Prevalence of hypertension in Ghanaian society: a systematic review, meta-analysis, and GRADE assessment

The major electronic archives for published research, Medline/PubMed, Web of Science, Embase (via Ovid), CINAHL, and African journals online (AJOL) were queried for publications reporting population estimates of hypertension in Ghana. For grey literature, the authors perused the first 200 results in an advanced Google Scholar search as well as local thesis repositories. In addition, a snowballing technique of scrutinizing bibliographies of all eligible studies was employed to identify additional potential reports. A modified approach with search terms specifying and targeting the defined population, intervention (defined appropriately for non-experimental scenarios), defined outcome of interest, and required setting for studies was used to screen all study titles and abstracts [39]. Terms specifying the intervention concept were “prevalence”, “proportion”, “survey”, “descriptive”, “cross-sectional”, “cohort”, “longitudinal”, “attributable fraction”, and “incidence”. Outcome of interest was coded in the search strategy as “hypertension”, “blood pressure”, “cardiovascular”, and “cardiometabolic”. Those for the settings were “Ghana” and “Ghanaians”. Search terms were framed with the “OR” and “AND” operators as previously described [35].

The main outcome was the prevalence of elevated BP in the general and various strata of the population. Published articles on studies and follow-up studies published before October 2019 with hypertension, risk factors, management practices, and control in the Ghanaian population as an outcome were eligible for inclusion in this review. An additional search for the period spanning October 2019 to November 2020 yielded an additional 2 studies. Conference abstracts reporting adequately detailed information regarding the population, definition of hypertension, sample size, blood pressure data collection, and a point estimate of hypertension were also included for screening. Population-based studies involving individuals living in Ghana or with well-defined data analysis of Ghanaian cohorts were considered for inclusion. Multicenter studies that include Ghanaian participants were included if there was adequate statistical detail on data pertaining to Ghana. Reports fulfilling inclusion criteria were eligible for the initial screening (Fig. 1). Studies that failed to satisfy conditions for inclusion were excluded in this review to prevent impact of related confounding variables. These included case reports, reviews, expert commentaries, and data on Ghanaians living in foreign territories. Although reviews were not included in this work, a manual search of references found in review articles was performed. Additionally, the findings of this work were compared to previous work and in order to build upon existing knowledge. Clinical studies reporting on organ-specific elevated blood pressure such as pulmonary hypertension and studies that were based on hospitalized patients, pregnant women, or utilized hospital-based sampling protocols were excluded on grounds that they do not adequately represent the national population. Studies that did not report absolute population or sub-population estimates on hypertension were further excluded from the quantitative synthesis or meta-analysis.

Citations returned in the database query were imported to Mendeley Desktop 1.19.4, and duplicate reports from multiple sources were identified and removed. The titles and abstracts of articles returned in database queries using the above search strategy were accessed and pre-screened by two co-authors who worked independently to flag non-suitable or extraneous reports for exclusion. Subsequently, full-text reports of the pre-screened and promising studies were accessed and perused by two co-authors who worked independently to ascertain conformity to stated inclusion criteria. A checklist was used for this purpose. Only articles published in the English language were considered. Detected discrepancies in the outcome of independent screening was resolved by consensus of reviewers. Relevant information from selected studies was extracted into a standardized Microsoft Excel template. In accordance with the PRISMA guidelines, the basis for excluding any article during the extensive screening stage was documented and reported (see Additional file 1) [40].

Relevant data was retrieved from selected full-text manuscripts using a standardized extraction form [35]. Details of publication such as manuscript title, author names, date of publication, digital object identifiers (DOIs), and other vital variables were extracted. In addition, informative variables such as study size, target population, time period of data collection, geographical location, study objective, eligible participants, and other elements of design were captured into format. Furthermore, explanatory and outcome variables, details of data analysis, and the major findings were retrieved as well. Additional data on participant sociodemographic characteristics, anthropometrics, and blood pressure measurement protocols were also extracted. The overall prevalence as well as age-specific prevalence based on the classification of hypertension were of interest. In instances of replicate data reporting affecting the same study population and setting, relevant information was retrieved from the most informative manuscript(s). All such reports were regarded as one unique study report. Also, in such instances, the earliest year of publication was reported.

All studies were appraised for potential risk of bias according to a suitably validated tool [42]. The tool concerns itself with assessments of both the internal and external validity of reports from cross-sectional studies. Two independent assessors conducted this appraisal and scored the manuscripts as high, low, or very low for bias based on the presence or absence of each construct being assessed. Specific issues probed included whether participants were adequately representative of the defined population, randomly sampled, and were drawn from an adequate sampling frame with low non-response bias (external validity). Internal validity was assessed from impressions about instrument reliability and validity, uniform administration of instrument, definition of hypertension, primary data collection, and exposure bias. All disagreements were resolved by discussion and regular consultations with more experienced team members. Reports with scanty detail were classified as “limited,” and the primary investigators were contacted for particular information.

Meta-analysis was conducted in the R computing software using the “meta” package [43, 44] (Additional file 2). A random intercept logistic regression model, which is a form of generalized linear mixed model (GLMM), was utilized to estimate pooled estimates of hypertension prevalence in Ghana based on available studies [45]. The proportion of hypertensive individuals in each study was transformed using the logit transformation in order to stabilize the variances of the prevalence estimates. The choice of the logit transformation was made after a thorough review and consideration of the merits and demerits of the various transformation methods of proportions, including the arcsine and Freeman-Tukey double arcsine transformations. Schwarzer et al. [46] recommended “the use of inverse variance method with the arcsine or logit transformations for the meta-analysis of single proportions that require individual study weights, after reporting seriously misleading results in a meta-analysis with very different sample sizes due to problems with the back-transformation of the Freeman-Tukey transformation.” Gender-, age-, and region-specific estimates were presented.

The I² statistic is a measure of variability in pooled estimate attributed to heterogeneity of studies while the Q statistic is a check for homogeneity in effect estimates. Higgins and Thompson’s I² statistic [47] and Cochran’s Q statistic [48] were used to assess heterogeneity between the included studies. Subgroup analysis was performed by BP measurement device, year of publication, nature of study population, region, and geographical location of study site in order to determine whether prevalence of hypertension was modified by subgroup membership.

Meta-regression was performed in R to determine the extent to which study characteristics could explain the heterogeneity among prevalence estimates of hypertension between studies. Codes for the procedure are shown in Additional file 3. Whereas σ² is typically used to represent within study variance, τ² is used to represent the between study variance, also called study heterogeneity [49]. The estimate of τ² in meta-regression analysis with covariate in comparison to τ² when the covariate is omitted allows for the calculation of the proportion of study heterogeneity explained by the covariate [50]. Attention was also given to the R² value which indicates the percentage of the variance in the dependent variable that the independent variables explain collectively.

The database queries returned 381 hits which were screened down to 344 studies after exclusion of duplicates (n = 37). An initial screening of titles for relevance narrowed this number down to 132 reports. A further 105 were removed following abstract (n = 63) and full manuscript review (n = 42) for a number of reasons stated in Fig. 1. These reasons include false hits, failure to meet eligibility criteria, reporting on secondary data, hospital-based studies involving convenient samples, small sample size, unexpected definition of hypertension or self-reports, systematic reviews or letters to the editor, duplicate analyses on previous samples, and other risks of bias. The remaining 27 eligible reports retained for analysis represent unique population-based studies on Ghanaian subjects conducted from 1977 to 2020.

At the time of compiling this report, new administrative regions of the country were created out of existing ones. To facilitate comprehension among international audiences and avoid confusion in future, the administrative map used here has been given in Fig. 2. Most of these studies (n = 15) were conducted in the Greater Accra (n = 8 studies) and the Greater Kumasi (n = 8 studies) areas. There were 3 studies from the Upper East, 2 nationwide reports and one international study. In addition, the Bono, Bono East, Eastern, and Volta regions were represented by a study each. Details of the specific locations and identities of these studies are given in Fig. 2 and Additional file 4. Data collection ranged from 2 to 36 months. The earliest sampling was conducted in 1972 by Pobee et al. [30], and the latest was from November 2017 by Acheampong et al. [51]. There were more published studies available in the past decade (2010–2020) than the two decades (1990–2010) prior. A total of sixteen (16) studies were published after year 2010, and eleven (11) studies were published before this year. The year with the most included studies was 2017 with 5 studies.

All studies included in the review were original articles capturing cross-sectional studies and baseline surveys of prospective studies [52, 53]. Hypertension prevalence was reported as a primary outcome in all 24 studies included in the meta-analysis. A number of studies also reported on risk factors (n = 15), awareness (n = 9), management (n = 9), and control (n = 9).

Although a number of reports published on the same dataset were identified, all selected studies reported on unique datasets. These clusters of publications emerged from datasets generated by large-scale multinational population-based cardiovascular risk studies such as the modeling epidemiologic transition study (METS) [27, 54] and the WHO SAGE project [22, 23, 55]. One study also [21] reported results from the University of Witwatersrand’s INDEPTH collaboration with the US National Institutes for Health (NIH) [56] and H3Africa Consortium [57]. Six study sites in four sub-Saharan African countries, namely, South Africa, Kenya, Ghana, and Burkina Faso, are involved in the African genomics partnership (AWI-Gen) [21].

Other smaller, localized datasets informing selected studies included the Accra urban poverty project (UPHS) which examined associations between health and developmental indices from urban poor populations in the capital of Ghana [53, 58] and the second phase of the Women’s Health Survey (WHSA-2) conducted in Accra [52, 59, 60]. The WHSA was designed to measure the burden of tropical diseases among adult women. Notably, the Time Use and Health Study of Accra (TUHS) [61] was not selected on account of using 87% of data reported in the WHSA [52].

In general, a total of 45,470 subjects (n = 22,866 males and 22,604 females) enrolled from urban (n = 12) and rural populations (n = 8), as well as mixed populations (n = 7), were captured in this review. Studies from populations designated as mixed covered both urban and rural populations. The definition of “urban” and “rural” given by the original authors were applied. In cases where this was not stated, the entry category used for the location in question in the Ghana Demographic Heath Survey 2010 by the Ghana Statistical Service was used to classify the study. Two studies sampled female subjects exclusively [51, 52], and one study reported on males exclusively [62].

In general, BP was measured across studies according to a validated and clinically approved protocol by trained field workers or healthcare workers including nurses and physicians (Additional file 5). Mostly, all subjects had their BP readings taken by protocols that were standardized across study sites. Only a handful of studies reported partial fractions of study subjects evaluated for blood pressure. Some studies, per the definition of hypertension given, admitted patients who were on medication into this category [63, 64]. Most studies reportedly used the JNC VII criteria which defines elevated BP or hypertension as having either a systolic BP greater or equal to 140 mmHg or concurrent with a diastolic BP greater or equal to 90 mmHg or prescription use of antihypertensives [41], irrespective of BP [28]. However, a few studies conducted before this definition was published used the old WHO definition (BP greater or equal 160/95) [30, 65]. One study refrained from explicitly providing a prevalence of hypertension but reported the mean systolic and diastolic BP values [66].

In most studies (n = 21), participants were seated during the BP measurement, usually in a quiet place. Two (2) reports had patients in the supine position [52, 67], and five (5) did not report on posture. In general, study subjects were required to take at least 5 min rest prior to BP measurement, and in one instance, investigators insisted on avoidance of smoking [68]. One study reported an initial resting period of 3–5 min [21].

Half of the selected studies (n = 13, 52%), usually the more recent, reportedly used electronic BP monitors. Commonly, models of the Omron digital brand were employed. In addition, the Boso Medistar Wrist BP Monitor Model S and the semi-automated Microlife Watch BP home were also used. Two studies did not specify the BP measurement apparatus used, and up to 10 (40%) included studies also used manual sphygmomanometers. In such cases, the systolic and diastolic BPs were noted to coincide with the first and the fourth or fifth [30, 68] Korotkoff phase sounds heard during auscultation, respectively.

The preferred anatomical site for BP measurement was the upper arm. A dozen studies (n = 12) indicated using either large or small cuffs depending on which was appropriate for the subject. The Danfa study used a 14-cm large cuff size for all subjects [30] while the SAGE WAVE 1 used a wrist BP monitor [23]. BP measurements were performed by teams of trained study staff with varying degrees of experience. Teams were variously constituted with clinical and community health nurses, general field staff/research assistants, allied health workers, and students. Teams were usually supervised by medical officers or research officers. In line with national and international guidelines, most studies measured participant’s blood pressure from the arm, mostly the right upper arm. Minicuci et al. used a device that had to be worn around the wrist to measure the BP. In most cases, participants were seated in a quiet area. Koopman et al. reported blood pressure readings from supine individuals, while Hill et al. had some participants seated and some supine [52].

Except for four studies [20, 65, 69, 70], all studies performed one-time BP readings. Studies using BP protocols based on multiple visits performed secondary readings within 24 h [65], 3 weeks [20], or 1 month [60] to confirm an initial BP reading exceeding 140/90 mmHg [20, 60]. In Kunutsor et al., BP readings were replicated after 2 weeks for participants (n = 89 (16%)) in order to adjust for the phenomenon of regression dilution in BP studies where baseline/initial BP measurements are noted to underestimate the actual BP leading to underestimation of cardiovascular risk [69]. A regression dilution ratio may be calculated from repeat readings for the purpose of correcting for this error and to estimate the actual BP [71].

On sampling days, with the exception of one study in which BP was measured a minimum of three times [19], most studies conducted BP readings up to three times to allow confirmation of elevated BP readings [65, 69, 72]. In general, the intervening time between repeat BP measurements was not less than 1 min but less than 1 h in most studies and up to a full day in one study [72]. Majority of studies (n = 23) reported the nature of BP statistic used in classifying participants. Most studies computed an average BP from two readings [20, 21, 23, 26‐28, 51, 66, 68, 69, 72‐75] or three [19, 30, 53, 70, 76]. Frequently, studies would discard the initial reading and rely on the mean of the latter readings [11, 20, 21, 23, 73] or the fifth and sixth [27]. One study, Pobee et al. [65], used only one BP reading with confirmation for high-for-age readings.

Reported prevalence estimates on hypertension ranged from a low of 4.5% in a rural population from the Ashanti Region [27] to a high of 54.3% in adults above 65 years in the same Region [70]. A combined total of 30,033 participants across twenty studies reporting on the population prevalence of hypertension were pooled with 10,625 (35.4%) identified to satisfy study criteria for elevated BP. A few studies reporting only mean systolic and diastolic BP were excluded from pooled analysis. After fitting a random effects model to 24 representative studies, the composite prevalence of elevated BP in the general population of Ghanaians stood at 30.3% (95% CI 26.1–34.8%) (Fig. 3). Two nationwide studies (9195 participants) in adults above 50 years gave rise to a higher pooled estimate of hypertension of 49.5% (46.2–52.7%) [23, 25]. The pooled estimate did not change significantly between studies using manual [32.2% (95% CI 26.1–38.9%)] and electronic measuring devices [29.5% (24.5–35.1%)] (p = 0.527).

After fitting a random effects model to 17 exclusive studies with sex-stratified rates, the prevalence of elevated BP was 30.6% (95% CI 25.6–36.0%) among females (n = 15 data points) and 34.0% (95% CI 28.5–40.0%) among males (n = 16 data points) (Fig. 4). In general, hypertension prevalence was higher in males (see Additional file 6) [19, 20, 23, 26, 53, 67, 70, 72, 74, 76‐78]. This trend was reversed in two studies which reported narrowly higher hypertension prevalence in females [21, 68].

Temporal variations in the prevalence estimates were not statistically significant (p = 0.624). Two identical studies were published in 2000–2005 with a combined estimate of 28.0% (95% CI 25.9–30.3%) [74, 78]. A further six studies were included from the next 5-year period (2006–2010), giving a pooled estimate of 29.3% (95% CI 23.9–35.3%). In the next 5-year window (2011–2015), another 6 published studies gave a pooled estimate of 35.5% (95% CI 24.5–46.1%). Similarly, there was no significant time-trend when the earliest year of sampling was considered instead of the publication year in the random effects model (p = 0.829). The pooled estimate for eleven studies conducted in the past decade (2010–2019) was 30.4% (95% CI 24.4–37.2%). For comparison, another eleven studies were conducted before the 2010 population and housing census giving a pooled prevalence of 31.5% (95% CI 25.0–38.9%).

In terms of the developmental indices of populations studied, the highest pooled prevalence of hypertension was seen in studies comprised of a mix of urban and rural dwellers (38.7%; 95% CI 31.4–46.7%). The prevalence of elevated BP in urban populations (31.7%; 95% CI 28.1–35.5%) was significantly higher than in rural populations (23.4%; 95% 18.6–28.9%) (Fig. 5). Most of the studies in rural areas were among the general population: just one study in eight recruited participants who are above 50 years [67] with the potential of skewing the result.

Prevalence of hypertension increased in the north-to-south direction. The highest prevalence of hypertension was (30.7%; 95% CI 25.8–36.2%) observed for southern Ghana. This was significantly higher than that in the other northern belt 22.9% (95% CI 20.3–25.9%) of Ghana and only marginally higher than that for the middle-belt 30.1% (95% CI 25.4–35.4%).

There was a distinguishable age pattern of hypertension in which prevalence of elevated bp increased with age, peaking in middle-aged individuals [51, 53, 76]. Statistics on the age distribution of participants were extracted to classify studies based on the following criteria: (1) retirees where selection criteria specifying advanced age > 50 or > 65 was used, and (2) senior citizens where the mean ages range from 67.2 to 74.4 years, and (3) studies in the general population with mean ages ranging from 31 to 54.7 years. The pooled prevalence of hypertension from studies focussing on adults was 43.9% (95% CI 36.2–51.9%) while general population studies that recruited younger participants on average gave an estimated prevalence of 27.4% (95% CI 24.5–30.6%). In addition to overall prevalence estimates, some studies also provided age-specific prevalence rates for hypertension. In general, the prevalence of hypertension was always lower in the youngest age group than in the oldest age group [51, 53, 76].

A weighted regression technique was performed to understand systematic differences among studies that potentially explained the heterogeneity between study results (see Additional file 7). In univariate analysis, type of study population (rural vs. urban) (R² = 31.4%, p < 0.05) and age structure of the population (R² = 47.2%, p < 0.05) accounted for significant variability in the estimated prevalence of hypertension. All other study characteristics were not responsible for significant variation in hypertension prevalence. A multivariate model based on significant determinants of hypertension in preliminary analysis across studies accounted for 65% of the variability in hypertension prevalence across studies (p < 0.05).

A total of 69 full-text articles were perused for eligibility and assessment of bias according to a suitably validated tool [42]. Available articles were rated as having “low,” “moderate,” “high,” and “very high” risk of bias by consensus of two investigators. The results of this appraisal are presented in supplementary file (Additional file 1). Forty-two (42) full-text articles were adjudged to have a very high risk of bias and were discarded, leaving 27 for the meta-analysis. Out of this number, 7 studies [23, 51, 69, 70, 73, 74, 76] were rated low risk of bias, 18 studies [19‐21, 25‐28, 30, 52, 53, 65, 67, 72, 75, 77‐79] were rated moderate risk, and two studies [62, 68] were rated high risk of bias.

In terms of quality, the pooled estimate presented here can be adjudged to have a bias rating of 3 + on a scale ranging from 1 + representing an estimate with a low quality and 4 + representing the highest quality. This rating reflects that all the included studies were observational studies. Although some level of bias may arise from the lack of blinding, concealment, and randomization to treatment, selected studies presented adequate sample sizes and simple/systematic random sampling to mitigate bias to moderate levels. In addition, most of the samples were sufficiently representative of the study populations from which they were obtained. In terms of consistency, we assign a low rating based on the elevated I² statistic. An elevated I² statistic and low Q statistic were indicative of a high proportion of variability in the pooled estimate attributable to heterogeneity of reported estimates. A relatively small confidence interval indicated a moderate level of precision in the effect estimate. In terms of the risk of the influence of publication bias from included studies, after examination of funnel plots and based on the results of Egger’s tests, a moderate level of confidence may be exercised when interpreting the pooled prevalence. The authors have provided subgroup analyses to supplement the interpretation of the pooled prevalence by circumventing some degree of heterogeneity in the composite value.