Introduction

The pulmonary artery catheter (PAC) provides unique hemodynamic information which changes management in nearly 50% of patients [1]. Because of its perceived utility, the PAC is used commonly in intensive care units (ICU), despite efficacy studies that have come to differing conclusions [2, 3]. The debate was kindled again by a report that PAC use was associated with higher cost and poorer outcomes [3].

Conflicting outcomes of PAC studies may in part relate to variability in physician factors. There are reports detailing life-threatening complications of PAC insertion and maintenance [4] as well as poor physician performance in various aspects of PAC management, including PAC waveform interpretation [5, 6]. One issue not evaluated is whether PAC data leads to uniform management. Variable responses to PAC data would confound evaluation of PAC utility and may adversely affect outcomes; thus, we sought to determine if intensivists make uniform management decisions with PAC data.

To answer this question we performed a survey study in which certified medical and surgical intensivists responded to three clinical vignettes. Each vignette presented a common intensive care unit (ICU) patient and included PAC data. Fifty percent of the surveys also included transthoracic echocardiographic (TTE) information for each vignette in order to determine if additional non-invasive assessment of cardiac function affected physician management. The intent of the vignettes was to determine the homogeneity of responses, and not the frequency of a "correct" answer.

Methods

Survey questionnaires were mailed to board-certified intensivists who belonged to the Illinois chapter of the American College of Chest Physicians (ACCP) as well as to the Illinois members of the Society of Critical Care Medicine (SCCM). A second survey was mailed 7 weeks later to those who failed to respond. The questionnaires were designed to assess practice variation using pulmonary artery catheter data.

Patient simulation, whether using standardized patients or computerized patient simulators, has gained acceptance for evaluating various aspects of physician performance, including quality of care [7] and differences in practice patterns [8]. Standardized patients are not an option for evaluating physician performance in critical illness and patient simulators are expensive and require participants to be on site. Clinical vignettes have been used in a wide range of practice settings and have been shown to closely replicate results from standardized patients [9].

Page one of the questionnaire requested demographics of the practitioners and their medical practice. Page 2 consisted of three clinical vignettes. Each vignette was developed by a board-certified pulmonary intensivist and validated by two other intensivists. The patients were in shock and/or respiratory failure, the diagnoses most likely to result in PAC placement in a medical intensive care unit (MICU) [10].

Each vignette provided information regarding history, physical examination, arterial blood gases, ventilator settings, lung mechanics, chest radiograph interpretations, and detailed pulmonary artery catheter data. In addition, 50% of the surveys included trans-thoracic echocardiographic (TTE) information for each vignette. After each vignette, respondents were asked to specify whether a PAC was indicated and then to select a single answer from six choices [five common specific interventions, normal saline (NS) infusion, dobutamine, dopamine, nitroprusside, lasix, and a sixth none of the above (NA)].

We assumed that a homogenous response would result in one choice being selected by at least 80% of the respondents. The sample was assumed to consist of independent identically distributed observations from a uniform distribution with mean equal to p. A single clear choice implies p>0.8 for one of the choices. To avoid assuming one of the six choices is the single clear choice, six tests were run for each scenario, rotating one single clear choice as the outcome. The Bonferroni Joint Estimation method was used to establish an overall significance level to be done simultaneously. Significance was defined as p<0.05.

The Fisher's exact test was used to compare the six choice distributions of respondents with TTE information and those without TTE information and among different demographic groups. Significance was defined as p<0.05.

Results

We received responses from 164 (42%) intensivists from the 393 questionnaires mailed; 35 did not use PACs and 3 failed to select an intervention. Of the remaining 126 surveys, 53 (42%) had vignettes with TTE data and 73 (58%) did not. Seventy (56%) and 56 (44%) respondents belonged to the ACCP and SCCM, respectively.

Demographic and practice characteristics of the respondents revealed a broad spectrum of training, experience, and practice settings. Seventy-eight (66%) were pulmonary and 39 (31%) were surgery or anesthesia intensivists, respectively. Forty-nine (39%) worked at an academically affiliated hospital, 47(38%) at a community hospital, and 29 (23%) practiced at a university hospital. Additionally, 78 (63%) worked in a medical ICU and 110 (87%) had been in practice for more than 5 years. Finally, 74 (59%) intensivists placed more than ten PACs annually. There was no correlation between demographic variables and interventions selected.

More than two-thirds of respondents thought that a PAC was indicated for each vignette (Table 1). There was no significant difference in the proportion who thought that a PAC was indicated between the TTE and the non-TTE groups (data not shown).

Table 1. Demographic data
Full size table

In the first vignette, 82 (65%) respondents selected choice 6, none of the above (NA), and three interventions were each selected by approximately 10% of respondents (Table 2). In vignette 2, 45% of respondents chose dopamine, 40% chose normal saline (NS), and 13% chose NA. In vignette 3, four different interventions were each selected by at least 10% of respondents. There were significant differences in the distribution of selected interventions between the TTE and non-TTE groups for each vignette (p<0.05), but no difference in overall heterogeneity.

Table 2. Responses to clinical vignettes
Full size table

Discussion

In summary, this study demonstrated three points: a majority of intensivists value PAC data; intensivists responses to clinical vignettes with identical PAC data is heterogeneous; and presence of additional TTE data affected treatment choices but did not alter heterogeneity.

Variability in medical decision making is not unique to intensivists and is affected by many factors including patient differences [11] and the disparity in measuring the pulmonary wedge pressure using raw tracings [5]. We excluded both patient differences and waveform tracing interpretation as sources of variability by our survey design.

Treatment variability may reflect respondents coming to different diagnostic conclusions based on given data, or alternatively, they chose different interventions to manage the same diagnosed problem. The data do not allow one to distinguish between these possibilities. If the former is true, there are significant inconsistencies in PAC data interpretation underscoring the need for a comprehensive educational and/or credentialing process to ensure basic competency in using PACs. The latter possibility may also indicate insufficient training, but could also reflect a lack of consensus in managing common ICU problems.

Despite data that echocardiography provides close correlation of hemodynamic parameters with PAC values [12, 13], the addition of echocardiographic information to PAC data leads to management changes 44% of the time [14]. This is consistent with our data that intensivists with both PAC and echo data selected significantly different management options compared with intensivists with PAC data only. Furthermore, combining PAC and echo information did not reduce overall heterogeneity, suggesting that coupling of information from the two modalities for MICU management decisions is suboptimal.

This study has at least three limitations: low response rate; limited information in the vignettes; and the generalizability of the data. Despite the 42% response rate, our subjects probably represent a disproportionate number of PAC users, since perceived salience of subject matter is a strong predictor of survey response [15]. There may be unidentified factor(s) that could diminish both response rates and enhance homogenous selections to the vignettes, which if present would invalidate our findings. We think this is unlikely given the lack of correlation of any demographic variable and response choices. With respect to the vignettes, they may have excluded clinical data important in PAC decision making. While it is impossible to replicate in a vignette all the factors that may impact bedside decision making, the vignettes did contain the variables used most often in studies evaluating the impact of PAC data on patient management [1, 16]. Lastly, these results may not be generalizable to CCU and surgical ICU populations. We suspect that there may be less management heterogeneity as these populations often have less co-morbid illnesses and less organ system dysfunction.

Our data suggest that it may be difficult to interpret the efficacy of a PAC trial unless strict treatment protocols are used. In addition, it articulates the need for education in the interpretation of PAC data and subsequent management.