Ethical Dilemmas in a Randomized Trial of ... - Wiley Online Library

1128

BAYESIAN STATISTICS

Holley et al. • PHYSICIAN BIAS IN BiPAP STUDY

CLINICAL INVESTIGATIONS Ethical Dilemmas in a Randomized Trial of Asthma Treatment: Can Bayesian Statistical Analysis Explain the Results? MARK T. HOLLEY, MD, THOMAS K. MORRISSEY, MD, PHD, DAVID C. SEABERG, MD, BEKELE AFESSA, MD, ROBERT L. WEARS, MD, MS

Abstract. Objectives: The original objective was to determine whether the use of bilevel positive airway pressure (BiPAP) ventilation would reduce the need for endotracheal intubation, the length of hospital stay, and hospital charges in patients with status asthmaticus. The development of physician treatment bias made patient enrollment difficult. The article subsequently describes the use of Bayesian statistics to explain study results when this bias occurs. Methods: This study was a prospective, randomized controlled clinical trial conducted over a 34.5-month period at an urban university hospital with an emergency department census of 94,000 annual visits. Patients remaining in status asthmaticus after initial standard treatment with inhaled beta-agonists and steroids were randomized to receive BiPAP ventilation plus standard treatment versus standard treatment alone (non-BiPAP), with intubation for either group as needed. Patients with concurrent cardiac or other pulmonary diseases were excluded. The primary outcome measures were endotracheal intubation rate and length of hospital stay. Secondary outcome measures included vital signs (respiratory rate, pulse rate, blood pressure), changes in expiratory peak flow, changes in pulse oximetry values, and hospital charges. Data were analyzed using Fisher’s exact test, Mann-Whitney tests, and Bayesian statistics. For patients enrolled in the study more than once, data analysis was performed on the first enrollment only. Results: Nineteen patients were enrolled in the BiPAP group and 16 patients in the nonBiPAP group. Patients were frequently enrolled more than once and the data from the subsequent en-

From the Department of Emergency Medicine, University of Florida Health Science Center (MTH, TKM, DCS, BA, RLW), Jacksonville, FL. Received September 1, 2000; revision received July 24, 2001; accepted July 30, 2001. Presented at the SAEM annual meeting, Chicago, IL, May 1998. Address for correspondence and reprints: Thomas K. Morrissey, MD, PhD, Department of Emergency Medicine, 655 West 8th Street, Jacksonville, FL 32209. Fax: 904-244-4509; e-mail: [email protected] A related commentary appears on page 1179.

rollments were excluded from the analysis. A marked decrease in enrollment, due to physician treatment bias, led to a premature termination of the study. Demographics showed that the groups were similar in age, sex, initial peak flow rate, and arterial blood gas measurements. There was a 7.3% increase (95% CI = ⫺22 to ⫹45) in the intubation rate in the nonBiPAP group (n = 2) compared with that for the BiPAP group (n = 1). No significant difference was seen in length of hospital stay or hospital charges, although there was a favorable trend toward the BiPAP group. Complications encountered in the BiPAP group included one patient with discomfort associated with the nasal BiPAP mask. Bayesian analysis demonstrated that in order for the collected data to be convincing at the 95% confidence level, the prior conviction among treating physicians that BiPAP was a successful treatment modality would have had to be 98.9%. Conclusions: In this study, BiPAP appeared to have no deleterious effects in patients with status asthmaticus, with a trend toward decreased endotracheal intubation rate, decreased length of hospital stay, and decreased hospital charges. Although further study with more patients is needed to determine the clinical and statistical significance of this intervention, ethical concerns regarding withholding BiPAP treatment from the patients in the control group forced a premature termination of the study in the authors’ institution. Key words: asthma; status asthmaticus; BiPAP; positive pressure ventilation; Bayesian statistics. ACADEMIC EMERGENCY MEDICINE 2001; 8:1128–1135

E

NDOTRACHEAL intubation and mechanical ventilation have been the treatment of choice for acute respiratory failure in patients with status asthmaticus. However, endotracheal intubation and mechanical ventilation have significant associated morbidity, including barotrauma, pneumonia, and vocal cord dysfunction.1–3 Nasal bilevel positive airway pressure (BiPAP) offers an alternative noninvasive approach to mechanical venti-

ACADEMIC EMERGENCY MEDICINE • December 2001, Volume 8, Number 12

lation that can avoid these complications.4–7 Recent studies incorporating noninvasive mechanical ventilation have proven this intervention to be beneficial in the treatment of respiratory failure of patients with chronic obstructive pulmonary disease (COPD).8–14 It is unclear, however, whether these results can be extrapolated to the treatment of status asthmaticus. We conducted a prospective, randomized controlled clinical trial to compare the efficacies of BiPAP ventilation plus standard medical treatment versus standard medical treatment alone, for patients admitted with status asthmaticus, in reducing endotracheal intubation rate, length of hospital stay, and hospital charges. Our study was prematurely terminated and enrollment was well short of our goal. We discuss the factors that led to premature termination and use Bayesian statistical analysis to evaluate the initial data.

METHODS Study Design. This study was originally designed as a prospective, randomized controlled clinical trial comparing treatment strategies for status asthmaticus. During the course of the study, clinician bias developed for one of the treatment arms. We subsequently used Bayesian statistics to evaluate the study question. The study protocol was approved by the institutional review board, and the patients or their relatives gave informed consent. Study Setting and Population. Over a 34.5month period, adult patients presenting to the emergency department (ED) in status asthmaticus and requiring admission were evaluated prospectively. The study was undertaken at University of Florida Health Sciences Center–Jacksonville, an urban university hospital with an ED census of 92,000 annual visits. Enrollment was conducted 24 hours per day and left to the discretion of the treating physician. Criteria for including patients in this study were the following: 1) presentation to the ED with expiratory wheezing and/or shortness of breath; 2) history of asthma; 3) failure to respond adequately to aerosolized albuterol manifested by one or more of the following: respiratory rate higher than 30 breaths/min, peak flow of less than 70% of predicted peak flow, or continued retractions (supraclavicular, intercostal, subcostal) with continued use of accessory muscles of respiration; 4) age 18 to 55 years; 5) ability of patient or family member to give consent. Criteria for excluding patients from this study included any of the following: 1) history by patient

1129

or medical record of other cardiopulmonary disease (COPD, congestive heart failure, idiopathic pulmonary fibrosis, pneumonia, pneumothorax, sarcoidosis, etc.); 2) need for emergent endotracheal intubation to protect the patient’s airway or manage secretions; 3) life-threatening hypoxemia (arterial oxygen saturation below 85% or PaO2 below 55 mm Hg, on 100% FIO2 via non-rebreathing face mask); 4) systolic blood pressure less than 90 mm Hg or the use of vasopressors; 5) electrocardiographic (ECG) instability with evidence of ischemia or significant arrhythmias. Study Protocol. Patients who met selection criteria were computer-randomized to receive either BiPAP ventilation plus standard medical treatment or standard medical treatment alone. Sealed unmarked envelopes contained the group assignments. Standard Medical Treatment. Patients assigned to the standard medical treatment group received oxygen via nasal cannula to keep oxygen saturation above 92%. Medications included aerosolized albuterol 5 mg every 0.5 to 2 hours as needed and intravenous methylprednisolone 125 mg every 6 hours. BiPAP Ventilation. Patients assigned to the BiPAP group received the same medications as the standard treatment group with the addition of periodic BiPAP. Bilevel positive airway pressure was initially set at inspiratory positive airway pressure of 10 cm H2O and expiratory positive airway pressure of 5 cm H2O and adjusted to patient comfort. Patients underwent BiPAP each day of hospital stay. The length of time on BiPAP each day depended on patient tolerance (usually four to 18 hours per day). The patients were allowed to breathe spontaneously on oxygen without assistance for at least two hours each day. The overall total duration of BiPAP was highly variable and determined by the patient’s preference and clinical improvement. Criteria for treatment success or treatment failure and criteria for endotracheal intubation are outlined below. Outcome Measures. The primary outcome measures were: the need for endotracheal intubation during hospital stay, and the length of hospital stay. The secondary outcome measures included vital signs (respiratory rate, pulse rate, blood pressure), change in expiratory peak flow, change in pulse oximetry values, and hospital charges. Criteria for treatment success and endpoint of BiPAP ventilation plus standard medical treatment and standard medical treatment alone included at least three of the following: 1) arterial oxygen saturation ⱖ95% on room air; 2) respira-

1130

BAYESIAN STATISTICS

tory rate ⱕ25 breaths/min; 3) increase in peak flow to >70% of predicted peak flow; 4) no use of accessory muscles of respiration or retractions (supraclavicular, intercostal, subcostal). Criteria for treatment failure and endpoint of BiPAP ventilation plus standard medical treatment or standard medical treatment alone included any of the following: 1) inability to improve hypoxia, dyspnea, or lethargy; 2) development of conditions necessitating endotracheal intubation (see below); 3) hemodynamic or ECG instability; 4) inability to tolerate the nasal mask because of patient discomfort; 5) patient request. Criteria for Endotracheal Intubation. Criteria for endotracheal intubation included: cardiopulmonary arrest, coma, altered mental status, respiratory muscle fatigue, hemodynamic instability with systolic arterial pressure below 70 mm Hg, respiratory rate above 35 breaths/min despite two hours of therapy, an arterial oxygen below 45 mm Hg despite supplemental oxygen therapy, and arterial blood pH below 7.2 despite therapy. Data Collection. The respiratory rate, pulse rate, blood pressure, pulse oximetry value, arterial blood gas analysis, and peak expiratory flow rate were recorded at onset of therapy. In addition, the respiratory rate, pulse rate, blood pressure, pulse oximetry value, and peak flow were recorded at one, three, and 12 hours after the onset of therapy, and on each subsequent hospital day. Clinical and billing records were examined to determine intubation rates, length of stay, and total hospital charges. Data Analysis. Sample Size Calculations. Preliminary review of prior medical records showed that 518 patients had been admitted in the previous year with asthma exacerbations. Seventy-nine were admitted directly to the intensive care unit (ICU) and 37 were intubated, for a baseline endotracheal intubation rate of 7.5%. We thought that a decrease in endotracheal intubation rate from of 7.5% to 2.5% would be clinically important. For a power of 80% to detect an absolute difference of 5.0% in endotracheal intubation rate between the treatment and control groups (7.5% to 2.5%), we would need 336 patients in each group, assuming an alpha error of 0.05. Outcome Measures. The primary outcome measures (endotracheal intubation rate and length of hospital stay) were analyzed by Fisher’s exact test and Mann-Whitney tests. The secondary outcome measures (vital signs, expiratory peak flow rate, pulse oximetry values, and hospital charges) were


analyzed by Mann-Whitney tests. The effect of BiPAP was summarized by mean or median difference between groups and the 95% confidence interval (95% CI) of the difference. The statistical analysis was performed using SYSTAT version 7.01 (SPSS, Inc., Evanston, IL). The exact CI for the proportions was determined by STATXACT version 3.02 (Cytel Software, Cambridge, MA). For patients who were enrolled in the study more than one time, data analysis was performed on the first enrollment only. Because the design sample size was not achieved, Bayesian statistical methods were used to analyze the data gathered from patients prior to the termination of the study. Bayesian analysis of clinical trials data has been frequently advocated15–18 but seldom used, largely due to the difficulty of the computations and the lack of readily available software. Bayesian analysis is by its very nature sequential; in principle, a Bayesian analysis could be conducted as the outcome from each patient becomes known, regardless of the originally planned sample size.19,20 Thus, Bayesian analysis is ideally suited to the situation of a trial failing to attain its design sample size. In principle, Bayesian analysis is simple, and has been well described previously.15 It first requires the specification of prior knowledge in the form of a prior probability distribution. This distribution reflects the degree of certainty clinicians have about the probable outcomes of the trial. This prior distribution is then updated based on the accumulating evidence from the trial, to form a posterior distribution, reflecting the degree of certainty clinicians have about probable outcomes, given both their prior beliefs and the evidence. For a two-group clinical trial with a dichotomous outcome, these distributions will typically be based on the difference in the proportions of success in the groups. The mean of the posterior distribution is a point estimate of effect magnitude, and the interval containing some useful fraction of the distribution (say, 95%) is an interval estimate of effect magnitude. If the evidence is strong, the effect of prior beliefs will not be dramatic, and many different prior beliefs can lead to approximately the same posterior distribution. Conversely if the evidence is weak, prior beliefs will substantially affect the resulting posterior distribution. Although Bayesian methods are permissible in a situation where a trial has not obtained its design samples size (and frequentist, classic methods are not), they cannot overcome the the loss of power inherent in a decreased sample size, so it would be unlikely that the use of Bayesian methods in this study would result in conclusive evidence for or against BiPAP. But since clinicians declined to enroll patients as evidence accumulated,

1131


we can infer that the clinicians’ posterior distributions must have become convincing. Since the evidence is likely to be weak, this implies that clinicians’ prior beliefs in the efficacy of BiPAP might not have been neutral. To analyze this, we performed three separate Bayesian analyses, for optimistic (favoring BiPAP), neutral, and pessimistic prior beliefs. By adjusting the strength of the prior belief in an iterative fashion, we attempted to determine how strong prior beliefs must be in order to support a behavior change given the available evidence. The Bayesian models were analyzed using winBUGS 1.2 (www.mrc-bsu.cam.ac.uk/bugs/win bugs/contents.shtml) software. Models were produced to predict the effect the available data would have on swaying the opinion of the effectiveness of BiPAP therapy, given predetermined physician conceptions of the likelihood of BiPAP’s being an effective, or useful, therapy.

TABLE 1. Patient Characteristics (Mean ⫾ Standard Deviation) Are Similar between Groups upon Study Entrance BiPAP* Group Age (years)

34.4 ⫾ 9.8

39.7 ⫾ 9.2

Respiratory rate (breaths/min)

28.3 ⫾ 4.8

26.6 ⫾ 8.4

Systolic blood pressure (mm Hg)

142.9 ⫾ 20.3

141.3 ⫾ 29.6

Diastolic blood pressure (mm Hg)

84.9 ⫾ 17.1

89.3 ⫾ 17.8

114.2 ⫾ 18.2

121.2 ⫾ 18.0

86.6 ⫾ 28.9

88.9 ⫾ 23.6

184.2 ⫾ 90.0

210.0 ⫾ 129.9

Arterial pH

7.35 ⫾ 0.04

7.32 ⫾ 0.06

PaCO2 (torr)

39.8 ⫾ 10.6

44.1 ⫾ 11.9

121.2 ⫾ 62.6

101.8 ⫾ 44.8

95.4 ⫾ 4.7

93.8 ⫾ 7.2

Sex Male Female

7 12

3 13

Race White African American

10 9

8 8

Number of patients

19

16

Pulse rate (beats/min) Pulse oximetry value (%) Peak flow rate (L/min)

PaO2 (torr)

RESULTS

Arterial oxygen saturation (%)

Nineteen patients were enrolled in the BiPAP plus standard medical treatment group and 16 patients were enrolled in the standard medical treatment alone group. All patients were enrolled while in the ED. The two groups were similar in age, sex, initial expiratory peak flow value, and arterial blood gas measurements (Table 1). In the BiPAP group one patient required endotracheal intubation. Two patients in the standard medical treatment group required endotracheal intubation. This represented a 7.3% increase (95% CI = ⫺22 to ⫹45) in the endotracheal intubation rate. In the BiPAP group the median length of hospital stay was 46 hours compared with 74 hours in the standard medical treatment group. This represented a 28-hour increase in median hospital stay (p = 0.29 by Mann-Whitney U analysis). The median hospital charge in the BiPAP group was $6,041, compared with $7,572 for the standard medical group. This represents a $1,531 increase (p = 0.16 by Mann-Whitney U analysis). Analysis of the other secondary outcome measures (vital

No-BiPAP Group

*BiPAP = bilevel positive airway pressure.

signs, peak expiratory flow rates, pulse oximetry values) showed no significant difference between the two groups (Table 2). Complications encountered in the BiPAP group were minimal, including one patient’s discomfort associated with the nasal BiPAP mask. Six patients were enrolled more than once and the data from the subsequent enrollments were excluded from data analyses. The rate of enrollment began to decrease during the study. In the first

TABLE 2. Secondary Outcome Measures (Mean ⫾ Standard Deviation) Are Similar between Groups at Three and 12 Hours after Study Entrance BiPAP* Group 3 Hours Respiratory rate (breaths/min) Systolic blood pressure (mm Hg) Diastolic blood pressure (mm Hg) Pulse rate (beats/min) Pulse oxymetry (%) Peak flow rate (L/min)

24.7 133.7 75.7 109.1 98.1 241.0

*BiPAP = bilevel positive airway pressure.

⫾ ⫾ ⫾ ⫾ ⫾ ⫾

6.8 17.8 13.2 11.7 2.2 93.5

No-BiPAP Group 3 Hours 20.3 140.6 81.1 135.3 96.4 184.3

⫾ ⫾ ⫾ ⫾ ⫾ ⫾

5.8 30.8 25.5 74.4 4.6 117.9

BiPAP Group 12 Hours 21.7 129.5 74.5 102.6 97.0 237.9

⫾ ⫾ ⫾ ⫾ ⫾ ⫾

5.0 10.6 16.2 13.8 2.0 76.9

No-BiPAP Group 12 Hours 23.6 ⫾ 132.1 ⫾ 81.4 ⫾ 111.8 ⫾ 95.5 ⫾ 158.3 ⫾

9.0 21.6 22.2 26.0 5.6 60.5

1132

BAYESIAN STATISTICS


TABLE 3. Effects of Bayesian Analysis on Physicians’ Prior Beliefs of Bilevel Positive Airway Pressure (BiPAP) as a Successful Treatment Modality Prior Belief that BiPAP Is an Improvement over Standard Treatment* Naı¨ve

Difference in Intubation Rates (Control–BiPAP)

Posterior 95% High Probability Interval

Probability that BiPAP Is an Improvement over Standard Treatment

7%

⫺13% to ⫹29%

76%

Neutral

4%

⫺9% to ⫹13%

77%

Skeptical

2%

⫺6% to ⫹14%

66%

Favorable

5%

⫺5% to ⫹15%

86%

*Naı¨ve prior beliefs represent no preconceived notion of effect size, i.e., that all possible values for the difference (from ⫺100% to ⫹100%) are equally likely. Neutral prior beliefs represent the belief that the intervention will have no effect, i.e., that delta is close to zero. Skeptical and favorable prior beliefs represent beliefs that the intervention will have a positive or negative effect on intubation rates, i.e., that the difference is ⫺5% and ⫹5%, respectively.

year of the trial, 25 patients were enrolled (2.3 patients per month from February 14, 1997, to February 14, 1998). For the duration of the study, the enrollment rate was 0.4 patients per month (10 patients in 22.5 months). The reason for the decreased patient enrollment appeared to be twofold. There was a decrease in the presentation rate of novel (not previously enrolled) patients. There was also an unwillingness of enrolling physicians to withhold BiPAP treatment from the patients enrolled in the control arm of the study because of the belief that it was unethical to deny access to a treatment modality they believed might/would be effective and could prevent intubation (and any associated complications). The study was prematurely terminated secondary to this physician treatment bias. Bayesian statistical analysis was performed in an attempt to determine whether the physicians’ belief that BiPAP was an effective alternative treatment was purely illusory, or could be supported by the data thus far gathered in the trial. Bayesian analysis requires specification of prior beliefs of probability of success. Since we could not elicit prior beliefs from clinicians (because the study had already taken place), we explored their plausible prior beliefs by selecting neutral, optimistic, and pessimistic probabilities of success to see whether they would be affected by the results. For example, clinicians with an optimistic prior belief (favoring BiPAP) would find modest evidence in favor of BiPAP to be ‘‘more convincing’’ in the sense that their posterior distribution would show a higher estimate of effect magnitude. If this estimate were sufficiently large, they would feel uncomfortable in enrolling a new subject in the study, given that there would be a 50% chance of the subject’s being allocated to the control arm. Conversely, clinicians with a pessimistic prior view of BiPAP would not have such a high posterior estimate of effect, and if they remained sufficiently in equipoise,21 they would continue enrolling new subjects.

Table 3 demonstrates examples of these prior beliefs, and shows the effect the data had on the physicians’ probabilities of predicting a successful outcome. Extrapolation shows that in order for the collected data to be convincing at the 95% confidence level, the prior conviction of BiPAP as a successful treatment modality would have had to be 98.9%.

DISCUSSION Most studies of noninvasive mechanical ventilation have been performed on patients with COPD.8–14 Noninvasive mechanical ventilation has been shown to reduce endotracheal intubation rate, morbidity, mortality, and length of hospital stay in COPD exacerbations.14 Although few studies of noninvasive mechanical ventilation have been performed on patients with status asthmaticus, it appears highly effective in correcting gas exchange abnormalities in these patients.22–24 Further, BiPAP-delivered beta-agonists have been shown to increase peak expiratory flow rate compared with standard aerosolized beta-agonists in status asthmaticus.25 However, the potential benefits of BiPAP in status asthmaticus in reducing endotracheal intubation rate, length of hospital stay, and hospital charges have not been studied.26 We were unable to determine whether nasal BiPAP ventilation has any beneficial or deleterious effects in patients with status asthmaticus as assessed by intubation rate, length of hospital stay, or hospital charges. Further study with more patients will be needed to determine the clinical and statistical significance of this intervention. We encountered unexpected obstacles in enrolling enough patients in the study to provide the power for conventional statistical analyses. Poor enrollment appeared to be due to patients’ declining to participate and physicians’ declining to enroll subjects. Initially we intended to enroll patients only one time into the study, with randomization to either the experimental or the control


group. During the second winter season of the study period, it was noted that many of the patients presenting to the ED with an asthma exacerbation severe enough to warrant consideration for the study had been enrolled (or refused enrollment) during the previous year. Furthermore, during the initial history and evaluation, many patients volunteered that they had used BiPAP with success during previous admissions. Some patients declined enrollment for fear of a potential treatment modality’s being withheld. In the minds of many physicians it was unacceptable to not offer BiPAP therapy to patients in severe respiratory distress. Many physicians admitted to consenting and enrolling patients in the study, but would not withhold BiPAP treatment if the patient happened to fall in the control group. At this time it became clear that the randomization process was not valid and that goal enrollment would not be reached, and the study was discontinued. This phenomenon is not new to this trial.27 Because of decreased patient enrollment and physician bias, we were unable to determine whether there was a difference in outcomes between the groups with any reasonable power. However, this does not mean that these data do not contribute to our knowledge base and cannot be used to help physicians make decisions about appropriate treatment for patients in status asthmaticus. Frequentist statistics are conventionally but erroneously used to determine the strength with which a data set supports a hypothesis about a population.16 This is performed on the assumption that we have no insight or preconceived notion of whether the hypothesis is correct. However, the magnitude or strength of data needed to sway an opinion does not usually stand alone as an isolated number, but, rather, depends on clinician biases, prior experiences, and given clinical situations. Bayesian statistical methods were used to incorporate this prior belief. For example, if one already had a strong inclination to believe BiPAP beneficial, would these data support that conclusion? This is important to know because we are concerned whether the physician bias, which led to the premature termination of our study, was entirely unfounded or had a rational basis. For the data we have collected to be convincing that BiPAP is effective, the physicians’ preconceived notion of its effectiveness would have had to be nearly 100%. With this strong of a preconceived notion, it would not make sense to be asking the question in a scientific setting (i.e., not sufficient clinical equipoise to support a ‘‘need for the study’’). The fact that physicians changed their behavior before the evidence was convincing supports the idea that ‘‘equipoise will typically be disturbed long before we obtain the predetermined level of statistical signif-

1133

icance required to support . . . policy decision[s].’’ 28 Since prior beliefs were certainly not that strong, the question arises as to why the physicians became convinced that BiPAP was a useful (or justified) treatment modality. We believe several factors contributed to this. First, this study was conducted in the ED, with emergency physicians as the enrolling doctors. Emergency department settings foster a bias toward action, rather than expectant observation. We may be more willing to ‘‘try an intervention in the meantime’’ than physicians in a more controlled area of the hospital, such as the ICU. Second, many patients requested BiPAP after either having used it in the past or having seen it used on other patients. Patient satisfaction is a strong driving force in the selection of treatment and satisfaction is largely subjective, based on whether patients think they are being actively and appropriately treated, rather than on objective measures such as peak flow or respiratory rate. The perception of ‘‘hightech treatment administration’’ may serve to alleviate some of the apprehension associated with respiratory distress. Similar results have been observed with the administration of low-dose benzodiazepines. The decrease in apprehension may allow more time for the conventional treatments (nebulizers and steroids) to have an effect. Third, in this study scenario we are not blinded to the treatment arms and we are acutely aware of the consequences of our decisions. Anecdotal data can be powerfully persuasive under the proper circumstances, and seeing one or two patients in extremis experience clinical improvement on BiPAP treatment can make it difficult to withhold treatment for the next patient. Fourth, this clinical trial represented the clinician’s decision as a single irrevocable event, but the clinical reality is that a decision (to forgo intubation or to use BiPAP) at one point in time can and frequently is revisited or even reversed in the light of subsequent information about the patient’s course. While studies are designed to provide numerical predictions of success or failure of a certain intervention in a controlled environment, emergent clinical decisions are often much more complicated. Most clinical decisions are not made on a ‘‘95% confidence interval’’ but rather on a spontaneous and informal risk– benefit analysis performed at the bedside. Factors taken into account in this analysis include clinical status of the patient, availability of treatment modalities, clinician familiarity with treatment modalities, patient desires, potential risks of treatment, and (importantly) the surrounding environment. The use of BiPAP for a patient in severe respiratory distress may be entirely appropriate when the ED census is low and the patient can be

1134

BAYESIAN STATISTICS

closely monitored and minute-by-minute adjustments can be made. When the department is overwhelmed and close monitoring may be compromised, it may be extremely dangerous to try a new and somewhat unfamiliar treatment on the same patient. In this case definitive airway management with intubation may be a much more sound medical decision.

LIMITATIONS AND FUTURE QUESTIONS The design of this study placed patients into either a control or an experimental arm, with the assumption that the clinician would merely watch and record the outcome. Study designs such as this, while often necessary to control confounding variables and facilitate data analysis, do not accurately reflect the clinical decision-making process in the real world. The time course of treatment modalities enforced by the study protocol did not necessarily mimic the time course that would be used clinically. Ideally the clinician would be able to place and remove BiPAP periodically based on both patient and physician desires. This limitation on their discretionary judgment likely added to physicians’ reluctance to enroll patients in the study. Perhaps future studies should be designed so as to more closely mimic the normal clinical situation, with patient satisfaction as a primary measurement endpoint. Other current limitations to this study included difficulty in collecting data on all secondary outcome measure data points (peak expiratory flow rates, vital signs, arterial blood gases, pulse oximetry values), which hindered the ability to do analysis of variance (ANOVA) on these data. Potential directions for future study include continued analysis of larger patient sample size (perhaps in a multicenter trial), application to the pediatric asthma population, and early administration of BiPAP-ventilation-delivered beta-agonists to determine decrease in length of ED stay in less severe asthmatic exacerbations.

CONCLUSIONS In our study, BiPAP appeared to have no deleterious effects in patients with status asthmaticus, with a trend toward decreased endotracheal intubation rate, decreased length of hospital stay, and decreased hospital charges. Although further study with more patients is needed to determine the clinical and statistical significance of this intervention, ethical concerns regarding withholding BiPAP treatment from the patients in the control group forced a premature termination of the study in our institution.


References 1. Pingleton SK. Complications of acute respiratory failure. Am Rev Respir Dis. 1988; 137:1463–93. 2. Stauffer JC, Olson DE, Petty TL. Complications and consequences of endotracheal intubation and tracheotomy: a prospective study of 150 critically ill adult patients. Am J Med. 1981; 70:65–76. 3. Fagon JY, Chastre J, Hance AJ, et al. Nosocomial pneumonia in ventilated patients: a cohort study evaluation attributable mortality hospital stay. Am J Med. 1993; 94:281–8. 4. Meduri GU, Conoscenti CC, Menashe P, et al. Noninvasive face mask ventilation in patients with acute respiratory failure. Chest. 1989; 96:865–70. 5. Benhamou D, Girault C, Faure C, Portier F, Muir JF. Nasal mask ventilation in acute respiratory failure: experience in elderly patients. Chest. 1992; 102:912–7. 6. Meduri GU, Abou-Shala N, Fox RC, Jones CB, Leeper KV, Wunderink RG. Noninvasive face mask mechanical ventilation in patients with acute hypercapnic respiratory failure. Chest. 1991; 100:445–54. 7. Wysocki M, Tric L, Wolff MA, et al. Noninvasive pressure support ventilation in patients with acute respiratory failure. Chest. 1993; 103:907–13. 8. Leger P, Jennequin J, Gaussorgues P, Robert D. Acute respiratory failure in COPD patients treated with noninvasive intermittent mechanical ventilation (control mode) with nasal mask [abstract]. Am Rev Respir Dis. 1988; 137(suppl):63. 9. Brochard L, Isabey D, Piquet J, et al. Reversal of acute exacerbations of chronic obstructive lung disease by inspiratory assistance with a face mask. N Engl J Med. 1990; 323:1523– 30. 10. Foglio C, Vittaca M, Quadri A, et al. Acute exacerbations in severe COLD patients: treatment using positive pressure ventilation by face mask. Chest. 1992; 101:1533–8. 11. Vitacca M, Rubini F, Foglio K, et al. Noninvasive modalities of positive pressure ventilation improve the outcome of acute exacerbations in COLD patients. Intens Care Med. 1993; 19:450–5. 12. Fernandez R, Blanch LP, Valles J, et al. Pressure support ventilation via face mask in acute respiratory failure in hypercapnic COPD patients. Intens Care Med. 1993; 19:456–61. 13. Bott J, Carroll MP, Conway JH, et al. Randomized controlled trial of nasal ventilation in acute ventilatory failure due to chronic obstructive airways disease. Lancet. 1993; 341: 1555–7. 14. Brochard L, Mancebo J, Wysocki M, et al. Noninvasive ventilation for acute exacerbations of chronic obstructive pulmonary disease. N Engl J Med. 1995; 333:817–22. 15. Lewis RJ, Wears RL. An introduction to the Bayesian analysis of clinical trials. Ann Emerg Med. 1993; 22:1328–36. 16. Goodman SN. Toward evidence-based medical statistics. 1: The p value fallacy. Ann Intern Med. 1999; 130:995–1004. 17. Goodman SN. Toward evidence-based medical statistics. 2: The Bayes factor. Ann Intern Med. 1999; 130:1005–13. 18. Goodman SN, Berlin JA. The use of predicted confidence intervals when planning experiments and the misuse of power when interpreting results. Ann Intern Med. 1994; 121:200–6. 19. Lewis RJ. Bayesian hypothesis testing: interim analysis of a clinical trial evaluating phenytoin for the prophylaxis of early post-traumatic seizures in children. In: Berry DA (ed). Bayesian Biostatistics. New York, NY: Marcel Dekker, 1996, pp 279–96. 20. Lewis RJ, Berry DA. Group sequential clinical trials: a classical evaluation of Bayesian decision-theoretic designs. J Am Stat Assoc. 1994; 89:1528–34. 21. Freedman B. Equipoise and the ethics of clinical research. N Engl J Med. 1987; 317:141–5. 22. Meduri GU. Noninvasive positive pressure ventilation in status asthmaticus. Chest. 1996; 110:767–74. 23. Meyer TJ, Hill NS. Noninvasive positive pressure ventilation to treat respiratory failure. Ann Intern Med. 1994; 120: 760–70. 24. Shivaram U, Miro AM, Cash ME, et al. Cardiopulmonary responses to continuous positive airway pressure in acute asthma. J Crit Care. 1993; 8:87–92.


25. Pollak CV, Fleisch KB, Dowsey K. Treatment of acute bronchospasm with beta-adrenergic agonist aerosols delivered by nasal bilevel positive airway circuit. Ann Emerg Med. 1996; 26:552–7. 26. Martin JG, Shore S, Engl LA. Effect of continuous positive airway pressure on respiratory mechanics and pattern of breathing in induced asthma. Am Rev Respir Dis. 1982; 126:

1135

812–7. 27. Sackett DL. Why randomized controlled trials fail but needn’t: 1. Failure to gain ‘‘coal-face’’ commitment and to use the uncertainty principle. Can Med Assoc J. 2000; 162: 1311–4. 28. Gifford F, Lilford RJ, Djulbegovic B, Sackett DL. Uncertainty about clinical equipoise [letter]. BMJ. 2001; 322:795.

䢇