Article type
Abstract
"Background: In intensive care units (ICUs), there is growing interest in improving nursing-sensitive outcomes (NSOs). Predictive models using artificial intelligence (AI) and machine learning (ML) have been introduced, and nursing data generated by nurses has been identified to help predict NSOs.
Objectives: This study aimed to ascertain the utilization of nursing data within prediction models for NSOs in ICUs employing AI and ML.
Methods: This scoping review systematically searched seven electronic databases in July 2022. The inclusion criteria were: 1) population: adult patients within ICUs; 2) concept: prediction models for NSOs using AI and ML; 3) context: utilization of nursing data from electronic health records. The data extraction encompassed the publication year, geographical location, ICU type, model type, datasets, sample size, calibration, modeling techniques, type of NSOs, and type of nursing data.
Results: Among 9,414 studies, a total of 23 were selected. Studies were published from 2011 to 2022; most were published since 2018 (86.9%) (Figure 1). The studies were primarily conducted in Asia (60.9%) and North America (30.4%), with combined ICUs being the most common setting (78.3%). More than half (60.9%) used prediction models that only demonstrated internal validity. Most used hospital (64.0%) or public data (28%), such as MIMIC, eICU, and PhysioNet, to develop prediction models. The median sample size of the development model was 2,493 (902–5,175). Calibration for model performance was only approximately half (52.2%). Regarding modeling methods, regression analysis was used the most (82.6%), followed by decision trees (56.5%), neural networks (26.1%), and support vector machines (21.7%). Regarding nursing-sensitive outcomes, pressure injury (43.5%) was most frequently identified, succeeded by delirium (21.7%) and functional status (17.4%). Regarding nursing data in the prediction model, nursing scale was used in most (82.6%), followed by nursing assessment (39.1%) and nursing activity (30.4%); nursing notes were not used (Table 1).
Conclusions: Our findings demonstrate limited use of nursing data in models predicting NSOs associated with patient clinical deterioration. Given the potential of nursing data to enhance predictive accuracy and patient outcomes, it is suggested that future research focus on integrating these data more effectively."
Objectives: This study aimed to ascertain the utilization of nursing data within prediction models for NSOs in ICUs employing AI and ML.
Methods: This scoping review systematically searched seven electronic databases in July 2022. The inclusion criteria were: 1) population: adult patients within ICUs; 2) concept: prediction models for NSOs using AI and ML; 3) context: utilization of nursing data from electronic health records. The data extraction encompassed the publication year, geographical location, ICU type, model type, datasets, sample size, calibration, modeling techniques, type of NSOs, and type of nursing data.
Results: Among 9,414 studies, a total of 23 were selected. Studies were published from 2011 to 2022; most were published since 2018 (86.9%) (Figure 1). The studies were primarily conducted in Asia (60.9%) and North America (30.4%), with combined ICUs being the most common setting (78.3%). More than half (60.9%) used prediction models that only demonstrated internal validity. Most used hospital (64.0%) or public data (28%), such as MIMIC, eICU, and PhysioNet, to develop prediction models. The median sample size of the development model was 2,493 (902–5,175). Calibration for model performance was only approximately half (52.2%). Regarding modeling methods, regression analysis was used the most (82.6%), followed by decision trees (56.5%), neural networks (26.1%), and support vector machines (21.7%). Regarding nursing-sensitive outcomes, pressure injury (43.5%) was most frequently identified, succeeded by delirium (21.7%) and functional status (17.4%). Regarding nursing data in the prediction model, nursing scale was used in most (82.6%), followed by nursing assessment (39.1%) and nursing activity (30.4%); nursing notes were not used (Table 1).
Conclusions: Our findings demonstrate limited use of nursing data in models predicting NSOs associated with patient clinical deterioration. Given the potential of nursing data to enhance predictive accuracy and patient outcomes, it is suggested that future research focus on integrating these data more effectively."