Categorizing & Comparing Examples of SaMD

In the recent past, we reviewed examples of software as a medical device (SaMD) as well as how SaMD is categorized in two separate posts. Now we are going to put these together, outlining how some real world examples of SaMD are categorized and comparing how they differ. We’ll also take a quick look at how SaMD that uses artificial intelligence is categorized.

Software as a medical device is currently categorized into classes from IV to I. These categories are based on how vital the information provided by the SaMD is to the patient or to public health, with Class IV having the largest amount of impact and Class I having the least. One thing to note is that if SaMD can be used in various healthcare settings or situations, it will be categorized by whichever is the highest level.

One rule of thumb is that if the SaMD’s main function is to inform, it will likely be a Class I or II, depending on how serious the information is. SaMDs that treat or diagnose will likely be categorized as Class III or IV, again depending on the seriousness of the situation.

Looking at the function or purpose of SaMD is not enough to categorize it. For example, 2 different SaMDs could generally do the same thing, such as create a treatment plan. It’s what the plan is treating that makes the difference in categories. SaMD that analyzes and develops a treatment plan for a cancerous tumor would likely be categorized as Class III or CLass IV, while another SaMD that creates a treatment plan for acne would likely be in Class I.

Another factor that can change the category for SaMD is the condition of the ideal patient, or the progression of the illness or disease being treated or monitored. For example, there is SaMD that monitors vitals and other data while a patient sleeps. A use case with a lower class would be otherwise healthy patients who have sleep apnea. This would likely be categorized as Class II. If, however, the ideal patient being monitored was someone who recently went through great physical trauma, the class would likely be increased to a III or a IV.

One way to categorize SaMD is to look at the potential outcome if the information or data provided by the SaMD is wrong. If a patient would be subject to serious injury or death, it would likely be given a Class IV. If the software’s purpose is to calculate a patient’s BMI, even if the information is wrong, very little, if any, harm is done. This would fall into Class I or II.

Artificial intelligence (AI) in SaMD is a newer breakthrough in med tech. The first authorization of an AI-backed SaMD was just recently given by the FDA in February of 2020. This was cardiac ultrasound software in which the documentation contained a Predetermined Change Control Plan, which incorporated future modifications learned by the AI over time.

As of today, there still are no set guidelines for AI-backed SaMD, although progress has been made. They are still generally being treated as any other SaMD.

The future of software as a medical device and the use of artificial intelligence in healthcare is quite astounding, and developers have really only seen the “tip of the iceberg”. As technology progresses and we learn more, I think new breakthroughs will happen exponentially.