Do heart and respiratory rate variability improve prediction of extubation outcomes in critically ill patients?
1 Ottawa Hospital Research Institute, 725 Parkdale Avenue, Ottawa, ON K1Y 4E9, Canada
2 University of Ottawa, 75 Laurier Avenue East, Ottawa, ON K1N 6N5, Canada
3 University Hospital Network, University of Toronto, 190 Elizabeth Street, Toronto, ON M5G 2C4, Canada
4 Intermountain Medical Center (IMC), Shock Trauma ICU, 5121 Cottonwood Street, Murray, UT 84157, USA
5 Mt Sinai, University of Toronto, 600 University Avenue, Toronto, ON M5G 1X5, Canada
6 London Health Sciences Center, 339 Windermere Road, London, ON N6G 2V4, Canada
7 Sunnybrook Hospital, University of Toronto, 2075 Bayview Avenue, Toronto, ON M4N 3M5, Canada
8 University Hospital Case Medical Center, Case Western Reserve University, 11100 Euclid Avenue, Cleveland, OH 44106, USA
9 University of Oxford, Kellogg College, Banbury Road, Oxford OX2 6PN United Kingdom
10 St. Michaels Hospital, University of Toronto, 30 Bond Street, Toronto, ON M5B 1W8, Canada
11 Divisions of Thoracic Surgery & Critical Care Medicine, 501 Smyth Road, Ottawa, ON K1H 8L6, Canada
Critical Care 2014, 18:R65 doi:10.1186/cc13822Published: 8 April 2014
Prolonged ventilation and failed extubation are associated with increased harm and cost. The added value of heart and respiratory rate variability (HRV and RRV) during spontaneous breathing trials (SBTs) to predict extubation failure remains unknown.
We enrolled 721 patients in a multicenter (12 sites), prospective, observational study, evaluating clinical estimates of risk of extubation failure, physiologic measures recorded during SBTs, HRV and RRV recorded before and during the last SBT prior to extubation, and extubation outcomes. We excluded 287 patients because of protocol or technical violations, or poor data quality. Measures of variability (97 HRV, 82 RRV) were calculated from electrocardiogram and capnography waveforms followed by automated cleaning and variability analysis using Continuous Individualized Multiorgan Variability Analysis (CIMVA™) software. Repeated randomized subsampling with training, validation, and testing were used to derive and compare predictive models.
Of 434 patients with high-quality data, 51 (12%) failed extubation. Two HRV and eight RRV measures showed statistically significant association with extubation failure (P <0.0041, 5% false discovery rate). An ensemble average of five univariate logistic regression models using RRV during SBT, yielding a probability of extubation failure (called WAVE score), demonstrated optimal predictive capacity. With repeated random subsampling and testing, the model showed mean receiver operating characteristic area under the curve (ROC AUC) of 0.69, higher than heart rate (0.51), rapid shallow breathing index (RBSI; 0.61) and respiratory rate (0.63). After deriving a WAVE model based on all data, training-set performance demonstrated that the model increased its predictive power when applied to patients conventionally considered high risk: a WAVE score >0.5 in patients with RSBI >105 and perceived high risk of failure yielded a fold increase in risk of extubation failure of 3.0 (95% confidence interval (CI) 1.2 to 5.2) and 3.5 (95% CI 1.9 to 5.4), respectively.
Altered HRV and RRV (during the SBT prior to extubation) are significantly associated with extubation failure. A predictive model using RRV during the last SBT provided optimal accuracy of prediction in all patients, with improved accuracy when combined with clinical impression or RSBI. This model requires a validation cohort to evaluate accuracy and generalizability.
ClinicalTrials.gov NCT01237886. Registered 13 October 2010.