Selection of studies
Our search identified 167 reports through database searches as well as 3 additional reports identified by other resources. A total of 117 reports remained after the duplicates were removed. We rejected 105 reports during the initial selection of titles and abstracts because they clearly did not meet the eligibility criteria. We retrieved and evaluated the full text of 12 reports. We excluded 3 reports of non-randomized studies. We included 9 reports from 3 studies [15,16,17]. Figure 1 shows the study selection process. We did not identify any ongoing trials.
We have summarized the characteristics of the included studies (countries, participants, interventions, outcomes, study design, sample size, follow-up period) in Table 1.
Methods and settings
We included three RCTs. All three trials included a multi-center trial in the United States  and two studies at a single site in the United States  and Malawi .
The studies included a total of 1264 participants who were randomized for self-administration (651 women) versus administration by a provider (613 women). All studies included women of childbearing potential receiving DMPA-SC for contraception.
All three included studies randomized participants to receive either self-administered or provider-administered DMPA-SC. The duration of follow-up was 12 months.
All the included trials reported continuation of injectable contraception at 12 months as reported by the patient. [15,16,17]. Only one study verified the use of DMPA by measuring the levels of MPA in the blood . Two studies reported failure (unwanted pregnancy) and satisfaction [16, 17], two studies reported other adverse events (not serious) [15, 17], and three studies reported serious adverse events, if any [15,16,17].
Risk of bias within studies
We presented risk of bias data from each study in Figure 2. We performed an outcome level assessment for detection bias, as the absence of blinding can introduce bias into the measurement of women. declaring their satisfaction. We judged the studies to be at high risk of performance bias due to the lack of blinding. Lack of blinding associated with a fixed block size is a potential source of selection bias in two of the included studies [15, 17]. Details and justifications for our judgments are provided in additional file 1.
Results of individual studies and summary of results
Continuation of contraceptive use for 12 months was reported by the included RCTs (3 studies, 1,261 women).
Self-administration, compared to provider administration, improved the continuation of contraceptive use (RR 1.3495 [1.0953; 1.6626]; p = 0.0049); moderate certainty evidence), although there was substantial heterogeneity (Tau2= 0.0239; I2= 71.4% [3.1%; 91.6%]; p = 0.0301). We performed a sensitivity analysis using the fixed-effect model which returned a hazard ratio (M – H, fixed, 95% CI) of 1.4401 [1.3023; 1.5924], P
Additionally, we investigated the source of heterogeneity using the pre-defined subgroup analysis.
For the HIC versus LMIC subgroup analysis, the test for subgroup differences was significant (Chi2= 5.71, df = 1, p = 0.0168), Fig. 3. In HIC (2 studies, 530 women) the relative risk (M – H, random, 95% CI) was 1.22 [1.04, 1.43] (evidence of very low certainty). The NNT-b is 10 (95% CI 5–53). In LMIC (1 study, 731 women) the relative risk (M – H, random, 95% CI) was 1.59 [1.40, 1.81] (low-certainty evidence), the NNT-b is 4 (95% CI 3-5).
For the subgroup analysis by type of care provider (community health worker vs clinic provider), the test for subgroup differences was not significant (Chi2= 0.05, df = 1 (P = 0.83), i2= 0%).
We also performed a sensitivity analysis to assess the impact of attrition on the outcome of the lawsuit, showing no change in the estimate of the effect.
Satisfaction at 12 months has been reported by two studies.
In HIC (1 study, 398 women) the hazard ratio (M – H, fixed, 95% CI) was 0.95 [0.84, 1.07] (evidence of very low certainty). In LMIC (1 study, 731 women) the hazard ratio (M – H, fixed, 95% CI) was 1.83 [1.61–2.07] (low-certainty evidence). We did not pool the studies because there was substantial heterogeneity (Tau2= 0.22; Chi2= 55.34, df = 1 (P 2= 98%). Testing for subgroup differences for HIC versus LMIC: Chi2= 53.13, df = 1 (P 2= 98.1%. We also performed a sensitivity analysis to assess the impact of attrition on the satisfaction outcome, showing no change in the estimate of the effect.
Contraceptive failure (unwanted pregnancy) was reported in two studies (1,129 women).
The relative risk (M – H, fixed, 95% CI) was 0.47 [0.13, 1.67] (evidence of very low certainty.
Only one woman reported serious adverse events in the three included trials (1,261 women).
The risk ratio (M – H, fixed, 95% CI) was 0.34 [0.01, 8.22] (evidence of very low certainty). These serious adverse events were menorrhagia and anemia requiring hospitalization) reported by a woman in the provider group and resolved without sequelae.
Other adverse events were reported in two trials (863 women) with a risk ratio (M – H, fixed, 95% CI) of 0.59 [0.28, 1.28] (evidence of very low certainty). Other side effects included pain or irritation at the injection site, nausea, vomiting, irregular uterine bleeding, headache, amenorrhea, decreased libido, and weight changes. Not all of these non-serious adverse events required hospital treatment.
The details of the GRADE results summary table, all results, are shown in Table 2.