Almost every medical device program has a moment where someone puts a $4 sensor module on the table, shows a clean waveform on a bench scope, and declares the hard part solved. Six months later that sensor is producing plausible numbers that disagree with the reference instrument by more than the clinical requirement allows, drifts over an eight-hour session, and falls apart after the third autoclave cycle. Sensor selection is not a components decision. It sets the accuracy claim on your label, and with it the testing you owe.
Start from the clinical requirement, not the datasheet
Write down what the device must measure, over what range, with what accuracy, on which population, under what conditions, before you look at a single part. The useful form of that statement is uncomfortably specific: measure core body temperature between 90 and 108 degrees Fahrenheit, within plus or minus 0.4 degrees Fahrenheit, on adults and children over two, in ambient conditions from 60 to 95 degrees Fahrenheit, with a reading available within thirty seconds.
That sentence eliminates most candidate sensors immediately and exposes what datasheets hide. A thermistor with plus or minus 0.1 degree accuracy at one calibration point may be far worse across your full range. An optical sensor characterized on a benchtop phantom may behave differently on dark skin or a cold extremity. This specification becomes a design input in your design history file, and everything downstream traces to it.
The standards define what accuracy has to mean
For established measurements, a particular standard already dictates how accuracy is demonstrated, and it is almost never a bench comparison. Knowing which one applies before you choose hardware saves a redesign.
| Measurement | Governing standard | What it really requires |
|---|---|---|
| Pulse oximetry | ISO 80601-2-61 | A controlled desaturation study in human volunteers, reported as accuracy root mean square across the claimed range |
| Non-invasive blood pressure | ISO 81060-2 | A clinical validation against reference sphygmomanometry across a specified distribution of subjects, arm sizes, and pressures |
| Clinical thermometers | ISO 80601-2-56 | Clinical bias and repeatability established against a reference body site, not a water bath alone |
| ECG monitoring | IEC 60601-2-27 and related | Defined frequency response, noise, common-mode rejection, and pacemaker pulse handling |
| Glucose measurement | ISO 15197 and FDA guidance | Accuracy bands against a laboratory reference across the clinical range |
The recurring theme is that clinical accuracy is established on people, not on a bench. Budget for that study when the measurement falls into one of these families. It is frequently the largest single line item in a monitoring device program, and it is discussed further in our article on developing a vital signs monitoring device.
The interface to the body decides your real performance
In bench conditions, sensors are usually good enough. Performance collapses at the interface. The dominant error sources are almost always physical rather than electrical:
- Motion artifact. Optical and biopotential signals are small; motion is large. Mitigate mechanically first, with a stable mount and controlled contact pressure, then algorithmically. Adding an accelerometer purely as a motion reference is one of the highest-value additions available, and our guide to choosing an IMU for a wearable covers the criteria.
- Contact pressure. Too little and the optical path or electrode impedance is unstable; too much and you occlude perfusion and change what you are measuring. That is a mechanical target, not something left to how tightly the user pulls a strap.
- Thermal coupling. A temperature sensor reads its own junction, not the patient. Thermal mass, insulation from ambient air, and self-heating all shift the result; see our article on choosing a temperature and humidity sensor.
- Patient variability. Skin tone, perfusion, body composition, tattoos, hair, edema, and age all affect optical and impedance measurements. Test across a representative population before the design freeze.
Materials, cleaning, and sterilization survival
Any sensor that touches a patient becomes part of your biocompatibility evaluation. The relevant question is not whether the silicon is inert but what the housing, adhesive, potting compound, optical window, and cable jacket are made of, and whether the supplier will tell you. Many low-cost modules use undisclosed adhesives that make a toxicological assessment impossible. Our guide to ISO 10993 biocompatibility testing explains how contact type and duration set the required endpoints.
Reprocessing is the other filter. If the device or its sensor is reusable, the sensor must survive the cleaning chemistry, the number of cycles, and the sterilization method, with its calibration intact. Steam autoclaving destroys most consumer-grade optical and MEMS parts. Ethylene oxide is gentler but its residuals and vacuum cycles still matter. Gamma irradiation darkens some optical windows and shifts semiconductor characteristics. Choose the sterilization method and the sensor together, using our comparison of medical device sterilization methods, rather than discovering the incompatibility during validation.
Supply longevity is a regulatory problem, not just a purchasing one
A consumer sensor has a market life of a few years. A medical device may ship for a decade. When a sensor is discontinued, you do not simply drop in a replacement: you re-run the affected verification, potentially re-run clinical accuracy work, and file a change with your notified body or evaluate whether a new submission is required. That is why sensor longevity belongs in the selection criteria alongside accuracy.
Practical guardrails: prefer suppliers with a formal longevity program, ask for a written end-of-life notification period, avoid single-sourced parts with no functional equivalent, abstract the sensor behind a driver layer, and leave enough mechanical tolerance that a similar alternative can fit. Our article on handling a component that goes end-of-life covers the options once it happens.
Plan calibration and verification before you commit
Decide early how each unit will be calibrated in production, whether calibration is factory-only or field-adjustable, how drift is detected over the device's life, and what the device does when it can no longer trust its own reading. A sensor that needs a two-point calibration against a traceable reference adds real cost and takt time to every unit, and that belongs in the cost model before you choose it.
Then separate the two questions your test plan has to answer: does the device measure what the specification says (verification), and does that measurement actually serve the clinical need in real use (validation)? Sensor programs fail most often on the second one, having proven bench accuracy against a simulator and never against patients. Our article on verification versus validation covers how to structure both.
Projects House selects and integrates sensors for medical devices, covering the mechanical interface, signal chain, calibration strategy, and verification evidence. Describe what you need to measure through our contact form.