Discussion
Successful fall prevention programs include a combination of environmental measures, clinical interventions, care process interventions, cultural interventions, and technological/logistical interventions. Some of these interventions are resource intensive and costly to an organization. Traditional fall risk tools place fall prevention programs at a disadvantage due to inaccurate risk prediction that leads to preventable patient harm and increased nursing workload. These tools lack accuracy and reliability to identify patients most at risk. With inaccurate fall risk assessment tools and a high percentage of patients deemed high-risk, resource-intensive prevention interventions cannot be targeted to the patients that need them most and organizations may resort to applying a "one size fits all" approach to fall prevention.
Traditional fall risk tools place fall prevention programs at a disadvantage due to inaccurate risk prediction…
Despite the assumption that prevention interventions will prevent many falls that would otherwise have occurred, it remains important to reduce over-prediction of risk or false negatives when evaluating the performance of fall risk prediction models. The JHFRAT, as the benchmark tool, is shown to categorize over 50% of a patient population as high fall risk,17 while a typical fall rate stands at approximately 2%. A successful prevention program utilizing this risk assessment tool may be preventing some accurately high-risk patients from falling, but is also likely over-predicting high fall risk, and the resources needed to apply high-risk interventions to 50% of patients is not insignificant.
CDS tools have been introduced in many areas with the goal of improving processes of care. It is important though that CDS tools provide a clinical benefit to clinicians to avoid becoming noise that disrupts workflow. In the case of fall risk prediction tools, comparing the Epic ROF model to the Kinometrix K-FRAS shows important differences. The ROF model ultimately performed only as well as the JHFRAT, while the K-FRAS showed improved sensitivity without an increase in false negatives. This improved predictive accuracy is of great importance when balancing a patient safety program with the resources required to provide the needed interventions.
How a model is developed matters a great deal. Falls occur in a specific point in time during a patient’s hospital stay. Likewise, a patient’s condition changes on a continuum during their stay. If a model, like the Epic ROF, is built by identifying an entire patient stay as positive for a fall event, as well as only looking at key data points around the time of discharge, then inaccuracy should be expected. This limits the benefit of the Epic ROF model to reducing the documentation required for manual fall risk assessments, though Epic does state additional documentation may be required for the tool.12
By using more rigorous and objective patient data, the K-FRAS allows clinical teams to manage risk through specific, actionable fall risk drivers.
The K-FRAS model was developed using best-in-class machine learning techniques focused on identifying patient risk factors specific to the time of a fall event. This allows the model to be dynamic throughout a patient’s hospitalization, adjusting to the changing conditions that occur. In addition, the K-FRAS is a developing system that will continue to improve as it learns from additional cases. With EHR integration, the K-FRAS plugs in wherever clinicians currently view a patient’s fall risk, facilitating adoption and reducing the burden of fall risk assessment documentation. By using more rigorous and objective patient data, the K-FRAS allows clinical teams to manage risk through specific, actionable fall risk drivers. Not only do clinicians receive the information that their patient is a fall risk, but also the specific factors driving the risk so the care plan can be individualized to the patient. The K-FRAS achieves the desirable CDS outcome of not only reducing documentation burden, but also improving clinical processes by more accurately identifying fall risk patients, helping hospitals to direct resource-intensive interventions where they are needed most.
