Education level and misuse of antibiotics in the general population: a systematic review and dose–response meta-analysis

Numerous studies evaluated the association of education level with misuse of antibiotics by the general population, yet divergent findings were reported. Therefore, a meta-analysis was conducted to summarize this association. A categorical and continuous dose–response meta-analysis of the association of education level with antibiotic misuse was undertaken. Summary odds ratios (ORs) and their 95% confidence intervals (CIs) were estimated using random-effect model. The meta-analysis included 85 studies from 42 countries of different socioeconomic status. Compared to low education (≤ 9 years), medium education (> 9–12 years) is associated with 20% lower odds of antibiotic misuse in high-income countries (OR = 0.80; 95% CI 0.66, 0.97), while high education (> 12 years) is associated with 14% lower odds of any aspect of antibiotic misuse (OR = 0.86; 95% CI 0.72, 1.03). The association is more pronounced in Middle East (OR = 0.64; 95% CI 0.42, 1.00) and countries of lower-middle economies (OR = 0.67, 95% CI 0.41, 1.11). Inversely, in Europe, high education is associated with 25% higher odds of antibiotic misuse (OR = 1.25, 95% CI 1.00, 1.58). Each additional year of education was associated with 4% lower odds of any aspect of antibiotic misuse in lower-middle economies (OR = 0.96; 95% CI 0.92, 1.00) and in Middle East (OR = 0.96; 95% CI 0.93, 1.00). Conversely, it was associated with 3% higher odds of antibiotic storage, a specific type of misuse (OR = 1.03, 95% CI 1.01, 1.06). Individuals misuse antibiotics irrespective of their education level. Intervention programs to enhance the proper use of antibiotics should target all communities independent of their education level.


Background
Antibiotic resistance continues to represent an essential public health problem despite efforts exerted worldwide to reduce the causes of the resistance. It affects all regions irrespective of the country socioeconomic status [1], and cause heavy clinical, social and economic burdens. More than 700,000 individuals die each year due to antibiotic resistant bacteria, and it is expected that the annual mortality rate from antibiotic resistance will exceed that of major diseases by 2050 [2]. Furthermore, projections show that the economic shortfalls due to this problem will increase and will soon be equivalent to those seen in the 2008-2009 global financial downturn [3].
Antibiotic resistance is exacerbated by excessive use of antibiotics in agriculture, food and feed chain, imprecise antibiotic prescription by physicians, and misuse of antibiotics by the patients. Individuals tend to misuse antibiotics by consuming these drugs without medical prescription (self-prescription) or by using them based on medical advice but without adherence to the physician's instructions such as modifying the prescribed dose, truncating or prolonging the treatment duration or not taking the antibiotics on time [4]. The use of antibiotics without prescription is salient worldwide, with a pooled prevalence exceeding 75% in low-and middle-income countries [5], and reaching 66% in some regions of high-wealth countries such as the United States [6].
Previous reports also indicated that more than onethird of patients do not fully adhere to antibiotic treatment regimen [7], around 50% cease their antibiotic treatment upon improvement [8], and one-third store antibiotic leftover for future use [7].
Besides determinants of antibiotic misuse such as female gender, youth and old age, lack of access to healthcare facilities and easy access to antibiotics [6,9], educational level was suggested to be associated with misuse. Yet inconsistent findings exist regarding this association. Several studies reported an association between low education and antibiotic misuse [10]. Inversely, several other studies reported that high education level is associated with greater risk of antibiotic misuse [11], while some studies failed to find any association [12] Accordingly, to summarize those findings, we aimed in the present study to carry out a systematic review and dose-response meta-analysis of the association of education level with antibiotic misuse.

Methods
We registered this systematic review and metaanalysis in the PROSPERO database (Protocol ID: CRD42021233425) and carried it out according to PRISMA guidelines. The main outcome, antibiotic misuse in the general population, was defined as the occurrence of any of the following practices: unprescribed use of antibiotics (self-medication), non-adherence to treatment guidelines and storage of antibiotics leftover for future use.

Literature search and study selection
We searched Medline, EMBASE, the five regional bibliographic databases of the World Health Organization (WHO), the Conference Proceedings Citation Index-Science, and the Open Access Theses and Dissertations until January 2021. In Medline, we used the following search term without any language, date or other restrictions: (Socioeconomic Factors OR education) AND (antibiotic*) AND ((compliance) OR (adherence) OR (Nonprescription Drugs / administration & dosage* [MeSH]) OR (misuse) OR (irrational use) OR (left-over)). We also ran the search using related free-text words. Then, we adapted the syntaxis to complete the search in the other databases. We manually checked the reference lists of included studies and those of relevant review reports to supplement the electronic search. The list of the examined review reports is provided in Additional file 3. We included studies that (1) measured the association between education and any aspect of antibiotic misuse (i.e., unprescribed use of antibiotics for oneself or for another person, non-adherence to antibiotic treatment guidelines or storage of antibiotic leftover), (2) defined the measured level of education, (3) reported odds ratio (OR) or risk ratio (RR) and their 95% confidence intervals (CIs) or sufficient data for their calculation. We excluded from the meta-analysis studies that only compared students according to their university year.

Data extraction and synthesis
We collected information on: (1) study source: author's last name and publication year, (2) settings and participants' demographic characteristics, (3) study design, (4) exposure: levels of education, (5) for each education level: ORs and its 95% CIs, total number of participants, and number of individuals who reported antibiotic misuse. We extracted the ORs adjusted for the largest number of variables, (6) restriction, adjustment, or matching variables, and (7) type of antibiotic misuse. In studies reporting more than one type of antibiotic misuse, we extracted the data of all types of misuse, and treated each type of misuse as a separate study unit in the dose-response analysis. We contacted the authors to inquire about the number of individuals who misused and those who did not misuse antibiotics per each education level, when needed.
In addition to data reported in the included studies, we obtained the classification of countries wellness (low, lower-middle, upper middle and high income) from the World Bank [13] and used for geographic distribution the classification by region of the World Health Organization (African, Eastern-Mediterranean, European, Region of the Americas, South-East Asia and Western Pacific) [14].

Dose definition
We defined the term "dose" as the level of education in years. Education level classification varies between countries, hence, to standardize the education levels across studies of different parts of the world, we transformed it to years of education according to the education system used in each country. We set the dose as the midpoint of the upper and lower boundaries of each education level.

Statistical analysis
We performed a dose-response meta-analysis using a one-stage mixed-effects model taking into account heterogeneity across studies [15]. We carried out categorical and continuous approaches.

Categorical approach
To facilitate tabular presentation of the summary ORs, and in line with other studies, we further recategorized education level into low (≤ 9 years), medium (> 9-12 years), and high (> 12 years) levels and used low education level as a referent.

Continuous approach
We applied a linear function to estimate a summary OR of antibiotic misuse associated with an increase of 1 year of education. Then, we flexibly modelled education using restricted cubic splines with 3 knots fixed at 10th, 50th and 90th percentiles of its distribution, and tested departure of the second spline from linearity. We undertook stratified dose-response analyses of the level of education with antibiotic misuse. The analysis was stratified by study design, type of antibiotic misuse, geographic region, country wellness, methods of exposure ascertainment, comparability (adjustment for age and gender), and publication year, using 2015, the year of publication by the World Health Organization of the global action plan against antibiotic resistance, as a cutoff limit [16].

Quality assessment
We appraised the quality of the studies included in the meta-analysis using the Newcastle-Ottawa Scale for cohort and cross-sectional studies [17,18]. Two epidemiologists (AF and NM) performed the quality assessment, and disagreements were resolved by consensus through discussion with a third epidemiologist (BT). Seven criteria were evaluated. The following five criteria were common to cohort and cross-sectional designs: (1) justified sample size; (2) application of previously tested or validated questionnaire to ascertain education level; (3) use of external assessment in addition to questionnaire to ascertain antibiotic misuse; (4) described and appropriate statistical analysis; and (5) adjustment, matching or restriction for age and gender. Additionally, we evaluated two criteria that were specific to study design. For cohort studies we checked (1) if the study sample was representative of the general population and (2) if the response rate was more than 50%. For cross-sectional studies we examined (1) if the study population was defined; (2) if the response rate was reported. We gave one point for the fulfilment of each of the seven criteria, and then summed those points to obtain a quality score of a maximum of seven points. When the information on an item was absent in the publication, this item scored zero point.

Publication bias
We checked publication bias visually using funnel plot, and formally through Egger's test [19] and the trim and fill method [20].

Literature search and study selection
Out of 1458 identified studies, 85 fulfilled the inclusion criteria and were included in the meta-analysis (Fig. 1). The general characteristics of the included studies are summarized in Table 1 and Additional files 1 and 2 and their references are provided in Additional file 3. Out of all contacted authors to inquire about missing data in Table 1, three answered our inquiry [21][22][23]. Eighty-three studies were of cross-sectional design and the remaining two studies were cohort studies. They encompassed a total population of 85,789 subjects, out of whom 24,579 had misused antibiotics as follows: use without prescription (N = 15,780), storage of antibiotics (N = 6077), nonadherence to antibiotic treatment regimen (N = 2293) and several concomitant types of misuse (N = 429). When studies provided data for several types of misuse, each type was treated as a separate study, making a total of 94 studies introduced in the dose-response analysis. The studies were published between 2000 and 2021 and originated from 42 different countries. All studies were published in English, except five that were available in Croatian, Italian, and Spanish.

Subgroup analysis Type of misuse
The categorical and continuous approaches revealed that education level is not associated with unprescribed antibiotics use (OR of 1-year increment = 0.99; 95% CI: 0.97, 1.00) or non-adherence to treatment regimen (OR of 1-year increment = 0.99; 95% CI 0.96, 1.03). Nonetheless, high education level is associated with 41% higher odds of storage of antibiotics (OR = 1.41; 95% CI 1.22, 1.64), compared to low education. The association with antibiotic storage was also observed for medium education level, yet with smaller magnitude (OR = 1.17; 95% CI 0.93, 1.48) ( Table 2). These findings are in line with that of the continuous approach in which 1-year increment in education is associated with 3% higher odds of antibiotics storage (OR = 1.03; 95% CI 1.01, 1.06) ( Table 2). The association of education with several concomitant types of antibiotic misuse could not be determined using the categorical approach due to insufficient observations, yet the continuous approach showed that each 1-year increase in education is associated with 9% lower odds of antibiotic misuse in general (OR = 0.91; 95% CI 0.87, 0.95) ( Table 2).

Country economy
In high-wealth countries, the odds of antibiotic misuse are 20% lower in individuals with medium education than in those with low education level (OR = 0.80; 95% CI 0.66, 0.97). In countries with lower-middle economy, high education is associated with 30% reduced odds of antibiotic misuse, compared to low education (OR = 0.70; 95% CI; 0.44, 1.13). Moreover, in these countries, each  additional year in education is associated with 3% lower odds of antibiotic misuse (OR = 0.97; 95% CI 0.93, 1.01) ( Table 2).

WHO regions
In the Eastern Mediterranean region, in reference to individuals with low education level, high education is associated with 36% lower odds of antibiotic misuse (OR = 0.64; 95% CI 0.42, 1.00). In the European region, high education is associated with 25% higher odds of misuse (OR = 1.25; 95% CI 1.00, 1.58). Similarly, in the continuous approach, every additional year in education is associated with 4% lower odds in the Eastern Mediterranean (OR = 0.96; 95% CI 0.93, 1.00) and 2% higher odds in the European regions (OR = 1.02; 95% CI 1.00, 1.04) ( Table 2).

Publication year
Pooled estimates from studies published after 2015 showed 19% lower odds of antibiotic misuse by  Table 2). Studies that controlled for age and gender, as well as those considered of higher quality did not show any association between education and antibiotic misuse ( Table 2).

Publication bias
The funnel plot of studies reporting medium education level was slightly skewed to the left (Fig. 3A), but publication bias was neither confirmed by Eggers's test (p value = 0.065), nor by the trim-and-fill analysis that did not suggest the addition of any study. As for those studies that assessed high education level, the funnel plot was also slightly skewed to the left (Fig. 3B). Egger's test suggested the presence of publication bias (p value = 0.001), but the trim-and-fill analysis did not suggest the addition of any study.

Discussion
We carried out a systematic review and dose-response meta-analysis to summarize the association between education level as a dose and misuse of antibiotics by the general population. Data from studies included in the meta-analysis fit well with a linear association between education and antibiotic misuse. We found no association between a one-year increment in education and the occurrence of antibiotic misuse. In the categorical analysis, we observed that individuals with high education level are, in general, at lower odds of antibiotic misuse than those with low education level. However, these results were not confirmed for the European studies group and the storage-type group of misuse. The likelihood of using antibiotics without prescription as well as that of non-adherence to treatment is similar for low, medium and highly educated individuals. Notably, the odds of storing antibiotics at home for future need is larger for highly educated people than for those with low education level.
People with low education are more susceptible to comorbidities and thus, are more exposed to medicines than individuals with higher education [24]. Education is strongly associated with socioeconomic status, especially with income [25]. On the one hand, financially disadvantaged people regularly report forgone care [26], and shorten their treatment or buy fewer doses than prescribed, due to cost [27]. In addition, self-medication is most often the only available choice for people with limited financial resources, especially in countries with constrained access to health facilities [28]. On the other hand, individuals with higher socioeconomic status, i.e., higher education, have more social networking which favors their access to unprescribed antibiotics. Moreover, they are more likely to have better economic affordability to buy and store non-reimbursed antibiotics [29]. This could, at least partially, explain our findings concerning a higher misuse likelihood in European countries. Regulations to control the dispensing of antibiotics should be further enforced as more than half of the antibiotics worldwide are still dispensed without prescription [30].
Health literacy significantly contributes to health status and medicines use [31]. Individuals with low education level are characterized by poorer health literacy skills than those with high education [32]. The lack of access to healthcare of less educated people also reduces their health literacy [33]. Nonetheless, limited health literacy is not only restricted to people with low education [33]. In wealthy countries such as in Europe the prevalence of low health literacy ranges between 30 and 60% [34]. Indeed, the health literacy of the population is also influenced by factors other than education level, such as the ease of public understanding to the available health related information [35] as well as the proficiency of the healthcare provider in communicating the information to the patient [33,36]. Cultural differences and divergence in opinions and beliefs may also influence population´s behaviours towards a specific health issue, including towards the medicines used in it. In the context of antibiotics, it was reported that in certain settings health literacy concerning antibiotic use was insufficient among highly educated people [37]. Insufficient knowledge and misconceptions about antibiotics were also reported both in developed and developing countries [38]. The odds of antibiotic misuse by highly educated people decreased after 2015, which could be related to the global efforts exerted by WHO as well as to the educational campaigns and antibiotic stewardship programs undertaken in many parts of the world to increase the awareness about antibiotic resistance [16,39]. Highly educated people have better access to internet-based health information than socioeconomically disadvantaged individuals [40], including information on antibiotic use.
This meta-analysis has various strengths. To the best of our knowledge, it is the first to assess the association between education and antibiotic misuse. To allow for comparability across studies, we transformed the education level to years of education, adapting to the setting of each country, before establishing cut-off limits for categories of education levels. We also provided a summary measure of association per increase of 1 year of education. In addition, we presented stratified analysis by country wealth as proxy of socioeconomic status.
Nonetheless, our study suffers from limitations. All except two studies included in the meta-analysis are of cross-sectional design, a design that, theoretically, does not allow for causal inference, due to the fact that exposure and outcome are measured concomitantly. However, education is not a transient factor, but a cumulative characteristic. The fact that exposure and outcome are measured concomitantly does not imply that the level of education may have been acquired after the occurrence of misuse.
High amount of heterogeneity existed across studies. Meta-analysis experts highlight that heterogeneity is expected in any meta-analysis [41], especially in a case of meta-analyses such as ours with large variability in study setting, population characteristics, definition of education level, and period of antibiotic misuse. Therefore, we accounted for heterogeneity by applying random-effect models more adapted to meta-analyses with substantial amounts of heterogeneity.
Furthermore, not all studies provided measures of association adjusted for potential confounders or reported restriction, matching or confounding variables. Studies that controlled for age and gender yielded a summary estimate closer to the null value than studies that did not control for those variables. Likewise, around one-third of studies used non-validated questionnaires. These studies yielded a pooled estimated farther from the null value than with validated instruments.
Finally, although some elements of publication bias were observed in studies that assessed the association of high education with antibiotic misuse, this was unlikely to affect our results as showed by the absence of additional studies in the trim-and-fill analysis.

Conclusions
This meta-analysis shed the light on the importance of orienting intervention programs to improve the rationale use of antibiotics to all communities independent of their educational level. It also pointed out on the considerable need for cohort studies that examined the association between education and antibiotic misuse and control the measures of association for potentially confounding variables. Measuring the interaction between various socioeconomic indicators such as income and education on antibiotic misuse, would help understand better the socioeconomic properties of antibiotic misuse and thus allow for better control.