A Number-Needed-to-Treat Analysis of the Use of Respiratory Syncytial Virus Immune Globulin to Prevent Hospitalization FREE

INCREASED EFFORTS to contain health care costs have directed the focus of medical decision makers toward demonstrable benefits of proposed new therapies. The development of clinical trials methods has greatly simplified this task by enabling the identification of treatments that meet statistical criteria for efficacy. Although a statistically significant effect is accepted as one criterion for adopting a new therapeutic regimen, clinical decisions also involve perceived benefits to all patients, toxicity, cost, and the administrative burden to patients and physicians. A proposed intervention may be shown to be efficacious, but the effect is too small to make it worth implementing. In some contexts, patients and the public may be better served by withholding treatment for those at low risk while recommending treatment for those at higher risk. As an example, the administration of respiratory syncytial virus immune globulin (RSV-IG), a respiratory syncytial virus (RSV) hyperimmune globulin, has been shown to be efficacious, but it may offer marginal benefits for low-risk patients at substantial cost.¹

The use of RSV-IG has been approved for the prevention of lower respiratory tract infections caused by RSV. Its use has been shown to reduce the incidence of hospital admission for RSV infection among premature infants with and without a history of bronchopulmonary dysplasia (BPD). Prophylaxis with RSV-IG has been recommended by the Committee on Infectious Diseases and Committee on Fetus and Newborn of the American Academy of Pediatrics² for infants born at 32 weeks or less of gestational age and infants younger than 2 years with BPD. The cost of RSV-IG per RSV season may be more than $5000 per patient, however, and the potential effect of this treatment on pediatric practice is considerable. In this article, we consider evidence for the effectiveness and cost of this drug and apply a method of cost-benefit analysis to aid in recommendations for its use.

Estimates of the effectiveness of RSV-IG in preventing hospital admission for RSV infection have been reported in 3 randomized controlled trials.^{1 ,3 - 4} Data from these trials are used to estimate the number of infants needed to be treated (NNT) with RSV-IG to avoid 1 hospital admission for RSV infection. Further analyses are performed to estimate the threshold number needed to treat (T-NNT) that would justify the use of RSV-IG on economic grounds. Together these statistics form decision rules for guiding clinical recommendations.

PREVENT¹ is described as the pivotal trial to assess the safety and efficacy of RSV-IG⁵ and is the only randomized, placebo-controlled study of this drug. This trial was conducted on 510 children. Eligibility criteria included an age of 24 months or younger and a diagnosis of BPD, with a requirement of supplemental oxygen within the past 6 months, or an age of 6 months or younger and being born at 35 weeks or less of gestation. Children were randomly assigned to receive either 750 mg/kg of body weight of RSV-IG or placebo (1% albumin) intravenously every 30 days for at least 4 months. The primary end point of the trial was hospital admission for RSV lower respiratory tract infection.

An earlier study of the efficacy of RSV-IG,³ supported by the National Institute of Allergy and Infectious Diseases, enrolled 182 infants meeting inclusion criteria of younger than 48 months of age with congenital heart disease or BPD, or premature (≤35 weeks' gestational age) and younger than 6 months. Patients were randomly assigned to receive either 750 mg/kg of RSV-IG each month or to a control group that received no RSV-IG. Hospital admission for RSV infection was a primary end point. This trial was not completely blinded to treatment assignment,^{6 - 7} not placebo controlled, and did not include separate analyses for the important subgroups of premature infants and those with BPD.

A third randomized controlled trial of 750 mg/kg of RSV-IG was conducted on 416 infants with congenital heart disease or cardiomyopathy.⁴ Hospital admission for RSV infection was a primary end point of the cardiac trial. This trial was not placebo controlled and did not include separate analyses for premature infants and those with BPD.

The NNT is a clinically useful statistic of treatment effect.^{8 - 10} Algebraically, NNT is the reciprocal of the absolute risk reduction (ARR), or the difference in the rates of adverse outcome between treated and control patients divided into 1. It provides an estimate of the number of patients who would need to be treated with a given therapy (RSV-IG) to prevent 1 adverse outcome (RSV hospital admission).

In contrast to less intuitive probability-based statistics of treatment effect such as risk reduction and relative risk reduction, the NNT better facilitates the interpretation of patients treated and is being used more commonly as a tool for clinical decision making.^{9 ,11} To illustrate, if 100 high-risk infants are observed during the RSV season and the risk of hospital admission with RSV lower respiratory tract infection without treatment is 15.1%, we would expect 15.1 infants to be admitted to the hospital. If, however, 100 high-risk patients are treated with RSV-IG during the season and the risk of hospital admission is only 9%, we would expect 9 infants to be admitted to the hospital. Thus, treating 100 patients would be expected to prevent 6.1 (15.1−9) hospital admissions, meaning that 16.4 (100÷6.1) would need to be treated to prevent 1 hospital admission.

The T-NNT is calculated by considering the costs incurred by treating patients, the costs saved by preventing an adverse health outcome, the costs that might be incurred as a result of preventing the adverse health outcome, and the costs incurred by the treatment of adverse events resulting from treatment.¹² If T-NNT is sufficiently above the value of the calculated NNT, it would be economically reasonable to administer treatment. If a patient's risk without treatment is low enough, the treatment has only marginal benefit, or both, the T-NNT will be below the calculated NNT, and treatment would not be recommended.

The T-NNT is computed according to the following formula:

where Cost_target indicates the cost of treating 1 patient admitted to the hospital for RSV infection; Value_target, the willingness to pay to avoid hospital admission for RSV infection; Cost_treatment, the cost of treating 1 patient with RSV-IG; Cost_AEi, the cost of treating 1 adverse event resulting from RSV-IG infusion (multiple events indexed by i); Rate_AEi, the proportion of treated patients who will suffer an adverse event; and Value_AEi, the willingness to pay to avoid an adverse event.

The formula for the T-NNT is derived from principles of cost-benefit analysis. The threshold represents equality between costs and benefits of preventing hospital admissions for RSV. The cost side of the equation includes the cost of RSV-IG treatment plus the cost of adverse events attributable to treatment, adjusted for the probability of the adverse event occurring plus the willingness to pay to avoid the adverse event. Benefits include the cost savings from the averted hospital admission for RSV infection plus the value to the patient or the family of averting a hospital admission for RSV infection. In this analysis, the benefit of averting a hospital admission for RSV infection is measured by the willingness to pay to avoid the hospital admission—the theoretically correct measure of benefits. Compared with a standard cost-benefit analysis, the T-NNT gives the point at which the net benefits of the therapy are 0. For example, if the solution of the T-NNT formula gives a value of 9, then treating 9 patients to prevent 1 target event would result in exactly 0 net benefits, expressed in dollar terms. If more than 9 patients must be treated to prevent 1 target event, then the costs of treatment outweigh the benefits. If fewer than 9 patients must be treated, benefits exceed costs.

The estimates of the willingness to pay required for the T-NNT formula can be established in various ways. The ideal strategy would involve individual interviews with a sample of medically sophisticated parents or the general public who are familiar with the vocabulary of common childhood illnesses and complications of treatment. A valuable alternative that allows for estimates of the willingness to pay without exceeding the vocabulary and exhausting the patience of research subjects is to sample health professionals.

The subjects were 35 physicians and 4 nurses attending a workshop on the treatment of infectious diseases. A list of facts about RSV lower respiratory tract infection was presented, followed by a question about the maximum amount the subject would pay for a treatment that would prevent a child of theirs from being admitted to a hospital. The facts and questions are reproduced in Table 1.

In willingness-to-pay methods, it is assumed that the amount a person would be willing to pay to avoid an event incorporates all indirect costs of the event.¹³ Willingness to pay thus implicitly includes opportunity costs, costs of pain and suffering, and costs of lost productivity due to the event.

Existing data are combined with plausible assumptions in a sensitivity analysis to generate a range of T-NNTs. Sensitivity analysis allows decisions about the appropriate use of an investigational therapy when treatment costs and benefit variables are subject to substantial uncertainty. Computed T-NNTs are compared with observed NNTs to determine whether existing data support the use of RSV-IG.

If the NNT is less than the T-NNT, ie, if it is possible to treat fewer than the threshold number of patients to prevent 1 hospital admission, then the benefits of prophylaxis outweigh the costs. If, however, the NNT is greater than the T-NNT, ie, if more than the threshold number of patients must be treated to prevent 1 hospital admission, then the benefits of prophylaxis do not justify the costs.

No clinically important heterogeneity of treatment effects was noted across the 3 randomized controlled trials of RSV-IG. The relative risk reduction was 0.41 in the PREVENT study,¹ 0.31 in the cardiac study,⁴ and 0.57 in the National Institute of Allergy and Infectious Diseases study.³ Hospital admission rates for RSV infections among control and RSV-IG–treated infants as reported in the 3 trials are presented in Table 2. The ARR, NNT, and 95% confidence interval (CI) around the number needed to treat are presented. Data reported in the 3 studies are displayed for all patients and for subgroups of patients.

Based on data for all subjects in the 3 trials^{1 ,3 - 4} (row 1), 16.4 infants (95% CI, 12.5-23.8) will require treatment with RSV-IG to prevent 1 hospital admission for RSV infection. The NNT varies greatly among subgroups. Of infants from the PREVENT study who are premature, 6 months of age or younger, and without BPD (row 2), 62.5 must be treated to avoid 1 hospital admission. No upper limit on the NNT can be computed for this premature subgroup. The lower bound on the ARR CI is negative (−0.055), indicating that an increased risk of hospital admission with treatment cannot be ruled out. Of infants with BPD and a recent oxygen requirement from the PREVENT study (row 3), approximately 12 would need to be treated to prevent 1 hospital admission for RSV infection.

Because of its immunoglobulin action, RSV-IG may be effective in reducing the severity of respiratory illness due to causes other than RSV. Trial data indicate a significant reduction in the number of hospital admissions for non-RSV respiratory illness following the prophylactic administration of RSV-IG. Combining data from 2 trials for which hospital admissions for respiratory illnesses other than RSV are reported,^{1 ,4} non-RSV hospital admissions occurred in 12.4% of control or placebo-treated patients and 7.4% of RSV-IG–treated patients. In total, hospital admissions for all causes occurred in 26.6% of control or placebo-treated patients and 16.6% of RSV-IG patients. The NNT for preventing either RSV or non-RSV hospital admissions is therefore 10.0 ((95% CI, 6.6-21.0).

Estimates of the costs of care, the rates of adverse events, and the value of events are required for the calculation of the T-NNT. Table 3 presents plausible estimates of these variables under low, midrange, and high assumptions. Savings associated with RSV-IG require estimates of the average cost of treating 1 hospital admission for RSV infection for high-risk infants. Because RSV-IG treatment would be of interest primarily for high-risk infants who would likely entail hospital charges much greater than an average patient with RSV-related illness, estimates of the costs of care for these patients should be at the higher end of the range for all patients treated for RSV infection. We estimate that these costs will vary from $15000 to $25000. A thorough literature search reveals that little empirical evidence exists to validate these values. One published estimate corresponds to our lower value,¹⁴ and one estimate corresponds to our upper value.¹⁵ Both estimates include costs of intensive care for a proportion of patients. To put these estimates in perspective, at our institution (Arkansas Children's Hospital, Little Rock), 12% of all infants admitted to the hospital with diagnoses of RSV pneumonia or bronchiolitis incur charges greater than the upper value ($25000), and 16% of all infants incur charges greater than the lower value ($15000). Thus, we consider these estimates representative of high and low true charges for hospital admission for RSV infection in high-risk patients.

The midrange estimate of the value of avoiding a hospital admission for RSV was $5787. This figure represents the mean amount our health care provider sample would be willing to pay to avoid hospital admission. The low figure represents the median amount for this sample, and the high figure represents the top decile of this sample. Estimates ranged from $500 to $15000 for all but 1 subject, who would be willing to pay more than $70000. Analyses of the effect of a limited number of sample characteristics on the willingness to pay, including whether the respondent had a child who was admitted to a hospital with RSV infection, proved of little value. Thus, no attempt was made to adjust the willingness to pay for differences in sample characteristics.

Estimates of the cost of RSV-IG prophylaxis were computed from the charges for an average outpatient infusion session at our institution ($142), an average of 4.5 infusions per patient, and charges for RSV-IG at our institution ($691 for a 2.5-g vial, $327 for a 1-g vial). The midrange estimate is based on the treatment of a 4.6-kg infant, the mean weight of subjects in the PREVENT trial. Infants weighing 4.6 kg will require one 2.5-g vial and one 1-g vial of RSV-IG at each of 4.5 infusion visits (based on a 750-mL/kg dose). The low estimate is based on the treatment of an infant weighing 2.5 kg, or 1 SD (2.1 kg) below the mean of PREVENT subjects. These infants will require one 2.5-g vial of RSV-IG at each infusion. The high estimate is based on the treatment of an infant weighing 1 SD above the mean, or 6.7 kg. These infants will require two 2.5-g vials and one 1-g vial of RSV-IG at each of 4.5 infusions.

Three possible adverse events of RSV-IG infusion were identified. Because the immune globulin will interfere with the activity of some vaccines, immunization with measles, mumps, and rubella virus vaccine and varicella-vaccine should be deferred for 9 months after the last RSV-IG dose is administered.² The available data do not support the need for reimmunization with any routinely administered vaccine.² We estimate that between 80% and 95% of all infants treated with RSV-IG will have delays in their normal schedule of immunization.

Failed intravenous access is a second possible adverse event. In the National Institute of Allergy and Infectious Diseases trial of RSV-IG compared with no treatment, 60% of infants had at least 1 problem with intravenous access.³ We estimate that between 55% and 65% of infants will have multiple intravenous insertions.

The use of RSV-IG is also associated with drug-related adverse events. Complications related to fluid volume and flow rate were observed in the PREVENT trial and the National Institute of Allergy and Infectious Diseases trials and are known to occur with the infusion of immune globulin.^{16 - 17} Adverse events judged to be potentially related to the study drug in the PREVENT trial included fever (occurring in 6% of patients receiving RSV-IG), respiratory distress (2%), vomiting or emesis (2%), and wheezing (2%). Diarrhea, rales, fluid overload, tachycardia, rash, hypertension, hypoxia, tachypnea, gastroenteritis, injection site inflammation, and overdose effect were each observed in 1% of patients who were given RSV-IG.⁵ A total of 15% of patients in the RSV-IG arm of the PREVENT trial had 1 or more adverse events. Taking the 95% confidence interval of around 15%, we estimate the rate of adverse drug-related events to range from 11% to 19%.

The costs of treating adverse events are considered minimal. Delayed immunizations cost no more than timely ones, multiple attempts at intravenous access require only additional needles and alcohol swabs, complications related to fluid volume are easily managed and require only a short additional stay in the infusion clinic, and other adverse reactions are managed with few additional resources. The midrange value represents the increased cost of staying 50% longer in the infusion unit.

The values placed on avoiding RSV-IG–related adverse events were determined from responses of the health care provider sample. The midrange amounts represent the mean willingness to pay to avoid each adverse event. The low figures represent the median amounts they would be willing to pay, and the high figures represent the top decile for each adverse event.

Table 4 presents estimates of the T-NNT based on the parameter values reported in Table 3. The T-NNT formula has been solved for each combination of low, midrange, and high estimates of the numerator (benefits) and denominator (costs). Viewing the estimates in Table 4 as a matrix, the T-NNT values on the diagonal represent estimates based on the low, midrange, and high assumptions for both the numerator and the denominator. For example, using all midrange assumptions, as described in Table 3, produces a T-NNT estimate of 4.8 high-risk children. The largest T-NNT estimate is generated by assuming high parameter estimates for the numerator (high cost of hospital admissions for RSV infection plus high willingness to pay to avoid a hospital admission) and assuming low parameter estimates for the denominator (low costs of treatment with RSV-IG plus low rates of adverse events attributable to RSV-IG, and low willingness to pay to avoid adverse events). Under these assumptions, the T-NNT is 9.0 high-risk infants. This T-NNT estimate is the most liberal and favorable to recommendations for the use of RSV-IG.

Comparing the T-NNT estimates in Table 4 with the NNT estimates in Table 2 permits an assessment of treatment recommendations for the use of RSV-IG. In general terms, if the T-NNT is sufficiently above the value of the calculated NNT, it would be appropriate from an economic perspective to administer treatment. Four specific treatment recommendations can be derived through comparisons of the T-NNT, NNT, and the CI around the NNT. If the T-NNT exceeds the upper bound on the 95% CI around the NNT, there is strong inference that treatment will do more economic good than harm, and treatment is strongly recommended. Neither the overall sample of high-risk infants nor either subgroup meets this criterion based on the estimated T-NNT values in Table 4.

If the T-NNT is less than the lower bound on the CI around the NNT, there is strong inference that the costs and risks of the proposed procedure outweigh the benefits, and treatment is not recommended. A comparison of Table 2 and Table 4 reveals that this criterion is met for the NNT estimates associated with the entire sample of high-risk infants and for preterm infants.

Treatment is weakly recommended if the T-NNT falls between the NNT and the upper bound of the CI and weakly not recommended if the T-NNT falls between the NNT and the lower bound. Applying these criteria to the subgroups of Table 2 and using the 2 most liberal estimates of the T-NNT (6.6 and 9.0), a weak recommendation not to treat can be made for infants with BPD. The recommendation against the treatment of infants with BPD becomes stronger if any of the other T-NNT estimates are used.

If a reduction in the number of hospital admissions for non-RSV respiratory illness is included as an effect of RSV-IG prophylaxis, the NNT to prevent all hospital admissions is 10.0 (95% CI, 6.6-21.0). In this case, using the 2 most liberal T-NNT estimates results in a conclusion that prophylaxis is weakly not recommended. Recommendations against treatment are stronger if any other T-NNT estimate is used.

An additional way of viewing the economic value of using RSV-IG is to consider at what attack (hospital admission) rate the benefits of treatment would outweigh the costs of treatment. Figure 1 illustrates such an analysis. The graph is a curve of the NNT by the risk of hospital admission if untreated, considered for a drug of known efficacy. The efficacy of RSV-IG was estimated from all 3 trials as a relative risk reduction of 0.40. The point at which this curve crosses the T-NNT line is the economic break-even point. The benefits of treatment would outweigh the costs of treatment for patients at an untreated risk of hospital admission above the intersection of the curve and the T-NNT line. Considering the most liberal T-NNT of 9.0, this point is at an untreated risk of 0.28. Using the midrange T-NNT of 4.8, this point is at an untreated risk of 0.52. The risk of hospital admission for RSV lower respiratory tract illness if untreated must exceed 0.28 to justify prophylaxis on economic grounds.

A similar sensitivity analysis can be performed to determine the hospital admission costs that would be required to equate the NNT and the T-NNT, thereby justifying the use of RSV-IG on economic grounds. Using midrange T-NNT estimates, a cost of hospital admission of $82500 is required to raise the T-NNT to 16.4. Using T-NNT estimates that maximize the benefits of RSV-IG (high numerator-benefits, low denominator-costs) a hospital admission cost of $54000 is required to equate the NNT and the T-NNT for all subjects.

Respiratory syncytial virus is the most frequent cause of bronchiolitis and pneumonia in infants. About 1% of all infants who become ill with RSV lower respiratory tract disease require admission to a hospital.¹⁸ A small proportion of high-risk infants die of RSV disease.¹⁹ Antiviral agents have shown limited therapeutic value for ill children,²⁰ stimulating efforts to develop hyperimmune globulin prophylaxis for RSV infection.

Administering RSV-IG has been shown in 2 trials to reduce the number of hospital admissions for RSV infection among a combined sample of infants younger than 24 months with BPD and young infants with a history of prematurity (<35 weeks).^{1 ,3} In a third trial of infants with congenital heart disease or cardiomyopathy, administering RSV-IG did not result in significantly fewer hospital admissions but was associated with severe life-threatening surgically related adverse events and surgically related fatalities.⁴ The pediatric community is faced with the task of assessing these and other data to decide whether to implement large-scale administration of RSV-IG, limit its use to selected high-risk infants, or not use the medication at all.

We have applied cost-benefit procedures to assess suggested indications for the use of RSV-IG.^{2 ,21} We have estimated the number of high-risk infants who would need to be treated to prevent a single hospital admission for RSV infection. We have also calculated the T-NNT to justify the use of the medication on economic grounds. The calculation of these values permits clinicians to apply decision rules for whether to accept recommendations for a given treatment.¹² Our analysis indicates that RSV-IG is not sufficiently effective under the most generous assumptions to justify its use with premature infants younger than 6 months without BPD. Furthermore, the T-NNT is sufficiently low under plausible assumptions to result in a weak recommendation not to treat infants with BPD.

Consensus statements,²¹ the American Academy of Pediatrics,² PREVENT authors,¹ and many institutional guidelines are recommending the use of RSV-IG for infants who were born at 32 weeks or younger gestational age. This recommendation departs from the inclusion criterion of younger than 35 weeks' gestational age of the PREVENT trial. Even this more stringent indication does not appear to be justified from the PREVENT data. The mean (SE, SD) gestational age of infants in the PREVENT trial was 28.5 (0.21, 3.39) weeks. Although the exact percentage of patients with a gestational age of 32 weeks or younger is unknown, clearly a large percentage would have been 32 weeks or younger in gestational age. No effect of the use of RSV-IG was observed for this large premature subgroup.

Our recommendations differ from those of a recent cost-effectiveness analysis (CEA) of RSV-IG.¹⁴ A CEA differs from the methods used in this study in that it permits a ranking of cost-effectiveness ratios for a range of alternative therapies, assuming similar methods were used to estimate the ratios. Using a comparison of ratios, providers, third-party payers, and consumers can make relative judgments concerning the therapy in question. A decision rule to guide the acceptance of recommendations is not offered in a CEA.

Despite these differences, conclusions from both methods should be similar. Given that the CEA of the use of RSV-IG found the drug to be cost-effective for all high-risk infants, differences in study findings should be attributed to something other than differences in methods. The most important difference between the studies is that data from different randomized controlled trials of the use of RSV-IG are used. The CEA study used results from the initial non–placebo-controlled trial,³ which resulted in the conclusion that the use of RSV-IG would cost $24000 per life-year saved. This estimate includes a direct and indirect effect of RSV-IG on mortality. The authors assume that RSV-IG directly affects mortality by reducing the severity of illness among hospitalized infants and predict a reduction in mortality from 2% to 1%. If no difference in mortality is assumed (2% risk for treated and untreated groups), mortality is affected indirectly by a reduced need for hospital admission.

Changing assumptions have a large effect on cost per life-year saved. Substituting data from all 3 trials and assuming no effect on mortality from the use of RSV-IG gives a cost-effectiveness ratio more than 21/2 times that of the initial CEA estimates, or $66000 per life-year saved. Adjusting the discount rate from 3% used by the authors to 5% in accordance with the studies the authors compare for RSV-IG use generates a cost-effectiveness ratio of about $134000 per life-year saved. Under reasonable assumptions and parameter values, claims of cost-effectiveness for RSV-IG are more limited than those published.

The analyses described here are limited in 5 respects: willingness-to-pay estimates have not been validated against actual payments; the PREVENT trial was not designed with sufficient power to detect treatment effects in the subgroups of patients we analyzed separately; deaths due to RSV disease, known to occur in as many as 3.5% of high-risk infants,¹⁹ have not been considered; secondary benefits of RSV-IG use have not been fully explored²² ; and estimates have been based on a rate of hospital admissions for RSV infection among the placebo-treated group that may have been lower than expected.

The willingness-to-pay concept is an accepted and widely used component of cost-benefit analyses.^{23 - 25} Nonetheless, methods for assessing the willingness to pay vary across studies, and estimates based on hypothetical situations are rarely validated against what people, parents, or payers actually pay to avoid health risks. These limitations of the willingness to pay are shared with other methods for assessing the economic value of medical care, including a cost-effectiveness analysis that relies on a subjective valuation of quality-adjusted life-years.¹³

As the authors of the PREVENT trial state, it is not designed with the statistical power to detect significant treatment effects among subgroups. Our analyses of subgroup NNTs was not intended by the trial design. Nonetheless, the effect of the use of RSV-IG across subgroups is presented in the article and emphasized in the discussion of the PREVENT trial and in the product information supplied by the manufacturer.⁵ To allow for limited power to detect effects, our assessments of recommendations are based on estimates of the 95% CI around subgroup NNTs.

An analysis of RSV-IG efficacy among subgroups of high-risk infants is important because of the implications to the health care system should most of these infants require treatment. In Arkansas, almost 9% of all babies (>3000 per year) are born at a gestational age of 35 weeks or younger.²⁶ The Food and Drug Administration has approved the use of RSV-IG in all of these infants. Treating each of them would place an unreasonable burden on the state health care system. Restricting the administration of RSV-IG to infants who are at 32 weeks' gestational age or younger would allow more than 700 infants born in the state to be treated. This number would also present a major challenge to local providers.

Mortality due to RSV disease among high-risk infants has not been considered as an end point in this analysis. It is known that as many as 3.5% of hospitalized high-risk infants will die of RSV-related illness.¹⁹ Because mortality was not an end point in the trials considered and there is no evidence that RSV-IG is effective in preventing deaths due to RSV disease, analyses did not address mortality. In the 3 trials considered,^{1 ,3 - 4} the groups treated with RSV-IG had a combined mortality rate of 2.36% (13 deaths) compared with 0.36% (2 deaths) in the control or placebo-treated groups. No explanation can be given for the slightly higher fatality rates among RSV-IG recipients.²¹

In assessing the recommendations for the use of RSV-IG, the observed secondary benefits of treatment have not been considered. Administering RSV-IG was shown to significantly reduce the incidence of otitis media in the PREVENT trial.¹ The length of stay for hospital admissions for RSV infection among those who were admitted to a hospital was also significantly shorter among PREVENT subjects who received RSV-IG. These potential benefits and unanticipated costs may change indications for the use of RSV-IG and should be considered in further analyses.

The minimal benefit of RSV-IG in premature infants without lung disease observed in PREVENT may have been due to unusually low rates of hospital admission for RSV infection in the placebo arm of the trial. Only 8.1% of premature infants without BPD and 8.6% of infants younger than 6 months in the placebo arm of the study were admitted to a hospital for RSV infections. Other data suggest that the true rates of hospital admission may be somewhat higher among these placebo-treated groups. Cunningham et al²⁷ noted that 25% of a sample of infants born at a gestational age of 32 weeks or less without BPD required readmission to a hospital for any respiratory illness. Given data from PREVENT, about half of these hospital admissions would have been for RSV disease. The resulting estimate of 12.5% is somewhat higher than the 8.1% found in PREVENT. If we assume 12.5% to be the true rate of placebo-treated hospital admission for premature infants and the rate for the RSV-IG–treated group to be the observed 6.5%, the ARR becomes 6.0 and the NNT becomes 15.4 (95% CI, 7.2, the lower bound on the ARR is negative, and, therefore, the upper bound on the NNT cannot be calculated). This number is above all but a single T-NNT estimate. Under these revised assumptions, strong inference remains that treatment would not be recommended.

To be safe and to obtain the optimal potential benefit of the use of RSV-IG, many hospitals are electing to follow consensus statements^{2 ,21} in their use of RSV-IG with infants known to be at high risk for severe lower respiratory tract infection. Institutions will likely choose to treat young infants with BPD who have required supplemental oxygen in the past 6 months and infants younger than 24 months born at younger than 32 weeks of gestation. Our analyses would question but not strongly discourage the use of RSV-IG in infants with BPD. The application of cost-benefit principles on currently available data do not support its use in premature infants without respiratory morbidity. No data exist to address the value of RSV-IG with immunocompromised infants or infants with chronic lung disease who are not premature, although many institutions will likely elect to treat these infants.

Given the limited value of RSV-IG, prudent practice will continue to emphasize other means of preventing RSV infection. Preventive measures should include limiting exposure to contagious settings (eg, day care) when feasible, limiting exposure to other sick children, emphasizing good hand washing in the home, and understanding the role that smoking in the household has on these infections. The possible effect of these practices may account for the low rates of hospital admission for RSV among placebo-treated patients in controlled trials where families receive intensive education about RSV transmission.¹

Solutions of the T-NNT formula presented in this article depend greatly on the costs of care and the dollar value placed on health states. Data to support these figures are rarely available. Treatment decisions must be made with or without data. Such decisions will increasingly incorporate the cost and value of care. By stating assumptions about costs and values explicitly, others can vary assumptions to fit their own practice setting. If treatment contexts differ greatly from those we have considered, medical decision makers can generate a new T-NNT more consistent with their charge structure or value of health. Similarly, as new data are generated on the efficacy of RSV-IG in other high-risk groups, updated NNTs may change the conclusions presented here. Further development and application of the explicit indexes of treatment decisions should aid the tasks of both clinicians and health policymakers.

Accepted for publication November 25, 1997.

Presented at the Society for Pediatric Research meetings, Washington, DC, May 6, 1997.

We thank John Sinclair, MD, and Gordon Guyatt, MD, for their valuable suggestions.

Editor's Note: I expect to receive some "interesting" comments from those readers who have contributed to previous studies that reached a different conclusion—using different analyses. That's the cost and benefit of publishing studies like this one.—Catherine D. DeAngelis, MD

Reprints: James M. Robbins, PhD, Center for Applied Research and Evaluation, Department of Pediatrics, University of Arkansas for Medical Sciences, Arkansas Children's Hospital, 800 Marshall St, Little Rock, AR 72202 (e-mail: jrobbins@care.ach.uams.edu).