Although individual life insurance has moved toward accelerated underwriting, there is still a place at older ages and higher face amounts for medical evaluation. An ECG would be a valuable part of that, but the high cost, inconvenience, poor quality of insurance ECGs, plus limited expertise of many readers reduces its value. There may be an answer on the horizon.
AI is now used in evaluation of ECGs, and we heard exciting news on this from Emoke Posan MD (PartnerRe) at the 2023 AAIM meeting. AI can use 12-lead, 6-lead or even single-lead ECGs and read rhythm more accurately than physicians (not a huge surprise). More importantly, it can also identify various cardiac issues such as systolic and diastolic heart failure, cardiomyopathies, and valvular heart disease invisible to the human reader and even predict risk of intermittent atrial fibrillation from an ECG currently showing normal sinus rhythm often by using parts of the ECG tracing humans cannot. Right now, most of what is published on AI ECG reading are articles looking for specific impairments (showing AI going far beyond what a trained human could do) rather than complete ECG readings, but things are moving quickly. I anticipate having AI ECG screening software soon that will far outperform what we currently see from even electrophysiologists producing a reading that gives us probabilities of various impairments such as heart failure which was previously impossible. Much of that increased yield will still be available with ECGs limited to the 6 limb leads and some of it with a single lead. Mayo clinic has run a trial of using Apple watches and cell phones with proprietary software to transmit results to physicians and is a leader in AI/ECG research. Recent articles on AI reading of ECGs include reviews by Konstantinos, Attia (both at Mayo) and Martinez-Selles. Unfortunately, there is no way for a human reader to confirm many of the AI conclusions (usually expressed as probabilities) regarding an ECG as the AI uses parts of the ECG and identifies differences that we cannot. Heat maps can help us identify the parts the AI finds useful for various findings but not the magic of how it uses them.
I am not convinced that using a single lead from the applicant’s Apple watch or Fitbit is going to provide enough data of sufficient quality for a robust analysis and would likely still require additional software to be installed on a cell phone. On the other hand, having a paramed haul a 12-lead ECG machine around and hooking up leads to an applicant is a huge problem. Instead, there are new devices including the KardiaMobile 6L battery powered device which is smaller than an energy bar and much thinner. It has a place for the applicant to put their thumbs on 2 contacts on one side and hold the middle contact on the other side of the device to their left ankle generating the pictured 6-lead ECG (part of 30 seconds recorded). I performed this and other 6-lead ECGs on volunteers at home using the device purchased at the Kardia website with similar results, although getting the ankle lead to connect for everyone (an image on your phone of each contact turns green when properly connected) and having the subject relax while holding the device to the ankle for a smooth baseline could be tricky. Perhaps electrically conductive gel pads would help. The Kardia device sends 30 seconds of ECG to a smartphone having their downloaded software (free from app stores) and from there by email anywhere.
This combination of a mini 6-lead ECG device and an AI reading program should provide a far less expensive and invasive option (no chest leads) with much more robust report capable of providing the probability of dysrhythmias and cardiac dysfunctions including heart failure and hypertrophy. I don’t think it will be long until we see AI reporting of 12, 6 and single-lead ECGs being used without human review (at least if found “normal” or “low risk”).
It may be that this AI-read ECG could become part of a simplified paramedical collection or even applicant self-collection approach, but whether 12, 6 or single lead, I think human reading of ECGs will fade into history replaced by the far more informative and robust AI reading which provides a meaningful addition to risk evaluation.
About the Author
Michael Fulks, MD, Consulting Medical Director, is board-certified in internal and insurance medicine. After leaving practice, he served as a medical director, creating or editing several underwriting manuals and preferred programs. More recently, Mike has consulted for CRL participating in its mortality research on laboratory test results, BP and build, and in the development of risk-scoring tools for laboratory and non-laboratory data.