Intensive Care Unit Risk Score

A scoring system usually consists of two parts - a score (a number reflecting the severity of the disease) and a probability model (equation indicating the probability of an event, e.g. the death of the patient in hospital). Scoring systems have been used in intensive care medicine for decades and can help to assess the effectiveness of treatment or identify comparable patients for study purposes. Scoring systems that are used in intensive care medicine are for example

• Acute Physiology, Age, Chronic Health Evaluation II (APACHE II)

• Simplified Acute Physiology Score II (SAPS II)

• Multiple Organ Dysfunction Score (MODS)

• Sequential Organ Failure Assessment (SOFA)

• Logistic Organ Dysfunction System (LODS)

• MPM II-Admission (Mortality Probability Models (MPM II)

• Organ Dysfunction and Infection score (ODIN)

• Three-Day Recalibrating ICU Outcomes (TRIOS)

• Glasgow coma score (GCS)

• Discharge Readiness Score (DRS) The above-mentioned scoring systems are already being collected regularly in the respective hospital's departments. In a recent study by Badawi et al. it could be shown that scoring systems allow more accurate predictions when calculated continuously. However, due to the patient collectives investigated, these results can only be transferred to other patient groups to a limited extent. Furthermore, only the scoring systems APACHE, SOFA and DRS were analyzed.

Therefore, in the present study, all of the above scoring systems will be calculated continuously (once per minute) using routine data from the digital patient records and optimized by applying machine learning and methods of time series analysis.

On the anesthesiologically managed intensive care units of the respective hospital, there is no campus-wide standard with regard to alarm management. Accordingly, we estimate the rate of alarm fatigue (ignoring alarms due to many false alarms) to be very high. In order to optimize the alarm management, alarms from the patient monitoring devices will be evaluated retrospectively and combined with the data mentioned above to determine, for example, whether more frequent alarms are to be expected for certain types of diseases (e.g. sepsis), or scores (e.g., high APACHE score) and how the alarm limit setting can be optimized.