Data will be analyzed using Stata software (Stata Corp V.14, Texas, USA). Unadjusted prevalence/incidence and standard errors of MDD will be recalculated based on the information of crude numerators and denominators provided by individual studies. Only studies using the same tool for detection of MDD will be pooled in the meta-analysis. To keep the effect of studies with extremely small or extremely large prevalence estimates on the overall estimate to a minimum, the variance of the study-specific prevalence/incidence will be stabilized with the Freeman-Tukey single arc-sine transformation before pooling the data within a random-effects meta-analysis model [
19]. Heterogeneity will be evaluated by the Chi-squared test on Cochrane’s Q statistic [
20], which will be quantified by I-squared values, assuming that I-squared values of 25, 50, and 75% being representative of low, medium, and high heterogeneity, respectively [
21]. When substantial heterogeneity will be detected, we will perform meta-regression and subgroup analyses to investigate the possible sources of heterogeneity using the following grouping variables: mean or median age, sex (female, male), study setting (rural, semi-urban, urban), geographical area (Northern, Central, Western, Eastern, Southern Africa), countries, ART (yes, no), MDD diagnosis criteria (tools used for the measurement), and administration of the questionnaire (auto-evaluation, hetero-evaluation). We will conduct a sensitive analysis including only studies with low risk of bias. We will assess inter-rater agreement between investigators for study inclusion, data extraction, and methodological quality assessment using the Cohen’s Kappa coefficient [
22]. If the included studies differ significantly in design, settings, outcome measures, or otherwise, a narrative format will be used to summarize them.