Early Use of the Pulmonary Artery Catheter and Outcomes in Patients With Shock and Acute Respiratory Distress Syndrome: A Randomized Controlled Trial

Context Many physicians believe that the pulmonary artery catheter (PAC) is useful for the diagnosis and treatment of cardiopulmonary disturbances; however, observational studies suggest that its use may be harmful.

Objective To determine the effects on outcome of the early use of a PAC in patients with shock mainly of septic origin, acute respiratory distress syndrome (ARDS), or both.

Design, Setting, and Patients A multicenter randomized controlled study of 676 patients aged 18 years or older who fulfilled the standard criteria for shock, ARDS, or both conducted in 36 intensive care units in France from January 30, 1999, to June 29, 2001.

Intervention Patients were randomly assigned to either receive a PAC (n = 335) or not (n = 341). The treatment was left to the discretion of each individual physician.

Main Outcome Measures The primary end point was mortality at 28 days. The principal secondary end points were day 14 and 90 mortality; day 14 organ system, renal support, and vasoactive agents–free days; hospital, intensive care unit, and mechanical ventilation–free days at day 28.

Results The 2 groups were similar at baseline. There were no significant differences in mortality with or without the PAC at day 14: 49.9% vs 51.3% (mortality relative risk [RR], 0.97; 95% confidence interval [CI], 0.84-1.13; P = .70); day 28: 59.4% vs 61.0% (RR, 0.97; 95% CI, 0.86-1.10; P = .67); or day 90: 70.7% vs 72.0% (RR, 0.98; 95% CI, 0.89-1.08; P = .71). At day 14, the mean (SD) number of days free of organ system failures with or without the PAC (2.3 [3.6] vs 2.4 [3.5]), renal support (7.4 [6.0] vs 7.5 [5.9]), and vasoactive agents (3.8 [4.8] vs 3.9 [4.9]) did not differ. At day 28, mean (SD) days in hospital with or without the PAC (0.9 [3.6] vs 0.9 [3.3]), in the intensive care unit (3.4 [6.8] vs 3.3 [6.9]), or mechanical ventilation use (5.2 [8.5] vs 5.0 [8.5]) did not differ.

Conclusion Clinical management involving the early use of a PAC in patients with shock, ARDS, or both did not significantly affect mortality and morbidity.

Many physicians believe that the pulmonary artery catheter (PAC) is useful for the diagnosis and treatment of cardiopulmonary disturbances and for assessing volume status in critically ill patients.^1,2 However, investigators have raised doubts about the safety of the PAC^3,4 because its use may be associated with direct complications⁵ or harmful effects related to inappropriate decisions resulting from misinterpretation of the data.⁶ The most serious concern was raised by the retrospective study by Connors et al,⁴ which suggested that the use of PAC within 24 hours following admission may increase mortality, length of stay, and health care costs. Either a moratorium on its use or the performance of randomized clinical trials have been proposed,^3,7 particularly in patients with septic shock or acute respiratory distress syndrome (ARDS).⁸ One of the difficulties in performing such a trial is to decide whether a PAC-guided therapeutic protocol should be used. Most physicians use it as a diagnostic tool but not all physicians have the same therapeutic approach. However, the impact of a goal-oriented therapy using PAC has been evaluated in high-risk surgical patients and yielded inconsistent results, ranging from decreased mortality⁹ to increased morbidity and mortality.¹⁰

To avoid the methodological problems of these studies, the Canadian Critical Care Clinical Trials Group¹¹ had recently conducted a randomized trial comparing goal-oriented therapy guided by a PAC with standard care without the use of a PAC in elderly high-risk surgical patients requiring intensive care. This study did not find any effect of goal-oriented therapy by PAC over standard care.¹¹ Moreover, the authors reemphasized the findings of others^9,10,12 when they suggested that it is difficult to achieve the physiological objectives of the goal-oriented therapy. Applied to critically ill patients, this practice of a goal-oriented therapy has not shown any consistent benefit.¹² In addition, no consensus yet exists for such a goal-oriented therapy in septic shock or in ARDS. For these reasons and because the harmful effects of the PAC were found in observational studies without formal protocols, we chose not to use any goal-oriented protocols in a study designed to assess catheter safety.

We report the results of a randomized clinical trial performed in 36 intensive care unit (ICU) centers in France, in which we assessed the effect of the early insertion of a PAC without goal-oriented therapy on mortality and morbidity in patients with shock mainly of septic origin, ARDS, or both.

Patients were eligible for inclusion if they met criteria for shock, ARDS, or both. Shock was defined by the presence of less than 12 hours of 4 criteria: heart rate of at least 90/min; a respiratory rate of at least 20/min or a PaCO₂ of 32 mm Hg or lower or the use of mechanical ventilation; the use of vasopressors to maintain a systolic blood pressure of at least 90 mm Hg despite fluid resuscitation, low dose of dopamine (≤5 µg/kg per minute), or dobutamine; and at least 1 of 3 signs of hypoperfusion (urine output <0.5 mL/kg of body weight per hour for 1 hour or more; neurologic dysfunction defined by confusion, psychosis, or a Glasgow coma scale score of ≤6; plasma lactate higher than the upper limit of the normal value). Patients with shock were excluded if more than 12 hours had elapsed since the presence of the 4 criteria had occurred or if the last 2 criteria were present for more than 12 hours.

Acute respiratory distress syndrome was defined by the presence of more than 24 hours of 4 criteria: acute decrease in PaO₂/FIO₂ to 200 mm Hg or lower, whatever the level of positive end-expiratory pressure; bilateral pulmonary infiltrates or a chest radiograph consistent with edema; no clinical evidence of left atrial hypertension; and requirement for positive pressure ventilation. Patients with ARDS were excluded if less than 24 hours had elapsed since the presence of the 4 criteria had occurred.

Patients were excluded if they were younger than 18 years, experienced hemorrhagic shock, myocardial infarction complicated by cardiogenic shock requiring revascularization, or thrombocytopenia (≤10.0 × 10⁹/L), participated in other trials within the last 30 days, were moribund, or if their physician refused to agree with the use of full life support.

Patients were enrolled from January 30, 1999, to June 29, 2001, at 36 French centers. In each center, at least 1 member of the medical staff was affiliated with the Société de Réanimation de Langue Française. The protocol was approved by the institutional review board of Hôpital de Bicêtre. Written informed consent was obtained from the patients or surrogates. As patients could understand, they were informed of their right to withdraw from the study. The trial was monitored by an independent data and safety monitoring board. Randomization was conducted centrally by telephone on a 24-hour-a-day, 7-day-a-week basis and based on a permuted-block algorithm, allowing stratification for each center.

No standardized protocols for managing patients were proposed; therefore, each patient was treated by his/her physician. The centers agreed on the following principles: optimization of circulating blood volume; vasoactive support if necessary at a mean arterial pressure of at least 60 mm Hg when fluid balance was optimal; no objective for maximization of oxygen transport; free access to echocardiography; assisted control ventilation with a maximum plateau pressure of 35 cm H₂O and SpO₂ of more than 90%; and prevention of thromboembolism with low-molecular-weight heparin, if not contraindicated. The PAC had to be inserted during the 2-hour period following randomization. The type of PAC was decided at the discretion of the physician. The type, site, and duration of PAC insertion as well as the onset of complications were recorded. The decision to remove the PAC or to replace it was at the discretion of the ICU team.

The primary end point was mortality at 28 days. The secondary end points were mortality at 14 and 90 days, duration of ICU and hospital stay, ICU and hospital free days, defined as the number of days from day 1 to day 28 without ICU stay or without hospital stay; ventilator-free days as the number of days from day 1 to day 28 during which the patient breathed spontaneously; and renal support, organ system, and vasoactive agents–free days as the number of days from day 1 to day 14 without renal support, organ system failure, or use of vasoactive agents.

At ICU admission, we recorded age; sex; severity of underlying medical condition (McCabe and Jackson)¹³; the Simplified Acute Physiology Score II (SAPS II)¹⁴; the Sepsis-related Organ Failure Assessment score¹⁵; and the organ dysfunction and infection score.¹⁶ Patients were classified as medical, scheduled, or unscheduled surgical patients. Signs of organ failure according to the Brussels score were simultaneously recorded.¹⁷ At baseline, we recorded mechanical ventilation, SAPS II, Sepsis-related Organ Failure Assessment score, organ dysfunction and infection score, presence and cause of shock and ARDS, PaO₂/FIO₂, and organ failure according to the Brussels score.¹⁷ Patients were monitored daily for 14 days for organ failure. Complications related to the PAC insertion, maintenance, and removal were recorded. Arrhythmia was determined by electrocardiography. Sepsis from the PAC was defined by inflammation at the insertion site and systemic sepsis by sepsis plus a positive culture of blood or catheter tip that resolved with removal of the catheter.

Since use of the PAC could not be blinded, an extensive on-site monitoring procedure was used to avoid bias evaluating morbidity. Calculation of severity scores was not performed by the investigators but centrally calculated by the statisticians from raw data recorded.

We calculated the sample size to detect a 10% difference in the mortality at day 28 between the 2 groups with a 2-tailed test, a significance level of 5%, and a power of 90%. For a global mortality with balanced groups of 40% (35% vs 45%), we planned to enroll 1100 patients, taking into account 1 interim analysis after the inclusion of 500 patients.¹⁸ Prior to our study, it was difficult to have an accurate estimate of the expected mortality based on our specific inclusion criteria. As per our protocol, the global mortality was estimated after the first 20 validated deaths; thereafter, we did not change our sample size calculation.

The study was scheduled for an 18-month period. This limit was chosen to avoid changes in practice regarding the PAC since the trial could not be blinded. A slower rate of enrollment than expected was observed because of the severity of the inclusion criteria (presence of shock for <12 hours and presence of ARDS for >24 hours). Because of this slow rate of inclusions, the steering committee recommended limiting the power of the study to 80% instead of 90%. With regard to the observed rate of inclusion, this decision required inclusion of 754 patients during a 30-month period. This decision was approved by the data and safety monitoring board. The interim analysis was conducted according to the protocol. Finally, the rate of inclusions further declined, so by the end of the 30-month period, 681 patients were included and the data and safety monitoring board decided to cease the study. Under these conditions, the power of the study to detect a 10% absolute difference, using the mortality observed in our control group and taking into account the interim safety analysis, was 78%.

The baseline characteristics, complications, and outcomes in the 2 groups were compared with the use of the χ², Fisher exact, and t tests. All tests were 2-sided. The primary analysis consisted of a comparison of mortality at day 28 by a χ² test on an intention-to-treat basis. The crude mortality relative risk (RR) at day 28 was estimated. Overall survival curves were obtained using the Kaplan-Meier method and compared using the log-rank test. A Mantel-Haenszel χ² test of homogeneity was performed to compare the mortality RR among all centers that included 10 or more patients.

A multivariate analysis was performed, using the Cox proportional hazards regression model stratified at hospital centers, to estimate hazard ratio (HR) adjusted for age, sex, SAPS II, shock or ARDS at inclusion, the medical origin of the patients (medical, scheduled surgical, or unscheduled surgical), the use of echocardiography, and the insertion of a central venous catheter at baseline. All the statistical tests were performed with the Stata statistical software version 6.0 (Stata Corp, College Station, Tex) and P≤.05 was considered statistically significant.

From January 30, 1999, to June 29, 2001, 681 patients were enrolled at 36 ICU centers (Figure 1). Five patients withdrew their consent, leaving a total of 676 patients randomized to the PAC group (n = 335) and the control group (n = 341). Among the 36 centers, 15 included fewer than 10 patients (1-9) and 12 included more than 20 patients (21-112). The percentages of patients in the PAC vs control groups randomized for shock (63.1% and 64.2%, respectively), ARDS (29.5% and 31.2%), or both conditions (7.4% and 4.6%) did not differ significantly (P = .32). Admission and baseline characteristics did not differ between the 2 groups with the exception of the classification of medical, scheduled surgical, or unscheduled surgical patients (P = .02) (Table 1).

A PAC was inserted in 15 (4.4%) of 341 patients in the control group. In contrast, 8 (2.4%) of 335 patients in the PAC group did not receive it; 6 died before insertion and in 2 placement was not possible. Three patients were lost to follow-up between day 28 and day 90 (1 patient in the PAC group and 2 in the control group) (Figure 1).

Overall mortality at day 28 was 60.2%. On an intention-to-treat basis, this percentage did not differ significantly between the PAC and the control group at day 28: 59.4% vs 61.0% (199 vs 208 deaths, with an absolute difference of –1.6%; 95% confidence interval [CI], –9.0 to 5.8) and a mortality RR associated with the PAC compared with the control group of 0.97 (95% CI, 0.86-1.10; P = .67) (Table 2). Figure 2 shows the Kaplan-Meier estimates of survival. The crude HR, estimated using the Cox proportional hazards regression model, was 1.0 (95% CI, 0.82-1.22). Similar results to those shown in Table 2 were obtained when the 23 patients who did not receive the allocated treatment were excluded.

The mortality RRs for all centers that included 10 or more patients did not differ significantly (P = .37) and a term for treatment × center interaction was not included in the analysis. In 1 center (n = 112), a significant difference in mortality in favor of the PAC group was observed. After adjustment for the severity on admission (SAPS II), this difference was no longer significant with an adjusted HR of 0.69 (95% CI, 0.42-1.13; P = .14).

Table 3 shows the results of the between-group comparison concerning the secondary end points. Day 14 mortality was 49.9% in the PAC group vs 51.3% in the control group (RR, 0.97; 95% CI, 0.84-1.13; P = .70) and day 90 mortality was 70.7% vs 72.0% (RR, 0.98; 95% CI, 0.89-1.08; P = .71), respectively. No statistically significant differences were observed for any of the secondary end points according to the intention-to-treat analysis.

The mortality RRs at day 28 were approximately equal to 1, whatever the reason for inclusion and the severity assessed by quartile of SAPS II (Table 2). With adjustment for the other variables listed in Table 4, the adjusted HR associated with the PAC group was 0.97 (95% CI, 0.79-1.20; P = .78). An increased mortality risk was associated with increased SAPS II (P<.001) and with shock and ARDS compared with ARDS alone (P = .02).

The catheter was left indwelling for a mean 2.3 days (range, 1-10) in the entire population and a mean 2.6 days in the survivors (range, 1-10). Ninety percent of survivors had an indwelling PAC for less than 5 days. Complications during pulmonary catheter insertion included arterial puncture (n = 17), hemothorax (n = 1), arrhythmias and conduction disturbances (n = 60), and knots (n = 6). No death attributable to ventricular fibrillation or to arrhythmia was reported. No pulmonary embolism and deep venous thrombosis were recorded. In 2 of 10 patients with positive PAC culture after insertion, the blood cultures were primary positive (once with Enterobacter cloacae and once with Staphylococcus epidermidis) and adequate antibiotic therapy was administered. For the other 8 patients, local clinical signs and sepsis resolved with removal of the PAC without antibiotics. For these patients, the microorganisms were: 5 S epidermidis, 1 Pseudomonas aeruginosa, 1 Klebsiella pneumoniae, and 1 Citrobacter freundii.

The major findings of our trial are that clinical management involving early use of a PAC was not associated with significant changes in mortality and morbidity among patients with shock, ARDS, or both. Severe complications were infrequent.

These findings do not confirm those reported by Connors et al.⁴ Inclusion of a more heterogeneous and less severely ill population might account for this discrepancy. To explain these findings, Connors et al⁴ reported complications related to maintenance of the PAC, the role of errors in measurements or misinterpretation of data that could have resulted in erroneous decisions, and also a possible aggressive therapeutic approach. Previous studies showed that deliberate increase in oxygen delivery resulted in either no benefit^12,19 or harmful effect.¹⁰ However, it is also possible that the study performed by Connors et al⁴ may have overestimated the mortality in the group with a PAC because of the limitation of retrospective matching of patients. Connors et al⁴ chose to use a propensity score but this score did not take into account the intensity of treatment used to sustain hemodynamics. This approach could have masked a higher illness severity in the patients with a PAC. This crucial issue was illustrated in a retrospective cohort study of patients with ARDS, which showed that the apparent increased mortality associated with the use of the PAC disappeared when the use of vasoactive agents was taken into account in the multivariate analysis.²⁰

We chose to enroll a selected population of patients who were not submitted to a goal-oriented protocol to define the role of the PAC. Our decision to perform a study without any goal-oriented therapy was based on the harmful results found in observational studies without use of formal protocols and the lack of consensus regarding hemodynamic support in septic shock or in ARDS.^4,21

No significant differences in mortality between patients with and without PAC either at day 14, 28, or 90 were observed. There was also no difference between the 2 groups in day 14 organ system failure, renal support, vasoactive-free days, or in day 28 ICU hospital or ventilator-free days. The overall day 28 mortality rate was very high (60.2%), as in the PAC group (59.4%) and the control group (61.0%). The higher mortality than previously planned in our protocol (60% vs 40%) may be partly explained by the following. When we planned our study, it was difficult to have a realistic idea of the mortality rate in patients with such strict inclusion criteria identifying seriously ill patients. In a recent randomized clinical trial involving new drugs,²² very careful selection of patients was performed to avoid a potentially inflated rate of adverse effects. In our protocol, we choose to include patients with an accurate diagnosis of shock, ARDS, or both. Thus, we did not include patients with shock for more than 12 hours or patients with ARDS for less than 24 hours. This decision precluded studying partly resolved shock or respiratory failure. The high inclusion rate in our study, 87.4% in the PAC group and 81.8% in the control group (Table 1), could also explain the high mortality because comorbidities are often more frequently observed in medical rather than in surgical patients. The calculation of the standardized mortality ratio confirmed the lack of influence of PAC on outcome: 1.16, 1.09, and 1.23 for the global population, PAC, and control groups, respectively.¹⁴ At least 1 other randomized trial is now in progress to define the role of the PAC.²³

Randomization, stratified by center, was conducted centrally by telephone. The details of each telephone call were recorded in a database by patient eligibility and randomization group. This approach was useful to avoid patient protocol violations. Our study shows that a majority of physicians were willing to have their patients participate in this type of controlled trial²⁴ because a PAC was inserted in only 15 of 341 patients in the control group. These data are different from a trial performed a few years ago in Canada that was stopped after only 33 patients were included.²⁵ Physicians may have had less uncertainty about the use of PAC at the time that this initial Canadian trial was conducted, and thus may have been less likely to allow their individual patients to undergo randomization.²⁵ In contrast, our results were comparable with a recently published study.²⁶ The exclusion of patients who violated the study protocol did not affect our results.

Our data, which are in agreement with those of a recently published single-center study,²⁶ strongly suggest that PAC is not associated with increased mortality or morbidity. The power of our study to detect a 10% absolute difference, using the mortality observed in our control group and taking into account the interim analysis performed, was 78%. Our study was underpowered to detect an absolute difference of 5% corresponding to the odds ratio of 1.24 observed in the study by Connors et al.⁴ However, the design-power method does not take into account the observed difference in the 2 groups²⁷ once the study was completed. In our study, we found an RR difference in day 28 mortality associated with a narrow CI (RR, 0.97; 95% CI, 0.86-1.10), suggesting an absolute difference of –1.6% with an upper bound of the 95% CI of less than 6% (95% CI, –9% to 5.8%). Following the CI method,²⁸ we can conclude at an α risk of 5% that the absolute difference in mortality rate between the 2 groups is no more than 7.8%. This result was close to the 5% difference in mortality rate reported in the study by Connors et al.⁴ We did not observe a center × treatment interaction because most RRs were near or equal to 1. The RRs remained near or equal to 1 among the different subgroups of baseline predictors of mortality, such as SAPS II and the type of diagnosis at inclusion (Table 2). Taking into account the use of echocardiography at day 1, the PAC was not associated with mortality (Table 2). No differences were found in morbidity criteria between PAC and control groups (Table 3).

We performed a careful review of the complications linked to insertion and maintenance of the PAC. The incidence of complications was low. Concerning the major complications, our results are in accordance with previous studies reporting rates of 0.1% to 0.5%.^5,29,30 We observed PAC-related infections in 10 patients (2.8%). This incidence is also in accordance with the most precise reported estimate of PAC and/or introducer sheath–related infections (5.9% to 29.1%) and with that of PAC bacteriemia (0% to 4.6%).³¹ However, our study may underestimate the true incidence of this complication because efforts to diagnose infection were in response to suggestive clinical findings instead of through screening for possible infection.

In contrast with the recently reported study in high-risk surgical patients, we did not observe a higher incidence of pulmonary embolism in the group assigned to PACs.¹¹ Thromboembolic complications range from less than 1% to up to 11% of patients with indwelling PACs. This incidence is obviously difficult to determine in critically ill patients in whom the suggested clinical signs of pulmonary embolism are not easy to detect. This was the case in our study in which we did not anticipate the diagnosis of pulmonary embolism in the presence of a PAC by scheduled diagnostic testing. However, to prevent the onset of this complication, 4 preventive measures were suggested in our protocol. Persistent wedging of the PAC and prolonged occlusion of a proximal artery by an inflated balloon was forbidden. The use of heparin-coated catheter was routinely performed. Prevention of thromboembolism by low-molecular-weight heparin was used. The removal of the PAC was suggested as soon at it was no longer required. This last recommendation partly explains the relatively short duration of pulmonary artery catheterization in our study (mean, 2.3 days) compared with observational studies.⁴ Despite the low incidence of complications observed, it is possible that some of the complications we observed, such as ventricular arrhythmia, catheter-related sepsis, and central venous access complications, may have offset the potential benefits of the PAC.

Even if the purpose of monitoring with PAC is ultimately to save lives, it would be unrealistic to believe that the prognosis of patients could be improved by its presence alone. An influence on prognosis without goal-oriented therapy could only be suggested when the presence of a PAC results in significant changes in treatment with fluid loading and vasoactive agents.³²

In our study, in both patient groups, the physicians were able to obtain relevant information by using echocardiography, thereby influencing diagnosis and treatment.³³ The ICU centers chosen for this multicenter randomized study have free access to echocardiography. This technique was used to complete evaluation of cardiovascular status by noninvasive determination of left ventricle ejection fraction. At least one echocardiographic examination was performed in 64% of the PAC group and 78% in the control group during the ICU stay. In this latter group, echocardiography was used as a morphologic tool to assess ejection fraction and also as a dynamic tool with Doppler analysis to evaluate cardiac output and estimate pulmonary artery pressure and left ventricle end-diastolic pressure. Echocardiography can be used instead of a PAC but also as a complementary technique to PAC when available.

In conclusion, our multicenter randomized trial demonstrates that the PAC remains a safe procedure for the management of patients with shock, ARDS, or both. However, we did not observe a morbidity or mortality benefit. Our results, which do not preclude the potential impact of a goal-oriented therapy with a PAC, strengthen the suggestion of the consensus statement made by the National Heart, Lung, and Blood Institute and the Food and Drug Administration that a randomized clinical trial with this design can be ethically performed in this population of critically ill patients.⁸