Protocol and registration: The presentation of the systematic review followed the Preferred Reporting Items for Systematic Review and Meta-Analyses (PRISMA) grid [3] , and the protocol was registered in the International Prospective Register of Systematic Reviews (PROSPERO) database under the number PROSPERO 2019 CRD42019118801 available on: http://www.crd.york.ac.uk/PROSPERO/display_record.php?ID=CRD42019118801.

Eligibility criteria: Articles included were those relating to the evaluation of road safety interventions implemented in Africa, published in English or French and for which full versions were available. These articles covered a range of interventions such as the adoption or revision of a law, controls or enforcement measures to prevent speeding and alcohol, the use of protective equipment (helmets, seatbelts, child restraints), user awareness, user (driver) training, school education (school modules), distribution of protective equipment, improvement of road conditions, and traffic control.

Abstracts without full articles, literature reviews, systematic reviews, meta-analyses, commentaries, and articles with unworkable data (not available in French or English) were excluded.

Information sources: Articles were searched on electronic databases: PubMed, Scopus, Lissa, Google Scholar, and African Journals Online.

Article search strategy: Articles were retrieved from 22 to 27 February 2019 using search equations tailored to each database. Two pairs of reviewers were formed to independently conduct the article selection and data extraction.

Article selection: For the selection, all articles from the database search and their abstracts were imported into EndNoteX9 regardless of the language of the full text. Each reviewer examined the articles on their own and then reviewed them within the pair and between the pairs to compare the articles selected. Different steps were followed, namely the elimination of duplicates, the exclusion of abstracts and other articles that did not meet the inclusion criteria, the exclusion of articles on the basis of the title and abstract, and then on the basis of the full text.

Data extraction: For this purpose, an Excel tool was designed containing information on the title of the study, the references (journal, year, publication number), the authors, the country, the type of study, the types of road safety interventions, the users involved, the outcomes to be measured, the duration of the intervention, and the assessment of the quality of the data.

Quality of the studies: Data quality was checked according to the type of study. Different approaches were used: 1) Strengthening The Reporting of OBservational Studies in Epidemiology (STROBE) indicators were checked for cross-sectional studies [4] ; 2) for case-control and cohort studies the Newcastle-Ottawa Scale (NOS) scale was complemented with the STROBE indicators [4] [5] ; 3) A Cochrane Risk Of Bias Assessment Tool for Non Randomized Studies of Interventions (ACROBAT-NRSi) scale [5] , and the Standard Protocol Items: Recommendations for Interventional Trials (SPRIT) guidelines for non-randomised trials and quasi-experimental studies [6] ; 4) the Cochrane Grading of Recommendations, Assessment, Development and Evaluations (GRADE) mechanism [7] , and the Consolidated Standards Of Reporting Trials (CONSORT) guidelines for randomised studies [8] . We assigned risk of bias as recommended by Cochrane, referring to the Cochrane Handbook (low, moderate, and high risk) [9] . The risk of bias is high if several indicators of bias are present on all scales; the risk is moderate when there is a single indicator of bias on several scales. It is low when all indicators measure low or no bias.

Type of data analysis: A qualitative analysis of the data was carried out and a narrative approach was adopted to describe and synthesise the results.

At the end of the search, a total of 3106 articles were obtained of which 848 were from PubMed, 1262 from Scopus, 543 from AJOL, 332 from LiSSa, and 121 from Google Scholar. The selection process resulted in twenty-nine articles (Figure 1).

The extracted data were presented according to the characteristics of the studies (Table 1). Considering the type of study, twenty-one articles were quasi-experimental studies, six of which were randomised and fifteen non-randomised. One study was cross-sectional and seven were time series.

The type of intervention evaluated was varied. Two studies concerned institutional strengthening [10] [11] . Nine interventions concerned law enforcement, including two on helmet wearing [12] [13] , one on seatbelt wearing [14] , three

on speed limits [15] [16] [17] , one on alcohol and other road safety measures [18] , one on banning motorbikes from the capital [19] , and one on banning the use of motorbikes as a means of public transport [20] . Eight studies concerned awareness raising, training, or user education interventions [21] - [28] . One study concerned the development of an Uber transport service in South Africa [29] , and one concerned the distribution of helmets and reflective vests [30] . Eight studies combined several interventions such as enforcement, monitoring, education and awareness raising, provision of equipment (helmets and hairnets), vehicle maintenance, and infrastructure improvements [31] - [38] .

The users benefiting from these interventions were pedestrians (three studies), drivers or passengers of two- or three-wheeled vehicles (eight studies), drivers or passengers of vehicles with more than three wheels (nine studies), all motorised vehicle users (two studies), or all types of users (seven studies).

The expected outputs or outcomes of these studies were diverse. These included a reduction in road traffic fatalities, injuries, crash cases, or emergency room admissions for trauma. Other outputs included reductions in the severity of road crashes, hospital stays, or speeding. Some outputs were related to helmet use, improved knowledge and behaviour regarding traffic rules, laws and protective measures, or insurance claims data (Figure 2).

In the implementation, twenty-two studies involved interventions that were implemented for three months or more, while six interventions were one-offs lasting less than a week, and one intervention was repeated more than once.

The risk of bias was assessed as high for twelve studies, moderate for thirteen, and low for four. No studies were excluded due to risk of bias.

The results obtained have been presented by intervention (Table 2). Most of the interventions implemented showed evidence of effectiveness in terms of the results obtained. Indeed, for fourteen studies all the expected indicators had changed favourably and for eleven the indicators had partially changed. Only four studies did not show the expected main results. The studies that showed significant change were institutional strengthening [10] [11] , law enforcement [15] - [20] , awareness raising, training, or user education interventions [21] - [28] , development of Uber transport services [29] , distribution of equipment [30] , and a combination of interventions [31] [33] - [38] . The studies were randomised [22] [23] [26] [27] [28] [30] , non-randomised [15] [16] [17] [19] [21] [24] [25] [33] [35] [36] [38] , time series [10] [11] [18] [20] [29] [31] [37] , or cross-sectional studies [34] . The inconclusive studies were non-randomised [12] [13] [14] [32] . Of them, three were law enforcement studies [12] [13] [14] , and one combined several interventions [32] .

Countries of studies implementation: Nigeria (twelve studies), South Africa (seven studies) and Kenya (five studies) embedded more than 80% of studies selected (Figure 3).

Institution building studies: The two studies concerned were implemented in Nigeria and assessed the influence of the creation of a road safety institution on the trend in road crashes, injuries and deaths. Both studies found a significant reduction in road crashes and injuries according to the indicators measured (average annual number for one, proportion and number per capita for the other) [10] [11] . However, while one study found a significant reduction in fatalities [11] , the difference was not significant in the second study [10] .

Law enforcement studies: Five of these interventions improved all the indicators predicted by the studies [15] [16] [18] [19] [20] , while one improved partially indicators [17] , and three were inconclusive [12] [13] [14] . The implementation of these laws resulted in a reduction in the frequency and severity of injuries [13] [15] - [20] , a significant decrease in the hospital prevalence of road traffic accidents (RTAs) [20] , a decrease in RTAs resulting from motorbike collisions [19] , a reduction in mean speed [17] , an increase of subjects wearing helmets [12] , a reduction of passenger vehicle crash [17] , a decrease in road traffic deaths [16] [17] [18] [19] , and a reduction in the total length of hospital stay for patients admitted for head injury after an RTA [16] . The laws concerned were speed limits [15] [16] [17] , alcohol control [18] , helmet use [12] , a ban on motorbikes in the capital [19] , and a ban on the use of motorbikes as a means of public transport [20] . In contrast, evaluation of the influence of helmet enforcement in Nigeria, rather than a reduction, noted an increase in the proportion of motorcyclists who died from RTAs after the intervention [12] . Furthermore, they were no improvements in helmet use among motorcycle drivers and passengers after helmets enforcement in Kenya [13] , and non-significant reduction in accident cases after seat-belt enforcement in Nigeria [14] . Authors of these studies believed that the influence of other confounding factors as lack of education on the correct use of the crash helmet, or differences in population in different periods were responsible for these outcomes. Opportunity to further increase both awareness as well as enforcement of the legislation would improve outcomes of law enforcement.

Studies on awareness raising, training or user education interventions: Four studies noted the achievement of all expected outputs [23] [25] [27] [28] , and for another four, there was partial achievement [21] [22] [24] [26] . Training drivers in rescue techniques, basic evacuation procedures and specific first aid skills resulted in a significant increase in the first aid knowledge score in the intervention group [22] as well as in the first aid skill score [22] [23] . However, the knowledge score was not improved in one of the two studies that evaluated this intervention [22] . Similarly, the intervention to improve knowledge of traffic signs resulted in a significant change in learners’ or drivers’ knowledge after the intervention [21] [25] , and a significant increase in the proportion of drivers reporting compliance with traffic signs in the group [25] . However, the effects varied according to the categories of traffic signs [21] .

Communication about road safety practices led to some expected results, such as an increase in the proportion of drivers reporting resting after hours of driving or observing speed limits, and a decrease in the proportion of drivers reporting fatigue. This study did not find significant differences in the possession of a valid driver’s licence, substance use, carrying more than one person, and a decrease in the prevalence of RTAs [24] .

Health and safety education on various road safety topics resulted in a significant increase in the proportion of drivers with correct knowledge of licensing requirements, and adequate knowledge of road signs and maximum speed limits in the intervention group. However, the proportion of drivers reporting compliance with speed limits in the intervention group decreased [26] . The introduction of stickers in public transport vehicles led to a reduction in the number of incidents resulting in insurance claims [27] [28] . Stickers had messages encouraging passengers to react if the driver did not follow the rules of the road, to report it [27] [28] , or messages on the possible consequences of driver misbehaviour if the driver did not listen to the advice [28] . There was also a reduction in the average maximum speed of drivers in the intervention group, a reduction in crashes (140 fewer RTAs) and fatalities (55 lives saved) [28] .

Study of the development of Uber transport services: This intervention resulted in a slight reduction in weekly RTA deaths in the province with Uber in South Africa compared to provinces without Uber [29] .

Equipment distribution study: Free distribution of reflective vests, combined with brief road safety education in Tanzania, led to a significant increase in the use of reflective vests among motorbike taxi drivers. However, there was no significant change in the use of white helmets and daytime running lights among motorbike taxi drivers [30] .

Multi-intervention study: The combination of various interventions among public transport drivers in Kenya did not result in significant changes in injury severity levels (including fatalities) or patient hospital stays [32] . A combination of interventions to reduce the risk of accidents among schoolchildren resulted in a decrease in the prevalence of RTAs, the frequency of pedestrian-car collisions among schoolchildren, and the frequency of minor injuries in the intervention group [35] [37] . There was a non-significant decrease in the frequency of pedestrian-motorcycle collisions (the most frequent type of collision) in the intervention group [35] , and in the number of motor vehicle collisions in one of the two zones, but an increase was recorded in the second zone [37] . These interventions reduced the frequency of speeding, braking, and sudden acceleration among drivers in the intervention group, but they drove more at night [36] . The construction of a footbridge and educating schoolchildren to use it in Nigeria showed a “significant” increase in the number of pupils using the footbridge to cross the road [38] .

Studies evaluating the effectiveness of the enforcement of various road safety laws and the monitoring of compliance have shown reductions in fatalities in terms of lives saved [33] , or reductions in fatalities per 10,000 vehicles [31] . There was also a significant reduction in the number of non-injury RTAs, but the number of injuries per 10,000 vehicles did not decrease as expected [31] . A study implementing activities to increase compliance with road safety measures and encourage helmet use resulted in improved helmet use (reported or directly observed), reduced prevalence of RTAs, speeding, carrying more than one passenger, and an increase in the proportion of subjects reporting protective behaviours [34] .