Conflicting evidence exists regarding the risks and benefits of inotropic therapies during cardiac surgery, and the extent of variation in clinical practice remains understudied. Therefore, the authors sought to quantify patient-, anesthesiologist-, and hospital-related contributions to variation in inotrope use.
In this observational study, nonemergent adult cardiac surgeries using cardiopulmonary bypass were reviewed across a multicenter cohort of academic and community hospitals from 2014 to 2019. Patients who were moribund, receiving mechanical circulatory support, or receiving preoperative or home inotropes were excluded. The primary outcome was an inotrope infusion (epinephrine, dobutamine, milrinone, dopamine) administered for greater than 60 consecutive min intraoperatively or ongoing upon transport from the operating room. Institution-, clinician-, and patient-level variance components were studied.
Among 51,085 cases across 611 attending anesthesiologists and 29 hospitals, 27,033 (52.9%) cases received at least one intraoperative inotrope, including 21,796 (42.7%) epinephrine, 6,360 (12.4%) milrinone, 2,000 (3.9%) dobutamine, and 602 (1.2%) dopamine (non–mutually exclusive). Variation in inotrope use was 22.6% attributable to the institution, 6.8% attributable to the primary attending anesthesiologist, and 70.6% attributable to the patient. The adjusted median odds ratio for the same patient receiving inotropes was 1.73 between 2 randomly selected clinicians and 3.55 between 2 randomly selected institutions. Factors most strongly associated with increased likelihood of inotrope use were institutional medical school affiliation (adjusted odds ratio, 6.2; 95% CI, 1.39 to 27.8), heart failure (adjusted odds ratio, 2.60; 95% CI, 2.46 to 2.76), pulmonary circulation disorder (adjusted odds ratio, 1.72; 95% CI, 1.58 to 1.87), loop diuretic home medication (adjusted odds ratio, 1.55; 95% CI, 1.42 to 1.69), Black race (adjusted odds ratio, 1.49; 95% CI, 1.32 to 1.68), and digoxin home medication (adjusted odds ratio, 1.48; 95% CI, 1.18 to 1.86).
Variation in inotrope use during cardiac surgery is attributable to the institution and to the clinician, in addition to the patient. Variation across institutions and clinicians suggests a need for future quantitative and qualitative research to understand variation in inotrope use affecting outcomes and develop evidence-based, patient-centered inotrope therapies.
While administration of inotropic infusions is sometimes necessary to support cardiac output after cardiac surgery with cardiopulmonary bypass, inotrope use is counterbalanced by risks of unwanted ischemic and arrhythmogenic effects
Little is known about the degree of variability of inotrope use in cardiac surgical patients and what the key drivers of such variability might be
While presenting patient comorbidities and characteristics are related to variability in inotrope use during cardiac surgery with cardiopulmonary bypass, the institution where surgery is performed and the attending anesthesiologist caring for the cardiac surgical patient are also factors associated with increased variability in inotrope usage
Among more than 300,000 cardiac surgeries performed in the United States annually,1 variation in clinical decision-making for blood transfusions,2 hemodynamic management,3 and anesthetic techniques4,5 are well described. However, one knowledge gap remaining in perioperative care variation for cardiac surgery is the use of inotropic therapies. While inotropes may achieve their intended physiologic effect and objectively improve cardiac contractility, such medications may also expose patients to potentially severe unintended consequences, including myocardial ischemia and malignant arrhythmia6,7 and increasing mortality.8 Furthermore, as inotropes often require administration via invasive central lines and skilled intensive care unit nursing, these medications are associated with adjusted hospital and intensive care unit length of stays prolonged by 1 to 3 days9,10 and $17,000 increased adjusted total inpatient hospital costs per patient.10 These findings have contributed to variable practice patterns described in high-risk cardiac surgical subpopulations,11 surveys exploring clinician decision-making,12 and single-center assessments of factors influencing inotrope use.13 However, variation has not been sufficiently assessed in a broad cardiac surgical population across multiple centers using perioperative electronic medical record data.
Understanding the factors driving such variation remains important for informing strategies to reduce variation if subsequently found to negatively affect patient outcomes.14–16 In other healthcare contexts, variation ideally reflects precision medicine, yet it can also reflect local culture, lack of agreement on optimal care, or care departing from established guidelines.17 Given increased needs for sustainable and high-value care, clinicians and policymakers have developed quality improvement initiatives and clinical practice guidelines aimed at reducing variation if found to negatively affect outcomes.18,19 To characterize where additional knowledge gaps may exist, clinical database registries can highlight sources of variation not clearly explained by adjusting for surgical and patient characteristics, which may serve as targets for further exploration and—if found to negatively affect outcomes—quality improvement initiatives.1,20 Sources of variation can occur at multiple levels, including: (1) the patient level, influenced by factors such as demographics, comorbidities, or access to care; (2) the clinician level, influenced by factors such as training, experience, or preference; and (3) the institutional level, influenced by factors such as resource availability, hospital operations, institutional preference or culture, and the setting of healthcare delivery.21 Whereas variation in care can be explained at different levels, understanding the relative contribution of each remains critically important, as each source raises unique issues about health equity, quality, and appropriateness of care allocation, and each implies different strategies for reducing any given component of variation if negatively affecting outcomes.22,23
To inform efforts to identify and measure sources of practice variation, we performed this multilevel observational cohort study across multiple centers, characterizing relative contributions of institution-, clinician-, and patient-level factors influencing the use of intraoperative inotrope infusions during cardiac surgical procedures. We hypothesized that potentially meaningful variation in inotrope use (greater than 5%) occurred at the clinician and institution levels and that characteristics influencing a patient’s likelihood to receive intraoperative inotropes spanned multiple perioperative data types including demographics, comorbidities, surgical procedure details, home medications, and clinician- and institution-level characteristics.
Materials and Methods
We followed the REporting of studies Conducted using Observational Routinely collected health Data (RECORD) extension of the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines throughout conducting this study (Supplemental Digital Content 1, https://links.lww.com/ALN/D142).24 Institutional review board approval (HUM00181872) was obtained for this observational study, and patient consent was waived. An a priori study protocol for the patient population, data collection and handling, primary outcome, and statistical methods was approved within a peer-review forum25 before statistical analysis and made publicly available on Open Science Framework.26
We studied nonemergent cardiac surgical procedures with cardiopulmonary bypass performed on adult patients older than 18 yr old at U.S. institutions from January 1, 2014, to August 1, 2019 (start date selected based upon data available, and end date selected to mitigate impact of unmeasured confounders related to the Coronavirus Disease 2019 pandemic), within the Multicenter Perioperative Outcomes Group registry. Institutions submitting valid inotrope data (details in “Data Handling”) contributing greater than 20 cardiac cases per year were eligible for inclusion. To develop a cohort reflective of typical cardiac surgical procedures, we restricted the study population to coronary artery bypass and valve procedures performed in isolation or combination. We excluded cardiac surgical procedures with pre-existing or implanted mechanical circulatory support (e.g., intra-aortic balloon pump, ventricular assist device, or extracorporeal membrane oxygenation), transcatheter or off-pump procedures, procedures with circulatory arrest, myectomies, patients receiving inotrope infusions before surgical incision, and American Society of Anesthesiologists Physical Status Classification V or VI patients. For patients undergoing multiple cardiac surgical procedures meeting inclusion criteria, only the index case was used. Finally, we excluded procedures with a case duration of less than 120 min or without invasive arterial blood pressure monitoring, as such surgeries were unlikely to be a complete cardiac surgery.
After study approval and registration, data were extracted from the Multicenter Perioperative Outcomes Group data set. The methods for local electronic health record data acquisition, validation, mapping to semantically interoperable universal Multicenter Perioperative Outcomes Group concepts, and secure transfer to the coordinating center have been previously described.20,27
We defined the primary binary outcome of interest as an inotrope infusion (epinephrine, dobutamine, milrinone, or dopamine) for more than 60 continuous intraoperative min or ongoing upon transport from the operating room and arrival to the intensive care unit.
To further characterize the extent of inotrope use among the cases studied, we defined a secondary outcome as the total number of intraoperative simultaneous inotrope infusions used. Inotrope infusions were considered simultaneous if used together for greater than 60 consecutive min or if used together during transport from the operating room.
We collected data on covariates available within the Multicenter Perioperative Outcomes Group postulated by the authors to influence intraoperative inotrope administration based upon literature review,9,11–13,28 as well as other factors (e.g., electronic health record data–reported sex and race) that have been previously observed to drive variation in other aspects of perioperative care29 (table 1). These included characteristics of the patient (demographics, anthropometrics, comorbidities, home medications, preoperative studies, and preoperative status), surgical case (type, times, and intraoperative events), clinician (primary attending anesthesiologist, clinician case volume), and institution (medical school affiliation, institutional case volume, number of attending cardiac anesthesiologists, and percentage of cases involving nurse anesthetists). Patient comorbidity data were collected using the Elixhauser Comorbidity Enhanced International Classification of Diseases, Ninth and Tenth Revisions, Clinical Modification algorithm.30 Demographics, laboratory values, and surgical case characteristics were validated utilizing precomputed, Multicenter Perioperative Outcomes Group–specific, publicly available perioperative electronic health record data phenotype algorithms.31 Relevant cardiovascular home medications were collected via natural language processing of home medication free-text entries within the preoperative history and physical and with classification via Veterans Affairs national formulary codes.32 Academic institutions were determined by whether or not the institution had an associated medical school; a detailed list is available via the “medical school affiliation” perioperative electronic health record data phenotype.31
To assess the accuracy of the inotrope administration primary outcome, within each eligible Multicenter Perioperative Outcomes Group institution, a sample of 10 cases receiving intraoperative inotropes and 10 cases not receiving inotropes were hand-reviewed by a cardiac anesthesiologist (M.R.M.); institutions with less than 95% agreement between the query algorithm and the manual review were excluded from the analysis. For each covariate, outlier values were handled as missing if outside of valid ranges described in the prespecified phenotype specifications.31 Missing data patterns were assessed, and the percentage of missing data was calculated. For missingness rates less than 10%, complete case analyses were performed; otherwise, multiple imputation techniques were to be used to complete the data.33
Statistical analyses were performed using SAS version 9.4 (SAS Institute, USA). Distributions of variables were assessed graphically, and potential outliers were identified via histograms, Q-Q plots, box plots, and basic descriptive statistics (mean, SD, median, and interquartile range). These were also used to determine appropriate covariate transformations and modeling strategies. For descriptive purposes, we characterized (1) per-case rates of inotrope administration at the clinician and institution levels via caterpillar plots; (2) temporal trends in inotrope administration via linear plots; (3) per-case rates of individual inotrope administration (epinephrine, dobutamine, milrinone, and dopamine) across institutions via density plots; and (4) per-case rates of single versus multiple simultaneous inotrope infusions via stacked bar charts.
Associations between covariates and the primary outcome were assessed via univariate analyses using standardized differences. Covariates showing standardized differences larger than 0.2 in absolute value between groups or plausibly influencing intraoperative inotrope administration were considered for inclusion in multivariable models. In addition, multicollinearity was assessed for using Pearson’s correlation coefficient and variance inflation analyses. In cases of covariate pairs leading to a correlation coefficient greater than 0.70 and variance inflation factor greater than 10, one covariate was removed from the multivariable model based on the clinical judgment of the investigators.
Unadjusted (null) generalized linear mixed models were next constructed, using a hierarchical nesting structure with patients nested within clinicians, nested within institutions. Modified intraclass correlation estimates were used to assess relative contributions of institution-, clinician-, and patient-level factors to the total variance of intraoperative inotrope infusions.34 Intraclass correlation estimates can be used to express the percentage of variation observed within a specific group of variables, relative to the total variation observed. For example, an intraclass correlation estimate of 80% for patient-level factors is interpreted as 80% of the total observed variation in inotrope use being attributable to characteristics of the patient and the remaining 20% attributable to characteristics of the clinician or institution. Intraclass correlation estimates can be used to ascertain the validity of a nested, multilevel approach to modeling the data observed; for example, if less than 5% of the total variability is explained by upper-level units, then limited empirical support for a multilevel analysis exists, favoring a single-level model using a generalized estimating equation approach.
Given that more than 5% of the total variance was explained by institution and clinician upper levels (as described in the Results section), adjusted generalized linear mixed models were next constructed to (1) compute adjusted median odds ratios for receiving intraoperative inotropes across clinicians and institutions and (2) analyze independent associations between covariates and intraoperative inotropes. The median odds ratio is the median value obtained from comparing adjusted odds of having received intraoperative inotropes, if the same patient underwent cardiac surgery at two different randomly selected institutions or under the care of two different randomly selected clinicians. For example, a median odds ratio of 1.3 is interpreted as the median odds of receiving an intraoperative inotrope infusion would be 30% higher if the same patient underwent cardiac surgery at one randomly selected institution versus another or under the care of one randomly selected clinician versus another. Pseudo-likelihood ratios and generalized chi-square statistics were used to characterize the suitability of multilevel multivariable models as the proper approach.
To assess the impact of varying inclusion criteria and intraoperative inotrope definitions on the results, we performed several sensitivity analyses. First, we repeated the primary analysis using an alternative intraoperative inotrope definition including norepinephrine. We then repeated the primary analysis using separate alternative intraoperative inotrope definitions restricted to each individual inotrope. Additionally, we repeated the primary analysis, restricting the intraoperative inotrope definition to (1) only cases with inotrope infusions ongoing during transport from the operating room and (2) only considering the time period between the end of the final cardiopulmonary bypass and transport from the operating room. Finally, we repeated the primary analysis, categorizing case types based upon current procedural terminology code rather than surgical procedure text.
Sample Size Calculation
An a priori minimum sample size was determined based on the desired precision of the prevalence of intraoperative inotrope administration to be within 3%. To estimate the true proportion of intraoperative inotrope administration, we used a 95% CI based upon the standard error (SE) of the sample proportion (p) using the following formula:
P ± 1.96*SE(p)
Supplemental Digital Content 2 (https://links.lww.com/ALN/D143) outlines estimated sample sizes with variable sample proportions and precision. Among a preliminary sample of data available on cardiac surgeries performed across all Multicenter Perioperative Outcomes Group centers from 2014 to 2019, we determined the proportion of cases receiving any inotrope administration (irrespective of duration) to be 56% and thus would require a sample size ranging between 5,500 and 5,800 cases to achieve 3% precision, assuming 25% loss of data due to additional data cleaning protocols. In addition, simulation-based sample size estimates for multilevel models show that a sample size of 27,000 patients nested within clinicians and institutions would yield 80% or more power to determine statistically significant associations at the 5% significance level. Given an estimated sample size of greater than 40,000 patients analyzed, this study achieved adequate power.
Patient Population: Baseline Characteristics
Of the 119,044 cardiac surgical cases reviewed, 51,085 met inclusion criteria (fig. 1). Cases meeting inclusion criteria comprised 611 attending anesthesiologist clinicians across 29 U.S. hospitals. The hospitals studied are listed in appendix 2 and included 24 academic hospitals (82.8%) contributing 47,500 cases (93.0%) and 5 community hospitals (18.2%) contributing 3,585 cases (7.0%). Manual case audits of inotrope infusion data across each of the 29 included hospitals demonstrated greater than 95% accuracy, and therefore, all 29 hospitals were included in the analysis. The population had a mean age of 64 yr, 31.4% were women, and 76.5% were White (table 1). Cardiac surgeries performed included valve (22,987, 45.0% of cases), coronary artery bypass (20,681, 40.5%), and valve/coronary artery bypass combination (7,414, 14.5%).
Across the study population, 27,033 (52.9%) received intraoperative inotropes. Individual non–mutually exclusive inotropes included epinephrine (21,796, 42.7% of cases), milrinone (6,360, 12.4%), dobutamine (2,000, 3.9%), and dopamine (602, 1.2%). Compared to patients without inotropes, those receiving inotropes more commonly (1) had preoperative comorbidities of heart failure, pulmonary circulation disorders, and renal failure; (2) underwent valve/coronary artery bypass combination surgeries; (3) had longer cardiopulmonary bypass durations; (4) had greater rates of blood product transfusion; (5) had lower preoperative estimated glomerular filtration rates and hemoglobin concentrations; and (6) were prescribed loop diuretic home medications.
Institution and Clinician Inotrope Use Patterns
Density plots of inotrope use per institution are shown in figure 2 (total inotrope use) and Supplemental Digital Content 3 (https://links.lww.com/ALN/D144; per-institution relative inotrope use). Inotrope use per institution ranged from 6.9 to 85.0% (median, 47.0%; interquartile range, 33.1 to 67.8%; fig. 3A). Simultaneous use of 2 or more inotropes per institution ranged from 0 to 1.4%, with 14 institutions (48.3% of institutions) never using 2 or more simultaneous inotropes (fig. 4). Among the 15 institutions ever using 2 or more simultaneous inotropes, a median of 0.4% and interquartile range of 0.2 to 0.7% of cases used 2 or more simultaneous inotropes. Inotrope use per clinician ranged from 0 to 100% (median, 45.2%; interquartile range, 30.0 to 67.0%), with 6.8% of clinicians using inotropes for all cases and 8.5% for no cases (fig. 3B). Temporal variation in institution-, clinician-, or patient-level inotrope use followed no discernible pattern (Supplemental Digital Content 4, https://links.lww.com/ALN/D145).
Across the study population, 3,805 (7.4%) had missing data, resulting in a complete case analysis cohort of 47,280 patients, 611 anesthesiologist attending clinicians, and 29 institutions for the multilevel model. A missing data analysis demonstrated that minimal additional bias was introduced when cases with high rates of missing data were excluded (Supplemental Digital Content 5, https://links.lww.com/ALN/D146).
Nested Multilevel Modeling
Goodness-of-fit testing for a multivariable mixed-effect model with random effects of clinician and institution demonstrated that the variability in inotrope use was properly modeled without residual overdispersion (likelihood ratio test, P < 0.001; generalized chi-square test, 1.1). Within the unadjusted model, 22.6% of the variation in inotrope use was attributable to the institution, 6.8% to the primary anesthesiologist attending clinician, and 70.6% to the patient (table 2). After adjustment, the amount of variation attributed to the institution and clinician increased (35.1% and 9.2%, respectively), while it decreased to 55.6% for the patient. The adjusted median odds ratio for a patient receiving inotropes was 1.73 at the clinician level and 3.55 at the institution level. Put into context, for any given patient, the median odds of receiving inotropes during cardiac surgery differed by three- to four-fold between two randomly selected institutions and by nearly two-fold between two randomly selected attending anesthesiologists, after adjustment for baseline characteristics. For further illustration, for a hypothetical patient whose characteristics were associated with a 50% chance of receiving an inotrope by a given anesthesiologist and institution, the median chance of receiving an inotrope would increase to 63.4% or decrease to 36.6% if receiving care from another randomly selected anesthesiologist and would increase to 78.0% or decrease to 22.0% if receiving care at another randomly selected institution.
After nested multilevel modeling adjusting for clinician- and institution-level factors, the patient-level factors independently associated with a statistically and clinically significant increased likelihood of inotrope use (P < 0.05 and adjusted odds ratio greater than 1.25) were heart failure (adjusted odds ratio, 2.60; 95% CI, 2.46 to 2.76), pulmonary hypertension/embolism (adjusted odds ratio, 1.72; 95% CI, 1.58 to 1.87), loop diuretic home medication (adjusted odds ratio, 1.55; 95% CI, 1.42 to 1.69), Black race (adjusted odds ratio, 1.49; 95% CI, 1.32 to 1.68), digoxin home medication (adjusted odds ratio, 1.48; 95% CI, 1.18 to 1.86), preoperative heart rate higher than 90 beats/min (versus 60 to 75 beats/min, adjusted odds ratio, 1.42; 95% CI, 1.30 to 1.55), and lower preoperative estimated glomerular filtration rate versus at least 90 ml · min–1 · 1.73 m–2 (less than 30 ml · min–1 · 1.73 m–2, adjusted odds ratio, 1.28; 95% CI, 1.06 to 1.54; 30 to 59 ml · min–1 · 1.73 m–2, adjusted odds ratio, 1.31; 95% CI, 1.19 to 1.44; and 60 to 89 ml · min–1 · 1.73 m–2, adjusted odds ratio, 1.11; 95% CI, 1.03 to 1.19; table 3). Additionally, continuous variables independently associated with increased likelihood of inotrope use were prolonged case duration (adjusted odds ratio, 1.31; 95% CI, 1.27 to 1.34) and prolonged cardiopulmonary bypass duration (adjusted odds ratio, 1.21; 95% CI, 1.17 to 1.26). At the clinician level, no association was observed between attending anesthesiologist case volume and the likelihood of inotrope use; however, variation in inotrope use was significantly greater for low-volume attending anesthesiologists (quintile 1; fewer than 19 cases in data set annually) compared to all other anesthesiologist subgroups with higher case volumes (quintiles 2 to 5), Supplemental Digital Content 6 (https://links.lww.com/ALN/D147). At the institution level, 24 of 29 hospitals (83%) were medical school affiliated (i.e., teaching hospital), and medical school affiliation was strongly associated with an increased odds of inotrope use (adjusted odds ratio, 6.2; 95% CI, 1.39 to 27.8); however, no other associations were observed between other institution-level characteristics and inotrope use.
When including norepinephrine infusions as part of the cardiac inotrope infusion outcome, norepinephrine infusions were used in 39.5% of cases (simultaneously with other inotropes in 24.4% of all cases) and an outcome incidence (any inotrope used, including norepinephrine) of 71.9% was observed. Compared to the primary model, cardiac inotrope variance estimates using the norepinephrine-included inotrope primary outcome definition were observed to have similar relative proportions of variance attributable to the patient, clinician, and institution levels, and similar covariates were independently associated with cardiac inotropes (Supplemental Digital Content 7, https://links.lww.com/ALN/D148).
When restricting the cardiac inotrope infusion outcome to (1) consider only inotrope infusions ongoing at the time of transport from the operating room or (2) consider only inotrope infusions occurring after cardiopulmonary bypass, variance estimates attributable to the patient, clinician, and institution levels were also observed to be similarly distributed as with the overall analysis, and similar covariates were independently associated with cardiac inotropes (Supplemental Digital Content 8, https://links.lww.com/ALN/D149).
Additionally, when restricting the cardiac inotrope infusion outcome to individual inotropes, variance estimates attributable to the patient, clinician, and institution levels were observed to be similarly distributed as with the overall analysis. Multivariable models converged for epinephrine and norepinephrine, but not dobutamine, milrinone, or dopamine; model covariate-associated epinephrine and norepinephrine primary outcomes were similar to the primary analysis (Supplemental Digital Content 9, https://links.lww.com/ALN/D150). Finally, when categorizing case types based upon current procedural terminology codes rather than surgical procedure text, variance estimates and covariates associated with cardiac inotropes were similar to the primary analysis (Supplemental Digital Content 10, https://links.lww.com/ALN/D151).
In this multicenter study of cardiac surgeries across 29 U.S. academic and community hospitals, we report wide variation in intraoperative inotrope use across clinicians and institutions, with significant variation attributable to the anesthesiologist attending clinician and institution rather than solely the patient or surgery. Factors driving such clinician- and institution-level differences are complex and multifactorial, potentially explained by clinician training, institutional or regional protocols, cultural dogma, resource availability, the setting of healthcare delivery, or patient factors that cluster by clinician or institution but remain unmeasured and therefore appear to be otherwise unexplained.21,35 However, as our findings were similar to other studies of cardiac anesthesiology practice patterns,5,36 the advantages of multicenter over single-center analyses continues to be underscored as they more completely capture the diversity of practices and more accurately reflect patterns in which clinical care is delivered.
The wide variation in inotrope use in the modern, broad cardiac surgical population studied was consistent with historic analyses of high-risk cardiac subpopulations.11 Similarly, factors independently associated with inotrope use paralleled previous studies, with the exception of medical school affiliation (i.e., teaching hospital) as the strongest factor observed in our analysis.11,13 Whereas the lack of association observed between institutional case volume and inotrope usage was consistent with previous findings,11 our divergent finding that institutions affiliated with a medical school were strongly and independently associated with inotrope use demands further investigation through qualitative research of clinician attitudes and institutional protocols toward inotrope use. Factors conceivably explaining this association include (1) the greater diversity of cases in our study, (2) potentially greater degrees of recent changes to historic practice patterns at medical school-affiliated institutions compared to community hospitals, and/or (3) unmeasured confounders that differed across cardiac surgical cases at medical school-affiliated institutions compared to community hospitals. Regarding the relationship between higher attending anesthesiologist case volume and lower clinician-level variance in inotrope use, further explanation of such findings remained beyond the scope of this study, although it may have reflected a more patient-centered approach to inotrope use among higher-volume anesthesiologists.
Although variation in inotrope use remained wide, patient-level multivariable associations between perioperative characteristics and inotrope use observed in this study were largely consistent with known predictors of low cardiac output syndrome, including heart failure, renal insufficiency, low preoperative hemoglobin, and prolonged cardiopulmonary bypass time.37,38 Notably, however, we observed that despite previous evidence suggesting that females are at higher risk of low cardiac output syndrome,39 female sex was independently associated with lower rates of inotrope use in our study. Conversely, an even stronger independent association—in the opposite direction—was observed for Black non-Hispanic patients, who had a nearly 50% increased adjusted odds of receiving inotropes. Such findings may be explainable by unmeasured confounders clustering within each subgroup and associated with severity of cardiovascular disease (e.g., social determinants of health, access to healthcare, or under-recognized inequities in cardiac surgical care for such patients) or may be a function of clinician bias.40,41
Potentially underpinning the variation in inotrope use we observed, which remained robust to multiple sensitivity analyses, are under-quantified risks versus benefits to such therapies. Whereas inotropes may achieve their intended physiologic effect and objectively improve a patient’s hemodynamics or oxygen delivery to end organs, such medications also expose patients to potentially severe unintended consequences including myocardial ischemia,6,7 malignant dysrhythmia,6,7 and central line-associated bloodstream infections as the need for central venous access may be prolonged in the setting of specific inotropes.42 Taken together, the variation in inotrope use observed in our study and variation in outcomes observed in previous studies, suggests a need for prospective trials to investigate optimal inotrope strategies for improving cardiac surgery outcomes. Should such trials be pursued, it should be noted that one-size-fits-all strategies to inotrope use after cardiac surgery, which have historically yielded indeterminate or conflicting conclusions, are unlikely to be effective.43–45 Rather, inotrope administration strategies that account for the heterogeneity of treatment effects and dynamic patient recovery trajectories across diverse surgical populations may be necessary to guide optimal inotrope use for cardiac surgery. Indeed, within a broader perioperative care context, the need to reduce components of variation negatively affecting outcomes not by introducing more standardized therapies but rather by introducing more patient-centered therapies has been underscored in shortcomings to recent clinical trial designs comparing interventions in a one-size-fits all fashion. In such trials, a lack of superiority of any one standardized intervention across all outcomes was found, offering the conclusion that the “best” treatment may be less based upon objective study results and more on how each individual—clinicians and patients alike—values each outcome.46
More broadly still, what constitutes warranted versus unwarranted variation in health care has been a topic of recent debate.47 A modern synthesis of the literature has suggested that clinical care variation can arise from (1) patients’ and clinicians’ agency, (2) scientific and clinical evidence, and (3) personal and organizational capacity.23 As related to agency, warranted versus unwarranted variation dichotomizes when patients’ preferences (driving warranted variation) are adequately informed yet superseded by solely clinician preferences (unwarranted). Related to evidence, variation dichotomizes when judgment in applying evidence into a local context (warranted) is absent (unwarranted). Finally, related to capacity, variation dichotomizes when intractable resource constraints and unpredictable events lead to clinician adaptation (warranted variation) versus when clinicians have varying levels of competency or technical proficiency despite local availability of training resources (unwarranted). Although our study decomposes variation in inotrope use into the patient, clinician, and institution levels, it should be noted that whereas patient-level variation might theoretically be conceived as more likely to be warranted and clinician- and institution-level variation more likely to be unwarranted, this is not necessarily the case. To make progress on understanding components of variation that are unwarranted and—if found to negatively affect patient outcomes—could be reduced, an important next step includes defining optimal medical decision-making in a way that considers not only population-level evidence but also patient and clinician preferences, heterogeneity of treatment effects, and local institutional policies and resource availability.
Our study has multiple important limitations which must be carefully considered when interpreting results. First, although using intraoperative data available via a robust multicenter data set, we were unable to fully capture all factors potentially influencing a clinician’s decision to administer inotropes, and therefore components of the observed variation remained unexplained. Most notably, quantitative preoperative and intraoperative structured data describing cardiac function (e.g., left ventricular ejection fraction, right ventricular systolic function, cardiac index, mixed venous oxygen saturation, and so forth), attending surgeon identifiers, or surgical details beyond valve/coronary artery bypass and cardiopulmonary bypass duration (e.g., previous sternotomy, cardioplegia type/dose, cannulation strategy, and so forth) were not routinely available. The covariates collected, including cardiovascular comorbidities, medications, and intraoperative events indicative of case complexity, were used but did not comprise any previously developed risk score and likely incompletely accounted for such factors.
Second, this study involved secondary use of routinely collected electronic health record data across institutions with heterogeneous documentation patterns. Although we leveraged a novel perioperative data set across multiple U.S. institutions and used validated, semantically interoperable Multicenter Perioperative Outcomes Group concepts with advanced techniques for handling aberrations in data,20 clinical rationales for inotrope administration were unavailable, and our study remained subject to a level of data quality inherent to observational research. Third, although detailed intraoperative documentation of inotrope administration was available, data describing postoperative intensive care unit use of inotropes were unavailable. Although data were captured on inotrope infusions continued at the time of transport from the operating room, conclusions regarding variation in the postoperative continuation or new administration of inotropes cannot be drawn from our study.
Next, although both academic medical centers and community hospitals were included in our multicenter data set, the data were primarily from academic centers, and data from institutions outside of the United States were not available for this study, precluding more detailed analyses. However, hospital-level case volumes for the cardiac surgical procedures included in our study were similar to other studies reporting outcomes across a wide range of U.S. centers.48,49 Lastly, our study did not investigate outcomes after inotrope use, and therefore, insights regarding whether variation was warranted versus unwarranted remain unknown. Such analyses were not performed, given the high likelihood for unaddressed confounding within the causal structure underlying potential analyses relating inotrope use to cardiac surgical outcomes, which may have yielded misguided conclusions. Prospective studies of individualized, dynamic inotrope interventions versus routine care are needed to adequately assess any putative associations between inotrope use and patient outcomes.
Within a national, multicenter cohort of cardiac surgeries across academic and community hospitals, half of the patients received intraoperative inotrope infusions. Variation in inotrope use was explained by clinician- and institution-level factors in addition to patient factors. These data provide insight into the extent of cardiac anesthesiology practice variation and suggest a need for future prospective trials of patient-centered inotrope use, seeking to understand whether cardiac surgery outcomes can be improved and whether unwarranted variation can be reduced.
The authors gratefully acknowledge Robert Coleman, B.S. (Department of Anesthesiology, University of Michigan Medical School, Ann Arbor, Michigan) for his contributions in data acquisition and electronic search query programming for this project.
Supported by departmental and institutional resources at each contributing site. In addition, the underlying electronic health record data collection into the Multicenter Perioperative Outcomes Group registry was provided by Blue Cross Blue Shield of Michigan/Blue Care Network as part of the Blue Cross Blue Shield of Michigan/Blue Care Network Value Partnerships program (Detroit, Michigan). Although Blue Cross Blue Shield of Michigan/Blue Care Network and Multicenter Perioperative Outcomes Group work collaboratively, the opinions, beliefs, and viewpoints expressed by the authors do not necessarily reflect the opinions, beliefs, and viewpoints of Blue Cross Blue Shield of Michigan/Blue Care Network or any of its employees. The project was additionally supported in part by U.S. National Institutes of Health (Bethesda, Maryland) grant Nos. K01HL141701 and R01DK133226. Industry contributors have had no role in the study.
Dr. Mathis received research grant No. K01HL141701 from the U.S. National Institutes of Health (Bethesda, Maryland). Dr. Janda received research grant No. T32GM103730 from the U.S. National Institutes of Health and received research support paid to the University of Michigan (Ann Arbor, Michigan) and (unrelated to the current work) from Becton, Dickinson and Company (Franklin Lakes, New Jersey). Dr. Schonberger received research grant No. R01AG059607 from the U.S. National Institutes of Health. Dr. Schonberger reports owning stock in Johnson & Johnson (New Brunswick, New Jersey) unrelated to the current work. Dr. Schonberger also reports that his organization receives funding from Merck, Inc. (Rahway, New Jersey), for a study in which he is coinvestigator unrelated to the current work. Dr. Shook reports support from Edwards Lifesciences (Irvine, California) and UpToDate (Waltham, Massachusetts) unrelated to the current work and has provided expert review and testimony unrelated to the current work. Dr. Muehlschlegel has received research grant Nos. R01HL14998 and R01HL150401 from the U.S. National Institutes of Health unrelated to the current work. The other authors declare no competing interests.
Supplemental Digital Content
Supplemental Digital Content 1. RECORD Extension of STROBE Checklist, https://links.lww.com/ALN/D142
Supplemental Digital Content 2. Power analysis, https://links.lww.com/ALN/D143
Supplemental Digital Content 3. Additional density plots of inotrope use, https://links.lww.com/ALN/D144
Supplemental Digital Content 4. Inotrope use temporal variation, https://links.lww.com/ALN/D145
Supplemental Digital Content 5. Missing data analysis, https://links.lww.com/ALN/D146
Supplemental Digital Content 6. Anesthesiologist case volume and inotrope use, https://links.lww.com/ALN/D147
Supplemental Digital Content 7. Sensitivity analysis: Inotropes and norepinephrine, https://links.lww.com/ALN/D148
Supplemental Digital Content 8. Sensitivity analyses: Intraoperative inotrope timing, https://links.lww.com/ALN/D149
Supplemental Digital Content 9. Sensitivity analyses: Individual inotropes, https://links.lww.com/ALN/D150
Supplemental Digital Content 10. Sensitivity analysis: Cases characterized using procedural codes, https://links.lww.com/ALN/D151
Appendix 1: Multicenter Perioperative Outcomes Group Collaborators
Ruth Cassidy, Ph.D.: Department of Anesthesiology, University of Michigan Medical School, Ann Arbor, Michigan.
David J. Clark, M.D., Ph.D.: Department of Anesthesiology, Stanford University, Palo Alto, California.
Douglas A. Colquhoun, M.B.Ch.B., M.Sc., M.P.H.: Department of Anesthesiology, University of Michigan Medical School, Ann Arbor, Michigan.
Robert E. Freundlich, M.D., M.Sc.: Department of Anesthesiology, Vanderbilt University Medical Center, Nashville, Tennessee.
Elizabeth S. Jewell, M.S.: Department of Anesthesiology, University of Michigan Medical School, Ann Arbor, Michigan.
Appendix 2: Multicenter Perioperative Outcomes Group – Study Institutions
Beaumont Hospital of Dearborn, Dearborn, Michigan
Beaumont Hospital of Royal Oak, Royal Oak, Michigan
Beaumont Hospital of Troy, Troy, Michigan
Brigham and Women’s Hospital, Boston, Massachusetts
Bronson Healthcare Group, Battle Creek, Michigan, and Kalamazoo, Michigan
Cleveland Clinic, Cleveland, Ohio
Duke University Hospital, Durham, North Carolina
Henry Hord Health System, Detroit, Michigan
Massachusetts General Hospital, Boston, Massachusetts
New York University Langone Medical Center, New York, New York
Oregon Health and Science University, Portland, Oregon
Sparrow Health System, Lansing, Michigan
Stanford Health Care, Palo Alto, California
Trinity Health Muskegon Hospital, Muskegon, Michigan
Trinity Health Ann Arbor Hospital, Ann Arbor, Michigan
University of California Los Angeles Medical Center, Los Angeles, California
University of California San Francisco Medical Center, San Francisco, California
University of Colorado Denver Health Medical Center, Denver, Colorado
University of Michigan Health System, Ann Arbor, Michigan
University of Oklahoma Health Sciences Center, Oklahoma City, Oklahoma
University of Pennsylvania Health System, Philadelphia, Pennsylvania
University of Tennessee Medical Center, Knoxville, Tennessee
University of Utah Health Care, Salt Lake City, Utah
University of Vermont Health Network, Burlington, Vermont
University of Virginia Health System, Charlottesville, Virginia
Washington University of St. Louis School of Medicine, St. Louis, Missouri
Atrium Health Wake Forest Baptist Medical Center, Winston-Salem, North Carolina
Weill Cornell Medical College New York, New York
Yale New Haven Hospital, New Haven, Connecticut