Open randomised trial of prescribing strategies in managing sore throatBMJ 1997; 314 doi: http://dx.doi.org/10.1136/bmj.314.7082.722 (Published 08 March 1997) Cite this as: BMJ 1997;314:722
- P Little, wellcome training fellowa,
- I Williamson, senior lecturer in primary carea,
- G Warner, general practitionerb,
- C Gould, research assistanta,
- M Gantley, anthropologista,
- A L Kinmonth, professor of primary medical carea
- a Primary Medical Care, Faculty of Health, Medicine, and Biological Sciences, Aldermoor Health Centre, Southampton University, Southampton SO16 5ST
- b Nightingale Surgery, Romsey, Hampshire
- correspondence to: Dr Little
- Accepted 18 December 1996
Objective: To assess three prescribing strategies for sore throat.
Design: Randomised follow up study.
Setting: 11 general practices in the South and West region.
Subjects: 716 patients aged 4 years and over with sore throat and an abnormal physical sign in the throat; 84% had tonsillitis or pharyngitis. Patients were randomised to three groups: prescription for antibiotics for 10 days (group 1, 246 patients); no prescription (group 2, 230 patients); or prescription for antibiotics if symptoms were not starting to settle after three days (group 3; 238 patients).
Main outcome measures: Duration of symptoms; satisfaction and compliance with and perceived efficacy of antibiotics; time off school or work. Outcomes were documented in 582 subjects (81%).
Results: Median duration of antibiotic use differed significantly in the three groups (10 v 0 v 0 days, P<0.001); 69% of patients in group 3 did not use their prescription. The proportion of patients better by day 3 did not differ significantly (37% v 35% v 30%, P=0.28), nor did the duration of illness (median 4 v 5 v 5 days, P=0.39), days off work or school (median 2 v 2 v 1, P=0.13), or proportion of patients satisfied (96% v 90% v 93%, P=0.09), although group 1 had fewer days of fever (median 1 v 2 v 2 days, P=0.04). More patients in group 1 thought the antibiotics were effective (87% v 55% v 60%, P<0.001) and intended coming to the doctor in future attacks (79% v 54% v 57%, P<0.001). “Legitimation” of illness—to explain to work or school (60%) or family or friends (37%)–was an important reason for consultation. Patients who were more satisfied got better more quickly, and satisfaction related strongly to how well the doctor dealt with patient's concerns.
Conclusion: Prescribing antibiotics for sore throat only marginally affects the resolution of symptoms but enhances belief in antibiotics and intention to consult in future when compared with the acceptable strategies of no prescription or delayed prescription. Psychosocial factors are important in the decision to see a general practitioner and in predicting the duration of illness.
Sore throat is one of the commonest presentations of upper respiratory illness to general practitioners, and attendance is increasing
Prescribing antibiotics for sore throat does not reduce the extent and duration of symptoms
Prescribing antibiotics enhances belief in antibiotics and intention to consult
Legitimation of illness is an important reason for attending the doctor
Satisfaction predicts duration of illness and closely relates to how well concerns are dealt with—unless patients are very ill, general practitioners should consider exploring concerns and should avoid or delay prescribing antibiotics
Patients with sore throat are commonly seen in primary care.1 Though recent guidelines advocate using results from throat swabs—or clusters of symptoms or signs—to determine which patients should be given antibiotics,2 management is still controversial.3 4 Throat swabs have poor validation characteristics,5 are expensive, and may not alter prescribing decisions.6 Further evidence and a recent systematic review showed that antibiotics give only marginal benefit in resolving symptoms or preventing complications—both suppurative and non-suppurative (rheumatic fever, glomerulonephritis)7 8 9. The largest primary care trial found that antibiotics did not shorten the duration of symptoms.10
Though double blind trials give the best evidence for efficacy by controlling the placebo effect, the behaviour and perceptions of patients and doctors may not be generalisable to the normal setting. By more closely approximating everyday practice, open trials provide important evidence of effectiveness; they are essential when outcome measures include patients' perceptions and choices in response to different strategies—for example, whether “delayed” prescriptions are collected, the perceived efficacy of antibiotics, or the likelihood of future attendance when symptoms have resolved without treatment.
An open randomised controlled trial of different management strategies in upper respiratory conditions has not yet been carried out. Case descriptions for otitis media and a non-randomised open trial of strategies for managing sore throat have been published.11 12 13 We report an open trial comparing three common plausible strategies (antibiotic prescription, no prescription, or delayed prescription) in patients with sore throat.
Development of advice packages–The advice package given to patients in each group had six or seven standard statements supporting the particular strategy and included advice to take analgesics or antipyretics. Provisional versions of the packages were modified after tape recorded interviews (five with general practitioners and 15 with patients presenting with sore throat) which identified important content and appropriate wording for the packages and outcome measures.
Setting–General practitioners expressing an interest in ear, nose, and throat research on the Wessex research network register were contacted: 25 general practitioners (17 men) in 11 general practices agreed to participate. Six practices were training practices, eight were wholly or predominantly rural (OPCS classification), and three were fundholding.
Patients–During the study period (September 1994 to May 1996), 714 patients aged 4 years and over presented to their general practitioners with sore throat either as principal or subsidiary symptom and showed an abnormal physical sign localising to the throat (inflamed tonsils or pharynx, purulent exudate, faucial or palatal inflammation, cervical adenopathy). For children (under 12 years), who are less likely to complain of sore throat, abnormal signs in the throat were sufficient.
Exclusions–Patients were excluded if they had other explanations of sore throat (drugs, aphthous ulcers, candida, etc), were very ill (when not giving antibiotics might be unethical), had suspected or previous rheumatic fever, had had multiple attacks of tonsillitis (>5/year), had had severe local complication (quinsy), or were pregnant.
Calculation of sample size–We calculated sample size for 80% power and significance at the 5% level of significance. Data from previous studies showed that 50 patients per group were required to detect one day's difference in resolution of symptoms with antibiotics (4 days with antibiotics, 5 days with no antibiotics or delayed antibiotics; standard deviation 2 days10).14 However, with the standard deviation from the current study (3.15 days), detecting one day's difference would require 150 patients in each group, and detecting 1.5 days' difference would require 50 patients in each group. Detecting a 15% difference in the proportion of patients better after three days9 10 would require 173 patients in each group (using the sample size calculator from spida version 1.6–a total of 693 to allow for 25% drop out.
Randomisation–Patients gave written consent to a “study looking at how quickly sore throats settle.” To minimise contamination between the groups, general practitioners were asked not to discuss the efficacy of antibiotics before randomisation. During the consultation the general practitioner opened a sealed envelope containing one of the three randomised advice sheets based around one of three prescribing strategies. Patients assigned to group 1 were given a prescription for antibiotics (10 day prescription of penicillin V (or erythromycin if sensitive to penicillin), 250 mg four times daily (125 mg for 3-5 year olds). Patients assigned to group 2 were not offered antibiotics. Patients assigned to group 3 were offered antibiotics as in group 1, but the patient could collect the prescription from the surgery if symptoms were not starting to settle within three days.
Documentation–Each advice sheet had boxes to tick once the statement had been read or explained. The advice sheet was returned to the administrative centre for monitoring.
The general practitioner's documentation sheet showed days of illness, physical signs, and antibiotic prescription. At the end of the study general practitioners were sent a questionnaire asking them to rank the list of reasons for non-recruitment and how they had used the advice packages (read verbatim, used as a prompt, etc).
Patients were given a daily diary in which to record symptoms and temperature (using the Tempadot thermometer15). This form of data collection minimises the bias that could result from a researcher administered questionnaire. The diary was to be filled in each day until patients were both free of symptoms and had finished their medication. Patients answered written questions on four point Likert scales (very, moderately, slightly, not at all) at the beginning of treatment—about worries, satisfaction, and “legitimation” (attending the doctor to explain the illness to others)–and at the end—about antibiotic use, perceived efficacy, future intentions, and time off work and school. Within three days of the consultation, patients were contacted by a research assistant to check that there were no problems with filling in the diary. A total of 105 patients had not returned their diaries two weeks after entry into the study; these patients were telephoned and asked the questions addressed by the diary.
Validity of telephone information and reliability of questions used–One week after the diaries were received, 25 consecutive patients were telephoned and asked about the duration of sore throat; these data showed excellent agreement with the diary (Spearman r=0.94; P<0.001), median difference 0 days (interquartile range for the difference 0 to 1 day, range -2 to 2 days). The Likert scale questions were sent again to 25 respondents after two weeks: there was high agreement on the two occasions (>14/20), with good rank correlations (>0.64) and values (>0.65).
Data entry and analysis–Data were entered and analysed on an intention to treat basis with spss. Differences in proportions were assessed by the χ2 test. For continuous variables, data were presented as medians (interquartile range), and groups were compared by the Kruskal-Wallis test to avoid assumptions about the normality of data—which was skewed for some variables. Subgroup analysis was performed by selecting features that might predict presence of streptococcus (that is, predict antibiotic response). To limit type I error, we prioritised features to test: the first priority was for variables found important in a previous study10 –pharyngitis and enlarged cervical glands; second priority was for variables identified from other settings,2 16 17 18 including a cluster of symptoms or signs (three of the following: tender cervical glands, higher temperature, purulent exudate, no cough, dysphagia) and a Breese score of more than 25 (Breese scores are calculated from nine factors, each weighted according to likelihood of predicting that results from a throat swab would be positive; scores less than 25 are associated with a low incidence of positive swabs18). Multivariate analysis of the predictors of a prolonged course of antibiotic treatment will be presented in a subsequent paper.
Diary or phone responses were obtained in 582 (81%) subjects—a higher response rate than in the previous major study in primary care (70%).10 Sociodemographic and clinical features were similar in the three groups, both in the total group and in the responders (table 1).
Selection bias–To assess selection bias due to non-recruitment, we compared the responders from high recruiting general practitioners (those recruiting more than 80 patients—probably most patients presenting to them19 20) with those from lower recruiters. There was no difference in the percentage aged over 12 (76% in high recruiters v 76% in low recruiters, χ2=0.03, P=0.87), those who had had symptoms for more than three days before consultation (39% v 36%, χ2=1.14, P=0.29), sex (males 39% v 33%, χ2=3.2, P=0.14), or the number with three of the five symptoms and signs (24% v 22%, χ2=0.12, P=0.7). The generalisability of the results was supported by the general practitioners' questionnaire: the highest ranked reason for not recruiting was “no time” (ranked first by 60%); patient's refusal was the most uncommon reason (ranked first by 13%).
Group differentiation–General practitioners' initial compliance with each prescribing strategy was good. In group 1, 243 of 246 patients (99%) left the consultation with a prescription for antibiotics compared with 5/230 (2%) in group 2 and 11/237 (5%) in group 3; antibiotic use was reported in 210/211 (99%), 23/184 (13%), and 55/176 (31%) respectively. Duration of antibiotic use for the three groups was significantly different (median duration 10 v 0 v 0 days, Kruskal-Wallis χ2=329, P<0.001; mean 9 v 0.9 v 2.9 days).
Resolution of symptoms–The three groups did not differ in resolution of symptoms by three days (37%v 35% v 30%, χ2=2.50, P=0.28) or in the median duration of any symptom except fever (table 2). Figure 1) shows marginal differences between groups in duration of symptoms.
Subgroup analysis–Advice groups did not differ significantly in the most important variables selected in advance (enlarged cervical glands or pharyngitis) (table 3). The only significant differences were found for high temperature (>37.5°C) and having three of the five symptoms and signs. For both these variables, however, the mean ranks for initial antibiotics and no antibiotics were similar (44 v 42 v 62 for higher temperature; 42 v 42 v 57 for symptoms and signs). Since the group given antibiotics immediately was so similar to the group given no prescription, antibiotics per se are not likely to be important.
Differences in patients' beliefs and intentions–Patients given antibiotics immediately were more likely to believe that antibiotics were effective. They were also more likely to intend consulting in future episodes (table 4).
Satisfaction–Most patients indicated they were very satisfied or moderately satisfied and had their worries dealt with very well or moderately well (table 4). In group 1 more patients indicated that they were very satisfied, rather than the other categories, with both the consultation (84% v 58% v 65%, χ2=33, P<0.001) and how their worries were dealt with (81% v 61% v 63%, χ2=22, P<0.001). Satisfied patients got better more quickly (median duration 4.0 days for very satisfied, 5.0 days for moderately, 5.5 days for not very, 6.5 days for not at all; Kruskal-Wallis χ2=15.4, P=0.002). Satisfaction and dealing with concerns were closely associated (agreement 497/572 (87%); χ2 for trend=385, P<0.001; κ=0.72).
This is the first open randomised controlled trial of prescribing strategies for sore throat, and the largest dataset (n=582) from a randomised controlled trial for sore throat from primary care. Despite several limitations the study provides reasonable estimates of the effectiveness of prescribing strategies for sore throat.
One potential limitation is that the trial excluded very ill patients and thus cannot show the efficacy of antibiotic prescribing strategies for them. Nevertheless, the trial considers the large group for which general practitioners might prescribe (or withhold) antibiotics.21
General practitioners switching between prescribing strategies according to randomisation of patients may be uncomfortable, and if the assigned strategies were not adhered to, the groups would be differentiated inadequately (contamination bias). However, randomisation by practice leads to potential selection bias, as the strategy (and thus the group) would be known in advance. Furthermore, contamination bias was minimised by structuring advice packages. In fact groups were well differentiated, as shown by patients' perceptions and antibiotic use.
Selective overall recruitment may limit generalisability of results. A comparison of patients from high and low recruiters, however, showed no obvious bias, consistent with general practitioners reporting lack of time as the commonest reason for non-recruitment. In addition, selective recruitment between groups may threaten randomisation—but in fact the groups had similar characteristics. Variable or low response rates can jeopardise generalisability and threaten randomisation; however, a response rate of more than 80% was achieved, and the responders had similar characteristics to the original randomised group, which did not differ significantly between groups.
Though the open design may cause bias owing to the placebo effect, it was chosen because it mimicked perceptions and behaviour of everyday practice more closely and could assess outcomes where patients' knowledge of treatment is essential (perception of efficacy, collecting scripts, satisfaction, intention to consult). By supporting each proposed strategy, the doctor functioned as a placebo in each group.
Role of antibiotics
The small differences between the three groups in the proportion of patients better by day 3 and in all symptoms (except fever) are consistent with findings of the previous major primary care trial10, and suggest that antibiotics only marginally affect resolution of symptoms.
Our findings suggest that identifying broad subgroups is unlikely to predict antibiotic response. Perhaps only more extreme groups—people with Breese scores of 30, or with all features of a complex of symptoms and signs—should be included in trials.2 16 17 18 However, in such subgroups it is probably only patients with positive throat swabs who benefit from antibiotics.22 Extreme subgroups are made up of a small minority of patients presenting to general practitioners, and a minority of those whose throat swabs test positive.5 23 5 Also, throat swabs are also neither sensitive nor specific for serologically confirmed infection,5 24 25 18 27 considerably increase costs, may medicalise illness, and alter few management decisions.6 Thus using symptom clusters or throat swabs to target antibiotic prescribing is not well justified by current evidence.
Antibiotics may increase recurrence by altering pharyngeal flora and limiting the development of immunity.28 Antibiotics would need to shorten the illness by more than 1.5 days to offset the likely 20% increase in relapse or recurrence with early antibiotics compared to delayed antibiotics (pooled results from three recent trials yield a conservative unweighted estimate of 90/220 (41%) for early antibiotics and 55/236 (23%) for delayed antibiotics28 29 30). Thus the net symptom burden is unlikely to be significantly improved by prescribing antibiotics for patients who are not very ill.
This study shows that “legitimation” of illness—either for work purposes or to family and friends—is an important factor in attending the doctor. This is consistent with Parsons's classic model where the doctor provides gatekeeping to the sick role.31 The prevalence of these reasons for attendance suggests that further research could evaluate the effect of exploring these issues with patients or employers.
Patients prescribed antibiotics initially were more likely to think antibiotics were effective and to intend coming back in the future. Such side effects of prescribing—in addition to the medical side effects of antibiotics3 32 –must be taken into account when the consultation rate for minor illness is rapidly increasing.19 Given the difficulty of changing perceptions in a single consultation, the effect of reinforcement by repeated messages over the longer term is likely to be greater than documented here.
All three groups had similar high percentages of patients who were very satisfied or moderately satisfied, but prescribing antibiotics resulted in more patients who were very satisfied. This suggests that some patients have considerable expectations for antibiotics—despite evidence that doctors overestimate such expectations.33 Satisfaction is important because it contributes to the “humanity” of an intervention. Satisfaction also predicted the duration of illness—which was probably not mediated by antibiotic prescribing strategy since strategy itself did not significantly predict duration—and was closely related to dealing with patients' concerns. These findings are supported by evidence that satisfaction predicts compliance,34 that psychological factors mediate development of upper respiratory illness,35 that the doctor's approach to the consultation affects the patient's recovery,36 and that more effective doctor-patient communication improves health outcomes.37
For most patients with sore throat presenting to their general practitioner, antibiotics only marginally affect the resolution of symptoms. Even after one consultation, prescribing antibiotics significantly enhances belief in antibiotics and intention to consult in the future when compared with the acceptable strategies of delayed or no antibiotic prescription. Psychosocial issues in the consultation are important: legitimation of illness is an important reason why people attend their general practitioner, and satisfaction is strongly related to effective management of patients' concerns and also to duration of illness. Therefore, unless patients with sore throat are very ill, doctors should consider exploring patient's concerns and avoid prescribing antibiotics—or offer antibiotics if symptoms persist for a few days. Immediate prescription is likely to encourage the medicalisation of a self limiting illness without altering the extent and duration of symptoms.
We are grateful to the following general practitioners for their help in recruitment, constructive comments, and enthusiasm: Drs Nigel Dickson, Graham Newman, Peter Willicombe, Peter White, Sue Tippett, Richard Peace, Julie Chinn, Katie Warner, Neil Ball, Tim Taylor, Deidre Durrant, Mark Rickenbach, Sally Bacon, Tim Whelan, Peter Markby, Simon Goodison, D Traynor, R Briggs, Evelyn Beale, Fiona Bradley, Simon Smith, and Stephen Morgan.
Funding: This work was supported by Wessex NHS regional research and development funds. PL is supported by the Wellcome Trust.
Conflict of interest: None.