Fall Detection is Hard. Here’s What We Can Do.

The ability for wearables to accurately and reliably detect falls has been the subject of much skepticism in the senior care world, and rightfully so. Falls are extremely difficult to detect.

Here are today’s options:

[1] “PERS-quality” fall detection

These are the super-simplified “Help I’ve Fallen and Can’t Get Up” type companies, which are selling no more than a pendant with a button for a $20-$50/month fee. A few of these companies claim their pendant can detect falls, but you should take that with a grain of salt. The only sensor these pendants have is an accelerometer, which measures acceleration (aka impact). However, acceleration is not enough to reliably detect a fall. To an accelerometer, there may be no difference between the impact of a fall and the impact of the pendant moving quickly from a fast movement, like a bump against a chair (think: false alarm).

[2] “Big Brother is watching” fall detection

Camera and light sensors on the walls offer another crack at fall detection. In the most sophisticated cases, they use computer vision to analyze a person’s movements to detect a fall. This technology performs much better than the PERS pendants. Some companies are boasting up to 95% detection. But it’s not without a catch. This approach works well only if you can capture every angle without obstruction. Once a resident is out of frame, it provides zero detection. It can also get pretty costly to set up that kind of coverage.

More importantly, feedback from the market suggests that this approach is met with strong resistance from seniors as an invasion of privacy, and from communities as a liability concern. If the two end users — seniors and senior care communities — don’t want cameras on the walls, it’s not a viable option.

[3] “Wearable” fall detection

Finally, you have the more sophisticated algorithms that run on devices like the Apple Watch or Somatix Smartband. These use an overlap of a multitude of sensors to more accurately detect and differentiate actions that are a fall from other high-impact movements. These types of algorithms took years and thousands of falls to develop. They yield better results, but they’re not perfect. In most cases, the best you will get is 6 or 7 out of every 10 falls. And while these have significantly reduced false alert rates, you can still expect one every few weeks.

Why is it so hard to detect falls?

(1) There is a lot of variability in a fall. Falls can happen forwards, backwards, sideways, from a bed, from a chair, slipping, tripping, fast or slowly, etc. And you can land on a wrist or elbow or hip. Every variation ends up with the resident on the floor, but there are many ways to get there.

(2) Falling is a rare event. Our algorithms are developed by training the sensors on real falls. That means a person needs to fall while wearing our smartband for every additional data point that we can use. Unlike our other algorithms, like drink detection and step tracking, which result in tens to thousands of data points per day per person, the fall detection algorithm can only be fueled by a few falls per week until mass scale is reached. It’s a chicken and egg problem.

Figure: Overlapping sensor data collected from a single fall

Here’s what we can do:

First of all, we can continue to improve the algorithms. With every new fall that happens while a senior is wearing our band, the algorithm gets smarter.

Next, we can add extra data inputs to improve the accuracy rates. Our newest smartband coming out in December will have additional sensors, including a barometer that can measure altitude to the accuracy of 5 cm. This will allow us to detect the change in height during a fall, not just the impact and motion of falling. In addition to measuring a change in altitude, we can also measure changes in vitals, like a spike in HR, to offer more confirmation of a fall. But the specifics on performance improvement are TBD.

Finally, we need to work together as an industry to improve this performance with better data. Senior care standards are evolving, and so is technology. Unlike pendants, which have barely upgraded in decades, modern senior care technology is upgrading on a monthly basis, give or take. Technology companies can produce better products when senior care communities adopt early, test new features, and collaborate on product improvements.

What do you think are senior care communities’ responsibilities and capacities to serve as testing sites for new technologies for residents?