Apple just came out with a new watch. In addition to telling time and reading text messages and email, the new watch has electrocardiogram (EKG) capabilities. When the EKG detects an irregular heart beat (atrial fibrillation, Afib) it alerts the wearer. How well will this new technology perform in the real world?
Using Bayes calculation and assuming (a big assumption) that the Apple watch can detect Afib with a sensitivity of 99% and a specificity of 99%, we need to determine the prevalence of Afib in the population. This is not easy and there is scant data. Using Dr John Mandrola’s recent article in JAMAnetwork (“Screening for Atrial Fibrillation Comes with Many Snags”, August 2018), we can get a starting number. A study screening 75 to 76 year old Swedes found new Afib in 0.5% of the screened population.
If we put these numbers into Bayes formula, a pretest probability of 0.5% yields a post-test probability of 33%. In other words, if the chance a person has Afib is 1 in 200 and the watch detects Afib, the posttest probability of Afib is 1 in 3. This is a substantial increase and may be worth additional testing and perhaps treatment.
But…
We don’t know the true sensitivity and specificity. A sensitivity and specificity of 99% is almost never achieved by a diagnostic test, even the best tests come in at 90% to 95%. Testing the sensitivity and specificity of the Apple watch’s ability to detect Afib can be done and it won’t be a difficult undertaking. For example, Apple watches can be given to patients with pacemakers who go in and out of Afib. The watch can then be compared to the pacemaker interrogation to see if it accurately detects Afib and then the sensitivity and specificity can be calculated. This study would not require too many patients, can be done in a short time and would not be that costly.
The true prevalence of Afib, however, is another story. This will require a much larger and more extensive study. It would require many thousands of people, monitored for long periods of time (likely years) and would be very costly. In addition, the prevalence noted above is for older patients. The prevalence will be lower in younger unselected patients. If we cut the prevalence in half, to 0.25% or 1 in 400 people, then the posttest probability becomes 20%, still a substantial increase, but less impressive and more likely the watch will have produced a false positive. If the true prevalence is even lower, say by an order of 10 (1 in 2000 people or 0.05%), then the posttest probability is only 5%. It would be much more likely that the result would be a false positive.
The bottom line is that we don't know how accurate the Apple watch will be in detecting Afib. However, even if it's detection rate is nearly perfect, it ability to find Afib is dependent on the prevalence of Afib in the community. If there is a higher prevalence (as would be the case in older patients or heart patients who are hospitalized) then the chance of finding Afib is higher. If the prevalence is low, an irregular heart beat on the watch is more likely a false positive and not Afib. So, if your Apple watch says that you are in Afib, try to keep this in perspective.
No comments:
Post a Comment