In the systematic search, a total of 875 citations were identified in databases, and the study selection process is illustrated in Fig. 1. After duplicate removal, 705 articles were screened by title and abstract for potential eligibility. In addition, 29 unpublished records were identified. No additional studies were identified in references of previously published systematic reviews. After screening, twenty-four studies were assessed for eligibility, and 13 studies were excluded [16,17,18,19,20,21,22,23,24,25,26,27,28]. Non-comparative studies were the main cause of exclusion, and all reasons are shown in the Suppl. 3. Inter-rater agreement was suitable (kappa = 0.83). Eleven studies fulfilled the inclusion criteria of the systematic review, and the characteristics of the included studies are summarized in Table 1. Two authors were contacted for data to be included in the meta-analyses [29, 30], only one responded, however the information could not be pooled. Eight studies were included in the meta-analysis [29, 31,32,33,34,35,36,37].

Flowchart for systematic review and meta-analysis (PRISMA) of educational intervention in pharmacovigilance, screening of articles, and selection process

For country, RCTs were principally conducted in Portugal (four articles) and Sweden (two articles). Geographically, all the studies were conducted in Europe and Asia. The EI varied from one day to nine months, and follow-up ranged from 0 to 20 months. The average participation rate (a healthcare professional who agreed to participate into the study) varies in each study between 7.9 and 84.0%, and participants had more adherence to combined interventions and electronic ADR information.

Four studies involved physicians [29, 31, 33, 35], two involved nurses [29, 36], four involved pharmacists [32, 34, 37, 38], and two studies evaluated primary healthcare units that included physicians and nurses [30, 39]. The professionals mainly studied were physicians (six studies with 5097 participants and 136 primary healthcare units), followed by the pharmacist (four studies with 887 participants) (Table 1).

Workshops were the most common educative interventions used into studies [31, 33, 34, 36], followed by intervention combined (session group and educative material) [29, 32, 35], telephone-based interventions [31, 34], lecture [36], educational material (transparencies, brochures, and posters) [38], electronic information sheet of ADR [37], E-mail interventions [39] and one-page ADR information letter [30]. Three studies included continuing education by the pharmacovigilance unit as a control group [32, 35, 37], while eight studies did nothing [29,30,31,32,33,34, 36, 38, 39] (Table 1).

Ten studies informed the number of ADR reports [29,30,31,32,33,34,35, 37,38,39]. Five studies were excluded from the meta-analysis because these have incomplete data such as number of participants, or the total number of ADR reports [29, 30, 37,38,39]. Five studies present complete data for meta-analysis, and classified ADR as total, serious, high probability, unexpected, and new drugs by control and intervention groups [31,32,33,34,35]. Two studies presented three arms (workshop, telephone-based interventions, and control group) [31, 34], and three studies with two arms (combined intervention or workshop vs. control group) [32, 33, 35].

Educational interventions increased the reporting of all ADRs in comparison with control group (OR = 4.74, [95%CI, 2.46 to 9.12], I2 = 93%, 5 studies). In the sensitivity analysis, after removed Herdeiro et al. [31], educational interventions showed consistency in increasing ADR reporting (OR = 6.06 [95%CI, 2.50 to 14.71], I2 = 94%, 4 studies). In subgroup analysis, workshops (OR = 6.26, [95%CI, 4.03 to 9.73], I2 = 57%, 3 studies) increased ADR reporting, more than combined interventions (OR = 5.14, [95%CI, 0.97 to 27.26], I2 = 98%, 3 studies), while telephone-based interventions no showed a difference (OR = 2.59, [95%CI, 0.77 to 8.73], I2 = 29%, 2 studies) (Figs. 2).

Forest plot of total adverse drug reactions reported for each educational intervention at the end of the study. Sub-analysis was performed by type of intervention

ADR reporting change over time is shown in Table 2. In the workshop intervention, the increase in the number of reports was significant up to 16 months after IE for total and severe ADRs, but only increased over 12 months for unexpected, high-causality, and new drug ADRs. In contrast, telephone-based interventions only increased the number of total reports and serious ADRs by 4 months. Interestingly, the combined interventions increased the number of unexpected and new drug ADRs for at least 12 months, although for total, serious, and high-causality ADRs, the effect was seen from 12 months onwards.

Regarding the change in knowledge in pharmacovigilance, three studies [29, 36, 37] evaluated 4 educative interventions. The meta-analysis results showed a tendency to increase pharmacovigilance knowledge mean scores in participants who received EI in comparison with the control group (SMD = 1.12, [95%CI, -0.12 to 2.36], I2 = 98%, 4 studies). After removing the highest risk of bias study [29], participants in EI group shown an augmented their pharmacovigilance knowledge (SMD = 1.53 [95%CI, 0.58 to 2.47, I2 = 92%, 3 studies]). In subgroup analysis, the participants who received lecture (SMD = 2.23 [95%CI, 1.81 to 2.65], 1 study) and workshop (SMD = 1.85 [95%CI, 1.44 to 2.27], 1 study) increased their knowledge; this effect was not observed in those who received the combined intervention or letter with ADR information (Fig. 3).

Forest plot of the difference in means of the effect of EI in the score of knowledge about pharmacovigilance at the end of the study. Sub-analysis was performed by type of intervention

Two studies evaluated ADR reporting attitudes among health professionals (Table 1), however, the measurement scales obtained by the questionnaire are different, so it was not possible to perform a meta-analysis. One study conducted in pharmacist showed a positive attitude toward ADR reporting after the intervention [38]. Likewise, a positive effect in behavior related to reporting was observed in physicians and nurses after educative intervention [29].

In risk of bias assessment (Fig. 4), 73% of studies had adequate random sequence generation [29, 32,33,34,35,36,37,38]. Only 54% describe the randomization process completely [29, 30, 35,36,37, 39], presenting low-risk allocation concealment, because the randomization was carried out by a person outside the study, or they avoided contamination between groups by randomizing health centers.

Risk of bias graph, review authors’ judgments about each risk of bias item presented as percentages across all included studies. A Risk of bias overall assessment, the proportion of assessment studies. B Risk of bias assessment summary for included studies

The performance bias had a high risk in at least 81% of articles, due to differences in interventions ranging from a phone call to a combined intervention [29,30,31,32,33,34,35,