The launch of the Apple Watch’s hypertension detection algorithm has received a lot of media attention, with many posts citing its sensitivity and specificity. For most of us who may be a little rusty on statistics, those numbers can feel abstract.
So I took 100 middle-aged men and ran them through the same sensitivity and specificity Apple reported in its FDA submission. To build the model, I had to make a few assumptions, mainly about how common undiagnosed hypertension really is in that age group. For that, I leaned on historical estimates of 35%.
I focused on middle-aged men because they represent a group with high rates of undiagnosed hypertension, low primary care engagement, and a high burden of cardiovascular risk.
Key Insights
1. It’s a Screening Tool
It doesn’t make a diagnosis and that’s appropriate and intentional.
2. It Minimizes False Alarms
Apple knows credibility matters. If physicians start seeing a flood of referrals from “hypertension alerts” in patients with normal blood pressure, the feature loses trust.
3. It Misses a Significant Number of Cases
This part concerns me most. Missing cases may create a false sense of “health security”.
Why It Matters
One question I keep asking is how many false positives we’re willing to accept.
Every screening tool lives in tension between sensitivity and specificity. Make it more sensitive, and you’ll catch more true cases, but you’ll also generate more false alarms.
Tighten specificity, and you’ll reduce false alerts, but you’ll miss people who actually have the condition. So the question becomes:
The real question is, what would you rather live with the false alarm or the silent miss? I’d love to hear your perspective.