Dental caries and associated factors in Ethiopia: systematic review and meta-analysis

Studies delineating the burden of dental caries in Ethiopia were consistently and exhaustively searched using medical specialty databases of ScienceDirect, PubMed, Google scholar, Embase, HINARI, and Cochrane Library based on the Preferred Reporting Items for Systematic Reviews and Meta-analysis (PRISMA) guideline [28]. This study was not preregistered. Published and unpublished data (grey pieces of literatures) available on the local university shelves with epidemiological data of dental caries and associated factors done in the Federal Democratic Republic of Ethiopia from 2000 up to 30 September 2020 were incorporated into the review. Citations known by our search terms were taken to EndNote–X7, and duplicate articles were removed. The complete texts of chosen articles were retrieved and read meticulously to determine the quality of the paper. The EndNote software statistical package was employed to cite and download articles. Studies were searched by exploiting the next keywords one by one or in combination: “prevalence of dental caries”, “magnitude of dental caries”, “epidemiology of tooth decay”, “Ethiopia”, “factors related to dental caries”. The search terms were used individually and along with exploitation Boolean operators like “OR” or “AND”.

We included studies delineating the burden of dental caries and associated factors no matter the measurement tool employed; publication type: journal articles; study subjects: all peoples, irrespective of their occupation and sex as well as articles revealed online in the English language were considered; and place of study: solely carried out in Ethiopia from 2000 to 30 September 2020 was included. Articles with methodological issues, incomplete data, and complete text not accessible were excluded from the analysis.

The extraction of information (data) was performed by four researchers separately by employing a format ready in Microsoft Excel spreadsheet. In the Microsoft Excel spreadsheet, the following information were incorporated: name of author, publication year, area/region, employed study design, age range, actual sample size, response rate and number of study subjects with the outcome of interest, prevalence rates as well as associated factors considerably related to dental caries. Those articles identified by abstracts and titles were closely reviewed to get back studies with the burden of dental caries. Relevant articles by titles and abstracts were selected for complete text review for its quality and inclusion. Quality of eligible studies was check by employing Newcastle-Ottawa Scale before analysis [29]. Prime quality articles were selected if the scale score was more than half out of 10. Four researchers performed the selection and quality of incorporated articles individually. Discrepancy among the researchers have been resolved through discussion, and articles were included once accord.

The outcome of interest was dental caries dichotomized as present or absent after rigorous oral examination. Physical examination was performed by dental health professionals by employing necessary instruments. The prevalence was found by dividing the total number of study subjects with the outcome of interest by the total number of study subjects involved in the study multiplied by 100%.

The 2nd outcome of interest in this meta-analysis was factors associated with dental caries closely revealed in three or more studies. The adjusted odds ratio of contributing factors were taken seriously from included studies to determine the most related factor with dental caries in meta-analysis.

Extracted data from primary study by employing a format ready in Microsoft Excel spreadsheet were imported to the STATA version-13 statistical software package for meta-analysis. A meta-analysis of dental caries was performed using random-effects (DerSimonian and Laird) method to adjust for the determined variability [30, 31]. In studies that did not delineate standard error (SE), a SE was calculated in Microsoft excel. And then, the calculated standard error as well as prevalence of every study was imported into the STATA version 13 software to calculate the pooled prevalence rate with 95% CI.

Publication bias was assessed by using a funnel plot through visual assessment. Asymmetry of the funnel plot showed the existence of potential publication bias [32]. Egger’s and Begg’s tests at 5% significant level were also performed; distribution of every study and a P-value of less than 0.05 were employed to announce clear existence of publication bias [33]. The non-uniformity among studies was assessed employing Cochran’s Q test (P-value less than 0.1 revealed as there was statistically considerable heterogeneity) and inverse variance (I²) test statistics (which was used to quantify the percentage of total variation in the study estimate because of non-uniformity).

I² worth ranges from 0 to 100%. I² ≥ 75% indicates high heterogeneousness across studies, and P-value < 0.05 was used to announce the presence of a statistically significant heterogeneity [34, 35]. Pooled effect size was calculated, and subgroup analysis was done based on geographical setting (regions) to reduce the random variations between the point effects of the original study. To know the source of non-uniformity, meta-regression was done. Furthermore, point prevalence with 95% CI was displayed on forest plot. On forest plot, the dimension of every box shows the weight of the study. To know the possible associated factors, overall effect was connected in the form of adjusted odds ratio. To identify the effect of every study on the pooled effect size, sensitivity analysis was performed. Four investigators individually done the statistical analysis, and the results were crosschecked for consistency. The included studies’ risk of bias was assessed by employing a 10-item rating scale prepared by Hoy et al. for burden studies [36].

Information collection, sampling, reliability, validity of measurement tool, definition of case, and burden periods of studies were checked. Investigators classified every study as having low risk of bias or “yes” answers to domain questions, high risk of bias or “no” answers to domain questions. Every study was allotted a score of 1 (yes) or 0 (no) for every domain, and these domain scores were summed up to supply the total study quality score. Scores of 8–10 were thought of as having a “low risk of bias,” 6–7 a “moderate risk,” and 0–5 a “high risk”. For the ultimate risk of bias categorization, differences among the reviewers were resolved through agreement.

Electronic medical specialty database search engine produced a cumulative of 451 records; from these, 70 duplicate records were identified and rejected. Title and abstract selection triggers rejection of 350 unwanted articles. Then, 31 records underwent complete text review. From 31 records, 18 articles were excluded based on the eligibility criteria; from these, one article was not accessible in full text [9‐12, 37‐50]. Lastly, 13 articles were incorporated in the final meta-analysis; two of them were pre-prints (Fig. 1). These thirteen studies had a cumulative of 6950 study subjects which were considered in this meta-analysis to determine the overall prevalence of dental caries. Of those, two [20, 25] were conducted in Addis Ababa, six [16‐19, 21, 23] in Amhara region, two [14, 24] in Tigray region, one [15] in Oromia region, one [13] in Southern Nations and Nationalities of Ethiopia (SNNPR) region, and one [22] in Harar town. Cross-sectional study design was employed in every study, and studies were conducted among different age groups (their age range 6–100 years) (Table 1). A large proportion of studies were conducted in Amhara region. Ten studies had a sample size above three hundred. The least reported response rate was 82% [18]. The minimum and maximum sample size was 147 and 1736 respectively [18, 25]. All studies were carried out between 2000 and 2020.

Overall pooled prevalence of dental caries was found to be 40.98 (31.62, 50.34). The test statistic showed high non-uniformity among every study (I² = 98.6%, P = 0.000) (Fig. 2). Due to this reason, random effects model was accustomed to estimate the DerSimonian and Laird overall effect. Furthermore, during this meta-analysis, subgroup analysis was strictly performed depending on the region of the studies conducted. The highest pooled prevalence of dental caries was determined in Tigray region of Ethiopia (46.59% (24.64, 68.54)) and the lowest prevalence of dental caries was determined in Addis Ababa (34.20% (8.42, 59.97)) (Fig. 3). On the subgroup analysis, the output still revealed as there was heterogeneity across studies. We carried out meta-regression analysis employing sample size and publication year as a covariate (Table 2). None of these variables were statistically considerable source of non-uniformity. Also, sensitivity analysis was carried out to identify the effect of every study on the overall effect size. The result of this analysis did not show a single study that considerably influenced the cumulative pooled prevalence of dental caries (Fig. 4). We have checked publication bias by looking at the funnel plot for its symmetry (Fig. 5).

The funnel plot revealed that there was symmetrical distribution of incorporated studies. Begg’s test showed absence of publication bias (Pr > |z| = 0.393).

However, the output of Egger’s test was statistically considerable for the existence of publication bias (P = 0.030). And also, we have done trim and fill analysis (Fig. 6). After meta-trim and fill analysis, the pooled value is 40.984 with 95% CI (31.623, 50.344). There is no significant change because the confidence intervals of the two findings (before and after meta-trim) overlapped.

Four studies have been evaluated to check the presence of association between sweet food consumption and caries [19‐22]. Dental caries prevalence was considerably high among study subjects who consumed sweet food (OR= 2.4 (95% CI (1.91, 3.01))). The analysis of three studies [15, 18, 19] found that presence of dental plaque was not considerably associated with dental caries (OR = 5.14 (95% CI (0.67, 39.39))) (Fig. 7). The pooled regression analysis of five articles [16, 18, 20, 22, 25] showed no association between habit of tooth cleaning and dental caries (OR =0.71(95 %( CI (0.17, 2.96))). Non-uniformity tests disclosed that there was high non-uniformity in studies that evaluate habit of tooth cleaning (I² = 97.5%, P = 0.000). No publication bias was detected for all factors as manifested by Egger’s regression test (Egger’s test = 0.074).