Fuzzy logic based multimodality clinical alarm system for intensive care unit

Abstract

In high workload areas such as the Intensive Care Units (ICU), clinicians are burdened with too many alarms and false alarms, leading to alarm fatigue that causes poor user response or no response. The alarm fatigue issue has become a top patient safety hazard issue in healthcare institution. The current clinical alarms design lacks critical information, difficult to identify, distinguish, create confusions and lacks human factor engineering (HFE) principles. As such in this research, HFE principles and fuzzy logic techniques are identified to reduce user-related hazards and false alarm to improve clinicians’ responses. Fuzzy logic techniques are the most frequently used Artificial Intelligence techniques to monitor the patient physiological condition, which requires a clinical decision based on vital sign information. Fuzzy logic has benefits over other algorithmic approaches, as it has the potential to incorporate values from ordinal, nominal and continuous datasets within its rules. The vital sign information to determine patients’ risk or alarm conditions has been established, machine learning-based algorithms is unnecessary to train to classify patient risk . Fuzzy logic techniques were identified to diagnose alarm conditions and classify vital sign alarm signals based on risk. This research aims to develop a multimodality clinical alarm software based on HFE principles and artificial intelligence to improve user response and performance of alarm systems. Observation study, focus group, task analysis, simulation testing, and root cause analysis were conducted to identify the root cause of alarm hazards in ICU. The information such as patient condition, device condition, risk-based alarm classification and urgency mapping were identified to mitigate this alarm hazard. Based on this, a new earcon-based multimodality alarms were developed for technical and clinical alarm. A pilot study was conducted using a newly developed alarm simulator to test, verify and validate the new multimodality alarm. The findings identified 1250-1500Hz earcons and 2750-3000Hz earcons to represent medium and high priority clinical alarms respectively. Whereas a combination of sequence of earcons, 525Hz and 550Hz to represent technical alarms. The interburst interval of the alarm waveform (tb) of 5.0sec and 0.5sec for medium and high risk urgency mapping were identified. In this research, four vital sign alarms in patient monitor, Heart Rate (HR), Respiration (RESP), Noninvasive blood pressure (NIBP) and Oxygen Saturation (SPO2) were identified and developed as fuzzy logic based multimodality alarms. Fuzzy logic techniques were used classify these alarms as high risk, medium risk, and normal conditions. These alarms were tested using MIMIC II real patient data, compared and validated with the medical professional evaluation. The results indicated that the sensitivity and specificity of blood pressure and heart rate alarm algorithms are 100.00%. The sensitivity and specificity for respiratory alarm algorithm are at 97.59% and 99.68%, whereas sensitivity and specificity for oxygen saturation at 100.00% and 98.04% respectively. The research concluded that incorporating alarm information based on risk, human factor engineering principles, and fuzzy logic in the alarm system significantly reduce the number of false alarms, improves user response, reduce alarm hazards and improve patient safety in ICU.