Fig. 1) is concerning and underscores the impor- aged 50 years or older, the prevalence of abdominal obesity tance of obesity prevention in Africa. However, to be noted among men was 54.4% [23]. Similarly, among South African is that our controls were urban men hospitalized for other men, the prevalence of central obesity has been reported to conditions including hypertension and cardiovascular dis- be between 36% and 54% [32]. Abdominal fat, especially eases, and therefore their prevalence of obesity (16%) might visceral fat, is metabolically more active, and poses higher be higher compared to population-based controls or men risk than other fat for many cancers, including PCa [14, 15], in the rural areas [23, 36]. In 2016, the WHO reported that as well as other chronic conditions, including cardiovas- the prevalence of general obesity ranged from 2.5% to 6.6% cular disease, hypertension, and diabetes. Reasons for the in West Africa men, but was almost 31% in black South extremely high prevalence of abdominal obesity in urban 1 3 Cancer Causes & Control Table 6 Body size Body size measurements Controls D’Amico Risk Group measurements and risk of prostate cancer according to Intermediate r iska  High riska D’Amico risk classification b among all participating centers Cases OR 95% CI Cases OR b 95% CI BMI (kg/m2)   < 18.5 109 10 0.53 0.27–1.07 109 1.11 0.83–1.49  18.5–24.9 879 129 1.00 Ref 763 1.00 Ref  25–29.9 563 121 1.28 0.96–1.71 391 0.82 0.69–0.97   ≥ 30 308 85 1.38 0.99–1.93 195 0.77 0.61–0.95  p for trendh 0.05 0.01 Waist circumference (WC, cm)c   ≤ 82.5 450 55 1.00 Ref 343 1.00 Ref  82.6–90.0 481 70 1.04 0.7–1.54 456 1.21 0.99–1.47  90.1–97.0 456 76 1.13 0.76–1.67 327 0.90 0.73–1.11  97.1–158 475 144 1.60 1.10–2.33 341 0.98 0.78–1.22  p for trend 0.01 0.25 Waist-to-hip ratio (WHR)   ≤ 0.95 695 153 1.00 Ref 615 1.00 Ref  0.96–0.99 774 122 0.79 0.60–1.04 567 0.77 0.66–0.91   ≥ 1 391 58 0.56 0.39–0.80 269 0.74 0.60–0.92  p for trend 0.001 0.002 Waist-to-height ratio (WHtR)   ≤ 0.54 1032 129 1.00 Ref 843 1.00 Ref  0.54–0.59 433 94 1.40 1.03–1.90 325 0.90 0.75–1.08   > 0.59 392 122 1.68 1.24–2.29 291 0.95 0.78–1.15  p for trend < 0.001 0.40 a D'Amico risk category: intermediate-risk (T2b, GS = 7, PSA 10–20) and high-risk (≥ T2c, GS 8–10, or PSA > 20); n = 74 cases with low-risk (T0–T2a and GS ≤ 6 and PSA ≤ 10) were excluded from stratification analysis b Multinomial logistic regression models were adjusted for age at enrollment (5-year group), hospital cent- ers, occupation, smoking status, hypertension, and diabetes c Cutoff points were based on the quartile distribution among controls African men might be related to genetics or increased west- loss in comparison to low-risk cases (11.8%) or to PCa ernization and lifestyle changes [24]. patients with intermediate-risk (15.5%; p < 0.001), suggest- The relatively consistent associations of body size meas- ing weight loss/cachexia related to cancer progression (dura- ures with intermediate-risk PCa suggest a potential link tion of PCa) that is consistent with reverse causation, rather between general and central obesity with PCa in African than an effect of body weight/size on disease risk. men that warrants further investigation. The less consistent Although PCa is the most common cancer in men in most findings for low-risk and high-risk GS are not completely African countries, and obesity rates are rising in Africa [23, surprising. In this population with little PCa screening (rela- 24], few studies of body size and PCa risk have focused tive to the US), very few cases had low-risk PCa (D’Amico on African or Afro-Caribbean men. A recently published low risk n = 74); thus, the analyses among low-risk group study in Ghana, which included 566 PCa cases and 964 were underpowered. Analyses of the high-risk groups (GS controls reported a 1.9-fold increased risk of PCa (95% CI of 8–10: n = 778, D’Amico high risk: n = 1,590) were not 1.1–3.1) among men associated with general obesity and underpowered, but were likely to have been affected by the a 1.8-fold increased risk associated with larger waist cir- presence of cancer, exemplifying reverse causation. The cumference (95% CI 1.2–2.5) [39]. In this study, most cases prevalence of general obesity was twice as high among (87%) were recruited from the Korle-Bu Teaching Hospital D’Amico intermediate-risk cases (25%) as among high-risk in Ghana (one of the centers of this MADCaP consortium cases (13.5%). Moreover, a higher proportion of PCa cases study; although none of the cases reported in that earlier with GS 8–10 (25.4%), advanced stage T4 cancer (29%) or study were included in the present analyses), but the con- those with high-risk PCa (24%) reported the highest weight trols were drawn from a population-based sample of 1,037 1 3 Cancer Causes & Control 1 3 Table 7 Body size measurements and risk of prostate cancer according to D’Amico risk classification stratified by geographic region (West vs. South Africa) Body size measurements West Africa South Africa Controls D’Amico intermediate riska D’Amico high riska Controls D’Amico intermediate riska D’Amico high riska Cases ORb 95% CI Cases ORb 95% CI Cases ORb 95% CI Cases ORb 95% CI BMI (kg/m2)   < 18.5 57 2 0.37 0.08–1.64 82 1.38 0.95 – 2.00 52 8 0.64 0.29–1.43 27 0.85 0.50–1.43  18.5–24.9 529 44 1.00 ref 514 1.00 ref 350 85 1.00 ref 249 1.00 ref  25–29.9 301 31 1.28 0.77–2.12 213 0.77 0.61–0.96 262 90 1.28 0.90–1.83 178 0.92 0.70–1.21   ≥ 30 109 13 1.66 0.81–3.40 76 0.78 0.56–1.09 199 72 1.36 0.93–2.00 119 0.78 0.57–1.06  p for trendh 0.16 0.03 0.10 0.12 Waist circumference (WC, cm)c   ≤ 82.5 284 19 1.00 ref 236 1.00 ref 166 36 1.00 ref 107 1.00 ref  82.6–90.0 296 21 0.93 0.47–1.81 327 1.33 1.04–1.70 185 49 1.05 0.64–1.73 129 0.90 0.63–1.28  90.1–97.0 269 18 0.84 0.42–1.69 203 0.89 0.69–1.17 187 58 1.22 0.75–2.00 124 0.81 0.56–1.16  97.1–158 151 32 2.71 1.41–5.21 127 1.03 0.76–1.41 324 112 1.31 0.82–2.07 214 0.85 0.61–1.19  p for trend 0.01 0.46 0.18 0.33 Waist-to-hip ratio (WHR)   ≤ 0.95 363 36 1.00 ref 362 1.00 ref 332 117 1.00 ref 253 1.00 ref  0.96–0.99 478 42 1.05 0.63–1.76 388 0.78 0.62–0.97 296 80 0.74 0.53–1.03 179 0.76 0.59–0.99   ≥ 1 158 12 1.09 0.48–2.50 139 0.81 0.59–1.12 233 46 0.49 0.33–0.72 130 0.67 0.50–0.90  p for trend 0.80 0.08 < 0.001 0.01 Waist-to-height ratio (WHtR)   ≤ 0.54 658 42 1.00 ref 598 1.00 ref 374 87 1.00 ref 245 1.00 ref  0.54–0.59 218 24 1.58 0.91–2.75 179 0.88 0.69–1.12 215 70 1.28 0.88–1.87 146 0.90 0.68–1.20   > 0.59 120 24 3.38 1.85–6.17 109 1.00 0.74–1.35 272 98 1.36 0.95–1.95 182 0.91 0.69–1.20  p for trend < 0.001 0.69 0.09 0.50 a D'Amico category: intermediate-risk (T2b, GS = 7, PSA 10–20) and high-risk (≥ T2c, GS 8–10, or PSA > 20). n = 74 cases with low-risk (T0–T2a and GS ≤ 6 and PSA ≤ 10) were excluded from the stratification analysis b Multinomial logistic regression models were adjusted for age at enrollment (5-year group), hospital centers in either West or South Africa, occupation, smoking status, hypertension, and diabe- tes c Cutoff points were based on the quartile distribution among controls Cancer Causes & Control men recruited for a PCa screening study [39]. In the ear- were selected from the same hospitals, differential selec- lier Ghana study, the prevalence of general obesity in the tion bias was probably minimal. Since PSA screening same catchment population was much lower (13% of cases is seldom used in Africa, all PCa cases were clinically and 9% of controls) than in the current study (43% of cases diagnosed (not PSA screened), and all controls were also and 25% of controls), since the earlier study was conducted not screened via a serum PSA test. The use of hospital 16 years ago when obesity was emerging as a problem there. controls, who are usually more ill than population-based It is reassuring that in both the earlier and current studies controls, might have also affected the direction or strength in Ghana, the prevalence of general obesity in cases is higher of the associations. To minimize this bias, we selected than that in controls. In a separate study in Barbados, West hospital controls primarily from departments with less Indies, several measures of central obesity were associated apparently serious conditions, including Ophthalmol- with increased risk of PCa: WHR ≥ 0.96 versus < 0.87 with ogy (40%), Internal and Family Medicine (32%), and OR = 2.11 (95% CI, 1.54–2.88) and waist size ≥ 99 cm with Orthopedics (15%). It should be noted that some control an OR = 1.84 (95% CI 1.19–2.85) [20]. subjects, especially those recruited in internal medicine, The few studies that have evaluated the relationship of may have been hospitalized because of diabetes, hyper- general obesity and PCa in African American men have tension, or other cardiovascular disease, related to higher yielded conflicting results [40–42]. A case–control study[40] BMI/obesity, which could potentially have affected our among African American men in Maryland reported inverse results. Although results of several stratified analyses did associations between obesity (BMI > 30) and risks of non- not reveal statistically significant differences in associa- aggressive (OR = 0.62) or aggressive PCa (OR= 0.41). tions of body size measures with PCa risk across strata The North Carolina/Louisiana prostate cancer (PCaP) pro- of comorbidities, some of the obesity-related conditions ject that included 991 African American cases reported no among controls could have potentially underestimated association between obesity and aggressive PCa (OR = the ORs for those associations. As noted earlier, we used 1.09; 95% CI 0.71, 1.67), although the comparison group standardized procedures and protocols at all centers but in this study were non-aggressive PCa cases, and not con- had to make adjustments at each center based on the needs trols [43]. Similarly, the Multiethnic Cohort Study, which of clinical care locally. These variations may have had a included 9,284 African American men, reported no asso- slight impact on the completeness of tumor staging and ciation between obesity and overall PCa risk (RR = 1.05, grading of PCa patients. Finally, our results are not gener- 95%CI 0.81–1.36 for BMI ≥ 35 vs. < 25 kg/m2)[44]. By con- alizable to African population living outside Africa, given trast, among African American men who participated in the differences in screening patterns, migration or changes in SELECT trial[41], BMI was positively associated with total dietary patterns. PCa risk [BMI ≥ 35 vs. < 25 kg/m2: hazard ratio (HR) 1.49; 95% CI 0.95–2.34, p for trend = 0.03]. Our study has several strengths. It is the first to exam- ine associations of body size measurements with risks Conclusion of total and aggressive PCa in African men, with a large sample size and patients recruited from seven clinical In conclusion, in this large multi-center case–control study centers in four countries in West and South Africa. The of African men, we found that general obesity and several study used standardized protocols across all participat- measures of central adiposity (e.g., waist size and WHtR) ing centers collecting high-quality detailed information were positively associated with intermediate-risk PCa. on demographic, social and lifestyle factors, as well as Given the high prevalence of general and central obesity in anthropometric measures, and abstracted relevant clini- our study population, and their rising prevalence in Africa, cal information on PCa and comorbidities from medical large cohort studies are needed to better clarify the role records. Only a small percentage of data (median of 5%) of obesity and PCa in various African populations. Our were missing. Anthropometric factors were measured dur- results support policies that target a potentially modifiable ing in-person interviews of both cases and controls. How- risk factor for many diseases including PCa, in order to ever, although the procedures were standardized across improve public health in Africa. centers, and the field teams used the same protocols for all patients, body size and shape at diagnosis could have been Supplementary Information The online version contains supplemen- affected by cachexia, among PCa patients with advanced tary material available at https://d oi.o rg/1 0.1 007/s 10552-0 21-0 1515-0. stage or high-grade cancer. Selection and referral bias are Acknowledgments This work was supported by Public Health Service also possible, because all clinical centers included in the (PHS) Grant U01-CA184374 from the U.S. National Cancer Institute study were tertiary-care hospitals. However, since most (NCI), National Institute of Health (NIH). We thank study participants cancers are treated in tertiary hospital centers and controls as well as research and clinical staff of the participating hospitals in 1 3 Cancer Causes & Control Senegal (Hôpital Général de Grand Yoff, Dakar), Ghana (Korle-Bu aggressiveness, and mortality in men of African descent. Prostate Teaching Hospital/University of Ghana, and 37 Military Hospital, Cancer 2013:560857. https://d oi.o rg/ 10. 1155/ 2013/ 560857 both in Accra), Nigeria (University College Hospital/University of 4. Bray FCM, Mery L, Piñeros M, Znaor A, Zanetti R, Ferlay J Ibadan, Ibadan, and University of Abuja Teaching Hospital/University (2017) Cancer incidence in five continents. Interational Agency of Abuja, Abuja), and South Africa (Tygerberg Hospital/Stellenbosch for Research on Cancer, Lyon University, Cape Town, and the Chris Hani Baragwanath Academic 5. Hsing AW, Yeboah E, Biritwum R, Tettey Y, De Marzo AM, Adjei Hospital/University of the Witwatersrand (Wits) and Wits Health Con- A et al (2014) High prevalence of screen detected prostate cancer sortium, Johannesburg). The four twinning centers in the United States in West Africans: implications for racial disparity of prostate can- were: Albert Einstein College of Medicine (Bronx, New York), Colum- cer. J Urol 192(3):730–735. https:// doi. org/ 10. 1016/j. juro. 2014. bia University Irving Medical Center (New York, New York), Dana- 04. 017 Farber Cancer Institute (DFCI, Boston, Massachusetts); and Stanford 6. Chu LW, Ritchey J, Devesa SS, Quraishi SM, Zhang H, Hsing AW Cancer Institute, Stanford University (Stanford, California). We also (2011) Prostate cancer incidence rates in Africa. Prostate Cancer thank Dana-Farber Cancer Institute/Harvard Cancer Center, for the use 2011:947870. https:// doi. org/ 10. 1155/ 2011/ 947870 of the Survey and Data Management Core, which provided database 7. Global Cancer Observatory (2018) Cancer tomorrow. Interna- services and support for this project; these centers were supported in tional Agency for Research on Cancer, Lyon part by an NCI Cancer Center Support Grant (P30 CA06516). 8. Bruner DW, Moore D, Parlanti A, Dorgan J, Engstrom P (2003) Relative risk of prostate cancer for men with affected relatives: Author contributions IA, AOA, AWH had full access to all the data systematic review and meta-analysis. Int J Cancer 107(5):797– in the study and take responsibility for the integrity of the data and the 803. https:// doi. org/ 10.1 002/ ijc.1 1466 accuracy of the data analysis. Concept and design: IA, AWH, AOA, 9. Hsing AW, Chokkalingam AP (2006) Prostate cancer epidemiol- TRR, JSJ, TER. Data acquisition and management: All authors. Sta- ogy. Front Biosci 11:1388–1413. https://d oi.o rg/1 0. 2741/ 1891 tistical analysis: WKL, IA, AWH. Interpretation of data: All authors. 1 0. Zhang X, Zhou G, Sun B, Zhao G, Liu D, Sun J et al (2015) Drafting of manuscript: IA, WKL, JSZ, AOA, AWH, JSJ, TER. Critical Impact of obesity upon prostate cancer-associated mortality: a revision of manuscript for important intellectual content: All authors. meta-analysis of 17 cohort studies. Oncol Lett 9(3):1307–1312. Supervision: IA, AWH, AOA, TRR. https:// doi. org/1 0. 3892/ ol. 2014. 2841 1 1. Fang X, Wei J, He X, Lian J, Han D, An P et al (2018) Quantita- Supported by Grant U01-CA184374 from the U.S. National tive association between body mass index and the risk of cancer: Funding Cancer Institute, NIH. a global meta-analysis of prospective cohort studies. Int J Cancer 143(7):1595–1603. https://d oi.o rg/ 10.1 002/i jc. 31553 12. Cao Y, Ma J (2011) Body mass index, prostate cancer-specific Data availability The data that support the findings of this study are mortality, and biochemical recurrence: a systematic review and available upon request from the corresponding authors or the Principal meta-analysis. Cancer Prev Res 4(4):486–501. https://d oi.o rg/1 0. Investigator of the MADCaP Network. The data are not publicly avail- 1158/1 940- 6207 able due to privacy or ethical restrictions. Requests for data access 13. Genkinger JM, Wu K, Wang M, Albanes D, Black A, van den can be submitted via the MADCaP Network website at: https:// www. Brandt PA et al (2020) Measures of body fatness and height in madcap netw ork. org/. early and mid-to-late adulthood and prostate cancer: risk and mor- tality in the pooling project of prospective studies of diet and can- Code availability R coding and Stata programs that were used for data cer. Ann Oncol 31(1):103–114. https://d oi.o rg/1 0.1 016/j.a nnonc. analyses are available upon request from the corresponding authors. 2019.0 9. 007 14. Lee MJ, Wu Y, Fried SK (2013) Adipose tissue heterogeneity: Declarations implication of depot differences in adipose tissue for obesity com-plications. Mol Asp Med 34(1):1–11. https:// doi. org/ 10. 1016/j. mam.2 012. 10. 001 Conflict of interest The authors declare that they have no conflict of 15. Dickerman BA, Torfadottir JE, Valdimarsdottir UA, Giovannucci interest. E, Wilson KM, Aspelund T et al (2019) Body fat distribution on computed tomography imaging and prostate cancer risk and mor- Ethical approval The study protocol and procedures were approved tality in the AGES-Reykjavik study. Cancer 125(16):2877–2885. by the Institutional Ethical Review Boards (IRBs) of all participating https:// doi. org/ 10. 1002/c ncr. 32167 institutions. 1 6. Boehm K, Sun M, Larcher A, Blanc-Lapierre A, Schiffmann J, Graefen M et al (2015) Waist circumference, waist-hip ratio, Consent to participate All cases and controls provided written body mass index, and prostate cancer risk: results from the North- informed consent to participate in the study. American case–control study prostate cancer & environment study. Urol Oncol 33(11):494. https:// doi. org/ 10. 1016/j. urolo nc. 2015. 07.0 06 17. Guerrios-Rivera L, Howard L, Frank J, De Hoedt A, Beverly D, References Grant DJ et al (2017) Is body mass index the best adiposity meas- ure for prostate cancer risk? Results from a veterans affairs biopsy 1. Bray F, Ferlay J, Soerjomataram I, Siegel RL, Torre LA, Jemal A cohort. Urology 105:129–135. https:// doi. org/1 0. 1016/j. urolog y. (2018) Global cancer statistics 2018: GLOBOCAN estimates of 2017.0 3. 042 incidence and mortality worldwide for 36 cancers in 185 coun- 18. Hsing AW, Deng J, Sesterhenn IA, Mostofi FK, Stanczyk FZ, Ben- tries. CA 68(6):394–424. https:// doi.o rg/ 10. 3322/c aac. 21492 ichou J et al (2000) Body size and prostate cancer: a population- 2. Wild CP, Weiderpass E, Stewart BW (2020) World cancer report: based case-control study in China. Cancer Epidemiol Biomark cancer research for cancer prevention. International Agency for Prev 9(12):1335–1341 Research on Cancer, Lyon 19. Lavalette C, Tretarre B, Rebillard X, Lamy PJ, Cenee S, Men- 3. Rebbeck TR, Devesa SS, Chang BL, Bunker CH, Cheng I, Cooney egaux F (2018) Abdominal obesity and prostate cancer risk: K et  al (2013) Global patterns of prostate cancer incidence, 1 3 Cancer Causes & Control epidemiological evidence from the EPICAP study. Oncotarget 33. Klienbaum DG, Nizam A, Kupper L, Muller KE (2007) Applied 9(77):34485–34494. https:// doi.o rg/ 10. 18632/o ncot arget. 26128 regression analysis and multivariate methods, 4th edn. Duxbury 2 0. Nemesure B, Wu SY, Hennis A, Leske MC (2012) mCentral adi- Press, Pacific Grove posity and Prostate Cancer in a Black Population. Cancer Epi- 34. D’Amico AV, Whittington R, Malkowicz SB, Schultz D, Blank demiol Biomark Prev 21(5):851–858. https:// doi. org/ 10. 1158/ K, Broderick GA et al (1998) Biochemical outcome after radi- 1055- 9965 cal prostatectomy, external beam radiation therapy, or interstitial 21. Krakauer NY, Krakauer JC (2012) (2012) A new body shape index radiation therapy for clinically localized prostate cancer. JAMA predicts mortality hazard independently of body mass index. PLoS 280(11):969–974. https:// doi. org/1 0. 1001/ jama. 280. 11.9 69 ONE 7(7):e39504. https://d oi.o rg/1 0.1 371/j ourna l.p one.0 03950 4 35. Dubin N, Pasternack BS (1986) Risk assessment for case-control 22. Harding JL, Shaw JE, Anstey KJ, Adams R, Balkau B, Brennan- subgroups by polychotomous logistic regression. Am J Epidemiol Olsen SL et al (2015) Comparison of anthropometric measures 123(6):1101–1117 as predictors of cancer incidence: a pooled collaborative analysis 36. Adeboye B, Bermano G, Rolland C (2012) Obesity and its of 11 Australian cohorts. Int J Cancer 137(7):1699–1708. https:// health impact in Africa: a systematic review. Cardiovasc J Afr doi. org/1 0.1 002/ ijc.2 9529 23(9):512–521. https://d oi.o rg/ 10. 5830/ cvja- 2012- 040 23. Lartey ST, Magnussen CG, Si L, Boateng GO, de Graaff B, Birit- 37. Cois A, Day C (2015) Obesity trends and risk factors in the South wum RB et al (2019) Rapidly increasing prevalence of overweight African adult population. BMC Obes 2:42. https:// doi. org/ 10. and obesity in older Ghanaian adults from 2007–2015: evidence 1186/ s40608-0 15-0 072-2 from WHO-SAGE waves 1 & 2. PLoS ONE 14(8):e0215045. 3 8. Hales CM, Carroll MD, Fryar CD, Ogden CL (2020) Prevalence https://d oi. org/ 10.1 371/ journ al. pone.0 21504 5 of obesity and severe obesity among adults: United States, 2017– 2 4. NCD Risk Factor Collaboration (NCD-RisC) – Africa Working 2018. NCHS Data Brief 360:1–8 Group (2017) Trends in obesity and diabetes across Africa from 3 9. Hurwitz LM, Yeboah ED, Biritwum RB, Tettey Y, Adjei AA, 1980 to 2014: an analysis of pooled population-based studies. Int Mensah JE et al (2020) Overall and abdominal obesity and pros- J Epidemiol 46(5):1421–1432. https://d oi.o rg/1 0.1 093/i je/d yx078 tate cancer risk in a West African population: an analysis of the 2 5. Andrews C, Fortier B, Hayward A, Lederman R, Petersen L, Ghana prostate study. Int J Cancer. https:// doi. org/ 10. 1002/ ijc. McBride J et al (2018) Development, evaluation, and implemen- 33026 tation of a Pan-African cancer research network: men of African 40. Pichardo MS, Smith CJ, Dorsey TH, Loffredo CA, Ambs S descent and carcinoma of the prostate. J Glob Oncol 4(4):1–14. (2018) Association of anthropometric measures with prostate https:// doi. org/1 0. 1200/ JGO. 18.0 0063 cancer among African American men in the NCI-Maryland pros- 26. Odiaka E, Lounsbury DW, Jalloh M, Adusei B, Diallo TA, Kane tate cancer case–control study. Cancer Epidemiol Biomark Prev PMS et al (2018) Effective project management of a Pan-African 27(8):936–944. https://d oi. org/ 10.1 158/1 055-9 965.E PI- 18-0 242 cancer research network: men of African descent and carcinoma 41. Barrington WE, Schenk JM, Etzioni R, Arnold KB, Neuhouser of the prostate (MADCaP). J Glob Oncol 4:1–12. https:// doi. org/ ML, Thompson IM Jr et al (2015) Difference in association of 10.1 200/J GO.1 8. 00062 obesity with prostate cancer risk between US African American 2 7. Breslow NE, Day NE (1980) Statistical methods in cancer and non-hispanic white men in the selenium and vitamin E cancer research. Volume 1-the analysis of case–control studies. Interna- prevention trial (SELECT). JAMA Oncol 1(3):342–349. https:// tional Agency for Research on Cancer, Lyon doi. org/1 0. 1001/j amao ncol. 2015.0 513 2 8. World Health Organization (WHO) (2011) Waist circumference 42. Su LJ, Arab L, Steck SE, Fontham ET, Schroeder JC, Bensen JT and waist-hip ratio: report of a WHO expert consultation, Geneva, et al (2011) Obesity and prostate cancer aggressiveness among 8–11 December 2008. World Health Organization (WHO), African and Caucasian Americans in a population-based study. Geneva Cancer Epidemiol Biomark Prev 20(5):844–853. https:// doi. org/ 29. Ashwell M, Gibson S (2016) Waist-to-height ratio as an indica- 10. 1158/ 1055- 9965. EPI-1 0-0 684 tor of “early health risk”: simpler and more predictive than using 4 3. Khan S, Cai J, Nielsen ME, Troester MA, Mohler JL, Fontham a “matrix” based on BMI and waist circumference. BMJ Open ETH et al (2016) The association of diabetes and obesity with 6(3):e010159. https:// doi. org/1 0. 1136/ bmjop en-2 015- 010159 prostate cancer aggressiveness among Black Americans and White 30. Swainson MG, Batterham AM, Tsakirides C, Rutherford ZH, Hind Americans in a population-based study. Cancer Causes Control K (2017) Prediction of whole-body fat percentage and visceral 27(12):1475–1485. https:// doi.o rg/ 10.1 007/ s10552- 016- 0828-0 adipose tissue mass from five anthropometric variables. PLoS 44. Park S-Y, Haiman CA, Cheng I, Park SL, Wilkens LR, Kolonel ONE 12(5):e0177175. https://d oi.o rg/1 0.1 371/j ourna l.p one.0 1771 LN et al (2015) Racial/ethnic differences in lifestyle-related fac- 75 tors and prostate cancer risk: the multiethnic cohort study. Can- 3 1. Browning LM, Hsieh SD, Ashwell M (2010) A systematic review cer Causes Control 26(10):1507–1515. https:// doi. org/ 10. 1007/ of waist-to-height ratio as a screening tool for the prediction of s10552-0 15-0 644-y cardiovascular disease and diabetes: 0.5 could be a suitable global boundary value. Nutr Res Rev 23(2):247–269. https://d oi.o rg/1 0. Publisher's Note Springer Nature remains neutral with regard to 1017/ s09544 2241 00001 44 jurisdictional claims in published maps and institutional affiliations. 32. Owolabi EO, Ter Goon D, Adeniyi OV (2017) Central obesity and normal-weight central obesity among adults attending healthcare facilities in Buffalo City metropolitan municipality, South Africa: a cross-sectional study. J Health Popul Nutr 36(1):54. https://d oi. org/1 0. 1186/ s41043- 017- 0133-x 1 3 Cancer Causes & Control Authors and Affiliations Ilir Agalliu1  · Wei‑Kaung Jerry Lin2 · Janice S. Zhang1,2 · Judith S. Jacobson3 · Thomas E. Rohan1 · Ben Adusei4 · Nana Yaa F. Snyper4 · Caroline Andrews5 · Elkhansa Sidahmed5,6 · James E. Mensah7 · Richard Biritwum7 · Andrew A. Adjei8 · Victoria Okyne7 · Joana Ainuson‑Quampah9 · Pedro Fernandez10 · Hayley Irusen10 · Emeka Odiaka11 · Oluyemisi Folake Folasire11 · Makinde Gabriel Ifeoluwa11 · Oseremen I. Aisuodionoe‑Shadrach12 · Maxwell Madueke Nwegbu12 · Audrey Pentz13 · Wenlong Carl Chen13,14,15 · Maureen Joffe13,16 · Alfred I. Neugut17 · Thierno Amadou Diallo18 · Mohamed Jalloh18 · Timothy R. Rebbeck5,6 · Akindele Olupelumi Adebiyi11 · Ann W. Hsi ng2,19,20,21 1 Department of Epidemiology and Population Health, Albert 13 Non-Communicable Diseases Research Division, Wits Einstein College of Medicine, 1300 Morris Park Ave., Health Consortium (Pty) Ltd, Johannesburg, South Africa Bronx, NY 10461, USA 14 National Cancer Registry, National Health Laboratory 2 Stanford School of Medicine, Stanford Cancer Institute, Service, Johannesburg, South Africa Stanford University, Stanford, CA, USA 15 Sydney Brenner Institute for Molecular Bioscience, Faculty 3 Department of Epidemiology, Mailman School of Public of Health Sciences, University of the Witwatersrand, Health, Columbia University, New York, NY, USA Johannesburg, South Africa 4 37 Military Hospital, Accra, Ghana 16 SAMRC/Wits Developmental Pathways for Health Research 5 Dana Farber Cancer Institute, Boston, MA, USA Unit, Department of Paediatrics, Faculty of Health Sciences, University of the Witwatersrand, Johannesburg, South Africa 6 Harvard T.H. Chan School of Public Health, Boston, MA, 17 USA Departments of Medicine and Epidemiology, Columbia University Irving Medical Center, New York, NY, USA 7 Korle-Bu Teaching Hospital and University of Ghana, Accra, 18 Ghana Institut de Formation et de la Recherche en Urologie et de la Santé Familiale, Hôpital Général de Grand Yoff, Dakar, 8 College of Health Sciences, University of Ghana Medical Senegal School, Accra, Ghana 19 Stanford Prevention Research Center, Department 9 College of Health Sciences, School of Biomedical and Allied of Medicine, Stanford School of Medicine, Stanford Health Sciences, University of Ghana, Accra, Ghana University, Stanford, CA, USA 10 Stellenbosch University, Cape Town, South Africa 20 Department of Epidemiology and Population Health, 11 College of Medicine and University College Hospital, Stanford School of Medicine, Stanford University, Stanford, University of Ibadan, Ibadan, Nigeria CA, USA 21 12 College of Health Sciences, University of Abuja, Stanford Cancer Institute, 780 Welch Road, Room 250D, and University of Abuja Teaching Hospital and Cancer Stanford, CA 94305, USA Science Centre, Abuja, Nigeria 1 3