Analysing Nursing Workload in Intensive Care Unit by Using a Novel Objective Tracking System
Thesis DisciplineMechanical Engineering
Degree GrantorUniversity of Canterbury
Degree NameDoctor of Philosophy
Quantifying nursing workload in Intensive Care Unit (ICU) can help optimize nursing resources, allowing allocation of nurses according to patient demand. This can benefit the ICU patients and nurses through the provision of adequate nursing care, prevention of nurse burnout, and reducing of ICU fixed costs. However, quantifying nursing workload is extremely challenging. Current nursing workload assessment tools, such as the therapeutic interventions scoring system (TISS) and nursing activities score (NAS), are subjective and laborious, requiring experienced nurses and researchers to fill out forms. Therefore, an automatic system that can objectively quantify nursing workload is required. The development of computer imaging and tracking technology offer possible solutions to track nursing activities. This research focuses on developing a novel tracking system that can continuously track bedside nursing interventions and quantify nursing workload. Nursing workload is then compared with patient clinical data to analyse which factors strongly influence patient nursing demands. The first part of this thesis discusses the development of a tracking system, and the second part discusses the correlation between quantified nursing workload and patient clinical conditions. Facial detection, color detection, infrared detection, and local position measurement (LPM) are the 4 possible approaches to continuously track nurse bedside interventions. These 4 approaches are evaluated, and infrared detection was the optimal non-invasive approach that most suited implementation in the Christchurch Hospital ICU environment. A clinical activities tracking system (CATS) was developed to track bedside nursing interventions continuously. The CATS hardware contains a Microsoft Kinect with image and infrared depth sensors, controlled by a portable laptop. The CATS software was designed under Microsoft Express C++ environment, implementing OpenNI and OpenCV libraries. The Kinect sensor is placed onto the ceiling to monitor a defined detection area. When an object enters the detection area, it is converted into an unidentifiable unrecognizable blob to protect privacy, and its location over time is recorded. CATS was tested in an experimental environment using two metrics, distance and dwell time to quantify nurse-patient interaction. CATS was then implemented in the Christchurch Hospital ICU. A trained ICU researcher performed manual observations on nursing workload at the bedside and compared with data collected using CATS. The researcher calculated the direct nursing intervention time for each observed hour for 30 hours. The observed direct nursing intervention time was then compared to CATS recorded nursing intervention time. It was found that the CATS recorded nursing intervention is highly correlated with manual observed intervention, and thus CATS was able to record nursing intervention objectively. A preliminary study shows that nursing intervention density is higher during the day compared to night time. Clinical trials include all patients admitted and allocated to a monitored bed between 04/08/2014 to 03/05/2015. 23 patients, with a total of 104 patient days, were recorded. Patient demographics, various patient acuity assessment scores, and workload assessment scores, including APACHE-II, APACHE-III, SAPS-II, SOFA, TISS-28, and NEMS, were calculated and compared with nursing intervention recorded by CATS. The patient’s sedation level, as quantified by GCS, RASS, and sedation drug dose, were also assessed and compared with the nursing intervention recorded by CATS. In this study, APACHE-III and SAPS-II were found to have better resolution in describing patient acuity compared to APACHE-II and SOFA. Both TISS-28 and NEMS display poor sensitivity to different patient specific nursing demands because only 36% of TISS-28 score varies from patient to patient. Equally, no significant trend was found between nursing intervention and sedative dose or sedation level assessed by GCS or RASS. Results showed that the nursing intervention is highly patient-specific and conventional generalised approaches were not able to capture the specificity. CATS was able to capture specificity automatically and objectively. Overall, the objective nursing intervention tracking system provides an objective approach to automatically quantify nursing intervention. This system is validated in clinical trials, indicating its high accuracy and robustness. Nursing intervention captured by CATS shows that during the day nursing intensity is higher than at night time. In addition, none of patient sedation level, acuity level, TISS-28, NEMS, age, length of stay, admission type, or intubation condition shows a strong clinical correlation with nursing time. The difficulty of quantifying nursing intervention using conventional scores revealed a need for an objective system to evaluate nursing workload. At this stage, it is almost impossible to link nursing invention assessed directly by CATS to any existing assessment system based on tasks, patient severity or similar clinical data. As a result, CATS has the potential to standardise nursing workload quantification objectively, and is much less invasive and labor intensive than current assessment systems and scoring based approaches.