Acute sinusitis: Which factors do FPs believe are most diagnostic and best predict antibiotic efficacy?
A questionnaire-and-case-vignette study reveals misjudgment of infection and antibiotic overprescribing.
Questionnaire distribution
The questionnaires were piloted on a small group of doctors before mailing to 1024 family doctors in the Hampshire Health Authority. Each doctor received a different example of the 64 possible factor combinations for patient A, which were randomly paired with 1 of the 64 possible factor combinations for patient B. We produced 16 replicates of these pairings, and then assigned these scenario-pairs randomly to the 1024 doctors in the Wessex locality (Hampshire, Dorset, and Wiltshire).
After 3 mailings, 557 replies were received; 537 were fully completed and used in the analysis (54.4% response rate). To assess response bias, we compared the Townsend Index (material deprivation score). This is an area level unweighted sum of 4 census variables standardized as Z-scores. It was derived from the practice/premise zip codes of all participating and nonparticipating doctors.21
Statistical methods and analysis
Diagnostic factors. We present the mean scores with 95% confidence intervals for each of the variables (since the median is insensitive to small changes when the range is narrow). Factor analysis with varimax rotation was also performed on the scores to identify clusters of variables that doctors use in diagnosis (using STATA version 7). Factor analysis is a special statistical method for dealing with multi-way frequency tables, and which involves reducing correlated variables to a single factor (such as a symptom/sign cluster). Varimax rotation is a method of extracting the principal component factors by using a variance maximizing method on the data set.22
Case vignettes. The doctors’ likelihood of prescribing antibiotics for vignettes A and B was reduced to a dichotomous variable (likely vs unlikely to prescribe to correspond to a treatment decision). Multiple logistic regression was used to investigate this variable with the psycho social variables, demographic variables (including the Townsend Index of deprivation) and doctors’ beliefs in antibiotics. Extracted factors from the factor analysis of the symptoms and signs were also included in the analysis. Backwards elimination was used with variables at the P≤.05 level remaining in the models.
Results
Of 537 completed replies from Wessex family doctors, 60.7% were male and 37.8% were female. The median age was 45 years, with 15 years as a family doctor. 65.4% had professional collegiate membership and 8.4% had a special interest in otorhinolaryngology. The health center characteristics were 37% urban, 43% suburban, and 18% rural, with a median number of partners of 5.5 and list size of approximately 10,000 registered patients. The number of sessions per week was 7.4. The Townsend Index of material deprivation ranged from –6.0 to 9.92.
In our analysis of nonresponders we found that nonresponders were from practices in more deprived areas (t-test; t995=2.433, P<.05). The mean score for responders was –0.24 and for nonresponders 0.26. The demographics of our sample was identical or very close to that for Hampshire and England for gender, individual list size, and age, but differed in average number of partners from 5.5 (sample), 4.3 for Hampshire, and 3.2 for England.23 Since number of partners was not significant in any analysis these results should nevertheless be generalizable.
Diagnostic factors
The importance (mean score) of each symptom and sign used by doctors in diagnosing acute sinusitis are shown in TABLE 1. Factor analysis with varimax rotation identified four groups of variables used—nasal symptoms (without purulence), purulence, being unwell, and pain (TABLE 2).
TABLE 2
Four key groups of variables: Nasal symptoms (without purulence), purulence, being unwell, pain
| Components (with loadings) of the 4 factors from factor analysis (varimax rotation) with Cronbach's α for internal reliability | |||
| Factor 1 Nasal (α=0.75) | Factor 2 Purulent (α=0.77) | Factor 3 Unwell (α=0.76) | Factor 4 Pain (α=0.63) |
| Blocked nose/nasal congestion on both sides (0.773) | Purulent rhinorrhea more marked on one side (0.785) (0.832) | Malaise/unwell | Tender on facial pressure or percussion (0.732) |
| Blocked nose/nasal congestion on one side (0.738) | Purulent rhinorrhea on both sides (0.766) | Temperature > 38°C (0.801) | Pain in face on bending forward (0.686) |
| Purulent secretions in nasal cavity on inspection (0.75) | Frontal pain (0.584) | ||
| Pus exuding from ostium (0.626) | Severity of facial pain (0.537) | ||
Case vignette—patient A (unlikely bacterial sinusitis)
In this vignette, 37% of doctors would prescribe an antibiotic. From multivariate analysis the most important variable in the decision to prescribe was belief in the effectiveness of antibiotics in treating symptoms (TABLE 3). Nasal symptoms without purulence (factor 1), and pain (factor 4) were independently significant in the model, as was Townsend Index of deprivation. Two variables were predictive in univariate analysis—the practice type/urban vs suburban (P=.019), and number of years as a family doctor (P=.03)—but were not independently predictive in multivariate analysis.
Case vignette—patient B (likely bacterial sinusitis)
In this vignette 82% of doctors would prescribe an antibiotic. The most important variable in the decision to prescribe was, again, belief in the effectiveness of antibiotics to treat symptoms, followed by belief in their effectiveness in treating complications (TABLE 4). Family history was predictive in univariate analysis but not in multivariate analysis.
