Deciphering the explanatory potential of blood pressure variables on post-operative length of stay through hierarchical clustering: A retrospective monocentric study

Objective: Mean arterial pressure is widely used as the variable to
monitor during anesthesia. But there are many other variables proposed to
define intraoperative arterial hypotension. The goal of the present study
was to search arterial pressure variables linked with prolonged
postoperative length of stay (pLOS). Design: Retrospective cohort study of
adult patients having received general  for a scheduled non
cardiac surgical procedure between 15th July 2017 and 31st December 2019.
Methods: pLOS was defined as a stay longer than the median (main outcome),
adjusted for surgery type and duration. 330 arterial pressure variables
were analyzed and organized through a clustering approach. An unsupervised
hierarchical aggregation method for optimal cluster determination,
employing Kendall’s tau coefficients and a penalized Bayes information
criterion was used. Variables were ranked using the absolute standardized
mean distance (aSMD) to measure their effect on pLOS. Finally, after
multivariate independence analysis, the number of variables was reduced to
three. Results: Our study examined 9,516 patients. When LOS is defined as
strictly greater than the median, 34% of patients experienced pLOS. Key
arterial pressure variables linked with this definition of pLOS included
the difference between the highest and lowest pulse pressure values
computed throughout the surgery (aSMD[95%CI] =0.39[0.31-0.40],
p<0.001), the accumulated time pulse pressure above 61mmHg (aSMD =
0.21[0.17-0.25], p<0.001), and the lowest MAP during surgery (aSMD=
0.20[0.16-0.24], p<0.001). Conclusions: By applying a clustering
approach, three arterial pressure variables were associated with pLOS.
This scalable method can be applied to various dichotomized outcomes.