Why Trustworthy Clinical Trial Data Starts As Analog, Not Digital

By John Oncea, Chief Editor, Clinical Tech Leader

Think back to the last conversation you had about clinical trial technology, and I’d imagine at some point, it turned to dashboards, algorithms, and cloud platforms. Almost none of these start where the data actually starts, though: a physical sensor trying to capture something happening inside a human body.

That gap in attention is exactly what Yasir Shafiullah, a Ph.D. researcher in the University of Oulu’s CWC-RF department who has spent 11 years designing analog and mixed-signal integrated circuits, thinks the clinical research industry is missing. Shafiullah’s background runs through 2G/3G/4G RF optimization work in the field and into VCO and IC design for next-generation wireless systems. Along the way, he started asking a question most clinical operations teams never think to ask: what has to happen, physically, before a heartbeat becomes a data point in a trial database?

A Signal Before It’s Data

“The sensor captures the data, uses low-energy Bluetooth to send it over the air to a mobile phone or Wi-Fi connection, and then the phone or connected app processes the data,” Shafiullah explained, describing a typical wearable heart rate or SpO2 device. Before transmission happens, though, an analog front end has to detect a physiological signal – a heartbeat, a light absorption pattern, a temperature change – and convert it into something electrical. That conversion is where reliability is either built in or lost.

He pointed to the color-based light sensing used in most wrist-worn optical sensors as an example: different wavelengths of light interact with blood and tissue differently, and each pulse produces a distinct electrical response the hardware has to interpret correctly before software ever touches it. Get that step wrong, and no amount of downstream processing will fix it.

Why Placement And Averaging Matter More Than People Assume

Sponsors and CROs evaluating wearable data quality tend to focus on specifications and certifications. Shafiullah’s answer to what actually drives reliability was more practical: correct placement and how the sensor samples over time. A device that averages a measurement over 10 seconds before transmitting, he said, is more dependable than one reporting a single reading every second. Improper placement – a sensor a nurse attaches slightly wrong, or poor skin contact – introduces noise that no algorithm downstream can fully untangle from the real signal.

That distinction matters for decentralized trials specifically, where there’s no clinician in the room to catch a badly placed sensor in real time.

Two Devices, Two Answers

One detail worth sitting with: even well-regarded consumer wearables measuring the same physiological parameter won’t necessarily agree. Shafiullah estimated a routine two-to-five-percent variance between devices from different manufacturers, driven partly by motion: a watch moves relative to skin in ways a fixed skin-contact sensor doesn’t. His practical litmus test for trial-grade reliability isn’t brand reputation; it’s repeatability. If a measurement produces the same result under the same conditions each time, the sensor is behaving reliably. If it doesn’t, something in the hardware needs investigation before the data gets trusted.

The Environment The Hardware Has To Survive

The piece of this conversation clinical technology buyers are least likely to have considered is temperature. Shafiullah described the difference between a decentralized trial unit deployed in Texas heat and one deployed in Alaska cold as a genuine engineering problem, not a logistics footnote. An analog front end has to down-convert and process a signal reliably whether it’s operating at 120 degrees or minus 25. If it can’t, the software layer above it is working from corrupted input no matter how sophisticated that software is.

And critically, that’s not a fixable-later problem. “Bad design is bad design, and we cannot fully compensate for it later,” Shafiullah said. Once decentralized hardware is deployed across a trial’s geography, a poor analog design decision can’t be patched with a firmware update; the device may need to be replaced or redesigned entirely.

What This Means For Evaluating Remote Monitoring Technology

For readers vetting wearable or decentralized monitoring vendors, this reframes the diligence questions worth asking. Instead of leading with software capabilities, it’s worth asking how a device’s analog front end was validated across the temperature and environmental range your trial sites will actually see, how sensor data is sampled and averaged before transmission, and what repeatability testing has been done across units. None of that shows up in a typical vendor pitch deck, but according to an engineer who has spent his career on exactly this layer of the stack, it’s where trustworthy data is won or lost before a single algorithm ever runs.

This is the first layer of a larger technology stack that determines whether decentralized trial data can be trusted. The next challenge – what happens when multiple sensors on the same patient need to agree on when something happened – is even less discussed. That’s the subject of the companion piece, The Synchronization Problem Clinical Research Isn’t Watching, coming August 3.