Fall detection wearables: interpreting sensitivity and specificity correctly

My first brush with a “fall detected” alert felt oddly cinematic—heart racing, phone buzzing, a brief freeze while I decided whether to tap “I’m OK.” Later that night I opened my notes app because something about that moment nagged at me. I kept asking myself a simple question that turned into a rabbit hole: if my wearable is “accurate,” what does that actually mean for real life? The more I traced the math behind the promise, the more I realized how easy it is to misread numbers like sensitivity and specificity. This post is my attempt to translate those stats into what they mean on the kitchen floor, not in a lab. I’m keeping the tone personal because this is how I learn—curious, sometimes skeptical, always trying to connect the dots between data and daily routines.

The moment a notification became a math problem

I used to think a high “accuracy” number was the whole story. Then I learned the first high-value takeaway: sensitivity and specificity are different, and you feel their trade-off in daily life. Sensitivity is about how many real falls the device catches. Specificity is about how many non-falls it correctly ignores. If a device prioritizes sensitivity, you’re less likely to be missed when you truly fall—but you might get more false alarms. Favor specificity, and you’ll be spared nuisance alerts—at the risk of missing a real event. This is not abstract: older adults and caregivers live with the consequences of that balance every day. For context on how common and serious falls can be, the CDC’s overview of older adult falls underscores why detection even matters in the first place (CDC on older adult falls).

• Practical point: Ask the seller or look up the spec sheet for both sensitivity and specificity. If only “accuracy” is shown, it’s not enough.
• Check whether numbers come from lab drops or real-world trials. My own alerts behaved differently on my wrist versus clipped to my waistband.
• Remember individual differences: gait speed, use of a cane, medications that cause dizziness, and even flooring type can shift how algorithms behave in the wild (NIA on falls & fractures).

Why a 95 percent sensitivity device can still miss your worst moment

It sounds paradoxical, but here’s the rub: “95% sensitivity” means that out of 100 true falls, the device flags about 95—under the circumstances studied. That qualifier matters. If a study defined a fall as a scripted, padded drop on a gym mat, that’s different from a sideways slip on a wet bathroom tile. I learned quickly to ask what definition of ‘fall’ the device uses and whether the test scenarios match my life. A sideways stumble while grabbing a counter can generate a different acceleration pattern than a straight backward drop. The same device might shine in one scenario and struggle in another.

Then there’s timing. Many algorithms require the phone or watch to be worn consistently, and some need movement afterward to confirm you’re responsive. If you take it off for a shower, or you nap on the couch and the device thinks you’re “still,” it may interpret signals differently. In other words, sensitivity isn’t purely “device skill,” it’s also “match with your routines.”

The false alarm tax I kept forgetting to count

False alarms are not just annoying; they create what I call the false alarm tax—the social, emotional, and time costs that accumulate. Each false alert may trigger phone calls, startle family members, or tempt you to disable features you actually need. That’s where specificity earns its keep. With higher specificity, the device rejects more non-fall events, like plopping onto a couch or setting your watch down on a table. But even “high specificity” can feel different at home than in a paper—because your home environment produces endless non-fall movements. As the WHO points out, falls are a major public health issue, but the environments that raise risk are diverse, from cluttered hallways to uneven outdoor surfaces (WHO facts on falls). That diversity also feeds the false alarm problem.

Here’s a simple way I taught my brain to think about it. Imagine that in a given month you generate 1,000 “decision moments” for your wearable—bursts of motion that the algorithm evaluates. Suppose only 10 are real falls (1% prevalence). If the device has 90% sensitivity, it catches 9 of those 10. If its specificity is 95%, it correctly ignores 95% of the 990 non-falls—meaning it still raises about 50 false alarms (5% of 990). The positive predictive value (PPV) becomes 9 true alerts out of about 59 total alerts (~15%). Translation: even a seemingly “excellent” device can produce mostly false alarms when true events are rare. That doesn’t make it bad—it makes it predictable math.

Kitchen-table math for PPV and NPV

I started doing napkin math: in my household, how rare are actual falls compared with all the jostles and drops my watch experiences? Two more ideas help:

• Positive Predictive Value (PPV) answers “When my device alarms, how likely is it that a real fall happened?” PPV goes up when true events are more common, and when specificity is higher.
• Negative Predictive Value (NPV) answers “When my device doesn’t alarm, how likely is it that no fall occurred?” NPV tends to be high when falls are rare and sensitivity is strong.

Prevalence—the base rate of falls in your life—drives both. That’s why a watch that works well for my neighbor who has frequent balance issues might behave very differently for me. If you want a clean, non-jargony definition of sensitivity and specificity, the NCI’s glossary is a friendly doorway (NCI glossary).

Device position and everyday variability

I experimented with wearing the device on my dominant versus non-dominant wrist, tried a belt-clip sensor for a week, and noticed how position shifts patterns. Wrist-worn devices see arm swings and sudden bracing against countertops. Hip or torso sensors see center-of-mass changes more directly. Neither is “better” in every situation; they’re different lenses on the same event. A few little nudges improved my results:

• Pick one default wearing position and stick to it. Retraining days after big changes (e.g., switching sides) can help your own expectations calibrate.
• Look for settings that let you tune sensitivity. If you’re getting frequent false alarms from vigorous housework or fitness, a slight dial-down may be worth it.
• Keep the app updated—algorithm refinements often target edge cases discovered in the field.

What numbers on the box don’t tell you

Labels rarely mention who the device was tested on. Was it primarily healthy adults? Older adults using walkers? People post-stroke? That matters. A device can be both “accurate” and “mismatched” for your gait. Consider the denominator: how many non-fall movements were counted when specificity was measured? A study with a tiny denominator can produce a deceptively clean specificity number. I learned to scan for sample size, age range, and whether tests included real-world scenarios like icy sidewalks or bathroom slips.

Also worth noting: not every fall-detection wearable is a regulated medical device. Some are consumer safety features. That’s not a critique—just a reminder that claims and testing standards can vary. If you want to dive into how software-based devices are evaluated clinically, the FDA has a helpful primer on Software as a Medical Device clinical evaluation (FDA SaMD clinical evaluation).

A simple mental checklist I keep on my phone

Here’s the three-step framework that finally cut through the noise for me:

• Step 1 — Notice: What are my real goals? Immediate caregiver alerts? Automatic 911 calls? Quiet logging for later? Understanding the goal clarifies whether I should prioritize sensitivity or specificity.
• Step 2 — Compare: For devices I’m considering, do sensitivity/specificity estimates come from comparable populations and settings? Are there options to adjust thresholds? Check if any documentation references independent testing (even small observational studies can be informative).
• Step 3 — Confirm: Trial the device for two to four weeks. Keep a tiny log: date/time, what you were doing, whether the device alerted, and whether it was right. If alerts are frequent and wrong, tweak settings or placement. If genuine events are going unrecognized, that’s a red flag to escalate and discuss backup plans with a clinician.

Little habits I’m testing in real life

Because I’m not a lab, I went for realistic, boring routines that add up:

• Weekly “quiet drills” with a neighbor: we check whether our devices can still recognize motion after firmware updates. It builds confidence without pressure.
• Bathroom safety first: I laid a non-slip mat and added a reachable towel bar. If environment risks drop, the wearable can be a second net—not the only one (MedlinePlus fall prevention).
• Protected alert circle: I list two contacts who understand false alarms are part of the game. We agreed on a simple follow-up script so no one panics.

Signals that tell me to slow down and double-check

My rule is to listen when patterns change. These are the cues that make me pause:

• Frequent false alarms that make me want to turn the feature off. Instead of switching it off, I review settings, wearing position, and whether a recent update changed default thresholds.
• Missed real events—even one is a serious signal. I review what happened, how the device was worn, and whether I had connectivity issues. If it’s not fixable via settings, I consider a different device category (e.g., hip sensor vs watch) and talk with a clinician about alternatives.
• New symptoms like dizziness, fainting, or unsteadiness. That’s a health issue first, technology second. Resources from public health organizations help frame next steps (CDC).

What I’m keeping and what I’m letting go

I’m keeping three principles on an index card:

• Match the metric to the mission: If I live alone and want rapid help, I’ll lean toward higher sensitivity and accept some false alarms. If I have attentive caregivers nearby, I may prioritize specificity to avoid alarm fatigue.
• Base rates rule: PPV and NPV ride on how often real falls happen in my life. The rarer the event, the gentler I should read an “excellent” PPV—and the more I should consider broader safety improvements at home.
• Iterate in the open: I tell my alert contacts that we’re learning together. No blame for false alarms; no shame for misses. We adjust and move on.

And here’s what I’m letting go: the idea that a single number on a product page can settle anything. Real-world safety feels more like layering reasonable protections than chasing perfection. If I can understand the trade-offs, keep the environment safer, and choose settings that fit my day, the wearable becomes a good teammate instead of a magic shield.

FAQ

1) Is a higher sensitivity always better?
Answer: Not always. Higher sensitivity catches more true falls but may trigger more false alarms. If false alarms cause you to disable alerts, the net safety can drop. Aim for a balance that supports your goals and tolerance for interruptions.

2) What’s a “good” PPV for fall detection?
Answer: It depends on how often falls actually happen in your life. Even with excellent sensitivity/specificity, PPV can be modest when real falls are rare. Look at both numbers and consider your prevalence. Try a trial period and track your own PPV.

3) Do I need an FDA-cleared device?
Answer: Some fall detection systems are consumer safety features, others are medical devices. FDA-cleared options have specific evidence behind them, which can be valuable in clinical contexts. For everyday safety, consumer features can still help, but read claims carefully (FDA overview).

4) Where should I wear the device?
Answer: Follow the manufacturer’s guidance. Wrist devices are convenient but may see arm-swing noise; hip or torso sensors may track body movement more directly. Pick one position, stick with it, and reassess after a few weeks of logging.

5) How can I reduce false alarms without risking misses?
Answer: Start by adjusting thresholds if the app allows, fine-tune your wearing habits, and cut environmental risks (lighting, mats, decluttering). If nuisance alerts persist, consider a device with higher specificity or a different form factor, and share your log with a clinician for tailored advice (MedlinePlus tips).

Sources & References

• CDC — Older Adult Falls
• WHO — Falls Fact Sheet
• NIA — Falls and Fractures
• NCI — Sensitivity and Specificity
• FDA — SaMD Clinical Evaluation

This blog is a personal journal and for general information only. It is not a substitute for professional medical advice, diagnosis, or treatment, and it does not create a doctor–patient relationship. Always seek the advice of a licensed clinician for questions about your health. If you may be experiencing an emergency, call your local emergency number immediately (e.g., 911 [US], 119).