Different online databases (PubMed, HINARI, Scopus, EMBASE, Science direct, Cochrane library database, and African Journals online), Google Scholar, other open gray literatures, and online university repositories were retrieved to include articles conducted on preventive practice toward COVID-19. We have developed different Boolean operators to have comprehensive datasets on preventive practice toward COVID-19. Search MeSH terms: Wuhan coronavirus” OR “COVID-19” OR “novel coronavirus” OR “2019-nCoV” OR “coronavirus disease” OR “SARS-CoV-2” OR “SARS2” OR “severe acute respiratory syndrome coronavirus 2” AND “preventive practice” OR “practice” AND “associated factors” OR “risk factors” OR “determinants”) [Additional file 2]. This study involved studies conducted across the globe to assess the level of practice toward preventive measures of COVID-19 among the adult population.

In this systematic review, all observational studies (i.e., cross-sectional, case-control, and cohort), studies reported the level of preventive practice toward COVID-19 and its determinants, and studies published in English language were eligible for this study. Besides, it also included all studies involving the adult population without restriction on the year of publication. However, we excluded studies other than observational studies (case reports, conference reports, and expert opinions) and the studies did not undergo a peer review process.

This study had two main outcomes. The first outcome was to determine the pooled level of practice toward preventive measures of COVID-19 among the general population. In this study, the adequate practice level was measured by including studies that were correctly classified the level of practice using the median (50%) score or above. Then, the pooled estimate of preventive practice was calculated by dividing the number of population with adequate practice by the total sample size multiplied by 100. The second outcome of the study was to identify determinants of preventive practice using the pooled odds ratio with the corresponding 95% confidence interval.

Note: in this study, the general population is defined as all population other than health care professionals who are assumed to have better knowledge and practice toward COVID-19 compared to the general population.

We had collected the findings of all online databases and eligible articles and it was exported to Microsoft Excel 2016 spreadsheet. Two authors (AWT and SB) extracted the data and reviewed all screened articles. The quality of the included articles was assessed using the Newcastle Ottawa Quality Assessment Scale (NOS) for observational studies (i.e., cross-sectional, case-control, and cohort) was employed (19, 20). The studies with NOS scores of six or more were considered a “good” quality study (low risk) while studies scored less than six were considered as “poor” quality study (high risk) (20) [Additional File 3]. However, all retrieved articles had a score of six or more NOS scores.

The extracted data were entered into a Microsoft Excel Database and then it was imported into STATA version 14.0 software with meta-analysis package for further analysis. We had performed a narrative description of the study population, the studies included, the risk factors identified, and the determinants of preventive practice toward COVID-19. The pooled estimate of the level of preventive practice toward COVID-19 was calculated using the random-effects models (21) at 95% confidence intervals. Moreover, the pooled odds ratios were determined with the corresponding 95% confidence intervals for its determinants.

The Cochran's-Q statistic and I² statistic tests with the corresponding p-values (22) were used to determine the existence of heterogeneity between studies. In this study, a value of I² 25, 50, and 75% were used to declare the heterogeneity test as low, moderate, and high heterogeneity, respectively (22). As a result, we had conducted subgroup analyses, meta-regression, and sensitivity analysis to handle the heterogeneity.

Publication bias was examined by visual inspection of funnel plots (23) and Egger's test (24). Hence, a p-value of < 0.05 was considered indicative of statistically significant publication bias.

In this review, 1,431 studies were retrieved from international databases, African Journals, online, Google Scholar, open gray, and online repositories. The accessed articles were focused on the level of practice toward the preventive measures of COVID-19 and its determinants among the general population. Furthermore, extended references were reached from the published articles. All of he retrieved articles were exported into endnote X8 reference managers and 1,180 articles were removed due to duplication and 152 articles were excluded after review of their titles and abstracts. Therefore, 99 full-text articles were assessed for eligibility and 48 articles were also excluded due to different reasons (i.e., abstracts, case-reports, conference reports, language, and experimental studies). Finally, 51 studies were met the inclusion criteria to undergo the final systematic review and meta-analysis (Figure 1). In this review study, 88,255 study participants were included from 51 observational studies conducted across the world.

The level of practice toward the preventive measures of COVID-19 varies from country to country. In this study, 51 observational studies conducted across the world were included to estimate the level of practice toward preventive measures of COVID-19 among the general population (13, 25–74). Thus, the overall pooled level of adequate practice level toward the preventive measures of COVID-19 was 74.4% (95% CI: 70.2–78.6%, I² = 99.7%, P < 0.001) using a random effects model [Additional File 4].

Different techniques were applied to handle the high level of heterogeneity between the included studies. These include using random effects model, subgroup analysis, meta-regression, and sensitivity analysis.

In this study, subgroup analysis was done based on the region category (i.e., low-income, middle-income, and high-income) of the countries where the included studies were conducted and sample size category (i.e., sample size <380 and sample size ≥380). As a result, the pooled level of practice toward the preventive measures of COVID-19 in low-income, middle-income, and high-income countries was 69.0% [95% CI 62–76: I² = 99.1%, p < 0.001), 81.0% (95% CI 75.0–87.0: I² = 99.4%, p < 0.001), and 78.0% (95% CI 70–86: I² = 99.8%, p < 0.001) respectively [Additional File 5]. Regarding sample size, the pooled level of practice on the preventive measures of COVID-19 was 81.0% (95%CI 75.1–86.0: I² =75.5%, P < 0.01) and 74.0% (95%CI 70.0–78.0: I² = 99.7%, P < 0.001) among studies involving fewer than 380 and 380 or more study participants, respectively [Additional File 6].

To identify the presence of publication bias, both a funnel plot and Egger's test were performed. Visual inspection of the funnel plot showed an asymmetrical distribution, which indicated the presence of publication bias (Figure 2). The finding of publication bias was confirmed following the Egger's test (p <0.013).

In this review, the authors confirmed that the presence of significant publication bias that may be subjected to unpublished small studies. Thus, to handle this problem, the authors did trim and fill analysis and 19 studies were filled (Figure 3).

We conducted a sensitivity analysis to assess the effect of any individual study on the pooled effect size. However, the sensitivity analysis done using a random effects model revealed that no single study affected the overall level of practice with the preventive measures of COVID-19 (Additional File 7).

To investigate the possible source(s) of variation across the included studies, we performed meta-regression analysis using region (high, middle, and low-income), sample size, and quality of the score as covariates of interest. Thus, the results of this meta-regression analysis showed that region category was significantly associated with the presence of heterogeneity (p <0.028) (Table 1).

In this study, thirteen studies were included to assess knowledge level as a predictor of practice of preventive measures toward COVID-19 (26, 28, 34, 37, 41, 42, 44, 46, 49, 53, 65, 73, 74) by using random-effect model analysis. However, still we were unable to handle the heterogeneity between studies [I² = 86.2%, P < 0.001]. Hence, subgroup analysis was done using country development category [i.e., low-and high-income]. Hence, participants from high-income countries and with adequate knowledge had 22% more likely to have adequate practice toward COVID-19 compared to their counterparts [OR = 1.22: 95% CI 1.09, 1.36; I² = 47.3%, P = 0.077] (Figure 5).

Eleven studies were included to assess residence in rural areas as a negative predictor of adequate practice of preventive measures toward COVID-19 (13, 26, 39, 44, 45, 48, 49, 51–53, 74) by using a random-effect model that was applied to handle the variation between studies [I² = 71.8%, P < 0.001]. Hence, people living in rural areas had 47% less likely to have adequate practice on the preventive measures of COVID-19 compared to their counterparts [OR = 0.53: 95% CI 0.44, 0.65; I² = 73.5%, P < 0.001] [Figure 6].

Seven studies were included to assess education level as an independent predictor of level of practice toward COVID-19 the preventive measures (13, 26, 33, 37, 41, 46, 50). The meta-analysis results revealed that participants who had attended higher education levels were 47% more likely to practice preventive measures toward COVID-19 compared to their counterparts [OR = 1.47: 95% CI 1.18, 1.83: I² = 75.4%, P < 0.001]. We applied random-effects model to handle the variation between studies (Figure 7).

In this study, eight studies were included to assess being employed as a predictor of adequate practice toward the preventive measures of COVID-19 globe (27, 28, 33, 41, 42, 48, 50, 53). As a result, employed participants had twice higher odds of adequate practice compared to unemployed participants [OR = 2.12: 95% CI 1.44, 3.12, I² = 91.2%, P < 0.001] using a random-effects model analysis (Figure 8).

In this study, nine studies were included to assess the association between the age of the participants and the level of practice of preventive measures toward COVID-19 (13, 27, 28, 34, 37, 46, 50, 54, 55). Hence, young age [<30 years of age] participants had 27% lower odds of adequate practice toward the preventive measures of COVID-19 compared to their counterparts [OR = 0.73: 95% CI 0.60, 0.89, I² = 73.9%, P < 0.001]. Besides, the random-effect model was employed to handle the heterogeneity between studies [Figure 9].

The first limitation of the study was only English articles or reports were considered to carry out the analysis. Even though the quality of each study was assessed by using The Newcastle-Ottawa Scale, inter-author bias might be occurred with the leveling of the scale of each article. Reviewing of different characteristics of the involved cases with different sampling methods was also the other limitation of this study.