Causal determinants of postoperative length of stay in cardiac surgery using causal graphical learning

Abstract

Objective: We aimed to learn the causal determinants of postoperative length of stay in cardiac surgery patients undergoing isolated coronary artery bypass grafting or aortic valve replacement surgery.

Methods: For patients undergoing isolated coronary artery bypass grafting or isolated aortic valve replacement surgeries between 2011 and 2016, we used causal graphical modeling on electronic health record data. The Fast Causal Inference (FCI) algorithm from the Tetrad software was used on data to estimate a Partial Ancestral Graph (PAG) depicting direct and indirect causes of postoperative length of stay, given background clinical knowledge. Then, we used the latent variable intervention-calculus when the directed acyclic graph is absent (LV-IDA) algorithm to estimate strengths of causal effects of interest. Finally, we ran a linear regression for postoperative length of stay to contrast statistical associations with what was learned by our causal analysis.

Results: In our cohort of 2610 patients, the mean postoperative length of stay was 219 hours compared with the Society of Thoracic Surgeons 2016 national mean postoperative length of stay of approximately 168 hours. Most variables that clinicians believe to be predictors of postoperative length of stay were found to be causes, but some were direct (eg, age, diabetes, hematocrit, total operating time, and postoperative complications), and others were indirect (including gender, race, and operating surgeon). The strongest average causal effects on postoperative length of stay were exhibited by preoperative dialysis (209 hours); neuro-, pulmonary-, and infection-related postoperative complications (315 hours, 89 hours, and 131 hours, respectively); reintubation (61 hours); extubation in operating room (-47 hours); and total operating room duration (48 hours). Linear regression coefficients diverged from causal effects in magnitude (eg, dialysis) and direction (eg, crossclamp time).

Conclusions: By using retrospective electronic health record data and background clinical knowledge, causal graphical modeling retrieved direct and indirect causes of postoperative length of stay and their relative strengths. These insights will be useful in designing clinical protocols and targeting improvements in patient management.

Keywords: aortic valve replacement; causal graph learning; causality; coronary artery bypass grafting; postoperative length of stay.

Central Picture

Causal graph learning uncovers insights that can reduce postoperative length of stay (pLOS).

Figure 1:

Causal Graphical Learning improves upon traditional associational analyses of EHR data by uncovering direct, indirect, and confounded relationships among various clinical factors. The FCI algorithm uses EHR data and clinical knowledge to learn a causal graph of all variables, including postoperative length of stay. The LV-IDA algorithm estimates average causal effects on the learned causal graph. Top determinants from our method are found to be distinct from associative ones. Targeted interventions to address causal determinants can improve pLOS in cardiac surgery.

Figure 2:

Adjacency matrix for the PAG obtained from FCI. Blue cells represent direct edges (indicating a direct causal relationship from one variable to another), red cells represent bidirected edges (indicating unobserved confounding), purple cells indicate ambiguity as to whether a direct or bidirected edge exists, and gray indicates no edge.

Figure 3:

(a) depicts extubation (extOR) and reintubation (reint) mechanisms. (b) depicts intra-(iBld) and post-operative (pBld) blood product mechanisms.

Figure 4:

Relationship between cross-clamp duration (xcTm), perfusion duration (pfTm), total operating time (ORTm) and pLOS.

Figure 5:

Select gender (gen) pathways to pLOS via total OR duration (ORTm) and intra-operative blood products (iBld). The asterisks indicate that those effects are not identified in at least one causal explanation consistent with data. The causal effect value provided, however, is the minimum causal effect amongst all explanations where that effect is identified.