January, 3rd, 2023
The COmmunicating Narrative Concerns Entered by Registered Nurses (CONCERN) system is a tool that uses machine learning and natural language processing to extract and analyze nursing documentation patterns from electronic health records (EHRs) in order to predict when patients are at risk of clinical deterioration.
It is currently being tested in a clinical trial at Columbia University Medical Center, as well as three other hospital systems: Mass General Brigham, Vanderbilt University Medical Center, and Washington University School of Medicine/Barnes-Jewish Hospital. The CONCERN implementation toolkit, developed to support large-scale adoption of the tool, is also being tested for effectiveness. The initiative recently received funding from the American Nurses Foundation through the Reimagining Nursing Initiative. The goal of CONCERN is to improve patient outcomes by utilizing the expertise of nurses and combining it with AI technology.
Why It Matters?
It is important to have a predictive early warning score for patients at risk of detection in the hospital because it allows healthcare professionals to identify patients who may be at risk for deterioration earlier, potentially allowing for earlier interventions and improved patient outcomes. Additionally, having a tool that can extract and analyze data from nurses’ assessments and observations can help to better utilize their expertise and knowledge, and can help to engage the care team when nurses feel that patients are at risk. In the current COVID-19 pandemic, having a tool like CONCERN that can provide clinical decision support for patients at risk of deterioration is especially important due to the increased severity of illness and decreased resources. Utilizing this toolkit to implement CONCERN in multiple hospital systems can also help to scale its success and potentially improve patient outcomes on a larger scale.
How CONCERN is using Machine Learning
The COmmunicating Narrative Concerns Entered by Registered Nurses (CONCERN) system is a novel approach to predicting patients at risk of clinical deterioration in hospitals using artificial intelligence (AI) and natural language processing (NLP). The system was developed to convert real-time patient data into consumable information that clinicians can use to identify patients at risk of deterioration and prevent mortality and improve patient outcomes.
The development of the CONCERN AI system was divided into three stages: feature selection and preprocessing, feature modeling, and assignment of colors and postprocessing. The initial features were selected from prior research and included vital signs and vital sign comments frequency. These were combined with additional features selected by experts, including PRN medications administered, medications withheld, frequency of nursing notes written, and nursing note content, as well as the times these actions were performed. These features were aggregated over the past 12 hours and combined using machine learning techniques (NLP, decision trees, and logistic regression) with outcomes such as rapid response, cardiac arrest, sepsis, unanticipated intensive care unit transfer, and death.
A logistic regression-based model was chosen for implementation because focus groups with clinicians indicated that model explainability was important. The weights derived from the machine learning techniques were used to combine the features into a single score that reflected clinical concern for patient deterioration. The score was then color-coded to represent the degree of risk for deterioration.
The color-coded CONCERN score is intended to be used as a tool for clinical decision support within electronic health record systems. When a patient’s score changes from green to yellow or red, it serves as a warning for clinicians to pay closer attention to the patient and possibly initiate interventions to prevent further deterioration. The CONCERN system was developed through a participatory design process involving user-centered design sessions, focus group interviews, and simulation testing sessions with nurses and physicians. This approach ensured that the final product was both clinically relevant and user-friendly.
In a cluster randomized pragmatic clinical trial, the CONCERN system was implemented and evaluated at two different study sites. The trial used a multiple time-series intervention consisting of three phases: pilot and trial testing, activation of optimized versions of the CONCERN CDS, and a silent release mode where the CDS was not viewable to the end user. The mixed methods approach used for the evaluation included assessments of the system and clinician perspectives.
Preliminary results have shown that the CONCERN system is effective at identifying patients at risk of clinical deterioration. In a previous study, the system identified at-risk patients 5-24 hours earlier than other early warning scores. Further research is needed to determine the long-term impact of the CONCERN system on patient outcomes such as in-hospital mortality, length of stay, and hospital readmissions. However, the CONCERN system has the potential to significantly improve patient care by providing clinical decision support and enabling timely interventions to prevent deterioration and improve patient health outcomes.
Overall
The CONCERN CDS system is a promising approach to improving patient outcomes by utilizing the expertise of nurses and combining it with AI technology. If the clinical trial is successful, the CONCERN CDS system has the potential to be implemented in hospitals and other healthcare settings, helping to identify patients at risk of deterioration and preventing mortality and harm.
