Following approval of the study protocol by the Institutional Review Boards of two large academic tertiary hospitals— both with active transplant programs and surgical volumes in excess of 40,000 cases annually— electronically recorded pulse oximetry data were obtained from the anesthesia information management systems at both centres utilizing a structured query language. Data were collected over a three-year period (Hospital A: April 2006 to April 2009; Hospital B: November 2005 to January 2009) for each patient who underwent general anesthesia and for whom a completed electronic anesthesia record was available.
For each patient, the database query returned all pulse oximetry values, patient demographics (age, sex, American Society of Anesthesiologists' [ASA] physical status classification), surgical procedure, and case duration. Pediatric patients (age < 16 yr), patients undergoing cardiopulmonary bypass, and procedures performed outside of the operating room (e.g., magnetic resonance imaging, endos-copy, radiology) were excluded from the study. Case milestones (induction of anesthesia, start of surgery, end of surgery, patient departure from the operating room) were also obtained from the electronic anesthesia databases in order to separate pulse oximetry signals into discrete case phases.
Current generation pulse oximeters (Hospital A: Radical 7 - Masimo, Irvine, CA, USA; Hospital B: Nellcor - Boulder, CO, USA) utilizing conventional red and infrared photoplethysmography, digital signal processing, and adaptive filtration were used during the study period. The pulse oximeters were configured to provide either an eight-second (Hospital A) or 20-sec (Hospital B) average of oxygenation (SpO2) values. This information was polled by the information management systems either every 60 sec (Hospital A) or every 30 sec (Hospital B). All values recorded by the information management systems were included in the analysis.
We calculated the number and duration of episodes of both hypoxemia (SpO2
< 90) and severe hypoxemia (SpO2
≤ 85) for each patient. We chose these definitions on the basis of the accepted definition of hypoxemia (SpO2
< 90 that correlates with a PaO2
of < 60 mmHg) and previous studies that have defined severe hypoxemia as SpO2
Episodes of hypoxemia and severe hypoxemia were then categorized according to maximum duration per patient and according to the phase of intraoperative care during which they occurred: induction of anesthesia to start of surgery (induction), start of surgery to end of surgery (surgery), or end of surgery to departure from the operating room (emergence).
When calculating episode durations, we categorized hypoxemic events as lasting one, two, three, four, five, or ≥ six minutes. We treated each individual value recorded in the anesthesia information management systems as persisting until replaced by a new value. Thus, at Hospital A, two consecutive abnormal values sampled every 60 sec represented a two-minute hypoxemic episode. Three consecutive abnormal readings represented a three-minute hypoxemic event. At Hospital B, four consecutive abnormal values sampled every 30 sec represented a two-minute hypoxemic episode, while six consecutive abnormal values represented a three-minute hypoxemic event. In order to allow comparison of data between Hospital A and Hospital B, episodes at Hospital B that consisted of an odd number of values were divided evenly into adjacent bins (e.g., the group of patients with three consecutive abnormal values representing 1.5 min was split evenly into either the one-minute bin or the two-minute bin). In order to understand the effect of our treatment of the odd number of consecutive readings at Hospital B, a sensitivity analysis was performed by modelling the effects of downscaling data recorded at 30-sec intervals into data recorded at one-minute intervals using a Weibull distribution.
For each episode of hypoxemia and severe hypoxemia that lasted two minutes or longer, we then calculated the average number of hours between instances. Our goal was to report how often providers encounter hypoxemia in routine practice. We performed this calculation only on episodes that lasted two minutes or longer, because we wished to increase our specificity for detecting episodes that were truly reflective of hypoxemia by including episodes where multiple abnormal SpO2 recordings occurred sequentially—and excluding non-consecutive readings that might be more reflective of artefact.
Mean age, sex, ASA physical status classification, number of emergency cases, and average surgical times for the entire sample were also calculated. To validate the accuracy of the electronic data query, a random sample of print-outs of anesthesia information management system records from 100 patients at each hospital was manually compared with the information returned from the data query.