In this chapter, we provide discussions and conclusions on the five key questions addressed in this evidence report. In addition, in translating the evidence into practice, we discuss the implications and limitations of patient responses to issues regarding the diagnosis and treatment of acute bacterial rhinosinusitis and conclude by highlighting the results of the decision and cost-effectiveness analyses, along with their key "take-home" messages.

1. What is the prevalence of acute bacterial infection in patients presenting with acute rhinosinusitis in primary care and specialty settings?

As seen from the evidence, the range of estimates of the prevalence of acute bacterial rhinosinusitis is wide. Several factors appear to contribute to this variation, in particular, the patient populations selected for study and the criteria used to identify cases of acute bacterial rhinosinusitis. Even when the diagnostic reference of maxillary sinus puncture is used, several potential confounding factors, including patient age, geographic and socioeconomic conditions, seasonal variation, sampling technique, sample processing, and bacteriologic evaluation could have an effect on prevalence determinations (Axelsson and Brorson, 1972). Among the studies we reviewed in detail -- all of which used the reference standard of maxillary sinus aspiration for diagnosis -- the prevalence estimates differed greatly.

Much of the variation also may result from the initial screening of the populations for study, which may change the denominator for the prevalence estimates. For example, in a study of children in a primary care practice, the overall prevalence was estimated at 9.3 percent, although in patients presenting with cold and cough symptoms, the prevalence was 17.3 percent (Aitkin and Taylor, 1998). Two studies of adults documented efforts to evaluate relatively unselected patients (Savolainen, Pietola, Kiukaanniemi, et al., 1997; van Buchem, Peeters, Beaumont, et al., 1995) and reported a prevalence of 50 percent and 83 percent, respectively. The latter study was carried out in a specialty clinic. Many of the patients in this study may have been referred by their primary care physicians, which may have added an additional population selection bias.

These findings demonstrate the need for clear description of the populations enrolled in clinical trials because prescreening (to varying extents) can substantially change estimates of prevalence. These changes, in turn, can affect the predictive values of various diagnostic tests and, as noted in the decision and cost-effective analyses, may affect clinical decisionmaking.

The prevalence of acute bacterial rhinosinusitis in actual primary care settings could be lower than that observed in clinic-based studies. Good evidence on the prevalence of acute bacterial rhinosinusitis in the general population and in general primary care practices is lacking. Studies to assess community prevalence (e.g., the National Hospital Discharge Survey and the National Ambulatory Medical Survey) rely on much less rigorous criteria for diagnosis. These inconsistencies and the expected variations (for example, in general practice vs. referral or specialty clinics) require more detailed descriptions of inclusion criteria, populations studied, and diagnostic methods. Although sinus puncture is used as the "gold standard" diagnostic procedure for research, its limitations preclude its widespread use, and patient and researcher acceptance also may limit its use in research settings. As reviewed in this report, knowing how well other tests perform in comparison to sinus puncture is critical to enable studies in more widespread clinical settings. As seen from variations in the evidence, the selection of patients for these comparative studies also may influence test evaluations, emphasizing the need for detailed reporting of study protocols and populations.

2. What is the diagnostic value of clinical features/imaging modalities for identifying acute rhinosinusitis and acute bacterial rhinosinusitis?

Sinus radiography offers fair-to-good discrimination between patients who have puncture-documented acute bacterial rhinosinusitis and those who do not. Interobserver agreement is good for four-view sinus radiographs, particularly in the diagnosis of maxillary sinusitis (Williams, Roberts, Distell, et al., 1992). Using the summary estimates of sensitivity and specificity based on the radiographic criterion of "fluid or opacity," for a prevalence of 50 percent, a positive sinus radiogram increases the posttest probability of acute bacterial rhinosinusitis to 78 percent, whereas a negative radiogram decreases it to 25 percent.

Studies included in this report also evaluated the radiographic criterion "sinus fluid or opacity or mucous membrane thickening." However, these data leave the specificity of radiographs based on this criterion imprecisely estimated (95 percent confidence interval [CI], 0.20 to 0.91). Therefore, likelihood ratios and posttest probabilities based on this threshold also are relatively imprecise, and it remains difficult to know how to interpret radiographically documented mucous membrane thickening.

The eight studies that compared sinus ultrasonography with either sinus puncture (Kuusela, Kurri, and Sirola, 1982; Laine, Maättä, Varonen, et al., 1998; Revonta, 1980; Savolainen, Pietola, Kiukaanniemi, et al., 1997; van Buchem, Peeters, Beaumont, et al., 1995) or radiography (Berg and Carenfelt, 1985; Jensen and von Sydow, 1987; Rohr, Spector, Siegel, et al., 1986) offered inconclusive information about how well ultrasound identifies patients with acute bacterial rhinosinusitis. As illustrated by the five studies comparing ultrasonography with sinus puncture, the results of ultrasonography may vary substantially beyond what is expected simply from different test thresholds (see Figure 9) (Littenberg and Moses, 1993). This added variability perhaps may arise through differences in patient populations, ultrasonography techniques, or medical personnel involved in diagnostic testing, and it calls into question the reliability of ultrasound. As documented by the poor performance of ultrasound in the study by Laine, Maättä, Varonen, et al. (1998), careful training and experience with ultrasonography may be necessary before clinicians can accurately interpret test results. Before ultrasonography can be accepted as a useful and reliable diagnostic tool, further studies, ideally comparing it with sinus puncture in a variety of research and clinical settings, are necessary.

Five studies looked at clinical examination criteria as a diagnostic tool in identifying patients with acute bacterial rhinosinusitis, comparing it with sinus puncture (Berg and Carenfelt, 1988), radiography (Axelsson and Runze, 1976; Jannert, Andreasson, Helin, et al., 1982; Williams, Simel, Roberts, et al., 1992), or ultrasound (van Duijn, Brouwer, and Lamberts, 1992). The clinical examination, composed of items from the patient's history and physical examination, provides a rapid, readily available, and reasonably inexpensive approach to the diagnosis of acute bacterial rhinosinusitis. Unfortunately, assessment of the ability of clinical criteria to diagnose acute bacterial rhinosinusitis is limited by methodologic problems with two studies, namely, inadequate description of the reference test (Berg and Carenfelt, 1988), and inadequate description of the clinical criteria (Berg and Carenfelt, 1988; van Duijn, Brouwer, and Lamberts, 1992).

Because of these methodologic problems in the study by Berg and Carenfelt (1988), there are no reliable data for how well clinical examination compares with sinus puncture as a reference test. Furthermore, data comparing clinical examination with radiography or ultrasound cannot be easily interpreted as "true" estimates of sensitivity and specificity because radiography and ultrasound themselves are imperfect tests and not ideal reference standards. However, based on data from three studies, it is possible to conclude that the clinical examination does offer moderate ability to identify patients with a positive radiograph (see Figure 8, area under SROC curve, 0.74). In turn, this ability might imply that some components of the clinical examination may be useful for determining which patients would benefit most from radiography as part of the diagnostic evaluation. An approach in which clinical findings dictate which patients receive sinus radiography may have important cost implications and deserves further evaluation.

An important type of clinical examination tool is the risk score. With a risk score, a clinician determines the presence or absence of each of a series of patient symptoms and signs; the probability of acute bacterial rhinosinusitis increases with each additional finding, and the clinician can use this information to help decide whether acute bacterial rhinosinusitis is present. Three studies presented data on how well different risk scores, consisting of three to five separate clinical symptoms and signs, identify acute bacterial rhinosinusitis. One of these studies suggested that a risk score performs better than a physician's overall clinical impression (Berg and Carenfelt, 1988). On the other hand, two studies (one of which evaluated diagnosis of bacterial rhinosinusitis in children) suggested equivalent discrimination between a clinician's overall impression and a risk score (Jannert, Andreasson, Helin, et al., 1982; Williams, Simel, Roberts, et al., 1992). Because a risk score may be simple to apply and because it may depend less on clinical experience and acumen than does a clinician's overall impression, further work is needed to develop risk scores that can assist clinicians by adding to or performing better than their overall impressions. Ideally, new risk scores would be carefully described, easily reproducible, and prospectively validated against a reference test.

Two limitations to conclusions about diagnostic test performance should be noted. First, some studies of diagnostic tests were of poor quality. In the studies included in this report, the most common of these problems were inadequately described study populations, poorly described test methods and criteria for a positive test (for both reference tests and tests under evaluation), and lack of blinding of investigators to results of one test when they were performing and interpreting another test. The information from Evidence Tables 3 to 8 highlights the need in future work for more careful attention to study design and reporting. Second, because data on test performance were extracted from each study for all available thresholds, the observations used to calculate the SROC curve were not independent. This lack of independence, in turn, precludes calculation of confidence intervals for the SROC curve. However, few studies were available for each comparison of diagnostic tests, and including several observations from each study had the important advantage of allowing for a relatively comprehensive evaluation of tests over a range of possible thresholds.

3. Given a (clinical) diagnosis of acute bacterial rhinosinusitis, are antibiotics effective in resolving symptoms and in preventing complications or recurrence?

About two-thirds of patients with acute bacterial rhinosinusitis improve or are cured without antibiotics. In patient populations defined by clinical symptoms alone, without a firm radiographic or bacteriologic diagnosis, this rate may even be higher. Treatment with any antibiotic, regardless of type, reduces the rate of clinical failures by about one-half. For the large number of patients with uncomplicated acute bacterial rhinosinusitis, a course of inexpensive antibiotics is probably adequate first-line treatment if antibiotics are to be given.

A concern in interpreting the results of this meta-analysis is the comparability of patients included in these trials with current patient populations, particularly in terms of the rates of antimicrobial resistance in sinusitis pathogens. Some of the included studies were conducted before the widespread emergence of resistance in common causes of acute bacterial rhinosinusitis, such as H. influenzae, M. catarrhalis, and most recently, S. pneumoniae. Yet we found no evidence of a difference in results between recent and older studies or between studies that included or excluded patients with organisms resistant to their allocated antibiotic. It is possible that the clinical resolution of acute bacterial rhinosinusitis may not be strongly influenced by the resistance profile of the pathogen, whatever the antibiotic, although data gathered from sinus puncture were not sufficient to permit an analysis of clinical outcomes in patients infected with resistant organisms in the studies included in our meta-analysis. To further investigate this question, we performed a MEDLINE search of the past 5 years looking for clinical outcomes of acute bacterial rhinosinusitis patients infected with resistant pathogens isolated from sinus puncture. We were unable to find any other relevant data. Such data should be collected in future studies.

Complications of bacterial rhinosinusitis can be serious and include brain abscess, orbital cellulitis, subdural empyema, and meningitis. We found no specific mention of such complications in more than 2,700 patients in the 28 analyzed trials. Large referral hospitals covering populations over wide geographic areas have reported only a handful of such complications over periods of 10 or more years (Johnson, Markle, Wiedermann, et al., 1988; Skelton, Maixner, and Isaacs, 1992). To our knowledge, there are no data showing that in clinical practice the use of more expensive, broad-spectrum antibiotics in uncomplicated cases would abort the development of these rare complications. It would be useful to collect such data in large-scale field studies because serious complications are important to prevent. However, avoiding delays in instituting therapy, obtaining adequate blood levels of antibiotics, and draining pus are certainly more important in aborting complications than the initial choice of antibiotic. The results of our meta-analysis should not be extrapolated beyond uncomplicated community-acquired acute bacteria rhinosinusitis. Serious cases and patients with predisposing factors should get aggressive treatment. On the other hand, evidence suggests that extensive use of antibiotics is associated with widespread development of resistant microorganisms in the community (Arason, Kristinsson, Sigurdsson, et al., 1996; Nissinen, Gronroos, Huovinen, et al., 1995; Seppala, Klaukka, Lehtonen, et al., 1995). This association would reduce the utility of new antibiotics in serious infections when they are most needed. Such potential misuse of these agents for the uncomplicated presentations of a common and benign syndrome may pose an ethical issue.

4a. In treatment of acute bacterial rhinosinusitis, what is the efficacy of antibiotics compared with that of placebo, and among the various antibiotics, what is their comparative efficacy?

The use of newer antibiotics may not reduce treatment failure rates substantially beyond what drugs such as amoxicillin or folate inhibitors can achieve. The evidence does not suggest a clinically meaningful superiority of newer, more expensive antibiotics over amoxicillin or folate inhibitors in treating uncomplicated acute and acute-on-chronic rhinosinusitis. It could be argued that by grouping all other antibiotics together we might have obscured some important and systematic differences between drug classes. There were too few studies in any single antibiotic group (penicillins, cephalosporins, macrolides, tetracyclines, and lincomycin) to allow for a meaningful meta-analysis of each class. However, simple inspection of Figures 12 and 13 suggests that there is no consistent superiority of any of the drug classes over the inexpensive reference agents. It is important to recognize, too, that many of the trials included in our analysis were carried out before the reports of rapidly rising levels of antibiotic-resistant organisms being isolated from community-acquired infections. As such, the significance with regard to treatment efficacy in the face of laboratory-diagnosed antibiotic resistance is unclear. This issue needs further and expedient investigation.

The trials in this meta-analysis did not include fluoroquinolones because we found no trial comparing these agents with an inexpensive reference drug, with the exception of one that addressed a variety of upper respiratory infections (Falser, Mittermayer, and Weuta, 1988). On the other end of the range of antimicrobial agents, the main analysis did not include penicillin VK, which is a favored choice in some practices, particularly in Scandinavia. Nevertheless, a sensitivity analysis including a large study involving penicillin VK gave similar results. The data with penicillin VK (Haye, Lingaas, Hoivik, et al., 1996) are more limited than the evidence available for amoxicillin. Amoxicillin is more active against susceptible strains of H. influenzae than is penicillin VK, and in a direct comparison of the two antibiotics, amoxicillin performed marginally better (Lindbaek, Hjortdahl, and Johnsen, 1996a). The cost of both antibiotics is approximately the same. Amoxicillin is theoretically preferable on the grounds of its antimicrobial spectrum of activity, but more clinical evidence is needed.

Most studies included in this meta-analysis were of small size. The total evidence was more substantial and of better quality for amoxicillin than for folate inhibitors. There were trends suggesting that newer antibiotics may offer some advantage over amoxicillin, and formal statistical significance might have been reached if more data were available. However, the chance of this advantage being large enough to be clinically important is small. One would need to treat 118 patients with newer, more expensive antibiotics instead of amoxicillin to prevent one case of clinical failure. Even if newer agents would avert 1 clinical failure per 32 patients treated with amoxicillin or 1 clinical failure per 16 patients treated with a folate inhibitor (the most extreme scenarios for the 95 percent confidence intervals obtained from the meta-analyses), this result is unlikely to be meaningful enough to justify their use as first-line therapy.

These results also should be viewed in the context of possible bias. Publication bias (Ioannidis, 1998) is unavoidable in this domain, where mostly small studies of modest quality are conducted, and there may be some unpublished studies on acute bacterial rhinosinusitis. However, in contrast to the usual situation for most meta-analyses, in which publication bias would lead to spuriously positive pooled results, the effect of publication bias in our meta-analysis would be to show that amoxicillin and trimethoprim-sulfamethoxazole are not better than newer agents. The same might have been true for bias related to poor methodologic quality, which as empirical evidence suggests, tends to favor new, experimental treatments (Schulz, Chalmers, Hayes, et al., 1995). That this bias is not shown even with the available data only strengthens the hypothesis that these inexpensive agents are probably adequate in the treatment of acute bacterial rhinosinusitis. We performed sensitivity analyses and found that, when only good quality trials with a Jadad score of at least 3 were considered, the estimates for all major endpoints of treatment effect were similar for all the main comparisons. For example, the risk ratio (RR) for clinical failures was 0.53 (95 percent CI, 0.34 to 0.84) for antibiotics vs. placebo, 0.86 (95 percent CI, 0.58 to 1.28) for other antibiotics vs. amoxicillin, and 1.01 (95 percent CI, 0.52 to 1.97) for other antibiotics vs. folate inhibitors.

In this meta-analysis, both groups of reference medications were generally as effective as other antibiotics. We found only two direct comparisons of the efficacy of folate inhibitors with amoxicillin that had adequate data, but with a total of 113 patients, the data were too few to reliably compare the agents, as indicated by the very wide confidence intervals of the pooled estimate for these two comparisons (folate inhibitors treatment failure RR, 0.5; 95 percent CI, 0.08 to 3.01) (Hamory, Sande, Sydnor, et al., 1979; Nyffenegger, Riebenfeld, Macciocchi, 1991). Trimethoprim-sulfamethoxazole has a broader spectrum than amoxicillin, covering amoxicillin-resistant H. influenzae and M. catarrhalis, and its use should largely satisfy those who would like to take antimicrobial resistance into account in prescribing treatment for acute community-acquired bacterial rhinosinusitis in the absence of extensive, conclusive data.

4b. What evidence do these comparative studies provide regarding side effects?

Major side effects from antibiotics are uncommon. Minor rashes and gastrointestinal complaints (e.g., nausea and diarrhea) are most common and were the most commonly reported side effects listed in the trials. Although not statistically evaluated in this report, the nature and incidence of side effects from antibiotics in the treatment of rhinosinusitis did not appear to differ from reported profiles of the various antibiotics for the treatment of infections at other sites at similar treatment dosage and duration.

5a. Are there data to support the use of other types of treatments for acute rhinosinusitis and acute bacterial rhinosinusitis, specifically: decongestants, steroids, antihistamines, drainage, sinus irrigation, others?

5b. What is the efficacy of antibiotics compared with that of other types of treatment?

5c. What evidence do any comparative studies provide regarding side effects?

Studies of ancillary treatments for acute bacterial rhinosinusitis are limited. All but one study tested ancillary treatments in the presence of antibiotic therapies. Although many of the antibiotic effectiveness studies also reported patient use of ancillary medications, this use was often not systematic and not the focus of the trials. No randomized trials assessing nondrug interventions for acute bacterial rhinosinusitis were found. None of the studies directly compared antibiotic treatment with nonantibiotic interventions. Several studies reported side effects in patients participating in these trials of ancillary therapies, but the major effects were attributed to the antibiotics also used in the trials and not to the ancillary treatments. Cost data suggest that patients are using large amounts of nonantibiotic treatments to address the symptoms of rhinosinusitis, but the evidence is not available to assess the potential benefits of the various treatments, either qualitatively or quantitatively.