01 / Research article
Research article
Citation
Mathis, M. R., Janda, A. M., Kheterpal, S., Schonberger, R. B., Pagani, F. D., Engoren, M. C., Mentz, G. B., Shook, D. C., Muehlschlegel, J. D., & the Multicenter Perioperative Outcomes Group. (2023). Patient-, clinician-, and institution-level variation in inotrope use for cardiac surgery: A multicenter observational analysis. Anesthesiology, 139(2), 122–141. https://doi.org/10.1097/ALN.0000000000004593 Free accepted manuscript: https://pmc.ncbi.nlm.nih.gov/articles/PMC10524016/
Chapter Summary
This is a 20-page journal article in the standard structure of a clinical research paper: introduction, methods, results, discussion, limitations and conclusions, with supplementary material for the reporting checklist, sample-size calculation and sensitivity analyses. It asks a single question about everyday cardiac anesthesia: when a patient has heart surgery on cardiopulmonary bypass, how much of the decision to start an inotrope infusion is explained by the patient, how much by the attending anesthesiologist, and how much by the hospital?
Introduction
The authors begin from the observation that variation in transfusion, hemodynamic management and anesthetic technique in cardiac surgery is already documented, whereas inotrope use has not been examined across many centers with electronic perioperative records. Inotropes improve contractility but are linked in the paper to myocardial ischemia, malignant arrhythmia, longer hospital and intensive care stays and higher costs. The introduction then separates variation into three levels (patient, clinician, institution) and argues that each level carries different implications for equity, quality and what could be done about it. The stated hypothesis is that more than 5% of the variation sits at the clinician and institution levels.
Methods
The cohort comes from the Multicenter Perioperative Outcomes Group registry: non-emergent adult coronary bypass and valve operations on cardiopulmonary bypass, January 2014 to August 2019. Exclusions include mechanical circulatory support, off-pump and transcatheter procedures, circulatory arrest, patients already on inotropes, ASA class 5 or 6, and very short cases. The primary outcome is an infusion of epinephrine, dobutamine, milrinone or dopamine lasting more than 60 continuous minutes, or still running at transport to the intensive care unit. The reporting follows the RECORD extension of STROBE, and the protocol was published before analysis. The analysis uses nested generalized linear mixed models (patients within clinicians within institutions), intraclass correlation to apportion variance, and the median odds ratio to express clinician and institution effects. A manual audit of 10 inotrope and 10 non-inotrope cases per hospital checked the outcome definition.
Results
Of 119,044 cases reviewed, 51,085 qualified, spanning 611 attending anesthesiologists at 29 hospitals (24 academic, 5 community). Just over half (52.9%) received an intraoperative inotrope, most often epinephrine (42.7%), then milrinone, dobutamine and dopamine. Rates ranged from 6.9% to 85.0% across hospitals and from 0% to 100% across clinicians. In the unadjusted model, 22.6% of the variation was attributable to the institution, 6.8% to the attending anesthesiologist and 70.6% to the patient; after adjustment the institution and clinician shares rose to 35.1% and 9.2%. The adjusted median odds ratio was 1.73 between two random clinicians and 3.55 between two random hospitals. Strongest associations with inotrope use were medical-school affiliation, heart failure, pulmonary circulation disorders, home loop diuretics, Black race and home digoxin. Sensitivity analyses (adding norepinephrine, restricting to particular time windows or to single drugs, recoding procedure types) gave similar variance patterns.
Discussion, Limitations and Conclusions
The discussion reads the findings as evidence of wide, partly unexplained practice variation and calls for qualitative work on why teaching hospitals differ. It notes that female sex was associated with less inotrope use despite higher expected risk of low cardiac output, and that the association with Black race may reflect unmeasured clustering of disease severity or access, or clinician bias. It distinguishes warranted from unwarranted variation and warns that clinician- and institution-level variation is not automatically the unwarranted kind. The limitations section lists missing measures of cardiac function and surgeon identity, unavailable postoperative inotrope data, a mostly academic sample from the United States, and the absence of patient outcomes. The conclusion calls for prospective trials of patient-centered inotrope strategies.
Key Concepts
Introduction and Methods
- Inotrope infusion (primary outcome): a continuous infusion of epinephrine, dobutamine, milrinone or dopamine for more than 60 minutes or ongoing at transfer to intensive care.
- Three levels of variation: patient characteristics, the attending anesthesiologist and the hospital, treated as nested sources of difference in practice.
- Warranted versus unwarranted variation: differences that reflect patient preferences or sensible adaptation, as opposed to differences without a clinical rationale.
- Multilevel (nested) model: a statistical model that recognizes patients as clustered within clinicians and hospitals.
Results and Discussion
- Intraclass correlation: the share of total variance in the outcome associated with a given level of the hierarchy.
- Median odds ratio: the typical change in a patient's odds of receiving an inotrope if the same patient moved to another randomly chosen clinician or hospital.
- Adjusted versus unadjusted variance shares: after accounting for case mix, the clinician and hospital shares grew, which is why adjustment matters for interpretation.
- Low cardiac output syndrome: the postoperative state of inadequate cardiac output that inotropes are meant to treat, used to judge whether the patient-level predictors make clinical sense.
Critical Analysis
Strengths
- A clear question and a pre-specified design. The hypothesis (more than 5% of variance above the patient level), outcome definition and statistical plan were set out in a protocol approved and made public before analysis, and the paper follows a recognized reporting guideline.
- Scale and data quality checks. More than 51,000 cases from 29 hospitals, with a manual audit of the outcome at every site and a missing-data analysis, give the variance estimates more weight than a single-center review could.
- Method suited to the question. Variance partitioning with nested models and the median odds ratio translate abstract percentages into a concrete statement: the same patient's odds differ by roughly three-and-a-half-fold between two random hospitals.
- Robustness and candor. Several sensitivity analyses (a broader and several narrower outcome definitions, different procedure coding) point the same way, and the discussion states what the design cannot show, namely whether the variation is harmful.
Limitations
- No outcomes. The study deliberately does not relate inotrope use to patient outcomes, so it cannot say which practice is better or which variation is unwarranted.
- Missing clinical detail. Ejection fraction, right ventricular function, cardiac index and surgeon identity were not routinely available, so part of the "institution" or "clinician" effect may be unmeasured patient or surgical complexity.
- Sample composition. Most cases came from academic centers, all in the United States; only five community hospitals contribute, and the medical-school finding rests on a small number of non-affiliated hospitals (its confidence interval, 1.39 to 27.8, is very wide).
- Interpretive limits on equity findings. The race and sex associations are reported but cannot be explained with the available data, as the authors acknowledge.
Where It Fits in the Course
This article is the one in the course list that analyzes routinely collected perioperative data from many hospitals with a multilevel model. The cholecystectomy cost study in the same list is also an observational database analysis, but of a drug-route comparison and its costs, while the hip arthroplasty paper is a single-center retrospective cohort of laboratory changes after nerve blocks. Read together, the three show different observational designs and their typical weaknesses: confounding, missing variables and the gap between statistical and clinical significance.
For method, the books in the list are the natural companions: How to Read a Paper is a short guide to appraising each kind of study, Evidence-Based Medicine treats appraisal and application in more depth, and Testing Treatments explains in plain language why fair comparisons are needed. Use them to place this paper: it describes variation but is not a trial and does not test an intervention.
How to Use This Book
This article suits readers who already know basic cardiac anesthesia terms and want to practice reading a multilevel observational study: residents and fellows in anesthesiology, cardiac surgery or critical care, and students in epidemiology or health services research. Familiarity with logistic regression and odds ratios is enough to follow the main text; the mixed-model details can be read more lightly at first.
A practical reading plan: read the abstract and the introduction, then go straight to the results paragraph on nested modeling and the discussion of warranted and unwarranted variation. Next return to the methods and ask three questions: how is the outcome defined, who is excluded, and what covariates are missing? Finally read the limitations and write down what additional data would be needed to say whether the variation harms patients. As an exercise, restate the 22.6%, 6.8% and 70.6% split and the two median odds ratios in plain language for a clinician, and note which statement would change if the unadjusted and adjusted shares were reversed. Use the supplementary material (protocol, sample-size calculation, sensitivity tables) to see how a study of this type is planned.
Edition and Sources
Version covered: the accepted author manuscript of the article as published in Anesthesiology 139(2), 122–141 (2023), freely available through PubMed Central at https://pmc.ncbi.nlm.nih.gov/articles/PMC10524016/ (NIH public access). This guide draws on the full text of that manuscript: abstract, introduction, methods, results, discussion, limitations and conclusions. Tables, figures and supplementary files were not consulted.