What the Failure of Wearables Can Teach Us About AI

“My watch caught my atrial fibrillation.” 

As a physician, these rare episodes can feel like proof that the long-promised consumer tech revolution in healthcare has arrived. It sticks in your mind because it feels miraculous. It feels like something you might tell your friends or family over dinner. But moments like this are memorable precisely because they’re incredibly rare. Consumer devices can sometimes spot something real, but anecdotes aren’t evidence of system-level transformation. 

The fact is, despite billions invested, wearables have yet to demonstrate consistent, population-level improvements in clinical outcomes. They collect physiology data, but critically, they don’t understand context. Devices produce torrents of heart rate, HRV, respiratory rate, heart rhythm data, and more, but they do so outside the context clinicians need to make decisions.

The burden of deciding what data matters and what doesn’t falls, then, on the consumer.  People buy these devices with the best of intentions for their health, but we know that false positives go up when you screen people with low probability of actually having a specific condition. It’s similar to how people in the South should not dive for cover from an impending earthquake when the ground rumbles. 

The impact of wearables on healthcare has been minimal because they prioritize signal collection over clinical context. In medicine, context is the foundation of meaning.

The hierarchy that actually produces valuable clinical insight begins with longitudinal clinical data like charts, labs, imaging, and medication history. Second to longitudinal clinical data is AI synthesis, the intelligence layer that can parse clinical data and generate insight. Then comes wearables, as a secondary input to longitudinal clinical data. 

If you feed a foundational LLM like ChatGPT or Claude a raw stream of heart rate data, it will generate a summary. You might even create a custom GPT to be your personal clinical data analyst. But without the context of your medical story – linked diagnoses, medication lists, prior imaging – the summary it produces will often just be plausible-sounding nonsense. Worst case, it will hallucinate clinical meaning, inducing fear or panic. 

Wearables data can be valuable, but only once the foundation and synthesis exist. Reverse that hierarchy, and you just get a lot of noise.

Getting to the heart of the matter

If wearables are changing outcomes at scale, we would be seeing it in cardiology. This is a specialty that produces some of the richest clinical signals in medicine, from imaging and ECGs to troponins and natriuretic peptides, vitals, medication changes, procedures – the list goes on. All of those data points are signals, fragmented across years, systems, and clinicians who all document differently. 

In practice, we are often asked to reconstruct years of cardiac history in minutes, to spot subtle shifts in heart-failure severity or arrhythmia burden, and to make high-stakes decisions under time pressure. Consumer wearables don’t help us do any of that. If anything, they add more cardiac data. Without clinical synthesis, very little of that data is actionable. 

A rapid heart rate, for example, can be benign or life-threatening depending on medical history. Has the patient been on beta-blockers, have they had a procedure recently? Take an arrhythmia alert, common among consumer wearables, whose signal is only meaningful against the backdrop of prior diagnoses, anticoagulation status, and prior cardiology assessments. 

The bottom line is that these raw wearable signals don’t provide the contextual scaffolding clinicians need to make decisions. At worst, they can even distract us from being able to spot real risk quickly. 

This is not just a wearables problem. Somewhere, at some point, the healthcare system became obsessed with recording more numbers, when we should be focused on weaving the data we have into actionable bedside insights. We can’t keep mistaking the act of monitoring data for the action of improving outcomes. 

Clinicians don’t need more dashboards, more notifications or alarms. We need AI that can synthesize fragmented longitudinal data, flag the markers of clinical severity, and connect all of the data that exists in the EHR into digestible content that supports our decision-making.

Once we have that, only then can wearable signals be interpreted meaningfully, either as confirmatory inputs or early indicators that enrich an already-established clinical picture. 

What this means for the future of AI

The next phase of AI in healthcare must be clinician-centered, EHR-native, and built around longitudinal understanding of the patient. 

The healthcare system cannot ask consumers to be their own diagnostician. Most don’t have the full clinical foundation AI needs to make useful inferences. Wearables data can be beneficial, but the workflow and evaluation should be through established clinical pathways. 

Catching an occasional atrial fibrillation makes us raise an eyebrow in interest, but mostly these stories just sell more devices. As ubiquitous as wearables are, as much of the massive amounts of data they catch, and considering how much people spend on these devices, the impact is currently unimpressive. 

The takeaway from the current failure of wearables is that signal without synthesis doesn’t change outcomes. If healthcare treats AI as just another way to collect or repackage data, it will repeat the same mistake. Yes, wearables can be a potentially valuable data source, but ultimately those potential signals must be integrated into the vast symphony of inputs required to make an accurate diagnosis.

Photo: exdez, Getty Images

This post appears through the MedCity Influencers program. Anyone can publish their perspective on business and innovation in healthcare on MedCity News through MedCity Influencers. Click here to find out how.