In this video, Walt Maclay discusses the issues involved with bringing wearable medical devices from an idea to FDA approval based on his firms many experiences. He explores areas including user interface issues, battery life, different sensing technologies, data security, and verification to meet FDA and CE requirements.
Below is an edited transcript of the video:
Building a Better Medical Device and Getting It Cleared Through FDA
Voler offers electronic design services for wearable devices and IOT devices. We have helped many firms build medical devices that have been approved by the FDA, but here I will be focusing mainly on wearable devices and some of the particular challenges in obtaining accurate physiological measurements, which are usually the key mission for a wearable medical device. I will explain how battery limitations are major drivers in the design of wearable devices and how to navigate the trade-offs often encountered with limited power. Finally, I will address data security and conclude with some observations on the FDA approval process.
Right now wearable devices are “hot”. Fitbit was launched in 2007, but since then new innovations have been coming along, being driven by the need for data. Data companies come to us and say, “We don’t want to build a piece of hardware, but we need data, so we will have to build one.“ Hospitals also want to keep patients at home because, if patients go back with the same condition within 30 days, the hospitals don’t get paid. But they also don’t want their patients to get sick while staying at home so what do they do? They monitor their patients at home.
Ten Common Physiological Measurements
Let’s take a look at ten common parameters doctors want to measure on a patient’s body, how that is done, and some of the challenges in making accurate measurements.
1. Body Temperature
Measuring body temperature seems straightforward since medical thermometers have been in use for more than two centuries. One difficulty is the doctor’s need for measuring body core temperature rather than skin temperature. Skin temperature, particularly on extremities like the wrist, can be different from body core temperature in a cold or hot environment and other additional measurements may be necessary. For example, if you know that the air temperature is cold, then it’s not a good time to use the value that you’re measuring on the skin. If the temperature is moderate, it’s a better time, but if the skin is wet because the person has been exercising, that’s not a good time either. The skin could also be wet for other reasons. An accelerometer could detect if the person has been exercising or moving rapidly from exercise. When you combine inputs from several different sensors, the measurement can be much more accurate. This is commonly called “sensor fusion”.
Contact with the skin is also important because the device is actually measuring heat flow from inside the body towards the outside surface, and if the sensor doesn’t make good contact, the temperature measurement will not be accurate. A wristwatch may not make good contact if it is worn loose; therefore how people use the device is important.
Motion has been popular for step counts, but it measures other things as well, and manufacturers of motion chips provide the necessary algorithms. For example, gait can be measured and can be an indication of various illnesses. A device on a wrist or ankle can determine if the person is standing, sitting or walking and the algorithms for wrist or ankle measurement already exist. Dead reckoning can be helpful to show a person’s actual position and movement direction. Motion sensor chips now have nine axes of motion detection in a single chip. They are good for short term position measurement but lose accuracy over time. If you need to know where somebody is for hours or days then you need GPS, but GPS is power hungry while motion chips consume very little power.
3. Heart Rate
Heart rate is a very important health parameter that can be measured in one of three ways:
- Basic ECG: Because a complete ECG is not necessary, heart rate can be measured with just two ECG electrodes, and even on the wrist it works fine. Dry electrodes are more convenient than wet electrodes, but they need to make good contact. This can be challenging, particularly on very dry skin.
- Pulse Oximeter: While a pulse oximeter is used for measuring oxygen in the blood, mainly it measures the pulse. The common approach shines light through a thin part of the body like a finger or earlobe, and is very accurate. The less common reflective approach reflects light off any body part but may be less accurate.
- Pressure Sensor: Another way to measure pulse is with pressure. If you put your finger on your wrist,you can feel your pulse, and a pressure sensor can also sense it.
4. Blood Oxygen
Pulse oximeters to measure blood Oxygen were developed originally in the 1940’s and the technology is well understood. Two infrared light sources pass light through the body. One of them is sensitive to the Oxygen, while both are sensitive to the pulse. When one is subtracted from the other, the pulse measurement is removed. The Oxygen measurement remains. The transmissive type has been in use for a long time. The light passes through a thin part body part like a finger or earlobe, or an infant’s foot. Other parts of the body are generally too thick for enough light to pass through; therefore the alternative is a reflective approach. The amount of light that’s reflected is much smaller than what gets transmitted and the sensitivity and the accuracy is often a lot less. The challenge is to make them accurate enough to meet FDA requirements.
5-7. ECG / EMG / EEG
ECG for the heart, EMG for muscles and EEG for the brain are all measuring electrical signals. This seems straightforward, but again you need electrodes, ideally dry electrodes, because applying a wet electrode with a wearable device is inconvenient. For ECG it’s important to have separation between the locations of the electrodes. On the chest I’ve seen one and a half inches work, and that’s small. If you take a look at the Apple watch, they’re measuring ECG on the wrist, but what you may not have noticed is that you have to reach over with the other hand to touch the watch. The measurement is actually from one hand to the other, far apart; therefore it doesn’t measure ECG continuously which is a limitation.
The standard is 12 leads for an ECG. which is what you get in the laboratory or when you go to the doctor. There are companies that now claim they are getting equally good results for detecting disease with just two leads. This would typically be on the chest, or alternately from arm to arm by reaching over to touch.
EMG measures the same sort of electrical signal but on muscles. A common place to measure is on the forearm. You can detect individual muscles in the forearm and detect which finger is moving from an EMG signal. However, if you don’t have an accurate reference on the arm and are off by a few millimeters, it’s sensing the wrong muscle, and you’re detecting the wrong finger. So how do you position something on the wrist that’s accurate within one or two millimeters? You could put an ink dot there, but that will wear away and will be a problem.
For an EEG, measuring brain signals, the only place to do it is on the head. The forehead or the temples are convenient places. There are helmets to do these measurements with a large number of leads. It would be nice if there were another location enabling the device to be worn for a long time. A hat is okay, but not typically indoors.
Respiration is another important parameter and the standard is to put on a chest strap. There are devices that do this, and it’s easy to measure. But a chest strap is inconvenient for many types of wearable device. Another way of making the measurement is with thoracic impedance. You’re actually measuring the changes in the impedance of the chest as you breathe in and out. This can be quite readily picked up, but the electrodes need to be on the chest. Attempts have been made to do it on the wrist but I haven’t seen this actually work reliably. The wrist has this big impedance called your arm between it and the chest so the signal is much weaker, and that makes this method difficult.
9. Blood Pressure
Right now the only way to get an accurate blood pressure measurement is with a cuff on the arm. There are devices that work on the wrist but their accuracy is sensitive to position and therefore questionable. Ideally you would like to measure it anywhere on the body without an inconvenient cuff. Today it’s commonly done with pulse transit time, measuring the time difference between the heartbeat to when that pulse reaches an extremity. This can be done on the wrist with two electrodes, picking up the ECG signal, which arrives instantly from the heart beat, and a photoplethysmograph (PPG) signal arriving when the pulse of blood reaches the extremity. Blood pressure is proportional to the change in the time difference. The difficulty is to get it accurate enough. There are some firms that claim they’ve gotten enough accuracy to meet FDA requirements for diagnosis, but I haven’t actually seen it. It works fine for an indication that blood pressure has changed. With calibration for each person the measurement can be accurate. Blood pressure taken with a cuff is entered into the device and it will give an accurate measurement from then on. There are firms working to improve the algorithm.
10. Blood Glucose
Glucose measurement is a very common need. Today there are devices with micro needles that are worn as a patch. They don’t actually feel like needles because they go no further than the outer layer of the skin, which is deep enough to pick up a glucose signal. Wearable patches require calibration with a finger prick once or twice a day because they are not accurate over a long period of time. A good perfusion of blood is necessary for measuring glucose and unfortunately the wrist doesn’t have that, although the place most people would like to wear a device is the wrist. Today there are sensors to measure blood glucose coupled with pumps to inject insulin and regulate blood glucose, so you can now have a wearable pancreas. There’s been a lot of progress there.
Batteries have been progressing, but slowly. If batteries had progressed like semiconductors over the last 50 years you would have a battery the size of the head of a pin that would cost a penny and would power your car. We’re not even remotely close and never will be. We’re limited by the space required for chemical storage of energy. Today’s batteries are about 10% of the ultimate, which would be something like gasoline. However, gasoline has a bit of a problem with safety. I don’t want to carry gasoline in my pocket. Another option that is more efficient is nuclear energy, but I don’t want to have a nuclear reactor in my pocket either. We are going to see incremental improvements in batteries in the future, but they will not become a lot smaller.
Managing The Trade-offs With Limited Power
Every time we design a wearable device or any battery operated device, people say:
- I want it to operate for a long time,
- I want to transmit lots of data a long distance, and
- I want the battery to be very small.
Of course, now we have to make trade-offs.
Power and Wireless Transmission
Let’s talk about the impact of the limited power. The way that people would like to send data from a wearable device is directly to the cloud like a cell phone. But how long does a large cell phone battery last? Obviously this isn’t practical for many applications. Another way would be to send the data to a gateway like Wi-Fi, which uses less power. However most wearable devices are in a third category where they use typically Bluetooth LE to send data to a cell phone in order to reach the cloud. The wearable device only works when a cell phone is nearby.
Wireless transmission has a big impact on power usage. In wireless transmission, there are three trade-offs that must be managed: the power required to transmit, data rate and transmission distance.
Here is a detailed summary of power requirements for popular wireless options. For a full description of the wireless standards, see our Whitepaper “Best wireless standard for your application?“.
To highlight one option, there is technology out now that enables low speed data to be sent long distances, but it is not widely available. Soon there will be wearable devices sending data directly to the cloud at nearly the same low energy level as Bluetooth LE.
How much power do sensors use?
Taking a look look at the first two here, a camera uses a lot of power, 300 milliwatts, which will drain your cell phone quickly, and cell phones have large batteries compared to most wearable devices. The illumination for night time use, would almost double that. The police wear cameras, but they require large batteries that are recharged after each shift.
GPS is moderately power hungry so you don’t want to run it all the time. A load cell which can measure weight or force is also moderate. Pulse oximeters are a little lower, but you don’t have to run them all the time. How often do you have to know what your oxygen level is? Maybe once an hour in some applications. EKG and heart rate? Now we’re getting to quite a low power. At one milliwatt, that’s 300 times lower than the camera. The nine axis motion sensor is also very low. Microphones can be extremely low or moderate. Light intensity is also extremely low to moderate.
But the winner in this is the three axis accelerometer which is what Fitbit had to begin with. It uses far less than a milliwatt, just a few microwatts. Devices are often designed to leave the accelerometer running all the time and shut everything else down. When the user moves, it wakes up the processor which takes more power than the accelerometer. The processor quickly determines if that movement was something important, and if it wasn’t it goes back to sleep. This is an important way to save power.
Data security is a big deal for wearable devices. The FDA weighed in on this in 2016, so it’s been a while and the surprising thing is that many people aren’t terribly worried about it. I would expect more of our customers to be more worried about data security. The FDA encourages end to end security in a guidance document. But a guidance document is not something to be ignored. The data security challenges for a wearable device are much higher than for an endpoint in a fixed location:
- The wearer can wander around and be almost anywhere.
- The device may not be the correct device.
- How do you determine if it’s authorized to send data?
- How do you know the device is on the right person?
- Has it been spoofed? Is there a different device sending data?
- Is it sending the right data? Is it sending it accurately, is it taken at the right time?
- And then the data gets to the cloud and you have to store it accurately and safely in the cloud.
Software Updates are the True Achilles Heel
It’s easy to have problems with these devices if they are not under the control of a medical professional. Many of these devices can have their software updated over the air. If the device is unable to detect where that update is coming from, I can update that device myself. I can put new software in it. Now it’ll do whatever I want it to do. That’s a serious breach of security and most devices today don’t do a good job in this area.
Fortunately there are companies that offer software and hardware that help you solve these security problems, so you don’t have to do it all from scratch. Secure RF has an encryption that runs on a tiny processor. Intrinsic ID and other companies use the random state of RAM at power-up to make a unique key. You don’t have to put the key in the device, it already has it, and every device has a different key. Companies like Secure Push can provide an end-to-end solution for sending data to the cloud, so it’s not as daunting as it might seem, but it still requires great care.
Conclusion: Some Observations on FDA Clearance
There’s an interesting thing that the FDA has done for apps that run on smart phones, and presumably watches. Apps that help people self manage their disease without providing a specific treatment and without diagnosing are not a medical device. Apps that help patients send data to their medical professional are not a medical device. Apps that help patients get their data from their health records are not a medical device. Apps that simply automate tasks for health care workers are not a medical device. However, if you create a wearable device that’s not a phone and it does the same thing, it may be a medical device. There’s definitely a gray area in the definition of a medical device. It’s very easy to end up making a medical device without realizing it. We help our customers determine if their device needs to be regulated by the FDA or CE.
Thank you for your time.