Effect of antibiotic prescribing in primary care on
antimicrobial resistance in individual patients: systematic review and meta-analysis Ceire Costelloe, research associate, 1 Chris Metcalfe, senior lecturer in medical statistics, 2 Andrew Lovering, consultant clinical scientist, 3 David Mant, professor of general practice, 4 Alastair D Hay, consultant senior lecturer in primary health care 1 ABSTRACT Objective To systematically review the literature and, where appropriate, meta-analyse studies investigating subsequent antibiotic resistance in individuals prescribed antibiotics in primary care. Design Systematic review with meta-analysis. Data sources Observational and experimental studies identified through Medline, Embase, and Cochrane searches. Review methods Electronic searches using MeSH terms and text words identified 4373 papers. Two independent reviewers assessed quality of eligible studies and extracteddata. Meta-analyses were conductedfor studies presenting similar outcomes. Results The review included 24 studies; 22 involved patients with symptomatic infection and two involved healthy volunteers; 19 were observational studies (of which two were prospective) and five were randomised trials. In five studies of urinary tract bacteria (14348 participants), the pooled odds ratio (OR) for resistance was 2.5 (95% confidence interval 2.1 to 2.9) within 2 months of antibiotic treatment and 1.33 (1.2 to 1.5) within 12 months. In seven studies of respiratory tract bacteria (2605 participants), pooled ORs were 2.4 (1.4 to 3.9) and 2.4 (1.3 to 4.5) for the same periods, respectively. Studies reporting the quantity of antibiotic prescribed found that longer duration and multiple courses were associated with higher rates of resistance. Studies comparing the potential for different antibiotics to induce resistance showed no consistent effects. Only one prospective study reported changes in resistance over a long period; pooled ORs fell from12.2 (6.8 to 22.1) at 1week to6.1(2.8to13.4) at 1month, 3.6(2.2to6.0) at 2 months, and 2.2 (1.3 to 3.6) at 6 months. Conclusions Individuals prescribed an antibiotic in primary care for a respiratory or urinary infection develop bacterial resistance to that antibiotic. The effect is greatest in the month immediately after treatment but may persist for up to 12 months. This effect not only increases the population carriage of organisms resistant to first line antibiotics, but also creates the conditions for increased use of second line antibiotics in the community. INTRODUCTION One of the most pressing problems facedby healthcare services is the increasing prevalence of antimicrobial resistance. Compounded by a diminishing number of newagents entering clinical practice, such resistance is widely recognised as a major threat to public health. 1 In general practice, there are concerns that some com- mon infections are becoming increasingly difficult to treat and that illnesses due to antibiotic resistant bac- teria may take longer to resolve. 2 Some antimicrobial resistance mayresult fromindis- criminate or poor use of antibiotics. In response, initia- tives at the local, national, and international levels, are trying to promote antibiotic stewardship, with the goal of improving the appropriateness of antimicrobial use. However, such initiatives rely for success on the continuing education of prescribers and patients, which needs to be supported by high quality evidence linking antimicrobial use to the emergence of resis- tance. Although many countries have been successful in reducing primary care prescribing of antimicrobials, primary care is still responsible for the majority of anti- biotics prescribedtopeople. 3 4 Much of this use is inthe treatment of suspected respiratory infection and levels of prescribing vary widely within 5 and between countries, 6 suggesting that further reductions are pos- sible. However, there are many barriers to reducing the inappropriate use of antimicrobials, including: patient and practitioner expectations, 7 lack of patient awareness of the problems caused by antimicrobial resistance, 8 and a perception in primary care clinicians and patients that antibiotic resistance is only a theoretical 9 or minimal 10 risk. Although the reason for such views being held is unclear, it may in part be because some previous studies have only investigated the relation between prescribing and resistance with population level data. 11-13 Consequently for clinicians, whose primary concern is the unwell individual, the 1 Academic Unit of Primary Health Care, NIHR National School for Primary Care Research, Department of Community Based Medicine, University of Bristol, Bristol BS8 2AA 2 Department of Social Medicine, Canynge Hall, University of Bristol, Bristol BS8 2PS 3 Department of Microbiology, Bristol Centre for Antimicrobial Research and Enterprise, Southmead Hospital, Bristol BS9 3HU 4 Department of Primary Health Care, NIHR National School for Primary Care Research, Oxford University, Headington, Oxford OX3 7LF Correspondence to: A Hay [email protected] Cite this as: BMJ 2010;340:c2096 doi:10.1136/bmj.c2096 BMJ | ONLINE FIRST | bmj.com page 1 of 11 impact of antimicrobial use on the prevalence of socie- tal resistance may not be an important consideration. 14 To reduce prescribing, it may therefore be important to highlight the effect of antimicrobial use on emergent resistance for individuals. 15 To date, a limited number of good quality studies have reported on the relation between prescribing and prevalence of resistance for individuals treated in primary care, 16 and to the best of our knowledge no systematic reviews have been published in this area. 17-19 We have therefore undertaken a systematic review and meta-analysis of studies where the effect of antimicrobial use on the emergence of resistance has been assessed for individual patients in primary care. We were particularly interested in quantifying the strength and duration of any association as well as identifying which antibiotics were most and least likely to cause resistance. METHODS Search strategy The search strategy was designed to identify observa- tional and experimental studies: conducted in any country; investigating relations between primary care prescribed antibiotics and antimicrobial resistance in bacteria sampled from any body site; analysed at the level of the individual; and published in any language. We searched the MEDLINE (1955 to May 2009), EMBASE (1980 to May 2009), and Cochrane data- bases using the OVID interrogation software. We also searched for grey literature and unpublished work using the ISI web of knowledge, which identifies journal articles, patents, websites, conference proceed- ings, and open access material. MeSH terms used included ambulatory care, drug resistance, anti- microbial resistance, and bacterial resistance. We combined these terms with selected text word searches that included: primary care, ambulatory care, family practice, and antibiotics (see box for full search strategy). Additionally, we screened the refer- ence lists of selected papers and wrote to authors who appeared more than once in our search asking for details of other published and unpublished studies. We performed citation searches of all full text papers. Study selection Studies were eligible for inclusion if they investigated and reported quantitative relationships between pri- mary care prescribed antibiotics and subsequent anti- microbial resistance at the level of the individual. Studies were excluded if they were not original research, did not measure antibiotics prescribed in pri- mary care, or were ecological studies. Two independent reviewers (CC, ADH) screened the title and abstract of papers identified by the electro- nic searches, completing an inclusion/exclusion form for all papers. We retrieved full copies of included papers, each of which was independently reviewed for eligibility by two authors (CC and either DM, CM, ADH, or AL). Disagreements were resolved by discussion with a third author. Data extraction and quality assessment Full articles were independently reviewed for quality and data extracted using a purpose-designed form by two reviewers (CC and one other). Disagreements were resolved by discussion with a third author (ADH). Where data extraction was difficult or unclear, the papers author was contacted for clarification. The explanatory variables extracted included: study design; description of participants; recruitment loca- tion; prescribed antibiotic types; dose and number of courses; and time between antibiotic exposure and measurement of resistance. Time was measured as the exact time at which individuals took antibiotics, or time period during which antibiotic prescribing was recorded, before measurement of resistance. The outcome data extracted were: bacteria type; sampling location; the antibiotics to which resistance was measured; and the method of measuring resis- tance. We agreed that the most important quality criteria for studies quantifying the relation between prescrib- ing and resistance were (1) a reliable measure of anti- biotic exposure; (2) a reliable measure of resistance; (3) unbiased control selection; (4) ability to identify inci- dent cases (that is, patients bacteria were known to be non-resistant before exposure toantibiotics); (5) adjust- ment for key confounders such as recent hospitalisa- tion or instrumentation of the urinary tract. Each study was rated according to these criteria and any stu- dies that met fewer than three criteria present were not included in the meta-analyses. Data synthesis and analysis All analyses were done using STATA version 10. The outcome measure was the odds ratio (OR) of resistance in participants exposed to antibiotics compared with those who were unexposed. The ORs and associated Papers identified by database search (n=4373) Papers exported into Refman2 (n=1134) Excluded on the basis of title (n=3239) Excluded on the basis of title and abstract (n=728): No primary care prescribing data (n=514) Not original research (n=146) Not individual level data (n=68) Papers exported into Access database (n=926) Papers to be read in full (n=198) Duplicates removed (n=208) Excluded on full reading (n=174): No primary care prescribing data (n=143) Not original research (n=23) Insufficient data to determine resistance risk (n=8) Papers included in systematic review (n=24) Fig 1 | Stages of inclusion/exclusion and evaluation of studies in the systematic review RESEARCH page 2 of 11 BMJ | ONLINE FIRST | bmj.com 95% confidence intervals (CI) were tabulated by bac- terium type and sampling location (for example, Escherichia coli from the urinary tract) and by time since exposure to antibiotic. Heterogeneity was mea- sured using the I 2 statistic and the null hypothesis of no heterogeneity was testedusing the Qstatistic generated from the 2 test. For analyses in which we found evi- dence of heterogeneity, a random effects model was used to estimate the pooled OR. Where meta-analyses involved non-randomised studies the unadjusted OR was calculated or used, since unadjusted ORs were available for many more studies than the adjusted OR. We carried out a meta-analysis of the adjusted results, although we only had sufficient data to do this for the studies examining resistance in urinary tract bacteria at the 0-3 month period. Meta-regression was used to investigate differences in the OR between exposures and resistance across different time periods. The meta-regression analyses could each include sev- eral estimates based on overlapping data from the same study; a sensitivity analysis based on the boot- strap method was used to check for overestimation of the precision of estimates. Finally, we produced funnel plots tolook at the possibility of small study effects, one cause of which is publication bias. All methods were undertaken according to the MOOSE 20 and QUOROM 21 guidelines for the conduct and reporting of systematic reviews. During the course of the study the PRISMA 22 guidelines were developedand we have also adhered to these where relevant. RESULTS Study characteristics Database searches identified 4373 potential studies of which 3239 were excluded on the basis of title (fig 1). Assessment of title and abstract led to the identification of 208 duplicate studies and the exclusion of 728 stu- dies not meeting eligibility criteria. For 514 studies, no primary care prescribing data were presented and 146 of the studies identified were reviews, and not original research. Sixty-eight articles were ecological studies and did not report on resistance individuals. The remaining 198 papers were read in full, and of these 174 were excluded on the basis of not including pri- marycare prescribing data (143), not reporting original research (23), and not reporting sufficient evidence to determine resistance risk (8). Twenty-four papers were included in the review. Table 1 summarises the characteristics of the final 24 studies includedin the review. 17-19 23-43 These consisted of five randomised controlled trials (RCTs) and 19 observational studies, two of which were prospective, and 17 retrospective controlled observational or case- control studies. These studies investigated effects in 15 505 adults and 12 103 children. Although not an inclusion criterion, all studies were based in countries where antibiotics are available by prescription only. Twenty-two studies sampled bacteria from patients with symptomatic infection: urinary tract infection (seven studies); 17-19 23-25 27 respiratory tract infections (seven); 28-34 otitis media (two); 36 42 chronic obstructive pulmonary disease (one); 37 meticillin resistant Staphy- lococcus aureus (MRSA) infection (four); 38-41 and tra- choma in children (one). 43 Two studies examined asymptomatic healthy adult volunteers. 18 35 Studies presented a wide range of antibiotic exposure analyses including those for: macrolides (eight studies); 19 24 29 31 33-35 37 42 43 penicillins (five); 17 27 32 34 35 sulphonamides and trimethoprim (six); 17-19 23-25 cepha- losporins (six); 27 33 34 36 37 42 tetracyclines (two); 33 37 qui- nolones (two); 33 37 nalidixic acid (one); 27 metronidazole (one); 24 nitrofurantoin (one); 27 and any antibiotic (seven) 18 25 30 38-41 given between two and 104 weeks before measurement of antibiotic resis- tance. Adherence to antibiotic regime was not assessed in most of the studies as they were retrospective and researchers were only able to measure prescribing from patients records or questionnaires, though in one of the RCTs 32 adherence was measured by record- ing medicine bottle weights at various time points throughout the study. Table 2 reports the quality assessment of the studies. Only one study 35 met all five criteria. Three studies, 25-27 all of urinary tract bacteria, met fewer than three criteria and were excluded from the meta- analyses. Resistance in urinary bacteria Of the eight studies investigating effects on urinary bacteria, one was an antibiotic comparison study 21 (table 1). The remaining seven studies reported com- parisons between no treatment and prescription of any antibiotic, 18 25 23 amoxicillin, 19 trimethoprim, 17 19 24 and ofloxacin 25 in relation to resistance to ampicillin, 19 trimethoprim, 17-19 23-25 sulphonamides, 24 or ciprofloxacin. 25 Five were of sufficient quality to be included in the meta-analysis. 17-19 24 23 Estimates from these studies were grouped according to the periods over which exposure was measured and represented on a forest plot (fig 2). This plot shows that at all time MEDLINE and EMBASE search strategy 1 exp Drug Resistance, Microbial/145030 2 bacterial resistan$.tw. 7742 3 antibiotic resistan$.tw. 36098 4 family practice (no related terms) 1530 5 antimicrobial resistance (no related terms) 2042 6 drug resistan$.tw. 77383 7 primary care (no related terms) 10339 8 ambulatory care (no related terms) 913 9 family practice (no related terms) 1530 10 exp Primary health care/91893 11 antibiotics (no related terms) 10404 12 2 or 3 or 4 or 6 or 7 232539 13 5 or 8 or 9 or 10 or 11 97137 Combination of 12 and 13 4373 $ indicates truncation. tw=text word. No related terms specifies search words only RESEARCH BMJ | ONLINE FIRST | bmj.com page 3 of 11 Table 1 | Study characteristics (by reference number) Study Design n Participants Recruitment Bacteria Sample site Method AB used AB to which resistance measured Time* (months) Crude or adjusted OR (95% CI), P for primary comparison Urinary bacteria 17 O 8833 SA MLR Ec U PDT Tr Tr 0-6 1.22 (1.16 to 1.28) P<0.001 19 CC 903 SA GP Ec U PDT Am Am 0-12 1.7 (1.24 to 2.32) Ec U PDT Tr Tr 0-12 2.39 (1.62 to 3.53) 23 CC 3435 SA MLR Ec U PDT AA Tr 0-6 1.36 (1.14 to 1.60) Ec U PDT Tr Tr 0-6 3.95 (3.04 to 5.12) 24 CC 559 SA OP Ec U PDT ST, Q, P, AA ST 0-12 4.1 (2.20 to 7.50) P=0.03 18 O 618 ASA GP or POST Ec U PDT AA Tr,Am 0-12 1.13 (0.79 to 1.63) Ec U PDT AA Tr, Am 0-2 1.93 (1.06 to 3.51) 25 CC 300 SA GP Ec U M O or Am Q 0-6 0.26 (0.09 to 0.74) 26 OP 35 SC GP Ec U PDT Su Am, Str, Te, Ch, K, Su, G, Tr, N NR 0.14 (0.03 to 0.56) P=0.9 27 RCT 131 SA women with recurrent UTI OP Ec U PDT N or Cp N 2-6 W 0.67 (0.30 to 1.6) P=0.01 Respiratory bacteria 28 OP 119 SC GP H T MIC Am Am 2 1.9 (1.3 to 2.7) p<0.001 29 CC 145 SA GP Sp T PDT Ma Ma 0-6W 2.11 (1.05 to 4.26) 30 CC 412 SA GP Hi S PDT AA Am 0-3 2.1 (1.2 to 36.0) P<0.002 31 O 263 SC GP Spy T PDT Ma Ma 0-3 4.19 (1.23 to 14.26) 32 RCT 795 SC OP Sp NP PDT Sh Am or StAm P, Am, Tr-Su 0-2 1.5 (1.2 to 2.8) P=0.001 33 O 737 SC OP Sp NP E Cp P, Ce, E, Tr, Su, To, Cf, Cef, Te 0-4 2.3 (1.04 to 5.1) P<0.05 Sp U PDT Not Ma Ma 0-6W 1.27 (0.64 to 2.54) 34 CC 134 SC and ASC PED Sp NP,OC,AN E B P 0-1 Sp NP,OR,AN Not B P 0-1 6.0 (0.06 to 516.4) Sp NP,OR,AN B P 0-6 1.0 (0.003 to 255.6) Sp NP,OR,AN Not B P 0-6 13.7 (0.6 to 311.48) Sp NP,OR,AN B + Oth P 0-6 0.02 (0.0007 to 1.08) 35 RCT 224 ASA U Sp TON GAP Az Ma 8D 1.86 (0.9 to 3.75) P<0.05 Sp TON GAP Cl Ma 8D 2.47 (1.2 to 5.07) 36 CC 448 Cwithsuspected otitis media DCC Pn NP PDT B, Co, or E P 0-12 8.8 (2.69 to 28.8) P<0.001 37 CC 100 A with COPD OP Hi S PDT P B 0-24 2.15 (0.94 to 4.89) Hi S PDT Cp B 0-24 4.04 (1.22 to 13.4) Hi S PDT Te B 0-24 2.45 (1.09 to 5.49) Hi S PDT Ma B 0-24 4.04 (1.22 to 13.4) Hi S PDT Co B 0-24 1.94 (0.82 to 4.6) Hi S PDT Q B 0-24 2.13 (0.5 to 9.07) Hi S PDT Oth B 0-24 0.69 (0.29 to 1.61) MRSA 38 O 167 A with skin infection GP Sa SK PDT AA Me 0-1 2.7 (0.8 to 15.0) P<0.1 Sa SK PDT Me 0-6 3.1 (1.1 to 8.6) P<0.01 Sa SK PDT Me 0-12 1.5 (0.6 to 4.0) P<0.4 39 O 206 A with skin infection MTC Sa SK PDT AA Me 0-12 0.147(0.05to0.4) P=0.123 40 CC 131 C with MRSA positive N samples PED Sa N PDT AA Me 0-12 16.13 (6.38 to 40.76) P<0.001 41 O 840 C with Sa skin infection PED Sa SK PDT AA Me 0-3 0.98 (0.67 to 1.42) 42 RCT 1009 C with acute otitis media PED GN ST PDT Cf or Amor Az CEFT, Cp, Am 3-5D P<0.01 43 RCT 121 C with early trachoma PED Ct OC E Az Az 2W P=0.1 Design: O=observational, OP=observational prospective, CC=case-control, RCT=randomised controlled trial. Participants: A=adults, C=children, SA/SC=symptomatic adults/children, ASA/ ASC=asymptomatic adults/children. Recruitment: GP=general practice, POST=postal recruitment, PED=paediatric clinic, OP=outpatient clinic, DCC=day care centre, MLR=medical lab records (routine laboratory samples linked to regional prescribing database), MTC=military training centre, U=university. Bacteria: Ct=Chlamydia trachomatis, Ec=E coli, GN=gram negative, H=Haemophilus, Hi=Haemophilus influenzae, Pn= pneumococci,Sp=S pneumoniae Sa=S aureus, Spy=S pyogenes.Antibiotics (AB): AA=any antibiotic, Az=azithromycin, Am=amoxicillin, B= lactam, Ce=ceftriaxone, Cf=cefprozil, Cef=cefaclorR, Cp=cephalexin, Cl=clarithromycin, Co=co-trimoxazole, E=erythromycin, G=gentamicin, K=kenamycin, Ma=macrolide, Me=meticillin, N=naladixic acid, O=ofloxacin, P=penicillin, Q=quinolone, ShAm=short course high dose amoxicillin, StAm=standard amoxicillin, Str=streptomycin, Su=sulphonamide, ST=sulphamethoxazole- trimethoprim, Te=tetracycline, Tr=trimethoprim, To= trimethoprimsulfamethoxazole. Method (of measuring resistance): PDT=paper disk testing, MIC=minimum inhibitory concentration, E=E testing (AB Biodisk, Sweden), GAR=grid agar plate. Sample site: U=urine, T=throat, NP=nasopharyngeal, N=nasal, OC=ocular, ST=stool, SK=skin, S=sputum, TON=tonsil, OR=oral cavity, AN=anterior nares. *Either exact time at which individuals took antibiotics, or time period during which antibiotic prescribing was recorded, before measurement of resistance. Insufficient data to calculate OR and/or 95% CI and/or P value. RESEARCH page 4 of 11 BMJ | ONLINE FIRST | bmj.com periods the odds of resistance were greater in patients exposed to these antibiotics than in those who were unexposed and that the strongest association was at 0-1 months, with reduced association at subsequent time points, and a small but important residual associa- tion within 12 months. In participants who were unex- posed to antibiotics, the pooled odds of resistance variedlittle betweentime periods (pooledodds of resis- tance among unexposed participants was 0.44 at 0-1 month, fig 2). The coefficient for each month increase in the exposure period was 0.33 (95% CI 0.49 to 0.17, P<0.001) from the meta-regression showing a clear time trend. There was no evidence of within group heterogeneity in the 0-1 and 0-3 month periods, but some evidence of heterogeneity in the 0-6 month and 0-12 month periods. For studies in whichadjustedORs were presentedwe dida meta-ana- lysis of these results, although we only had sufficient data to do this for the studies examining resistance in urinary tract bacteria at the 0-3 month period. The resulting pooled OR did not differ markedly from the unadjusted or crude ORused in the meta-regression at the same time point (pooled adjusted OR2.17, 95%CI 1.49 to 3.22), compared with 2.48 (2.06 to 2.98) Resistance in respiratory bacteria Nine studies examined resistance in respiratory tract bacteria. 28-35 37 Since the Malhotra-Kumar 35 andChung studies 28 were prospective and measured resistance at specific time points, their data are presented separately from the remaining retrospective studies measuring antibiotic exposure during periods of time. The Sportel study 37 measured resistance over a 0-24 month period ina groupof patients with chronic obstructive pulmon- ary disease (in which the lower respiratory tract becomes colonised with different bacteria from those usually resident in the upper respiratory tract 44 ) so it was not included in the meta-analysis. The remaining seven studies were meta-analysed according to time periods. 29-34 36 Figure 3 shows that although there was some evidence of an association between antibiotics and resistance between 0 and 1 month (with an OR of 2.1, 95%CI 1.0 to4.2), 0 and2 months (pooledOR2.4, 95% CI 1.4 to 3.9), and 0 and 12 months (pooled OR 2.4, 95%CI 1.3to4.5), intervening periods showedless evidence of such associations, and no association between resistance and time with a coefficient of 0.01 (95% CI 0.26 to 0.24, P=0.91). We found little within group heterogeneity with the most heterogene- ity present in the 0-12 month period with an I 2 value of 57.3% (P=0.04). Among participants who were unex- posed to antibiotics the pooled odds of resistance var- ied across time periods (fig 3) from 0.08 in the 0-2 month period to 0.51 during the 0-3 month period. Some studies did not report raw data and for these stu- dies the odds of resistance in the control group could not be calculated. The Chung study 28 used minimum inhibitory con- centrations of ampicillin as a measure of antibiotic resistance and also presence of the integrative and con- junctive element ICEHin 1056 that encodes -lacta- mase and circulates among nasopharyngeal Haemophilus species. Prescribing amoxicillin to a child in general practice more than tripled the mean inhibitory concentration for ampicillin (9.2 g/ml compared with 2.7 g/ml, P=0.005) and doubled the risk of isolation of Haemophilus isolates possessing homologues of ICEHin1056 (67%, 42 of 63, compared with 36%, 14 of 39 in patients who were not exposed to antibiotic) with a relative risk of 1.9 (95%CI 1.2 to 2.9) at two weeks post-exposure and 1.0 (0.5 to 1.7) at 12 weeks post-exposure. 28 The Malhotra-Kumar study 35 was the only rando- misedcontrolledtrial toexamine resistance (associated with azithromycin or clarithromycin) at specific time points. Figure 4 and the meta-analysis of the pooled 0-1 month Donnan 17 Hillier 19 Hillier 19 Pooled odds ratio Test for heterogeneity: I 2 =0.0%, P=0.576 0-3 months Donnan 17 Hillier 19 Hillier 19 Hay 18 Pooled odds ratio Test for heterogeneity: I 2 =0.0%, P=0.796 0-6 months Steinke 23 Donnan 17 Steinke 23 Hillier 19 Donnan 17 Hillier 19 Metlay 24 Pooled odds ratio Test for heterogeneity: I 2 =89.2%, P=0.000 0-12 months Donnan 17 Donnan 17 Hillier 19 Hay 18 Hillier 19 Pooled odds ratio Test for heterogeneity: I 2 =71.9%, P=0.007 4.45 (3.78 to 5.21) 4.85 (2.63 to 8.94) 3.11 (1.57 to 6.17) 4.40 (3.78 to 5.12) 2.60 (2.04 to 3.33) 2.62 (1.69 to 4.07) 2.26 (1.41 to 3.62) 1.93 (1.06 to 3.51) 2.48 (2.06 to 2.98) 1.36 (1.14 to 1.61) 1.67 (1.32 to 2.10) 3.95 (3.04 to 5.12) 1.83 (1.39 to 2.42) 1.65 (1.10 to 2.46) 2.57 (1.83 to 3.61) 4.10 (2.20 to 7.50) 2.18 (1.57 to 3.03) 1.22 (1.16 to 1.28) 1.18 (1.06 to 1.32) 1.62 (1.18 to 2.23) 1.13 (0.79 to 1.63) 2.36 (1.59 to 3.50) 1.33 (1.15 to 1.53) Trimethoprim Trimethoprim Amoxicillin Trimethoprim Trimethoprim Amoxicillin Any antibiotic Any antibiotic* Trimethoprim Trimethoprim Amoxicillin Any antibiotic* Trimethoprim ST Trimethoprim Any antibiotic* Amoxicillin Any antibiotic* Trimethoprim 0.1 * Any antibiotic other than trimethoprim. ST=sulfamethoxazole-trimethoprim. NR=not reported 1 10 Time period, study Antibiotic use associated with susceptibility Antibiotic use associated with resistance Odds ratio (95% CI) Odds ratio (95% CI) Antibiotic exposure NR 20 20 NR 39 39 20 19 NR 19 28 NR 28 28 NR NR 19 38 19 Resistance in unexposed (control) group (%) Fig 2 | Forest plot showing individual study and pooled ORs (log scale) for resistance in urinary tract bacteria (E coli) and antibiotic exposure. Studies grouped according to time period during which exposure was measured and ordered within each time period by increasing standard error RESEARCH BMJ | ONLINE FIRST | bmj.com page 5 of 11 OR showed a decaying association with resistance to macrolides at all time points upto6 months withstrong evidence of a time trend ( coefficient 0.25 (95% CI 0.39 to 0.11, p=0.004). 35 Resistance over time in MRSA studies We found few studies investigating effects on MRSA; three studies in skin samples 38 39 41 and one study in nasal samples. 40 Paganini et al 41 examined community acquired MRSA in children. These isolates were obtained from skin and soft tissue infections, and some invasive infection sites. The study found that 10% (26 of 273) of isolates were resistant to clindamy- cin as well as meticillin and 1%(two of 272) were resis- tant to trimethoprim-sulfamethoxazole. Raw data obtained from the authors allowed the calculation of an OR for resistance of 0.98 (95%CI 0.67 to 1.42) sug- gesting that previous antibiotic use is not an important risk factor for community acquired MRSA isolated from childrens skin infections. However, exposure data for this study relied on parental reports only. Campbell et al 39 examined community acquired MRSA in skin infections in healthy military trainees. Previous antibiotic use was not associated with MRSA infection (OR 0.7 (95% CI 0.2 to 1.9). Baggett et al 38 investigated a large outbreak of com- munity acquired MRSA in a rural community and found a strong association (OR 3.1, 95% CI 1.1 to 8.6) between this infection and the prescription of any antibiotic in the previous 0-6 months. This association disappeared (1.5, 0.6 to 4.0) when the time period was broadened to include any antibiotic prescription in the preceding 12 months. Lo et al 40 examined resistance associated with the use of any antibiotic in the 12 months preceding resis- tance testing. This study reported a strong association of OR16.1 (95%CI 6.4 to 40.8) between previous anti- biotic use and nasal colonisation of Panton-Valentine leukocidin positive MRSA in healthy children. We did a meta-analysis of the three studies investi- gating MRSA and resistance in bacteria sampled from skin abrasions 38 39 41 in which individuals had been exposed to antibiotics in the previous 12 months; the pooled OR for these studies was 1.04, with the confi- dence interval crossing the null (95% CI 0.47 to 2.29). Comparisons between antibiotics Many of the studies included in the review looked at head to head comparisons between antibiotics, although their statistical power was often limited. For example, the Ghaffar study 34 examined resistance to penicillin in Streptococcus pneumoniae following use of lactam drugs or lactam in combination with another antibiotic class but the 95% confidence intervals around the estimated OR varied from 0.1 to 516. Table 2 | Study quality (by reference number) Study Data on antibiotic exposure complete and accurate? Resistance well defined? Resistance threshold described? Unbiased control selection? Resistance status known before exposure? Adjustments for confounders? Adjusted OR (95% CI) P 17 Y Y N Y Unknown Y 1.22 (1.16 to 1.28) ) <0.001 19 Y Y Y Y Unknown Y Not reported 23 Y Y N Y Unknown Y 4.35 (3.03 to 5.73) 1.32 (1.10 to 1.60)
24 Y Y Y Y Unknown Y 4.1 (2.2 to 7.5, ) 0.03
18 Y Y Y Y Unknown Y 1.12 (0.77 to 1.65) Y Y Y Unknown Y 1.95 (1.08 to 3.49) 25 N Y Y NA Unknown Y Ciprofloxacin 20.6 (2.4 to 179.2), ofloxacin 7.6 (2.9 to 19.5) 0.006, 0.0001 26 N Y Y NA Unknown Unknown Not reported 27 Y Y Y NA Y Unknown Not reported 28 Y Y Y Y Y Y Not reported 29 N Y Y Y Unknown Y 1.0003 (1.0001 to 1.0004) <0.0001 30 Y Y Y Y Unknown Y 3.4 (1.3 to 9.2) <0.01 31 Y Y Y Y Unknown Y 2.16 (0.55 to 8.45) 0.26 32 Y Y Y Y Unknown Y Not reported 33 Y Y Y Y Unknown Y 2.7 (1.1 to 6.6 ) <0.001 34 Y Y Y Y Unknown Unknown Not reported 35 Y Y Y Y Y Y Not reported 36 Y Y Y Y Unknown Y Not reported 37 Y Y Y Y Unknown Y 1.15 (1.04 to 1.28) <0.01 38 Y Y Y Y Unknown Y Not reported 39 Y Y Y Y Unknown Y Not reported 40 Y Y Y Y Unknown Y 29.37 (10.72 to 80.50) <0.001 41 N Y Y Y Unknown Y Not reported 42 N Y Y No controls Unknown No Not reported 43 Y Y Not reported Y Unknown Y Not reported RESEARCH page 6 of 11 BMJ | ONLINE FIRST | bmj.com The Sportel study 37 allowed greater precision and showed that there was no greater association between penicillin use and ampicillin resistance than cephalos- porinuse inHaemophilus influenzae (OR0.4, 95%CI 0.1 to 1.4). The Malhotra-Kumar study 35 also reported no difference in macrolide resistance in Streptococcus pneu- moniae after use of azithromycin or clarithromycin at any previous time, although use of clarithromycin was associated with greater expression of the erm(B) gene, which confers high level macrolide resistance. To summarise, in comparisons of different anti- biotics in the same antibiotic classes for effects on resis- tance, we found no evidence that one class led to reduced resistance compared with another, although we were unable to adequately address the issue owing to the limited number of studies available. Effects of antibiotic dose, duration, and number of courses on resistance Several studies assessedthe relationbetweenresistance andincreasingcourses of or doses of antibiotic(table 3). Schrag et al 32 compared the effect of standard dose and duration of amoxicillinwith that of high dose and short durationamoxicillinonresistance at 5, 10, and28days. They reported some evidence of reduced resistance at 28 days associated with this treatment (OR 0.77, 95% CI 0.06 to 0.97), possibly attributable to better compli- ance. Hillier et al 19 found greater rates of resistance associated with higher doses of amoxicillin (OR 2.3, 95% CI 1.1 to 4.6) and longer courses of trimethoprim (2.9, 1.4 to 5.8), but no differences associated with dif- ferent course durations for amoxicillin (1.5, 0.7 to 2.9). Hillier also found associations between number of courses of amoxicillin (three or more v one; OR 3.9, 95%CI 1.0 to 14.7) and trimethoprim(three or more v one; 3.6, 1.2 to 10.5) and resistance. 19 The Hay study 18 showed no differences in resistance rates with differing numbers of antibiotic courses, but did find an increase of 1% in the odds of resistance for each 200 mg tri- methoprim tablet prescribed (OR 1.01, 95% CI 1.01 to 1.02). The same type of association was not found for increasing numbers of 500 mg amoxicillin tablets prescribed. Publication bias We were able to assess publication bias in the urinary bacteria studies investigating resistance in E coli and antibiotic exposure in the previous six months. The Funnel plot in fig 5 shows some evidence of positive publication bias. There were too few studies to assess publication bias for respiratory flora. DISCUSSION Principal findings Our reviewidentifieda number of studies that together provide strong evidence of an association at the indivi- dual patient level between the prescribing of anti- biotics in primary care and antimicrobial resistance in bacteria at different sites, including the urinary and respiratory tracts and the skin. Effects were strongest in the month directly after prescription but were detectable for up to12 months. This residual effect is likely to be an important driver for the high endemic levels of antibiotic resistance in the community. 28 Moreover, we found evidence of a dose-response rela- tion for two commonly prescribed first line antibiotics in primary care, amoxicillin and trimethoprim. Prescribing time periods Most studies that reported resistance in urinary and respiratory bacteria reported the association between resistance and antibiotics prescribed within overlap- ping time periods. This means that associations with longer time periods could reflect long or short term relations. However, the prospective studies did not suf- fer fromthis methodological weakness and did suggest persistence of resistance over a number of months. 28 35 0-1 month Beekmann 29 Pooled odds ratio 0-2 months Seaton 30 Ciftci 31 Pooled odds ratio Test for heterogeneity: I 2 =1.6%, P=0.313 0-3 months Schrag 32 Samore 33 Samore 33 Samore 33 Pooled odds ratio Test for heterogeneity: I 2 =44.2%, P=0.146 0-6 months Ghaffar 34 Ghaffar 34 Pooled odds ratio Test for heterogeneity: I 2 =0.0%, P=0.463 0-12 months Beekmann 29 Samore 33 Samore 33 Arason 36 Arason 36 Arason 36 Pooled odds ratio Test for heterogeneity: I 2 =57.3%, P=0.039 2.10 (1.05 to 4.26) 2.10 (1.04 to 4.23) 2.10 (1.20 to 3.60) 4.19 (1.23 to 14.26) 2.37 (1.42 to 3.95) 1.50 (1.20 to 1.80) 2.30 (1.04 to 5.10) 1.80 (0.80 to 4.20) 0.40 (0.10 to 1.30) 1.48 (0.95 to 2.32) 1.56 (0.50 to 4.86) 3.93 (0.44 to 35.28) 1.90 (0.69 to 5.21) 1.28 (0.64 to 2.54) 1.20 (0.50 to 2.50) 1.60 (0.80 to 3.50) 6.75 (1.78 to 25.51) 7.22 (1.73 to 30.05) 8.56 (1.14 to 64.04) 2.37 (1.25 to 4.50) Any antibiotic Any antibiotic Macrolide lactam Cephalosporin Penicillin Macrolide lactam lactam* Any antibiotic Penicillin Cephalosporin lactam Co-trimoxazole Erythromycin 0.1 * lactam plus another antibiotic. NR=not reported 1 10 Time period, study Antibiotic use associated with susceptibility Antibiotic use associated with resistance Odds ratio (95% CI) Odds ratio (95% CI) Antibiotic exposure 13 13 2 33 17 17 17 14 14 13 NR NR NR NR NR Resistance in unexposed (control) group (%) Fig 3 | Forest plot showing individual study and pooled ORs (log scale) for resistance in respiratory tract bacteria and previous antibiotic prescribing. Studies grouped according to time period during which exposure was measured and ordered within each time period by increasing standard error RESEARCH BMJ | ONLINE FIRST | bmj.com page 7 of 11 Reverse causality and confounding If bacterial samples are takenonly if the illness does not respond to first line antibiotics (as is standard practice in many parts of the UK) then retrospective case-con- trol analyses will show a spuriously strong association with previous antibiotic prescribing. This bias is avoided if only incident cases are included. For most of the studies reviewed it was not possible to ascertain whether non-incident cases had been excluded. This bias is less likely, although not impossible, in prospective studies than in retrospective ones. How- ever, bothof the prospective studies eliminatedreverse causality as a possible explanation by collecting base- line resistance data; they showed substantial increases inresistance withindays of prescribing andsubsequent decay in effects over three months in the Chung study (28) and six months in the Malhotra-Kumar study. 35 Other confounding associations, such as the relation between community prescribing and recent hospital admission, could also have introduced bias. However, the studies that attempted to adjust for potential con- founders such as age, sex, comorbidities, catheter use, and smoking status seldom demonstrated substantial difference between crude and adjusted estimates of association. Heterogeneity between studies The observed differences between studies may well reflect the difficulties of overlapping time periods and confounding, but could also reflect the differences in populations studied (which must have varied in base- line prevalence of antibiotic resistance and transmis- sion potential), the definition of resistance applied, and the different antibiotic-organismrelationships stu- died. Residual heterogeneity was a particular problem in the pooled analyses of urinary bacteria in the 0-6 month and 0-12 month periods and in the analyses of respiratory bacteria in the 0-12 period. As pre- viously noted, this heterogeneity existed despite some evidence of publication bias. MRSA studies Hospital MRSA strains are becoming feral, persisting in the community, and non-hospital epidemic strains are being acquired in the community. 45 Although two of the studies suggest that the selective pressure pro- duced by antibiotic prescribing in the community may contribute to this problem, the other two showed no effect. The mechanism of MRSA transmission is clearly complex and to explore this issue further, repeated screening of large numbers of individuals (mostly non-carriers) would be necessary over a long period. In the meantime, minimisation of unnecessary community prescribing for skin infections seems a rea- sonable precautionary principle. Clinical and research implications This review provides the evidence needed by clini- cians responsible for the prescription of antibiotics in primary care to quantify the link between individual prescribing decisions and the problem of antibiotic resistance. Although the clinical impact of isolating antibiotic resistant bacteria warrants further research in its own right, resistance is not simply a characteristic of the infecting organism. It is also related to the indi- viduals bacterial gene pool, since resistance carried on plasmids and integrons can be transferred between commensal organisms and potential pathogens. And because both transmission of commensal organisms between individuals and antibiotic prescribing in the At 1 week post-antibiotic Azithromycin Clarithromycin Pooled odds ratio At 2 weeks post-antibiotic Azithromycin Clarithromycin Pooled odds ratio At 1 month post-antibiotic Azithromycin Clarithromycin Pooled odds ratio At 2 months post-antibiotic Azithromycin Clarithromycin Pooled odds ratio At 6 months post-antibiotic Azithromycin Clarithromycin Pooled odds ratio 14.67 (6.14 to 35.00) 10.44 (4.65 to 23.44) 12.22 (6.76 to 22.10) 11.59 (5.08 to 26.47) 5.37 (2.58 to 11.17) 7.70 (3.63 to 16.34) 9.25 (4.21 to 20.31) 4.13 (2.01 to 8.46) 6.08 (2.76 to 13.39) 4.37 (2.13 to 8.98) 3.00 (1.48 to 6.08) 3.61 (2.18 to 5.97) 1.89 (0.88 to 3.77) 2.47 (1.20 to 5.07) 2.16 (1.30 to 3.61) 0.1 1 10 100 Time period Antibiotic use associated with susceptibility Antibiotic use associated with resistance Odds ratio (95% CI) Odds ratio (95% CI) Fig 4 | Forest plot showing individual analytic and pooled ORs (log scale) for resistance in respiratory tract streptococci of healthy volunteers from the Malhotra-Kumar study 35 and previous antibiotic prescribing Log odds ratio S t a n d a r d
e r r o r -0.5 0 0.5 1.0 1.5 0.5 0.4 0.3 0.2 0.1 0 Fig 5 | Funnel plot with pseudo 95% confidence limits for studies investigating the relation between resistance in urinary bacteria and antibiotic use in the previous six months RESEARCH page 8 of 11 BMJ | ONLINE FIRST | bmj.com community remain frequent events, even a transient effect of antibiotic use on the carriage of resistant organisms by an individual could have a major impact on the endemic level of resistance in the population. 28 Our findings also provide evidence to support the Standing Medical Advisory Committee report recom- mendations that the fewest number of antibiotic courses should be prescribed for the shortest period possible. 3 And they draw attention to the increased risk of resistance to commonly used first line anti- biotics: if a patient has received one or more courses of such antibiotics in the previous 12 months and further antibiotic treatment is necessary, for a subse- quent respiratory or urinary infection, consideration should be given to choosing a different antibiotic. This final implication serves to highlight that the only way to avoid the vicious cycle of resistance leading to the ever greater use of more powerful broad spectrum antibiotics is to avoid their initial use whenever possible. The main research implication is the need to strengthen the current evidence base, which is heavily reliant on observational studies, with more clinical trials. We believe that the opportunity to assess the effects of antibiotics on antimicrobial resistance should be considered whenever a placebo or no treatment controlled trial is being designed. Further research is also needed to assess relations between antibiotics pre- scribed in primary care and more serious infections that require secondary care treatment, as well as to further clarify the effects of interactions between anti- biotic dose, duration, and adherence on resistance. In conclusion, we have summarised and synthesised evidence fromaroundthe worldthat primarycare anti- biotics make an important contribution to the problem of antimicrobial resistance. Primary care clinicians and patients may wish to consider this evidence when dis- cussing the benefits and risks of prescribing and con- suming antibiotics. Contributors: CC performed the searches. CC and ADH identified eligible studies. All authors appraised study quality and extracted data. CC and CM transformed data and performed the meta-analyses. CC and ADH drafted first sections of the text. All authors contributed to the final draft. ADH and CC are the guarantors. All authors had full access to all of the data (including statistical reports and tables) in the study and can take responsibility for the integrity of the data and the accuracy of the data analysis. Funding: This work was undertaken by the University of Bristol in collaboration with the University of Oxford which both received a proportion of their funding from the Department of Healths NIHR School for Primary Care Research. The views expressed in this publication are those of the authors and not necessarily those of the Department of Health. The funder had no role in the study design; data collection, data analysis, and interpretation of data; in the writing of the report; or in the Table 3 | Relations between antibiotic dose and resistance (by reference number) Study High dose/ concentration Standard dose and duration Antibiotic to which resistance was measured Time* OR (95% CI) Schrag 2001 32 High dose amoxicillin 90 mg/kg for 5 days Normal dose amoxicillin 40 mg/kg per day for 10 days Penicillin 5 days 0.9 (0.6 to 1.2) 10 days 1.16 (0.88 to 1.53) 28 days 0.77 (0.06 to 0.97) Hillier 2007 19 High dose amoxicillin 500 mg Normal dose amoxicillin 250 mg Ampicillin 0-12 months 2.26 (1.13 to 4.55) Amoxicillin >7 days Amoxicillin <7 days Ampicillin 0-12 months 1.50 (0.76 to 2.92) Trimethoprim >7 days Trimethoprim <7 days Ampicillin 0-12 months 2.89 (1.44 to 5.78) Hillier 2007 19 2 courses amoxicillin 1 course amoxicillin Ampicillin 0-12 months 1.58 (0.77 to 3.27) 3 courses amoxicillin 1 course amoxicillin Ampicillin 0-12 months 3.95 (1.06 to 14.72) 2 courses trimethoprim 1 course trimethoprim Trimethoprim 0-12 months 0.98 (0.39 to 2.49) 3 courses trimethoprim 1 course trimethoprim Trimethoprim 0-12 months 3.62 (1.25 to 10.48) Hay 2005 18 2 courses any antibiotic 1 course any antibiotic Trimethoprim 0-12 months 1.18 (0.53 to 2.37) 3 courses any antibiotic 1 course any antibiotic Trimethoprim 0-12 months 0.4 (0.12 to 1.31) 4 courses any antibiotic 1 course any antibiotic Trimethoprim 0-12 months 2.77 (0.94 to 8.15) Increasing dose of trimethoprim by 200 mg Normal dose of trimethoprim Trimethoprim 0-12 months 1.01 (1.01 to 1.02) Increasing dose of lactam by 500 mg Normal dose of lactam Amoxicillin 0-12 months 1.00 (0.99 to 1.01) *Either exact time at which individuals took antibiotics, or time period during which antibiotic prescribing was recorded, before measurement of resistance. WHAT IS ALREADY KNOWN ON THIS TOPIC Worldwide, primary care is responsible for the majority of antibiotic use by human beings Although many countries have reduced prescribing rates, substantial variations remain between countries Many clinicians and patients do not see antibiotic resistance as a reason to refrain from antibiotic use WHAT THIS STUDY ADDS Antibiotics prescribed to an individual in primary care were consistently found to be associated with resistance of urinary and respiratory bacteria to those antibiotics in that individual Antibiotics prescribed in primary care may impact on bacterial resistance in a patient for up to 12 months The greater the number or duration of antibiotic courses prescribed in the previous 12 months, the greater the likelihood that resistant bacteria would be isolated from that patient RESEARCH BMJ | ONLINE FIRST | bmj.com page 9 of 11 decision to submit the article for publication. The researchers were independent of the funder. Competing interests: All authors have completed the Unified Competing Interest form at www.icmje.org/coi_disclosure.pdf (available on request from the corresponding author) and declare that (1) CC, ADH, DM, AL, and CM have support from the Universities of Bristol and Oxford for the submitted work; (2) None of the authors has relationships with any companies that might have an interest in the submitted work in the previous 3 years; (3) None of the authors spouses, partners, or children have any financial relationships that may be relevant to the submitted work; and (4) None of the authors has any non-financial interests that may be relevant to the submitted work. Ethical approval: Not required. Data sharing: No additional data available. 1 Department of Health. UK antimicrobial resistance strategy and action plan. 2000. www.dh.gov.uk/prod_consum_dh/groups/ dh_digitalassets/@dh/@en/documents/digitalasset/ dh_4078448.pdf. 2 Butler CC, Hillier S, Roberts Z, Dunstan F, Howard A, Palmer S. Antibiotic-resistant infections in primary care are symptomatic for longer andincrease workload: outcomes for patients withE coli UTIs. Br J Gen Pract 2006;56:686-92. 3 Department of Health. The path of least resistance. 1998. www.dh. gov.uk/prod_consum_dh/groups/dh_digitalassets/@dh/@en/ documents/digitalasset/dh_4120729.pdf. 4 Prescription Pricing Authority. Trends in antibiotic prescribing in England. 2006. www.ppa.org.uk/news/pact-102004.htm. 5 Ashworth M, Charlton J, Ballard K, Latinovic R, Gulliford M. Variations in antibiotic prescribing and consultation rates for acute respiratory infection in UK practices 1995-2000. Br J Gen Pract 2005;55:603-8. 6 Ferech M, Coenen S, Malhotra-Kumar S, Dvorakova K, Hendrickx E, Suetens C, et al. Europeansurveillanceof antimicrobial consumption (ESAC): outpatient antibiotic use in Europe. J Antimicrob Chemother 2006;58:401-7. 7 Macfarlane J, Holmes W, Macfarlane R, Britten N. Influence of patients expectations on antibiotic management of acute lower respiratory tract illness in general practice: questionnaire study. BMJ 1997;315:1211-4. 8 Brooks L, ShawA, Sharp D, Hay AD. Towards a better understanding of patients perspectives of antibiotic resistance and MRSA: a qualitative study. FamPract 2008;25:341-8. 9 Butler CC, Rollnick S, Pill R, Maggs-Rapport F, Stott N. Understanding the culture of prescribing: qualitative study of general practitioners and patients perceptions of antibiotics for sore throats. BMJ 1998;317:637-42. 10 Kumar S, Little P, Britten N. Why do general practitioners prescribe antibiotics for sore throat? Grounded theory interview study. BMJ 2003;326:138. 11 Seppala H, Klaukka T, Vuopio-Varkila J, Muotiala A, Helenius H, Lager K, et al. The effect of changes in the consumption of macrolide antibiotics on erythromycin resistance in group A streptococci in Finland. N Engl J Med 1997;337:441-6. 12 Priest P, Yudkin P, McNulty C, Mant D, Wise R. Antibacterial prescribing and antibacterial resistance in English general practice: cross sectional study. Commentary: antibiotic resistance is a dynamic process. BMJ 2001;323:1037-41. 13 Lipsitch M, Samore MH. Antimicrobial use and antimicrobial resistance: a population perspective. Emerg Infect Dis 2002;8:347-54. 14 Simpson SA, Wood F, Butler CC. General practitioners perceptions of antimicrobial resistance: a qualitative study. J AntimicrobChemother 2007;59:292-6. 15 NICE. Respiratory tract infections: prescribing of antibiotics for self- limiting respiratory tract infections in adults and children in primary care. 2008. www.nice.org.uk/CG69. 16 Hillier SL, Magee JT, Howard AJ, Palmer SR. How strong is the evidence that antibiotic use is a risk factor for antibiotic-resistant, community-acquired urinary tract infection? J Antimicrob Chemother 2002;50:241-7. 17 Donnan PT, Wei L, Steinke DT, Phillips G, Clarke R, Noone A, et al. Presence of bacteriuria caused by trimethoprimresistant bacteria in patients prescribed antibiotics: multilevel model with practice and individual patient data. BMJ 2004;328:1297-300. 18 Hay AD, Thomas M, Montgomery A, Wetherell M, Lovering A, McNulty C, et al. The relationship between primary care antibiotic prescribing and bacterial resistance in adults in the community: a controlled observational study using individual patient data. J Antimicrob Chemother 2005;56:146-53. 19 Hillier S, Roberts Z, Dunstan F, Butler C, Howard A, Palmer S. Prior antibiotics and risk of antibiotic-resistant community-acquired urinary tract infection: a case-control study. J Antimicrob Chemother 2007;60:92-9. 20 Stroup DF, Berlin JA, Morton SC, Olkin I, Williamson GD, Rennie D, et al. Meta-analysis of observational studies in epidemiology: a proposal for reporting. JAMA 2000;283:2008-12. 21 Moher D, Cook DJ, Eastwood S, Olkin I, Rennie D, Stroup DF. Improving the quality of reports of meta-analyses of randomised controlled trials: the QUOROMstatement. Onkologie 2000;23:597-602. 22 Moher D, Liberati A, Tetzlaff J, Altman DG. Preferred reporting items for systematic reviews and meta-analyses: the PRISMA statement. J Clin Epidemiol 2009;62:1006-12. 23 Steinke DT, Seaton RA, Phillips G, MacDonald TM, Davey PG. Prior trimethoprimuseandtrimethoprim-resistant urinary tract infection: a nested case-control study with multivariate analysis for other risk factors. J Antimicrob Chemother 2001;47:781-7. 24 Metlay JP, StromBL, Asch DA. Prior antimicrobial drug exposure: a risk factor for trimethoprim-sulfamethoxazole-resistant urinary tract infections. J Antimicrob Chemother 2003;51:963. 25 Colodner R, Kometiani I, ChazanB, Raz R. Risk factors for community- acquired urinary tract infection due to quinolone-resistant E coli. Infection 2008;36:41-5. 26 Grunberg RN, Shaw EJ. The influence of antibiotic treatment on resistance patterns of coliformbacilli in childhood urinary-tract infection. J Med Microbiol 1976;9:233-7. 27 Preiksaitis JK, Thompson L, Harding GKM, Marie TJ, Hoban S, Ronald AR. A comparison of the efficacy of nalidixic acid and cephalexin in bacteriuric women and their effect on fecal and periurethral carriage of enterobacteriaceae. J Infect Dis 1981;143:603-8. 28 Chung A, Perera R, Brueggemann AB, Elamin AE, Harnden A, Mayon-White R, et al. Effect of antibiotic prescribing on antibiotic resistance in individual children in primary care: prospective cohort study. BMJ 2007;335:429-34. 29 Beekmann SE, Diekema DJ, Heilmann KP, Richter SS, Doern GV. Macrolide use identified as risk factor for macrolide-resistant Streptococcus pneumoniae in a 17-center case-control study. Eur J Clin Microbiol Infect Dis 2006;25:335-9. 30 Seaton RA, Steinke DT, Phillips G, MacDonald T, Davey PG. Community antibiotic therapy, hospitalization and subsequent respiratory tract isolation of Haemophilus influenzae resistant to amoxycillin: a nested case-control study. J Antimicrob Chemother 2000;46:307-9. 31 Ciftci E, Dogru U, Guriz H, Aysev D, Ince E. Investigation of risk factors for tonsillopharyngitis with macrolide resistant Streptococcus pyogenes in Turkish children. Pediatr Int 2002;44:647-51. 32 Schrag SJ, Pena C, Fernandez J, Sanchez J, Gomez V, Perez E, et al. Effect of short-course, high-dose amoxicillin therapy on resistant pneumococcal carriage: a randomized trial. JAMA 2001;286:49-56. 33 Samore MH, Magill MK, Alder SC, Severina E, Morrison-De Boer L, Lyon JL, et al. High rates of multiple antibiotic resistance in Streptococcus pneumoniae fromhealthy children living in isolated rural communities: association with cephalosporin use and intrafamilial transmission. Pediatrics 2001;108:856-65. 34 Ghaffar F, Muniz LS, Katz K, Smith JL, Shouse T, Davis P, et al. Effects of large dosages of amoxicillin/clavulanate or azithromycin on nasopharyngeal carriage of Streptococcus pneumoniae, Haemophilus influenzae, nonpneumococcal alpha-hemolytic streptococci, andStaphylococcus aureus inchildrenwithacute otitis media. Clin Infect Dis 2002;34:1301-9. 35 Malhotra-Kumar S, Lammens C, Coenen S, Van Herck K, Goossens H. Effect of azithromycin and clarithromycin therapy on pharyngeal carriage of macrolide-resistant streptococci in healthy volunteers: a randomised, double-blind, placebo-controlled study. Lancet 2007;369:482-90. 36 ArasonVA, KristinssonKG, SigurdssonJA, Stefansdottir G, MolstadS, Gudmundsson S. Do antimicrobials increase the carriage rate of penicillin resistant pneumococci in children? Cross sectional prevalence study. BMJ 1996;313:387-91. 37 Sportel JH, Koeter GH, van Altena R, Lowenberg A, Boersma WG. Relation between beta-lactamase producing bacteria and patient characteristics in chronic obstructive pulmonary disease (COPD). Thorax 1995;50:249-53. 38 Baggett HC, Hennessy TW, Rudolph K, Bruden D, Reasonover A, Parkinson A, et al. Community-onset methicillin-resistant Staphylococcus aureus associated with antibiotic use and the cytotoxin Panton-Valentine leukocidin during a furunculosis outbreak in rural Alaska. J Infect Dis 2004;189:1565-73. 39 Campbell KM, Vaughn AF, Russell KL, Smith B, Jimenez DL, Barrozo CP, et al. Risk factors for community-associated methicillin- resistant Staphylococcus aureus infections inanoutbreakof disease among military trainees in San Diego, California, in 2002. J Clin Microbiol 2004;42:4050-3. 40 LoWT, LinWJ, Tseng MH, LuJJ, Lee SY, ChuML, et al. Nasal carriage of a single clone of community-acquired methicillin-resistant Staphylococcus aureus among kindergarten attendees in northern Taiwan. BMC Infect Dis 2007;7:51. RESEARCH page 10 of 11 BMJ | ONLINE FIRST | bmj.com 41 Paganini H, la Latta MP, Muller OB, Ezcurra G, Uranga M, Aguirre C, et al. [Community-acquired methicillin-resistant Staphylococcus aureus infections in children: multicenter trial]. Arch Argent Pediatr 2008;106:397-403. 42 Toltzis P, Dul M, ORiordan MA, Toltzis H, Blumer JL. Comparative effects of single-dose ceftriaxone versus three oral antibiotic regimens onstool colonizationby resistant bacilli inchildren. Pediatr Infect Dis J 2007;26:25-30. 43 Chern KC, Shrestha SK, Cevallos V, Dhami HL, Tiwari P, Chern L, et al. Alterations in the conjunctival bacterial flora following a single dose of azithromycin in a trachoma endemic area. Br J Ophthalmol 1999;83:1332-5. 44 Sethi S. Bacteria in exacerbations of chronic obstructive pulmonary disease: phenomenon or epiphenomenon? Proc AmThorac Soc 2004;1:109-14. 45 Miller R, Walker AS, Knox K, Wyllie D, Paul J, Haworth E, et al. Feral and wild-type methicillin-resistant Staphylococcus aureus in the United Kingdom. Epidemiol Infect 2009;1-11. Accepted: 19 March 2010 RESEARCH BMJ | ONLINE FIRST | bmj.com page 11 of 11