Ethics statement

Ethical approval was not needed as the aim of this paper was to systematically review the literature on the prevalence of T2DM in the GCC countries.

Review questions

A literature search was used to identify material relevant to the following review question:

What is the prevalence of T2DM in the GCC countries?

Search

We developed a systematic review protocol (see appendix S1) using the Centre for Reviews and Dissemination guidelines [9]. Medline and Embase were searched separately on 15/07/2009 and the search was repeated on 03/07/2010 (via Dialog and Ovid, respectively; 1950 to July week 1 2010, and 1947 to July 2010) using terms identified from PICOS deconstruction of the above review questions such as diabetes mellitus, non-insulin-dependent, hyperglycemia, prevalence, epidemiology and Gulf States, and database- and manually-derived alternatives (appendix S2). The search strategy (see appendix S3) was trialled, reviewed by independent professional colleagues (E.H., K.P.), and updated before use. Further relevant studies were identified by searching the reference lists of the database-derived papers, contacting expert investigators, screening conference proceedings, citation searching and hand searching the International Journal of Diabetes and Metabolism and the Saudi Medical Journal, for the periods 1993–2009 and 2000–2010, respectively.

Selection

The initial search produced 792 studies. After excluding duplicated studies (17 studies), the titles and abstracts were evaluated by one reviewer (L.A) to determine eligibility for full text screening. No limitations on publication type, publication status, study design or language of publication were imposed. However, we did not include secondary reports of prevalence, such as review articles without novel data synthesis. The inclusion criteria required that the study population be of a GCC country, but otherwise all ages, sexes and ethnicities were included, resident and migrant populations, urban and rural, of all socioeconomic and educational backgrounds. Studies of general-, working-, university- and healthcare attending- populations were included. We did not specify diagnostic criteria, but required that they would detect at least predominantly type 2 (rather than other forms of) diabetes, and they were incorporated into our data synthesis.

Twenty-eight studies were identified. The full texts of these studies were each examined by two reviewers (L.A and A.Mc). One study [10] was excluded as the data were already included in other studies [11]; [12], and no further (relevant) synthesis had been performed. The full text of a further study [13] could not be accessed, thus the abstract alone was used for review. Additionally, we could not fully access the data published in 3 studies [14]–[16], and the extracted data were therefore similarly limited. The selection process is summarised in Figure 1 and Figure S1.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 1. Flow chart of study selection process.

https://doi.org/10.1371/journal.pone.0040948.g001

Data extraction/quality assessment

The data captured for each study included data relating to, (1) methods (study design, recruitment, measurement tools, analysis), (2) participant characteristics, (3) setting, and (4) outcomes (those observed, their definitions, results of analysis). Study quality was assessed using a checklist adapted from the Centre for Reviews and Dissemination guidelines [9]. Data extraction was performed, in duplication, by two reviewers (L.A and A.M) (see Table S1).

Data synthesis

We were looking to estimate the prevalence of T2DM in the GCC countries between 1980 and 2009. To estimate the prevalence of T2DM, the related data was entered to STATA version 11 for statistical analysis. To assess the difference in prevalence of T2DM between different GCC countries and years, and to investigate the reasons for heterogeneity between the studies included in the review, a subgroup analysis was carried out. Subgroup analysis was performed for each country separately, and for years, which were classified as: (1) 1980–1989; (2) 1990–1999; and (3) 2000–2009. Publication year was used instead of the definite start year, as the last was not indicated in many studies. Furthermore, the measures of uncertainty (95% confidence intervals (CIs)) were calculated as meta-analysis was not appropriate to be conducted for the included studies in this systematic review due to the due to sampling, methodological and statistical variation of the studies [17].

Further data synthesis was designed around several proposals produced, for the most part, a priori, but also included an appraisal of potential association between diabetes and urban/rural residency, after preliminary scoping searches demonstrated that data pertaining to this were commonly reported.

These proposals were therefore:

1. prevalence of T2DM is increasing.
2. rates of T2DM in the GCC states are similar.
3. prevalence increases with advancing age.
4. there is a sex difference in prevalence.
5. there are differences in prevalence between urban and rural populations.
6. there are differences in prevalence between national and expatriate populations.

In addition, prevalence in children was separately considered.

Association of diabetes with BMI and relationship to income/socioeconomic status were not examined in this review.

In consideration of the above proposals, synthesis included summarising the results of the data extraction process, considering the strength of evidence relating to each of our questions, and examination of results inconsistent with our formed suggestions (Figure S2).

Prevalence of T2DM and association with time and countries

21 studies were included in the sub-analysis based on methodological consideration [16], [18]–[38]]. The sub-analysis suggested that estimated prevalence had increased across the three time periods listed in the data analysis section respectively (3.58% [95% CIs, 1.94–5.23; 2 studies]17, 18; vs. 4.01% [95% CIs, 3.58–4.43; 10 studies] 11,16,19,20,21,24–28; vs. 5.06% [95% CIs, 4.02–6.09%;10 studies ] 12, 22, 29, 30–33, 35, 36, 37). The differences in the estimated prevalence rate of T2DM in the GCC countries between the three periods was not statistically significant p = 0.9.

Subgroup analysis by country indicated that the estimated prevalence rates of T2DM between GCC countries are comparable. The lowest estimated prevalence rate was found in KSA 4.01% [95% CIs, 3.60–4.43; 10 studies] 16, 18–21, 25–28, 31; followed by Oman 4.5% [95% CIs, 3.16–5.85; 4 studies] 33, 11, 12, 22. Bahrain, in contrast, had the highest estimated prevalence rate of T2DM among GCC countries at 5.17% [95% CIs, 2.48–8.93; 2 studies] 28, 29. However, the estimated prevalence rates between Qatar, UAE and Kuwait were close (5.12 [95% CIs, 0.39–9.85; 1 study] 38; vs. 5.10% [95% CIs, 2.90–7.30; 3 studies] 35, 36, 31; vs. 5.14% [95% CIs, 1.45–8.82; 1 study] 32; respectively).

Prevalence of DM and age

Four studies (all studies in which testing was well described) demonstrated a significant association between advancing age and prevalence of diabetes [25], [36]–[38]. There was otherwise, where reported, an apparently similar association of unclear significance, or in some cases, such an association until 40 – 49 [38], 59 [13] or 60 [15] years, after which point the prevalence appeared to decrease, or fluctuate. Fatani et al [19], report an association (multiple logistic regression analysis; P<0.0001) between age and blood glucose levels.

Prevalence of DM and sex

Significant sex differences were reported in 6 studies (including that of schoolchildren). All except one relatively old report [19] were in favour of a male predominance [21], [24], [27], [31], [35]. In 9 further studies, however, higher prevalence, of undetermined significance (or close to significance: [29]), was observed in females. This was the case for males in 2 studies. A further 3 studies showed no significant gender difference.

Prevalence of DM and residential environs

Urban versus rural prevalence was commented on in 5 studies [27], [22], [31], [34], [21]. All except the oldest study [21] demonstrated higher prevalence in urban.

Prevalence of DM in children

Prevalence in children was consistently reported to be low: 0.035% [35], 0.027% [24], 0.033% and 0.099% (in urban and rural populations, respectively; [22]).

Prevalence of DM in national/expatriate populations

The prevalence of diabetes in UAE-resident expatriate populations, versus that in UAE citizens, was considered in only one study [32]. The UAE citizens appeared to have relatively high rates of disease, although no statistical methods were employed to test this suggestion.

Quality consideration

There were some inconsistencies in our tabulated results: both generally, and within the country of investigation. The studies of El-Hazmi et al [16] and Mahfouz et al [23] produced relatively low results, inconsistent with the general trend. The El-Hazmi et al [16] sample is 39.1% children <14 years, which may account for the low rates. The authors report a ‘significant’ increase in prevalence with age, but we could not access the full data and the statistical methods used were not well-described. We have suggested a higher prevalence with advancing age of unclear significance, but with rates of 0.12% and 0.79% in those <14 and 14 – 29 years, respectively, and rates of 28.82% (males) and 24.92% (females) in those >60 years, this is potentially rather conservative. We believe that this study is consistent with the others listed, but it does highlight that a number of other prevalences we have tabulated are also relatively low because they include children [19], [20], [21]. Indeed, the prevalences in these populations are much higher when children are removed from the calculation, although insufficiently to interrupt the general trends observed. This cannot however, explain the low rates reported by Mahfouz et al [23], and so we consider that these may be due to inclusion of only previously diagnosed people with diabetes (and omission of the often substantial ‘undiagnosed’ population), but concede that the result could still be relatively low, and of importance given the sample size.

We cannot extend our observed association between type 2 diabetes and age to children. Low prevalences of T2DM in children have been reported since the late 1990s (date of first identified study). However, the data are few; insufficient to evaluate the possibility that prevalence in children is increasing, as has been observed in other countries [42], [43].

The relationship between T2DM and sex was unclear. We noted, where tested, predominance in males. Wild et al [7] have reported this to be the case (‘globally’) for individuals <60 years. Even where these differences may exist, however, they generally appear to be slight. By contrast, we did observe higher prevalence associated with urban (cf. rural) residence, which again has been observed by others e.g. [44], [45].

Only a few of our included studies excluded patients with type 1 diabetes (including the study in schoolchildren) and/or pregnant women. It is therefore likely that in the majority of samples tested, the prevalence suggested includes small numbers of type 1, gestational, and potentially other forms of diabetes.

The majority of studies relied at least in part on the various World Health Organization (WHO) criteria for diagnosis. There is mild variation in definition by edition of WHO criteria, with discrepancies producing differences in estimations of the extent demonstrated by Al-Lawati and Mohammed [24], and only the later editions (1998 onwards) are consistent with those of the American Diabetes Association. Some studies, however, used definitions of diabetes and methodological approaches that led to results relatively difficult to use comparatively. Some relied on previous records alone to make diagnoses (where the diagnostic criteria used were often unclear), and so potentially missed an ‘undiagnosed’ section of the diabetes population, which has been reported to be potentially substantial [12], [14], [30], [33], [38], [39]. By contrast, relying purely on blood sampling to estimate of prevalence may have missed a significant number of cases of treated, well-controlled disease [21]. In addition, one study [37] omitted patients with secondary dyslipidaemias and may therefore have underestimated prevalence; another [15] measured prevalence in primary care consultations rather than prevalence within the population. There were also concerns that loss of difficult cases to secondary care [29], or identification only of cases sufficiently severe to merit secondary care [23], may have resulted in estimated prevalences providing relatively poor estimates for the general population.

Limitations of review

The heterogeneity of the reviewed studies, and variable availability of sub-group data, was a major limitation in our review process. All of our reviewed studies were published in English. Clarity of reporting was a relatively frequent quality issue, but we did not exclude any studies on this basis. Indeed, with such paucity of data and inability to draw more than broad conclusions anyway, we included even studies without full data availability, and one where only the abstract could be accessed.

Although five studies [26], [29], [30], [36], [38] had high rate of loss of follow-up (>20%), they were included in the review. In four of these studies [26], [29], [30], [34], the target number of subjects that were supposed to take part in these studies was unreachable. For instance, in two studies (30 and 36) 382 and 861 subjects respectively were not resident at the address given. Other reasons for high loss to follow-up were participants' death, travelling abroad, refusal to participate in the studies, and exclusion based on health grounds. Bener et al. [38] was included although details on the high rate of loss of follow up was not mentioned because it is the single study carried out in Qatar, and excluding it from the review would not help us draw an estimated prevalence rate of T2DM in this country. All of these factors impact on the strength and confidence of our proposals.

Implications

The relatively high levels of T2DM in the GCC region, and increasing prevalence, suggest that novel, or more widespread, management strategies will be important to averting an increasingly unmanageable problem. This may be particularly so given the observed associations with urban residence and age, within a context of continued urbanisation and unfavourable trends in population demography. The nature of the problem is probably similar across the different GCC states (with the possible exception of Oman, for which data are limited). Potentially, then, cross-implementation of management strategies would provide relatively high levels of success, and a co-ordination of effort would likely be relatively cost-effective. Cost is particularly important given the size of the problem, the observed impact on the working population, and the nature of migrant populations within the GCC region.

The migrant populations contribute greatly to the currently high rates of population growth in these countries. General prevalence could thus be hugely influenced by differential disease rates between national and expatriate populations. This is particularly important to estimates of future rates of disease as these are usually based – at least in part – on predicted changes in population demographics. As the GCC countries have strict naturalisation policies, and the vast majority of expatriate workers are not national citizens, these countries are at relatively high risk of fluctuation in population size and structure, and predictions regarding demography are thus relatively difficult to make. Economic change could have a particularly strong impact on population structure, and building such possibilities into strategies for disease management, when this is itself of significant economic status, is important.

Given these issues, we find the observed infrequent consideration of ethnicity as a variable particularly striking, and anticipate that continued study of this issue would be useful. Study of physical inactivity – another risk factor for T2DM – may also be useful. Finally, we expect that study of prevalence in children would be helpful, particularly given the recent rise in childhood prevalence reported elsewhere, as the available data are minimal. Longitudinal studies in both children and adults are desirable, as longitudinal data are lacking and such studies would be the optimal way to observe changes in prevalence with time.