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

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.

Journal of Clinical Monitoring and Computing end of year summary 2019: hemodynamic monitoring and management

“…In line with the aspects of the review, the editorial by Vistisen et al [45] accompanying the paper by Yoon et al [40] also highlighted the need for sensible methodology and clear reporting of studies applying machine learning to physiologic data. Yoon et al [40] developed a model to predict tachycardia based on 1-min trending values of vital signs such as heart rate, blood pressure, respiratory rate, and spectral features of these. Based on 787 episodes of tachycardia (cases) and 705 control periods without tachycardia (non-cases), the authors predicted tachycardia with an AUC ROC of 0.81 with their developed algorithm.…”

“…Last year, we saw five original papers applying various types of machine learning techniques/algorithms to clinical data. All studies had a retrospective design and authors predominantly tried to predict hemodynamic events/derangements such as tachycardia [40], hypotension [41] and cardiac arrest [42], but we also saw one before/after implementation study [43] and another study trying to identify patterns between clinical practice and outcomes advised by machine learning techniques [44]. For one of the original papers [40], an accompanying editorial was published highlighting important aspects of the reporting of such papers [45].…”

“…All studies had a retrospective design and authors predominantly tried to predict hemodynamic events/derangements such as tachycardia [40], hypotension [41] and cardiac arrest [42], but we also saw one before/after implementation study [43] and another study trying to identify patterns between clinical practice and outcomes advised by machine learning techniques [44]. For one of the original papers [40], an accompanying editorial was published highlighting important aspects of the reporting of such papers [45]. Finally, general and future aspects of applying machine learning to continuously monitored physiological data was highlighted in an excellent narrative review [46], where Rush et al stated in their introduction that "Machine learning is a term likely spoken of more than understood.…”

“…The editorial discussed how to identify the data sets for cases and noncases, a selection that is fundamentally acausal if the existence of an event is used to define the data set, i.e. if analyzing only selected temporal windows preceding events and non-events, which Yoon et al [40] did. In turn, one would expect a dramatically different performance of such an algorithm, once set free in a prospective validation study, likely associated with an unacceptably high false positive rate, i.e.…”

Artificial Intelligence in the Intensive Care Unit

“…Yoon et al [27] developed a method to predict instability in the ICU based on logistic regression and random forest models of electrocardiogram (EKG) measures of tachycardia, reporting an accuracy of 0.81 and AUC of 0.87. The publication of the study is accompanied by an excellent and highly recommended editorial by Vistisen et al [28] that thoroughly analyzes the strengths and pitfalls of machine learning methods as predictors of complications in the ICU.…”