Prevalence

As discussed in meta-analysis of diagnostic test studies, most of the estimates of prevalence of acute bacterial rhinosinusitis in patients seen by providers with sinus symptoms are in the range of 50 percent (Kuusela, Kurri, and Sirola, 1982; Laine, Maättä, Varonen, et al., 1998;van Buchem, Peeters, Beaumont, et al., 1995). We therefore use this prevalence for our baseline estimates. The estimates are tested across the full range of prevalence (0-100 percent).

Complications due to sinus puncture

No data are available as to the complication rate of sinus puncture. We therefore used consensus opinion of our technical experts to arrive at an estimate of a 1 percent complication rate.

Diagnostic tests

We derived estimates of diagnostic test performance from various sources. The values of the sensitivity and specificity of each diagnostic test used are shown in Table 17. For each test we used one set of test performance estimates for the baseline case. For our sensitivity analyses, we tested an available estimate that would bias against the performance of the test (i.e., relatively low sensitivity and specificity, thus increasing the proportion of patients not given the appropriate treatment). We also tested the idealized situation where each test has perfect sensitivity and specificity.

For the clinical criteria strategy we used the approach taken in the one trial that compared clinical signs and symptoms with sinus puncture (Berg and Carenfelt, 1988 ). As described in meta-analysis of diagnostic test studies (clinical examination compared with sinus puncture), Berg's paper provides data from which we were able to derive the sensitivity and specificity for four-item risk scores calculated by the presence of: (1) purulent rhinorrhea with unilateral predominance, (2) local pain with unilateral predominance, (3) bilateral purulent rhinorrhea, and (4) pus in the nasal cavity.

We used a "Berg score" of three or more (three or four of the signs or symptoms are present) which allowed for a moderately high sensitivity and high specificity. For comparison, we used the test performance estimates derived from a study that compared the physician's "overall clinical impression" with sinus radiography (Axelsson and Runze, 1976), which had poorer test performance.

As discussed in meta-analysis of diagnostic test studies, data exist for estimates of sinus radiography sensitivity and specificity for various methods of reading the films. All the methods include a series of three or four views including the occipitomental (Water's) view. However, the definition of a positive radiograph can include "opacification of sinus" only, "sinus fluid or opacity," or "sinus fluid or opacity or mucous membrane thickening." For the baseline decision analysis, we chose the definition of a positive radiograph that yielded the highest sensitivity while allowing for a moderate specificity, namely, "sinus fluid or opacity or mucous membrane thickening." We also performed sensitivity analysis using the lower bounds of the 95 percent confidence interval limits of the random effects pooled results.

We used the random effects pooled estimates of test performance for sinus ultrasonography from our meta-analysis of diagnostic test studies. Most of the available data are from studies in which ultrasonograms were performed by otolaryngologists. The results of the ultrasonography study with the poorest test performance (Laine, Maättä, Varonen, et al., 1998) were used in the sensitivity analysis.

No trials are available that compare either sinus CT or MRI to sinus puncture. Thus, no data are available from the literature of these tests' sensitivity or specificity. We therefore used consensus opinion of our technical experts to arrive at estimates of CT and MRI test performances. The consensus was that CT and MRI have high sensitivity but lower specificity. Lower test performances were also tested in the sensitivity analysis.

Sinus puncture includes endoscopic evaluation of the maxillary sinuses, sampling of any fluid present, and bacteriologic culture and sensitivity of the aspirated fluid. This strategy in conjunction with clinical history and examination is the most reliable method of diagnosing acute bacterial rhinosinusitis. However, because of its invasive nature, it is not routinely used. We included it in our model to provide an assumed perfect reference standard for comparison. Thus, the sensitivity and specificity of sinus puncture and culture are set at 100 percent.

Antibiotic side effects

Reviews of the literature of penicillin allergy generally quote an incidence of reactions or side effects at between 1 and 10 percent (Lin, 1992). The three studies found that discussed allergic reactions to amoxicillin (or ampicillin) (Bigby, Jick, Jick, et al., 1986; Caldwell and Cluff, 1974; Saxon, Beall, Rohr, et al., 1987) all report a rate of cutaneous or more severe drug reactions of approximately 5 percent in hospitalized patients. We used an estimate of 5 percent for severe drug allergies requiring change of antibiotic and tested the estimate in a wide range of 0 to 20 percent.

No data were found estimating the rate of minor side effects, which do not require a change in antibiotic. By consensus expert opinion, we estimated the rate of minor side effect to be somewhat lower than that of major side effect, or 4 percent. This estimate was also tested across a wide range of 0 to 20 percent.

Disease complications

No explicit evidence exists about the rate of complications of community-acquired, acute bacterial rhinosinusitis. However, using Gwaltney's (1996) estimate of approximately 1 billion cases annually of viral rhinosinusitis and a middle estimate that 1 percent of these cases are complicated by acute bacterial infection (Berg, Carenfelt, Rystedt, et al., 1986; Dingle, Badger, and Jordan, 1964), we can estimate 10 million cases of acute bacterial rhinosinusitis annually. The 1994 National Hospital Discharge Survey reported approximately 5,000 or fewer cases of intracranial abcesses. We estimated that approximately 20 percent of these cases were a result of bacterial rhinosinusitis (Bradley and Shaw, 1983; Small and Dale,1984). Extrapolating from these data, we used an estimate of major complication rate due to acute bacterial rhinosinusitis of 1/10,000 cases. To further bias our model toward treatment with antibiotics (specifically to avoid complications), we used this estimate for the complication rate only in patients not receiving antibiotics. We assumed that patients treated with antibiotics are fully protected against complications (their complication rate is 0). Patients who do not in fact have acute bacterial rhinosinusitis cannot develop a complication due to the disease.

Cure rates

Cure and improvement rate estimates were derived from the antibiotic treatment meta-analyses. We derived cure rate estimates for treatment with amoxicillin and folate inhibitors from the meta-analyses of the relevant trials. Cure rate estimates for symptomatic treatment were derived from the meta-analysis of the placebo arms of relevant trials. The cure rates for antibiotic and symptomatic treatments were varied across the 95 percent confidence intervals of the meta-analyses estimates.

We used consensus estimates of 67 percent cure and 17 percent improvement for patients who are seen with symptoms of rhinosinusitis but do not, in fact, have acute bacterial rhinosinusitis.

Outcome utilities

An important issue in the management of acute bacterial rhinosinusitis is the quality of life of the symptomatic patient. We used an arbitrary utility scale from 0 to 1 to assign a quality-of-life value to each of the clinical outcomes. The scale ranged from the value of death (0) to the value of full health (1). Cure without any adverse outcomes at the end of 2 weeks was set to a value of 1. Intermediate health states such as "improvement," "major side effect," or "complication due to disease" were assigned values shown in Table 17. The values were chosen from expert consultation and after patient interviews. In cases where more than one health state applied to a patient (such as "improvement" and "major side effect"), the quality-adjustment values were multiplied together to arrive at the final utility. For example, if a patient had a sinus puncture complication (quality adjustment = 0.6), a minor antibiotic side effect (0.9), and was cured (1.0), that patient's final utility would be 0.6 X 0.9 X 1.0, or 0.54.

In our interviews with patients, we were told that the effect on quality of life of rhinosinusitis symptoms is likely to vary significantly both between patients and in the same patient between episodes of acute bacterial rhinosinusitis. Some episodes are accompanied by more severe symptoms than others thus resulting in lower quality of life. As our baseline estimates of quality adjustments for improvement and no improvement were set for relatively more severe symptoms (which bias the results toward empiric antibiotic treatment), we also tested a scenario where symptoms of disease are relatively mild (milder than antibiotic side-effect symptoms) that would bias the findings away from empiric antibiotic treatment.

Costs

Cost estimates used in our analyses are presented in Table 17. Costs for antibiotics were estimated drug costs (Cardinale, 1997), including the pharmacy handling charges. All patients will have an office visit and will use prescription or over-the-counter symptomatic therapies; no additional costs are accrued because of initial office visit or symptomatic treatment.

Costs of diagnostic tests were derived from maximum allowable reimbursements from a managed care company as of November 1996, including radiologists' fees. Applying clinical criteria as a decision tool for treatment has no additional cost. The cost of a sinus puncture includes the reimbursement cost of obtaining a culture and antimicrobial sensitivities of the fluid sample.

The cost of a major antibiotic side effect includes the reimbursement cost of an additional office visit, the cost of therapy to treat the side effect, and the cost of switching antibiotics. The cost of a minor antibiotic side effect includes the cost of therapy to treat the side effect.

The cost of sinus puncture complication and disease complication includes the total costs of hospitalization including surgery, intravenous antibiotics, and so on. The actual figure used is from consensus expert opinion.

Through consensus expert opinion, we estimated that only 50 percent of patients who improve return for an office visit; the rest either continue symptomatic treatment on their own or speak to their provider without an office visit. The cost of improvement, thus, includes 50 percent of the cost of a return office visit and the cost of additional over-the-counter symptomatic therapy. It is assumed that these patients will not require an additional course of antibiotics.

We assumed that all patients who did not improve would return for an office visit. Those patients who had not been treated initially (because they were in the symptomatic treatment cohort or had a negative diagnostic test) would be given a course of amoxicillin or folate inhibitor. Those who had been treated initially (because they were in the empiric treatment cohort or had a positive diagnostic test) would be given a course of a newer, more expensive antibiotic. The cost of no improvement, thus, includes the costs of a return office visit and the cost of a course of antibiotics. We varied costs of improvement and no improvement, as shown in Table 17, to estimate the effect of using different approaches toward managing patients whose symptom did not resolve.

In addition, we tested the model under the assumption that the first antibiotic given to the patient was a newer, more expensive choice than amoxicillin. However, under this scenario we used the symptom resolution and side effect proportions of amoxicillin.

Cost per patient

At a 50 percent prevalence of bacterial rhinosinusitis, symptomatic treatment alone is the least costly strategy (see Table 18), with treatment guided by clinical criteria being less than $1.00 more costly. At this prevalence, the cost of treating the majority of patients with rhinosinusitis and some without (as dictated by the clinical criteria) approximately balances the cost of followup for the relatively high percentage of symptomatically treated patients whose symptoms fail to resolve. The additional cost of treating all patients (in the empiric treatment strategy) is greater than the savings of preventing the followup costs of the untreated patients who fail to resolve (in the symptomatic treatment and clinical criteria strategies).

As all the diagnostic procedures are fairly costly, the total cost per patient for each of the diagnostic procedure strategies was significantly more expensive than for the other strategies. The costs per patient were in direct relation to the cost of the diagnostic procedure performed.

Cost-effectiveness

The cost-effectiveness (cost divided by outcome utility, or effectiveness) allows us to determine the value of choosing each strategy, accounting for both the cost and effectiveness of each strategy. For each strategy, the cost (in dollars) per cure (utility = 1) is estimated (see Table 18). Clinical criteria-guided treatment was the most cost-effective, followed closely by symptomatic treatment alone. As the outcome utility for empiric antibiotic treatment was very similar to that of clinical criteria although the cost was greater, empiric antibiotic treatment was less cost effective.

The cost per healthy outcome of all the other diagnostic tests was much greater in direct relation to the cost of the diagnostic test.

Marginal cost-effectiveness

The costs and effectiveness of the strategies compared are graphically represented in Figure 19. The marginal cost-effectiveness of one strategy over another is the additional cost of the first strategy that achieves an additional quality-adjusted outcome of 1 over the second strategy [(Cost Strategy A - Cost Strategy B) / (Effectiveness Strategy A - Effectiveness Strategy B)]. In Figure 19, the more cost-effective strategies will be toward the upper left corner of the figure, having greater outcome utility and lower costs. The marginal cost-effectiveness between any two strategies is the inverse of the slope connecting the two points representing the strategies in the graph above. Only additional costs per additional utility are considered; thus the first strategy being considered must be to the upper right of the second, with greater cost and greater utility. Strategies that are more costly and less effective are "eliminated by strict dominance."

Figure

Figure 19. Treatment cost and effectiveness of each strategy in the multiple-strategies comparison model.
Note: CT - computerized tomography; MRI - magnetic resonance imaging

As presented in Table 18, symptomatic treatment is the least costly strategy at a prevalence of 50 percent. The marginal cost-effectiveness of clinical criteria-guided treatment is $18; thus, by using clinical criteria instead of symptomatic treatment, it would cost an additional $18 per patient to achieve one additional healthy outcome (outcome utility of 1). Empiric treatment, however, would require an additional cost of $3,677, and sinus puncture-guided treatment would require $35,865 beyond that to achieve an additional healthy outcome.

Risk profile

Presented in Table 19 are estimates of percentage of patients that would be cured, improve, not improve, and/or suffer a complication due to rhinosinusitis for each of the strategies studied. These estimates are shown for three different prevalences of acute bacterial rhinosinusitis: 25, 50, and 75 percent. In addition, for each prevalence, Table 19 shows the percentage of patients in each cohort given an antibiotic prescription, the number of prescriptions for amoxicillin written per case of real acute bacterial rhinosinusitis, the percentage of all patients who receive a prescription inappropriately (because they do not have acute bacterial rhinosinusitis), and the percentage of all patients who do not receive a prescription inappropriately (because they do have acute bacterial rhinosinusitis). For comparison purposes, the cost per patient of each strategy (including the costs of side effects, complications, and incomplete cures) is shown.

Table

Table 19. Risk profile of multiple strategies comparison model.

Symptomatic treatment alone (withholding antibiotics for all patients) is inferior to all other strategies at all prevalences shown in terms of cure rate, complication rate, and, by definition, percentage of patients with disease who are left untreated. All other strategies have very similar cure and complication rates, and all others but the idealized sinus puncture-guided treatment have similar numbers of prescriptions written per true case of acute bacterial rhinosinusitis and similar numbers of unnecessary prescriptions written for patients without acute bacterial rhinosinusitis.

Effectiveness

The effects of the various strategies across the full range of acute bacterial rhinosinusitis are shown in Figure 20. As it correctly diagnoses acute bacterial rhinosinusitis all the time, sinus puncture and culture avoids unnecessary treatment and maximizes appropriate treatment; it, thus, is always most effective. The effectiveness of the diagnostic test-guided treatment strategies are directly related to the test performance (sensitivity and specificity) of each test. As the test performances modeled are generally fairly high, the effectiveness of the tests is relatively high and stable across the range of prevalence.

Because of the increasing failure rate with symptomatic treatment, its effectiveness falls rapidly with increasing acute bacterial rhinosinusitis prevalence. The effectiveness of empiric antibiotic therapy rises with increasing prevalence, as the proportion of patients benefitting from antibiotic treatment rises. At low prevalence, the value of the empiric antibiotic therapy is low, but the rate of unnecessary side effects is high compared with that of the other strategies. As prevalence increases, the value of empiric treatment rises until it is equal to that of sinus puncture-guided treatment at 100 percent prevalence where all patients are being treated by both strategies.

Cost per treatment

As shown in Figure 21 (average cost per treatment), the costs for strategies involving diagnostic procedures (those lines above $100/treatment) do not vary considerably with acute bacterial rhinosinusitis prevalence. This is a result of the high costs of these procedures relative to the costs of further workup and treatment of the few patients who do not improve with empiric treatment.

Figure 22, an expanded version of Figure 21, reveals that the relative costs of empiric treatment, symptomatic treatment, and treatment guided by a set of clinical criteria differ depending on the underlying prevalence of acute bacterial rhinosinusitis in the cohort of patients. Symptomatic treatment and clinical criteria-guided treatment are very close to each other in cost across the full range of prevalence. Symptomatic treatment is marginally less costly at a prevalence lower than 66 percent; clinical criteria-guided treatment is less costly at a higher prevalence. The cost of empiric treatment is relatively flat across prevalence. Only at very high prevalence (>99 percent) is empiric treatment less costly than clinical criteria.

Cost-effectiveness

Figure 23 shows the relative cost-effectiveness for the four least costly strategies (symptomatic, clinical criteria-guided, empiric, and radiography-guided treatment). As in the graph of cost across prevalence (Figure 22), symptomatic treatment and clinical criteria-guided treatment are similar across the range of prevalence. Clinical criteria-guided treatment is the most cost-effective strategy when prevalence of acute bacterial rhinosinusitis is between 41 and 95 percent.

Below a prevalence of 41 percent, symptomatic treatment is most cost effective; and at a prevalence greater than 95 percent, empiric treatment is most cost effective. At a low prevalence, most patients' symptoms will respond spontaneously; thus, antibiotic treatment can be deferred and still achieve good average outcome and minimal cost of followup for patients whose symptoms fail to resolve. At very high prevalence, the cost of not treating those patients with acute rhinosinusitis who have negative clinical criteria is greater than the cost of treating everyone, and the outcome is poorer. Therefore, empiric treatment is most cost effective at very high prevalence.

Sinus radiograph

We used the same estimates for radiography test performance that were used in the multiple-strategies comparison model. These assumed that a sinus radiograph is read as positive for acute bacterial rhinosinusitis if there is "sinus fluid or opacity of mucous membrane thickening." For the sensitivity analysis, we used the lower bounds of the 95 percent confidence interval limits of the random effects pooled results and the idealized situation of perfect sensitivity and specificity.

Clinical criteria

We used the same estimates for test performance of applying a set of clinical criteria to diagnose acute rhinosinusitis that was used in the multiple-strategies comparison model. These used a "Berg score" of three or more as described above (Berg and Carenfelt, 1988). For sensitivity analysis, we used an estimate of lower test performance from expert opinion.

Effectiveness

Whether symptom-free days or quality-adjusted days with mild, moderate, or severe symptoms are estimated, the effectiveness of empiric antibiotic treatment, clinical criteria-guided treatment, and radiography-guided treatment were very similar. These strategies outperform symptomatic treatment because of the latter's high rate of failure to cure. However, in the scenario of mild rhinosinusitis symptoms (where rhinosinusitis symptoms are milder than symptoms of antibiotic side effect) all strategies yield similar estimates of effectiveness.

Cost per patient

At a 50 percent acute bacterial rhinosinusitis prevalence, the cost per patient seen was least for those given symptomatic treatment alone. Because of the cost of amoxicillin for all patients, empiric treatment was somewhat more costly. The cost of clinical criteria-guided treatment was only slightly higher than symptomatic treatment. It was less costly than empiric treatment because the cost of "excess" amoxicillin given to all patients without acute bacterial rhinosinusitis was greater than the cost of the few patients with acute bacterial rhinosinusitis left untreated because of inaccurate clinical criteria. Because of its high cost compared with amoxicillin, sinus radiography-guided treatment is by far the costliest strategy. (Note that the cost per patient is independent of the outcome utility or the severity of disease symptoms.)

Cost-effectiveness

At the given prevalence of 50 percent, clinical criteria-guided treatment is the most cost-effective strategy in all the given scenarios. This strategy incurs no additional up-front cost, yields fewer treatment failures and disease complications than symptomatic treatment alone, yet avoids the additional cost and lessened effectiveness due to additional antibiotic side effects in patients who do not have acute bacterial rhinosinusitis. However, as the severity of a patient's rhinosinusitis symptoms lessen, the cost-effectiveness of symptomatic treatment alone improves, such that with mild and moderate disease symptoms the cost-effectiveness of symptomatic treatment and clinical criteria-guided treatment are very similar.

Marginal cost-effectiveness

The cost and effectiveness for each strategy for the symptom-free-days scenario are shown in Figure 27 and Table 23. The marginal cost-effectiveness for the other scenarios is shown in Table 24.

Figure

Figure 27. Cost and effectiveness of strategies in the symptom-duration model.
Note: Effectiveness measured in symptom-free days.

Although for each scenario symptomatic treatment is the least costly, it is also the least effective. As the cost of clinical criteria-guided treatment is only slightly greater than symptomatic treatment and is more effective, the marginal cost-effectiveness (the additional cost required of the second strategy compared with the first to achieve an additional unit of effectiveness) is small. Empiric antibiotic use, on the other hand, achieves greater effectiveness but at a much greater cost.

As radiography-guided treatment is both less effective than symptomatic treatment and more costly, it is eliminated by strict dominance.

The more severe the rhinosinusitis symptoms, the smaller the marginal cost-effectiveness of both clinical criteria-guided treatment and empiric treatment.

Effectiveness

At low acute bacterial rhinosinusitis prevalence, the effectiveness (number of symptom-free or quality-adjusted days over a 14-day course) of all strategies is very similar as the course of symptoms is driven primarily by the cure rate of diseases mimicking acute bacterial rhinosinusitis. The number of symptom-free days estimated across the range of prevalence for the four strategies is shown in Figure 28.

With rising prevalence, symptomatic treatment alone rapidly becomes less effective than the other strategies as no patients with acute bacterial rhinosinusitis are being treated with antibiotics. In the other three strategies (empiric, radiography-guided, and clinical criteria-guided treatment) the relative effectiveness remains similar across the range of prevalence. As more patients with disease are left untreated with clinical criteria-guided treatment than with radiography because of poorer test performance (sensitivity and specificity, see Table 17), the effectiveness of the former is poorer. Likewise, all patients with disease are treated with empiric treatment; thus, the effectiveness of this strategy is always greatest.

The relative effectiveness across prevalence for the symptom severity scenarios is similar to that shown in Figure 28 for mild, moderate, and severe rhinosinusitis symptoms. The milder the symptoms, the closer the relative effectiveness is and the greater the absolute value of the quality-adjusted symptom days.

Cost per treatment

As shown in Figure 29, the cost per patient of radiography-guided treatment rises with increased prevalence, as the proportion of patients being treated (and thus having side effects), and also not being cured and having disease complications rises. At no prevalence is sinus radiography-guided treatment less costly than the other strategies.

Figure

Figure 29. Average cost per treatment of strategies in the symptom-duration model.
Note: Tx - treatment

The cost of empiric treatment rises only minimally with increasing prevalence as the cost of treatment and side effects are stable and only the proportions of uncured patients and disease complications rise; however, these proportions remain relatively low as all patients are being treated.

The cost of symptomatic treatment rises rapidly with increasing acute bacterial rhinosinusitis prevalence. This is because of the increasing percentage of patients with acute bacterial rhinosinusitis who are not being cured and thus require a return office visit and the patients who are developing disease complications (at a higher rate than if they were treated). The cost of clinical criteria-guided and radiography-guided treatment rises rapidly with increasing prevalence, since at higher prevalence, there is increased use of amoxicillin which includes increased side effects and an increased overall rate of disease complication and treatment failures.

At acute bacterial rhinosinusitis prevalence below 24 percent, symptomatic treatment per patient is least costly. Clinical criteria-guided treatment is least costly from 24 to 91 percent. Thereafter, empiric treatment is least costly. However, the costs of symptomatic and clinical criteria-guided treatment are very similar at lower prevalence, as are the costs of empiric treatment and clinical criteria-guided treatment at high prevalence.

Cost-effectiveness

The cost-effectiveness of the symptom-free-day scenario across the full range of prevalence for the four strategies is represented in Figure 30. For each strategy, as the prevalence of acute bacterial rhinosinusitis rises, the cost increases (as a result of the increased persistence of symptoms requiring followup, the increased rate of disease complications, and the increased use of antibiotics in the test-guided strategies) and the effectiveness decreases (as a result of the longer duration of symptoms and the likelihood of disease complications with acute bacterial rhinosinusitis than without). Thus the value of the cost-effectiveness rises for each strategy with increased prevalence. (Note that the lower the value, in dollars per symptom-free or quality-adjusted day, the more cost effective the strategy.)

In the symptom-free-day scenario, the cost-effectiveness of symptomatic treatment and clinical criteria-guided treatment are similar across the range of prevalence. Symptomatic treatment is more cost effective (fewer dollars per symptom-free day) below a prevalence of 25 percent. Clinical criteria-guided treatment is the most cost effective at a prevalence between 25 and 83 percent. Empiric treatment is most cost effective at prevalence above 83 percent.

The range of cost-effectiveness for each strategy under each scenario (symptom-free and quality-adjusted days) is presented in Table 25 under "base case scenarios." As radiography is never most cost effective, it is left off the table. When rhinosinusitis symptoms are mild, empiric treatment is most cost effective only at prevalence rates higher than when symptoms are more severe. With mild symptoms, empiric treatment is most cost effective only when prevalence exceeds 97 percent; however, with severe symptoms, empiric treatment is most cost effective when prevalence exceeds 90 percent. Likewise, symptomatic treatment is most cost effective at higher prevalence rates when symptoms are less severe. With mild symptoms, symptomatic treatment is most cost effective when prevalence is less than 49 percent, while with severe symptoms, symptomatic treatment is most cost effective only at prevalence rates less than 31 percent.

Table

Table 25. Range of cost-effectiveness in various scenarios.

Clinical criteria test performance

Lowering the test performance (sensitivity and specificity) of the clinical criteria test shrank the range of prevalence in which using clinical criteria to discriminate between those who should be treated symptomatically and those who should be given amoxicillin.

Cure rates with and without amoxicillin

We used the extremes of the 95 percent confidence intervals for the Kaplan-Meier curves of cure rate with and without amoxicillin from the Lindbaek, Hjortdahl, and Johnsen (1996a) study. In one scenario, we assumed amoxicillin is maximally more effective than placebo (or symptomatic treatment) by using the high cure rate estimate for amoxicillin and the low cure rate estimate for placebo. This scenario implies that empiric treatment is most cost effective at lower prevalence than the base case scenario at high prevalence and that symptomatic treatment is most cost effective only at very low prevalence.

The opposite assumption, that the cure rate with amoxicillin is more similar to that of placebo resulted in the conclusion that symptomatic treatment, although not always most effective, is most cost effective across the full range of prevalence, no matter the severity of disease.

To reproduce the effect of antibiotic resistance, we lowered the cure rate with amoxicillin and left that of placebo at its base level. Again, symptomatic treatment was most cost effective across prevalence for disease severity-adjusted outcome. This is because of the lessened effectiveness of the antibiotic. Only if the goal of treatment is the minimization of any symptom is clinical criteria-guided treatment or empiric antibiotic treatment cost effective at very high prevalence.

Antibiotic side-effect rate

If the risk of antibiotic side effects is nil (as opposed to 5 percent in the base case scenario), then use of antibiotics becomes more cost effective as costs (of treating side effects) are lower and effectiveness is higher (as outcome utility is not decreased for any patients because of side effects). The range of prevalence in which both clinical criteria-guided treatment and empiric treatment are most cost effective shift downward somewhat.

If the risk of antibiotic side effect is very high (at 20 percent), then symptomatic treatment is always most cost effective in mild disease (where the symptoms of rhinosinusitis are less severe than those of side effect would be). The range of prevalence in which both clinical criteria-guided treatment and empiric treatment are most cost effective shift upward and, as in all scenarios, the more severe the symptoms (in comparison to the quality of life of having a side effect), the more cost-effective the use of antibiotics becomes.

Cost of initial antibiotic

In our final sensitivity analysis, we tested the scenario in which a newer, more expensive antibiotic is used as initial treatment (either empirically or after applying clinical criteria). As the literature shows that cure rates for newer, more expensive antibiotics are similar to those of amoxicillin and folate inhibitors (see Attachment C), we used the same cure rates estimated for amoxicillin. As a more expensive antibiotic was used initially, the followup costs for those who fail treatment with the medication are similarly more expensive, as shown in Table 21.

Because of the high cost of treatment, symptomatic treatment for mild and moderate disease is most cost effective across the range of prevalence. Even for rhinosinusitis with severe symptoms, the cost-effectiveness of symptomatic treatment is greatest until very high prevalence. Empiric treatment with amoxicillin ranges in cost-effectiveness from approximately $2 to $5 per quality-adjusted day (across the full range of prevalence and in all base case scenarios). Our hypothetical expensive antibiotic meanwhile ranges in cost-effectiveness from $4 to $20 per quality-adjusted day.

Form of the cure-rate function

The model was also tested using linear and exponential functions instead of the Weibull function to describe the change in daily cure rate (as described above). Although there were small quantitative differences in the costs, effectiveness, and cost-effectiveness thresholds, qualitatively the results were the same (data not shown). The Weibull function more closely describes the biology of infectious diseases (Kalbfleisch and Prentice, 1980) and more closely fits the Kaplan-Meier curves from Lindbaek's trial (1996a) than do linear or exponential functions. Therefore, we believe that the results from the Markov model using the Weibull function are more robust, and that the results from models using the other curves can be disregarded.