Predicting Intensive Care Unit Length of Stay and Mortality Using Patient Vital Signs: Machine Learning Model Development and Validation

Abstract

Background Patient monitoring is vital in all stages of care. In particular, intensive care unit (ICU) patient monitoring has the potential to reduce complications and morbidity, and to increase the quality of care by enabling hospitals to deliver higher-quality, cost-effective patient care, and improve the quality of medical services in the ICU. Objective We here report the development and validation of ICU length of stay and mortality prediction models. The models will be used in an intelligent ICU patient monitoring module of an Intelligent Remote Patient Monitoring (IRPM) framework that monitors the health status of patients, and generates timely alerts, maneuver guidance, or reports when adverse medical conditions are predicted. Methods We utilized the publicly available Medical Information Mart for Intensive Care (MIMIC) database to extract ICU stay data for adult patients to build two prediction models: one for mortality prediction and another for ICU length of stay. For the mortality model, we applied six commonly used machine learning (ML) binary classification algorithms for predicting the discharge status (survived or not). For the length of stay model, we applied the same six ML algorithms for binary classification using the median patient population ICU stay of 2.64 days. For the regression-based classification, we used two ML algorithms for predicting the number of days. We built two variations of each prediction model: one using 12 baseline demographic and vital sign features, and the other based on our proposed quantiles approach, in which we use 21 extra features engineered from the baseline vital sign features, including their modified means, standard deviations, and quantile percentages. Results We could perform predictive modeling with minimal features while maintaining reasonable performance using the quantiles approach. The best accuracy achieved in the mortality model was approximately 89% using the random forest algorithm. The highest accuracy achieved in the length of stay model, based on the population median ICU stay (2.64 days), was approximately 65% using the random forest algorithm. Conclusions The novelty in our approach is that we built models to predict ICU length of stay and mortality with reasonable accuracy based on a combination of ML and the quantiles approach that utilizes only vital signs available from the patient’s profile without the need to use any external features. This approach is based on feature engineering of the vital signs by including their modified means, standard deviations, and quantile percentages of the original features, which provided a richer dataset to achieve better predictive power in our models.

Machine Learning-based Prediction of Prolonged Intensive Care Unit Stay for Critical Patients with Spinal Cord Injury

“…8,39 -41 To maximize the potential of prediction classifiers for LoS-ICU or LoS-hospital, lots of studies [12][13][14]42,43 endeavored to train more classifiers with various ML algorithms and multi-dimensional data. Logistic regression, LDA, RF, KNN, and SVM were five frequently used ML algorithms in the studies [12][13][14]42,43 mentioned above. In the present study, we also adopted other ML algorithms like DNN, MLP, gradient boosting, AdaBoost, bagging, extra tree, decision tree, and NB.…”

“…It turned out that REF and Embed-dingLR were the top two feature-selection methods, as they constituted the top three initial classifiers for both the LoS-ICU and the LoS-hospital. Most available studies [12][13][14]44 included demographic data, vital signs, clinical examinations and laboratory values as potential predictors to develop ML classifiers for predicting LoS-ICU or LoS-hospital. Except for the data mentioned above, only a few studies included treatment information, 43 source information 45 to develop ML classifiers.…”

“…To maximize the potential of prediction classifiers for LoS-ICU or LoS-hospital, lots of studies 12–14,42,43 endeavored to train more classifiers with various ML algorithms and multi-dimensional data. Logistic regression, LDA, RF, KNN, and SVM were five frequently used ML algorithms in the studies 12–14,42,43 mentioned above.…”

Untitled

“…In general, high compliance was observed from the results of classifying and prioritizing variables in reviewed studies with the most common variables in the current study (Table 5). So far, multiple studies have been conducted on the application of ML techniques to predict the LOS in hospitalized patients [11,32,34,[45][46][47]…”

WildWood: a new Random Forest algorithm

“…Related works. Since their introduction [9], RF algorithms have become one of the most popular supervised learning algorithm thanks to their ease of use, robustness to hyperparameters [7,55] and applicability to a wide range of domains, recent examples include bioinformatics [57], genomic data [19], predictive medicine [65,1], intrusion detection [20], astronomy [35], car safety [67], differential privacy [53], COVID-19 [64] among many others. A non-exhaustive list of developments about RF methodology include soft-pruning [12], extremely randomized forests [32], decision forests [24], prediction intervals [60,74,13], ranking [76], nonparametric smoothing [70], variable importance [44,37,45], combination with boosting [33], generalized RF [3], robust forest [41], global refinement [59], online learning [39,50] and results aiming at a better theoretical understanding of RF [6,5,31,2,63,61,62,49,48,75].…”