Cochrane Review: Different antibiotic treatments for group A streptococcal pharyngitis
Description of the condition
Pharyngitis is a common upper respiratory tract infection. Antibiotics are often prescribed to treat this condition. Patients usually consult a physician with the complaint of sore throat. A previous Cochrane review comparing the effect of antibiotics to placebo in participants with or without group A beta-haemolytic streptococci (GABHS) sore throat (Del Mar 2009) pointed to the self-limiting nature of an acute sore throat (even in case of positive GABHS culture). Antibiotics provide only modest benefit when prescribed for the condition 'sore throat'. The effect of antibiotic treatment was increased in participants with positive throat swabs for GABHS. The streptococci-positive participants are only a small proportion of all participants with 'sore throat'. Nevertheless, in many countries antibiotics are prescribed for most people who have a 'sore throat' (Cars 2001; Linder 2001). Given the high consumption of antibiotics for this condition a rational approach would be to reserve a treatment with antibiotics for participants with proven presence or a high likelihood of group A streptococci (Cooper 2001; Snow 2001). But clinical scoring systems are somewhat limited in their ability to correctly target GABHS positive patients (McIsaac 1998) and the usefulness of rapid assay tests depends on the prevalence of GABHS in the population (Sonnad 1999) and justification of its cost-effectiveness is unclear (Gerber 2004; Neuner 2003).
Description of the intervention
The slight benefit of treatment with antibiotics in patients with GABHS sore throat may be considered relevant. When antibiotics are indicated a choice needs to be made. In that case several aspects need to be considered, such as comparative benefit-harm balance, costs and local antimicrobial resistance patterns. Many guidelines recommend penicillin as a first choice, with erythromycin preferred for penicillin-allergic participants (Cooper 2001; Snow 2001). To date, resistance of GABHS to penicillin has not been documented (Gerber 2009) and resistance to erythromycin is still low (Cooper 2001). Considering the growing problem of antibiotic resistance for other pathogens, this responsiveness of group A streptococci should not be endangered (Wise 1998). Penicillin and erythromycin are cheap and the most cost-effective option. In spite of this, physicians continue to prescribe broad-spectrum antibiotics, including recently marketed ones. It is not clear if these antibiotics have any substantial clinical benefit over penicillin (and erythromycin).
Why it is important to do this review
Internationally, guidelines recommend using penicillin as first choice when choosing to treat acute sore throat (suspected to be caused by GABHS) with antibiotics (Matthys 2007). However, some argue that cephalosporins are more effective and should therefore be preferred (Casey 2004). Many physicians argue that occurrence of penicillin allergy should be taken into account when making a choice for an antibiotic. This review looked for evidence of penicillin allergy occurring in the available trials. In addition, in the presence of documented penicillin allergy, the side effect profile of the eligible antibiotics can guide choice. Therefore, in order to provide healthcare providers with sufficient information to make an evidence-based choice, both treatment benefits and adverse events are compared.
Description of studies
See: Characteristics of included studies; Characteristics of excluded studies.
We retrieved 136 search results from our electronic searches. A total of 60 trials were considered for the review. One additional trial was identified through a Google search (Muller 1992). Of these, 18 met the predefined inclusion criteria. Two of the 18 papers reported different outcomes of the same study and are thus considered as one single study (Norrby 2002). Thus, 17 trials were included in the review.
Most of the included trials were conducted in the 1990s, two in the 1980s (Henness 1982a; Randolph 1985) and two in the 1970s (Jackson 1973; Trickett 1973). Only one trial was more recent (Norrby 2002).
Contacting pharmaceutical companies did not result in any additional published or unpublished data (only one company replied), neither did contacting authors or experts in the field.
All included studies compared penicillin with another antibiotic class. None of the identified studies compared macrolides with antibiotics other than penicillin.
The included trials investigated a total of 5352 participants with acute GABHS tonsillopharyngitis. The age of participants ranged from one month to 80 years. Seven trials included only or predominantly paediatric participants (Disney 1992a; Disney 1992b; Henness 1982a; Henness 1982b; O'Doherty 1996; Randolph 1985; Reed 1991), nine trials included participants who were at least 12 years or older (Bachand 1991; Carbon 1995; Levenstein 1991; McCarty 1992a; Nemeth 1999; Norrby 2002; Stein 1991; Trickett 1973; Watkins 1997). Two other trials included a wider range of participants aged one month or older. In the study by Reed approximately 80% of participants were under 15 years of age (Reed 1991) and therefore included in the subgroup analysis for children. In Muller 1992 90% of participants were over 12 years old and as results were not stratified per age group this study was included in the adult subgroup analysis.
All trials included only participants with confirmed acute GABHS tonsillopharyngitis. Confirmation of the presence of GABHS in participants with clinical signs of tonsillopharyngitis was in most cases performed first by a rapid immunoassay test and then reconfirmed with a throat culture. In five trials the confirmation of GABHS tonsillopharyngitis was carried out only by a throat culture (Henness 1982a; Henness 1982b; Jackson 1973; Randolph 1985; Trickett 1973) and in two trials only with a rapid immunoassay test (O'Doherty 1996; Stein 1991). All but one trial reported on clinical outcomes. Trickett 1973 only reported bacteriological outcomes, but was included in the meta-analysis on adverse effects.
Clinical outcomes, in most studies defined as complete resolution of signs and symptoms (Characteristics of included studies) were assessed at various time points, but mostly measured between five to 10 days following the end of antibiotic treatment. Therefore, post-treatment the outcome "post-treatment clinical efficacy" (i.e. assessment of signs and symptoms after completion of the treatment course) was pooled. One trial reported clinical effect within the first 24 hours of treatment (Randolph 1985). Three trials reported on specific symptoms, such as sore throat and fever (Bachand 1991; Levenstein 1991; Randolph 1985). None of the trials reported data on the duration of illness.
Ten trials reported the incidence of clinical relapse (Bachand 1991; Carbon 1995; Disney 1992a; Disney 1992b; McCarty 1992a; Muller 1992; Nemeth 1999; O'Doherty 1996; Reed 1991; Stein 1991). The definition of clinical relapse varies slightly between trials; from "pretreatment signs & symptoms resolved but reappeared" Bachand 1991; Carbon 1995; Disney 1992b; Levenstein 1991; McCarty 1992a; Muller 1992; Nemeth 1999; Norrby 2002; Stein 1991) or "initial improvement or alleviation of symptoms, but subsequent worsening or recurrence" (McCarty 1992a; Watkins 1997) to "new infection with different serotype" (Disney 1992a). One study defined clinical cure as "clinical improvement within first 24 hours of therapy and all follow-up cultures no S. pyogenes" (Henness 1982a). Two studies used the physician's assessment of symptoms as outcome (Randolph 1985; Reed 1991).
Four trials reported complications occurring during longer follow-up (Carbon 1995; Jackson 1973; McCarty 1992a; Muller 1992). Fifteen trials mentioned adverse effects reported during treatment.
The use of antipyretic analgesics was allowed in four trials (Bachand 1991; Disney 1992b; Muller 1992; Watkins 1997), prohibited in two (Carbon 1995; Randolph 1985) and not stated in the other 11 trials.
Forty-nine trials were excluded from analysis. The most common reason for exclusion (35 trials) was no or inadequate blinding (Adam 1994; Adam 1995; Adam 1996; Adam 2000a; Adam 2000b; Adam 2001; Aujard 1995; Cohen 2002; Denny 1953; Dykhuizen 1996; Esposito 2002; Feder 1999; Gerber 1986; Gooch 1993; Hamill 1993; Holm 1991; Howe 1997; Lennon 2008; McCarty 1992b; McCarty 1994; Milatovic 1991; Milatovic 1993; Pacifico 1996; Perkins 1969; Pichichero 2000; Pichichero 2008; Portier 1990; Portier 1994; Sakata 2008; Shapera 1973; Shvartzman 1993; Stillerman 1986; Tack 1997; Tack 1998; Uysal 2000). Six trials did not compare at least two different classes of antibiotics (Breese 1974; Disney 1979; Matsen 1974; McIsaac 2004; Siegel 1961; Zwart 2000). In two trials the included participants did not exclusively have acute GABHS tonsillopharyngitis (Davies 1995; Standaert 1997) and one trial included patients with recurrent tonsillitis (Roos 1997). One trial did not report any clinical outcomes (Gerber 1999) and four trials were not randomised controlled trials (RCTs) (Del Mar 2008; De Meyere 1992; Granizio 2008; Haverkorn 1971).
Risk of bias in included studies
Risk of bias assessment is reported in the Characteristics of included studies table. Only three trials (Disney 1992a; Norrby 2002; Randolph 1985) reported an ITT analysis for the efficacy outcomes. One trial reported carrying out an ITT analysis, but post-randomisation exclusions were not included in the efficacy analysis (Carbon 1995). All trial authors used an ITT analysis for adverse effects.
All trials were randomised, but only three described the method of randomisation and/or allocation concealment (Jackson 1973; Randolph 1985; Watkins 1997).
Incomplete outcome data
The post-randomisation dropout rate was high in most trials. In 11 trials (Bachand 1991; Henness 1982a; Jackson 1973; Levenstein 1991; McCarty 1992a; Nemeth 1999; Muller 1992; Norrby 2002; O'Doherty 1996; Stein 1991; Watkins 1997) the proportion of dropouts was more than 20%, ranging from 21.5% (McCarty 1992a) to 48.5% (Levenstein 1991). In the outcome analysis most trials included only participants with complete outcome data. This may have had an important impact on the effect measured. Only three trials performed an ITT analysis with all randomised participants included in the analysis of the clinical outcome (Disney 1992a; Norrby 2002; Randolph 1985). These three trials all have minimal to no dropouts (0 or 1 participant).
Most trials reported a composite outcome of "clinical cure". This is a relevant outcome for clinical practice, but the definition of cure may have differed in the included trials.
Other potential sources of bias
Ten trials reported that they were sponsored by a pharmaceutical company (Disney 1992b; Jackson 1973; McCarty 1992a; Muller 1992; Nemeth 1999; Norrby 2002; Randolph 1985; Reed 1991; Trickett 1973; Watkins 1997). Authors of six trials were reported as employees of a pharmaceutical company (Bachand 1991; Disney 1992b; Henness 1982a; Henness 1982b; Nemeth 1999; Watkins 1997); and in three of those trials the employing pharmaceutical company was not reported as a funding source (Bachand 1991; Henness 1982a; Henness 1982b). The remaining five trials did not mention their funding source.
Six trials mentioned that ethics approval was obtained for the study (Bachand 1991; Levenstein 1991; Muller 1992; Nemeth 1999; Norrby 2002; O'Doherty 1996) and seven trials reported- that informed consent was obtained from participants or guardians (Levenstein 1991; McCarty 1992a; Muller 1992; Nemeth 1999; Norrby 2002; O'Doherty 1996; Reed 1991).
Effects of interventions
1. Cephalosporin versus penicillin
Resolution of symptoms
Six trials (Carbon 1995; Disney 1992a; Nemeth 1999; Henness 1982a; Randolph 1985; Reed 1991) reported on the resolution of symptoms at various points in time.
Five trials measured resolution of symptoms at the end of treatment (two to 15 days or more post-treatment), two trials in adults (Carbon 1995; Nemeth 1999) and three in children (Disney 1992b; Henness 1982a; Reed 1991). The ITT analysis included 2018 participants and showed no difference between treatments (OR 0.79; 95% CI 0.55 to 1.12). The effect in adults (N = 2; n = 1163; OR 0.78; 95% CI 0.60 to 1.01) was similar to that in children (N = 3; n = 855; OR 0.83; 95% CI 0.40 to 1.73).
The result of the analysis of evaluable participants only (n = 1660) showed an effect in favour of treatment with cephalosporins (OR 0.51; 95% CI 0.27 to 0.97; absolute risk difference (ARD) 0.05; NNTB 20). However, the estimates of effect in adults (N = 2; n = 880; OR 0.56; 95% CI 0.24 to 1.32) and in children (N = 3; n = 780; OR 0.46; 95% CI 0.14 to 1.52) analysed separately revealed no statistically significant differences between treatment groups.
We analysed the studies with reported pharmaceutical company sponsorship separate from the studies that did not mention any industry involvement for the outcome resolution of symptoms post-treatment. The two studies that did not report their funding source (Carbon 1995; Disney 1992a) showed a statistically significant effect in favour of cephalosporins (OR 0.47; 95% CI 0.27 to 0.81; ARD 0.02; NNTB 50). The sponsored studies pooled together (Henness 1982a; Nemeth 1999; Reed 1991) did not result in a significant difference between the two groups of antibiotics.
One trial in children (n = 138) also reported the resolution of symptoms within 24 hours of treatment (Randolph 1985) and found no difference between treatment groups (OR 0.97; 95% CI 0.34 to 2.74).
One trial in children (Randolph 1985) found no difference between treatment groups for resolution of sore throat (n = 138; OR 0.97; 95% CI 0.23 to 4.04).
One trial in children (Randolph 1985) found no difference between treatment groups for resolution of fever (n = 138; OR 0.97; 95% CI 0.19 to 4.98).
Incidence of relapse
In four trials (n = 1386) that reported the incidence of clinical relapse in evaluated participants (Carbon 1995; Disney 1992a; Nemeth 1999; Reed 1991) there was a benefit of treatment with cephalosporins over penicillin in the total population (OR 0.55; 95% CI 0.31 to 0.99; ARD 0.02; NNTB 50). This was due to a difference in the two trials in adults (Carbon 1995; Nemeth 1999) (n = 770; OR 0.42; 95% CI 0.20 to 0.88; ARD 0.03; NNTB 33.3). There was no difference in the two trials in children (Disney 1992a; Reed 1991) (n = 616; OR 0.89; 95% CI 0.33 to 2.43).
In one trial in adults (Carbon 1995) no complications were reported in the cephalosporin group (119 participants) or the penicillin group (125 participants).
Three trials in adults reported the incidence of adverse effects (Carbon 1995; Nemeth 1999; Reed 1991). There was significant heterogeneity between the trials. In the cephalosporin group 212 of 788 participants reported adverse events, compared to 87 of 491 in the penicillin group. There was no difference between the two treatments (OR 0.99; 95% CI 0.31 to 3.16).
2. Macrolide versus penicillin
Resolution of symptoms post-treatment
Five trials in adults (Bachand 1991; Levenstein 1991; Norrby 2002; Stein 1991; Watkins 1997) and one trial in children (O'Doherty 1996) investigated the resolution of symptoms at various points in time post-treatment. In the ITT analysis of 1728 participants there were no differences between the two treatment groups (OR 1.11; 95% CI 0.92 to 1.35). The estimate of effect in adults (N = 5; n = 1239; OR 1.07; 95% CI 0.86 to 1.34) was similar to that in the trial in children (n = 489; OR 1.25; 95% CI 0.85 to 1.84). The analysis of evaluable participants only did not result in any significant differences between treatment groups (n = 1159; OR 0.79; 95% CI 0.57 to 1.09). The estimate for the five trials in adults (n = 801) was OR 0.88; 95% CI 0.59 to 1.31 and one trial in children (n = 358) was OR 0.64; 95% CI 0.36 to 1.11.
ITT analysis of pharmaceutical industry sponsored trials versus trials that did not report their funding source does not show significant differences in results.
Two trials in adults (n = 371) reported the resolution of sore throat in adults (n = 371) (Bachand 1991; Levenstein 1991) and found no difference between the two treatments (OR 0.97; 95% CI 0.64 to 1.46).
Resolution of fever at two to 10 days post-treatment was reported in two trials with 371 adult participants (Bachand 1991; Levenstein 1991). All participants in both groups were free of fever at the time they were evaluated (45 participants in the macrolide group and 39 in the penicillin group; OR 1.05; 95% CI 0.69 to 1.59).
Incidence of relapse
Incidence of clinical relapse was evaluated in six trials, five trials in adults (Bachand 1991; Levenstein 1991; Norrby 2002; Stein 1991; Watkins 1997) and one in children (O'Doherty 1996). Twenty-two of 441 participants in the macrolide group and 16 of 361 in the penicillin group reported a relapse at day 15 to 56 post-treatment. The difference was not statistically significant (OR 1.21; 95% CI 0.48 to 3.03).
In the six trials (n = 1727), five in adults and one in children (O'Doherty 1996), that reported on the incidence of adverse events, there were no statistically significant differences between the treatment groups: 282 events were reported in the macrolide group and 251 in the penicillin group (OR 1.19; 95% CI 0.82 to 1.73). In the trial in children (n = 489) macrolides seemed to cause more adverse events than penicillin (OR 2.33; 95% CI 1.06 to 5.15; NNTH 17.2).
3. Carbacephem versus penicillin
Three trials are included in this comparison (n = 795): one in children (Disney 1992b), one in adults (McCarty 1992a) and one in a population of adults and children (but predominantly adults as 90% were older than 12 years) (Muller 1992).
Resolution of symptoms post-treatment
In the ITT analysis more participants reported resolution of symptoms in the carbacephem group than in the penicillin group (n = 795; OR for absence of symptom resolution post-treatment 0.70; 95% CI 0.49 to 0.99; ARD 0.07; NNTB 14.3). In adult participants there was no difference (n = 562; OR 0.75; 95% CI 0.46 to 1.22) and in children there was a beneficial effect of carbacephem (n = 233; OR 0.57; 95% CI 0.33 to 0.99; ARD 0.12; NNTB 8.3). The analysis of evaluable participants showed no differences between treatment groups (n = 602; OR 0.62; 95% CI 0.38 to 1.01).
Incidence of relapse
There were no differences in the incidence of clinical relapse between groups treated with carbacephem or penicillin (21 events in 267 participants treated with carbacephem and 16 in 256 participants treated with penicillin; OR 1.27; 95% CI 0.64 to 2.50).
There were no differences in reported adverse events between the treatments (75 events reported in 396 participants treated with carbacephem and 71 in 399 participants treated with penicillin; OR 1.08; 95% CI 0.75 to 1.55). Muller reported that one participant was hospitalised for surgical drainage of a tonsillar abscess in the group treated with loracarbef one day after initiating therapy (Muller 1992).
4. Clindamycin versus ampicillin
One trial compared treatment with clindamycin to ampicillin (Jackson 1973) (n = 314). The only outcome reported is adverse events. Six participants reported adverse events in the group treated with clindamycin (156 participants) and 14 participants experienced adverse events in the ampicillin group (158 participants). The difference was not statistically significant (OR 0.41; 95% CI 0.15 to 1.10). No other clinical outcomes were reported.
5. Sulfonamide versus penicillin
One trial in adults was included in this comparison (Trickett 1973). It reported only on adverse events (eight events reported in participants treated with sulphonamides and six events in the penicillin group) and found no difference between sulphonamide and penicillin (OR 1.37; 95% CI 0.43 to 4.34).
6. Penicillin allergy
Muller reports that one patient developed a rash and one patient experienced vomiting, both attributed to use of penicillin (although the patient was then successfully switched to amoxicillin/clavulanate). However, in the loracarbef group also one participant discontinued treatment because of a rash (Muller 1992). None of the other included trials reported on penicillin allergy.
Summary of main results
Our meta-analysis shows that there is generally no strong evidence for clinically important differences in clinical outcomes when comparing different classes of antibiotics with penicillin in adults and children with pharyngitis caused by GABHS.
Resolution of symptoms
Intention to treat (ITT) analysis does not show any difference in resolution of symptoms between cephalosporins and penicillin. When only evaluable participants are included in the analysis (i.e. participants for whom an outcome was known) there seems to be a benefit of cephalosporins over penicillin with regard to resolution of symptoms after treatment (NNTB 20). Subgroup analysis of adults and children did not reveal any significant differences, but this can be attributed to lack of sufficient power.
ITT analysis of the comparison between carbacephem and penicillin showed a benefit of carbacephem with regard to resolution of symptoms after treatment with a NNTB of 14.3. There is no significant benefit in the (large) adult subgroup, and the effect may thus be largely based on an observed effect in children (NNTB 8.3). The analysis of evaluable participants only does not reach statistical significance (but the estimated NNTB is likely to be high).
Other comparisons with penicillin (macrolides or clindamycin or sulfonamides) did not report clinical outcomes for this meta-analysis.
The incidence of relapse in evaluable participants seems to be lower in participants treated with cephalosporins compared with penicillin, but the event rate is low (approximately 3.5%) and the NNTB is quite high (NNTB 50). There were no differences in relapse rate between other antibiotics and penicillin.
Adverse events occurred at a similar rate in all treatment groups, except children treated with macrolides seemed to experience more adverse events than children treated with penicillin (although this difference was not statistically significant, most likely due to insufficient power).
The results of our meta-analysis are not clear cut and need to be discussed in the context of morbidity (including serious complications) prevalence, concerns for rising antibiotic resistance and economic constraints in all healthcare systems.
Overall completeness and applicability of evidence
Although we have searched several databases and scrutinised all references listed in identified reviews and publications of trials, we may have missed some trials. We have contacted experts and pharmaceutical companies. One pharmaceutical company responded, but this did not result in additional data. This additional search did not yield any new published or unpublished trials. As an analysis of unpublished data used in Cochrane Reviews suggests that generally, searching for unpublished data does not uncover new data that are important to the conclusion of the review (van Driel 2009), the lack of unpublished data may not have had an important impact on the results of our review.
Our meta-analysis focuses on clinical outcomes only. Reviews that report bacteriological outcomes point to the superiority of cephalosporins over penicillin with regard to eradication of GABHS (Brunton 2006; Casey 2004). However, this does not take the clinical presentation into account. Gerber et al found no difference in bacteriologic treatment success rates between cefadroxil and penicillin groups among participants classified clinically as likely to have true GABHS pharyngitis, but cephalosporins seemed to be more successful in eradicating GABHS is patients classified clinically as likely to be streptococcal carriers (Gerber 1999). Contamination of treatment groups by such chronic GABHS carriers contributes to the apparent superiority of cephalosporins in studies focusing on bacteriological outcomes (Shulman 2004). This is of very limited clinical relevance. To our knowledge chronic streptococcal carriage is not linked to higher risk of developing GABHS pharyngitis and hence eradication of streptococci in carriers is not a treatment goal. Information on complications is scarcely reported and therefore we cannot draw any conclusions concerning this outcome.
Quality of the evidence
A strong point of our review is that we included only randomised and double-blinded trials. This was intended to minimise the risk of bias related to selection of participants and reporting of outcomes. However, in spite of the lower risk of bias due to methodology, reporting of the findings and transparency of the analyses in the trials were often unsatisfactory. Patient characteristics were poorly reported and outcomes poorly or not at all defined. Dropout rates in some studies were very high (> 20%).
The overall risk of bias in included studies is difficult to assess because the process of randomisation and blinding is not described in most studies. For instance, only four studies (Jackson 1973; Randolph 1985; Reed 1991; Watkins 1997) described the method used to conceal allocation.
It is surprising that "resolution of sore throat", a key symptom in GABHS pharyngitis, is only reported as a separate outcome in one study (McCarty 1992a). Most studies assess the "whole clinical picture" of the clinical presentation of pharyngitis, which is a combination of symptoms including sore throat, fever and feeling unwell. Assessment of the effect of antibiotics on the full range of signs and symptoms is therefore clinically relevant.
Potential biases in the review process
Pooling of the different outcomes is hampered by the differences of outcome definitions across studies. As most trials measure clinical outcome within two weeks of the end of antibiotic treatment, they were pooled for the outcome 'resolution of symptoms post-treatment'. The trial that reported symptom resolution within the first 24 hours of treatment is considered separately. Very few trials report on specific symptoms related to acute GABHS tonsillopharyngitis. As 'symptom resolution' is a subjective outcome, the interpretation may be different across trials and pooling may therefore be inappropriate. However, differences between comparison groups in the same trial will not be affected (as they are measured in the same population).
The results of this meta-analysis are based on ITT analysis of the selected outcomes. However, this may underestimate the efficacy of treatment. Most trials reported the number of participants randomised, but included only the evaluated participants in the outcome analysis. When reported, a common reason for post-randomisation exclusion is a negative throat culture, suggesting that another pathogen caused the signs and symptoms of acute tonsillopharyngitis. Including these GABHS-negative participants in the analysis could bias the results if exclusion is not similar in both treatment groups. Some trials reported exclusions per group and show that this is not the case. When comparing two efficacious treatments this potential underestimation does not seem relevant as it will not influence the conclusions. However, for the trials that do not report this, it is not possible to know if selective exclusions occurred. We checked if the method of analysis influenced outcome by performing both ITT and analysis of evaluable participants for the outcome resolution of symptoms post-treatment. This showed different results in two comparisons. When comparing cephalosporins and penicillin, ITT analysis for this outcome yielded a non-significant result, whereas analysis of evaluable participants showed a benefit of cephalosporins over penicillin. The opposite occurred in the analysis of effect on the same outcome in participants treated with carbacephem versus penicillin; where ITT analysis showed a statistically significant difference and the evaluable participants analysis did not, most likely due to a reduction in the number of participants included in the analysis (resulting in reduced statistical power). Analysing only evaluable participants implies a high risk of bias as there may have been a selective dropout. On the other hand, the ITT analysis can be considered as a conservative estimate of the true effect.
The estimated odds ratios suggest that large benefits can be expected when treating patients with cephalosporins or carbacephems. But these supposedly impressive effects expressed as a relative measure of risk (expressed as an OR) do not always translate into a clinically meaningful difference. For example, the estimated OR of 0.55 for the incidence of relapse in cephalosporins compared with penicillin, suggests that the risk of relapse could be halved by treating patients with cephalosporins. However, the associated absolute risk difference is 0.02, resulting in a NNTB of 50, which means that 50 patients need to be treated with broad-spectrum more expensive antibiotics to prevent one additional relapse.
Calculating the absolute risk difference and the NNTB is therefore a useful method to assess the clinical importance of a relative risk. The interpretation of the NNTBs (how many patients needed to treat is acceptable) is however, not clear cut and depends on assessment of benefit and harm and also cost-effectiveness.
All the trials in our review were performed in high-income countries. The incidence of suppurative and other complications (which are rare in high-income countries) as well as antimicrobial resistance rates may be different in low-income countries or specific communities with high prevalences of GABHS tonsillitis (Hanna 2010). Therefore, studies performed in low-income and high prevalence communities are needed.
Agreements and disagreements with other studies or reviews
Our review shows that although there seems to be some benefit of antibiotics with a wider spectrum, i.e. cephalosporins and carbacephem, this observed effect is not consistent across analysis methods and studied subgroups. Cephalosporins show a benefit regarding resolution of symptoms only in the analysis of evaluable participants and carbacephem is superior to penicillin for this outcome only in the ITT analysis (attributable to an effect in children treated with a carbacephem). The NNTBs associated with the observed effects are relatively high (20 for treatment with cephalosporins compared with penicillin), except perhaps for the effect of carbacephem in children (NNTB 8.3). There is no clinically meaningful difference between penicillin and the other classes of antibiotics that have been studied with regard to rate of clinical relapse. However, cephalosporins seem to reduce relapse rate (NNTB 50), especially in adults (NNTB 30).
The effects observed in cephalosporins and carbacephems and not in the other antibiotic classes can be explained by the fact that although they are considered different classes of antibiotics, carbacephems chemically closely resemble cephalosporins (Cooper 1992).
Interpretation of these findings for clinical practice is not straightforward. One could argue that our meta-analysis points to a superior efficacy of cephalosporins over penicillin, especially in adults where the upper limit of the 95% CI is 1.01 (P = 0.06) in the ITT analysis. The population size may not have been large enough to reach statistical significance. This finding is in line with an earlier meta-analysis concluding that cephalosporins are superior to penicillin in treating GABHS pharyngitis and therefore cephalosporins should be considered first choice (Casey 2004). But in our meta-analysis the absolute difference between the two groups (cephalosporin or penicillin) although not statistically significant is only 2.5% which implies a NNTB of 40. Treating 40 patients with cephalosporins instead of penicillin would incur an additional cost to the healthcare system as well as add to the risk of developing antibiotic resistance, especially in broad spectrum antibiotics such as cephalosporins.
The observed superior effect of cephalosporins in reducing the rate of relapse has also been reported in another meta-analysis (Casey 2004). However, in our meta-analysis it is only observed in adults and may be biased by the rather liberal definition of relapse in the study that accounts for 49% of the weighting in the meta-analysis (Nemeth 1999); "worsening of, or absence of significant remission of, signs & symptoms 17 to 24 days post-therapy or need for further AB therapy", whereas in other studies "recurrence of symptoms" after initial remission was required. The NNTB of 33 participants that need to be treated with cephalosporins rather than penicillin to prevent one participant experiencing relapse illustrates the limited clinical relevance of this statistically significant result.
How can the differences between Casey's meta-analysis and ours be explained? Casey included 35 trials; two thirds of those were not blinded and reporting of randomisation and losses to follow up was very poor implying a high risk of bias (Gerber 2004b). By restricting the inclusion to double-blinded trials we ruled out one source of potential bias and improved the methodological rigour of the meta-analysis. Casey's subgroup analysis of double-blinded studies generated an OR similar to ours (although with a much narrower CI, OR 0.43; 95% CI 0.25 to 0.71), but it included studies with carbacephems, which have been advertised as a separate class of antibiotics (Cooper 1992). Casey reports an analysis of evaluable patients, whereas ITT analysis may be more appropriate especially with important numbers of dropouts (which is the case in many of the trials included in our meta-analysis). The trial populations included in Casey's review, as in ours, may have been contaminated with chronic carriers of GABHS who had intercurrent viral pharyngitis (Gerber 2004b) but it is not clear if this has implications for clinical practice.
We found no differences in the incidence of adverse events and data on long-term follow up and the occurrence of complications was insufficient. Therefore, costs and antimicrobial resistance patterns are important in making a choice.
Implications for practice
Although there seem to be indications that carbacephems and cephalosporins might have some benefit over penicillin in terms of resolution of symptoms and incidence of relapse, the findings are inconsistent across analysis methods and the NNTB are substantial. This is insufficiently convincing evidence to alter current guideline recommendations for the treatment of patients with GABHS tonsillopharyngitis. Moreover, we found no clinically important differences in occurrence of adverse events and data on the incidence of complications are too scarce to draw conclusions.
Antibiotics have a limited effect in the treatment of patients with acute sore throat, even in the presence of GABHS. However, if antibiotics are to be prescribed, based on these results and taking into consideration the costs and antimicrobial resistance patterns of the different antibiotics, penicillin can still be considered first choice in both adults and children.
Implications for research
The observed differences in clinical efficacy between adults and children needs further exploration. Prevention of serious complications such as acute rheumatic fever and acute glomerulonephritis are often mentioned as arguments in favour of antibiotic use. However, the current data do not provide information about the impact of different antibiotics on the prevention of complications. Further studies with longer follow up might be able to address this issue. As these complications seem to be more prevalent in low-income and high-risk communities (for example, Australian Indigenous communities), studies in these specific high-risk communities are needed. Economic analysis of the cost-effectiveness of the different treatment options can provide additional guidance for making a choice.
We thank the Cochrane Acute Respiratory Infections Review Group and, in particular, Liz Dooley, Sarah Thorning and Chris Del Mar for their support. We thank Warren McIsaac, Amy Zelmer, Mark Jones and Paul Little for their valuable comments.
Appendix 1. Embase.com search strategy
28. #24 AND #27
27. #25 OR #26
26. random*:ab,ti OR placebo*:ab,ti OR factorial*:ab,ti OR crossover*:ab,ti OR 'cross over':ab,ti OR 'cross-over':ab,ti OR assign*:ab,ti OR allocat*:ab,ti OR volunteer*:ab,ti OR ((singl* OR doubl*) NEAR/2 (mask* OR blind*)):ab,ti
25. 'randomized controlled trial'/exp OR 'single blind procedure'/exp OR 'double blind procedure'/exp OR 'crossover procedure'/exp
24. #20 AND #23
23. #21 OR #22
22. antibiotic*:ab,ti OR antibacterial*:ab,ti OR (anti NEAR/1 bacterial*):ab,ti
21. 'antibiotic agent'/exp
20. #1 OR #2 OR #3 OR #4 OR #5 OR #6 OR #7 OR #8 OR #9 OR #10 OR #11 OR #12 OR #19
19. #17 AND #18
17. #13 OR #14 OR #15 OR #16
15. ('group a beta haemolytic' NEAR/1 streptococc*):ab,ti
14. ('group a beta hemolytic' NEAR/1 streptococc*):ab,ti
13. 'streptococcus infection'/de OR 'group a streptococcal infection'/de
12. (strep* NEAR/3 throat*):ab,ti
11. (sore NEAR/1 throat*):ab,ti
10. 'sore throat'/exp
8. 'streptococcal pharyngitis'/exp
Protocol first published: Issue 3, 2003
Review first published: Issue 10, 2010
| 6 October 2010 || Amended || Contact details updated. |
| 31 August 2008 || Amended || Converted to new review format. |
Contributions of authors
MVD wrote the protocol. All authors contributed to final editing of the protocol.
MVD and NK selected trials.
MVD and NK independently performed quality assessment.
MVD and NK performed data extraction with support from ADS. MVD analysed the data.
MVD wrote the draft review and addressed the reviewers' comments. All review authors contributed to the discussion and the editing.