Strategies Used for Implementing and Promoting Adherence to Antibiotic Guidelines in Low- and Lower-Middle-Income Countries: A Systematic Review

Containing antimicrobial resistance and reducing high levels of antibiotic consumption in low- and lower middle-income countries are a major challenge. Clinical guidelines targeting antibiotic prescribing can reduce consumption, however, the degrees to which clinical guidelines are adopted and adhered to are challenging for developers, policy makers and users. The aim of this study was to review the strategies used for implementing and promoting antibiotic guideline adherence in low- and lower middle-income countries. A review of published literature was conducted using PubMed, Cochrane Library, SCOPUS and the information systems of the World Health Organization and the Australian National University according to PRISMA guidelines and our PROSPERO protocol. The strategies were grouped into five broad categories based on the Cochrane Effective Practice and Organization of Care taxonomy. The 33 selected studies, representing 16 countries varied widely in design, setting, disease focus, methods, intervention components, outcomes and effects. The majority of interventions were multifaceted and resulted in a positive direction of effect. The nature of the interventions and study variability made it impossible to tease out which strategies had the greatest impact on improving CG compliance. Audit and feedback coupled with either workshops and/or focus group discussions were the most frequently used intervention components. All the reported strategies are established practices used in antimicrobial stewardship programs in high-income countries. We recommend interrupted time series studies be used as an alternative design to pre- and post-intervention studies, information about the clinical guidelines be made more transparent, and prescriber confidence be investigated.


Introduction
Curbing rising levels of antimicrobial resistance (AMR) and maintaining the effectiveness of antibiotics are major global public health concerns. The inappropriate use of antibiotics is a key driver of AMR. Research assessing global trends in antibiotic consumption found total consumption (defined daily doses (DDDs) per 1000 inhabitants per day) increased 65% between 2000 and 2015 and the rate of antibiotic consumption by 39% (DDDs) [1]. In LMICs the increased consumption rate was substantial: (77% (DDDs) and is explained by rising income levels due to rapid economic growth which is providing greater access to antibiotics. A greater proportion of this increase in antibiotic consumption was accounted for by low-and middle-income countries (LLMICs) than by upper antibacterial agents to treat infections in humans). We have used the term antibiotic rather than the broader term antimicrobial because the burden of AMR is with bacteria in LLMICs and, the review only focussed on bacteria.

Search Methods for Identification of Studies
We conducted literature searches using PubMed, Cochrane Library, SCOPUS and the information systems of the Australian National University and the WHO website. Google Scholar was searched for additional studies. We developed search strategies for the PubMed database which were translated using appropriate syntax, subject headings and vocabulary for the other databases. The reference lists of the results retrieved were searched manually for additional items. Search strategies for PubMed can be found in supplementary file Table S1.

Study Screening and Inclusion/Exclusion Criteria
The titles and abstracts of the search results were subjected to an initial screen for potential eligibility by one reviewer (N.F.). Two reviewers, (N.F., C.H.) read the full text of the selected studies independently and assessed the studies for inclusion by applying the inclusion and exclusion criteria. A third reviewer (N.T.) made the final decision in cases where agreement could not be reached.

•
The LLMICs include those listed by the Organisation of Economic Cooperation and Development's (OECD) Development Assistance Committee for 2018 to 2020 [22]. • Health workers in LLMICs who prescribe antibiotic therapy.

•
Patients from all age groups in LLMICs who receive antibiotic therapy. • Any strategy which was aimed at promoting CG uptake or compliance for the purpose of improving rational antibiotic prescription. • Studies published in English language between 2000 and July 2020.

•
The primary outcomes included health worker performance based on appropriateness of prescribing including: Correct agent, correct dose, correct duration, correct route of administration or time of administration. Proportion of antibiotics prescribed in accordance with CG. Consumption of antibiotics expressed as defined daily doses per 100 or 1000 patient days. Patient encounters with an antibiotic. Patient outcomes-mortality and hospital re-admission rates. Adverse effects impacting patient outcomes.

Exclusion
• Commentaries, conference proceedings and literature reviews • Languages other than English.

Study Quality Assessment
The included studies were appraised for risk of bias using one of two risk of bias tools. The Downs and Black Risk of Bias tool was used for all studies except the ITS studies [23]. The ITS studies were appraised using criteria specifically developed by Ramsay et al. to strengthen reviews which include ITS designs [24]. Two reviewers independently applied the risk of bias tools (N.F., C.H.). When discrepancies occurred, the independent assessment of a third reviewer (K.R.) provided consensus. The criteria comprising the risk of bias tools were used to calculate the scores for high, medium, and low risk for each study.

Data Extraction Method
An Excel spreadsheet was used by the first author for extraction and storage of the data. The following information was recorded about each study: author, date of publication, study title and design, country where the study took place, description of context, aim of study, number and age group of participants, period of study, infectious disease focus, CG source, intervention description, data summary, outcome measure and effect size. Tables were prepared to display the information which was reported according to study design.

Data Synthesis and Presentation
The selected studies varied widely in study design and quality, range of intervention components, settings, infectious disease focus and measures of effect. A bubble plot (supplementary Figure S1) was used to display study diversity using the number of intervention components (strategies) implemented per study on the y axis and studies by design type identified by bubble colour on the x axis with study quality indicated by bubble size [25]. Given the diversity of the studies and the complex nature of the interventions, our results are presented using a narrative synthesis approach, supported by an effect direction plot [26,27].
The CGs were classified into groups based on an approach devised by Dizon, JM et. al. [28]. to customise a CG to suit local conditions: developed from scratch by local health workers and obtained from the WHO, other international organizations or medical associations. The latter group was further divided into CGs which were adopted and CGs which were adapted. An adopted CG was implemented in its entirety with some contextualisation to suit local issues (i.e., staffing, patient's access to care, training). If the CG was adapted, the recommendations were modified to suit the local environment (e.g., need to substitute one antibiotic for another due to cost and availability). Whilst every effort was made to assign the CGs to one of the groups, there was limited information about the source of the CG in some studies, hence an additional category, the CG was pre-existing with no additional information included.
The Cochrane Effective Practice and Organisation of Care (EPOC) taxonomy for implementation strategies was deemed an appropriate tool to use to develop broad categories for grouping and displaying the intervention strategies in a table [18,29]. The EPOC taxonomy has been applied to meta-reviews of health system interventions which are relevant to LLMICs [18]. Supplementary file Box S1. provides a detailed explanation of these categories and the strategies included in each category.
The methods used to measure the intervention outcomes in the studies were grouped into five broad outcome domains: (i) encounters with an antibiotic; (ii) antibiotics prescribed appropriately: dose, timing, and duration; (iii) defined daily doses per 100 bed-days; (iv) rate of clinical failure; and (v) CG indicator scores. Clinical guideline indicators were used to measure appropriate antibiotic use and practical competencies in clinical examination, diagnosis and treatment. The domains were used to develop an effect direction plot to synthesise and report the direction of study effects [27]. The effect direction plot provides a method of synthesising the data when meta-analysis is not possible [27]. The evidence for improvement, deterioration or no change/mixed effects indicated by each study's primary intervention outcome is represented by the use of up, down or bi-directional arrows. The nonparametric sign test was used to support the synthesis of effect direction across outcome domains not limited by a small number of studies (≤5). The sign test includes studies with positive or negative effect direction for an outcome domain. Studies with unclear/mixed or no effect were excluded. The power of the sign test is limited when the number of studies included in a domain is small because the number of studies may be further reduced when those with unclear/mixed or no effect are excluded [29].

Search Results and Study Quality
The database search yielded 6567 articles. After removing duplicates, we screened 4045 titles and abstracts, selected 127 full text articles to review and included 33 studies in our systematic review ( Figure 1). The study designs included five RCTs, five cluster RCTs, two quasi-experimental studies, five ITS studies, three CPPI studies and 13 PPI studies. Five studies had been included in an earlier systematic review [19]. Overall, the quality of the studies was generally low with only 15% (n = 5) scoring low risk of bias. Three RCTs, one cluster RCT and one PPI study were assessed as low risk. Medium risk of bias was attributed to 42% (n = 14) of studies and 42% (n = 14) scored high risk of bias. The main risks were related to internal validity: the lack of a control group (n = 13); the selection of participants into the study was non-randomised; and the lack of clarity around how the interventions were assessed (n = 5). The authors who used a PPI study design did not report on attempts to address threats inherent in the design (e.g., unrelated historical events, dropouts, Hawthorne effect). Few studies reported on measures taken to ensure persons measuring the research outcomes were blinded (n = 5). Three studies were limited by small sample sizes and in one study no sample size was provided. The risk of bias results for all studies can be found in supplementary file Table S2a,b. direction across outcome domains not limited by a small number of studies (≤5). The sign test includes studies with positive or negative effect direction for an outcome domain. Studies with unclear/mixed or no effect were excluded. The power of the sign test is limited when the number of studies included in a domain is small because the number of studies may be further reduced when those with unclear/mixed or no effect are excluded [29].

Search Results and Study Quality
The database search yielded 6567 articles. After removing duplicates, we screened 4045 titles and abstracts, selected 127 full text articles to review and included 33 studies in our systematic review ( Figure 1). The study designs included five RCTs, five cluster RCTs, two quasi-experimental studies, five ITS studies, three CPPI studies and 13 PPI studies. Five studies had been included in an earlier systematic review [19]. Overall, the quality of the studies was generally low with only 15% (n = 5) scoring low risk of bias. Three RCTs, one cluster RCT and one PPI study were assessed as low risk. Medium risk of bias was attributed to 42% (n = 14) of studies and 42% (n = 14) scored high risk of bias. The main risks were related to internal validity: the lack of a control group (n = 13); the selection of participants into the study was non-randomised; and the lack of clarity around how the interventions were assessed (n = 5). The authors who used a PPI study design did not report on attempts to address threats inherent in the design (e.g., unrelated historical events, dropouts, Hawthorne effect). Few studies reported on measures taken to ensure persons measuring the research outcomes were blinded (n = 5). Three studies were limited by small sample sizes and in one study no sample size was provided. The risk of bias results for all studies can be found in supplementary file Table S2a,b.
All studies reported on antibiotic prescribing in association with one or a combination of infectious diseases. Diarrhoeal disease was investigated in seven studies and, acute respiratory tract infections (ARTIs), healthcare-associated infections (HAIs) and hospital ASPs were examined in six studies each. One study focussed on community-acquired (CA) urinary tract infections (CA-UTIs) and the remaining studies investigated strategies to optimise antibiotic prescribing without specifying a particular type of infection.

Clinical Guidelines
Seventy-nine percent (n = 26) of studies provided information about the origin of the CG. The CGs were developed from scratch in 28% (n = 9) of studies. Twenty-one percent (n = 7) and 30% (n = 10) of studies, respectively, reported adopting and contextualising the CG or adapting a guideline that had been developed by WHO, other international organizations or medical associations. The remaining 21% (n = 7) referred to CGs which were already in place at the time of the study and information about their origin was not reported. In eight studies prescribers were either key participants in CG development or were invited to contribute feedback during the process.

Strategies Used
There were 18 different strategies implemented to increase compliance with CG recommendations across the 33 studies. These 18 strategies were classified into five broad categories: organisational, capacity building, monitoring and review, clinical decision support systems (CDSS) and persuasive strategies. Table 2 details the categories and associated strategies used either singularly or as components of multifaceted interventions in each study to promote and improve CG adherence.    Antimicrobial stewardship to optimize the use of antimicrobials for surgical prophylaxis in Egypt: a multicentre pilot intervention study.

Egypt
Five tertiary acute care surgical hospitals with infection control programs and ASPs teams. Use of rapid influenza testing to reduce antibiotic prescriptions among outpatients with influenza-like illness in Southern Sri Lanka.

Sri Lanka
Outpatient department in a 1500-bed teaching hospital in Karapitiya with >1000 patients daily.

Stakeholder Consensus
X X X X

Institution
Incentives X X AMS Programme X X X X X

Persua-Sive Activities
Sharing audit results across depts. The interventions in 85% of studies (n = 29) were multifaceted, combining strategies from either the same category or across categories. Three or more categories were combined in 27% (n = 9) of studies, 58% (n = 19) combined two categories and the remaining five studies investigated strategies from one category each. Most (70%) studies employed strategies from the categories of capacity building (n = 23) or monitoring and review (n = 23). Organisational and CDSS strategies were implemented in 38% (n = 13) and 42% (n = 14) of studies, respectively. Persuasive strategies were used in just two studies. Only one study which implemented a multifaceted intervention compared the impact of the individual strategies.

Organisational Strategies
Organisational strategies including management endorsement, stakeholder consensus, engaging a champion, institutional incentives and hospital ASPs were all components of multifaceted interventions. Clinical guidelines which had management endorsement (n = 3) were made compulsory.
Studies which obtained stakeholder consensus for the CG (n = 6), achieved this by involving prescribers in CG development. Researchers in Lao PDR involved the Lao Paediatric Network in translating the guideline, the WHO Pocketbook of Hospital Care for Children, into the local language and then engaged local opinion leaders amongst the Network to help drive the intervention [49]. Likewise, in an Indian study using the Plan, Do, Study, Act (PDSA) model to improve prescribing for URTIs, staff feedback was obtained during initial awareness-raising discussions about the problem and channelled into followup training. Subsequently, the quality improvement team collaborated with staff to design best practice recommendations for the new CG [39]. Two studies reported using champions to advocate for the CG. In a study in Nepal seven physician champions, one for each department, were specially trained to conduct the intervention in their departments. Their tasks included audit and feedback; modify/de-escalate/stop treatment according to the CG; attend all staff training; and keep a log of all associated activities [51]. In an Egyptian study, senior surgeons were nominated as champions to audit prescriptions in patient charts and give one-to-one feedback about the prescription plan with the prescriber when needed [35].
Hospital-based ASPs were reported in six studies. The ASP committees drove the interventions which included one or more activities from across the categories. The following three studies provide examples. In Egypt, the infection control teams in seven hospitals delivered training workshops and conducted audit and feedback to improve adherence to surgical prophylaxis [35]. In Indonesia, HAIs and prescribing were assessed after a new CG and hand hygiene campaign, including educational seminars, reminders and weekly audit and feedback had been delivered by infection control staff [45]. Pharmacists were also involved in leading ASP interventions. In Ethiopia, a CG based on the latest antibiogram data was loaded onto a mobile stewardship application. Pharmacists supported by laboratory staff conducted training in AMR, antibiotic prescribing, laboratory services and report interpretation and then carried out weekly AMS audit and feedback rounds [36].

Capacity-Building Strategies
Capacity-building strategies involved developing prescriber skills and competencies. Workshops and seminars were used in 91% (n = 21) of studies, focus group discussions in 35% (n = 8) and one third of the studies (n = 7) combined both strategies. Follow-up training was used in three studies and academic detailing (onsite training and follow-up by a clinical expert) in just two. In total, 33% (n = 12) of studies in the capacity-building category found positive effect change in the use of antibiotics and CG adherence.
The following studies illustrated how these activities were used. In Laos, an intervention conducted across seven hospitals included regular educational presentations, group discussions focusing on diagnosis and treatment of key infectious diseases and follow-up audit and feedback [50]. Two studies combined training seminars with face-to-face educa-tional outreach by a pharmacist and one added supervision of the CG implementation to the mix [48]. A multi-disciplinary panel of experts in Kenya adapted a CG for the treatment of CA-UTI and developed indicators for quality of care for screening, diagnosis and treatment to measure compliance over a nine-month period. The multifaceted intervention comprised interactive educational workshops based on the CG, peer-to-peer review and feedback of patient charts, according to CG compliance and focus group discussions on research into AMR patterns in uropathogens. The primary outcome-appropriate antibiotic prescription-improved from 19% at baseline to 68% by the end of the study period [47].
A study from Sudan combined both capacity-building and monitoring and review strategies and, unlike other studies in this review, measured the impact of the components [56]. When audit and feedback, a monitoring and review strategy was used alone antibiotic consumption did not change. However, audit and feedback coupled with either academic detailing or educational seminars showed positive effect change at both one-and three-month post-intervention. [56] Further details of the outcome results can be found in supplementary file Table S3a,b.

Monitoring and Review Strategy
Monitoring and review strategies were frequently components of study interventions (n = 23). Audit and feedback and practice supervision were components of 13 and seven studies, respectively. Antimicrobial restriction was used in four studies and reminders in two. Fourteen of the studies also incorporated a capacity building strategy and 64% (n = 9) of these showed positive effect change in CG uptake. Audit and feedback was coupled with focus group discussions (n = 6) or workshops and seminars (n = 6) or both (n = 5). All, but one study using practice supervision (n = 7) was combined with workshops and seminars (n = 6).
The following examples demonstrated the use of monitoring and review activities. A RCT in Zimbabwe investigated supervision, together with audit and feedback across different types of infections [62]. Pharmacy staff attended 14 days of training in the theory and practice of supervision before auditing prescriptions and holding on the spot discussions with health workers to improve knowledge and performance. The results were mixed: CG adherence improved in treatment for non-bloody diarrhoea and ARTI, but not genital infections. An antibiotic restriction policy was used in four studies [37,41,52,53]. All were implemented in a similar way. The policies involved completing an antibiotic justification form before commencing use of a restricted antibiotic and, if after reviewing the culture report, it was decided to continue treatment, approval had to be sought from a senior clinician or ASP committee member. Three of the studies found a reduction in the use of antibiotics. Additional information about the results can be found in supplementary file Table S3a,b.

Clinical Decision Support Systems
Forty-one percent (n = 14) of studies used CDSS strategies. The WHO's clinical algorithm, ALMANACH (Algorithms for the Management of Acute Childhood illnesses) was used in three studies [30,57,58]. Two of the studies, one in Afghanistan and the other in Tanzania adapted ALMANACH for use on mobile technology [30,58]. A third study, also from Tanzania, compared ALMANACH with ALMANACH and e-POCT, a smartphonebased algorithm that incorporated point of care tests (oximetry, haemoglobin, C-reactive protein, and procalcitonin) for the management of febrile illness [57]. Rapid diagnostic testing tools were also used to promote the rational use of antibiotics in influenza-like illness, acute respiratory tract infections and diarrheal disease [33,55,60]. All involved smartphone technology, but the level of training clinicians received varied. In a study from Sri Lanka clinicians received no training but were referred to the CG disseminated by the Ministry of Health [54]. Clinicians in a Vietnamese study attended initial workshop presentations in a central location and follow-up onsite training with leaflets and posters and were given a telephone contact number for further queries [60]. All of the studies found positive improvement in adherence to CG recommendations.
Quick reference material was reported as an intervention component in 24% of studies (n = 8). This included wall charts, leaflets, posters, drug lists and booklets available in print, on stand-alone computers, hand-held devices or via hospital intranets. Only one study, an Indian ITS study used multiple disseminations of the CG as the intervention strategy. The CG was distributed four times following initial stakeholder participation in development, revisions to content and changes in format. Adherence improved after the CG was made available online during the final stage of development [38].

Persuasive Strategies
Only two studies used persuasive strategies: one through peer pressure, and the other by way of formal contractual obligations [43,59]. In an Indian study, the prescribing decisions of each clinician were considered to be a group decision of the unit. Each unit (n = 35) received a monthly prescribing score based on the amount of antibiotics consumed relative to that consumed by all units. Scores were then shared across all units and discussion followed. At three-months post-intervention 43% of units (n = 15) had reduced their antibiotic consumption [43]. In the Vietnamese study, health officials were required to sign contracts and pledge their commitment to carry out supervision. Staff received training and reminders and were required to support the CG. Funds and equipment were donated to community health hubs on the condition monthly supervision was deemed adequate, and prescribing had improved. The primary outcome, adequate antibiotic dose improved from 30% to 90% post-intervention [59]. Table 3 displays the effect direction plot that summarises the direction of effects of the intervention outcomes according to outcome domain for all studies according to risk of bias score.

Outcomes
Overall, 67% (n = 22) of the studies provided evidence to indicate improvement in CG adherence: the outcome effects indicating a positive direction. The study designs in this group included four RCTs, one cluster RCT, two QE studies, two ITS studies, three CPPI and 10 PPI studies. The outcomes of 33% (n = 11) of the studies indicated either mixed (n = 5), unclear (n = 3) or no change in effects (n = 3), thus, suggesting no overall change in CG adherence. This group included one RCT, four cluster RCTs, three ITS studies and three PPI studies.
The effect direction plot displayed in Table 3 shows all studies reported a positive direction of effect for measures of reduction in encounters with an antibiotic with four studies reporting no change or unclear effect. The p-value for the sign test for this domain is 0.0005 at the 0.05 level. The nine studies which scored either low or medium risk of bias for this domain show a positive direction of effect with one study indicating no change or unclear effect and the p-value of 0.0027 at the 0.05 level is slightly lower than that when all studies are included. For antibiotics prescribed appropriately according to dose, timing, and duration, the 10 included studies reported a positive effect direction with one finding a negative effect and four studies finding mixed or unclear effects (p-value for the sign test is p 0.0066). When the seven studies which scored high risk of bias were excluded, five studies showed a positive direction of effect, one a negative direction and one study reported no change or unclear effect (p-value 0.1024 at the 0.05 level). Studies with no change/unclear effect could not be included in the sign tests. The three remaining domains: reduction in defined daily doses per 100 bed-days; reduction in clinical failure; and CG knowledge scores all show studies with positive directions of effect, however, the total number of studies in these categories is too small (<5) to apply the sign test to. For further information about the study outcome measures and effect sizes see supplementary file Table S3a,b which provides a summary of outcome effects as reported by the authors. Table 3. Effect direction plot summarising direction of effects of intervention outcomes of strategies used for implementing and promoting adherence to antibiotic guidelines in LLMICs.  Table 3. Effect direction plot summarising direction of effects of intervention outcomes of strategies used for implementing and promoting adherence to antibiotic guidelines in LLMICs.  [57] RCT ction plot summarising direction of effects of intervention outcomes of strategies used for implementadherence to antibiotic guidelines in LLMICs. Test: for all studies P = 0.0005 P = 0.0066 randomised controlled trial; CRCT: cluster RCT; QR; quasi experimental; ITS: interrupted time series; e-post-intervention; PPI: pre-post-intervention. Effect direction: upward arrow ▲ = positive impact, = negative impact, sideways arrow ◄► = no change/mixed effects. Sample size in intervention group: ; medium arrow ▲ > 100 to 300; arrow with hat ▲^ ≤ 100. Study quality denoted by row colour: green ber = medium risk; red = high risk of bias. §Two tailed sign test p value with hypothetical probability bject being 0.5 calculated for domains with sufficient studies to do so (<5). † L = low and M = medium = no change/mixed effects. Sample size in intervention group: large arrow > 300; medium arrow > 100 to 300; arrow with hat ˆ≤ 100. Study quality denoted by row colour: green = low risk of bias; amber = medium risk; red = high risk of bias. §Two tailed sign test p value with hypothetical probability of success in each subject being 0.5 calculated for domains with sufficient studies to do so (<5). † L = low and M = medium risk of bias.

Discussion
Our review identified 33 studies from 16 LLMICs that were published between 2000 and 2020 and examined strategies for implementing and promoting antibiotic guidelines. This collection of studies represents just 21% of LLMICs. The studies varied widely in terms of design, settings, target groups, strategy types, intervention components and timeframes, implementation methods, outcome measures and effects. The quality of the studies was generally poor with 40% using an uncontrolled pre-and post-intervention design, many of which, scored a high risk of bias.
It was not possible to tease out which strategies had the greatest impact on improving CG compliance, because of the complex nature of the interventions. Interventions in the majority of studies were multifaceted and only one study with a multifaceted intervention assessed the individual components. Therefore, for most studies, it was not possible to assess the contribution any one strategy made to the outcome or to establish how vital a single strategy was to the success of the entire intervention. The Institute of Medicine (USA) and others recommend employing multifaceted interventions over single strategies to promote adherence to CG [7,63,64]. However, whilst there is evidence for the effectiveness of particular strategies, such as audit and feedback, antimicrobial restriction or reminders, when it comes to which specific components are associated with increased effectiveness in a multifaceted intervention, the evidence is not available.
Clinical decision support systems which used digital technology to deliver the intervention, stood apart from other strategies. Rapid diagnostic testing tools and digital algorithms employed smartphone technology, were implemented with little or no support from other strategies, and all found a measure of improvement in antibiotic use. In LLMICs where access to laboratories is limited, broad spectrum antibiotics are routinely used empirically for patients presenting with acute fever, for example. Ascertaining whether fever symptoms are the result of a bacterial infection based on clinical presentation alone is challenging. However, using antibiotics unnecessarily drives AMR. Thus, POCTs have the potential to reduce antibiotic consumption by supporting prescribing decision-making in LLMICs. A qualitative study in South Africa reported most clinicians regarded POCTs as having potential for common infections: aiding diagnosis, indicating when an antibiotic is not needed, enabling earlier treatment and managing patient expectations. However, resource issues were identified as a barrier [65]. In 2019, the WHO reviewed the first Essential In Vitro Diagnostics (IVDs) List (EDL) to provide guidance to Member States developing interventions for EDLs and for selecting and using IVDs [66]. Access to digital technologies is growing in LMICs: the median rate of smart phone ownership was 37% in 2015, having risen from 21% in 2013 [67]. However, whilst digital technologies have the potential to transform the delivery of health care in resource-poor settings, major challenges (e.g., funding, ownership, privacy) need to be overcome.
Antimicrobial stewardship programmes were implemented in just six studies. Even though progress is slow, the evidence suggests LLMICs are moving towards implementing ASPs by engaging with ASP strategies in their efforts to reduce antibiotic consumption and improve CG compliance [19]. A wide range of intervention strategies were implemented, and all are used in ASPs in HICs. Most of the interventions were driven by senior clinicians, infection control experts and pharmacists. This is similar to the approach taken in HICs where a multi-disciplinary team of experts is an integral component of an ASP [68]. It is noteworthy that the studies reporting on ASPs called for further research into ASPs: initiating, implementing and maintaining ASPs, benefits of ASPs and involving pharmacists in ASP initiatives.
At least one third of CGs were developed from scratch. Developing CGs based on evidence is time-consuming, costly and requires research expertise and commitment to keep the recommendations up to date [69]. Achieving all of these elements may be overly ambitious for many LLMICs. Improving access to freely available trustworthy evidence based CGs from international guideline repositories (NHMRC, NICE), medical associations and WHO will benefit LLMICs. Regardless of the source, CGs must suit the local situation, be trusted by the end-users and easily accessible. Only eight studies reported engaging stakeholders in CG development. There is an association between barriers to uptake (e.g., complexity, end user trust in the CG) and stakeholder participation in CG development [12,41]. Stakeholders differ in education and experience. Therefore, involving stakeholders in CG development, evaluation and implementation, allows for trust and a sense of ownership to be built, and differences to be accommodated [16,41].
Mobile applications (apps) are commonly used to provide access to evidence-based CG in HICs, though little is known about this in LLMICs. A recent study in four African countries investigated prescriber perceptions and assessment of CGs used on a smartphone app [70]. Prescribers (n = 38) reported that the app increased their awareness of antimicrobial stewardship, was the "best way" to access CGs, caused them to re-appraise their prescribing as well as document the patient drug chart. Further research into the use and effectiveness of mobile technology for CGs in resource-poor settings is needed [70].
There are several limitations to this work. The outcomes measured were limited; only three studies examined clinical failure and no studies reported on prescriber confidence even though several investigated capacity building strategies. Slightly more than half of the studies overall found a positive direction of effect in guideline adherence, suggesting the research on this topic may be limited by publication bias. It is not unexpected that the studies finding positive effect change also included three-quarters of the studies which used the bias-prone PPI design [24]. The more robust ITS design, which collects data at multiple time-points before and after the intervention is implemented is recommended as an alternative to the PPI design [24]. The inclusion criteria allowed for a wide range of study designs, strategies, settings, methodologies, outcomes and CG foci. Heterogeneity is often cited as a characteristic of systematic reviews and hinders meta-analysis being conducted, comparisons being made, and the ability to generalise from the results [19,71].

Conclusions
In this review most interventions showed a positive direction of effect. It was not possible to recommend any one strategy or combination of strategies in the selection of intervention components to improve uptake of CGs in LLMICs, because of the complex nature of the interventions and the limitations of the studies, even though some strategies were particularly notable. Audit and feedback coupled with either educational workshops and/or focus group discussions were the most frequently used intervention components. Clinical decision support systems which made use of mobile technologies proved they could be implemented with little or no support from other strategies. The implementation of ASPs remains slow in LLMICs; however, LLMICs are moving in the right direction by engaging with antibiotic stewardship strategies. Our review suggests other LLMICs need to conduct similar studies. We recommend ITS studies be used as an alternative design to PPI studies, information about the CG be made more transparent, and prescriber confidence be investigated to aid future research Supplementary Materials: The following are available online at https://www.mdpi.com/article/10 .3390/tropicalmed6030166/s1, Table S1: PubMed search strategies used to identify studies investigating strategies for implementing and promoting antibiotic guidelines in LLMICs; Box S1: Description of the intervention strategies used in studies for implementing and promoting antibiotic guidelines in LLMICs; Table S2: (a) Results of risk of bias assessment of studies investigating strategies for implementing and promoting antibiotic guidelines in LLMICs; (b) Risk of bias assessment of interrupted time series studies investigating strategies for implementing and promoting antibiotic guidelines in LLMICs [24]. Figure S1: Bubble plot showing the variation in study design, research quality and number of intervention strategies implemented across studies investigating strategies for implementing and promoting antibiotic guidelines in LLMICs; Table S3: (a): Interventions, outcome measures and effect sizes for studies investigating strategies for implementing and promoting antibiotic guidelines in LLMICs; (b): Interventions, data summaries, outcome measures and trend changes reported in ITS studies; Table S4: PRISMA 2009 Checklist [21].