The literature search strategy employed

The literature search strategy used both text words and medical subject heading (MeSH) terms (all fields). Key terms such as “antimicrobial” or “antibiotic” or “anti-infective agent” were used to develop a search strategy. No time and language restrictions were applied during database searches.

Example of the search strategy developed and on PubMed database

(("primary health care"[mesh] OR primary care[tw] OR primary health*[tw] OR community health*[tw] OR community care[tw] OR community worker*[tw] OR clinic[tw] OR clinics[tw] OR “general practitioners” [mesh] OR general practi*[tw] OR family medicine[tw] OR family practi*[tw] OR “physicians, family” [mesh] OR family physician*[tw] OR family doctor*[tw] OR "physicians, primary care"[mesh])) AND (("anti-bacterial agents"[Pharmacological Action] OR "anti-bacterial agents"[MeSH Terms] OR "anti-infective agents"[Pharmacological Action] OR "anti-infective agents"[MeSH Terms] OR antibiotic*[tw] OR antimicrobial*[tw] OR antibacterial*[tw] OR anti bacterial*[tw] OR anti-infective*[tw])) AND ("therapeutic use"[sh] OR "drug prescriptions"[mesh] OR "drug utilization"[mesh] OR “inappropriate prescribing” [mesh] OR "drug utilization review" [mesh] OR "practice patterns, physicians'"[mesh] OR use[tiab] OR user*[tiab] OR used[tiab] OR overuse*[tiab] OR underuse*[tiab] OR misuse*[tiab] OR utiliz*[tiab] OR overutili*[tiab] OR underutili*[tiab] OR prescri*[tw] OR overprescri*[tiab] OR underprescri*[tiab])

Types of studies included in the review

We included in this review studies conducted in East Africa that reported the proportion of patients receiving any antibiotic prescription irrespective of facility setting or level. The following study types were included in the review: cross-sectional studies, cohort studies, and RCTs (randomized controlled trials).

Reviews of all kinds, economic evaluation studies, qualitative studies, mathematical modeling, and non-primary research publications such as commentaries, editorials, and conference proceedings were excluded. Studies reporting antibiotic use in animals, i.e., those focused on veterinary use of antibiotics and those focused on special cohorts of patients such as surgical prophylaxis where the use of antibiotics is justified, were excluded.

Data collection and analysisSelection of studies

All electronic database outputs were imported into Rayyan software for screening and selection. The first and second authors independently screened 100% titles and abstracts for inclusion of potentially eligible studies sourced from the database searches. Titles and abstracts in non-English languages were translated into English using Google Translate. The first reviewer (J. A.) collected full-text articles/publications of potentially eligible studies, and then J. A. and the second reviewer (PK) independently screened 100% of full-text articles for inclusion. Where disagreement occurred between the two reviewers, the last author (EM) was consulted. Each step of the study selection process was documented, and where a study was excluded, the reason(s) for exclusion was recorded and entered into the PRISMA flow diagram.

Data extraction and management

Data were independently extracted in text, tables, and figures of the included studies by the first and second authors and recorded on a standardized, pre-designed extraction form designed in Excel. In the case of unclear data, reviewers contacted corresponding authors listed in the articles for clarifications. Data management was the duty of the first author (J. A.) in consultation with the second author (P. K.). Completed data extraction forms were maintained on both a password-secured laptop and USB memory stick and exported to STATA for analysis.

The following data points were extracted from the included studies:

Study characteristics: Year(s) of data collection, study design, source of data, population or participants, and objectives of the study

Study setting: Country, income level, and health facility level

Outcome measures: Number of individuals receiving at least one antimicrobial prescription to the number of persons attending a given health facility within a specified period

Risk-of-bias assessment

The methodological quality of studies including the risk of bias was assessed using a checklist to assess for internal and external validity. A modified checklist originally developed by Hoy and colleagues was used to score; sampling strategies used, outcome assessment, outcome measurement, and statistical reporting and higher overall scores represented higher methodological quality [10]. Each article was independently scored by the first and second authors in consultation with the last author, attached as supplemental file, S1.

Treatment of missing data

Authors of articles with missing data were contacted to provide the missing data points. In cases where the missing data were impossible to obtain, full descriptions are provided about the nature of the missing data and the implications on the results in the reporting of this systematic review.

Assessment of heterogeneity

Forest plots were used to assess the presence of statistical heterogeneity. We assessed heterogeneity by calculating χ2 (threshold P > 0.1) and I2 statistics (threshold I2 > 40%). The values of I2 were categorized for heterogeneity as follows: “not important” (0 to 40%), “moderate” (41 to 60%), “substantial” (61 to 80%), and “considerable” (81 to 100%). Where “not important” or “moderate” heterogeneity exists between studies (I2≤ 40%), the outcomes were pooled in a meta-analysis and reported using forest plots. Where “substantial” or “considerable” heterogeneity exists between studies (I2 > 40%), the outcomes were pooled and reported in narrative form using forest plots.

Data synthesis

Data from the included studies were combined using a random-effects model to account for variability between studies. This is because substantial between-studies heterogeneity was anticipated considering the different study designs included in the systematic review. STATA software (College Station, Texas 77845, USA) was used to perform the meta-analysis.

Sensitivity analysis

Sensitivity analyses were performed to assess if methodological differences in outcome measurement influenced the review results.