Arterial catheters are used frequently in intensive care units (ICUs). Clinical effectiveness and adverse events associated with the use of the catheters have not been formally evaluated in clinical studies.
To determine whether an association exists between arterial catheter use and hospital mortality in ICU patients.
Design, Setting, and Participants
Propensity-matched cohort analysis of data in the Project IMPACT database, from 2001 to 2008. A total of 139 ICUs in the United States were included. Participants were ICU patients 18 years or older.
Arterial catheter use.
Main Outcomes and Measures
Our main outcome was hospital mortality. We assessed a primary cohort of medical patients requiring mechanical ventilation and 9 secondary cohorts. We used propensity score–matched pairs as the primary analytic strategy. Sensitivity analyses included 4 alternative methods of comparison in the primary cohort: multivariate modeling without propensity adjustment, mixed-effects multivariate logistic regression without propensity adjustment, multivariate modeling with propensity adjustment, and stratification based on propensity quintiles.
Our primary cohort consisted of 60 975 patients; 24 126 of these patients (39.6%) had an arterial catheter in place during their ICU stay, and analyses were based on 13 603 propensity score–matched pairs. We found no association between arterial catheter use and hospital mortality in medical patients requiring mechanical ventilation in the primary analysis (odds ratio [OR], 0.98; 95% CI, 0.93-1.03; P = .40) or the 4 sensitivity analyses (P ≥ .58 for all). In 8 of 9 secondary cohorts we were unable to detect an association between arterial catheter use and hospital mortality. In the cohort of patients receiving vasopressors, arterial catheter use was associated with an increased odds of death (OR, 1.08; 95% CI, 1.02-1.14; P = .008).
Conclusions and Relevance
In this propensity-matched cohort analysis, arterial catheters were not associated with improvements in hospital mortality in medical ICU patients requiring mechanical ventilation. Given the costs and potential harms associated with invasive catheters, randomized clinical trials are needed to further evaluate the usefulness of these frequently used devices.
The use of diagnostic and therapeutic interventions in modern medical practice should ideally be supported by evidence demonstrating net benefit. However, many interventions in use over long intervals are assumed to be beneficial without data to support them.1-3
Arterial catheterization is a commonly used intervention in intensive care units (ICUs) worldwide. In the United States, 36% of all patients receive an arterial catheter (AC) during their ICU stay.4 Arterial catheters are inserted to facilitate diagnostic phlebotomy, augment hemodynamic monitoring, or monitor arterial blood gases.5-8 However, ACs are associated with measurable risks, such as limb ischemia,9 pseudoaneursyms,10,11 and catheter infections,12-18 as well as costs for insertion and maintenance.19,20 Significant variability in AC use exists across ICUs in the United States that is more attributable to an individual ICU than to patient-specific characteristics.4 Practice variability may stem partially from a paucity of data regarding the effect of ACs on patient-centered outcomes in the critically ill population. We, therefore, conducted a cohort study to examine the association between AC use and clinical outcomes in critically ill patients. We hypothesized that we would observe no association between AC use and hospital mortality.
We performed an analysis of all adults (≥18 years) admitted from January 1, 2001, through December 31, 2008, to ICUs in the United States participating in Project IMPACT (Cerner Corporation).21 Institutional review board exemption was obtained from Albert Einstein College of Medicine. Project IMPACT was developed to provide regular performance audits and feedback to participating ICUs. Participation was voluntary and ICUs paid for the service. Information was collected by on-site data collectors who were certified to ensure standardization in data definitions and entry. Hospitals participating in Project IMPACT tended to be larger and more urban than general-population hospitals, but were diverse in size and location. Data were either from consecutive admissions to each ICU or a random sample of admissions. Sites using the latter method collected information on 50% or 75% of all patients; the percentage was determined quarterly before data collection commenced.
We included only the patient’s initial ICU admission for a given hospital stay. Available data included patient demographics (age, sex, race/ethnicity, and insurance provider), severity of illness as described by the Mortality Probability Model III–predicted hospital mortality at ICU admission (MPM0-III),22 preference for cardiopulmonary resuscitation at ICU admission and during the ICU stay, Acute Physiology and Chronic Health Evaluation II (APACHE II) diagnostic category,23 location before ICU arrival, need for cardiopulmonary resuscitation within 24 hours before ICU arrival, vital signs on ICU admission (heart rate >150 beats/min, systolic blood pressure <90 mm Hg, fraction of inspired oxygen required >50%), patient categorization (medical, emergency surgery, or elective surgery), comorbidities, year of ICU admission, and number of organs failing during the ICU admission. Comorbidities were defined by a list of 16 chronic conditions used for severity-of-illness score calculations. Explicit definitions for organ failure were provided by Project IMPACT.24 Data on interventions included the use of invasive mechanical ventilation, the use of intravascular catheters (arterial, central venous, and/or pulmonary artery), and vasopressor administration by continuous intravenous infusion (epinephrine, norepinephrine, dopamine, phenylephrine, and/or vasopressin). Intensive care units and hospitals were characterized according to ICU type, ICU model (closed vs open), mean nurse to patient ratio, and hospital organization (city, state, or federal government; community; or academic).
Reporting of data on intravascular catheter use, including ACs, was mandatory in Project IMPACT. Patients were considered to have had an intravascular catheter if it was in place for any portion of the ICU stay, regardless of whether it was placed in the ICU or before ICU arrival. Arterial catheters were included if they were placed in the dorsalis pedis, radial, axial, femoral, or brachial arteries; documentation of the site was not required and was not consistently available.
Our primary outcome was hospital mortality. Other patient-centered outcomes included days requiring vasopressor support, days of mechanical ventilation, and ICU length of stay. We additionally evaluated the rate of packed red blood cell (PRBC) transfusions. The PRBC transfusions were hypothesized to be a negative consequence of AC use because of additional phlebotomy and/or bleeding at the catheter site.25 We also conducted falsification analyses by testing for associations between AC use and 4 other interventions that were not presupposed to be related to AC use: platelet transfusions, paracentesis, lumbar puncture, and transesophageal echocardiography.
To reduce patient-level heterogeneity, for our primary analysis we assessed a primary cohort of medical patients (no surgery within the 7 days before ICU admission) who arrived in the ICU from any location other than the operating room or postanesthesia care unit and who required mechanical ventilation at any point during their ICU stay. Prior work established that there was large variability in the use of ACs in this population.4 To evaluate the generalizability of our findings, we repeated the main analyses on 9 secondary cohorts: medical patients requiring mechanical ventilation in the (1) lowest, (2) middle, and (3) highest tertiles of MPM0-III; (4) medical patients requiring mechanical ventilation admitted to mixed medical-surgical ICUs in which AC use for surgical patients was within the 25th to 75th percentile of all mixed ICU use (to address issues of possible AC documentation errors); (5) all ICU patients (ie, not limited to medical patients or patients requiring mechanical ventilation); (6) all patients admitted to mixed medical-surgical ICUs; (7) all patients requiring vasopressors at any point during their ICU stay; (8) all patients with septic shock (as defined by the sepsis APACHE II diagnostic category plus the need for vasopressors at any point during the ICU stay); and (9) surgical patients (underwent surgery within the 7 days before ICU admission and were admitted to the ICU from the operating room or postanesthesia care unit) who required mechanical ventilation at any point during their ICU stay. Owing to the high rates of ACs placed for intraoperative purposes, for the surgical mechanically ventilated cohort, we defined AC use in the ICU as either placement in the ICU or placement before arrival in the ICU with removal more than 1 day after arrival in the ICU. In this analysis, all patients who died within 24 hours of ICU admission were excluded and comparison was made with surgical patients requiring mechanical ventilation who never had an AC.
The statistical plan was determined a priori. Additional analyses were conducted as requested upon manuscript review.
Association Between AC Use and Hospital Mortality
Five distinct analytic approaches were taken to assess the relationship between AC use and hospital mortality: (1) propensity matching (our primary analysis), (2) multivariate logistic regression without propensity adjustment, (3) mixed-effects multivariate regression without propensity adjustment, (4) multivariate logistic regression with propensity adjustment, and (5) stratification based on propensity quintiles.26-28 Only the primary analysis (propensity matching) was performed for the secondary cohorts.
Propensity Score Calculation
A propensity score is a probabilistic measure that reflects the propensity of a patient, based on other characteristics, to receive an AC. This score is not a marker of whether particular patients received an AC but rather whether they were likely to do so given their characteristics. For each cohort, the propensity to receive an AC was calculated using multivariate logistic regression with AC use as the dependent variable and all available patient characteristics as the independent variables.29-31 Clustering of patients within individual ICUs, which we a priori assumed would have a significant effect on the use of ACs,4 was modeled by including the individual ICU as a fixed effect in the model.32,33 To allow for nonlinear relationships, continuous variables were included in the model as 4-knot cubic splines.34 The success of the propensity score is determined by its ability to balance independent covariates between the patients who received ACs and those who did not. This covariate balance was assessed using standard mean differences.35
Propensity-Matched Analysis (Primary Analysis)
Propensity matching is a technique in which quasi-case/control pairs are produced from a retrospective cohort. For our cohorts, patients who received an AC were matched to patients who did not receive an AC but had a similar propensity to do so. Matching was performed, after randomly ordering patients, using the psmatch2 algorithm36 in Stata, version 11.1 (StataCorp), with 1 to 1 nearest-neighbor matching without replacement and with maximal caliper distance of 25% of the SD of all propensity scores. In addition, exact matching was used for each covariate for which the propensity score did not achieve appropriate balance.
Standard summary statistics were used to compare the baseline patient-, ICU-, and hospital-level characteristics between groups of propensity-matched cases and controls. The odds of hospital mortality were compared between the 2 matched groups using χ2 statistics.
Multivariable Logistic Regression Without Propensity Adjustment
We used multivariable logistic regression to assess the association between hospital mortality and AC use after adjusting for all covariates. The individual ICU was included as a fixed effect.
Mixed-Effects Multivariable Logistic Regression Without Propensity Adjustment
We used mixed-effects multivariable logistic regression to assess the association between hospital mortality and AC use after adjusting for all covariates. All patient-level independent variables were entered into the model as fixed effects; the individual ICU was included as a random effect.
Multivariable Logistic Regression With Propensity Adjustment
The addition of propensity adjustment to a multivariate model has been advocated as a method to further adjust for confounders that may be related to the independent and the dependent variables.37 In this analysis, we performed the same multivariate logistic regression outlined in the previous section with the addition of the propensity score as an independent variable to calculate the adjusted odds of hospital mortality associated with AC use.
Stratification Based on Propensity Quintiles
Stratification based on propensity percentiles is a method for comparing outcomes among patients with a similar likelihood of receiving the intervention (eg, AC use). Stratification into quintiles is commonly used because it is adequate to remove 90% of all bias.38 We divided our primary cohort into quintiles of propensity score and evaluated the odds of hospital mortality associated with AC use within each quintile. Combination of these odds ratios (ORs) into a single OR across the entire cohort was done using the Mantel-Haenszel method.39
Sensitivity Analysis to Address Potential Immortal Time Bias
Immortality bias arises when patients in a retrospective analysis who are exposed to the intervention of interest have their exposure at varying times after the start of the study period.40,41 These individuals must have survived until their intervention, making them “immortal” in the time preceding it. Thus, exposed patients are biased toward better survival outcomes than their nonexposed counterparts who do not experience this immortal time. To address this potential bias, we conducted a sensitivity analysis of patients in the primary cohort who survived for at least 2 days, comparing matched pairs of those who received an AC during their first ICU day vs those who never received an AC.
Association Between AC Use and Other Patient-Centered Outcomes
Using the primary analytic strategy (propensity matching), we evaluated the association between AC use with vasopressor days (the total number of ICU days on which the patient received ≥1 vasopressor agents), mechanical ventilation duration (the first episode as well as the total number of days requiring mechanical ventilation), and ICU length of stay. Because these outcomes are also subject to immortal time bias, we conducted these analyses for the primary cohort of medical patients requiring mechanical ventilation as well as the subset of these patients who were in the ICU on day 2 and with or without an AC on day 1.
Association Between AC Use and Other Interventions
We assessed the association between AC use and PRBC transfusions using propensity matching applied to the primary cohort of medical patients requiring mechanical ventilation. To exclude patients who received transfusions for treatment of gross hemorrhage, we restricted transfusion outcomes to patients who received at most 2 U of PRBCs on any given day. Because we assessed the number of PRBCs transfused per day, we excluded patients in whom ICU length of stay was not recorded. A total of 3481 patients (5.7% of the primary cohort) were excluded. After these exclusions, we performed the propensity scoring and matching. For falsification analyses,42 we also evaluated 4 interventions that we hypothesized would not be associated with AC use: platelet transfusion, paracentesis, lumbar puncture, and transesophageal echocardiography. Because many patients received none of these interventions, comparisons used zero-inflated negative binomial regression modeling on the matched patients, with the ICU length of stay as an offset and the AC status being the sole independent variable; in such a model, the exponentiated coefficient represents the rate ratio of units of the intervention per ICU day associated with having had an AC.43 Database management and statistical analyses were performed using Excel (Microsoft) and Stata, version 11.1 (StataCorp).
Characteristics of Matched Cohort
Our primary cohort consisted of 60 975 medical patients who required mechanical ventilation in 139 ICUs in 119 hospitals; 24 126 (39.6%) of the patients had an AC (eTable 1 in the Supplement). Propensity score matching yielded 13 603 pairs of patients who did not have an AC and patients who did have an AC (Table 1 and eTable 2 and eTable 3 in the Supplement). The initial propensity score model for the primary cohort achieved balance on all but 5 covariates (pulmonary artery catheter use, central venous catheter use, vasopressor use, initial fraction of inspired oxygen greater than 50%, and mechanical ventilation on ICU admission); these 5 factors were exactly matched upon to make the final matches. All patient-level characteristics were statistically balanced after matching with the exception of minor differences in age (mean [SD], 59.0 [18.3] in the no-AC group vs 58.2 [18.4] in the AC group) and primary insurance (Medicare 48.5% vs 46.6%, respectively). Two ICU/hospital-level characteristics differed, again by small amounts, between the 2 groups: ICU type (P < .001) and hospital organization (P = .003). The matched cohort was 78.4% white and 56.6% male with a mean (SD) MPM0-III–predicted hospital mortality of 26.9% (23.6%). On ICU admission, 63.5% of the patients were receiving mechanical ventilation; at some point during their ICU stay, 44.5% required vasopressors, 72.8% had a central venous catheter, and 5.3% had a pulmonary artery catheter. Approximately 53% of the patients were admitted to mixed medical-surgical ICUs.
Using propensity matching, we demonstrated no association between AC use and hospital mortality in the primary cohort of medical patients requiring mechanical ventilation: the OR for hospital mortality associated with having an AC was 0.98 (95% CI, 0.93-1.03) (Table 2). The hospital mortality was 35.5% for patients who received an AC and 36.0% for patients who did not. Each of the 4 alternative statistical analyses yielded similar findings, as did the sensitivity analysis focused on minimizing immortal time bias (eTable 4 in the Supplement).
In 8 of the 9 secondary cohorts, propensity-matched analysis revealed no association between AC use and mortality (Figure). In patients requiring vasopressors, the odds of death was increased in patients who received an AC (OR, 1.08; 95% CI, 1.02-1.14; P = .008).
Secondary Patient-Centered Outcomes
Days requiring vasopressors, duration of mechanical ventilation, and ICU length of stay were all greater for patients who received ACs (eTable 5 in the Supplement). Restriction to the subset of patients who had ACs on day 1 and survived at least until day 2 did not alter the results.
Association With Other Interventions
In the primary cohort of medical patients receiving mechanical ventilation, we found no association between PRBC transfusions and AC use (rate ratio, 0.99; 95% CI, 0.82-1.19; P = .91) (Table 3). Similarly, we found no association between platelet transfusion, paracentesis, lumbar puncture, or transesophageal echocardiography and AC use.
In this propensity-matched cohort analysis, we found no association between AC use and hospital mortality in medical ICU patients who require mechanical ventilation. Our results were robust through 4 different modeling methods. Similarly, in the analyses of 8 of 9 secondary cohorts, we found no association between AC use and hospital mortality. In one secondary cohort (patients requiring vasopressors), AC use was associated with an 8% increase in the odds of death.
We found no other study that reports clinically meaningful outcomes associated with AC use. There is a growing number of studies, however, evaluating the impact of other monitoring devices that are generally made available with less scrutiny than is required for pharmacotherapeutics.44 First, pulmonary artery catheters, once commonly used in the ICU, are being used less frequently.45-47 Although replacement with noninvasive monitors likely accounts for some of this decline, it likely is also because of the growing evidence that these catheters do not improve outcomes.48-52 Second, a 2009 Cochrane review53 of 5 randomized (and quasirandomized) trials examined the usefulness of continuous-pulse oximetry during anesthesia. Although the American Society of Anesthesiologists mandates the use of continuous pulse oximetry,54 the Cochrane review53 found that although oximetry led to a reduction of hypoxemic events in the recovery room, it was not associated with improved morbidity or mortality. Third, data on the use of continuous telemetry monitoring are conflicting. Two cohort studies55,56 reported that hospitalized patients who have cardiac arrests while undergoing telemetry have higher survival rates than do those who are not monitored. However, a third study57demonstrated that only 56% of cardiac arrests were signaled by telemetry and that only 0.02% of patients survived telemetry-signaled arrests. Finally, 2 recent randomized clinical trials investigated the usefulness of intracranial pressure monitors for traumatic brain injury; one found that monitors improved survival58 and the other found no impact of the device.59
For our primary cohort and 8 of the 9 investigated secondary cohorts, we found no association between AC use and outcomes. There are 2 potential interpretations of these results. The first is that there is no mortality benefit associated with AC use for ICU patients. Alternatively, despite attempts to adjust for confounders, residual confounding remains. Were this to be the case, it is plausible that no mortality effect was detected with ACs because patients who receive ACs are more likely to die, which use of ACs ameliorates. The replication of our results across multiple analyses for the primary cohort and multiple secondary cohorts makes this latter explanation less likely. However, concerns regarding residual confounding can never be eliminated in observational analyses.
Monitoring devices come with increased risks and costs. Prior studies60,61 have found AC use to be associated with more phlebotomy and laboratory testing in the ICU. We did not have access to blood sampling practices for our patients; instead, we used a surrogate marker—PRBC transfusions—to evaluate the potential morbidity of AC use as a result of increased phlebotomy. Arterial catheter use did not result in increased PRBC transfusions in our cohort. However, it is possible that increased phlebotomy—and laboratory testing—occurred but was not sufficient to result in excess transfusions. Increased phlebotomy in the ICU has been shown62 to raise costs without conferring any benefit. In addition, all diagnostic tests have false-positive rates, which frequently result in unnecessary, often invasive, additional testing.63-67
The strength of our study stems from the robustness of our results across multiple cohorts and analytic methods. Our study is limited by the fact that receipt of ACs by individual patients was not random, and we cannot exclude the potential for residual confounding, treatment indication bias, or immortal time bias, which would be required to establish causality. Therefore, our findings should be viewed as hypothesis generating similar to the work by Connors et al48 on pulmonary artery catheters. Because of that initial propensity-matched cohort analysis, numerous randomized clinical trials were performed, none of which supported the belief that pulmonary artery catheters save lives in the ICU.49-52 Nearly 2 decades later, it may be prudent to embark on a similar set of investigations into the usefulness of ACs.
In addition, our data set did not allow us to assess why patients had ACs placed or how information from ACs was used. Having these data would not have changed our main results, but may have allowed us to better understand possible drivers of use. Our data set included patients admitted from January 1, 2001, to December 31, 2008. Although we know that AC use patterns in ICUs in the United States did not change substantially during that time,4 it is possible that meaningful change that we could not capture has occurred more recently. Finally, we were unable to identify matching pairs for 43.6% of the patients who had an AC in our primary cohort. The patients with an AC who we could match had statistically significantly different baseline characteristics (eg, fewer vital sign abnormalities on ICU admission, less use of vasopressors and other intravascular catheters, and less multiorgan failure) (eTable 6 in the Supplement) compared with those who had an AC and were unable to be matched. Our final matched patient cohort, however, was large and representative of ICU patients. Moreover, our matching rates are comparable to those of similar studies.48,68,69
Our results suggest that ACs do not improve the ability to care for ICU patients. Monitoring devices are not without cost and potential harms. Therefore, the possibility that a monitoring device may not help patients highlights the need for randomized clinical trials to evaluate this topic. The results from our subgroup analyses may help to identify populations for enrollment, and the magnitude of the associations that we found may assist planners in determining enrollment targets. With careful planning, such randomized clinical trials could address the effect of AC use on mortality as well as on numerous other meaningful patient-centered outcomes.
Accepted for Publication: June 1, 2014.
Corresponding Author: Hayley B. Gershengorn, MD, Division of Critical Care Medicine, Montefiore Medical Center, 111 E 210th St, Gold Zone, Main Floor, Bronx, NY 10467 (firstname.lastname@example.org).
Published Online: September 8, 2014. doi:10.1001/jamainternmed.2014.3297.
Author Contributions: Dr Gershengorn had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Study concept and design: Gershengorn, Wunsch, Scales, Rubenfeld, Garland.
Acquisition, analysis, or interpretation of data: Gershengorn, Wunsch, Zarychanski, Rubenfeld, Garland.
Drafting of the manuscript: Gershengorn, Wunsch, Garland.
Critical revision of the manuscript for important intellectual content: All authors.
Statistical analysis: All authors.
Study supervision: Wunsch, Garland.
Conflict of Interest Disclosures: None reported.
Funding/Support: Dr Scales is supported by a fellowship in Translational Health Research from the Physicians' Services Incorporated Foundation.
Role of the Sponsor: The funding source had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.
Blood-Letting. BMJ. 1871;1(533):283-284.
RA. An evaluation of internal-mammary-artery ligation by a double-blind technic. N Engl J Med
. 1959;260(22):1115-1118.PubMedGoogle ScholarCrossref
J, van der Sanden
A. Surgical removal versus retention for the management of asymptomatic impacted wisdom teeth. Cochrane Database Syst Rev
. 2012;6:CD003879. doi:10.1002/14651858.PubMedGoogle Scholar
H. Variation of arterial and central venous catheter use in United States intensive care units. Anesthesiology
. 2014;120(3):650-664.PubMedGoogle ScholarCrossref
et al. Comparison of noninvasive oscillometric and intra-arterial blood pressure measurements in hyperacute stroke. Blood Press Monit
. 2007;12(3):149-156.PubMedGoogle ScholarCrossref
et al. Accuracy of oscillometric blood pressure measurement according to the relation between cuff size and upper-arm circumference in critically ill patients. Crit Care Med
. 2000;28(2):371-376.PubMedGoogle ScholarCrossref
et al. Factors influencing the accuracy of oscillometric blood pressure measurement in critically ill patients. Crit Care Med
. 2003;31(3):793-799.PubMedGoogle ScholarCrossref
X. Noninvasive monitoring of blood pressure in the critically ill: reliability according to the cuff site (arm, thigh, or ankle). Crit Care Med
. 2012;40(4):1207-1213.PubMedGoogle ScholarCrossref
TD. Radial artery pseudoaneurysm in the intensive care unit. Ann Vasc Surg.
2010;24(4):554.e13-e16. doi:10.1016/j.avsg.2009.07.039.PubMedGoogle ScholarCrossref
SO. A late complication of percutaneous radial artery cannulation. Acta Clin Belg
. 2011;66(3):223-225.PubMedGoogle Scholar
et al. Bacteremia related with arterial catheter in critically ill patients. J Infect
. 2011;63(2):139-143.PubMedGoogle ScholarCrossref
et al. Infectious risk associated with arterial catheters compared with central venous catheters. Crit Care Med
. 2010;38(4):1030-1035.PubMedGoogle ScholarCrossref
CM. Assessment of peripheral arterial catheters as a source of sepsis in the critically ill: a narrative review. J Hosp Infect
. 2010;75(1):12-18.PubMedGoogle ScholarCrossref
A. Prospective study of peripheral arterial catheter infection and comparison with concurrently sited central venous catheters. Crit Care Med
. 2008;36(2):397-402.PubMedGoogle ScholarCrossref
CJ. The risk of bloodstream infection in adults with different intravascular devices: a systematic review of 200 published prospective studies. Mayo Clin Proc
. 2006;81(9):1159-1171.PubMedGoogle ScholarCrossref
B. Prospective study of arterial and central venous catheter colonization and of arterial- and central venous catheter–related bacteremia in intensive care units. Crit Care Med
. 2005;33(6):1276-1280.PubMedGoogle ScholarCrossref
N. Arterial catheters as a source of bloodstream infection: a systematic review and meta-analysis. Crit Care Med
. 2014;42(6):1334-1339..PubMedGoogle ScholarCrossref
D. Bacterial contamination of arterial lines: a prospective study. JAMA
. 1983;249(2):223-225.PubMedGoogle ScholarCrossref
Jr. Indwelling arterial catheters in the intensive care unit: necessary and beneficial, or a harmful crutch? Am J Respir Crit Care Med
. 2010;182(2):133-134.PubMedGoogle ScholarCrossref
RL; Project IMPACT Clinical Implementation Committee. Project IMPACT: results from a pilot validity study of a new observational database. Crit Care Med
. 2002;30(12):2765-2770.PubMedGoogle ScholarCrossref
AA. Assessing contemporary intensive care unit outcome: an updated Mortality Probability Admission Model (MPM0
-III). Crit Care Med
. 2007;35(3):827-835.PubMedGoogle ScholarCrossref
JE. APACHE II: a severity of disease classification system. Crit Care Med
. 1985;13(10):818-829.PubMedGoogle ScholarCrossref
Data Collection, Therapies and Drugs. Project IMPACT Participation Manual. North Kansas City, MO: Cerner Corp; 2003:77-96.
GF. ABGs and arterial lines: the relationship to unnecessarily drawn arterial blood gas samples. J Trauma.
DB. The central role of the propensity score in observational studies for causal effects. Biometrika
. 1983;70(1):45-51.Google ScholarCrossref
M. Propensity Score Analysis: Statistical Methods and Applications. Thousand Oaks, CA: Sage Publications; 2010.
PC. A tutorial and case study in propensity score analysis: an application to estimating the effect of in-hospital smoking cessation counseling on mortality. Multivariate Behav Res
. 2011;46(1):119-151.PubMedGoogle ScholarCrossref
T. Variable selection for propensity score models. Am J Epidemiol
. 2006;163(12):1149-1156.PubMedGoogle ScholarCrossref
PE. Does matching overcome LaLonde's critique of nonexperimental estimators? J Econom
. 2005;125:305-353.Google ScholarCrossref
F. The specification of the propensity score in multilevel observational studies. Comput Stat Data Anal
. 2011;55(4):1770-1780.Google ScholarCrossref
S. The use of propensity scores for nonrandomized designs with clustered data. Multivariate Behav Res
. 2011;46(3):514-543.Google ScholarCrossref
A. Quantile regression and restricted cubic splines are useful for exploring relationships between continuous variables. J Clin Epidemiol.
B. psmatch2: Stata module which implements full Mahalanobis matching and a variety of propensity score matching methods to adjust for pre-treatment observable differences between a group of treated and a group of untreated.http://ideas.repec.org/c/boc/bocode/s432001.html
. Accessed November 14, 2013.
Jr. Propensity score methods for bias reduction in the comparison of a treatment to a non-randomized control group. Stat Med
. 1998;17(19):2265-2281.PubMedGoogle ScholarCrossref
WG. The effectiveness of adjustment by subclassification in removing bias in observational studies. Biometrics
. 1968;24(2):295-313.PubMedGoogle ScholarCrossref
W. Statistical aspects of the analysis of data from retrospective studies of disease. J Natl Cancer Inst
. 1959;22(4):719-748.PubMedGoogle Scholar
L. Survival bias associated with time-to-treatment initiation in drug effectiveness evaluation: a comparison of methods. Am J Epidemiol
. 2005;162(10):1016-1023.PubMedGoogle ScholarCrossref
AB. Prespecified falsification end points: can they validate true observational associations? JAMA
. 2013;309(3):241-242.PubMedGoogle ScholarCrossref
G. Zero-Truncated and Zero-Inflated Models for Count Data: Mixed Effects Models and Extensions in Ecology with R. New York, NY: Springer; 2003:261-293.
DM. Review of the processes for FDA oversight of drugs, medical devices, and combination products. J Manag Care Pharm
. 2011;17(1):40-50.PubMedGoogle Scholar
HG. Trends in the use of the pulmonary artery catheter in the United States, 1993-2004. JAMA
. 2007;298(4):423-429.PubMedGoogle ScholarCrossref
et al. Pulmonary artery catheters: evolving rates and reasons for use. Crit Care Med
. 2011;39(7):1613-1618.PubMedGoogle ScholarCrossref
H. Understanding changes in established practice: pulmonary artery catheter use in critically ill patients. Crit Care Med
. 2013;41(12):2667-2676.PubMedGoogle ScholarCrossref
et al; SUPPORT Investigators. The effectiveness of right heart catheterization in the initial care of critically ill patients. JAMA
. 1996;276(11):889-897.PubMedGoogle ScholarCrossref
et al; Canadian Critical Care Clinical Trials Group. A randomized, controlled trial of the use of pulmonary-artery catheters in high-risk surgical patients. N Engl J Med
. 2003;348(1):5-14.PubMedGoogle ScholarCrossref
et al; French Pulmonary Artery Catheter Study Group. Early use of the pulmonary artery catheter and outcomes in patients with shock and acute respiratory distress syndrome: a randomized controlled trial. JAMA
. 2003;290(20):2713-2720.PubMedGoogle ScholarCrossref
et al; PAC-Man study collaboration. Assessment of the clinical effectiveness of pulmonary artery catheters in management of patients in intensive care (PAC-Man): a randomised controlled trial. Lancet
. 2005;366(9484):472-477.PubMedGoogle ScholarCrossref
et al; National Heart, Lung, and Blood Institute Acute Respiratory Distress Syndrome (ARDS) Clinical Trials Network. Pulmonary-artery versus central venous catheter to guide treatment of acute lung injury. N Engl J Med
. 2006;354(21):2213-2224.PubMedGoogle ScholarCrossref
K. Pulse oximetry for perioperative monitoring. Cochrane Database Syst Rev
. 2009;(4):CD002013. doi:10.1002/14651858.CD002013.pub2.PubMedGoogle Scholar
RE; American Heart Association’s Get with the Guidelines (formerly, NRCPR) Investigators. In-hospital cardiac arrest: impact of monitoring and witnessed event on patient survival and neurologic status at hospital discharge. Resuscitation
. 2011;82(7):845-852.PubMedGoogle ScholarCrossref
et al. The impact of telemetry on survival of in-hospital cardiac arrests in non-critical care patients. Resuscitation
. 2013;84(7):878-882.PubMedGoogle ScholarCrossref
DA. Continuous electrocardiographic monitoring and cardiac arrest outcomes in 8,932 telemetry ward patients. Acad Emerg Med
. 2000;7(6):647-652.PubMedGoogle ScholarCrossref
J. Increased mortality in patients with severe traumatic brain injury treated without intracranial pressure monitoring. J Neurosurg
. 2012;117(4):729-734.PubMedGoogle ScholarCrossref
et al; Global Neurotrauma Research Group. A trial of intracranial-pressure monitoring in traumatic brain injury. N Engl J Med
. 2012;367(26):2471-2481.PubMedGoogle ScholarCrossref
DP. The effect of arterial lines on blood-drawing practices and costs in intensive care units. Chest
. 1995;108(1):216-219.PubMedGoogle ScholarCrossref
WA. Evaluating laboratory usage in the intensive care unit: patient and institutional characteristics that influence frequency of blood sampling. Crit Care Med
. 1997;25(5):737-748.PubMedGoogle ScholarCrossref
et al. The increasing burden of phlebotomy in the development of anaemia and need for blood transfusion amongst trauma patients. Injury
. 2012;43(1):78-83.PubMedGoogle ScholarCrossref
DL. Cumulative probability of false-positive recall or biopsy recommendation after 10 years of screening mammography: a cohort study. Ann Intern Med
. 2011;155(8):481-492.PubMedGoogle ScholarCrossref
N. False-positive screening mammograms and biopsies among women participating in a Canadian provincial breast screening program. Can J Public Health
. 2012;103(6):e420-e424.PubMedGoogle Scholar
et al. Unknown primary cancer presenting as an adrenal mass: frequency and implications for diagnostic evaluation of adrenal incidentalomas. Surgery
. 1998;124(6):1115-1122.PubMedGoogle ScholarCrossref
LM. Needle biopsy of incidentally discovered adrenal masses is rarely informative and potentially hazardous. Surgery.
et al. High risk of malignancy in patients with incidentally discovered adrenal masses: accuracy of adrenal imaging and image-guided fine-needle aspiration cytology. Tumori
. 2007;93(3):269-274.PubMedGoogle Scholar
et al. Early intravenous unfractionated heparin and mortality in septic shock. Crit Care Med
. 2008;36(11):2973-2979.PubMedGoogle ScholarCrossref
et al; Cooperative Antimicrobial Therapy of Septic Shock (CATSS) Database Research Group. Early combination antibiotic therapy yields improved survival compared with monotherapy in septic shock: a propensity-matched analysis. Crit Care Med
. 2010;38(9):1773-1785.PubMedGoogle ScholarCrossref