RESEARCH BMJ: first published as 10.1136/bmj.n808 on 28 April 2021. Downloaded from http://www.bmj.com/ on 9 March 2023 at Sant Paul Hospital Consortia. Protected by copyright. Delayed antibiotic prescribing for respiratory tract infections: individual patient data meta-analysis Beth Stuart,1 Hilda Hounkpatin,1 Taeko Becque,1 Guiqing Yao,2 Shihua Zhu,1 Pablo Alonso-Coello,3 Attila Altiner,4 Bruce Arroll,5 Dankmar Böhning,6 Jennifer Bostock,7 Heiner C Bucher,8 Jennifer Chao,9 Mariam de la Poza,10 Nick Francis,1 David Gillespie,11 Alastair D Hay,12 Timothy Kenealy,5 Christin Löffler,4 David P McCormick,13 Gemma Mas-Dalmau,14 Laura Muñoz,15 Kirsty Samuel,16 Michael Moore,1 Paul Little1 For numbered affiliations see Abstract Results end of the article. Objective Data were obtained from nine randomised controlled Correspondence to: B Stuart To assess the overall effect of delayed antibiotic trials and four observational studies, totalling Academic Unit of Primary Care, prescribing on average symptom severity for patients 55 682 patients. No difference was found in follow- Population Sciences and Medical Education, Faculty of Medicine, with respiratory tract infections in the community, and up symptom severity (seven point scale) for delayed University of Southampton, to identify any factors modifying this effect. versus immediate antibiotics (adjusted mean Southampton SO16 5ST, UK difference −0.003, 95% confidence interval −0.12 to bls1@soton.ac.uk Design (ORCID 0000-0001-5432-7437) Systematic review and individual patient data meta- 0.11) or delayed versus no antibiotics (0.02, −0.11 Additional material is published analysis. to 0.15). Symptom duration was slightly longer in online only. To view please visit those given delayed versus immediate antibiotics the journal online. Data sources (11.4 v 10.9 days), but was similar for delayed versus Cite this as: BMJ 2021;372:n808 Cochrane Central Register of Controlled Trials, Ovid no antibiotics. Complications resulting in hospital http://dx.doi.org/10.1136/bmj.n808 Medline, Ovid Embase, EBSCO CINAHL Plus, and Web admission or death were lower with delayed versus Accepted: 15 March 2021 of Science. no antibiotics (odds ratio 0.62, 95% confidence Eligibility criteria for study selection interval 0.30 to 1.27) and delayed versus immediate Randomised controlled trials and observational antibiotics (0.78, 0.53 to 1.13). A significant cohort studies in a community setting that allowed reduction in reconsultation rates (odds ratio 0.72, comparison between delayed versus no antibiotic 95% confidence interval 0.60 to 0.87) and an increase prescribing, and delayed versus immediate antibiotic in patient satisfaction (adjusted mean difference prescribing. 0.09, 0.06 to 0.11) were observed in delayed versus Main outcome measures no antibiotics. The effect of delayed versus immediate The primary outcome was the average symptom antibiotics and delayed versus no antibiotics was severity two to four days after the initial consultation not modified by previous duration of illness, fever, measured on a seven item scale (ranging from normal comorbidity, or severity of symptoms. Children to as bad as could be). Secondary outcomes were younger than 5 years had a slightly higher follow-up duration of illness after the initial consultation, symptom severity with delayed antibiotics than with complications resulting in admission to hospital or immediate antibiotics (adjusted mean difference death, reconsultation with the same or worsening 0.10, 95% confidence interval 0.03 to 0.18), but no illness, and patient satisfaction rated on a Likert increased severity was found in the older age group. scale. Conclusions Delayed antibiotic prescribing is a safe and effective strategy for most patients, including those in higher risk subgroups. Delayed prescribing was associated What is already known on this topic with similar symptom duration as no antibiotic Clinical trials have suggested that delayed prescribing for respiratory tract prescribing and is unlikely to lead to poorer symptom infections is probably safe and effective for most patients control than immediate antibiotic prescribing. These clinical trials have been underpowered to look at subgroups or harms, and Delayed prescribing could reduce reconsultation might be subject to selection bias rates and is unlikely to be associated with an increase in symptoms or illness duration, except in young What this study adds children. Individual patient data from randomised controlled trials and observational Study registration studies were used to investigate the effectiveness of delayed antibiotic PROSPERO CRD42018079400. prescribing (compared with no antibiotics or immediate antibiotic prescribing), overall and for subgroups such as children and those with comorbidities Introduction Delayed prescribing was associated with similar symptom severity and duration Antimicrobial resistance is an important public health as no antibiotics, but patient satisfaction was higher and reconsultation rates concern.1 2 The burden of antimicrobial resistance has were lower; the effectiveness did not differ for any of the high risk subgroups increased substantially in recent years,3 and resistance Delayed prescribing is unlikely to lead to poorer symptom control than to second and third line antibiotics is predicted to immediate prescribing; older age was associated with increasing benefit on increase by 70% by 2030 if effective public health symptom severity two to four days after consultation measures are not implemented.4 the bmj | BMJ 2021;372:n808 | doi: 10.1136/bmj.n808 1 RESEARCH BMJ: first published as 10.1136/bmj.n808 on 28 April 2021. Downloaded from http://www.bmj.com/ on 9 March 2023 at Sant Paul Hospital Consortia. Protected by copyright. Reducing unnecessary and inappropriate use Study identification and selection of antibiotics is crucial to reduce antimicrobial Two researchers (HH and TB) searched the Cochrane resistance, particularly in primary care where Central Register of Controlled Trials, Ovid Medline, antibiotics are most prescribed.2 5 However, antibiotics Ovid Embase, EBSCO CINAHL Plus, and Web of Science are commonly used to treat acute respiratory tract to identify eligible quantitative studies (observational infections, despite studies showing that antibiotics cohort studies and RCTs). For observational studies, have, at best, modest effects.6-9 Guidelines recommend these searches were undertaken from inception to that the fewest number of antibiotic courses should 23 October 2017. For RCTs, we updated the searches be prescribed for the shortest period possible.10 11 undertaken in the most recent Cochrane review, and However, in the United Kingdom and internationally, searched from 26 May 2017 to 9 November 2017. antibiotics are still being overprescribed.12-16 Delayed We searched the International Standard Randomised antibiotic prescribing is a useful strategy that can Controlled Trial Number Registry, performed additional be used to help reduce antibiotic use, especially searches through Google, reviewed reference lists during consultations when patients expect to receive of identified papers, and contacted collaborators to an antibiotic prescription.6 A Cochrane review of identify any additional relevant studies. No language 10 trials found that delayed prescribing was as restrictions were reported. The full search strategy is effective as immediate prescribing in terms of clinical available in the protocol.19 Searches were rerun on 8 outcomes for cough and cold, but less effective for October 2020 but no additional eligible studies were reducing fever, pain, and malaise in some studies, identified. and with lower antibiotic use.9 However, the review Two reviewers (HH and TB) independently screened noted a high level of heterogeneity between studies titles and abstracts to determine inclusion criteria. that made combining them in a traditional meta- Both reviewers independently assessed the full text of analysis difficult, and did not allow sufficient power potentially relevant studies and determined eligibility. for the examination of subgroups of participants or Discrepancies were resolved through discussion with a complications. third reviewer (BS). These problems can be addressed in part by evidence synthesis using raw individual level data Data collection processes from relevant studies.17 18 Therefore, we conducted IPD were requested from the chief investigator for each a collaborative individual patient data (IPD) meta- eligible trial and observational study, initially by email analysis of randomised controlled trials (RCTs) and and if no response was received after two emails by observational cohort studies to determine the clinical letter or telephone call. Once data had been received effectiveness of a delayed prescribing strategy on from the original authors, a complete database of all outcomes for respiratory tract infection, overall and for study data was prepared in Stata (version 15).21 TB and key subgroups of people. HH performed internal consistency checks against the published data to ensure the published analysis could Methods be replicated. While the protocol contained a provision Protocol and registration to contact study authors about any discrepancies, this The systematic review and IPD meta-analysis did not prove necessary. were performed according to the published study Data relating to the general characteristics of the protocol that was registered with PROSPERO study were extracted, such as study design, country, (CRD42018079400) and was reported in line with setting, type of respiratory tract infection, average age, PRISMA-IPD (preferred reporting items for systematic and funding source. We requested all the variables that reviews and meta-analysis of individual participant had been collected in the individual studies from the data).19 20 authors and received the full datasets. These variables were used in the observational studies to calculate the Eligibility criteria propensity score. We included all observational cohort studies and RCTs The IPD dataset included baseline data on in a community setting that had a delayed antibiotic prescribing strategy, age (0-4, 5-15, 16-65, and >65 prescribing strategy (prescribed an antibiotic but years), fever at baseline consultation (greater than or advised the patient not to start taking the course less than 37.5ºC), previous duration of illness (above unless their condition deteriorated or failed to improve or below the median for each study), baseline severity after a set period), or a watchful waiting approach of symptoms (average severity across all symptoms (observation for a set period to allow spontaneous being above or below the median of each study), sex, symptom resolution before antibiotic prescription). smoking status (smoker or non-smoker), and lung Included studies also had a comparator group (no disease (asthma, coronary obstructive pulmonary antibiotic prescription or immediate prescription). We disease, or any other lung disease). Patients were excluded studies on antibiotic prescribing that were classified as having a comorbidity if they had any of not RCTs or observational cohorts (eg, cross sectional, the chronic conditions (eg, heart disease, diabetes) case-control, or survey studies), and studies on for which data were collected in the original study. patients in hospital. Follow-up data included symptom diaries (if collected) 2 doi: 10.1136/bmj.n808 | BMJ 2021;372:n808 | the bmj RESEARCH BMJ: first published as 10.1136/bmj.n808 on 28 April 2021. Downloaded from http://www.bmj.com/ on 9 March 2023 at Sant Paul Hospital Consortia. Protected by copyright. or days of illness determined by telephone interview, duration of illness, and a logistic regression model to complications resulting in admission to hospital or assess complications and reconsultation. All models death, reconsultation with same or worsening illness, controlled for baseline severity of illness, age, and and patient satisfaction. condition (acute sore throat, cough or chest infection, otalgia or otitis media, or upper respiratory tract Risk of bias assessment for included studies infection), and study type (RCT or observational). Two reviewers (HH and BS; TB and BS) independently All participants were included as randomised and assessed the risk of bias of each included study. RCTs the primary analysis was of complete cases (without were assessed using the Cochrane risk of bias tool imputation for missing data).27 for allocation bias (random sequence generation, We used inverse probability weighting by propensity allocation concealment, baseline imbalance), depar­ score analysis to adjust for baseline factor imbalance tures from intended interventions (participant and on measured covariates (such as age, sex, comorbid study personnel blinding, deviations from intended health conditions, and signs and symptoms at baseline interventions, and analysis in groups to which they consultation) in observational studies.28-30 Propensity were randomised), attrition bias and appropriate scores based on covariates associated with any of methods to account for missing data, detection bias the outcomes were derived for each observational (blinding of outcome assessors), and selective outcome study. We checked balance by using standardised reporting.22 RCTs were considered to have a high risk mean differences and the appendix figures show of bias if scored as such in more than one of the six the results. Propensity scores were also calculated domains. for the RCTs by using the probability of randomised We assessed observational cohort studies using intervention given baseline covariates.31 An inverse the ROBINS-I (risk of bias in non-randomised studies probability of treatment weighting regression was of interventions) tool for bias due to confounding, carried out for the combined observational and RCT selection bias, bias due to deviations from intended data to obtain a pooled estimate of treatment effect. intervention, and bias due to missing data and We assessed heterogeneity across studies with the selective reporting.23 Observational cohort studies I2 statistic (tested by Higgins I2 test).22 Substantial were considered to have a high risk of bias if judged statistical heterogeneity was considered to be present to be at serious or critical risk of bias in at least one of if the I2 statistic was greater than 50% and reasons for the domains. heterogeneity were explored.22 We repeated each model after including an intera­ Specification of outcome measures ction term between antibiotic prescribing strategy and The primary outcome of interest was the average subgroup characteristic to obtain summary estimates symptom severity two to four days after the initial of the subgroup effects (interactions) of interest, consultation. Symptom severity was measured on a which compared differential effects of interventions seven item scale (0-6: normal, very little problem, across the outcomes. The prespecified subgroups of slight problem, moderately bad, bad, very bad, interest were previous duration of illness (above or as bad as could be).24 Secondary outcomes were below median for the condition), age (<16, 16-64, duration of illness after the initial consultation, >65 years), fever at baseline consultation (>37.5°C), complications resulting in admission to hospital or comorbid conditions including lung comorbidity (such death, reconsultation with the same or worsening as asthma or chronic obstructive pulmonary disease), illness, and patient satisfaction rated on a Likert scale. and severity of symptoms at baseline consultation. Reconsultation and complications (defined as hospital We conducted several sensitivity analyses. All admission or death) were defined as binary outcomes analyses were repeated using a two stage approach: (yes or no). Patient satisfaction data were rescaled IPD for each study were first analysed separately and to a four point scale to allow comparison across then meta-analysed by using random effects models.18 studies.25 We performed a two stage meta-analysis of extracted study level data from RCTs that did not contribute Synthesis methods to the IPD to obtain summary estimates of effects of Study and patient level characteristics were described delayed antibiotic prescribing that combined IPD and for all studies that contributed IPD. We performed non-IPD RCT studies, and to assess IPD availability a one stage IPD meta-analysis to obtain summary bias.32 This process was not possible for observational estimates and 95% confidence intervals for delayed studies because papers did not control consistently antibiotic prescribing (compared with no antibiotic for the same confounding factors. Further sensitivity prescribing or immediate antibiotic prescribing) for analyses included repeating the analyses after each outcome measure.18 26 The one stage approach excluding studies with high risk of bias, and repeating combines all the data in a single meta-analysis based subgroup analyses with age, fever, and baseline on a suitable regression model, with a random effect severity treated as continuous variables. All meta- to account for individual studies. We used a linear analyses were undertaken with Stata software (version regression to model the severity of symptoms and 15)21 and statistical significance was considered at the patient satisfaction, a count model to assess the 5% level. the bmj | BMJ 2021;372:n808 | doi: 10.1136/bmj.n808 3 RESEARCH BMJ: first published as 10.1136/bmj.n808 on 28 April 2021. Downloaded from http://www.bmj.com/ on 9 March 2023 at Sant Paul Hospital Consortia. Protected by copyright. Certainty of evidence per outcome This feedback helped to inform our interpretation of We used the five GRADE (grading of recommenda­tions the findings. assessment, development and evaluation) considera­ tions (study limitations, consistency of effect, impre­ Results cision, indirectness, and publication bias) to assess the Study selection and IPD obtained quality of the evidence for our analysis of the primary We sought IPD from 22 eligible studies (14 RCTs outcomes.33 and eight observational studies), totalling 59 705 participants (fig 1).6 34-49 IPD were obtained from Patient and public involvement 13 studies (nine RCTs, four observational studies; Two patient and public involvement team members table 1), totalling 55 682 participants. We were (JB and KS) were involved in determining the research unable to obtain data from nine eligible studies question, defining outcome measures, study design, because of no response (n=6), researchers moving and implementation. They attended all study meetings on (n=2), or no response after initial agreement and are coauthors on this publication. We also shared (n=1).50-58 our research findings with a patient and public involvement panel, allowing them to feedback to us Study characteristics their interpretation of the evidence and how general Each study included between 129 and 28 856 practitioners might more effectively communicate this participants (median 557, interquartile range 316- information to patients. In the absence of published 2690). Participants belonged to a delayed antibiotic minimum clinically important differences for the prescription group and an immediate or no antibiotic outcomes considered in this study, it was particularly prescription group. Studies were conducted in the UK, helpful to discuss their interpretation as to whether the the United States, New Zealand, Spain, and one study differences observed represented a meaningful change. used data from multiple European studies. Most studies Randomised controlled trials Observational studies 204 3852 Records identified through database searching Records identified through database searching 1 191 Additional record identified through Records screened aer duplicates removed citation tracking, Google search, discussion with collaborators 187 Records excluded 1715 Records screened aer duplicates removed 4 Full text articles assessed for eligibility 1653 Records excluded 4 Full text articles excluded 62 1 Wrong study design Full text articles assessed for eligibility 3 Wrong comparator 54 14 Full text articles excluded Included 28 Wrong study design 11 Studies from previous Cochrane review 9 wrong outcomes 3 Additional records identified through 5 Duplicate discussion with collaborators 5 Commentary 5 Commentary to study already included 1 Wrong comparator 14 1 Wrong intervention Studies included in quantitative synthesis 8 14 Studies included in quantitative synthesis Studies included in quantitative synthesis 9 Individual patient data 5 Aggregate data 4 Studies included in quantitative synthesis 4 Individual patient data Fig 1 | PRISMA (preferred reporting items for systematic reviews and meta-analysis) flowchart for randomised controlled trials and observational studies on delayed antibiotic prescribing 4 doi: 10.1136/bmj.n808 | BMJ 2021;372:n808 | the bmj Table 1 | Characteristics of included studies No and age of Intervention and Outcome Reference Country and setting participants* Condition comparison group Type of delay SS SD R C PS Funding source Randomised controlled trials None, immediate, Prescription to be filled if symptoms did Little 1997 UK, primary care 716; 25.3 (17.0) Sore throat + + + + − Wessex NHS delayed not start to settle after 3 days None, immediate, Prescription collected after 72 hours if Little 2001 UK, primary care 315; 5.0 (2.8) Acute otitis media + + + + + NHS R&D delayed child still not improving New Zealand, primary Prescription to be filled after 3 days if Arroll 2002 129; 25.6 (23.0) Common cold Immediate, delayed − + − − + NZ Health Research Council care symptoms fail to improve No antibiotics unless returning with acute McCormick 2005 USA, paediatric clinic 223; 2.7 (2.7) Acute otitis media None, delayed + + + + + US National Institutes of Health ear symptoms within 30 days Lower respiratory tract None, immediate, Prescription to be filled if symptoms not Little 2005 UK, primary care 807; 39 (20.8) + + + + − Medical Research Council infection delayed resolved after 4 days USA, paediatric Advised to fill prescription if symptoms Chao 2008 232; 5.1 (2.4) Acute otitis media None, delayed − + + − + NR emergency department did not resolve in 2-3 days Acute respiratory tract Four types: recontact, postdated, Little 2014 UK, primary care 889; 31.0 (21.2) None, delayed + + + + + National Institute for Health Research the bmj | BMJ 2021;372:n808 | doi: 10.1136/bmj.n808 infection collection, patient led De La Poza Acute respiratory tract None, immediate, Collection or patient led if symptoms did Spain, primary care 405; 44.9 (16.6) + + + + + Spanish Ministry of Health Abad 2016 infection delayed not start to improve after a few days Acute respiratory tract None, immediate, Collection or patient led if symptoms did Mas-Dalmau 2021 Spain, primary care 437; 6.3 (3.1) + + + + + Spanish Ministry of Health infection delayed not start to improve after a few days Observational cohort studies European Commission; National Butler, Francis et al 13 European countries, 2690; 47.8 Cough or lower respira- None, immediate, Advised to fill if symptoms did + + + + + Institute for Health Research; Research 2012 primary care network (16.3) tory tract infection delayed not start to improve after 2-7 days Foundation-Flanders 12 626; 34.0 None, immediate, Medical Research Council; National Little 2013 UK, primary care Sore throat Patient led + + + + − (14.6) delayed Institute for Health Research 28 856; 51.7 Acute lower respiratory None, immediate, Advised to fill if symptoms did not start to Little 2017 UK, primary care − − + + − National Institute for Health Research (17.9) tract infection delayed improve; median advised delay=3 days Acute cough and res- None, immediate, Advised to fill if symptoms did not start to Hay 2016 UK, primary care 8320; 3.9 (3.7) + + + + − National Institute for Health Research piratory tract infection delayed improve; median advised delay=3 days Studies for which individual patient data unavailable 114; 7.5 (2.6); RobertWood Johnson Foundation, Eli Lilly & Pichichero 1987 USA, paediatric clinic Sore throat Immediate, delayed No detail + + + − − 7.8 (2.3) Co, and Elmwood Paediatric Research fund Gerber 1990 USA, paediatric clinic 113; NR Sore throat Immediate, delayed No detail + − − − − NR 229; 7.8 (0.23); Biochemie GmbH and Jordan University of El-Daher 1991 Jordan, paediatric clinic Sore throat Immediate, delayed No detail + − − − − 8.3 (0.24) Science and Technology Patient to pick up prescription Dowell 2001 UK, primary care 191; 41.6 Cough Immediate, delayed − + + − + Royal College of General Practitioners after 1 week of delay USA, emergency US National Institutes of Health and Yale Spiro 2006 283; 3.2 Acute otitis media Immediate, delayed No detail − + − − − department University School of Medicine To be filled only if symptoms did not Siegel 2003 USA, paediatric clinic 194; 5 Acute otitis media Immediate, delayed − − − − − Whitehall-Robins Healthcare improve within 2 days Marchetti 2005 Italy, primary care 1672; 4.8 Acute otitis media Immediate, delayed No detail − − − − − NR USA, emergency To be filled only if symptoms Fischer 2009 144; NR Acute otitis media Immediate, delayed − − − − − NR department did not improve within 2 days 120; 47.6 (16.3); Acute cough or sore Kavanagh 2011 Ireland, primary care Immediate, delayed No detail − − + − + WestREN, MSD 48 (17.8) throat C=complication; NR=nor reported; PS=patient satisfaction; R=reconsultation; SD=symptom duration; SS=symptom severity. *Mean (standard deviation). RESEARCH 5 BMJ: first published as 10.1136/bmj.n808 on 28 April 2021. Downloaded from http://www.bmj.com/ on 9 March 2023 at Sant Paul Hospital Consortia. Protected by copyright. RESEARCH BMJ: first published as 10.1136/bmj.n808 on 28 April 2021. Downloaded from http://www.bmj.com/ on 9 March 2023 at Sant Paul Hospital Consortia. Protected by copyright. were conducted in primary care settings (n=11/13). Mean symptom severity two to four days after Other settings included a paediatric emergency depart­ consultation ment (n=1) and paediatric clinic (n=1). Mean age of One stage random effects IPD meta-analysis of study participants ranged from 2.7 to 51.7 years. Six, individual RCTs and observational studies combined four, and three studies examined all age groups, adult found that, overall, there was no significant difference populations only, and paediatric populations only, in symptom severity between delayed antibiotics respectively. One study focused on the common cold, and no antibiotics (mean difference on seven point two studies each assessed sore throat and cough, three scale −0.003, 95% confidence interval −0.12 to 0.11; focused on acute otitis media, and seven included seven studies, 3907 participants; table 4, fig 3). No more than one respiratory tract infection. Eleven out significant difference was found in symptom severity of 13 studies (84.6%) reported symptom severity and between delayed and immediate antibiotics (0.02, complication outcomes, 12 studies (92.3%) reported −0.11 to 0.15; eight studies, 3752 participants; table data on symptom duration and reconsultation, and 4, fig 3). Consistent results were obtained using a two eight studies (61.5%) reported patient satisfaction. stage approach. Length of follow-up was 28-30 days. Eligible studies that did not contribute IPD data (five Subgroup effects trials and four observational studies) were generally None of the prespecified subgroup variables modified smaller, based on younger populations, and had a the effectiveness of delayed antibiotic prescribing higher proportion focused on acute otitis media and relative to no antibiotics (table 5). We found a signifi­ sore throat than IPD studies (table 2). Aggregate data cant overall interaction effect of age on the effectiveness were available for 930 patients from five RCTs that did of delayed relative to immediate antibiotic prescribing not contribute IPD. (mean difference −0.10, 95% confidence interval The mean age of IPD study participants was 38.7 −0.17 to −0.03). Children younger than 5 years had a years (table 3). Patients in the delayed antibiotics slightly higher follow-up symptom severity score two group were younger than those prescribed immediate to four days after consultation with delayed versus antibiotics. A lower proportion of patients in the immediate antibiotics (0.10, 0.03 to 0.18), whereas no delayed antibiotic group had high baseline severity, significant difference was found in severity between longer previous duration of illness, fever at baseline, delayed and immediate antibiotics for other age groups or lung disease compared with those in the immediate (table 5). antibiotics group (table 3). Secondary outcomes IPD integrity and risk of bias Time to symptom resolution was longer with For all included IPD, we were able to replicate aggregate delayed (11.4 days) than immediate antibiotics results that were reported in each of the associated (10.9 days; hazard ratio 1.04, 95% confidence publications. Each individual study contributing IPD interval 1.01 to 1.08). Reconsultation rates was deemed low or moderate risk of bias (fig 2), except were lower with delayed (13%) than with no for two RCTs that were judged to be high risk of bias antibiotics (17%; odds ratio 0.72, 95% confidence on two domains. We also assessed the risk of bias of interval 0.60 to 0.87), but were not statistically studies that did not contribute IPD. These studies were significantly different for delayed (16%) versus judged to have potentially high (n=6) or unclear (n=2) immediate antibiotics (22%; odds ratio 0.95, 95% risk of bias (fig 2), and were more likely to be at risk of confidence interval 0.74 to 1.22). Complications selection bias but also more likely to have been low risk resulting in hospital admission or death were lower of bias with respect to blinding. with delayed than with no antibiotics (odds ratio 0.62, 95% confidence interval 0.30 to 1.27) and lower in delayed than immediate antibiotics (0.78, Table 2 | Comparison of included and excluded study characteristics in individual patient 0.53 to 1.13), but neither result was statistically data (IPD) significant. Patient satisfaction was higher with No (%) of studies delayed (3.04 points) than no antibiotics (2.96), Eligible study characteristics Included in IPD Excluded from IPD but by a small difference (mean difference 0.09, Population source 95% confidence interval 0.06 to 0.11; table 4). Primary care 11 (84.6) 3 (33.3) Emergency department 1 (7.7) 2 (22.2) Paediatric office (USA) 1 (7.7) 4 (44.4) Quality of evidence across studies Condition Based on GRADE, the overall quality of the evidence Common cold 1 (7.7) 0 (0.0) for all outcomes in the IPD dataset was judged as Acute otitis media 3 (23.1) 4 (44.4) moderate, apart from patient satisfaction which was Sore throat 2 (15.4) 4 (44.4) low. Table 6 provides a full evidence profile. Two RCTs Cough 2 (15.4) 2 (22.2) were deemed higher risk because of lack of blinding Respiratory tract infection 7 (53.8) 0 (0.0) Antibiotic group and allocation concealment, which lowered the rating None 12 (92.3) 2 (22.2) for risk of bias to serious. However, consistent effects Immediate 11 (84.6) 7 (77.8) across RCTs suggest results are likely to be unbiased. Delayed 12 (92.3) 7 (77.8) Observational studies were considered high quality 6 doi: 10.1136/bmj.n808 | BMJ 2021;372:n808 | the bmj Table 3 | Summary statistics for IPD database. Values are numbers (%) unless stated otherwise Combined RCTs Observational studies No No No No of No of antibiotics Immediate Delayed No of No of antibiotics Immediate Delayed No of No of antibiotics Immediate Delayed Characteristics studies participants (n=19 360) (n=28 129) (n=8193) studies participants (n=985) (n=1082) (n=1745) studies participants (n=18 375) (n=27 047) (n=6448) Age, years 13 55 644 31.6 (23.4) 44.1 (22.3) 37.0 (22.4) 9 3779 27.2 (21.9) 23.4 (22.4) 24.2 (21.2) 4 51 865 31.8 (23.4) 44.9 (21.9) 40.4 (21.4) (mean (SD)) Sex 12 54 848 — — — 8 2992 — — — 4 51 856 — — — Male — — 7714 (40.4) 11 302 (40.6) 3047 (38.5) — — 281 (39.7) 352 (43.0) 621 (42.4) — — 7433 (40.5) 10 950 (40.5) 2426 (37.6) Female — — 11 363 (59.6) 16 557 (59.4) 4865 (61.5) — — 427 (60.3) 466 (67.0) 845 (67.6) — — 10 936 (59.5) 16 091 (59.5) 4020 (62.4) Condition 13 55 682 — — — 9 3812 — — — 4 51 870 — — — Cough 4 — 14 067 (72.7) 21 497 (76.4) 5031 (61.4) 1 — 273 (27.7) 262 (24.2) 271 (15.5) 3 — 13 794 (75.1) 21 235 (78.5) 4760 (73.8) Sore throat 2 — 4813 (24.9) 6058 (21.5) 1926 (23.5) 1 — 232 (23.6) 246 (22.7) 238 (13.6) 1 — 4581 (24.9) 5812 (21.5) 1688 (26.2) RTI 4 — 364 (1.9) 311 (1.1) 846 (10.3) 4 — 364 (40.0) 311 (28.7) 846 (48.5) — — — — — Otitis media 3 — 116 (0.6) 263 (0.9) 390 (4.8) 3 — 116 (11.8) 263 (24.3) 390 (22.3) — — — — — Baseline severity 1.5 1.5 1.2 average score, 13 54 956 0.7 (0.4-1.1) 0.9 (0.6-1.2) 0.8 (0.5-1.2) 8 3125 4 51 831 0.7 (0.4-1.0) 0.8 (0.6-1.2) 0.8 (0.5-1.1) (0.75-2.0) (0.75-2.1) (0.8-1.8) the bmj | BMJ 2021;372:n808 | doi: 10.1136/bmj.n808 0-4 (median (IQR)) Below median — — 14 387 (75.2) 18 048 (64.7) 5616 (70.8) — — 400 (51.3) 490 (57.2) 895 (60.1) — — 13 987 (76.2) 17 558 (65.0) 4721 (73.2) Median and — — — — — — 4750 (24.8) 9837 (35.3) 2318 (29.2) 380 (48.7) 367 (42.8) 593 (39.9) 4370 (23.8) 9470 (35.0) 1725 (26.8) above Previous — — — — — — — — — 11 54 462 6 2700 4 51 762 duration illness Below median — — 7794 (40.9) 10 388 (37.5) 3565 (46.3) — — 381 (52.0) 403 (57.4) 683 (53.9) — — 7413 (40.5) 9985 (37.0) 2882 (44.8) Median and — — — — — — 11 260 (59.1) 17 318 (62.5) 4137 (53.7) 351 (48.0) 299 (42.6) 583 (46.1) 10 909 (59.5) 17 019 (63.0) 3554 (55.2) above Fever at baseline 12 53 761 — — — 8 2896 — — — 4 50 865 — — — — — — — 1002 — — No 11 169 (59.4) 13 472 (49.3) 4483 (58.8) 604 (79.7) 668 (81.7) 10 565 (58.6) 12 804 (48.3) 3381 (53.7) (82.1) Yes — — 7634 (40.6) 13 865 (50.7) 3138 (41.2) — — 154 (20.3) 150 (18.3) 218 (17.9) — — 7480 (41.5) 13 715 (51.7) 2920 (46.3) Any lung disease 11 47 504 — — — 7 2389 — — — 4 45 115 — — — — — — — 1036 — — No 11 719 (80.1) 19 177 (74.2) 5618 (80.1) 514 (88.9) 560 (88.9) 11 205 (79.7) 18 617 (73.8) 4582 (78.5) (87.7) Yes — — 2914 (19.9) 6679 (25.8) 1397 (19.9) — — 64 (11.1) 70 (11.1) 145 (12.3) — — 2850 (20.3) 6609 (26.2) 1252 (21.5) Follow-up symptom severity 2.4 2.0 2.0 9 7074 1.8 (1-3) 2.03 (1.2-3.5) 2 (1.1-3.3) 6 2148 3 4926 1.7 (1.0-2.8) 2.1 (1.2-3.7) 2.0 (1.2-3.9) score, 0-6 (1.4-3.8) (0.8-3.0) (1.0-3.0) (median (IQR)) Symptom duration, days 11 8607 10 (5-28) 9 (4-28) 7 (3-12) 8 2918 5 (2-11) 6 (3-10) 6 (3-10) 3 5689 13 (6-28) 13 (5-28) 9 (4-28) (median (IQR)) Reconsultation 11 54 105 — — — 8 2840 — — — 4 51 265 — — — No — — 15 632 (83.2) 21 483 (77.3) 6433 (85.4) — — 556 (72.8) 678 (71.2) 842 (74.9) — — 15 076 (83.7) 20 805 (77.5) 5591 (87.2) Yes — — 3155 (16.8) 6302 (22.7) 1100 (14.6) — — 208 (27.2) 274 (28.8) 282 (25.1) — — 2947 (16.4) 6028 (22.5) 818 (12.8) Patient satisfaction score, 1-4 8 4584 4 (3-4) 4 (3-4) 3.2 (2.7-4) — 2100 4 (2.7-4) 4 (3-4) 3.2 (2.7-4) — 2484 3 (3-4) 4 (3-4) 4 (3-4) (median (IQR)) IPD=individual patient data; IQR=interquartile range; RCT=randomised controlled trial; RTI=respiratory tract infection; SD=standard deviation. RESEARCH 7 BMJ: first published as 10.1136/bmj.n808 on 28 April 2021. Downloaded from http://www.bmj.com/ on 9 March 2023 at Sant Paul Hospital Consortia. Protected by copyright. RESEARCH BMJ: first published as 10.1136/bmj.n808 on 28 April 2021. Downloaded from http://www.bmj.com/ on 9 March 2023 at Sant Paul Hospital Consortia. Protected by copyright. and achieved balance on potential covariates, but were was the only study to test several delayed strategies downrated because residual confounding could not be in a single trial. When data from the Little study were ruled out. excluded from the analyses, the heterogeneity within the RCTs was reduced (I2=0%), and also the overall Sensitivity analyses heterogeneity between observational studies and RCTs One stage versus two stage IPD analysis—consistent was reduced (P=0.25, I2=0%). The results remained results were observed for analyses using a two stage consistent with the main analysis (no significant approach. We found no significant difference in difference in treatment effect). We did not observe symptom severity between delayed and no antibiotics any important variability for analyses that explored (mean difference 0.04, 95% confidence interval −0.12 delayed versus immediate antibiotic comparison to 0.20) or between delayed and immediate antibiotics (heterogeneity between observational studies and (0.06, −0.05 to 0.17). RCTs; P=0.02, I2=24%). Exploring heterogeneity—additional sensitivity Subgroup analyses with continuous variables—when analyses explored the effect of heterogeneity across we replaced dichotomised variables with a continuous studies. For symptom severity analyses, heterogeneity variable for each subgroup, the subgroup results did was found within the RCTs (I2=65%), and also between not change. One exception was that patients with observational studies and RCTs (P<0.005, I2=68%), for lower baseline severity had lower follow-up symptom delayed versus no antibiotics. The forest plots clearly severity with delayed versus immediate antibiotics showed that the results for Little (2014) were different (mean difference −0.27, 95% confidence interval from the other included studies, perhaps because it −0.34 to −0.19). Randomised controlled trials Observation studies Bias due to deviations from intended interventions Blinding (performance bias and detection bias) Random sequence generation (selection bias) Incomplete outcome data (attrition bias) Allocation concealment (selection bias) Bias in classification of interventions Selective reporting (reporting bias) Measurement of outcome bias Bias due to confounding Selective reporting bias Missing data bias Selection bias Other bias Study Study Pichichero 1987 Siegel 2003 Dowell 2001 Marchetti 2005 El-Daher 1991 Fischer 2009 Gerber 1990 Kavanagh 2011 Little 1997 Francis 2012 Little 2001 Little 2013 Arroll 2002 Little 2017 McCormick 2005 Hay 2016 Little 2005 Spiro 2006 Chao 2008 Little 2014 (PIPS) Low risk Unclear risk De la Poza Abad 2016 Moderate risk Mas-Dalmau 2021 High risk Fig 2 | Risk of bias for randomised controlled trials (using the Cochrane risk of bias tool) and observation studies (risk of bias in non-randomised studies of interventions (ROBINS-I) tool). Bold studies contributed to individual patient data 8 doi: 10.1136/bmj.n808 | BMJ 2021;372:n808 | the bmj Table 4 | Effect of antibiotic prescribing strategy on all clinical outcomes in one stage random effects IPD meta-analysis. Values are means (95% confidence intervals) unless stated otherwise Delayed v no antibiotics Delayed v immediate antibiotics No of No of No of No of participants participants Adjusted* estimate participants participants Adjusted* estimate Outcome (none) (delayed) No antibiotics Delayed (95% CI) (immediate) (delayed) Immediate Delayed (95% CI) Follow-up symptom severity score on days 2-4 the bmj | BMJ 2021;372:n808 | doi: 10.1136/bmj.n808 RCT+OS 2716 1192 2.499 (1.44 to 3.56) 2.496 (1.47 to 3.52) −0.003 (−0.12 to 0.11) 2699 1053 2.68 (1.26 to 4.11) 2.70 (1.36 to 4.05) 0.02 (−0.11 to 0.15) RCT 484 815 2.77 (1.13 to 4.11) 2.86 (1.6 to 4.06) 0.09 (−0.10 to 0.28) 606 674 2.29 (0.66 to 3.93) 2.40 (0.78 to 4.03) 0.11 (−0.004 to 0.22) OS 2232 377 2.22 (1.86 to 2.59) 2.12 (1.74 to 2.50) −0.10 (−0.12 to −0.08)† 2093 379 2.75 (2.24 to 3.26) 2.63 (2.05 to 3.20) −0.12 (−0.33 to 0.07) Time to complete symptom resolution (days) RCT+OS 3091 1123 11.5 (6.4 to 16.6) 11.6 (6.4 to 16.6) 1.00 (0.82 to 1.23)‡ 3275 1442 10.9 (6.3 to 15.6) 11.4 (6.8 to 15.9) 1.04 (1.01 to 1.08)†‡ RCT 540 647 9.42 (4.22 to 14.62) 9.88 (4.37 to 15.38) 1.04 (0.95 to 1.13) 876 962 9.94 (7.17 to 12.70) 11.37 (8.19 to 14.54) 1.14 (1.06 to 1.22)† OS 2551 476 15.1 (14.8 to 15.4) 15.5 (15.2 to 15.8) 0.98 (0.94 to 1.01) 2399 480 11.96 (5.90 to 18.03) 12.37 (6.40 to 18.35) 1.02 (0.97 to 1.07)‡ Duration of moderately severe or severe symptoms (days) RCT+OS 2997 1251 6.2 (5.0 to 7.5) 6.3 (5.0 to 7.5) 1.01 (0.91 to 1.12) 2744 902 5.8 (4.5 to 7.2) 6.4 (5.3 to 7.6) 1.08 (1.01 to 1.17)† RCT 447 775 6.88 (4.21 to 9.56) 6.81 (4.16 to 9.46) 1.04 (0.94 to 1.15) 347 426 6.36 (5.85 to 6.86) 7.48 (6.94 to 8.02) 1.18 (1.06 to 1.31)†‡ OS 2550 476 6.59 (5.45 to 6.74) 6.74 (6.60 to 6.87) 0.97 (0.93 to 1.01) 2397 476 5.36 (4.02 to 6.70) 5.94 (4.86 to 7.02) 1.05 (0.96 to 1.16)‡ Reconsultation (%, 95% CI) RCT+OS 16 232 5901 17 (4 to 58) 13 (2 to 52) 0.72 (0.60 to 0.87)† 22 430 6157 22 (4 to 58) 16 (3 to 59) 0.95 (0.74 to 1.22) RCT 509 611 25 (15 to 22) 21 (10 to 31) 0.92 (0.67 to 1.26) 796 865 25 (5 to 41) 24 (8 to 41) 1.29 (0.84 to 1.99) OS 15 723 5290 16 (4 to 60) 12 (2 to 54) 0.54 (0.49 to 0.60)† 21 634 5292 22 (4 to 58) 15 (3 to 59) 0.70 (0.66 to 0.75)† Complication (hospital admission or death; %, 95% CI) RCT+OS 16 364 5827 0.6 (0.03 to 14) 0.4 (0.02 to 11) 0.62 (0.30 to 1.27) 22 371 6017 0.8 (0.1 to 5) 0.6 (0.1 to 4) 0.78 (0.53 to 1.13) RCT 431 530 0.6 (0.02 to 14) 2.09 (0.01 to 22) 0.35 (0.07 to 1.92) 650 719 0.9 (0.3 to 5) 1.0 (0.3 to 4) 1.25 (0.38 to 4.16) OS 15 933 5297 0.6 (0.01 to 3.7) 0.4 (0.01 to 2.2) 0.64 (0.28 to 1.43) 21 721 5298 0.7 (0.3 to 1.7) 0.6 (0.2 to 1.6) 0.22 (0.19 to 0.27)† Patient satisfaction score RCT+OS 1434 674 2.96 (2.55 to 3.36) 3.04 (2.64 to 3.44) 0.09 (0.06 to 0.11)† 1743 805 2.99 (2.33 to 3.64) 2.87 (2.25 to 3.49) −0.12 (−0.26 to 0.03) RCT 433 520 3.17 (2.82 to 3.51) 3.26 (2.92 to 3.60) 0.06 (−0.03 to 0.16) 563 649 3.20 (2.33 to 4.06) 3.07 (2.22 to 3.92) −0.13 (−0.31 to 0.05) OS 1001 154 2.77 (2.15 to 3.39) 2.85 (2.25 to 3.46) 0.10 (−0.03 to 0.23) 1180 156 2.88 (2.07 to 3.69) 2.77 (2.01 to 3.52) −0.06 (−0.18 to 0.06) Adjusted estimate=coefficient, odds ratio, or relative risk; IPD=individual patient data; OS=observational study; RCT=randomised controlled trial. *Adjusted for baseline severity, age, and condition. †Statistically significant result. The number of observations for each outcome varies as not all studies collected data on all outcomes; analyses used all available data for each outcome. ‡Estimated using fixed study specific effects and random treatment effect due to convergence difficulties. RESEARCH 9 BMJ: first published as 10.1136/bmj.n808 on 28 April 2021. Downloaded from http://www.bmj.com/ on 9 March 2023 at Sant Paul Hospital Consortia. Protected by copyright. RESEARCH BMJ: first published as 10.1136/bmj.n808 on 28 April 2021. Downloaded from http://www.bmj.com/ on 9 March 2023 at Sant Paul Hospital Consortia. Protected by copyright. Including data from studies that did not provide Patient and public involvement IPD—we carried out a further sensitivity analysis We approached our patient and public involvement that included aggregate data from published panel of 10 people with a history of respiratory tract estimates of studies that did not provide IPD. infections to discuss these results as they emerged. This sensitivity analysis compared the effect of They agreed that the results were reassuring and did immediate antibiotic prescribing with delayed not suggest a meaningful benefit to taking antibiotics. antibiotic prescribing. We observed a stronger They suggested that the way in which delayed effect favouring immediate antibiotics for symptom prescribing is communicated to patients is important. severity (mean difference 0.95, 95% confidence They felt that some patients might not easily assess interval 0.71 to 1.18) when including aggregate and gauge the severity or nature of their symptoms data, particularly the 1991 study by El-Daher, and would need clear guidance to determine whether compared with IPD only analysis (0.09, −0.01 to they needed to take antibiotics. Almost all contributors 0.18; fig 4). emphasised that general practitioners need to Delayed v no antibiotics Study ID Effect Weight Effect (95% CI) (%) (95% CI) Observational studies Francis et al 2012 18.71 -0.11 (-0.27 to 0.05) Hay et al 2016 17.31 -0.09 (-0.28 to 0.10) Little et al 2013 7.64 -0.13 (-0.60 to 0.34) Subgroup: I2=0% 43.66 -0.10 (-0.22 to 0.02) Randomised controlled trials Little et al 2005 15.61 0.10 (-0.13 to 0.33) Little et al 2014 18.77 0.38 (0.22 to 0.53) Mas-Dalmau et al 2021 11.14 -0.01 (-0.35 to 0.34) De la Poza Abad et al 2016 10.83 -0.00 (-0.35 to 0.35) Subgroup: I2=59.7% 56.34 0.16 (-0.05 to 0.36) Heterogeneity between groups: P=0.00 Overall: I2=67.8% 100.00 0.04 (-0.12 to 0.20) -1 0 1 Favours Favours no delayed antibiotics antibiotics Delayed v immediate antibiotics Study ID Effect Weight Effect (95% CI) (%) (95% CI) Observational studies Francis et al 2012 17.89 -0.07 (-0.22 to 0.09) Hay et al 2016 9.41 0.02 (-0.28 to 0.31) Little et al 2013 6.35 -0.34 (-0.73 to 0.05) Subgroup: I2=0% 33.64 -0.08 (-0.21 to 0.05) Randomised controlled trials Little et al 2001 20.20 0.23 (0.11 to 0.36) McCormick et al 2005 13.86 -0.00 (-0.21 to 0.21) Little et al 2005 14.21 0.14 (-0.06 to 0.35 ) Mas-Dalmau et al 2021 8.82 0.22 (-0.09 to 0.54) De la Poza Abad et al 2016 9.26 0.04 (-0.26 to 0.34) Subgroup: I2=22.7% 66.36 0.15 (0.04 to 0.25) Heterogeneity between groups: P=0.003 Overall: I2=50.6% 100.00 0.06 (-0.05 to 0.17) -1 0 1 Favours Favours delayed immediate antibiotics antibiotics Fig 3 | Unadjusted association between treatment group and symptom severity two to four days after consultation for delayed versus no antibiotics, and delayed versus immediate antibiotics. Weights are from random effects model 10 doi: 10.1136/bmj.n808 | BMJ 2021;372:n808 | the bmj RESEARCH BMJ: first published as 10.1136/bmj.n808 on 28 April 2021. Downloaded from http://www.bmj.com/ on 9 March 2023 at Sant Paul Hospital Consortia. Protected by copyright. Table 5 | Effect of antibiotic prescribing strategy subgroup variable interactions on mean symptom severity score two to four days after consultation for delayed versus no antibiotics, and delayed versus immediate antibiotics No of No of Interaction Adjusted* mean Subgroup studies participants (95% CI) P difference (95% CI) Delayed v no antibiotics Previous duration Median and above 6 1835 0.008 (−0.21 to 0.23) Below median 1589 0.05 (−0.23 to 0.34) 0.70 0.03 (−0.22 to 0.28) Age (years) 0-4 9 749 −0.20 (−0.24 to −0.15) 5-15 637 0.12 (0.07 to 0.16) 16-64 2153 −0.03 (−0.14 to 0.08) >65 368 0.11 (−0.02 to 0.24) 0.11 0.07 (−0.36 to 0.51) Fever ≥37.5°C 8 1436 −0.03 (−0.15 to 0.09) <37.5°C 2211 0.01 (−0.16 to 0.18) 0.88 −0.02 (−0.19 to 0.16) Comorbidity Any lung disease 9 438 0.15 (−0.12 to 0.42) No lung disease 2598 0.14 (−0.05 to 0.34) 0.15 −0.01 (−0.15 to 0.13) Baseline severity Median and above 9 1972 −0.09 (−0.31 to 0.13) Below median 1935 −0.05 (−0.28 to 0.19) 0.69 −0.14 (−0.33 to 0.04) Delayed v immediate antibiotics Previous duration Median and above 6 1516 −0.04 (−0.38 to 0.29) −0.07 (−0.41 to 0.27) 0.68 Below median 1526 0.02 (−0.17 to 0.21) Age (years) 0-4 9 729 0.10 (0.03 to 0.18) 5-15 548 0.09 (−0.11 to 0.30) −0.10 (−0.17 to −0.03) 0.005† 16-64 2107 −0.09 (−0.27 to 0.09) >65 366 −0.19 (−0.62 to 0.25) Fever ≥37.5°C 8 1765 −0.01 (−0.42 to 0.40) −0.05 (−0.32 to 0.23) 0.80 <37.5°C 1662 0.03 (−0.06 to 0.11) Comorbidity Any lung disease 9 483 0.13 (−0.27 to 0.53) 0.01 (−0.16 to 0.19) 0.87 No lung disease 2554 0.12 (−0.14 to 0.38) Baseline severity Median and above 9 2286 0.10 (−0.57 to 0.77) 0.35 (−0.23 to 0.93) 0.24 Below median 1466 −0.27 (−0.34 to −0.19) *Adjusted for baseline severity, age, and condition. †Statistically significant interaction term. be better at explaining the self-limiting nature of or it could be that secondary opportunistic bacterial respiratory tract infections and the harmful effects of infections that start later after an initial viral illness are inappropriate use of antibiotics, using examples or more effectively managed by the later start of a delayed pictures where necessary to relate to people and ensure prescription. The second suggestion is supported by a clear, simple, and effective message is delivered to findings from the large GRACE trial; one of the groups patients. Qualitative studies of patients support these that reported beneficial effects for antibiotics were observations of our patient and public involvement people for whom evidence was found of coinfection collaborators.59 with viruses and bacterial pathogens.60 Delayed antibiotics resulted in longer duration of Discussion symptoms than immediate antibiotics, but were as We used individual level data from 13 RCTs and effective for the remaining clinical outcomes. The observational cohort studies (55 682 patients) to literature suggests that delayed prescribing could assess the clinical effectiveness of delayed antibiotic also reduce antibiotic use by patients compared with prescribing in patients with respiratory tract infections immediate antibiotics by 23-75%.48 61 62 Consistent in the community setting. Overall, our findings suggest results were obtained in subgroups often considered delayed antibiotic prescribing is just as effective as to be at higher risk, which suggests that delayed no antibiotics for all clinical outcomes, but increa­ prescribing is unlikely to lead to poorer symptom sed patient satisfaction and reduced reconsultation control than immediate antibiotics. In children and complication rates. The reasons for reduced younger than 5 years and in those with higher symptom reconsultation rates are unclear, but one suggestion scores at baseline, we found statistically significant is that if a prescription is delayed, by the time the differences in the symptom severity scores two to four antibiotic course has finished, symptoms will have had days after consultation. However, the mean differences more time to settle and so reconsultation is less likely; were only 0.11 points higher on a scale from 0 to 6 (the the bmj | BMJ 2021;372:n808 | doi: 10.1136/bmj.n808 11 RESEARCH BMJ: first published as 10.1136/bmj.n808 on 28 April 2021. Downloaded from http://www.bmj.com/ on 9 March 2023 at Sant Paul Hospital Consortia. Protected by copyright. Table 6 | Evidence profiles based on GRADE (grading of recommendations assessment, development and evaluation) assessment Certainty assessment No of participants No of Study Risk of Other No Delayed Adjusted* studies design bias Inconsistency Indirectness Imprecision considerations antibiotics antibiotics estimate (95% CI) Quality Importance Delayed v no antibiotics Symptom severity 4 RCT Serious† Not serious‡ Not serious Not serious§ None 484 815 0.09 (−0.10 to 0.28) Moderate Critical 3 OS Serious¶ Not serious Not serious Not serious** None 2231 377 −0.10 (−0.12 to −0.08) Low Critical 7 RCT+OS Serious Not serious Not serious Not serious§ None 2715 1192 −0.003 (−0.12 to 0.11) Moderate Critical Time to complete symptom resolution (days) 5 RCT Serious† Not serious‡ Not serious Not serious§ None 540 647 1.04 (0.95 to 1.13) Moderate Important 3 OS Serious¶ Not serious Not serious Not serious** None 2551 476 0.98 (0.94 to 1.01) Low Important 8 RCT+OS Serious Not serious Not serious Not serious§ None 3091 1123 1.00 (0.82 to 1.23) Moderate Important Reconsultation (%) 5 RCT Serious† Not serious‡ Not serious Not serious§ None 509 611 0.92 (0.67 to 1.26) Moderate Important 4 OS Serious¶ Not serious Not serious Not serious** None 15 723 5290 0.54 (0.49 to 0.60) Low Important 9 RCT+OS Serious Not serious Not serious Not serious§ None 16 232 5901 0.72 (0.60 to 0.87) Moderate Important Complication (hospital admission or death; %) 6 RCT Serious† Not serious‡ Not serious Not serious§ None 431 530 0.35 (0.07 to 1.92) Moderate Important 4 OS Serious¶ Not serious Not serious Not serious** None 15 933 5297 0.60 (0.28 to 1.43) Low Important 10 RCT+OS Serious Not serious Not serious Not serious§ None 16 364 5827 0.62 (0.30 to 1.27) Moderate Important Patient satisfaction score 5 RCT Serious† Not serious‡ Not serious Not serious§ None 433 520 0.06 (−0.03 to 0.16) Moderate Important 1 OS Serious¶ NA Not serious Not serious** None 1001 154 0.10 (−0.03 to 0.23) Low Important 6 RCT+OS Serious Not serious Not serious Not serious§ None 1434 674 0.09 (0.06 to 0.11) Moderate Important Delayed v immediate antibiotics Symptom severity 5 RCT Serious† Not serious Not serious Not serious None 606 674 0.11 (−0.004 to 0.22) Moderate Critical 3 OS Serious Not serious Not serious Not serious None 2093 377 −0.12 (−0.33 to 0.07) Low Critical 8 RCT+OS Serious Not serious Not serious Not serious None 2699 1053 0.02 (−0.11 to 0.15) Moderate Critical Time to complete symptom resolution (days) 7 RCT Serious Serious‡ Not serious Not serious None 876 962 1.14 (1.06 to 1.22) Low Important 3 OS Serious Not serious Not serious Not serious None 2399 480 1.02 (0.97 to 1.07) Low Important 10 RCT+OS Serious Not serious Not serious Not serious None 3275 1442 1.04 (1.01 to 1.08) Low Important Reconsultation (%) 6 RCT Serious Not serious Not serious Not serious†† None 796 865 1.29 (0.84 to 1.99) Moderate Important 4 OS Serious Serious‡ Not serious Not serious None 21 634 5292 0.70 (0.66 to 0.75) Low Important 10 RCT+OS Serious Not serious Not serious Not serious None 22 430 6157 0.95 (0.74 to 1.22) Moderate Important Complication (hospital admission or death; %) 3 RCT Serious Not serious Not serious Not serious†† None 650 719 1.25 (0.38 to 4.16) Moderate Important 4 OS Serious Not serious Not serious Not serious None 21 721 5298 0.22 (0.19 to 0.27) Low Important 7 RCT+OS Serious Not serious Not serious Not serious None 22 371 6017 0.78 (0.53 to 1.13) Moderate Important Patient satisfaction score 5 RCT Serious Serious‡ Not serious Not serious None 563 649 −0.13 (−0.31 to 0.05) Low Important 1 OS Serious NA Not serious Not serious None 1180 156 −0.06 (−0.18 to 0.05) Low Important 6 RCT+OS Serious Serious Not serious Not serious None 1743 805 −0.12 (−0.26 to 0.03) Low Important Adjusted estimate=adjusted coefficient, odds ratio, or relative risk; NA=not applicable; OS=observational studies; RCT=randomised controlled trials. *Adjusted for baseline severity, age, and condition. †Most RCTs here were not blinded. However, results were not considered biased because similar evidence obtained for blinded studies and observational studies. ‡Statistical but not important heterogeneity. §Confidence intervals exclude important benefits and harms. ¶Balance achieved for key covariates but residual confounding is still possible. **Large enough sample size and the 95% confidence interval excludes no effect. ††Wide confidence intervals but not downgraded because overall same conclusion. equivalent of 1 in 10 participants rating symptoms include results obtained from RCTs and observational one point different; for example, as moderately bad studies. Selection bias associated with trials can limit rather than a slight problem). This finding suggests perceived external validity, therefore a strength of this that while the effect might be statistically significant, study was the ability to include observational data. these differences are not clinically significant, and our Therefore, the external validity was improved and patient and public involvement panel did not feel that the impact of delayed antibiotic prescribing could be they were likely to be meaningful to patients. assessed in a clinical trial and a real world setting.63 The studies included in the IPD comprised 93% of the Strengths and limitations population from all eligible studies. The observational This large study examined the clinical effectiveness of studies that did not provide data tended to be smaller the delayed antibiotic prescribing strategy. Strengths studies. The trials for which IPD were not available but include the ability to control for baseline severity, that were included in a sensitivity analysis were older to assess the quality of the studies based on the full studies (dating from 1991 or earlier). Therefore, the dataset, to explore heterogeneity across studies, and to difference between the results of the primary analysis 12 doi: 10.1136/bmj.n808 | BMJ 2021;372:n808 | the bmj RESEARCH BMJ: first published as 10.1136/bmj.n808 on 28 April 2021. Downloaded from http://www.bmj.com/ on 9 March 2023 at Sant Paul Hospital Consortia. Protected by copyright. and the sensitivity analysis might be because the trials and returned diaries might not be representative that did provide IPD are more likely to be relevant to of all study participants, which could also impact modern patient populations. This difference might generalisability. However, previously published esti­ also be partly due to eligible trials that did not mates from included studies suggest that those who contribute IPD being based on younger populations, completed diaries had broadly similar characteristics as highlighted in our subgroup analyses which showed to all recruited participants.65 66 Power was also low that children younger than 5 years might benefit more for the comparisons involving complications because from immediate antibiotics; however, this is unlikely this outcome is extremely rare, even in a dataset as because the size of the interaction was statistically large as the one we compiled. However, this extensive significant but not clinically important. The studies dataset enabled us to include large numbers of that were included with aggregate data only were also participants when analysing outcomes—even the at high risk of selection, attrition, and other biases. In smallest comparison contained 2108 participants— particular, the study by El-Daher favoured immediate and the rarity of severe complications should be antibiotics over delayed antibiotics. The Cochrane reassuring. review on this topic suggests that the El-Daher study Delayed prescribing is one of several strategies that was one of the less methodologically sound of the might help to safely reduce inappropriate antibiotic included studies. However, the El-Daher study is also prescribing and consumption. Other strategies, such as the only one undertaken in a lower income setting9 point of care diagnostic testing, patient decision aids, and it is not clear whether the results of the IPD would and specific training for health professionals, might generalise to that population. The illness spectrum also be helpful alone or in combination with delayed in a lower income setting might be different, and the prescribing.67 68 However, none of the studies included previous probability of more serious infection could in our IPD evaluated these strategies, which means we be higher as could the risk of complications. Different can only draw conclusions about delayed prescribing organisms might be more prevalent and underlying when used in isolation rather than in combination comorbid conditions (such as tuberculosis) could lead with other approaches that might be deployed in a to a different outcome. Delayed access to reassessment primary care setting. or secondary care in the event of deterioration might Looking across all the outcomes we included, we also be an important factor.64 Further research is found a tendency for the treatment effect estimates from needed in low to middle income countries to determine observational studies to be in the opposite direction whether delayed antibiotic prescribing would be a safe from those of RCTs. This finding could be because of and effective strategy in such settings. residual confounding (eg, use of other, or known or A further limitation relates to the statistical unmeasured covariates such as patient presence and power. Not all outcomes were collected in all compliance) in observational studies, differences in studies. Symptom severity data were not collected how delayed prescribing is implemented in real life for all studies, or were only collected for a subset of versus RCTs, and varying time periods. However, the participants in some studies, resulting in a smaller overall heterogeneity estimates for the combined RCT sample size for the outcome analysis. This outcome and observational study analyses were not high or was based on diary data and those who completed could be explained by individual studies at higher risk Study ID Effect Weight Effect (95% CI) (%) (95% CI) IPD Little et al 2001 37.33 0.09 (-0.06 to 0.25) McCormick et al 2005 22.35 0.02 (-0.18 to 0.23) Little et al 2005 22.94 0.12 (-0.08 to 0.32) Mas-Dalmau et al 2021 8.20 0.18 (-0.16 to 0.52) De la Poza Abad et al 2016 9.18 0.04 (-0.28 to 0.37) Subgroup: I2=22.7% 100.00 0.09 (-0.01 to 0.18) Aggregate data Pichichero 1987 64.00 0.30 (0.01 to 0.59) El-Daher 1991 36.00 2.10 (1.71 to 2.49) Subgroup: I2=98.1% 100.00 0.95 (0.71 to 1.18) -3 -2 -1 0 1 2 3 Favours Favours delayed immediate antibiotics antibiotics Fig 4 | Difference in mean symptom severity two to four days after consultation; aggregate meta-analysis including studies that did not provide individual patient data (IPD) the bmj | BMJ 2021;372:n808 | doi: 10.1136/bmj.n808 13 RESEARCH BMJ: first published as 10.1136/bmj.n808 on 28 April 2021. Downloaded from http://www.bmj.com/ on 9 March 2023 at Sant Paul Hospital Consortia. Protected by copyright. of bias. We recognise that our pooled effect estimates concept and design of the study, the interpretation of the results, and manuscript editing. All authors read, provided feedback, and were influenced by observational studies because these approved the final manuscript. The corresponding author attests that contributed large numbers of individual participants all listed authors meet authorship criteria and that no others meeting to the overall pooled dataset. The magnitude of the the criteria have been omitted. pooled treatment effect needs to be interpreted with Funding: This work was funded by the NIHR Research for Patient Benefit (RfPB) Programme (grant No PB-PG-0416-20005). This caution because, while propensity scores were used to funding supported the collation of the individual participant data, control for measured confounding, there might still be data management and analyses. The NIHR RfPB was not involved in residual confounding from unmeasured confounders. any other aspect of the project, such as the design of the project’s protocol and analysis plan, the collection and analyses. The funder had no input on the interpretation or publication of the study results. Conclusions and implications Competing interests: All authors have completed the ICMJE Delayed prescribing appears to be a safe and effective uniform disclosure form at www.icmje.org/coi_disclosure.pdf and antibiotic strategy for most patients, including declare: support from the NIHR Research for Patient Benefit (RfPB) Programme for the submitted work; no financial relationships with any those in higher risk subgroups. Compared with organisations that might have an interest in the submitted work in the a no prescription approach, delayed prescribing previous 3 years and no other relationships or activities that could probably reduces reconsultation rates, and therefore appear to have influenced the submitted work. the workload of general practitioners, with slightly Ethical approval: Ethical approval was obtained by the Faculty of Medicine Ethics Committee at the University of Southampton higher levels of patient satisfaction. Compared with (approval No 30068). immediate antibiotics, delayed prescribing does Data sharing: The IPD database can be shared subject to the not result in higher complication rates (if anything, approval of the study collaborators. they are lower) and it does not significantly decrease The lead author affirms that the manuscript is an honest, accurate, patient satisfaction. Delayed prescribing could be used and transparent account of the study being reported; that no important aspects of the study have been omitted; and that any as a standalone interventional approach, but it might discrepancies from the study as planned have been explained. also be a way of resolving mismatched expectations Dissemination to participants and related patient and public between clinician and patient. communities: The findings of the study were disseminated to all Author affiliations members of the patient and public involvement panel at the time 1 Academic Unit of Primary Care, Population Sciences and Medical of our discussion with them as reported in this paper. No members Education, Faculty of Medicine, University of Southampton, asked for further disseminated information. Southampton, UK Provenance and peer review: Not commissioned; externally peer 2 Biostatistics Research Group, Department of Health Sciences, reviewed. College of Life Sciences, University of Leicester, Leicester, UK This is an Open Access article distributed in accordance with the 3 Iberoamerican Cochrane Centre, Instituto de Investigación terms of the Creative Commons Attribution (CC BY 4.0) license, which Biomédica Sant Pau (IIB Sant Pau-CIBERESP), Barcelona, Spain permits others to distribute, remix, adapt and build upon this work, 4 for commercial use, provided the original work is properly cited. See: Institute of General Practice, Rostock University Medical Center, http://creativecommons.org/licenses/by/4.0/. Rostock, Germany 5 Department of General Practice and Primary Health Care, University 1 Ridge KW, Hand K, Sharland M, et al. Antimicrobial resistance. In: of Auckland, Auckland, New Zealand Chief Medical Officer annual report 2011: antimicrobial resistance. 6 2011. http://www.dh.gov.uk/health/2013/03/cmo-vol2/. Southampton Statistical Sciences Research Institute, University of 2 Public Health England. English surveillance programme for Southampton, Southampton, UK antimicrobial utilisation and resistance report. https://assets. 7 Division of Health and Social Care Research, King’s College publishing.service.gov.uk/government/uploads/system/uploads/ London, London, UK attachment_data/file/843129/English_Surveillance_Programme_ 8 Basel Institute for Clinical Epidemiology and Biostatistics (CEB), for_Antimicrobial_Utilisation_and_Resistance_2019.pdf (accessed 7 University Hospital Basel and University of Basel, Switzerland Jul 2020). 9 3 Cassini A, Högberg LD, Plachouras D, et al, Burden of AMR Pediatric Emergency Medicine, State University of New York Collaborative Group. Attributable deaths and disability-adjusted Downstate, Brooklyn, New York, USA life-years caused by infections with antibiotic-resistant bacteria in 10 the EU and the European Economic Area in 2015: a population-level Institut Català de la Salut, CAP Doctor Carles Ribas, Foc 112, Barcelona, Spain modelling analysis. Lancet Infect Dis 2019;19:56-66. doi:10.1016/ 11 S1473-3099(18)30605-4 Centre for Trials Research, School of Medicine, College of 4 Organisation for Economic Cooperation and Development. Stemming Biomedical & Life Sciences, Cardiff University, Cardiff, UK the Superbug Tide. OECD, 2018. 12 Centre for Academic Primary Care, Population Health Sciences, 5 Costelloe C, Metcalfe C, Lovering A, Mant D, Hay AD. Effect of Bristol Medical School, University of Bristol, Bristol, UK antibiotic prescribing in primary care on antimicrobial resistance 13 Department of Pediatrics, University of Texas Medical Branch at in individual patients: systematic review and meta-analysis. BMJ 2010;340:c2096. doi:10.1136/bmj.c2096 Galveston, Galveston, TX, USA 14 6 Little P, Stuart B, Hobbs FDR, et al, DESCARTE investigators. Instituto de Investigación Biomédica Sant Pau (IIB Sant Pau), Antibiotic prescription strategies for acute sore throat: a prospective Barcelona, Spain observational cohort study. Lancet Infect Dis 2014;14:213-9. 15 Agència de Qualitat i Avaluació Sanitàries de Catalunya (AQuAS), doi:10.1016/S1473-3099(13)70294-9 Barcelona, Spain 7 Spinks A, Glasziou PP, Del Mar CB. Antibiotics for sore throat. 16 Cochrane Database Syst Rev 2013;2013:CD000023. ASPIRE PPI Panel, Leeds Institute for Health Sciences, University of doi:10.1002/14651858.CD000023.pub4 Leeds, Leeds, UK 8 Kenealy T, Arroll B. Antibiotics for the common cold We thank the following collaborators who allowed us to use the data and acute purulent rhinitis. Cochrane Database Syst from their studies: Professor Ngaire Kerse, Ms Jacqueline Nuttall. Rev 2013;2013:CD000247. doi:10.1002/14651858.CD000247. pub3 Contributors: BS and HH contributed equally to this paper. BS is 9 Spurling GKP, Del Mar CB, Dooley L, Foxlee R, Farley R. Delayed the guarantor. BS conceived the original study concept and design. antibiotic prescriptions for respiratory infections. Cochrane Database BS, PL, MM, SZ, GY, DB, JB, KS, and HCB wrote the grant application Syst Rev 2017;9:CD004417. doi:10.1002/14651858.CD004417. and obtained funding from the National Institute for Health Research pub5 (NIHR) Research for Patient Benefit (RfPB) Programme. TB, HH, BS, 10 The path of least resistance: main report/Standing Medical Advisory DB, and HCB contributed to the data analysis. BS, HH, and TB wrote Committee Sub-Group on Antimicrobial Resistance. Department of the first draft of the manuscript. All listed authors contributed to the Health. 1998. 14 doi: 10.1136/bmj.n808 | BMJ 2021;372:n808 | the bmj RESEARCH BMJ: first published as 10.1136/bmj.n808 on 28 April 2021. Downloaded from http://www.bmj.com/ on 9 March 2023 at Sant Paul Hospital Consortia. Protected by copyright. 11 National Institute for Health and Care Excellence. Respiratory tract 36 Little P, Stuart B, Hobbs FD, et al, DESCARTE investigators. Predictors infections (self-limiting): prescribing antibiotics Clinical guideline. of suppurative complications for acute sore throat in primary 2008. www.nice.org.uk/guidance/cg69 (accessed 7 Jul 2020). care: prospective clinical cohort study. BMJ 2013;347:f6867. 12 Gulliford MC, Dregan A, Moore MV, et al. Continued high rates of doi:10.1136/bmj.f6867 antibiotic prescribing to adults with respiratory tract infection: 37 Moore M, Stuart B, Hobbs FR, et al, DESCARTE investigators. survey of 568 UK general practices. BMJ Open 2014;4:e006245. Symptom response to antibiotic prescribing strategies in acute doi:10.1136/bmjopen-2014-006245 sore throat in adults: the DESCARTE prospective cohort study in UK 13 Pouwels KB, Dolk FCK, Smith DRM, Robotham JV, Smieszek T. Actual general practice. Br J Gen Pract 2017;67:e634-42. doi:10.3399/ versus ‘ideal’ antibiotic prescribing for common conditions in English bjgp17X692321 primary care. J Antimicrob Chemother 2018;73(suppl_2):19-26. 38 Hay AD, Redmond NM, Turnbull S, et al. Development and internal doi:10.1093/jac/dkx502 validation of a clinical rule to improve antibiotic use in children 14 Smieszek T, Pouwels KB, Dolk FCK, et al. Potential for reducing presenting to primary care with acute respiratory tract infection and inappropriate antibiotic prescribing in English primary care. J cough: a prognostic cohort study. Lancet Respir Med 2016;4:902- Antimicrob Chemother 2018;73(suppl_2):ii36-43. doi:10.1093/jac/ 10. doi:10.1016/S2213-2600(16)30223-5 dkx500. 39 Little P, Stuart B, Smith S, et al. Antibiotic prescription strategies 15 Fiore DC, Fettic LP, Wright SD, Ferrara BR. Antibiotic overprescribing: and adverse outcome for uncomplicated lower respiratory tract still a major concern. J Fam Pract 2017;66:730-6. infections: prospective cough complication cohort (3C) study. 16 van der Velden A, Duerden M, Bell J, et al. Prescriber and BMJ 2017;357:j2148. doi:10.1136/bmj.j2148 patient responsibilities in treatment of acute respiratory tract 40 Mas-Dalmau G, Villanueva López C, Gorrotxategi Gorrotxategi P, infections—essential for conservation of antibiotics. Antibiotics et al, DAP PEDIATRICS GROUP*. Delayed antibiotic prescription (Basel) 2013;2:316-27. doi:10.3390/antibiotics2020316 for children with respiratory infections: A Randomized Trial. 17 Lambert PC, Sutton AJ, Abrams KR, Jones DR. A comparison of Pediatrics 2021;147:e20201323. summary patient-level covariates in meta-regression with individual 41 Butler CC, Hood K, Verheij T, et al. Variation in antibiotic prescribing patient data meta-analysis. J Clin Epidemiol 2002;55:86-94. and its impact on recovery in patients with acute cough in primary doi:10.1016/S0895-4356(01)00414-0 care: prospective study in 13 countries. BMJ 2009;338:b2242. 18 Riley RD, Lambert PC, Abo-Zaid G. Meta-analysis of individual doi:10.1136/bmj.b2242 participant data: rationale, conduct, and reporting. 42 Little P, Gould C, Williamson I, Warner G, Gantley M, Kinmonth AL. BMJ 2010;340:c221. doi:10.1136/bmj.c221 Reattendance and complications in a randomised trial of prescribing 19 Stuart B, Hounkpatin H, Becque T, et al. Delayed antibiotic prescribing strategies for sore throat: the medicalising effect of prescribing for respiratory tract infections: protocol of an individual patient antibiotics. BMJ 1997;315:350-2. doi:10.1136/bmj.315.7104.350 data meta-analysis. BMJ Open 2019;9:e026925. doi:10.1136/ 43 Little P, Gould C, Williamson I, Moore M, Warner G, Dunleavey J. bmjopen-2018-026925 Pragmatic randomised controlled trial of two prescribing strategies 20 Stewart LA, Clarke M, Rovers M, et al. Preferred Reporting Items for a for childhood acute otitis media. BMJ 2001;322:336-42. Systematic Review and Meta-analysis of Individual Participant Data: 44 Arroll B, Kenealy T, Kerse N. Do delayed prescriptions reduce the use The PRISMA-IPD Statement. JAMA 2015;313;1657-65. doi:10.1001/ of antibiotics for the common cold? A single-blind controlled trial. J jama.2015.3656 Fam Pract 2002;51:324-8. 21 StataCorp. Stata Statistical Software: Release 15. 2017. 45 McCormick DP, Chonmaitree T, Pittman C, et al. Nonsevere acute otitis 22 Higgins JPT, Altman DG, Gøtzsche PC, et al, Cochrane Bias Methods media: a clinical trial comparing outcomes of watchful waiting versus Group, Cochrane Statistical Methods Group. The Cochrane immediate antibiotic treatment. Pediatrics 2005;115:1455-65. Collaboration’s tool for assessing risk of bias in randomised trials. doi:10.1542/peds.2004-1665 BMJ 2011;343:d5928. doi:10.1136/bmj.d5928 46 Little P, Rumsby K, Kelly J, et al. Information leaflet and antibiotic 23 Sterne JA, Hernán MA, Reeves BC, et al. ROBINS-I: a tool for prescribing strategies for acute lower respiratory tract infection: a assessing risk of bias in non-randomised studies of interventions. randomized controlled trial. JAMA 2005;293:3029-35. doi:10.1001/ BMJ 2016;355:i4919. doi:10.1136/bmj.i4919 jama.293.24.3029 24 Watson L, Little P, Moore M, Warner G, Williamson I. Validation study 47 Chao JH, Kunkov S, Reyes LB, Lichten S, Crain EF. Comparison of of a diary for use in acute lower respiratory tract infection. Fam two approaches to observation therapy for acute otitis media Pract 2001;18:553-4. doi:10.1093/fampra/18.5.553 in the emergency department. Pediatrics 2008;121:e1352-6. 25 Schünemann HJ, Vist GE, Higgins JP, et al. Interpreting doi:10.1542/peds.2007-2278 results and drawing conclusions. In: Cochrane Handbook for 48 Little P, Moore M, Kelly J, et al, PIPS Investigators. Delayed Systematic Reviews of Interventions. Wiley, 2019: 403-31, antibiotic prescribing strategies for respiratory tract infections in doi:10.1002/9781119536604.ch15 primary care: pragmatic, factorial, randomised controlled trial. 26 Stewart GB, Altman DG, Askie LM, Duley L, Simmonds MC, Stewart BMJ 2014;348:g1606. doi:10.1136/bmj.g1606 LA. Statistical analysis of individual participant data meta-analyses: 49 de la Poza Abad M, Mas Dalmau G, Moreno Bakedano M, et al, a comparison of methods and recommendations for practice. PLoS Delayed Antibiotic Prescription (DAP) Group. Prescription strategies One 2012;7:e46042. doi:10.1371/journal.pone.0046042 in acute uncomplicated respiratory infections: a randomized 27 Akl EA, Kahale LA, Agoritsas T, et al. Handling trial participants clinical trial. JAMA Intern Med 2016;176:21-9. doi:10.1001/ with missing outcome data when conducting a meta-analysis: a jamainternmed.2015.7088 systematic survey of proposed approaches. Syst Rev 2015;4:98. 50 Pichichero ME, Disney FA, Talpey WB, et al. Adverse and beneficial doi:10.1186/s13643-015-0083-6 effects of immediate treatment of Group A beta-hemolytic 28 US Government Accountability Office. Cross-design synthesis: a new streptococcal pharyngitis with penicillin. Pediatr Infect Dis strategy for medical effectiveness research. 1992. https://www.gao. J 1987;6:635-43. doi:10.1097/00006454-198707000-00004 gov/products/PEMD-92-18 (accessed 7 Jul 2020). 51 Gerber MA, Randolph MF, DeMeo KK, Kaplan EL. Lack of impact of 29 Rosenbaum PR, Rubin DB. The central role of the propensity score in early antibiotic therapy for streptococcal pharyngitis on recurrence observational studies for causal effects. Biometrika 1983;60:41-55. rates. J Pediatr 1990;117:853-8. 30 D’Agostino RBJr. Propensity score methods for bias reduction 52 el-Daher NT, Hijazi SS, Rawashdeh NM, al-Khalil IA, Abu-Ektaish FM, in the comparison of a treatment to a non-randomized control Abdel-Latif DI. Immediate vs. delayed treatment of group A beta- group. Stat Med 1998;17:2265-81. doi:10.1002/(SICI)1097- hemolytic streptococcal pharyngitis with penicillin V. Pediatr Infect Dis 0258(19981015)17:19<2265::AID-SIM918>3.0.CO;2-B J 1991;10:126-30. doi:10.1097/00006454-199102000-00010 31 Williamson EJ, Forbes A, White IR. Variance reduction in randomised 53 Dowell J, Pitkethly M, Bain J, Martin S. A randomised controlled trials by inverse probability weighting using the propensity score. Stat trial of delayed antibiotic prescribing as a strategy for managing Med 2014;33:721-37. doi:10.1002/sim.5991 uncomplicated respiratory tract infection in primary care. Br J Gen 32 Riley RD, Simmonds MC, Look MP. Evidence synthesis combining Pract 2001;51:200-5. individual patient data and aggregate data: a systematic 54 Spiro DM, Tay KY, Arnold DH, Dziura JD, Baker MD, Shapiro ED. review identified current practice and possible methods. J Clin Wait-and-see prescription for the treatment of acute otitis media: a Epidemiol 2007;60:431-9. doi:10.1016/j.jclinepi.2006.09.009 randomized controlled trial. JAMA 2006;296:1235-41. doi:10.1001/ 33 Mustafa RA, Santesso N, Brozek J, et al. The GRADE approach is jama.296.10.1235 reproducible in assessing the quality of evidence of quantitative 55 Siegel RM, Kiely M, Bien JP, et al. Treatment of otitis media evidence syntheses. J Clin Epidemiol 2013;66:736-42, quiz 742. with observation and a safety-net antibiotic prescription. e1-5. doi:10.1016/j.jclinepi.2013.02.004 Pediatrics 2003;112:527-31. doi:10.1542/peds.112.3.527 34 Little P, Williamson I, Warner G, et al. Open randomised trial of 56 Marchetti F, Ronfani L, Nibali SC, Tamburlini G, Italian Study Group prescribing strategies in managing sore throat. BMJ1997;314:722-7. on Acute Otitis Media. Delayed prescription may reduce the use 35 Francis NA, Gillespie D, Nuttall J, et al. Delayed antibiotic prescribing of antibiotics for acute otitis media: a prospective observational and associated antibiotic consumption in adults with acute cough. Br study in primary care. Arch Pediatr Adolesc Med 2005;159:679-84. J Gen Pract 2012;62:e639-46. doi:10.3399/bjgp12X654614 doi:10.1001/archpedi.159.7.679 the bmj | BMJ 2021;372:n808 | doi: 10.1136/bmj.n808 15 RESEARCH BMJ: first published as 10.1136/bmj.n808 on 28 April 2021. Downloaded from http://www.bmj.com/ on 9 March 2023 at Sant Paul Hospital Consortia. Protected by copyright. 57 Fischer T, Singer AJ, Chale S. Observation option for acute otitis media 63 Faraoni D, Schaefer ST. Randomized controlled trials vs. observational in the emergency department. Pediatr Emerg Care 2009;25:575- studies: why not just live together?BMC Anesthesiol 2016;16:102. 8. https://journals.lww.com/pec-online/Fulltext/2009/09000/ doi:10.1186/s12871-016-0265-3 Observation_Option_for_Acute_Otitis_Media_in_the.8.aspx. 64 Jehan F, Nisar I, Kerai S, et al. Randomized trial of amoxicillin doi:10.1097/PEC.0b013e3181b91ff0 for pneumonia in Pakistan. N Engl J Med 2020;383:24-34. 58 Kavanagh KE, O’Shea E, Halloran R, Cantillon P, Murphy AW. A pilot study doi:10.1056/NEJMoa1911998 of the use of near-patient C-reactive protein testing in the treatment of 65 Moore M, Stuart B, Hobbs FR, et al, DESCARTE investigators. adult respiratory tract infections in one Irish general practice. BMC Fam Symptom response to antibiotic prescribing strategies in acute Pract 2011;12:93. doi:10.1186/1471-2296-12-93 sore throat in adults: the DESCARTE prospective cohort study in UK 59 McDermott L, Leydon GM, Halls A, et al, PIPS investigators. general practice. Br J Gen Pract 2017;67:e634-42. doi:10.3399/ Qualitative interview study of antibiotics and self-management bjgp17X692321 strategies for respiratory infections in primary care. BMJ 66 Wensaas KA, Heron J, Redmond N, et al. Post-consultation illness Open 2017;7:e016903. doi:10.1136/bmjopen-2017-016903 trajectories in children with acute cough and respiratory tract 60 Bruyndonckx R, Stuart B, Little P, et al, GRACE project group. infection: prospective cohort study. Fam Pract 2018;35:676-83. Amoxicillin for acute lower respiratory tract infection in primary care: doi:10.1093/fampra/cmy021 subgroup analysis by bacterial and viral aetiology. Clin Microbiol 67 Moore M. Antibiotics: time to act. Br J Gen Pract 2013;63:340-1. Infect 2018;24:871-6. doi:10.1016/j.cmi.2017.10.032 doi:10.3399/bjgp13X668447 61 Hungenberg M, Hansen-Guzman A, Jankousky KC, et al. Does a 68 Cooke J, Butler C, Hopstaken R, et al. Narrative review of primary care delayed antibiotic prescribing strategy for respiratory infections result point-of-care testing (POCT) and antibacterial use in respiratory tract in fewer prescriptions compared with an immediate prescribing infection (RTI). BMJ Open Respir Res 2015;2:e000086. doi:10.1136/ or no-prescribing strategy?Evidence-Based Pract. 2019;22:1-2. bmjresp-2015-000086 doi:10.1097/EBP.0000000000000508 62 Høye S, Frich JC, Lindbæk M. Use and feasibility of delayed prescribing for respiratory tract infections: a questionnaire survey. BMC Fam Pract 2011;12:34. doi:10.1186/1471-2296-12-34 Supplementary information: appendix figures No commercial reuse: See rights and reprints http://www.bmj.com/permissions Subscribe: http://www.bmj.com/subscribe