Bulletin of the World Health Organization

Differences by sex in the prevalence of diabetes mellitus, impaired fasting glycaemia and impaired glucose tolerance in sub-Saharan Africa: a systematic review and meta-analysis

Esayas Haregot Hilawe a, Hiroshi Yatsuya b, Leo Kawaguchi a & Atsuko Aoyama a

a. Department of Public Health and Health Systems, Nagoya University School of Medicine, 65 Tsurumai-cho, Showa-ku, Nagoya, 466-8550, Japan.
b. Fujita Health University School of Medicine, Toyoake, Japan.

Correspondence to Esayas Haregot Hilawe (e-mail: esayas@med.nagoya-u.ac.jp).

(Submitted: 13 November 2012 – Revised version received: 21 February 2013 – Accepted: 25 March 2013.)

Bulletin of the World Health Organization 2013;91:671-682D. doi: http://dx.doi.org/10.2471/BLT.12.113415

Introduction

Increasing urbanization and the accompanying changes in lifestyle are leading to a burgeoning epidemic of chronic noncommunicable diseases in sub-Saharan Africa.1,2 At the same time, the prevalence of many acute communicable diseases is decreasing.1,2 In consequence, the inhabitants of sub-Saharan Africa are generally living longer and this increasing longevity will result in a rise in the future incidence of noncommunicable diseases in the region.13

Diabetes mellitus is one of the most prominent noncommunicable diseases that are undermining the health of the people in sub-Saharan Africa and placing additional burdens on health systems that are often already strained.4,5 In 2011, 14.7 million adults in the African Region of the World Health Organization (WHO) were estimated to be living with diabetes mellitus.6 Of all of WHO’s regions, the African Region is expected to have the largest proportional increase (90.5%) in the number of adult diabetics by 2030.6

Sex-related differences in lifestyle may lead to differences in the risk of developing diabetes mellitus and, in consequence, to differences in the prevalence of this condition in women and men.3 However, the relationship between a known risk factor for diabetes mellitus – such as obesity – and the development of symptomatic diabetes mellitus may not be simple. For example, in many countries of sub-Saharan Africa, women are more likely to be obese or overweight than men and might therefore be expected to have higher prevalences of diabetes mellitus.3,7 Compared with the corresponding men, women in Cameroon8, South Africa9 and Uganda10 were indeed found to have higher prevalences of diabetes mellitus. However, women in Ghana,11 Nigeria,12 Sierra Leone13 and rural areas of the United Republic of Tanzania14 were found to have lower prevalences of diabetes mellitus than the men in the same study areas. No significant differences between men and women in the prevalence of diabetes mellitus were detected in studies in Guinea,15 Mali,16 Sudan17 and urban areas of the United Republic of Tanzania,18 or in a meta-analysis of data collected in several studies in West Africa.19 Although wide variations in the distribution of diabetes mellitus by sex have been documented in several review articles,35,7,20 the possible causes of this heterogeneity have never been examined in detail.

Like obesity, impaired fasting glycaemia and impaired glucose tolerance appear to be risk factors in the development of diabetes mellitus.21,22 According to the International Diabetes Federation, the estimated age-adjusted prevalence of impaired fasting glycaemia in WHO’s African Region was substantially higher in 2011 than the corresponding global mean value – 9.7% versus 6.5%, respectively – and is expected to have risen further by 2030.23

Impaired fasting glycaemia and impaired glucose tolerance are reported to be metabolically distinct entities that affect different subpopulations, albeit with some degree of overlap.22,24 In Mauritius, the prevalence of impaired fasting glycaemia was found to be significantly higher in men than in women, whereas the prevalence of impaired glucose tolerance was found to be higher in women than in men.24,25

Differences between men and women in the prevalence of diabetes mellitus, impaired fasting glycaemia and impaired glucose tolerance in much of sub-Saharan Africa have yet to be reviewed. Given the variation in health care, culture, environment, human behaviour and other determinants of health across sub-Saharan Africa,26 the conclusions drawn from a recent meta-analysis of data from West Africa19 should not be assumed to apply to the whole of sub-Saharan Africa. The sex-specific prevalence of at least one risk factor for diabetes mellitus – obesity – is known to differ across different parts of sub-Saharan Africa.7,27

The main aims of the present systematic review were to examine differences between men and women in the prevalence of three conditions – diabetes mellitus, impaired fasting glycaemia and impaired glucose tolerance – in Eastern, Middle and Southern Africa (i.e. all in sub-Saharan Africa according to the United Nations subregional classification for African countries),28 and to explore the possible causes of any variation observed. We followed the Meta-analysis of Observational Studies in Epidemiology (MOOSE) group’s guidelines for the reporting of systematic reviews of observational studies.29

Methods

Data sources

In September 2011, we searched PubMed and Web of Science for studies that presented the sex-specific prevalences of diabetes mellitus, impaired fasting glycaemia and/or impaired glucose tolerance in Eastern, Middle and/or Southern Africa (Table 1). The medical subject headings (MeSH) and search terms we used are described in Box 1. We limited our search to human studies but placed no restrictions on the language of publication. We also used Google, Google Scholar and WHO’s InfoBase to search the “grey” literature for relevant studies and reports. The citations in articles that appeared to be relevant were examined for other articles that might hold useful data. When it seemed possible that relevant data had been recorded but not published, the authors of published study reports were contacted via e-mail to see if they could provide such data.

Box 1. Strategy followed in searching PubMed and the Web of Science

Various medical subject headings (MeSH) and search terms, including “prevalence”, “incidence”, “epidemiology”, “proportion”, “rate”, “diabetes mellitus”, “hyperglycaemia”, “abnormal* blood glucose”, “glucose intolerance”, “dysglycaemia”, “insulin resistance”, “metabolic* syndrome”, “insulin resistance syndrome X”, “cardiovascular syndrome”, “hypertension”, “increase* blood pressure”, “obesity”, “overweight”, “hypercholesterolaemia”, “hyperlipidaemia”, “dyslipidaemia”, “physical inactivity”, “smoking”, “cardiovascular diseases risk factors” and “Africa South of the Sahara” – and alternative spellings such as “hyperglycemia” were used. Searches were combined with the names of each country in Eastern, Middle and Southern Africa (Table 1) – except Cameroon, which was included in a previous study on West Africa19 – by using the Boolean operators “OR” or “AND”.

Inclusion and exclusion criteria

Data were included in the meta-analysis if they came from studies that fulfilled all of the following criteria:

  • community-based;
  • cross-sectional;
  • reported prevalence of diabetes mellitus, impaired fasting glycaemia and/or impaired glucose tolerance;
  • reported either odds ratios (ORs) for differences between men and women in the prevalence of diabetes mellitus, impaired fasting glycaemia and/or impaired glucose tolerance or data that allowed the computation of such ORs;
  • conducted in apparently healthy, non-pregnant subjects;
  • most subjects are adults (i.e. aged ≥ 15 years) and residing in the UN-designated Eastern, Middle or Southern subregions of Africa;
  • both men and women investigated;
  • employed any of WHO’s diagnostic criteria – or the equivalent criteria of the American Diabetic Association – for diabetes mellitus, impaired fasting glycaemia and/or impaired glucose tolerance;3038
  • reported results either in English or in another language with an abstract in English.

When multiple reports of the same study were retrieved, only the most informative report was selected. Clinic-, hospital- and laboratory-based studies, anonymous reports, letters, commentaries, case studies and reviews were excluded.

Data abstraction

After reading each article that appeared relevant and met the inclusion criteria, one of the authors (EHH) made notes of the year of study and publication, sampling method, sample size, response rate, study design, diagnostic criteria, study area, mean age and/or age range of the subjects, mean blood glucose level, the recorded prevalences of diabetes mellitus, impaired fasting glycaemia and/or impaired glucose tolerance, and, if available, the OR and corresponding 95% confidence intervals (CIs) that indicated the type and significance of any differences in these prevalences by sex. When articles presented data separately for urban and rural subjects, information for these two groups of subjects was extracted separately. When articles presented data stratified by subject age, only the data for subjects aged 15 years or older were included in the analysis. All of the extracted data were independently reviewed by a second author (HY).

Quality appraisal

A checklist – adopted from one created by the University of Wisconsin39 – was used to assess the quality of the included studies. The checklist had eight questions relating to the research question, selection of study subjects, comparability of study groups, handling of withdrawals, measurement of outcomes, statistical analyses, results and conclusions, and funding or sponsorship. If the answers to five or more of these questions were positive, the study involved was categorized as “positive” and considered to be of good quality. If the answers to five or more of these questions were negative, the study involved was categorized as “negative” and considered to be of poor quality. All other studies were categorized as “neutral”.

Statistical analysis

ORs were used as “effect estimates” to quantify the relationship between sex and the prevalence of diabetes mellitus, impaired fasting glycaemia and impaired glucose tolerance. If no OR had been reported, it was calculated from the raw data. Since the studies included in the meta-analysis used different standard populations, crude prevalences were preferred to the age-adjusted values when both were available. The DerSimonian and Laird random-effects model was used to estimate the mean OR for all of the studies included in the meta-analysis.40

Statistical heterogeneity across the studies was evaluated using both the Q and I2 statistics.40 In the Q-tests, a P-value of < 0.1 was considered indicative of statistically significant heterogeneity. We performed subgroup analyses to assess the potential influence of the following study-level covariates on the OR for any sex-specific differences: area of residence (urban or rural), subregion of residence in sub-Saharan Africa (i.e. Eastern, Middle or Southern Africa), study year, ethnicity of the study subjects, and the World-Bank-determined income level of the study country.41 Random-effects univariate meta-regression analysis40 was also performed as an extension of the subgroup analyses.

The potential influence of each individual study on the overall summary estimates was assessed by rerunning the meta-analysis while omitting one study at a time. Sensitivity analysis was performed to assess the impact of the quality of the studies on the overall effect estimates. For those studies that reported both crude and age-adjusted prevalences, we also assessed if the effect estimates would have been substantially altered if the age-adjusted values had been used instead of the crude ones.

Publication bias40 was assessed using a funnel plot to examine the relationship between the effect size and study precision. Begg and Mazumdar’s rank-correlation test40 was then used to test this relationship statistically. Finally, Duval and Tweedie’s “trim and fill” analysis was used to assess the possible impact of publication bias on the effect size.40

Version 2 of the Comprehensive Meta-Analysis software package (Biostat, Englewood, United States of America) was used for all of the statistical analyses. All statistical tests were two-sided. A P-value of < 0.05 was generally considered indicative of statistical significance.

Results

Literature search

Although the PubMed and Web of Science searches revealed 5129 potentially useful reports, only 25 of these reports were found to satisfy all of the inclusion criteria (Fig. 1). Four additional reports that met all of the inclusion criteria were identified via a Google search (n = 2), a search of the WHO InfoBase (n = 1) or contact with authors (n = 1). The meta-analysis therefore included data from 29 reports that, together, covered 36 studies in which cross-sectional data were collected.14,17,4268

Fig. 1. Flow diagram of the study selection procedure
Fig. 1. Flow diagram of the study selection procedure
DM, diabetes mellitus; IFG, impaired fasting glycaemia; IGT, impaired glucose tolerance.

Study characteristics

Table 2 (available at: http://www.who.int/bulletin/volumes/91/9/12-113415) provides detailed descriptive information for the 36 studies included in the meta-analysis. These studies involved 75 928 subjects and were conducted between 1983 and 2009 in Angola, the Democratic Republic of the Congo, Kenya, Malawi, Mauritius, Mozambique, Seychelles, South Africa, Sudan, Uganda, the United Republic of Tanzania, Zambia or Zimbabwe. Most (92%) of the studies included in the meta-analysis employed probability- or census-sampling techniques and had response rates of 62–99%. Sex-specific prevalences of diabetes mellitus, impaired fasting glycaemia and impaired glucose tolerance were included in the reports of 35, 21 and 11 of the studies, respectively. Almost half (45%) of the studies were conducted in both urban and rural areas. The other studies were conducted exclusively in urban (26%), rural (23%) or periurban (6%) areas. In terms of quality, the studies were categorized as either “positive” (n = 31) or “neutral” (n = 5)42,49,58,61,63 (Appendix A, available at: http://www.med.nagoya-u.ac.jp/intnl-h/swfu/d/auto-UZzMJC.pdf).

Sex-specific prevalences

The prevalence of diabetes mellitus was 5.7% (95% CI: 4.8–6.8) overall, with a slight difference between the men (5.5%; 95% CI: 4.1–7.2) and women (5.9%; 95% CI: 4.6–7.6) included in the meta-analysis. The prevalence of impaired fasting glycaemia was 4.5% (95% CI: 3.3–6.1) overall – 5.7% (95% CI: 3.7–8.6) among the men and 3.5% (95% CI: 2.1–5.8) among the women – whereas the prevalence of impaired glucose tolerance was 7.9% (95% CI: 6.7–9.2) overall – 7.3% (95% CI: 6.0–8.8) among the men and 8.5% (95% CI: 6.7–10.7) among the women.

Odds ratios

The prevalence of diabetes mellitus among men was not significantly different from that among women (OR: 1.01; 95% CI: 0.91–1.11). However, impaired fasting glycaemia appeared to be significantly more common among men than among women (OR: 1.56; 95% CI: 1.20–2.03), whereas impaired glucose tolerance appeared to be significantly less common among men than among women (OR: 0.84; 95% CI: 0.72–0.98) (Fig. 2). These significant differences between the sexes were still observed when the analysis was restricted to those studies in which the prevalences of both impaired fasting glycaemia and impaired glucose tolerance were determined in the same study cohorts (data not shown). A moderate to substantial level of heterogeneity between studies was detected in the data for diabetes mellitus (I2 = 54.62%; P < 0.001 in Q-test), impaired fasting glycaemia (I2 = 85.38%; P < 0.001 in Q-test) and impaired glucose tolerance (I2 = 74.13%; P < 0.001 in Q-test).

Fig. 2. Forest plot of main meta-analysis results, showing sex-specific odds ratios for diabetes mellitus, impaired fasting glycaemia and impaired glucose tolerance in sub-Saharan Africa
Fig. 2. Forest plot of main meta-analysis results, showing sex-specific odds ratios for diabetes mellitus, impaired fasting glycaemia and impaired glucose tolerance in sub-Saharan Africa
DM, diabetes mellitus; IFG, impaired fasting glycaemia; IGT, impaired glucose tolerance; MWMoH, Malawi Ministry of Health; OR, odds ratio; ZWMoH, Zimbabwe Ministry of Health and Child Welfare.
Note: The ORs shown are for differences in prevalence between the sexes (i.e. odds in men versus odds in women). For each study, the plot indicates the mean OR (midpoint of the square), the corresponding 95% confidence interval (horizontal lines) and the weight given to the study (area of the square).

Subgroup analyses

Table 3 summarizes the results of the subgroup analyses. Significant heterogeneity in the OR for diabetes mellitus was observed by area of residence (i.e. urban or rural), subregion of residence in Africa, ethnicity of the study subjects, and country income level – each of which gave a P- value of < 0.05 in a Q-test. The prevalence of diabetes mellitus was found to be significantly higher in men than in women in studies conducted in a mix of urban and rural areas, in Middle or Eastern Africa or in low-income countries. However, in studies conducted in Southern Africa or among subjects of Indian ethnicity, the prevalence of diabetes mellitus was significantly higher among women than among the corresponding men.

Significant heterogeneity in the OR for impaired fasting glycaemia was observed by subregion of residence in Africa (P = 0.02) and country income level (P = 0.006). In studies conducted in Eastern Africa or upper-middle-income countries, impaired fasting glycaemia appeared to be significantly more common among men than among women.

With impaired glucose tolerance, significant heterogeneity in the OR was observed by area of residence (P < 0.001), subregion of residence in Africa (P = 0.001), ethnicity (P = 0.002), and country income level (P = 0.03). The odds of impaired glucose tolerance were found to be higher in men than in women in studies conducted on urban residents or subjects of Indian ethnicity.

Meta-regression

In general, the univariate random-effects meta-regression revealed similar associations – between the OR and study-level covariates – as seen in the subgroup analyses (Appendix A). For example, the OR for the sex-specific prevalences of diabetes mellitus appeared to be significantly affected by area of residence (rural versus urban; P = 0.018), subregion of residence in Africa (Southern and Middle Africa versus Eastern Africa; P < 0.001), ethnicity of the study subjects (multi-ethnic versus Indian; P = 0.013), study year (1990s versus 2000s; P = 0.039), and country income level (low versus upper middle; P < 0.001). Subregion of residence (Eastern versus Southern Africa; P = 0.047) and country income level (low versus upper-middle; P = 0.006) also had a significant effect on the OR for impaired fasting glycaemia, whereas subregion of residence (Eastern versus Southern Africa; P < 0.001), ethnicity of study subjects (multi-ethnic versus Indian; P < 0.001), country income level (low versus upper-middle; P < 0.001), and area of residence – both rural versus urban (P < 0.001) and rural versus urban and rural combined (P = 0.003) – had significant effects on the OR for impaired glucose tolerance.

Sensitivity and influence analyses

No meaningful change in the OR was evident when the meta-analysis was rerun either with the data from the five studies of “neutral” quality omitted or using age-adjusted prevalences instead of the crude values (data not shown).

The results of the influence analysis indicated that the omission of the data from any of seven studies – described in five reports43,44,47,48,57 – could eliminate the statistical significance of the overall differences between men and women in the prevalence of impaired glucose tolerance. However, even when the data from one of these studies were omitted, women still showed a higher prevalence of impaired glucose tolerance than the corresponding men, with a P-value of > 0.05 but < 0.1. The pooled results for diabetes or impaired fasting glycaemia were not substantially affected by the omission of the data from any one study.

Publication bias

The funnel plots for diabetes mellitus and impaired fasting glycaemia were asymmetric, indicating possible publication bias. However, the corresponding results from Begg and Mazumdar’s rank-correlation tests – P-values of 0.93 and 0.64, respectively – were not statistically significant. Duval and Tweedie’s “trim and fill” analysis indicated that the meta-analysis would have benefitted from the inclusion of data from more studies – nine for diabetes mellitus and one for impaired fasting glycaemia – and that, if the asymmetry seen in the funnel plots was the result of publication bias, the summary estimates of the sex-specific (i.e. men versus women) OR for diabetes mellitus and impaired fasting glycaemia should be 1.09 (95% CI: 0.98–1.20) and 1.65 (95% CI: 1.27–2.14), respectively (Appendix A).

There were no indications of publication bias in the data on impaired glucose tolerance.

Discussion

To our knowledge, this study is the first systematic review of possible associations between sex and the prevalences of impairments in glucose tolerance and fasting glycaemia in Eastern, Middle and Southern Africa. Previous narrative reviews have reported on the prevalence of diabetes mellitus and, briefly, on the variation in the sex distribution of this illness in sub-Saharan Africa.35,7,20 However, there appears to have been only one previous meta-analysis of data on the prevalence of diabetes mellitus in sub-Saharan Africa and that was limited to data collected in West Africa.19

The present results reveal considerable between-country variation in the prevalence of diabetes mellitus among adults. However, the relatively high value recorded for all of the studies combined (5.7%) is a reflection of the rapid transition – from a predominance of communicable disease to one of noncommunicable disease – that much of sub-Saharan Africa is facing. In this vast area of Africa, important risk factors for diabetes mellitus, such as impaired glucose tolerance, appear to be increasing in prevalence while humans are tending to live longer. The prevalence of diabetes mellitus in sub-Saharan Africa will therefore probably rise further unless prevention efforts are intensified. 23

In the present meta-analysis – as in most22,24,69 – but not all70 – previous studies on this risk factor for diabetes mellitus – impaired fasting glucose was found to be significantly more common among men than among women, irrespective of the subgroup that was investigated. One possible explanation for this difference is that men tend to have lower hepatic sensitivity to insulin and may, in consequence, have generally higher fasting levels of plasma glucose.69 Another possible explanation or contributing factor is that, within sub-Saharan Africa, men are more likely to smoke than women71 and smoking appears to increase the risk of impaired fasting glucose, by decreasing insulin sensitivity.7274

In earlier research, impaired glucose tolerance has generally been found to be more common among women than among men.22,24,69 The same difference between the sexes was detected in most of the subgroups that were investigated in the present meta-analysis. In general, women have a smaller mass of muscle than men and therefore less muscle available for the uptake of the fixed glucose load (75 g) used in the oral glucose-tolerance test.69,75 Women also have relatively high levels of estrogen and progesterone, both of which can reduce whole-body insulin sensitivity.76 Physical inactivity77 and unhealthy diet78 have also both been associated with impaired glucose tolerance. In many countries in sub-Saharan Africa, women are more likely to be physically inactive than the corresponding men.79,80

The differences in the sex distribution of both impaired fasting glycaemia and impaired glucose tolerance in sub-Saharan Africa need to be considered in evaluating the probability that individuals will develop diabetes mellitus and in efforts to prevent the disease. Impairments in glucose tolerance and in fasting glycaemia are not metabolically equivalent, and the people classified as having each condition are different as well.22,81 If screening programmes were based only on the measurement of “fasting plasma glucose”, most individuals with impaired glucose tolerance would go undetected and the population identified as being at risk would probably be biased towards males. The glycated haemoglobin (HbA1c) assay69 may offer a way of evaluating the risk of diabetes mellitus that is relatively sex-neutral, although this assay is currently too expensive for routine use in Africa and it can also be affected by disorders such as malaria.82 Screening for both impaired fasting glycaemia and impaired glucose tolerance might eliminate most of the sex bias in the identification of those who are at risk of developing diabetes mellitus. Even then, the dose of glucose used in the oral glucose-tolerance test may have to be made lower for women than for men – or tailored to the height of the individual to be tested – to allow for the lower mean muscle mass in women and so prevent the over-diagnosis of impaired glucose tolerance in women.72

In the present meta-analysis, despite the differences seen by sex in impaired fasting glycaemia and impaired glucose tolerance, the overall prevalence of diabetes mellitus in men was found to be very similar to that in women. However, subgroup analyses revealed that diabetes mellitus was more common in the men who lived in Middle and Eastern Africa than in the women who lived in the same African subregions, whereas the women who lived in Southern Africa were more likely to have diabetes mellitus than the corresponding men. Such differences between the sexes were not seen in the earlier study on diabetes mellitus in West Africa.19 Some of these differences may be related to differences between the sexes in the prevalence of central obesity, which, as a risk factor for diabetes mellitus, is more predictive than peripheral obesity.83 Central obesity has been found to be more common in men than in women in Eastern Africa84,85 and more common in women than men in Southern Africa.86 However, such obesity cannot be used to explain why the men of Middle Africa are more likely to have diabetes mellitus than the women, as central obesity is more common among the women in this area than among the men.87 Behavioural risk factors, such as smoking and alcohol use, which are more common among the men of sub-Saharan Africa than among the women,3,71 might contribute to the prevalence of diabetes mellitus among the men of Middle Africa.

In the present meta-analysis, the income level of the country of residence – a proxy indicator of the economic status of the people in the country – appeared to contribute to the heterogeneity seen in the association between sex and the prevalence of diabetes mellitus. Women of low socioeconomic status in Australia,88 Canada,89 Germany90 and the United States of America91 appear to be at markedly higher risk of diabetes mellitus than the corresponding men. In a recent meta-analysis, the incidence of Type 2 diabetes mellitus among adults with low socioeconomic status was found to be generally higher in women than in men; it was suggested that the women who lived in impoverished areas were more likely to be obese, physically inactive and under high levels of psychosocial stress than the men in the same areas.92 In contrast, the results of the present meta-analysis indicated that men who lived in the low-income countries of sub-Saharan Africa were more likely to be diagnosed with diabetes mellitus than the corresponding women. This difference between the sexes may be a consequence of differences between men and women in the distribution of risk factors for diabetes mellitus (e.g. obesity, physical inactivity, poor diet and smoking, etc.) in low-income countries. Another possibility is that women in low-income countries have particularly poor access to health-care services and therefore little chance of being diagnosed with diabetes.88,89,91,92 In addition, as Africa is one of the most inequitable parts of the world in terms of income,93 the income level recorded for an African country might not correlate with the socioeconomic status of a study cohort in that country. There appear to be no published data sets that would allow sex-based differences in the relationship between individual socioeconomic status and diabetes mellitus in sub-Saharan Africa to be investigated.

The present meta-analysis had several limitations. First, the studies that provided the data for the meta-analysis were conducted under different circumstances in different countries and the prevalences of diabetes mellitus, impaired fasting glycaemia and/or impaired glucose tolerance were not the primary outcomes of some of the studies. A random-effects model was therefore employed to embrace this considerable heterogeneity.40 Second, the studies had to be cross-sectional in design to be included in the meta-analysis and may therefore have been affected by confounding and biases. However, we attempted to minimize selection bias by employing predefined study selection criteria and a quality appraisal checklist. Potential sources of heterogeneity were also assessed in subgroup and meta-regression analyses. Third, since our subgroup and meta-regression analyses were entirely observational in nature, the relationships recorded – across all of the studies – between some study-level characteristics and the effect estimate could be subject to confounding by other study-level characteristics. Unfortunately, the studies included in the meta-analysis were too few to allow for a reasonable assessment of interactions between the study-level covariates. Fourth, we used the income levels of the countries of residence to stratify the studies because of a general lack of information on the socioeconomic status of study participants. The relationships that we observed between a country’s income level and the sex-specific prevalences of interest may therefore not reflect the relationships between the socioeconomic status of the subjects and their risks of impaired fasting glycaemia, impaired glucose tolerance or diabetes mellitus. Finally, our conclusions may have been affected by publication bias. The asymmetric funnel plots were indicative of possible publication bias in the data for diabetes mellitus and impaired fasting glucose. Furthermore, our study selection criteria excluded reports that did not have an abstract in English and may have excluded some reports that were not recorded in the PubMed or Web of Science databases, although we did try to search the “grey” literature for relevant data. The results of the “trim and fill” analyses indicated that the impact of any publication bias on our conclusions was probably trivial.

In summary, our meta-analysis demonstrated that, compared with the corresponding women, the men in Eastern, Middle and Southern Africa had a significantly higher prevalence of impaired fasting glycaemia and a lower prevalence of impaired glucose tolerance. Although the overall prevalence of diabetes mellitus did not significantly differ by sex, the prevalence of diabetes mellitus was found to be lower or higher in women than in men when analysed by African subregion. Sex-based differences in the relationship between individual socioeconomic status and impaired fasting glycaemia, impaired glucose tolerance and diabetes mellitus still need to be investigated in sub-Saharan Africa. Our observations may help in the targeting of appropriate – and perhaps sex-specific – interventions to prevent diabetes mellitus in sub-Saharan Africa.


Acknowledgements

We are grateful to the authors of the articles included in the meta-analysis, many of whom kindly provided us with additional information regarding their studies.

Competing interests:

None declared.

References

Share