Antibiotics for sore throat

Sore throat is a very common reason for people to seek medical care (ABS 1985). Moreover, four to six times as many people suffering sore throat do not seek care (Horder 1954; Goslings 1963). Sore throat is a disease that remits spontaneously, that is, 'cure' is not dependent on treatment (Del Mar 1992c). Nonetheless primary care doctors commonly prescribe antibiotics for sore throat and other upper respiratory tract infections. There are large differences in clinical practice between countries (Froom 1990) and between individual primary care doctors (Howie 1971).

Whether or not to prescribe antibiotics for sore throat is controversial. The issue is important because it is a very common disease. Therefore differences in prescribing result in large cost differences. Moreover, increased prescribing increases patient attendance rates (Howie 1978; Little 1997).

To assess the benefits of antibiotics in the management of acute sore throat.

We searched the following electronic databases:
The Cochrane Central Register of Controlled Trials (CENTRAL) (The Cochrane Library, issue 2, 2003)
MEDLINE WebSPIRS (January 1966 to May week 1, 2003)
EMBASE WebSPIRS (January 1990 to March 2003)

The following search strategies were run for MEDLINE and EMBASE , and combined with the Cochrane highly sensitive search strategy phases one and two as published in appendix 5c of the Cochrane Reviewers' Handbook (Clarke 2003).
MEDLINE (WebSPIRS)
#1 explode 'Pharyngitis-' / all subheadings in MIME,MJME
#2 pharyngit*
#3 explode 'Tonsillitis-' / all subheadings in MIME,MJME
#4 tonsillit*
#5 Sore adj throat*
#6 #1 or #2 or #3 or #4 or #5
#7 explode 'Antibiotics-' / all subheadings in MIME,MJME
#8 antibiot*
#9 #7 or #8
#10 #6 and #9

EMBASE (WebSPIRS)
#1 explode 'pharyngitis-' / all subheadings
#2 pharyngit*
#3 explode 'tonsillitis-' / all subheadings
#4 tonsillit*
#5 explode 'sore-throat' / all subheadings
#6 sore adj throat*
#7 #1 or #2 or #3 or #4 or #5 or #6
#8 explode 'antibiotic-agent' / all subheadings
#9 antibiot*
#10 #8 or #9
#11 #7 and #10

References of selected studies and relevant reviews were hand-checked to find additional studies. No language restrictions were applied.

Two reviewers independently screened the abstracts of potential studies, and retrieved the full articles for those that were trials. The two reviewers then examined the full articles, and either selected for inclusion or rejected, resolving differences of opinion by discussion. See the characteristics of excluded studies table.

Two reviewers independently extracted data based on the complications and symptoms listed above from the included studies. Data extraction involved reading from tables, graphs and, in some cases, contacting the study authors to request raw data (Dagnelie 1996; Little 1997; Zwart 2000).

We considered 57 studies for inclusion in the review. Of these, 26 controlled studies met our inclusion criteria and were included in the review (see the characteristics of included studies table). The most common reason for exclusion was lack of appropriate control group (n = 13). Other reasons for exclusion were: irrelevant or non-patient centred outcomes (n = six), main complaint other than acute sore throat (six), inappropriate or no randomisation to treatment (n = five), or that the study reported previously published data already included (n = one).

The included studies investigated a total of 12,669 cases of sore throat. The majority of studies were conducted in the 1950s, during which time the rates of serious complications (especially acute rheumatic fever) were much higher than today. Six recent studies were included (1996 to 2000), perhaps signalling renewed interest in this topic.

The age of participants ranged from less than one year to greater than 50 years. The participants of eight early studies were young male recruits from the United States air force. Six of the remaining studies recruited children up to 18 years of age only, three recruited only adults or adolescents aged 15 years or over, and eight studies had no age restrictions.

All studies recruited patients presenting with symptoms of sore throat. Sixteen studies did not distinguish between bacterial and viral aetiology, however eight studies included GABHS positive patients only, whilst two studies excluded patients who were GABHS positive.

Many of the studies were of poor quality. Only 17 studies were double blind: three were single blind. In most early studies subjects were randomised to treatment and control groups by methods that could potentially introduce bias (for example, air force serial number, drawing a card from a deck, hospital bed number) or not randomised at all. The generalisability of studies can be questioned. In five studies subjects were excluded if they did not yield a positive throat-swab culture for Group A Beta Haemolytic Streptococcus. In two studies subjects were excluded if they did yield a positive throat-swab culture for Group A Beta Haemolytic Streptococcus (Taylor 1977; Petersen 1997). The use of antipyretic analgesics was not stated in nine studies, administered routinely in five studies, and prohibited in four studies. The prohibition of analgesics might exaggerate any small symptomatic benefit of antibiotics over control if antipyretic analgesics are usually recommended in normal practice.

1. Non-suppurative complications (See MetaView)
Cases of acute glomerulonephritis occurred only in the control group, which suggests that antibiotics conferred protection. There were, however, only two cases, and only ten studies reported on acute glomerulonephritis as an end point. A very wide 95% confidence interval, (OR = 0.07; 95% CI 0.00-1.32) precludes the claim that antibiotics protect sore throat sufferers from acute glomerulonephritis.

Several studies found that antibiotics reduced acute rheumatic fever to about one third of that in the placebo group (OR = 0.30; 95% CI 0.20 to 0.45). Few studies examined antibiotics other than penicillin. Use of penicillin alone resulted in little difference in protection (OR = 0.27; 95% CI 0.18 to 0.41).

2. Suppurative complications (See MetaView)
Antibiotics reduced the incidence of acute otitis media to about one quarter of that in the placebo group, (OR = 0.22; 95% CI 0.11 to 0.43) and reduced the incidence of acute sinusitis to about one half of that in the placebo group (OR = 0.46; 95% CI 0.10 to 2.05). Data indicate that the incidence of quinsy was also reduced in relation to placebo group (OR = 0.16; 95% CI 0.07 to 0.35).

3. Symptoms (See MetaView)
At day three of illness, antibiotics reduced symptoms of sore throat (OR = 0.41; 95% CI 0.36 to 0.48), headache (OR = 0.70; 95% CI 0.52 to 0.94) and fever (OR = 0.62; 95% CI 0.46 to 0.85). Day three was the time of greatest benefit because the symptoms of only half the patients had settled. At one week (six to eight days) the odds ratio of experiencing sore throat was 0.51 (95% CI 0.40 to 0.64), although 85% of controls were better by this time.

A new trial is included in the 2003 update from Thailand (Leelarasamee 1999). This trial is especially important because it is one of the few trials from a non-Western, non-industrial country. Unfortunately we were unable to enter its data into the meta-analysis because of different ways of collecting the data (in particular no data were collected mid-way through the illness). Nevertheless, the use of antibiotics conferred no benefit (nor harm) on symptoms (or complications).

4. Subgroup analysis of symptom reduction (See MetaView)
a) Blind versus unblinded studies
There was no significant difference between blind and unblinded studies for symptoms of sore throat at day three (OR = 0.36; 95% CI 0.30 to 0.43; and OR = 0.58; 95% CI 0.45 to 0.75 respectively) or at one week (OR = 0.54; 95% CI 0.40 to 0.72 and OR = 0.57; 95% CI 0.38 to 0.84 respectively). Contrary to expectation the trend was for a greater effect of antibiotics for blind studies at day 3, see MetaView.

b) Antipyretics administered versus not administered (See MetaView)
There was no significant difference between studies in which antipyretics were offered and those in which they were not (OR = 0.28; 95% CI 0.19 to 0.41; and OR = 0.34; 95% CI 0.25 to 0.47 respectively).

c) Throat swabs positive for Streptococcus, versus negative for Streptococcus, versus not tested/inseparable combined data (See MetaView)

The probability of still experiencing pain on day three was only about one quarter (OR = 0.26; 95% CI 0.21 to 0.32) for those patients who had throat swabs positive for Group A Streptococcus, compared to about one half (OR = 0.46; 95% CI 0.33 to 0.64) for those with negative swabs. There was a similar effect at one week (OR = 0.16; 95% CI 0.005 to 0.26 and OR = 0.65; 95% CI 0.38 to 1.12 respectively). That is, the effectiveness of antibiotics is increased in people with Streptococci growing in the throat.

d) Children versus adults (See MetaView)
Few studies included children (< 13 years of age): there were only 61 children in total for evaluation of fever at day three. There was overlap of the OR 95% confidence intervals, so that the trend for children to not experience benefits was not significantly different to adults who did (OR = 1.87; 95% CI 0.48 to 7.23; and OR = 0.38; 95% CI 0.24 to 0.58, respectively).

Natural history
Symptoms of sore throat and fever had disappeared by three days in about 40% and 85% respectively in the placebo groups. Eighty-five percent of patients were free of symptoms by one week. This natural history was similar in Streptococcus positive, negative, and untested patients. About two per cent of placebo patients developed rheumatic fever. However, this complication occurred only in trials reporting before 1961. The background incidence of acute rheumatic fever has continued to decline in Western societies since then.

Benefits of treatment
The absolute benefit of antibiotics for the duration of symptoms was modest. The reduction of illness time is greatest in the middle of the illness period when the mean absolute reduction is about one day at around day three. There are not enough data to make conclusions about effects of treatment on children. The absolute reduction averaged over the whole illness can only be estimated from these data. The difference in the area under the survival curves of sore throat symptoms for those treated with placebo as opposed to antibiotic is about 16 hours for the first week.

Estimates of the number of people needed to treat (NNT) to resolve the symptoms of one by day three is about 3.5 for those with positive throat swabs for Streptococcus. For those with a negative swab, the NNT is 5.5, and for those in whom no swab has been taken it is 14.5. The last result is difficult to understand. One might have expected the NNT value to lie between the swab negative and swab positive results. Perhaps patients with less severe throat infections were recruited into the three studies in which swabs were not taken.

Antibiotics are effective in reducing the relative complication rate of people suffering sore throat. The relative benefit, however, exaggerates the absolute benefit because complication rates are low and the illness is short. Interpretation of these data is aided by estimating the absolute benefit, which we attempt below.

For every 100 patients treated with antibiotics rather than placebo in these trials (conducted mostly during the 1950s), there was one fewer case of acute rheumatic fever, two fewer cases of acute otitis media, and three fewer cases of quinsy. These figures need to be adapted to current circumstances and individuals. For example, the complication rate of acute otitis media among those with sore throats before 1975 was three per cent. A number needed to treat (NNT) of about 50 to prevent one case of acute otitis media can be estimated from the data. After 1975, this complication rate fell to 0.7%, and applying the odds of reducing the complication with antibiotics from the data table yields a NNT of nearly 200 to prevent one case of acute otitis media. Clinicians should exercise judgement in applying these data to their patients.
In particular in the modern times in the West (where absolute rates of complications are lower) the NNT will rise above a rate at which it might be regarded as worthwhile to treat. In developing countries where the absolute rate may be much higher, the lower NNT will mean antibiotics are more likely to be effective.

Adverse effects of treatment
We were unable to present the adverse effects of antibiotic use because of inconsistencies in recording these symptoms. In other studies these were principally diarrhoea, rashes and thrush (Glasziou 1997). Consideration of the side effects of antibiotics would have been useful in further defining their risk-benefits.

Special risk groups
Acute rheumatic fever is common among people living in some parts of the world (Australian Aborigines living in poor socioeconomic conditions, for example), and use of antibiotics may be justified in these settings to reduce the complication of acute rheumatic fever. In other parts of the world the incidence of acute rheumatic fever is so low (one estimate is that it took twelve general practitioners' working lifetimes to encounter one new case of acute rheumatic fever in Western Scotland in the 1980s (Howie 1985)), that the risks of serious complication arising from using antibiotics for this complication might be of the same order as that of acute rheumatic fever.

Prof. Jim Dickinson for helpful suggestions about dividing the studies into early and late last century to examine the idea that the pathogenesis of this illness, and, or, its sequelae, have changed with time.

Ian Thomas and Michael Thomas for research assistance.

Beth Clewer and Katie Farmer who, in January 1999, drew our attention to mistakes in the data extraction by their careful checking of original studies as part of their medical student project at the University of Bristol Medical School.

A previous update was completed with the help of a GlaxoSmithKline-sponsored educational support grant from the Australasian Cochrane Centre.

None known.

The Acute Respiratory Infections Group would like to thank Dr Dilruba Nasrin for reading and commenting on this review.

1. The objectives as they are stated in the abstract include an assessment of the harms associated with the use of antibiotics in the management of sore throat, but the objectives as stated in the text of the review no longer refer to any assessment of harm. Indeed, the review does not address any adverse effects of antibiotics [which are not unimportant] and does not provide a reasonable explanation as to why this is not done other than to state in the discussion that this was not possible because of inconsistencies in the way these data were recorded. In the absence of RCT data on harmful effects the authors might have considered whether usable information could be provided by other study designs.

2. Reviews on this subject should treat adults and children separately, but this review does not attempt to do this.

3. All clinically important outcomes have not been addressed by the review and others such as resource use, re-attendance and time off school or work are probably at least as important as those that were selected. It may have been more helpful to have collected data on all available outcomes provided that they are free from detection bias.

4. The question addressed by the review is not sufficiently well defined to allow the review to be executed systematically. Clear definitions are not given for the key elements of the question.

Most importantly, clear definitions of what is meant by primary care and sore throat are not given, leading to confusion around inclusion and exclusion decisions. Many of the control groups of the included studies do not involve a placebo but instead simply compare treatment with antibiotics to no treatment, so that some excluded studies would be eligible for inclusion, such as Catanzaro 1958 which was excluded because it compared antibiotics with sulfadiazine.

Apparent errors in inclusion and exclusion decisions have arisen probably as a result of the general lack of clarity discussed above. Specifically, the lack of a clear definition of what is meant by primary care appears to have led to the inclusion of an odd assortment of studies. For example, a couple of the included trials studied only people with sore throat who were admitted to hospital (Siegal 1961 and Bennike 1951). In addition, there appears to be an issue around the definition of a sore throat particularly in relation to positive or negative Streptococcus throat swabs. Streptococcal sore throats are a small sub-set of the total population of sore throats and the failure of the reviewers to address this in the inclusion criteria means that the results of pragmatic trials of sore throat are mixed in with those of
streptococcal sore throat.

There is a failure to always faithfully report the detailed results of the included studies, and there are several numerical errors in the data abstracted. For example, in Bennike 1951 the baseline numbers include patients in the "ulcerative tonsillitis" group even though most outcomes are not reported for this group.

5. The search strategy is restricted to a Medline search, a search of the Cochrane Library and citation checking. No attempt appears to have been made to search other databases. The reviewers are not explicit about the details of their searching activities nor about how they used the work of the Cochrane Acute Respiratory Infections Group.

6. References to the included and excluded studies were incomplete. Specifically they were not provided for Dagnelie 1996, Howie 1997, Little 1997 and Peterson 1997 (included) and Herx 1988, Howie 1970, Marlow 1989, McDonald 1985, Schalen 1993 and Todd 1984 (excluded).

7. Given the nature of the data presented, it is possible that a formal meta-analysis was inappropriate. A descriptive analysis may have been more appropriate and more informative.

8. There is considerable uncertainty around the effectiveness of antibiotics on sore throat on the basis of the existing research examined by this review and this is not emphasised by the authors. Particular problems exist around the relevance of the trials to the present day with regard to the outcomes examined (rheumatic fever and glomerulonephritis), the poor quality of the majority of the included trials and the generalisability of the trials with regard to the study populations (e.g. United States airforce recruits).

1. This is valid criticism: we need to describe the inadequacies of the information in the trials (after checking again) in the text.

2. A subgroup analysis on the basis of age is a good idea, and we will attempt this at the next major review.

3. This is a good idea, and we will attempt this at the next major review.

4. Certainly the issue of definitions is particularly difficult in this group of illnesses. One of us has written a paper on these difficulties (Del Mar C. Managing sore throat: a literature review. I. Making the diagnosis. The Medical Journal of Australia 1992;156:572-5.). There is a particular difficulty in the fact that primary care doctors use the terms 'sore throat' tonsillitis and pharyngitis in slightly different ways, including interchangeably. Moreover the notion that patients with positive swabs for Streptococcus have a different illness can be challenged. Nevertheless a subgroup analysis for this with swab-positive and swab-negative is a good idea, and one which we will incorporate with our next review.

Thank for pointing numerical errors out to us, and we will check on this. Please could you detail other numerical errors for us?

5. We are explicit about our search method. At the time we undertook the search the Cochrane Acute Respiratory Infections Group had no material to assist us. This will be reviewed at the next major update.

6. Thank you for drawing our attention to this.

7. As is often the case, there is considerable variation in the population groups, treatments, outcomes measures, etc in these trials. This does not make a synthesis inappropriate, but rather allows us to examine whether these factors appear to make a difference. We also felt it important to specifically attempt to calculate the SIZE of the benefits, as this is what clinicians are interested in, and what will persuade them to modify their practice. It is then important to recognise that the size of the effect will vary in different populations: as we point out, in groups at high risk of rheumatic fever - such as Australian aboriginals - the prevention of RF is important; we are also interested in trying to better predict which sub-groups will experience the most or least symptom relief, and plan to detail this in the next update.

8. We think we have discussed this in the Review. However we will reconsider what we have written in the overhaul.

Jackie Young (on behalf of an interdepartmental critical appraisal workshop based in the Department of Public Health and
Epidemiology, The University of Birmingham, UK) Email: j.m.young.20@bham.ac.uk

I noticed that trials with no events in either groups are not (cannot) be part of the pooled estimates. Although I see there is a statistical/technical problem here it does not seem right. It appears to imply that no events is no evidence. I wonder whether it is defensible to add one event in both groups and add the evidence as one would normally do?

I certify that I have no affiliations with or involvement in any organisation or entity with a direct financial interest in the subject matter of my criticisms.

Many thanks for this. We have gone back and checked with statisticians about your point. The issue seems to be:

1. Whether empty cells are a problem. The concern is that because one cannot divide anything by zero, this might represent a problem. We think not, because in no forest plots are there totals with zero - except for acute glomerulonephritis (there were no cases in the intervention arms of any trials, and only two in the control arms).

2. Whether the empty cells represent no evidence or evidence of no effect. We only recoded a zero where the study declared the outcome. Thus we assume that "no events" implies no events, rather than no reporting of events that might have occurred.

We have reported in Peto Odds ratios, the best measure for rare events.

Chris Del Mar

Gerben ter Riet