Outcome Assessment and Pulmonary Artery Catheterization

Over the last two decades, anesthesiologists and other physicians who manage critically ill patients have used pulmonary artery catheterization (PAC) with considerable frequency for the purpose of obtaining information on the adequacy of perfusion and for guiding management directed at improving perfusion. Since it was first introduced into clinical practice in 1970, the use of PAC has remained controversial despite the adoption of practice guidelines by some medical specialty societies, including the American Society of Anesthesiologists [1]. The economics of managed/capitated care heighten the concern for routine application of PAC, and the recent publication of the findings of the Study to Understand Prognoses and Preferences for Outcomes and Risks of Treatment (SUPPORT) [2] has intensified the controversy over this monitoring technique, even generating a passionate call for a moratorium on PAC [3]. However, thorough scrutiny of this study reveals that its design precludes definitive conclusions about the impact of PAC on outcome and that the published conclusions of the accompanying editorial commentary are unjustified [3]. We wish to focus on the science of the SUPPORT protocol and its implications. The SUPPORT study was a prospective, cohort-matched examination of outcome of 5735 critically ill patients (2184 of whom had PAC) with nine disease categories who were admitted to 15 intensive care units in five medical centers [2]. This was an observational study, and case matching was used to identify 1008 pairs of patients who matched for disease category and propensity for PAC. Despite "adjustment" for treatment selection bias using case matching and multivariate analysis, there was no demonstrable benefit of PAC on several outcomes. In fact, the SUPPORT investigators concluded that patients receiving PAC had an increased 30-day mortality rate (37.5% vs 33.8% without PAC), greater mean hospital costs ($49,300 vs $35,700 without PAC), and longer ICU and hospital stays. The finding of a subgroup analysis that the relative risk of death was more than 1.5 times greater in postoperative patients managed with PAC than in all other patients may be important for anesthesiologists, although this observation is not meaningful, since no data comparing risk factors in this subgroup with the remainder of the study population were provided. Close examination of the SUPPORT data does reveal several important but concerning findings. In this study, patients managed with PAC were more likely to enter the study with multiorgan system failure, acute respiratory failure, congestive heart failure, higher Acute Physiology and Chronic Health Evaluation (APACHE III) scores, and lower estimated survival. They also had significantly lower mean arterial blood pressures and baseline serum albumin concentrations than patients not receiving PAC. The decision to use PAC was based upon the judgment of individual clinicians, and the SUPPORT investigators therefore appropriately attempted to control for selection bias, since patients entering the study who received PAC were, on average, more critically ill than patients not receiving PAC. Despite attempts to control for the effects of underlying disease and illness severity, failure of the SUPPORT protocol to define or control the indications for PAC colors the validity and biases the results. Unfortunately, retrospective (or prospective) case matching can only account for the few known, measured variables; examples abound of errors caused by not recognizing that important variables are often not accounted for in such observational studies. Only true randomization, with chance alone dictating who receives the intervention in a large enough sample, can account for unknown variables that have a causal role in affecting outcome. Several other problems cloud the interpretation of the SUPPORT findings. Although subgroup analysis to determine the association of PAC with survival at five study sites found no significant difference in the relative risk of death at the five study institutions, the SUPPORT data do not define whether patient mix, patient management, or total number of patients managed with PAC varied significantly among centers in the subset of 1008 cohort-matched pairs of patients. Failure to specify the treatment algorithm used at each institution also prohibits the formation of definitive conclusions. Nonuniformity of patient management in response to hemodynamic data obtained from PAC both within and between institutions greatly impacts outcome and further distorts the association between PAC and outcome. PAC in itself is not a therapeutic intervention, and a number of steps link the use of any diagnostic technology and ultimate outcome. Failure to account for this complexity can result in profoundly erroneous conclusions. For example, failure to satisfactorily account for the differences in outcome resulting from different management paradigms for ischemic heart disease (and for different severity of coronary artery disease) might result in the dangerous conclusion that a moratorium should be placed on cardiac catheterization and coronary angiography since application of this technology to diagnose anatomical abnormalities and guide surgical or nonsurgical therapy is associated with a poorer outcome when compared with noninvasive testing alone. In the SUPPORT study, similar distortion may have occurred: the average intensity of care was 4-7 Therapeutic Intervention Scoring System points higher in the PAC group, although the SUPPORT protocol was not designed to determine whether this greater level of intervention was prompted by PAC or because patients with PAC were intrinsically sicker, despite accounting for a large number of risk variables with a multivariate analysis [2]. We propose that the SUPPORT data could readily have an alternative, nearly opposite interpretation since it is also plausible that the greater intensity of care given to the PAC group (higher TISS scores) actually prevented a higher mortality rate from being observed in patients with extremely high mortality risk. Debate over the correct interpretation is futile without a randomized, well-controlled clinical trial with specific therapeutic end points. The process by which care is provided after data are obtained from PAC is the substrate upon which any (future) study must be designed. A related problem is the common assumption that normalization of physiologic variables is a desired end point. A recent review of studies examining interventions designed to provide supranormal levels of oxygen delivery found that methodologic limitations of available studies precluded any definite evidence-based clinical recommendations [4]. In fact, it is terribly difficult to perform such studies [4], and the development of prospectively tested therapeutic end points based on PAC-derived data has therefore been slow. Without further studies, it will not be possible to ascertain whether some critically ill patients (including those in the SUPPORT protocol) have been undertreated while others receive excessive intervention. This is not a moot point since it is not the insertion of catheters and measurement of physiologic variables (which in themselves do not alter outcome) but rather the therapeutic interventions directed by these measurements that is the central issue with PAC. In fact, the SUPPORT investigators postulated that PAC prompted a more aggressive level of care, which in itself may have provided no incremental benefit but did increase the risk of morbidity and possibly mortality. The need to analyze the appropriateness of clinical responses to physiologic measurements is an important caveat for the interpretation of such studies [5], which must also control for adjunctive cointerventions (antibiotics, nutrition, ventilation, transfusion, etc.) to ensure similar application across groups. Because of the above limitations, the observations made by the SUPPORT group can not elucidate whether the patients studied were benefited or harmed by PAC. Previous expert consensus [1] outlined the research agenda necessary to answer questions about the efficacy of PAC-unfortunately, the SUPPORT protocol did not fulfill these needs. Despite a careful description of the study limitations by the SUPPORT investigators, the message of the accompanying editorial and the lay press have distorted the findings of the SUPPORT study [6]. The editorial called for the National Heart, Lung and Blood Institute to direct an immediate prospective study of PAC (with a control group not receiving such hemodynamic monitoring) or, failing that, Food and Drug Administration-initiated moratorium on PAC [3]. The conclusion of the editorialists are clearly unwarranted based on the SUPPORT findings or, for that matter, any other current data, and highlight the problem of permitting editorial writers with conflicting interests to opine in a prominent forum [7]. Where do we stand now with regard to knowledge about PAC and patient outcome, and where should we go from here? With careful attention to measurement technique and data interpretation, PAC provides measures of cardiovascular performance not otherwise readily available on a continual basis, and several studies have firmly established that as many as one half of significant hemodynamic abnormalities cannot be adequately assessed based on clinical experience, preoperative catheterization data, physical exam, chest radiography, or other invasive or noninvasive monitors [8,9]. In addition, PAC-derived data often lead to changes in therapy [10]. It is logical that prompt recognition and specific therapy for hemodynamic abnormalities using PAC-derived data might improve outcome, and, in fact, data from a few small randomized trials in which PAC-derived data were applied in specific fashions to manage certain patient populations strongly suggest a beneficial effect [11-13]. Nonetheless, there are still insufficient data from adequately powered, prospective, randomized studies to conclusively answer the question of whether PAC reduces morbidity or mortality. Such a study would require a large population of patients concurrently randomized to either a control or a PAC group over a relatively short time to avoid the confounding effects of changing anesthetic, medical, and surgical management. Further, such studies may require using comparable institutions for randomization since a learning contamination bias clearly exists [14]. Learning contamination bias includes the error introduced when clinicians who have used a technique (e.g., PAC) have learned from it and now potentially better manage patients even when not using that technique. If PAC use were forbidden, such learning (if it exists) would vanish, and patients without PAC would cease to benefit from such learning. To properly study whether PAC is of value, clinicians naive to PAC should care for those patients randomized to the non-PAC arm of the study. This requirement makes the study more difficult to conduct but reduces the bias common to most technology assessment studies, since technology teaches practitioners who use it for a small segment of patients, thereby potentially improving outcome for all patients. The ethical issues of testing the null hypothesis by continuing management without a monitor that clinicians believe should improve outcome once overt hemodynamic instability occurs has probably been the greatest impediment to the initiation and successful conduct of such a randomized trial. A previous attempt to recruit patients into a randomized trial of management with and without PAC illustrates how difficult it likely will be to conduct such studies. In the Ontario Intensive Care Group study [15], which was terminated early due to poor accrual, nearly one half of the enrolled controls had crossed over to PAC because clinicians deemed it unethical to withhold such invasive monitoring in the face of clinical deterioration (88% of these cross-overs eventually died), and more than 50 patients eligible for the study were excluded because of clinician belief that PAC was ethically mandated. Anticipating similar difficulties does not negate the pressing need for such studies and should not serve as discouragement for potential investigators-rather, it should indicate that sufficient resources and commitment must be made available (by human investigation committees and funding sources) if such elaborate studies are to provide meaningful results. The SUPPORT group as well as previous investigators have, not surprisingly, shown that PAC can be associated with greater direct costs of care [2]. Direct costs of care are, however, hardly the issue-they may represent only a small fraction of total cost or total benefit. While PAC definitely has risk (sepsis and thrombosis are the two greatest [1]), it may also prompt changes in care that result in faster discharge from the ICU or greater preservation of cardiac or other organ function, which can result in indirect cost savings. Thus the real question is not the added cost but whether sufficient incremental value is obtained by expending additional costs. Invasive monitoring costs are difficult to assess but must account for the indirect costs of use and, if not used, indirect costs of failure to provide effective cardiopulmonary support. For instance, if PAC allows us to keep alive longer patients who are extremely ill and who will die regardless of therapy, indirect costs will likely be enormous without a demonstrable outcome benefit. Conversely, if invasive hemodynamic monitoring facilitates precisely titrated care in patients who would have otherwise died, the benefit is obvious. Under such circumstances, the direct costs of use or non-use of PAC are trivial compared with the indirect cost. The value of PAC is highly dependent on whether therapy guided by PAC is effective for the disease state as well as the efficacy of PAC data interpretation and application. The SUPPORT protocol, in failing to evaluate such factors, precludes any estimation of cost-effectiveness. Rational use of high-cost, high-tech monitoring such as PAC or transesophageal echocardiography requires application of disciplined logic-identifying those patients and clinical circumstances in which the value will be greatest. Nonselective, routine use of PAC during all surgical procedures of a given type is not prudent application of an expensive resource, and many other factors must be considered. These factors include patient status and surgical skill and expertise as well as ability, interest, and resources available for proper application of PAC-derived data to guide perioperative therapy [16]. Since most practitioners no longer live in an environment in which cost determines market price, but rather market price influences the total costs that can be expended, such decisions not only have great impact on patient care but also have significant financial implications. Selective application of PAC may not only reduce costs but may also help identify populations for future studies evaluating the impact of PAC on outcome. While PAC, when properly used, probably does attenuate adverse outcome in certain clinical circumstances (by facilitating institution of preemptive therapeutic measures tailored to correct specific physiologic abnormalities), this has not been clearly demonstrated because of limitations of design or conduct of existing studies. Most studies to date, and definitely that of the SUPPORT group, are fatally flawed by a number of serious limitations, and these deficiencies make it difficult to use published data as a basis from which to draw meaningful conclusions about the effect of PAC on outcome. We fully support the call for properly designed, adequately powered, randomized trials of the effectiveness of PAC, without breakdown of randomization assignment. Efforts to vigorously pursue such studies are necessary to provide objective scientific evidence about the effectiveness of PAC. While it is not certain that such trials can actually be successfully conducted, rigorous analysis of current data does not justify any immediate change in the use of PAC to monitor critically ill patients, and calls for a moratorium on PAC [3] are unwarranted. Rather, well-controlled randomized clinical trials focusing on relevant outcomes after optimal application of PAC-derived information will better serve our patients. Until then, clinicians can critically evaluate their use of PAC and the information they derive from it using scientifically derived guidelines [1]. Without additional data, continued discussion of the value of PAC will not provide further enlightenment and will only serve to generate more controversy without resolving the fundamental issues.