Self-Medication Prevalence for Sexually Transmitted Diseases: Meta-Analysis and Meta-Regression of Population Level Determinants

Abstract

Background: 

Estimated prevalence of self-medication for sexually transmitted diseases (STDs) in observational studies differs between studies and it may vary according to population characteristics.

Methods: 

We investigated the extent of self-medication use for STD and explored population and study level factors correlated to the variability observed between populations. Medical databases were systematically searched for published studies. Crude prevalence data were extracted. Pooled estimates were calculated using random effects models. Univariate and multivariate meta-regression models with categorical population level covariates were constructed.

Results: 

Of the 35 studies that met the inclusion criteria for all self-medication use, 20 were also included in the analysis of antibiotic only use. There was significant heterogeneity present across prevalence estimates for all self-medication, Q = 3954.82, P <0.001; and among antibiotic only self-medication, Q = 3797.94, P <0.001. In meta-regression analyses, publication year seemed to explain 22.5% of variation among studies of self-medication with antibiotic only. No other significant associations with population or study characteristics were found.

Conclusions: 

No population level factors leading to high levels of self-medication use for STD were identified. Publications before year 2000 seem to present a higher prevalence of antibiotic use for self-medication. Greater risk awareness of antibiotic misuse and improved control of antibiotic availability at population level might explain this result. An analysis of individual level characteristics should indicate which are influential and whether their local prevalence can explain the observed heterogeneity better.

In Brief

Meta-analysis and meta-regression of observational studies show great unexplained heterogeneity in self-medication prevalence across populations. Local prevalence of individual, instead of population level, characteristics might explain high levels of self-medication.

SEXUALLY TRANSMITTED INFECTIONS LEAD TO diseases where feelings of shame and stigma play an important role in the healthcare seeking behavior of patients. Symptomatic patients may wait for the symptoms to resolve spontaneously or may try to treat themselves. The choice of self-treatment may, in some cases, exacerbate or mask symptoms, alter diagnostic test results, and may increase the length of time an individual waits before seeking treatment.^1,2 In other cases, it may also prove effective: antibiotics of the appropriate type can treat gonorrhea in a single dose. This choice of treatment is influenced by the choice of providers, with a clear distinction between medication obtained from the formal sector (e.g. through pharmacies) and the treatment accessed through the informal sector (e.g. friends, relatives, traditional healers).

Self-medication, considered as the treatment of oneself in the absence of medical advice, including the use of prescription drugs, over-the-counter products, and home or traditional remedies, is a common practice, especially in countries where medications are readily available over the counter. The practice could influence sexually transmitted disease (STD) prevention and control programs. On one hand, self-administered medications obtained through the informal sector could compromise STD prevention and control programs by increasing the indiscriminate use of antibiotics over insufficient lengths of time (thereby promoting antibiotic resistance) and delaying diagnosis, increasing the period of infectiousness, and the probability of further STD transmission and complications.³ However, it could be useful for STD prevention and control in resource poor settings, when obtained through the formal sector, if adequate training is provided to pharmacists.⁴

Both the extent and effectiveness of self-medication practices vary according to socio-cultural determinants with a large prevalence range being reported. This large variability observed in the prevalence of self-medication might be due to population level determinants such as development status of the country where the study took place, or deferential STD risk between populations. Several narrative reviews and editorials have been published that focus on or mention self-medication but, to our knowledge, a systematic approach to understanding the extent and the distribution of this practice has not been published.^1,3,5,6 Therefore, this study assesses the spread of self-medication use for STD in the published literature; and explores possible population and study level factors related to its variability.

Materials and Methods

Literature Searches

We searched PubMed/Medline, Embase/OVID, and Web of Science for articles published since 1987, using the following terms: “self medication,” or “self-treatment,” or “healthcare seeking,” or “delay behaviour/behavior,” and “sexually transmitted diseases” or “STD,” or “sexually transmitted infections” or “STI.” There was no language restriction. Additional publications were identified through a manual search that included references from relevant articles and database searches on selected authors. The last search was conducted on the 11th March 2008.

Articles were included if they fulfilled the following criteria: they were publications in peer reviewed journals, or conference papers of primary research, defining self-medication as the use of any treatment (herbs, traditional medicine, antibiotics, or home remedies) without medical advice to treat STD symptoms; and they provided a cross-sectional estimate (or sufficient data for this to be calculated) of the prevalence of self-medication, either as a primary or secondary outcome. By limiting the study selection to only published articles, we aimed to avoid poor quality studies. However, this might lead to some publication bias. We explored publication bias graphically plotting the prevalence estimates found in the studies against sample size.

Data Extraction

We extracted, from full text articles, study and population characteristics such as geographical location, year of publication, sample size, sampling method, and self-medication prevalence data. Year of publication was used instead of year of study because study year was not available for all studies included. Sample size was defined as the number of participants in the study who were symptomatic or reported STD symptoms for which they may or may not have self-medicated. Two groups of studies were defined: those reporting prevalence of all type of self-medication use (antibiotics and/or other type) and those specifying self-medication with antibiotics exclusively. Each study contributed to one estimate of self-medication prevalence in either all or antibiotic only use analyses. Crabbe et al.⁷ was the only exception and was included twice in the analysis because this study reported the prevalence of self-medication in 2 different sample populations.

Analysis

Crude prevalence estimates (number of participants self-medicating/sample size) and standard errors were calculated. For the meta-analysis and meta-regression calculations, the prevalence (p) was transformed to logits: logodds = log(p/1-p). It was later transformed back to prevalence in percentages for graphic representations.⁸ Cochran’s χ² test of heterogeneity (Q test) was used to explore differences between the prevalence estimates.⁹

Substantial heterogeneity was identified, preventing a reasonable overall summary statistic from being calculated. Forest plots were then used to explore graphically the heterogeneity found between the estimates of all studies and between the pooled estimates by subgroups.⁸ We grouped studies by study or population characteristics and determined pooled estimates using random effects models to be able to weigh the estimates with the inverse of the study variance taking into account both within and between study variances.¹⁰ Female sex workers (FSW) studies were analyzed as a separate subcategory from male and female in the gender group. For all self-medication studies, only 9 provided stratified prevalence data by gender. The gender subgroup analysis also included 8 FSW-only, 7 men-only, and 3 women-only studies. For antibiotic only self-medication studies, 4 provided stratified prevalence data by gender. We added to this subgroup analysis: 5 FSW-only, 2 men-only, and 2 women-only studies.

Population risk was another covariate explored. Low risk were those populations with no obvious STD risk, whilst high risk included populations with factors associated with an increased STD risk, such as buying or selling sex, attendance at an STD clinic, and injection drug use. Pooled estimates were also determined by geographical regions: Asia, Africa, USA/UK, and other (Uzbekistan and Iran); and by development status: developing or developed countries. Finally, publication year was explored by grouping studies into 2 groups: before year 2000 and year 2000 and after.

Meta-regression investigates whether particular covariates could explain any of the heterogeneity observed between studies in the prevalence estimates of self-medication.⁸ Univariate and multivariate random effects regression models with categorical study and population level covariates were constructed excluding gender because this variable was only available for half of the studies.¹⁰ The 2 multivariate models included: (a) risk, region, and year; or (b) risk, development, and year. Development and region were not included in the same model to avoid confounding. Both univariate and multivariate meta-regression models were constructed using the metareg command in STATA.¹¹ The regression coefficients obtained in the output represent the log odds ratios that were summarized as odds ratios and 95% confidence intervals. If a covariate proved to be significantly associated with prevalence estimates, the percentage of between study variance explained by the covariate was calculated as:

100*(1−(τ² regression model with covariate/τ² regression without covariate).

Results

Studies Identified

The literature searches yielded 217 articles (including duplicates). After screening abstracts and titles against the inclusion criteria, 60 articles were selected for a full text detailed examination. A total of 25 studies were excluded at this stage; 24 did not provide an estimate for self-medication and 1 was excluded because the sample was selected only from self-medicating participants (Fig. 1). Of the 35 studies that met the inclusion criteria for all self-medication use, 20 were included in the meta-analysis of antibiotic only self-medication.

Figure 1: Search results and study selection.

The sample sizes varied from 19 to 6603 participants. Most studies took place in Asia (predominantly South East but not exclusively) or in Africa. There were also studies from UK, US, Uzbekistan, and Iran. The study populations were described as adults in all but one study researching health seeking behavior in adolescent STD patients.¹² Seventeen studies recruited from community-based samples, whereas 18 recruited patients attending STD clinics. FSW were recruited for 8 studies. Four studies selected samples from specific populations also perceived to be at a higher STD risk: clients of sex workers,¹³ fishermen,¹⁴ injection drug users,¹⁵ and military conscripts.¹⁶ Characteristics of studies included in the analyses are described in Table 1. Figure 2 shows plots for (a) all self-medication and (b) antibiotic only self-medication studies. In a review of an intervention it is possible to detect publication bias if smaller studies tend to be significant. A similar phenomenon might be to publish smaller studies with a higher prevalence of the behavior of interest. However, there is no correlation between sample size and prevalence estimate, which is somewhat reassuring.

TABLE 1: Characteristics of Studies Included: All Self-Medication Use and Antibiotic Only Self-Medication

TABLE 1: (Continued)

Figure 2: Plots exploring bias in studies included.

All Self-Medication Use

As shown in Figure 3A, there is significant heterogeneity present across all self-medication prevalence estimates, Q = 3954.82, P <0.001. They varied from 7.1%¹⁷ to 74.5%.¹⁸

Figure 3: Forest plots of prevalence estimates and 95% confidence intervals from studies of all and antibiotic only self-medication use.

To explore the sources of heterogeneity graphically, Figure 4A shows the pooled estimates by subgroup. Not one potential division of the studies explained the significant heterogeneity and none was significant in isolation. Five subgroups were explored. Gender was explored separately from the other 4 subgroups, including 36 studies. There is an indication that FSW tend to self-medicate more than the other women subgroup and other women more than men, but confidence intervals overlap. High-risk populations compared to low-risk populations had a slightly higher pooled estimate of self-medication prevalence and studies published before year 2000 and those taking place in developing countries. There was some variation between regions of study, with studies from Uzbekistan and Iran presenting the highest pooled prevalence. All subgroup analyses revealed significant heterogeneity between studies.

Figure 4: Pooled prevalence estimates for all and antibiotic only self-medication use by subgroup categories.

Both univariate and multivariate meta-regression analyses showed that none of the covariates explored were significantly associated with the variation observed (Table 2).

TABLE 2: Odds Ratios From Meta-Regression Results of All Self Medication Use Studies (n = 35)

Antibiotic Only Self-Medication Use

Important heterogeneity was also observed among antibiotic only self-medication prevalence estimates, Q = 3797.94, P <0.001 (Fig. 3B), varying from 1.8%¹⁷ to 74.5%.¹⁸

In the subgroup analysis, 5 subgroups were explored (Fig. 4B). Gender was again explored separately from the other 4 subgroups, including 17 studies. There is no indication that FSW tend to self-medicate more or less with antibiotics than the other women or men subgroups. High risk populations compared to low risk populations had a slightly higher pooled estimate of self-medication prevalence and studies taking place in developing countries. There was some variation between regions of study, with studies from Asia presenting the highest pooled prevalence. The clearest evidence of a difference between pooled estimates is observed in the publication year subgroup. Those studies taking place before year 2000 have a higher pooled estimate than those taking place after year 2000. All the subgroup analyses revealed significant heterogeneity between studies.

Both univariate and multivariate meta-regression analyses showed that the only covariate significantly associated with the variation observed across prevalence estimates was the year of publication, P = 0.022 in univariate analysis, P = 0.027 and P = 0.015 for multivariate models (a) and (b), respectively (Table 3). In the regression model with no covariates τ² = 2.164, whilst in the regression model with year of publication as covariate τ² = 1.677, the percentage of between study variance explained by this covariate was 22.5%.

TABLE 3: Odds Ratios From Meta-Regression Results of Antibiotic Only Self-Medication Use Studies (n = 20)

Discussion

Self-medication undermines our ability to effectively treat and thereby control curable STD, allowing infections to persist, generate damage, and develop resistance to inappropriately used antibiotics. A large variation in prevalence estimates of self-medication was found across both all and antibiotic only self-medication use. This variability is not explained by the underlying risk of populations studied, by the geographical location, or gender of the participants. However, some of these variables did change the estimated prevalence of self-medication, but more studies will be required to see whether the differences are real.

Publication year did not explain the variability across all self-medication studies; however it did explain 22.5% of variation between antibiotic only self-medication studies. This suggests that the amount of antibiotic used as self-medication might have diminished since 2000. In 1998, the World Health Organization published a report on the role of pharmacists in self-medication and the importance of responsible self-medication in different settings.¹⁹ This report may be one of the multiple factors increasing awareness of the risk related to antibiotic misuse and its role in the emergence of antibiotic resistance at a population level, particularly in both the media and the public health community. The decrease in antibiotic use may also be due to a subsequent greater control of antibiotic availability without prescription.

The searches identified studies worldwide, with a majority of them taking place in Asia and Africa. No studies were identified from South America, Eastern Europe, or the Pacific area. This might indicate a publication bias. However, it seems unlikely that there would be a publication bias because both high and low prevalence of self-medication is of interest. Furthermore, we observed no correlation between prevalence estimate and sample size in the studies included.

In the meta-regression models, gender was not included because the data were only available for half of the studies. Characteristics of countries could not be evaluated, other than the geographical location, and place of residence of populations studied, rural or urban, could not be assessed from the majority of publications. These characteristics, in particular those determining healthcare accessibility, might play an important role in the variability of the prevalence estimates.

Some studies reported self-medication for a current STD episode, whilst other asked the patient if they had had an STD episode in the past and to recall if they self-medicated for it. This methodological difference in establishing self-medication prevalence estimates might also have an impact in its variability and was not studied here because of the limit of covariates to be included in the model. We guided the choice of parameters by using a theoretical framework to help avoid both multiple testing and to specifically test previously selected hypotheses.

Self-medication prevalence estimates might also differ according to the populations studied. Whilst samples recruited for population-based studies would determine prevalence for effective and ineffective self-treatment, those studies with samples recruited from STD clinics will provide prevalence estimates of self-medication that was not effective, representing patients whose symptoms persisted despite self-treatment. However, in this meta-analysis, these differences did not explain the variability of self-medication estimates.

We conducted a comprehensive systematic review of self-medication prevalence studies and a meta-regression to investigate the sources of variability across estimates. It showed that the use of meta-analytic approaches is an important tool to improve the understanding of study and population characteristics influence on the variation of self-medication prevalence estimates. Self-medication might be a behavior that is more related to cultural beliefs at individual level, with individual characteristics varying in the populations studied, than a phenomenon related to study or population characteristics.

The next step will be to explore individual level determinants and how they influence self-medication in populations: who engages in self-medication and how does this practice impacts STD transmission? Because the same individual level determinants may influence self-medication in a different way depending on the population studied or the cultural context, multilevel modeling with individual characteristics nested within group/population-level determinants might prove an interesting tool to explore these associations.