Artikel
Developing an algorithm to detect falls in the electronic health record: a diagnostic accuracy study
Suche in Medline nach
Autoren
-
Barbara Schärer
- Inselspital Bern University Hospital, Head office of Nursing, Medical-Technical and Medical-Therapeutic Ares, Nursing Development, Switzerland
-
Nicole Grossmann
- Inselspital Bern University Hospital, Department of Internal Medicine
-
Franziska Gratwohl
- Inselspital Bern University Hospital, Department of Internal Medicine
-
Jacques Donzé
- Inselspital Bern University Hospital, Department of Internal Medicine
-
Stefanie Bachnick
- Inselspital Bern University Hospital, Nursing & Midwifery Research Unit, Switzerland
-
Franziska Zuniga
- Institute of Nursing Science, University of Basel, Switzerland
-
Sarah N. Musy
- Inselspital Bern University Hospital, Nursing & Midwifery Research Unit, Switzerland
-
Michael Simon
- Inselspital Bern University Hospital, Nursing & Midwifery Research Unit, Switzerland
| Veröffentlicht: | 30. April 2018 |
|---|
Gliederung
Text
Background and Purpose: Fall events are among the most common adverse events and are linked to undesired outcomes, such as prolonged hospitalization, disability or even death. To reduce falls, it is essential to accurately measure them. Currently used methods (e.g. voluntary incident reporting) suffer from underreporting. Fall detection algorithms for electronic health records (EHRs) may facilitate the task in an efficient and cost effective way.
The objectives were:
- 1.
- to develop an algorithm to identify fall events in the EHRs of a Swiss University Hospital and
- 2.
- to determine the diagnostic accuracy of the algorithm using voluntary incident reporting and the Global Trigger Tool (GTT)
Methods: This retrospective study included a sample of 120 randomly selected patients in a general internal medicine department over 6 months. The algorithm was developed using structured query language (SQL) and text mining approaches. Sensitivity, specificity and predictive values were compared to falls identified with the GTT and voluntary incident reporting.
Results: The patients’ mean age and length of stay were 70 years and 14.6 days, respectively. The algorithm identified 11 fall events, whereas two events were missing in the GTT and seven were missing in the incident reporting.
Conclusions: The newly developed algorithm produced higher sensitivity and precision than GTT and voluntary incident reporting. Further evaluation with a larger sample are needed, with the goal of using it in real-time to monitor fall events in the whole hospital.
