Improving Mobility for the Elderly Using Artificial Intelligence
We take many of our daily privileges for granted.
Food on our tables, roof over our heads, water to drink, family to spend time with, and so on.
But what if your own body failed you?
What if those two legs of yours refused to hold you up? To take you from place A to B? What if they were your barriers opposed to anything external? They refused to let you go to the bathroom, take a shower, go for a walk without support. Your own body trapping yourself.
Mobility is a privilege.
A privilege that 1 in 5 seniors (and that’s only in the US & Canada) do not get to enjoy. And yours will begin to deteriorate too… in time.
“If you’re unable to get out then you can’t go shopping, you can’t go out with your friends to eat dinner or go to the movies, and you become dependent on other people to get you places. So you become a recluse, you stay home, you get depressed. With immobilization comes incontinence, because you can’t get to the bathroom, you can develop urinary infections, skin infections. The list goes on,” says geriatrician Dr. Suzanne Salamon, an instructor at Harvard Medical School.
Mobility is the ability to move.
Immobility and deteriorating ease of mobility results in an increased risk of falling.
Who does this affect?
Older people have the highest risk of death or serious injury arising from a fall and the risk increases with age.
This risk level may be in part due to physical, sensory, and cognitive changes associated with ageing, in combination with environments that are not adapted for an aging population.
- Falls are the second leading cause of accidental or unintentional injury deaths worldwide.
- Each year an estimated 646 000 individuals die from falls globally of which over 80% are in low- and middle-income countries.
- Adults older than 65 years of age suffer the greatest number of fatal falls.
- 37.3 million falls that are severe enough to require medical attention occur each year.
According to the U.S. Centers for Disease Control and Prevention: One in four Americans aged 65+ falls each year. Every 11 seconds, an older adult is treated in the emergency room for a fall; every 19 minutes, an older adult dies from a fall.
Here are some more numbers that prove that falling is a serious problem:
- Falls are the leading cause of fatal injury and the most common cause of nonfatal trauma-related hospital admissions among older adults
- Falls result in more than 2.8 million injuries treated in emergency departments annually, including over 800,000 hospitalizations and more than 27,000 deaths
- In 2015, the total cost of fall injuries was $50 billion. Medicare and Medicaid shouldered 75% of these costs
- The financial toll for older adult falls is expected to increase as the population ages and is expected to reach $67.7 billion by 2020
- Over 800,000 patients a year are hospitalized because of a fall injury, most often because of a head injury or hip fracture
- Each year at least 300,000 older people are hospitalized for hip fractures. More than 95% of hip fractures are caused by falling.
- Falls are the most common cause of traumatic brain injuries (TBI)
Physical Impacts of Falling on Seniors
- Falls can cause hip and thigh injuries: the most common reason for hip fracture hospital admissions (9 in 10 cases). Other injuries that result from falls include head injuries, wrist fractures, and other injuries.
- Hip fractures also impose a heavy long-term burden as older people become less independent, more reliant on family members and carers, and have an increased chance of moving into an aged care facility.
Psychological Impacts of Falling on Seniors
- After a fall, many older people become fearful of falling again and lose confidence in walking. You may start to do less physical activity. However, over time this reduced movement makes it more likely that you’ll have another fall because of poorer balance, weaker muscles, and stiffer joints.
Effects of Falls in the Elderly
The factors associated with aging result in decreased ease of mobility, such as…
- The decline in physical fitness: Reduced muscle strength, decreased bone mass, loss of balance and coordination, and reduced flexibility.
- Impaired Vision: age-related eye diseases make it difficult to detect fall hazards, such as steps, puddles, and thresholds. While these environmental factors can be limited in elderly areas, they are inevitable and pose definite
- Increased reliance on medications: Side-effects, such as drowsiness, dizziness, and low blood pressure, can all contribute to an accident. Sedatives, antidepressants, anti-psychotics, opioids, and some cardiovascular drugs are the most common culprits. The impact that first-generation antihistamines pose on the poses on the central nervous system indicates symptoms such as “anxiety, confusion, sedation, blurred vision, reduced mental alertness, urinary retention, and constipation.” According to the Merck Manual, just over 40 percent of seniors take at least five drugs per week.
- Chronic Diseases: Health conditions such as Parkinson’s disease, Alzheimer’s disease, and arthritis cause weakness in the extremities, poor grip strength, balance disorders, and cognitive impairment.
Peripheral neuropathy, or nerve damage, can cause numbness in the feet, making it very difficult for a senior to sense environmental hazards and get around safely.
- ⬆ Chances of prior surgical procedures: Hip replacements and other surgeries can leave an elderly person weak, in pain and discomfort, and less mobile. This can be temporary while a patient heals or a new and lasting problem.
- ⬆ Susceptibility to environmental hazards: The majority of falls in the elderly population occur in or around seniors’ homes. Environmental factors such as poor lighting, clutter, areas of disrepair, loose carpets, slick floors, and lack of safety equipment can jeopardize a senior’s safety in their home.
Common factors that lead to loss of mobility:
- old age
- low physical activity
- impaired strength and balance
- chronic diseases (e.g. diabetes & arthritis)
Given the numerous factors that increase one’s fall risk, let’s look at how professionals today are handling this problem.
How do we test an elder’s mobility today?
The consequences of falls are being suppressed once they occur. That is a reactive approach to addressing this problem.
In the medical scene, the analysis of one’s gait is typically approached through a qualitative method. While it is timed (adding to the quantitative methodology), the specific analysis is often based on the observations of the medical staff.
Timed Up and Go Test
- The person is sitting on the chair, and when the stopwatch starts, they have to walk 3 meters and turn around and walk back to the chair and sit down. Assistive footwear and a walker is also needed, but the observer must generally observe three variables.
- Postural stability — is the individual centered well?
- Gait (including speed, stride length).
- With age, gait speed increases while stride length decreases.
- Sway — with more sway, there is less balance around the center of mass.
- An elder adult which takes longer than 12 seconds is at a high risk of falling.
Functional Reach Test
- Identifies fall risk based on the balance of an individual try to reach a surface -> if they are greater than 10 inches (25cm) then they have a lower chance of falling, but if they reach 6–10 inches, then they have a high risk of falling.
- You must maintain a center of balance while doing this -> meant to assess dynamic balance
- Has a high re-test ability (gets similar levels upon retesting) but low accuracy and positive predictive value (cannot predict fall risk too well, but predicts negative predictive value — not having fall risk — extremely well). This is not effective for predicting fall risk, though.
POMA (Tinetti Performance Oriented Mobility Assessment
- Similarly to the Timed Up and Go Test, this utilizes a chair. The person has to stand up, turn 360 degrees, and sit down on the armless chair and the professional identifies the balance of the individual. After this, they are tested to walk over a distance: first at original gait velocity, and then at a faster speed. Professionals identify, and scale, their stride length, step height, and the symmetry and velocity of the steps, among other gait-related factors.
- Creates a quantitative index based on the observational data based on gait and balance.
- Makes them perform specific actions (walking, rising, sitting, turning around, step symmetry + stride length, etc.)
- Pros: about 20 minutes to perform and good inter-rater reliability (observers tend to generally agree), 93 percent of the fallers were identified (thus, a high specificity)
- Cons: poor identification of non-fallers (low sensitivity), quantification to specific rating scales do not account for small differences, test re-test error is high, and they cannot identify what specifically in gait/balance impacted the fall risk.
The general trend with the information gathered from these tests is that the scale is all one-size-fits-all — it does not analyze the little parts that differentiate the gaits of different people (and thus, cannot specifically find the cause of the fall risk). We need quantitative data.
Existing Solutions For Quantification of Fall Risk
Sensor-Based Analysis Of TUG Test
- Used inertial sensor-equipped shoes and passed an instrumental walkway with and without gait support (so with and without a walker).
- Walker improved their gait velocity, stride length, and swing time significantly. However, without, there was a lower walking rate for the people involved.
- This utilized 106 participants and multiple variables.
- Was classification based and analyzed how effective a walker was -> not effective in predicting fall risk.
- This follows the outline of the POMA test, and if the data is inputted into the algorithm correctly, we can utilize the POMA test quantitatively rather than qualitatively.
Machine Learning Analysis Using SVM, ANN, and Random Forests
This method uses a sensor to detect the specific data required for gait based analysis (including accelerometer) and typically use a principal component analysis to extract significant information, and implement either an SVM or an ANN on the data for gait analysis.
This can detect various age groups (specifically, young-middle age, older, and geriatric adults) and was mainly classification based, rather than a predictive model. Specifically, it detected gait abnormality that characterizes those age groups and classified those age groups based on that. This would be incredibly interesting to gain data from, specifically because the correlation between gait change is adequately quantified through this methodology.
However, utilizing three different algorithms, they got various accuracies, specificities, and sensitivities -> don’t know which would be the most accurate.
This was an incredibly detailed test, however, with 239 participants and 23 variables → this was associated typically with the German and Dutch elderly populations, which is an incredibly homogenous society and will not translate for all areas.
Gait Dynamics to Optimize Fall Risk Assessment
There was another study that utilized PCA for feature extraction (like the previous one) and was able to build a fall classification model using PLS-DA (Partial Least Squares Discriminant Analysis).
This worked on identifying pace, variability, and coordination as key properties to formally develop a gait analysis.
Utilizing a Receiver Operating Characteristic curve, this was able to create an accuracy between the correlation of True Positives and False Positives of 0.93 (Area Under Curve).
The specificity of the model was about 80 percent, especially after cognitive and gait outcomes were added.
The problem with this analysis was the sample size of n=61 and was based on specific data acquired from a hospital in Amsterdam. Therefore, it doesn’t extend well towards all the possible impacts because the sample size is so small.
The main problem, again, is that this cannot be widely distributed — however, the principle is great. What do we do now?
What’s the solution?
All of these studies have been done during a certain, set point of time, and do not take into account all of the factors. Even in the cases of quantitative data, while the classification + regression-based analysis was done, it predicted it way earlier and it did not analyze all the patterns that the seniors undertake. What about some daily patterns? What environmental factors are they exposed to? What can we do better?
Our team came up with a prototype of a device called MobileMe. Combining the use of a sensor and a perceptron neural network, we will perform real-time, proactive modeling of the gait and balance of an elderly individual to assess their gait fo fall risk. Rather than a reactive measure, this detects the risk during any given time of day — using a sensor to get the data.
Our sensor would involve a gyroscope and an accelerometer. Gait data (acceleration, walking patterns, and spatial) is run through a classification model to identify specific gait concerns, like propulsive gait, steppage gait, etc. This would be numerical data that would be inputted in a perceptron neural network.
The neural network is fed pre-labeled training data containing a graph of movement and acceleration over time corresponding with a specific gait problem. This data is collected from the IMU (inertial measurement unit) onboard the microprocessor device.
During the testing phase, new graphs will be run through the NN & the prediction of the gait problem(s) by the network will be compared to the true condition to minimize the loss value.
After having connected the Bluetooth module with the app (in order to share and display data) using a standard Bluetooth protocol, we collect data for every millisecond on the Serial Plotter. After 15 minutes of the MobileMe (POMA-style) test, data will be sent to the backend NN.
The NN will run on the data and identify patterns of gait problems and display the output of the specific gait problem and its confidence rate. To do this, we plot a Receiver Operating Characteristic curve, which compares the rate of false positives and true positives. If we have an area under the curve > 50%, we have a higher confidence interval and should send the data to the doctor. Of course, data will also be sent under the interval — this would show situations when fall risk is low and will provide more quantitative patterns hinting towards why that might be the case.
This will help doctors create a personalized treatment plan to improve the patient’s overall mobility and well-being. The device automatically sends out a movement graph, with a gait problem of a specific confidence rate. This is sent to the doctor, who analyzes the data and makes sure that the treatment works to the patient’s behest.
By emphasizing this, the treatments would not be cookie-cutter and perfectly identical for every single person — it would be catered based on an individual’s gait and balance performance. This way, we can optimize mobility treatments and predict fall risk at a much more effective, proactive pace.
Possibilities for the future
This is a predictive model — since causation does NOT = correlation, it may not be 100 percent effective all the time. Newer technologies like causal neural networks can definitely locate the cause of such gait problems — as it becomes more widespread, a shift to modeling based on this increases the confidence interval and increase the likeliness of the fall. We will be able to trace the fall risk to more specific reasons — and personalize the treatment even more.
Alongside that, the training data for the populations will vary, needing a large level of data collection per region. Right now, what we will do is start out small — as we gain more data, we can expand MobileMe to more areas across the whole world. In less developed areas, where medical infrastructure is gradually improving and we will be able to connect with more people, they tend to have the greatest elderly fall rates — especially in areas like Southeast Asia. By leveraging this technology in a widespread manner, fall risk in these areas would reduce greatly and the elderly can continue to live comfortably — to their best life!
⭐️ Harvard Medical School: Mobility Loss in Elderly
⭐️ Performance-Oriented Mobility Assessment
⭐️ What Causes Elderly to Fall? Identifying Risk Factors
⭐️ Research Paper: Personal Risk Factors Affecting Mobility
World Health Organization: Facts on Mobility-Related Falls
Research Paper: Genomic Factors Affecting Aging
Research Paper: Investigating the Root Causes of Falling
Research Paper: Age-Related Decline in Vestibular System
BioMed Central: Longevity & Lifestyle in Aging
Research Paper: Genetics in Human Aging
Geriatric Medicine: Risks of Antihistamine in Elderly Falling
Research Paper: Hypoglycemia in Elderly Falling
Research Paper: Exercise in Mitigating Pathogenesis of Aging
Research Paper: How Does Genome Instability Affect Aging?
Statistics Of Mobility-Loss and Elderly Falls Globally
Background research and pitching with: Neha Shukla, Hung Huynh, Orna Mukhopadhyay, Mir Ali Zain, Sarah Naghmi, Sriya Chintalapalli
Idea, Device and Neural Network built by Neha Shukla (patent-pending device and algorithm)
About the Author:
👋 Thanks for reading the article!
I’m Neha Shukla, a 15-year old innovator who’s passionate about using science and technology as a catalyst for social change. I write articles about my innovations, the latest developments in technology, and the right mindsets for success. Read more to join my journey of making the world a better place!