© 2003 American Thoracic Society
What to Learn from Decision Analysis of Diagnosis and Treatment of Ventilator-associated Pneumonia?University of Ottawa Centre for Transfusion and Critical Care Research Ottawa Hospital Research Institute Clinical Epidemiology Program Ottawa, Ontario, Canada Ventilator-associated pneumonia is a major threat to patients requiring prolonged mechanical ventilation (1–4). Mortality rate is approximately 50% and pneumonia increases the length of intensive care and hospital stay (4–7). Despite the frequency and importance of ventilator-associated pneumonia, there is still controversy about the optimal diagnostic test and optimal therapy. In this issue of the Journal (pp. 1060–1067), Ost and colleagues (8) used a decision analysis to simultaneously evaluate different diagnostic modalities followed by three different antibiotic regimens to determine the most cost-effective strategy for managing ventilator-associated pneumonia. More specifically, the authors used a decision analysis model in a hypothetical cohort of immunocompetent critically ill patients intubated for 7 days to evaluate 16 possible strategies associated with 4 diagnostic paths. The diagnostic options were: no diagnostic testing; endotracheal tube aspirate followed by quantitative cultures; bronchoscopic cultures; or nonbronchoscopic, mini-bronchoalveolar lavage. The treatment path options consisted of using zero to three initial antibiotics. As outcomes, the authors assessed dollar cost, antibiotic utilization, patient survival, cost-effectiveness, and antibiotic utilization per survivor. Probability estimates were obtained from a literature review of 111 articles. Using this technique, the authors found that survival rates were improved and costs were decreased using initial coverage with three antibiotics rather than expectant management or treatment with one or two antibiotics. Although mini-bronchoalveolar lavage did not improve the survival rate, it minimized costs and antibiotic utilization. From the perspective of minimizing cost, minimizing antibiotic usage, and maximizing survival, the best strategy was diagnostic testing using a mini-bronchoalveolar lavage followed by treatment with three antibiotics. Often overlooked as a technique, decision analysis may shed light on many clinical problems including ventilator-associated pneumonia. It is ideally suited for clinical problems when there are multiple diagnostic and therapeutic options, significant amounts of clinical information with substantial levels of remaining uncertainty, disagreement among published studies, a lack of head-to-head comparisons within randomized trials, or an inability to conduct the appropriate clinical trial. The study conducted by Ost and coworkers (8) satisfies many of these pre-conditions. One of the major strengths of decision analysis is that it addresses the many uncertainties related to the development of pneumonia in patients who require mechanical ventilation. In this setting, uncertainty arises from imperfections in the application of diagnostic criteria and testing, uncertainty about the natural history of disease, the choice of therapeutic strategies, the random nature of complications, and the impact of patient-specific factors or co-morbid conditions (9, 10). The ability to perform many sensitivity analyses within a decision model assists in characterizing aspects of uncertainty by evaluating multiple variables simultaneously and an extended range of values for each parameter (11). In this study, the authors used mean estimates of probabilities derived from secondary sources along with estimates calculated from the analysis of extremes using Monte Carlo simulations that sampled from a normal distribution. Decision analysis can also incorporate repetitive "loop" processes, such as Markov chain, for accurately representing various states of health and risks that change constantly over time (11, 12). For instance, the authors of this study could have modeled the ever-changing risks of developing pneumonia and death over the course of a stay in the intensive care unit using Markov techniques (12). The use of Bayesian statistics in decision analysis further strengthens the capacity to deal with uncertainties (13). The Bayesian approach allows investigators to incorporate a range of prior probabilities into the decision model without specifying whether the probability distribution is parametric or non-parametric. Estimates for probabilities may be derived from the results of individual studies, or preferably, the results of systematic reviews. By combining the literature estimates and decision process into one coherent model, the Bayesian approach incorporates uncertainty in incidental model parameters. The authors, however, chose not to make use of Bayesian statistics in their analytic model. A decision analysis will not provide realistic estimates of cost-effectiveness if there is little published literature such that investigators can incorporate only their best guess of important probabilities within the model (11, 12). This may indeed be the case with some of the probabilities incorporated into the model developed by Ost and colleagues. Even though the authors systematically searched the literature to identify estimates of major probabilities, there was little attempt to incorporate the quality of the evidence into the point estimates and sensitivity analyses used in the Monte Carlo simulations. The effectiveness of any empiric treatment strategy for ventilator-associated pneumonia will be related to a large number of variables including the prevalence, virulence, and resistance patterns of bacterial flora in each intensive care unit as well as host factors, such as the cause and severity of the critical illness, immunocompetency, underlying health of the patient, and previous exposure to antibiotics. Antibiotic utilization patterns will significantly influence the development of antibiotic resistance as well as costs. The impact of many such factors may be difficult to incorporate into decision models. Despite these limitations, the results from the decision analysis suggest that an antibiotic regimen using three initial drugs with bronchoscopy or a mini-bronchoalveolar lavage is worth further study. The results suggest that several other strategies are unlikely to be effective and do not warrant further investigation. We agree with Ost and colleagues that randomized controlled trials need to be conducted to evaluate dominant strategies identified in this decision analysis. A study of ventilator-associated pneumonia led by Heyland and colleagues (personal communication) for the Canadian Critical Care Trials Group has essentially incorporated one of the dominant options into its design. In a two-by-two factorial randomized trial of 740 critically ill patients, investigators are comparing the use of bronchoscopy guided diagnostic testing followed by the use of a meropenem alone versus meropenem plus ciprofloxacin tailored to culture results (D.K. Heyland for the Canadian Critical Care Trials Group, personal communication). We encourage greater use of decision analysis in guiding clinical, research, and policy-driven decision making. Acknowledgments The authors thank Dr. Shawn Aaron for reviewing this editorial. FOOTNOTES Conflict of Interest Statement: M.A. has no declared conflict of interest; P.C.H. has no declared conflict of interest. REFERENCES
This article has been cited by other articles:
|
||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
