Pulmonary artery catheter use in adult patients undergoing cardiac surgery: a retrospective, cohort study

Pulmonary artery catheters (PACs) were introduced in 1970 (Swan et al. 1970) and have advantages over clinical assessment alone for predicting certain cardiac indices (Connors Jr et al. 1983; Iberti and Fisher 1983), detecting hemodynamic abnormalities (Hines 1990), and facilitating oxygen delivery-based protocols (Lobo et al. 2000; Boyd et al. 1993) that may decrease mortality during major surgery (Gurgel and do Nascimento Jr 2011). A subsequent retrospective study showed increased mortality and utilization of health care resources in critically ill intensive care unit (ICU) patients with PACs (Connors Jr et al. 1996), but other studies have shown either no harm or benefit associated with PAC use in critically ill (Harvey et al. 2005; Murdoch et al. 2000), high-risk, elderly, and surgical patients (Sandham et al. 2003), or in patients with symptomatic heart failure (Binanay et al. 2005).

Multiple studies have addressed PAC use in cardiac surgery patients, who currently receive 30% of PACs (Bernard et al. 2000). These range from increased mortality and a greater risk of severe organ complications associated with PAC use during coronary artery bypass grafting (CABG) (Schwann et al. 2011), increased mortality in high-risk cardiac surgery patients (Chiang et al. 2015), and no additional risk of cardiac arrest intraoperatively combined with a non-significant decrease in mortality and a lower likelihood of blood transfusion in patients undergoing CABG (Brovman et al. 2016). In non-emergent CABG patients, PAC use was associated with increased mortality, longer lengths of stay, and higher costs; however, these outcomes were more likely to be seen in hospitals with lower PAC use (Ramsey et al. 2000).

The lack of consensus regarding PAC-based monitoring outcomes in cardiac surgery has many potential reasons, including inadequate study design (Harvey et al. 2008), misinterpretation of data (Parviainen et al. 2006), non-standardized treatments, and training bias (i.e., regular PAC use may influence management of similar patients without PACs (Tuman 1997)). The accuracy and clinical utility of hemodynamic measurements obtained with a PAC correlate with proper catheter placement (Eisenberg et al. 1984) and correct data interpretation by physicians with PAC expertise (Iberti et al. 1994). Despite a lack of consensus and mixed messages surrounding PAC utility, PAC use increased significantly from 2010 to 2014 (Brovman et al. 2016).

Studies investigating clinical outcomes associated with PAC in the cardiac surgical population were performed exclusively in patients undergoing CABG procedures (Schwann et al. 2011; Ramsey et al. 2000), were performed using administrative data (Chiang et al. 2015; Brovman et al. 2016), had limited outcomes, or focused on factors associated with PAC utilization (Brovman et al. 2016). Hence, there is rationale for a study utilizing an electronic health record (EHR) database (including laboratory results and medications) to evaluate clinical outcomes associated with PAC use within all major cardiac procedures, with a cohort matched for hospital and patient characteristics inclusive of risk of mortality. In the current study, adult patients monitored with and without PAC in US hospitals performing a minimum of 100 qualifying cardiac surgeries from January 1, 2011, to June 30, 2015, were evaluated for primary outcomes of 30-day in-hospital mortality, major morbidity, and length of stay. We tested the hypothesis that PACs cause no harm and may provide some benefit for cardiac surgical patients.

Since Connors et al. initially questioned the utility of the pulmonary artery catheter (Connors Jr et al. 1996) in critically ill patients, numerous studies have investigated the effect of PACs in various patient populations using both prospective and retrospective study designs (Schwann et al. 2011; Harvey et al. 2005). These studies have all been influenced to varying degrees by a number of factors, including historical controls, database limitations, population size, population heterogeneity, and PAC expertise. By extracting data from a contemporary US EHR database of over 62 million patients, we sought to mitigate these factors and re-evaluate the potential harms and benefits of PAC use in a real-world cardiac surgery population incorporating laboratory tests and medications, particularly in light of recent work reporting increasing PAC use in cardiac surgeries (Brovman et al. 2016).

This large database population permitted the propensity score matching of patients based on patient and hospital characteristics, including risk of mortality which incorporated lab results, medications, and vital signs (via EuroSCORE II) and the exclusion of patients treated at hospitals performing fewer than 100 qualifying cardiac procedures per year and from institutions without documented use of monitoring PACs to account for the impact of PAC familiarity and expertise on outcomes. We believe that observational studies, such as this study, that are designed to look for evidence of harm (as opposed to benefits) are more robust to indication and selection bias, because health care providers do not make treatment choices with the intent of causing harm. As might be expected based upon current guidelines (Practice guidelines for pulmonary artery catheterization 2003), the unmatched populations did reveal that PACs were preferentially administered to patients with a higher risk profile (median modified EuroSCORE II: PAC 0.016, no-PAC 0.015; p = 0.004).

Using the propensity score-matched cohorts of cardiac surgery patients with or without a PAC, we found that PAC use was not associated with increased risk of harm as measured by in-hospital mortality (30 days). However, despite the large size of our study (6844 patients), because of the low mortality rate in both arms (0.6% no-PAC group; 0.7% PAC group), our study did not have sufficient power to detect a definitive outcome (Pearson chi-square; power 0.0575 at α = 0.0125). PAC use was associated with a statistically significant decrease in length of hospital stay and a significant decrease in the cardiopulmonary morbidity composite, suggesting potential benefits associated with PAC monitoring. Our study also supports a previous study that reported a reduction in transfusions in PAC patients (Cohen et al. 2005). Although we did not observe a significant decrease in transfusion rates in the PAC population, we did find a significant decrease in hemorrhage, suggesting that PAC use may have a positive association with these two related outcome measures (it is important to note that the number of units transfused could not be evaluated within the database between the study arms).

Other retrospective analyses of PACs in cardiac surgeries have revealed conflicting results when assessing mortality. Earlier evaluations with small patient numbers (i.e., < 100) (Larson and Kyff 1989) found no impact, while studies with historical controls reported some benefit associated with PAC use (Schwann et al. 2002). More recently, two large, controlled, retrospective analyses of different national administrative databases (i.e., the National Inpatient Sample and the National Anesthesia Outcomes Registry) reached different conclusions on PAC use and mortality risk, with one study finding a significant increase in mortality risk with PAC use (Chiang et al. 2015) and the other reporting a non-significant decrease in mortality risk with PAC use (Brovman et al. 2016). Notably, the former study (Chiang et al. 2015) found higher rates of mortality in high-risk patient groups such as octogenarian patients, and those with congestive heart failure. Similar to other studies (Binanay et al. 2005; Elliott et al. 1979), we found that PAC use was associated with urinary tract infections (UTIs) and bacteremia. Insertion of a PAC has been shown to be associated with infection (Mermel et al. 1991), but the nature of the observed increase in infection in our study population is unclear. We did not observe differences in line infection between the groups; therefore, it is more likely that the increase in bacteremia in patients monitored with a PAC is due to the higher UTI rate, perhaps due to indwelling Foley catheters which are not typically removed if a PAC is still present. Unfortunately, a limitation of our study is the lack of information about the timing of bladder catheterization (presumably immediately after anesthesia induction), total surgery time, and total bladder catheter dwell time. As placement of central venous catheters (CVCs) may not be routinely coded, examining an association between CVCs and infection is difficult. Another minor limitation is that EuroSCORE II was not designed for heart transplant patients; however, in the current study, patients were matched on cardiac procedure type, and heart transplant patients comprised a small proportion of the population that was not significantly different between arms (no-PAC: 0.8%; PAC 0.7%; p = 0.669).

Proper use and interpretation of PAC-derived data requires expertise and knowledge acquired during training and thereafter supplemented by frequent use. There is evidence to suggest that initial training in PAC use is important for high-quality outcomes (Practice guidelines for pulmonary artery catheterization 2003). Notably, in contrast to the current study, the sources for prior studies do not allow for propensity matching based on hospital demographics or the selection for facilities with documented PAC use to account for provider expertise and experience. This study performed a propensity match which included patient demographics inclusive of pre-existing conditions as Elixhauser comorbidities (e.g., pulmonary hypertension) as well as via the modified EuroSCORE II (inclusive of “critical preoperative state”; Additional file 1 Information S2). However, despite attempts to select for PAC expertise through selection of high volume cardiac surgery hospitals with PAC use, this analysis of EHRs does not indicate the level of experience of the PAC provider, whether the PAC was correctly used, or if a patient’s treatment was informed by PAC readings, highlighting additional study limitations. Local hospital policies and economic considerations influence PAC use (Ranucci 2006), and lack of information about these variables is an additional limitation of our study. A further difference between the current and prior database analyses is our ability to use both coding and clinical parameters to capture outcome measures and better match study arms. In the current analysis, outcomes such as KDIGO AKI and infectious disease are defined by both coding and laboratory test values (and for infectious disease, medications) rather than isolated claims data. EuroSCORE II calculation includes use of laboratory, medication, and vital signs in addition to medical diagnosis and procedure codes. The timing surrounding some of the cardiac components (those defined by medical diagnosis codes) of the cardiopulmonary outcome composite is unknown. Therefore, for example, a new onset of heart failure that occurs in the early phase such as during PAC insertion, or in the late phase of post-op cardiac surgery, cannot be discerned within this study. The database did not allow reliable analysis of CVC utilization within the study cohort, and therefore, this remains a limitation.

Recent years have seen considerable commercial support for new, non-invasive hemodynamic monitoring technologies proposed as PAC alternatives. However, despite reports suggesting neutral or negative outcomes associated with PAC use, and the development of competing monitoring technologies, the use of the PACs in cardiac surgery has remained robust—and perhaps has even increased (Brovman et al. 2016; Judge et al. 2015). This study was performed to investigate this apparent paradox through analysis of patients undergoing a full range of major cardiac procedures using lab values, medications, and vital signs available within an EHR database to both match and derive clinical outcomes. We found that the PAC cohort demonstrated significantly decreased length of stay and cardiopulmonary morbidity versus the no-PAC cohort. These improvements, in the absence of significant changes in in-hospital mortality risk, may help explain the recently reported increase in PAC use in cardiac surgical patients (Brovman et al. 2016; Judge et al. 2015).

In this EHR database study of 6844 patients, we tested the hypothesis that PACs cause no harm and may provide some benefit in cardiac surgeries. We found that use of a PAC in adult cardiac surgery patients was associated with no increased risk of mortality (although insufficiently powered at the low observed mortality rates), a reduced length of stay, and reduced cardiopulmonary morbidity. However, PAC use was also associated with greater incidences of bacteremia and urinary infections, although not line infections. Overall, the lack of serious harm and potential for benefit associated with PAC-based monitoring should inform future prospective trials in well-defined patient populations (such as within those of high vs. low risk) in which patient outcomes for this familiar technology may be tested.