Skip to main page content
U.S. flag

An official website of the United States government

Dot gov

The .gov means it’s official.
Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

Https

The site is secure.
The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
. 2019 Dec;33(6):973-985.
doi: 10.1007/s10877-019-00277-0. Epub 2019 Feb 14.

Predicting tachycardia as a surrogate for instability in the intensive care unit

Joo Heung Yoon  1   2   3 , Lidan Mu  4 , Lujie Chen  4 , Artur Dubrawski  4 , Marilyn Hravnak  5 , Michael R Pinsky  6 , Gilles Clermont  6
Affiliations

Predicting tachycardia as a surrogate for instability in the intensive care unit

Joo Heung Yoon et al. J Clin Monit Comput. 2019 Dec.

Abstract

Tachycardia is a strong though non-specific marker of cardiovascular stress that proceeds hemodynamic instability. We designed a predictive model of tachycardia using multi-granular intensive care unit (ICU) data by creating a risk score and dynamic trajectory. A subset of clinical and numerical signals were extracted from the Multiparameter Intelligent Monitoring in Intensive Care II database. A tachycardia episode was defined as heart rate ≥ 130/min lasting for ≥ 5 min, with ≥ 10% density. Regularized logistic regression (LR) and random forest (RF) classifiers were trained to create a risk score for upcoming tachycardia. Three different risk score models were compared for tachycardia and control (non-tachycardia) groups. Risk trajectory was generated from time windows moving away at 1 min increments from the tachycardia episode. Trajectories were computed over 3 hours leading up to the episode for three different models. From 2809 subjects, 787 tachycardia episodes and 707 control periods were identified. Patients with tachycardia had increased vasopressor support, longer ICU stay, and increased ICU mortality than controls. In model evaluation, RF was slightly superior to LR, which accuracy ranged from 0.847 to 0.782, with area under the curve from 0.921 to 0.842. Risk trajectory analysis showed average risks for tachycardia group evolved to 0.78 prior to the tachycardia episodes, while control group risks remained < 0.3. Among the three models, the internal control model demonstrated evolving trajectory approximately 75 min before tachycardia episode. Clinically relevant tachycardia episodes can be predicted from vital sign time series using machine learning algorithms.

Keywords: Critical Care; Intensive care unit; Machine learning; Prediction; Tachycardia.

PubMed Disclaimer

Figures

Fig. 1
Fig. 1
Operational definition of target tachycardia episode. a. A pseudocode for selecting target tachycardia episode by operational definition. b Schematic illustration of tachycardia episodes by rate, length, and interval. The time between events less than 30 min were combined to form an episode. c An illustration for the concept of density (‘duty cycle’) with two examples of heart rate time series. Red dotted lines indicate the threshold for tachycardia for each episode, with shaded area on the bottom graph shows the time period satisfying the operational definition of a tachycardia episode. Note while upper graph showed much larger number of episodes, lower graph revealed a single, but much dense episode of continued tachycardia. The subject in the bottom panel eventually expired in the ICU
Fig. 1
Fig. 1
Operational definition of target tachycardia episode. a. A pseudocode for selecting target tachycardia episode by operational definition. b Schematic illustration of tachycardia episodes by rate, length, and interval. The time between events less than 30 min were combined to form an episode. c An illustration for the concept of density (‘duty cycle’) with two examples of heart rate time series. Red dotted lines indicate the threshold for tachycardia for each episode, with shaded area on the bottom graph shows the time period satisfying the operational definition of a tachycardia episode. Note while upper graph showed much larger number of episodes, lower graph revealed a single, but much dense episode of continued tachycardia. The subject in the bottom panel eventually expired in the ICU
Fig. 2
Fig. 2
The three models for tachycardia episodes with corresponding control groups
Fig. 3
Fig. 3
Ten-fold cross validation method for training and test models
Fig. 4
Fig. 4
Clinical relevance of target rate for tachycardia episode. a Selection of rate thresholds for target tachycardia episode. With using heart rate (HR) 110/min, 120/min, and 130/min cut-off, different adverse clinical outcome variables including the use of norepinephrine (%), ICU length of stay (days), and ICU mortality (%). Non-tachycardia comprises control group which had no tachycardia episode during ICU stay (n = 2376). b. Clinical adverse outcomes for tachycardia subjects who met operational definition ‘Tachycardia’ indicates subjects met operational definition of tachycardia during the ICU stay (n = 235). ‘Non-tachycardia analyzed’ subjects are control group without tachycardia episode (rate threshold of HR < 130) during the ICU stay (n = 2572). ‘All mimic2 without tachycardia’ stands for the rest of MIMIC2 patient (n = 39397). When appropriate, mean and the standard error (SEM) was produced with error bars. c Comparison of clinically important abrupt onset of non-sinus tachycardia episodes with other overall tachycardia episodes as well as non-tachycardia MIMIC 2 dataset
Fig. 5
Fig. 5
Comparison of the performance of the algorithm. Random Forest (RF, left plot) and Logistic Regression (LR, right plot) with L1 regularization term were tested with using 10-fold cross-validation method. RF slightly outperformed LR with L1 regularization, with overall higher accuracy and larger area under the curve (AUC)
Fig. 6
Fig. 6
Risk score trajectories for the three models of tachycardia group and control group comparison. a Evolving risks for any tachycardia episode in the future. Number of cases = 787, number of controls = 707. b Evolving risks for the first tachycardia episode in the future. Number of cases = 240, number of controls = 240. c Life score to detect the risk within the same subject prior to the first episode. Number of cases = 235
See this image and copyright information in PMC

Comment in

Similar articles

See all similar articles

Cited by

See all "Cited by" articles

References

    1. Morelli A, DEA, Passariello M. Annual update in intensive care and emergency medicine. Vincent JL, editor. Switzerland: Springer International Publishing; 2015. p 409.
    1. Hamosh P, Cohn JN. Left ventricular function in acute myocardial infarction. J Clin Invest. 1971;50(3):523–533. - PMC - PubMed
    1. Palatini P, et al. High heart rate: a risk factor for cardiovascular death in elderly men. Arch Intern Med. 1999;159(6):585–592. - PubMed
    1. Diaz A, et al. Long-term prognostic value of resting heart rate in patients with suspected or proven coronary artery disease. Eur Heart J. 2005;26(10):967–974. - PubMed
    1. Dyer AR, et al. Heart rate as a prognostic factor for coronary heart disease and mortality: findings in three Chicago epidemiologic studies. Am J Epidemiol. 1980;112(6):736–749. - PubMed

MeSH terms