Modeling techniques for predicting mortality in intensive care units

A medical modeling method for predicting a risk of a physiological condition for a patient, said medical modeling method comprising:

• receiving measurements of a plurality of predictive variables for the patient, the plurality of predictive variables predictive of the risk of the physiological condition;
• calculating the risk of the physiological condition by applying the received measurements to at least one model modeling the risk of the physiological condition using the plurality of predictive variables, the calculating including:
• receiving a time series of measurements for a predictive variable of the plurality of predictive variables;
• calculating a predicted risk of the physiological condition by applying the received time series of measurements to a hidden Markov model modeling the risk of the physiological condition;
• and
• calculating the risk of the physiological condition by applying the received measurements and the calculated predicted risk to a logistic regression model modeling the risk of the physiological condition;
• outputting an indication of the risk of the physiological condition to a clinician;
• training at least one of the hidden Markov model and the logistic regression model based on a training population before calculating the risk of the physiological condition;
• retraining the at least one of the hidden Markov model and the logistic regression model based on the received measurements and an outcome describing whether the physiological condition occurred in the patient,
• calculating the risk of the physiological condition for another patient using the retrained model;
• and
• with a clinical decision system:
• receive the calculated risk of the physiological condition;
• provide a clinician with clinical decision support based on the calculated risk of the physiological condition;
• monitoring the calculated risk of the physiological condition for an alarm condition;
• and
• in response to the alarm condition, controlling an alarm output unit to output an alarm indicative of the physiological condition to the clinician.

A medical system comprising:

• a medical modeling system for predicting a risk of a physiological condition for a patient, the medical modeling system configured to:
• receive measurements of a plurality of predictive variables for the patient, the plurality of predictive variables predictive of the risk of the physiological condition;
• and
• train at least one of a hidden Markov model and a logistic regression model based on a training population;
• calculate the risk of the physiological condition by applying the received measurements to at least one model modeling the risk of the physiological condition using the plurality of predictive variables, the at least one model including at least one of the hidden Markov model and a logistic regression model, the calculating including:
• receive a time series of measurements for a predictive variable of the plurality of predictive variables:
• calculating a predicted risk of the physiological condition by applying the received measurements to the hidden Markov model modeling the risk of the physiological condition;
• and
• calculating the risk of the physiological condition by applying the received measurements and the calculated predicted risk to the logistic regression model modeling the risk of the physiological condition;
• retrain the model based on the received measurements and an outcome describing whether the physiological condition occurred in the patient;
• and
• calculate the risk of the physiological condition for another patient using the retrained model;
• and
• a clinical decision system configured to:
• receive the calculated risk of the physiological condition;
• provide a clinician with clinical decision support based on the calculated risk of the physiological condition;
• monitor the calculated risk of the physiological condition for an alarm condition;
• and
• in response to the alarm condition, control an alarm output unit to output an alarm indicative of the physiological condition to the clinician.