Predicting Asymptomatic Paroxysmal Events through Continuous Monitoring

This direction is merged into another direction: asymptomatic event underreporting. Its last version stays readable and citable here.

Version 1 of 120 papers23 conceptsfrozen 2026-10-04

In clinical medicine, the most dangerous events are often the ones you cannot see. Paroxysmal conditions—such as Atrial Fibrillation (AFib) or non-convulsive seizures—occur intermittently and frequently without symptoms. Because these "silent" episodes leave no trace once they pass, standard diagnostic tools like periodic ECGs or bedside observations often miss them entirely.

Recent research is shifting toward a proactive paradigm: using continuous monitoring and machine learning to detect these events in real-time or even predict them before they manifest.

The Invisible Burden of Asymptomatic Events

The core challenge of paroxysmal conditions is the "detection gap." As Mei‐jun Liu and colleagues point out, relying on periodic Holter monitoring to assess heart rhythm can significantly overestimate the success of medical interventions because transient, asymptomatic episodes simply aren't captured. This isn't just a measurement error; it’s a clinical risk. Arjuna Scagnetto notes that patients are often unaware of AFib until they suffer a major complication like a stroke.

This "silence" extends beyond the heart. In neurology, D. J. Albers and Jan Claassen describe non-convulsive seizures in brain hemorrhage patients that have no physical manifestations. Similarly, Yiying Zhang explains that video-based observation in epilepsy research fails to capture subclinical electrical events that still contribute to long-term brain damage.

The common thread across these domains is that intermittent snapshots of health are insufficient for paroxysmal diseases.

From Snapshots to Continuous Streams

To close this gap, researchers are moving away from specialized clinical hardware toward "always-on" monitoring. The intuition is simple: if the event is unpredictable, the sensor must be omnipresent.

  1. Wearable PPG: Yichen Shen and Andrew Y. Ng demonstrated that deep learning models can use Photoplethysmography (PPG) from smartwatches to detect AFib with 95% accuracy. By training models to be robust against the "noise" of daily movement, they turn consumer devices into clinical-grade diagnostic tools.
  2. Acoustic Sensing: Perhaps the most novel approach comes from Haoxian Liu and Yi Huang, who developed MobileAF. This system uses a smartphone's speakers and microphones as a form of "active sonar" to detect minute skin deformations caused by the pulse at the wrist, removing the need for even a smartwatch.
  3. EEG Spectral Analysis: For "silent" seizures, D. J. Albers applied Empirical Orthogonal Functional (EOF) analysis to EEG data, uncovering patterns of seizure activity that are invisible to the naked eye or standard algorithms.

Predicting the Future with Machine Learning

Detection is reactive; prediction is transformative. The goal is to identify the "pre-ictal" state—the physiological window just before an episode begins.

Amir Aminifar and Jean-Marc Vésin developed a methodology to predict AFib onset up to 25 minutes in advance. Their key insight was patient-specificity: by tailoring feature selection to an individual’s unique heart profile rather than using a one-size-fits-all model, they achieved a 97.1% F1 score.

On a longer time horizon, Arjuna Scagnetto used deep convolutional neural networks (CNNs) to predict the onset of AFib up to three years in advance from a standard 12-lead ECG. Interestingly, the model found that leads traditionally ignored by cardiologists for this purpose (like avR) were highly predictive, suggesting that AI can see structural "remodeling" in the heart long before it becomes symptomatic.

The Challenge of "Alarm Fatigue"

While continuous monitoring solves the detection gap, it creates a data management problem. Jessica R Golbus and Aishwarya Pastapur warn of "alarm fatigue" and patient anxiety triggered by frequent notifications from wearables.

The research direction is therefore not just about better sensors, but about better triage. This involves combining electrical signals (ECG) with clinical context (EHR). For instance, Jiandong Zhou showed that multimodal models—those that look at both the heart's electrical "signature" and the patient's broader medical history—are significantly more accurate at predicting complications than models looking at sensors alone.

Why This Works

This research direction works because it recognizes that "asymptomatic" does not mean "invisible." Even when a patient feels nothing, the underlying pathology—whether it's systemic inflammation in diabetics Zexiong Li or the loss of neural pathways in heart transplant recipients Felipe Kazmirczak—leaves subtle traces in physiological data. By moving from intermittent clinical checks to continuous, machine-learning-augmented monitoring, we can finally see the silent signals of disease.

Papers in this version

History & citationVersion 1 · 2026-10-04

Merged into another direction · Permanent link

View frozen version 1

Cite this direction

Latest version (a living page):

@misc{lacuna_predicting_asymptomatic_paroxysmal_event_5a67379e,
  title = {Predicting Asymptomatic Paroxysmal Events through Continuous Monitoring},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/predicting-asymptomatic-paroxysmal-events-through-continuous-monitoring/rdr_5a67379ec8bd403396b3cf58dfe5b511},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-19},
}

This exact version (version 1, frozen text):

@misc{lacuna_predicting_asymptomatic_paroxysmal_event_5a67379e_v1,
  title = {Predicting Asymptomatic Paroxysmal Events through Continuous Monitoring},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/predicting-asymptomatic-paroxysmal-events-through-continuous-monitoring/rdr_5a67379ec8bd403396b3cf58dfe5b511/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-19},
}