Markov models to detect the deterioration of patients in a pediatric intensive care unit: analytical observational study Ledys M. Izquierdo MD, MSc, Luis F. Niño PhD Jhon S. Rojas (Preprint)

BACKGROUND In the field of continuous vital-sign monitoring in critical care settings, it has been observed that the “early-warning signs” of impending physiological deterioration can fail to be detected timely and sometimes by resource constrained clinical staff. OBJECTIVE to develop a probabilistic model to detect the deterioration of patients in a pediatric intensive care unit. METHODS cross-sectional cohort study, pediatric intensive care unit of the Central Military Hospital in the city of Bogota, Colombia. Children from 1 to 18 years old from January 2018 to January 2020. The CRISP-DM (CRoss-Industry Standard Process for Data Mining) methodology was used as a data mining process and then we used Markov chains to identify the clinical states through which the patient passes. Then, a Hidden Markov model (HMM) based approach is applied for classification and prediction of patient's deterioration by computing the probability of future clinical states. RESULTS Both learning models were trained and evaluated using six vital signs data from 94,678 patient records, collected from the database of real patients who were in the Pediatric Intensive Care Unit of the Central Military Hospital in the city of Bogota, Colombia. To obtain the HMM based classification model, 10-fold cross validation was performed. the confusion matrix showed, Accuracy :0,7, precision: 0.75 and the F1 score:0.65. CONCLUSIONS classification analysis in medical applications can be very useful if considered as a very significant support tool for health professionals. CLINICALTRIAL does not apply