There are two perfect examples of fully automated IoT medical devices: a pacemaker and a closed-loop insulin-delivery system. The first, a pacemaker, is a surgically implanted device that helps to control heartbeat when there is an arrhythmia (irregular heartbeat). However, the new generation of pacemakers uses an IoT architecture system, where embedded sensors monitor a patient’s vital signs (breathing, sinus node rate, and blood temperature).
When an irregularity is detected, the patient’s heart rate is altered (slowed or speeded up), depending on the patient’s current activity level. In addition, patients are now able to access their data through a mobile device to check device battery life and any correlations between their heart pace and activity level. In the past, this required an inquiry to their physician.4 The second example of a smart medical device is much more complex because of the necessity of around-the-clock treatment of T1D.
The closed-loop insulin-delivery system, also known as an artificial pancreas or a bionic pancreas, uses an IoT architecture to control insulin delivery, which, as mentioned earlier, is needed for patient survival. To appreciate this amazing IoT device, some knowledge of T1D care is necessary, given the risks of automation. The greatest risk with this type of device is that if too much insulin is dosed, the patient’s life is on the line (from low blood sugar). If too little is dosed, the patient’s organs could be damaged (from high blood sugar), and the risk of other autoimmune diseases could be increased. When patients with T1D use needle therapy (no pumps), they require two types of insulin: long acting (also called basal) and fast acting (also called bolus).
The challenge is that the amount of insulin one needs throughout a day varies and depends on many factors. To keep a person’s blood sugar level in a normal range, the dosed amount is dependent upon the amount and type of food consumed, body size, hormone levels, activity, current health status, time of day, current amount of insulin in the body, and even the weather at times Also, keep in mind that most T1D patients are diagnosed at an early age, so imagine the difficulty a caregiver has in calculating/predicting a child’s activity level.
A patient with T1D can dose the needed insulin via syringes/insulin pens, which require manual predictions and calculations based on the aforementioned factors, or they can dose using an insulin pump and infusion set—a tiny catheter injected under the skin to deliver insulin (replaced every three days). The patient programs the pump to calculate the insulin dosage, but the pump still requires manual input to dose when eating or to correct high blood sugar. In other words, the user would manually enter the blood glucose reading into the pump from a finger prick or a CGM device.
How does the pump become smart? Both devices (CGM and pump) are worn by the patient, and the CGM wirelessly sends the glucose level directly to the pump. That data connection closes the loop to deliver some of the needed insulin automatically. Specifically, the CGM number is received by the pump, and the pump uses an algorithm to detect when the glucose level is rising or falling.
The interstitial CGM reading will lag behind the actual blood glucose readings as it takes time for the glucose level to reach the interstitial fluid. Therefore, algorithms within the pump software account for this lag by interpreting the steepness of the slope as the numbers rise or fall. Then, depending on the glucose trend, the pump will automatically either release bolus insulin to address a spike in blood sugar or scale back on the basal insulin when the blood sugar level is dropping.
Note that since only one type of insulin can be stored in the pump, the fast-acting insulin is dosed to the patient at prescribed intervals to give the “same” results as if the long-acting type of insulin were dosed once by a needle. However, the patient still must bolus insulin for food as the insulin needed will vary based on how many carbohydrates are in the food (another complexity as not all carbohydrates digest the same way). This system still requires manual entry when the patient consumes food.
These systems are getting smarter; a recent addition is that the status of the pump can be viewed from a mobile device and shared with a caregiver. This provides amazing peace of mind to a parent caring for a child with T1D. Imagine the significance of this feature during the overnight hours or when the child is away from home or on a soccer field. The parent receives an alarm and can notify the child’s chaperone.
The newer systems also have settings to address activity and sleep as the patient may want less insulin to avoid a dangerous low. Some of these devices also have algorithms to predict blood sugar levels as much as 30 min in advance to begin adjusting the basal insulin setting or bolus for a correction of high blood sugar, all based on readings from the CGM.
The closed-loop, insulin-delivery system is effective. However, it could be completely automated in the future if it could address the heaviest burden of T1D care, which is not only meal reporting but maintaining a normal blood sugar range after heavy activity and especially meals. In addition, some adolescents frequently forget to bolus for meals.6 Untreated meals and miscalculated carbohydrates (the calculation is typically an educated guess) lead to hyperglycemia (high blood sugar), and overdosing based on an incorrect carbohydrate count could lead to hypoglycemia (low blood sugar).10 Thus, as stated in the beginning of this article, researchers are on the threshold of developing a completely automated IoT insulin-delivery system by adding a module to detect unannounced meals, thus eliminating the manual guess of carbohydrate entry at every meal.
Researchers are investigating using CGM data to automatically detect meals. They have developed several systems and are currently testing algorithms and methods to detect meals based on variations of CGM readings.6,8,10,13,14 Some algorithms were tested with simulated data, and some were tested on patients and have reported improvements of glucose control post meal. The challenge continues to be the lag in glucose level getting to the interstitial fluid. Ideally, a noninvasive way to detect glucose level in the blood quickly is needed.