Article Text
PDF
Abstract
Background Antibiotics are overprescribed for children with upper respiratory infections (URIs), leading to unnecessary expenditures, adverse events and antibiotic resistance. This study assesses whether interventions antibiotic prescription rates (APR) for childhood URIs can be reduced and what factors impact intervention effectiveness.
Methods MEDLINE, Embase, Google Scholar, Web of Science, Global Health, WHO website, United States CDC website and The Cochrane Central Register of Controlled Trials (CENTRAL) were searched by December 2015. Cluster or individual-patient randomised controlled trials (RCTs) and non-RCTs that examined interventions to change APR for children with URIs were selected for meta-analysis. Educational interventions for clinicians and/or parents were compared with usual care.
Results Of 6074 studies identified, 13 were included. All were conducted in high-income countries. Interventions were associated with lower APR versus usual care (OR 0.63 (95% CI 0.50 to 0.81, p<0.001). A patient–clinician communication approach was the most effective type of intervention, with a pooled OR 0.41 (95% CI 0.20 to 0.83; p<0.001) for clinicians and 0.26 (95% CI 0.08 to 0.91; p=0.04) for parents. Interventions that targeted clinicians and parents were significant, with a pooled OR of 0.52 (95% CI 0.35 to 0.78; p=0.002). Insignificant effects were observed for targeting clinicians and parents alone, with a pooled OR of 0.88 (95% CI 0.67 to 1.16; p=0.37) and 0.50 (95% CI 0.10 to 2.51, p=0.40), respectively.
Conclusions Educational interventions are effective in reducing antibiotic prescribing for childhood URIs. Interventions targeting clinicians and parents are more effective than those for either group alone. The most effective interventions address patient–clinician communication. Studies in low-income to middle-income countries are needed.
- CHILD HEALTH
- HEALTH PROMOTION
- PUBLIC HEALTH
- METHODOLOGY
Statistics from Altmetric.com
Introduction
Worldwide, inappropriate medication use is a major problem. According to the WHO,1 50% of medicines are prescribed, dispensed or sold inappropriately, whereas 50% of patients take their medicines incorrectly.2 Inappropriate antibiotic use can lead to antibiotic resistance, resulting in difficult or impossible to treat infections.3 Antibiotic resistance is more common in countries with high rates of antibiotic prescription.3
Childhood upper respiratory infections (URIs) are very common, but are usually viral and self-limiting. Nevertheless, prescribing antibiotics for childhood URIs is highly prevalent.4 Antibiotic resistance is frequently observed in young children and more invasive infections occur in this vulnerable population.5 In Asia, every 2 min a child aged under 5 years dies from antibiotic-resistant infections.6
There are two main factors influencing inappropriate antibiotic use for childhood URIs—clinician prescribing and parent knowledge, attitude and demand.7 Educational interventions addressing these factors could reduce inappropriate antibiotic use. A Cochrane review showed that interventions involving physicians/pharmacists could reduce antibiotic prescription rates (APR).8 However, this review did not assess specifics of interventions (eg, intervention method, the intervention target and intensity). Conflicting results were seen with parental interventions. One review found that parental interventions can influence knowledge and behaviour, reducing consultation rates by 13–40%.4 However, another review showed that caregiver education may not be effective.9 Many published studies are descriptive, involving adults and children with various diseases.10–13 This study aims to analyse the effectiveness of different intervention approaches, targeting different groups (clinicians, parents or both) and whether other factors—study setting, study design, study period—influence effectiveness for reducing antibiotic prescribing for childhood URIs.
Methods
Search strategy
The PROSPERO registration number for this study is CRD42015029255. We searched MEDLINE, Embase, Google Scholar, Web of Science, Global Health, WHO website, United States CDC website and The Cochrane Central Register of Controlled Trials (CENTRAL) from 1980 to December 2015 for published articles without language restriction. Search terms included URIs, respiratory infections (RI), education, antibiotic prescription/prescribing, children/paediatric and antibiotic prescription rates. Two independent reviewers screened candidate studies using a structured form based on the PRISMA 2009 four-phase flow diagram.
Study selection
We included studies according to PICOS (population, intervention, comparison, outcome and setting) characteristics by following professional interventions in the Effective Practice and Organization of Care (EPOC) group scope.14
Population
We included studies of children (aged ≤18 years) diagnosed with any URI including rhinitis, sinusitis, pharyngitis, tonsillitis, acute otitis media or URI as a general category. To reduce misclassification bias, for studies that classified URIs and these aforementioned specific categories as separate, we included all patients as they should be classified under the category of ‘URI’ in our analysis.
Intervention
Approaches for targeting clinicians, featuring: (1) APR feedback; (2) update and/or reinforcement of national guidelines; (3) promoting delayed prescriptions; (4) clinician–parent communication skills training and workshops. Intervention methods included: (1) face-to-face training such as seminars, workshops or group discussion by trained peer leaders; (2) indirect training, which included online workshops or pop-up messages through software or printed information related to appropriate antibiotic use. Approaches for targeting parents included: (1) printed educational materials including leaflets/pamphlet or posters; (2) mass media such as video, radio and newspapers; (3) clinician–patient communication, which included facilitating patient health literacy, explaning appropriate antibiotic use by clinicians.
Comparator
We included prospective studies with an intervention group compared to a control with usual care. Study designs were (clustered) randomised controlled trials (RCTs) and non-RCTs including cohort experimental studies.
Outcome
Studies with antibiotic prescription expressed either as a rate (%) or as numbers per person-time were included. APR was defined as the number of children who were prescribed one or more antibiotic classes divided by the total number of children assessed for URIs during a designated interval.
Setting
Studies from all geographic regions were eligible for inclusion. Study sites were included if they cared for children either in primary care/general practice or in specialty clinics.
Quality assessment
Quality was determined by two independent reviewers using the Cochrane risk of bias tool,14 including domains related to sequence generation, allocation concealment, blinding, incomplete outcome data, selective reporting, intervention contamination, seasonal data collection and reporting of clustering coefficient. Each item outcome was categorised as high risk, low risk or unclear risk according to information provided. Disagreements between reviewers were resolved by discussion and consensus.
Data extraction
For the characteristic table, we extracted the following variables: study design, setting, follow-up duration, children age, participants, details of interventions (target group, intervention content for clinician and parents, and intervention methods), details of the comparator and outcome measures. To calculate the intervention group APR OR, we extracted the number of children or visits prescribed any antibiotic from both groups as well as the total number of children or visits in each group. For studies that were designed as cluster RCTs, we also extracted the intracluster correlation coefficients (ICCs) to adjust for design effect according to the Cochrane handbook.14 Furthermore, for studies with unknown ICC, we estimated from similar trials or by an approximated ICC.14 ,15 Studies with more than one control or time point were treated as if controls and time points were independent of each other.16
Analysis
Only studies with a calculable or reported APR were included into the meta-analysis. OR and 95% CI were used to measure the effectiveness of intervention compared to usual care. Overall, p value was used for the interaction between intervention and the estimates. We pooled studies using a Bayesian random-effects model to account for variations across RCTs in populations, interventions, settings and other factors during meta-analyses.17 ,18 Heterogeneity across studies was evaluated using the I2 test. To explore heterogeneity source, meta-regressions were conducted for each potentially influential factor (target group, follow-up duration, region, design and year). For cluster studies, additional factors (ICC reported or unreported, number of clusters and whether sites were paediatric clinics) were included. We also conducted a sensitivity analysis on individual studies to identify potential effects of outliers. For studies designed as cluster RCTs, we conducted another sensitivity analysis on ICC factors to determine whether the design effect influenced the study outcome. According to the Cochrane handbook, we selected three different ICC values (0.004, 0.02 and 0.2) for this sensitivity analysis. For studies with more than one control group and time point within a study, we also compared the result by merging effects within each study.16 Finally, we used a funnel plot with Egger's regression test to assess for publication bias (p<0.1). All analyses used STATA V.13.
Results
Description of included interventions
Of 6074 articles, 373 were accessed with full text. After exclusions, 12 articles were eligible for meta-analysis (figure 1).
Summary of included and excluded relevant articles in the review process. APR, antibiotic prescription rates; CENTRAL, The Cochrane Central Register of Controlled Trials; URI, upper respiratory infections.
Of the 12 studies included into this systematic review and meta-analysis, 7 were cluster RCTs, 3 were non-RCTs and 2 were individual RCTs. One study had two control groups;19 six studies were conducted in the USA, two in Israel, two in Norway and one each in UK, Iran and Canada. Eight of 12 were conducted in primary care or general practices and the remaining four studies were conducted in paediatric practices. The study year ranged from 2000 to 2014, with follow-up duration lasting from 1 to 12 months. Finkelstein et al 20 had two outcomes for age 3–36 and 36–72 months. Gonzales et al19 had two control groups with one located near to the intervention site and the other far from it. Regev-Yochay et al21 had four time point outcomes for different intervention methods—year 1 was workshops for determinants of reducing antibiotic prescriptions, year 2 focused on patient–clinician communication, year 3 involved workshops for APR feedback and year 4 was for follow-up after intervention. The cluster numbers ranged from 2 to 286 units, and the number of participating clinicians and registered patients ranged from 27 to 578 and 81 to 97 699, respectively (table 1). Nine articles were found to have low risk of bias, and 3 had high risk (table 2).
Basic characteristics of the included studies (n=12)
Summary of risk of bias of included studies (n=12)
Intervention effects, all studies
On the basis of heterogeneity, we used random effects to deal with differences among studies (figure 2). We combined the control groups and combined the different time point outcomes within studies. The pooled OR of APR for the intervention group was 0.63 (95% CI 0.50 to 0.81; p<0.001). However, significant heterogeneity was observed (I2=66%) as a result of differences in design, population and intervention details.
Pooled OR and 95% CI for reducing antibiotic prescription on childhood upper respiratory infections.
Effects of intervention strategies
Among the clinician interventions, four of nine studies used guidelines for RI and two used APR feedback to clinicians (figure 3). Delayed prescription was used in one study. Three studies used patient–clinician communication skills training. Training and workshops lasted from 40 min to 2 days. Training lasted <1 day in six studies and >1 day for the remaining three studies. Printed leaflets/posters were used in three of eight studies involving parents. Three studies used patient–clinician communication intervention. Two studies used video in waiting areas, lasting 5–8 min.
Forest plot result (OR) across different intervention types for clinicians (A) and parents (B). APR, antibiotic prescription rates.
Meta-analysis showed that the pooled OR for all types of intervention approach with clinicians (figure 3A) was 0.65 (95% CI 0.54 to 0.79; I2=44%), for patient–clinician communication approach was 0.41 (95% CI 0.20 to 0.83; I2=73%), for APR feedback was 0.65 (95% CI 0.49 to 0.87; I2=0%), for delayed prescription was 0.86 (95% CI 0.65 to 1.13; n=1) and for guideline use was 0.68 (95% CI 0.53 to 0.88; I2=21%). Though there were overlaps in 95% CI, testing for subgroup interaction was insignificant (p=0.2).
For intervention approach with parents (figure 3B), the pooled OR for all was 0.55 (95% CI 0.36 to 0.84; I2=36%), for video was 0.86 (95% CI 0.56 to 1.31; I2=0%), for leaflets/posters the pooled OR was 0.74 (95% CI 0.49 to 1.12; I2=0%) and for patient–clinician conmmunication was 0.26 (95% CI 0.08 to 0.91; I2=86%). No significant subgroup difference was found.
Effects of interventions through targeted group, study design, study year, follow-up duration and intensity of intervention and ICC report
When studies were grouped according to the intervention target, four studies targeting clinicians achieved a pooled OR of 0.88 (95% CI 0.67 to 1.16; I2=73%; table 3). Three studies targeting parents achieved an OR of 0.50 (95% CI 0.10 to 2.51; I2=80%). A total of five studies that had interventions targeting both groups achieved a pooled OR of 0.52 (95% CI 0.34 to 0.79; I2=50%).
Results of meta-analysis of all included studies
For studies with different designs, the pooled OR for all RCTs (including cluster RCTs) was 0.56 (95% CI 0.41 to 0.78; I2=74%). Non-RCTs had a similar pooled OR of 0.84 (95% CI 0.61 to 1.17; I2=0%). Five studies were conducted in 2010s and seven were in 2000s. The pooled OR was for studies conducted in 2000s was 0.59 (95% CI 0.36 to 1.00; I2=66%) and for studies in 2010s was 0.66 (95% CI 0.49 to 0.89; I2=72%).
For studies with different follow-up durations, six had follow-up durations from 1 to 6 months, with a pooled OR of 0.62 (95% CI 0.43 to 0.90, I2=77%), whereas the pooled OR for the other six studies with follow-up durations from 7 to 12 months was 0.59 (95% CI 0.45 to 0.79, I2=17%). For the nine face-to-face training studies, the pooled RR was 0.77 (95% CI 0.65 to 0.92; I2=36%), whereas three studies with written or online training had a pooled OR of 0.38 (95% CI 0.21 to 0.70; I2=44%). However, subgroup interaction here was significant (p=0.03).
When evaluating only cluster trials, the results were similar to including all studies. Studies that reported ICC achieved a pooled OR of 0.52 (n=4, 95% CI 0.33 to 0.84; p=0.007; I2=76%), whereas studies with unreported ICC had a higher OR of 0.81 (n=5, 95% CI 0.67 to 0.98; I2=13%). Paediatric clinical settings achieved a pooled OR of 0.61 (95% CI 0.47 to 0.79; I2=0%), which was similar to non-paediatric settings 0.60 (95% CI 0.43 to 0.85; I2=74%).
Meta-regression
To explore factors that might contribute to heterogeneity between studies, we also conducted meta-regressions on target group, follow-up duration, study design, study setting and study years, respectively, for all included studies. However, none of these factors were associated with between-study heterogeneity. Also when only considering cluster studies, none of these variables or additionally, ICC reported or unreported or study setting were associated with between-study heterogeneity.
Sensitivity analysis
We used three different ICCs to assess all cluster studies (ICC=0.04; ICC=0.02; ICC=0.2) without a reported ICC according to the following factors: target of intervention, study design, follow-up duration and study year for all studies and only cluster studies. Results were consistent across the three ICCs.
Sensitivity analysis on individual studies revealed that there was no change in heterogeneity after omitting any of the included articles.
Publication bias
Results from Egger's regression test revealed that publication bias was not significant (p=0.214).
Discussion
We have found that clinician–parent communication intervention appeared to have the strongest effect compared to other approaches. APR feedback and updated guidelines were effective in reducing APR for childhood URIs. Targeting clinicians and parents had a stronger effect compared with usual care, while targeting clinicians and parents alone did not reach statistical significant effect.
Among cluster trials, those with reported ICC had a stronger effect. None of the following factors including intervention target, follow-up duration, design, years, clinical setting and reported or unreported ICC were associated with residual variation due to heterogeneity. This is probably due to the presence of multiple sources of heterogeneity as well as difficulty in measuring sources of heterogeneity. However, heterogeneity declined by subgroup with heterogeneity for intervention with clinicians at 44%, whereas that for interventions with parents was 36%.
Previous systematic reviews were conducted to explore the effectiveness of interventions to reduce irrational antibiotic prescription in adults and children with various diseases.4 ,5 ,8 ,22–25 The advantages of our review over others are that we focus on a specific population (children) and condition (URI) and stratified analyses according to the type of intervention, which enriched the existing literature as most of the previous studies were descriptive. Additionally, very few studies describe intervention approaches. We found that the lowest pooled ORs were seen in the clinician–parent communication approach, This is consistent with the review by Davey et al12 for hospital inpatients. Davey found that clinician-targeted intervention using interactive meetings appeared more effective than didactic lectures, improved laboratory resources and consultation with specialists.
Combined interventions were found to be more effective than a single intervention alone.8 ,23 ,11 ,26 This review and prior research reached similar conclusions, in particular that the involvement of physicians and parents was most effective. Vodicka et al25 and Boonacker et al26 examined interventions to improve childhood antibiotic prescription for RIs. However, neither review used meta-analysis to measure relative risk. Both concluded that multifaceted interventions can reduce antibiotic use; however, providing printed materials and targeting only parents had limited effects, our results had a low OR suggesting a strong effect, but this results failed to achieve statistical significance. A review conducted by Thoolen showed that education to decrease inappropriate antibiotic use was not effective despite increased patient knowledge, which is contrary to the findings from Arnold and Straus8 and our analysis. We found that the pooled OR for targeting parents was 0.50, though it was statistically insignificant with a higher I2. In these studies, the change in knowledge was the primary outcome, while very few studies had data for APR that could be included into the meta-analysis. Nevertheless, education for parents could have a synergistic effect with clinician training and further reduce APR.27
For cluster RCTs, within-study variation can also influence the effect through variation at the cluster level.28 We observed a stronger effect in studies where ICC was reported. A study that reports the ICC may have been more carefully designed with more consideration of study design.
Our review has several strengths. First, to the best of our knowledge, this is the first study to examine the effect of interventions through APR, which is ultimately the desired effect. Second, sensitivity analysis with three different ICCs found trends to be consistent. Sensitivity analysis on excluding cluster non-RCTs had similar results, suggesting that the study design had limited effect on the results. This suggests that results were reliable. Third, the two studies with the widest CIs targeted clinicians and parents, and both were cluster RCTs. This may actually have led to underestimation of the effect of this type of intervention, making the result more conservative. Fourth, we evaluated the association between interventions and APR under subgroups of (1) different approaches to clinicians and parents, respectively; (2) study design and (3) study settings rather than just the overall effect, thus making our findings more specific.
In terms of limitations, first, non-RCTs had a higher selection bias than RCTs, whereas cluster RCTs had cluster-level selection bias, although we calculated the cluster RCT studies' ICC to adjust for design effects. It is impossible to avoid the nature of the existing selection bias within studies. Second, most of the studies we reviewed had multifaceted interventions, which were mixed to maximise the effect, making it difficult to evaluate individual components. At the same time, prescription is not a single clinician behaviour, but is influenced by different factors. This also increases the complexity of study design, which made data analysis complex and we could not analyse which intervention component is superior to the other. As no study provided feedback on the intervention implementation, it is difficult to know which component might contribute the most. Third, no study was from a developing country, which limits generalisability.
For policymakers, we hope we have better characterised the optimal intervention design. Future efforts should focus on an interactive approach that includes parents and clinicians, in addition to providing clinicians with information through feedback on their APR and guidelines. Communication skills between clinicians and parents should also be enhanced. Education for parents might facilitate improved communication.
Future studies of the quality of intervention implementation are needed. Given the variability in the content and intensity of a given type of intervention, more research is needed to understand optimal content/intensity. This is needed for children, especially given the frequency of URI in this population and the paucity of available proven therapies. There is a need to evaluate future interventions in the context of continuously improving diagnostics and therapeutics. More evidence on intervention sustainability is needed and the acceptance of interventions should be further explored based on the local context, resources and cost-effectiveness studies. The interventions in these trials should be studied against a theoretical framework of health behaviour change. Given the extent of antibiotic overuse in developing countries, further studies of interventions in these settings are needed.
What is already known on this subject
Antibiotic resistance is a global public health crisis, due in part to overprescribing of antibiotics, which is common for childhood URIs. Systematic reviews show that interventions with providers reduce antibiotic prescription, but with conflicting results about interventions with parents. Little is known about the relative benefits of each intervention, and no meta-analyses have been published.
What the study adds
To reduce antibiotic prescription for childhood URIs, the most effective interventions involve clinicians and parents. Improved communication between clinicians and parents is an essential part of antibiotic stewardship for childhood URIs.
Acknowledgments
The authors thank Yiwen Huang, the former research assistant, from COMDIS-HSD, Nuffield Centre for International Health Shenzhen office for the contribution to literature review and earlier version of drafting, Professor Robbie Foy from the Leeds Institute of Health Science, University of Leeds and Dr Rebecca King from Nuffield Centre for International Health and Development, University of Leeds for their comments and reviews of this manuscript.
References
Footnotes
Contributors YH designed the study, screened the paper, extracted and analysed data, wrote the manuscript and approved the final manuscript as submitted. JW wrote the manuscript, interpreted the data and approved the final manuscript. JIH wrote the manuscript, interpreted the data and approved the final manuscript. RC reviewed the revised manuscript and approved the final manuscript as submitted. JDT reviewed the revised manuscript and approved the final manuscript as submitted. GZ reviewed the revised manuscript and approved the final manuscript as submitted. XW reviewed the revised the manuscript and approved the final manuscript as submitted. JY reviewed the revised manuscript and approved the final manuscript as submitted. XW designed the study, extracted data, analysed and interpreted the results, wrote the manuscript and approved the final manuscript as submitted.
Funding This work was supported by Medical Research Council, Global Health Trials developmental grant—funding reference number: MR/M022161/1.
Competing interests None declared.
Provenance and peer review Not commissioned; externally peer reviewed.