Let’s talk about smartwatches.

In recent weeks, a couple of very interesting papers have been published, showing, on one hand, the tremendous potential of these consumer-grade devices, and on the other hand, their limited translation into clinical practice.

For those who may not be familiar with wearable devices, sensors, and data, I can recommend a non-expert guide that we recently published ¹ (I hope you’ll forgive me for cheekily plugging our own work here).

A wearable-based aging clock associates with disease and behaviour

The main piece I want to discuss today is from researchers at Apple, and it has just been published in Nature Communications ². As the title of the paper suggests (“A wearable-based aging clock associates with disease and behavior”), this study presents an investigation into a novel deep learning model based on Apple Watch data that predicts biological age.

A standardized definition of biological age is somewhat complicated ³,  but put simply, it can be defined as the age someone should be, given the biological processes he or she has undergone. Estimating biological age is equivalent to reclassify age with respect to the age of a healthy cohort, rather than merely counting how many days have passed since birth (i.e. chronological age). The difference between biological age and chronological age is a measurement of how much younger or older someone appears, a qualitative assessment we are probably all guilty of making when meeting someone for the first time. The mismatch between biological and chronological age is considered a holistic measure of global health, and several studies have already shown that it predicts morbidity and mortality.  For example, a recent systematic review has identified 17 studies investigating associations between age gap derived from ECG data and cardiovascular outcomes ⁴.

However, the study from Apple is novel and noteworthy for two reasons. First, it is the first to use data from a consumer-grade smartwatch using the photoplethymogram (PPG), the signal collected by the optical sensor on the back of the watch, routinely used to measure heart rate, heart rate variability, and respiratory rate ⁵, which is rapidly becoming ubiquitous in our lives. Second, it stands out for the breadth of its results, which include associations with diagnoses, predictions of adverse events, and correlations with lifestyle factors (exercise, smoking, and sleep) and life events (e.g., pregnancy or medical procedures).

In the study, biological age was estimated from a series of 60-second PPG recordings using state-of-the-art deep learning models, including self-supervised (or foundation) models. The analysis drew on approximately 150 million participant-days of recordings collected over several years from 213,593 participants in the Apple Heart & Movement Study ⁶. The model was trained to predict age using data from participants who self-reported healthy status and habits (about 5% of the total) and tested on the remaining participants. The authors used only PPG data collected at rest, as their goal was to capture information related to cardiovascular function rather than behavioral activity.

The predicted age was termed PpgAge, and the difference between PpgAge and chronological age was called the PpgAge gap, with a larger PpgAge gap indicating older biological age.
The authors reported five main findings:

1. PpgAge accurately estimated chronological age across a range of demographic groups.
2. The PpgAge gap strongly correlated with the diagnosis of various chronic diseases.
3. The PpgAge gap predicted incident cardiovascular events and metabolic disease diagnoses.
4. The PpgAge gap was associated with behavioral factors, including sleep, exercise, and smoking status.
5. PpgAge exhibited sensitivity to longitudinal physiological changes, such as cardiac events.

The association between increased PpgAge gap and incidence cardiometabolic disease deserves attention. The authors defined four adverse outcomes:

1. Atherosclerotic cardiovascular disease events (new diagnosis of coronary artery disease, heart failure, or peripheral artery disease; or heart attack, stroke, angioplasty, stent, or coronary bypass),
2. Hypertension, 
3. Hyperlipidaemia, 
4. Diabetes.

The PpgAge gap was significantly associated with all these outcomes, with an effect size (hazard ratio) comparable to, or greater than, traditional risk factors such as hypertension, diabetes, and high cholesterol.
Beyond its potential applications in longevity research or in evaluating the efficacy of longevity interventions, this work also carries implications for clinical translation. For example, PpgAge could serve as a more informative marker of biological age for risk stratification and clinical decision-making. Moreover, it could enable remote monitoring of individuals, potentially signalling worsening disease (e.g., heart failure) or sudden, unexpected physiological changes.

In addition to providing a comprehensive set of results, the authors should be commended for their transparency in discussing the study’s limitations. These include concerns regarding cohort representativeness (possible selection bias), reliance on self-reported outcomes, relatively short follow-up duration (although some participants contributed data for up to four years), and uncertainty about the mechanisms driving these predictions—beyond general references to vascular aging, arterial stiffening, reduced cardiac output, and altered autonomic function affecting PPG waveforms.

Clinical translation of smartwatch data

Despite the compelling evidence presented in this paper about the link between an increasingly ubiquitous wearable technology (PPG) and both prevalent and incident cardiovascular disease, it remains unclear how smartwatches and other consumer-grade devices can be effectively harnessed to advance healthcare. In this regard, very recent studies offer both caution and the prospect of progress.

Caution comes from a systematic review on wearable technology for predicting and preventing heart failure ⁷ , recently published in the European Heart Journal – Digital Health. The review focused on clinical studies that used wearable sensors to predict or prevent hospitalization in patients with heart failure. It identified eight studies demonstrating an overall reduction in hospitalization rates compared with standard care. However, none of these studies used consumer-grade technology, highlighting the limited penetration of smartwatches into clinical practice.

The prospect of progress, however, comes from an American study using data from the All of Us cohort ⁸ , which showed that smartwatch data consistently improved outcome predictions beyond state-of-the-art models based on electronic health records. This improvement was observed across ten clinical outcomes, including hypertension and diabetes. The authors of that study ⁹ , still pending peer review, claim that it represents the first large-scale evaluation of the benefits of integrating wearable and hospital data, enabling more holistic and personalized health outcome predictions.

Only with hindsight will we know whether smartwatches will remain personal gadgets, tools for wellbeing, or devices capable of transforming healthcare. For now, we proceed with caution and the prospect of progress.