Project update SNOWDROP: preventing fall-related injuries for senior citizens

Fall-related injuries are a serious problem: among senior citizens in the Netherlands, it is the number one reason for visiting the ER (100,000 cases per year). Medication is an important risk-factor for fall incidents. The SNOWDROP project aims to develop and implement a prediction model that supports joint medication management by senior citizens and general practicioners (GPs). Since the kick-off in August 2019, there have been some exciting developments. Time to sit down with researchers Leonie Westerbeek and Noman Dormosh for an update!

Two project goals

Noman Dormosh works on the medical informatics side of the project. SNOWDROP has two main goals, he explains. “The first goal is to use big data to identify senior citizens at high risk of falling by means of a prediction model. That is mostly my area of expertise. The second goal is to implement this model and evaluate it in real practice, which is where Leonie comes in.”

Leonie Westerbeek is a communication scientist. “We involve GPs and senior citizens in de the development of the tool that will use the prediction model”, she adds. “I’m currently conducting qualitative research by means of focus groups and interviews. The aim is to have the new tool fit optimally in existing workflows and to find effective ways to communicate with patients.”

Multidisciplinary collaboration

Both Noman and Leonie are enthusiastic about the multidisciplinary approach of the project, as it combines expertise from communication science, medical informatics, geriatrics and general practice. Next to the academic collaborations, three other parties are involved. Leonie: “We work with Pharmeon Pharmeonfor implementation in their patient portal and similarly with ExpertDoc, the developers of a decision support system that is already in use by GPs. Elsevier is also involved.”

“We work with two types of data”, Noman clarifies: “Structured data like age, gender and medication is relatively easy to manage. The other data type is free text input by physicians, that contains large amount of information, including (among others) previous fall incidents. We work together with researchers from Elsevier to extract relevant information from the free text that can be used in the development of the prediction model. We believe this approach can be adapted to other projects as well.”

Data challenges

Noman elaborates on challenges on the data side: “Pseudonymization of free text is difficult. Additionally, we needed proper outcome labels (‘did fall’ or ‘did not fall’) to train the predictive algorithm which requires a detailed look at the free text – I had to manually mark more than 10,000 records!”

For the future, the SNOWDROP team is looking into the possibilities of using deep learning technology (such as neural networks) to extract information from more free text entries, as this may help improve the accuracy of the prediction model. There are facilities available (Noman mentions the Digital Research Environment) that have the necessary computing power, but the team is still looking for ways to transfer data to those facilities in a way that is compliant with GDPR and other regulations.

Current state and future ambitions

Currently, the prediction model is finished and the collection of qualitative data from GPs and senior citizens has been completed. Together with the partners in this project, the researchers are now working on the input for the patient portal and the GP decision support system, as well as implementing the prediction model within the associated software. Usability testing by GPs and senior citizens is scheduled for the second half of 2021. Proper trails will start in 2022.

And after that? “Ideally, this system would be used by all GPs so they can timely invite their patients and review their medication when they are at high risk. That would require implementation in other decision support systems, next to the one we are working with now,” says Leonie. Noman adds: “Further validation of the model is on the agenda as well: we will research the generalizability of the model by validating it using data sets from a different source.”

While her qualitative research on the implementation side has not exactly been hassle-free due to COVID-19 restrictions, Leonie is enthusiastic about the project: “I really believe in the preventive nature of our approach. The citizens we target with this tool are often not even aware they are at high risk, so there’s a world to win.” Noman agrees: “Preventing fall-related injuries is increasingly important in an aging population. It is great that we can look beyond the development of the model and really implement it in everyday practice.”

More information

03 June 2021


Sign up for our newsletter and stay informed of the latest news Commit2data