JMIR Publications

ABSTRACT

Background:

Automated acquisition of intraoperative patient temperature data by temperature probes leads to the possibility of incurring a number of artifacts related to probe positioning that may impact their utility for observational research.

Objective:

We sought to compare the performance of two de-novo algorithms to filter such artifacts.

Methods:

In this observational retrospective study intraoperative temperature data of adults who received general anesthesia for non-cardiac surgery were extracted from the Multicenter Perioperative Outcomes Group (MPOG) registry. Two algorithms were developed and were then compared to the reference standard of anesthesiologists’ manual artifact detection. Sensitivity and specificity for artifact detection were calculated for each algorithm, as were mean temperatures and Area Under the Curve (AUC) for hypothermia (below 36C) for each patient after artifact removal by each methodology.

Results:

A total of 27,683 temperature readings from 200 anesthetic records were analyzed. The overall agreement among the anesthesiologists was 92.1%. Both the algorithms had high specificity but moderate sensitivity (Specificity- Algorithm 1: 99.02 % vs. Algorithm 2: 99.54%; Sensitivity- Algorithm 1: 49.13% vs. Algorithm 2: 37.72%, F-score- Algorithm 1: 0.65 vs. Algorithm 2: 0.55). The timeX temperature hypothermic Area Under 36.0 degrees and the mean temperature per case after artifact removal were similar between the algorithms and the anesthesiologists.

Conclusions:

The tested algorithms provide an automated way to filter intraoperative temperature artifacts which closely approximate manual sorting by anesthesiologists.Our study provides evidence demonstrating the efficacy of a highly generalizable artifact reduction algorithm that can be readily employed by observational studies that rely on automated intraoperative data acquisition.