The long-term objective of the project are to develop a set of stable mortality prediction models (MPMs) for ICU patients that can be used to stratify ICU populations by severity of illness. Models that have high sensitivity and specificity, high total correct classification, and good inter-hospital validity should be useful for inter-institutional QA studies; for use by the ICU team, patients, families, and ethics committees for guiding and monitoring clinical case decisions; and for providing information germane to resource allocation activities.
Specific aims are first to identify conditions apparently associated with patients misclassified by the original MPMs (developed in previous research funded by NCHSR), and then to develop new models, using the same statistical data reduction methods used in the earlier work. The new models will then be applied to ICU patients in three hospitals with the expectation of demonstrating substantially lower misclassification, which in the original MPMs was related to relatively low sensitivity (patients who ar predicted to live but who die in the hospital). Another aim is to demonstrate high inter-hospital validity of the MPM system. In addition, a new 72-hour MPM will be developed. This, together with the MPM0, MPM24, and MPM48, should sharpen substantially the serial probability strategy for predicting individual patient outcome. Software will be developed to facilitate on-site calculations of MPMs. The following results (products) are anticipated: - a new or refined MPM0 (admission) that can be used to stratify admission ICU populations by severity of illness for inter-hospital QA studies - new MPM24, MPM48, and MPM72 that will have substantially lower misclassification (higher sensitivity without serious loss of specificity) and greater use for calculating serial probabilities. - software packages for on-site calculations of MPM probabilities All current ICU predictive models suffer from unacceptably low sensitivity and need to be refined before being applied in evaluative or clinical practice. This research is proposed as a definitive step in refinement.
