Are antibiotics beneficial for patients with sinusitis complaints?

Patients

Patients were eligible to participate if they were 18 years or older; had at least 1 cardinal feature described by the clinical prediction rule: 1) purulent nasal discharge predominating on one side, 2) local facial pain predominating on one side, 3) purulent nasal discharge on both sides, or 4) pus in the nasal cavity; and had symptoms for at least 7 days. Patients were excluded if their histories included antibiotic treatment within the past month, allergy to penicillin, sinus surgery, compromised immunity, pneumonia, or streptococcal pharyngitis.

Randomization

Permuted block randomization stratified for the 3 participating clinicians was used to determine treatment assignment. Patients were given an envelope containing 40 capsules, either a placebo medicine taken twice daily for 10 days or 1000 mg of amoxicillin (500 mg pills) taken twice daily for 10 days. The envelopes were opaque, and each had 40 identical-appearing pills (to ensure allocation concealment). The participating clinicians were naive to the treatment assignments.

Assessment of outcomes

Trained personnel, masked to treatment assignment, conducted follow-up telephone interviews on days 3, 7, and 14 following patients’ visits for sinusitis, to assess clinical improvement. Twelve follow-up questions were asked.

Sample size

The primary outcome used to determine sample size was dichotomous—either “improved” or “not improved” by the end of 2 weeks. Thus, patients were asked, “what day were you entirely improved.” The sample sizes obtained per group (67 for amoxicillin and 68 for placebo) provided 80% power for detecting a change of 25% in rates of improvement.

Statistical analysis

Basic descriptive statistics were used to describe the groups. Baseline characteristics were compared between the 2 groups using chi-square test and Fisher’s exact test for categorical variables. For continuous variables, the Student’s t-test was used; the Wilcoxon Rank Sum test was used for ordinal or skewed variables. Similar statistical tests were used to compare baseline characteristics between the providers and also at the conclusion of the study between the responders for each group.

For the outcome variables, we hypothesized no difference between the groups either in the rates of improvement, times to improvement, or in patients’ self-rating of sickness. The actual proportions improving between the 2 groups were compared using the chi-square test. Relative risk estimates and 95% confidence intervals were calculated to provide measures of risk and precision. Multiple logistic regression was used to compare the rates of improvement adjusting for the number of signs or symptoms classified as either 1, 2, or 3–4, obtained from the clinical prediction rule (Table 1).

The Kaplan-Meier method was used to construct the curves showing the time until patient improvement for each treatment group. The Wilcoxon test was then used to test the statistical significance in these curves (Figure). Cox’s Proportional Hazards regression was used to test for differences in the time to improvement between the groups adjusting for the number of signs or symptoms.

Additionally, a univariate repeated measures analysis of variance model was constructed to compare the 10-point Likert scale scores for the question, “How sick do you feel today?” In this model, the number of signs and symptoms was entered as a covariate in the analysis. Orthogonal contrasts were used as post-hoc tests to compare the difference between the groups within each time point (Table 2 ).

For the subgroup of patients who improved, analysis of covariance was used to compare the mean number of days to improvement between the groups. In this case the number of signs and symptoms was used as the covariate (Table 3). The Kaplan-Meier method and the Wilcoxon test were used to compare the cumulative rates of improvement (Figure). Unadjusted P-values are reported.

Primary analyses were performed using the intention-to-treat principle. All statistical analyses were performed using JMP Software (Product of SAS Institute Inc, Cary, NC). Statistical significance was set at 0.05 and exact P-values are reported.

TABLE 1
Baseline characteristics for amoxicillin and placebo groups