Umetrics Suite Blog

Data analytics could help reduce falls and fear of falling among elderly people

January 7, 2019

Many elderly people are afraid of falling – and for good reasons. Falls can have serious consequences for the individual but also the fear of falling could have serious effects on health and independence. A new research project at Luleå University of Technology in Sweden has taken a closer look at fall-related concerns among elderly people, using multivariate data analysis, MVDA, with the ultimate goal of finding diagnostic and training methods that could help reduce falls. Results from the first studies have given some interesting answers.

HOW MVDA can reduce risk of falling

The fear of falling reduces the quality of life

The prevalence of fall accidents among elderly people in Sweden is very high. Among people who are 70 years and older, there are as many as 5000-7000 fall accidents per 100,000 individuals per year, the prevalence varying somewhat between different regions. The cost for the society of all these falls amount to a staggering SEK 1,1 billion per year.

Falls can have serious consequences for the individual. Fractures can for example lead to hospitalization, but also the fear of falling can have very negative effects on the quality of life. If you are afraid of falling you become less active. This fear of falling has a huge impact on how we feel when we age.

A multivariate approach

A research team of physiotherapists and automatic control researchers at Luleå University of Technology are presently working together to gain more insight into postural control and balance and how it changes as we age – the BAHRT project. PhD student Mascha Pauelsen recently presented parts of the research (see link to recorded webinar below), emphasizing that balance consists of many subsystems that interact, which makes it challenging to analyze.

The standard method has been to look at one or two aspects of balance at a time and do linear regressions. However, as balance is multivariate and interconnected, the research team decided that they had to look at complete systems. They needed a multivariate approach and a means of handling predictors that are correlated.

Interesting findings from the first study

A common concept in this context is falls self-efficacy (FsE). Falls self-efficacy can be explained as the fall-related concerns that elderly people have when doing different activities. A large part of it is explained by physical performance, including balance.

In the first study, the aim was to gain more insight about falls and the fear of falling. 153 elderly people were asked to fill in a comprehensive questionnaire as well as having their physical performance measured. The study showed that as many as 70% of the participants reported some form of fall-related concern.

Physical performance was measured with a clinical instrument called the short physical performance battery, which is a number of clinical tests that measure things like walking speed, how fast you can get up from a chair, etc.


Based on the results, the researchers arrived at a model where falls self-efficacy is explained by three aspects – morale, physical performance and fear. Morale in this context is defined as our outlook on life and how pessimistic or optimistic we are about our situation as we age.

The second study: Building a base model using multivariate data analysis and OPLS

The clinical instrument that was used in the first study measures balance as a total. To make sure it was really the balance aspect that was the main explanatory factor of physical performance, and its effect on falls-self efficacy, the research team had to measure balance in a more exact way. 45 of the participants in the first study were thus invited to measure their balance with a force plate. A force plate measures how much we sway when we try to stand as still as possible. Measurements were taken when participants had their eyes open as well as closed while standing on stable and unstable ground.

The research team then used SIMCA from the Umetrics Suite to analyze the test results. More specifically, they used OPLS, an analytical method in the SIMCA software that has been developed for multivariate data analysis. The advantage of OPLS, compared to regular multivariate linear regressions, is that it can handle predictors that are correlated. Thus, by using the OPLS, the research team could come up with a base model by putting all the different measurements in the x block, without having to worry about correlations between the measurements, and use the variance of the falls-self efficacy from the previous test as the y.

The question was: How well can the balance tests on the force plate explain the variance in falls self-efficacy? The resulting OPLS model could in fact explain 39% of the variance. That might seem like a low number, but for human data it is a very good model, as human behavior is influenced by so many factors – including factors that we may not even know about.

As can be seen in the image below, the measurements where the participants had their eyes open had the strongest impact. This makes sense as the questionnaire that was used to measure falls self-efficacy asked questions about activities when the participants have their eyes open, for example “how worried are you when you walk up or down the stairs?” The limit of stability, how far you can lean in different directions without moving your feet, was also an important explanatory factor.


The multivariate OPLS base model: measurements of SEO (stable eyes open) and UEO (unstable eyes open) as well as limit of stability (AP stab, ML stab) are the strongest explanatory variables of falls-self efficacy.

The third study: Building a top model, again by using the OPLS method

The next step for the research team was to find out if any of the balance subsystems is more important than the other systems. Hence, a new set of tests was conducted with the same 45 participants. One of the balance subsystems is called proprioception, that is, we feel how our body is positioned. We can for example feel how much we bend our knees without looking at them. The other subsystems are eyesight, vestibular system (the balance system in the inner ear) and touch sensation. These systems are called our sensory systems. Our balance is also based on our motor systems, more specifically the strength in the major muscle groups of our lower limbs. Finally, reaction time is important for our balance and postural control.

In the third study, the sensory systems, the motor systems and reaction time were measured. The aim was to see if these measurements were correlated to the base model. As stated before, balance is multivariate and there are correlations between the measurements. Thus, the OPLS method was used again in order to handle both the multivariate aspects and the correlations. This time, the base model from the second study was used as the new y and the measurements from the third test were put into the x block. In this new model, the top model, the variance of the different subsystems could explain the previous base model by 40%.

The image below shows that lower limb strength is the strongest explanatory aspect of the base model. Earlier physiotherapy studies have shown that lower limb strength is very important for our balance. This top model shows that lower limb strength is also very important for falls-self efficacy. Furthermore, the top model shows that proprioception, touch sensation, eyesight and reaction time also play a part in the variance of falls self-efficacy as they have an effect on balance. In fact, these systems compensate for each other. If one system declines, the other systems will make up for it, with no effect on our balance. It is not until this compensation system does not work any longer that there is an effect on balance and on falls self-efficacy.

blog56-chart-2-basemodeul explained

The multivariate OPLS top model: measurements of lower limb strength (to the right) have the strongest load in the model, but also other factors such as reaction time, eyesight and proprioception (the JPS measurements) are important.

The next step

The ultimate goal for the research team is of course to help prevent elderly people from falling. Based on the results so far, the team is currently trying to build a diagnostic system. The idea is to identify which of the sensorimotor systems within postural control that is the major culprit for a specific person and to make individualized, targeted interventions. And not just interventions to improve balance, but also interventions that will influence the person’s level of falls self-efficacy, as falls self-efficacy has a strong effect on our outlook on life and how healthy we feel as we age.

Want to know more?

Watch the recorded webinar with Mascha Pauelsen.

WEbinar screenshot


About the BARTH project

The Balancing Human and Robot project (BAHRT) is a project at Luleå University of Technology which aims at gaining further knowledge about balance, fall risks and fear of falling. The project, which is funded by the Swedish Research Council, is a unique collaboration between physiotherapists and automatic control researchers. The results will be used to gain insight into fall-related concerns among elderly people and also to construct a humanoid robot designed to mimic human biomechanics.

The results of the studies have been published in the Journal of Electromyography & Kinesiology:

Decline in sensorimotor systems explains reduced falls self-efficacy, Pauelsen M, Vikman I, Strandkvist VJ, Larsson A, Roijezon U., J Electromyogr Kines. 2018;42:104-10.


Both psychological factors and physical performance are associated with fall-related concerns. Pauelsen M, Nyberg L, Roijezon U, Vikman I. Aging Clin Exp Res. 2017.



Topics: Multivariate Data Analysis, SIMCA, OPLS, Medicine/Health

Lennart Eriksson

Written by Lennart Eriksson

Sr Lecturer and Principal Data Scientist at Sartorius Stedim Data Analytics