Life Sciences Blog

Recall Prediction Model – Real World Performance

by | Mar 24, 2021

We began a project in early 2020, asking the question “With the available history of adverse event reports, medical device recalls and in some cases, reported patient problems, can these data points be modeled to help understand if recall events are predictable in nature?

Leveraging publicly available data from the FDA, we are training a proprietary machine-learning algorithm to ‘predict’ recalls. The results are promising, providing some interesting insights.  

When we look back at the starting baseline to now, some very compelling results are becoming clear. On May 28th, 2020, our recall prediction model predicted which medical devices would be at risk of having a recall in the next 365 days. At that date, we had 104,356 class II & III medical devices in our database. Out of these 104,356 devices, the model identified 13,359 products that were the most at risk of having a recall.

Between May 28th, 2020, and March 03, 2021, we have seen 520 new recalls linked to the original 104,356 products. Out of these 520 new recalls, 435 of them (83.7%) were linked to products that were originally identified as being at risk of recall.

To get a visualization of the results of the modeling, we show a simple diagram.

  • Each square represents 100 products (each square, red and green).
  • The red squares represent the products that have been recalled.
  • Those squares within the orange outline show the subset of products the predictive recall model identified as carrying a high risk of recall.

When we study the activity related to adverse events in the last decade, medical device recalls have increased. What are the main reasons for this? Recalls have cumulatively increased, but the rate of change has remained relatively constant:

Our hypothesis is simply that with more devices entering the market, we will continue to see more recalls. We expect newer devices are more likely to have recalls than devices that are established in the market.

Top 10 Reasons for Medical Device Recalls

Additionally, there are patterns in the data that pinpoint the most common reasons a medical product has to be recalled. According to the data, the most common root cause is device design followed by Nonconforming Material. The Top 10 root causes are listed below:

What are the main challenges when trying to determine the safety and quality of a medical device?

One of the ways to determine the safety and quality of a medical device is to look at historic data such as adverse event and recall reports. The main challenge is with determining which variables increase the probability of a future safety event. It’s also challenging to connect events from the event databases to the products in the approval and clearance database.

Recalls may be inevitable, but the impact can be mitigated by being aware ahead of time

Having access to tools that can identify products that are more at risk and focusing more on these products may prevent some of the recalls. By using predictive recalls methodology, manufacturers can have improved insight for pro-active actions to protect patient outcomes.

Learn more about the Predictive Recalls methodology, analysis and current insights by downloading the white paper.

Questions? We are here to help. Visit our Knowledge Center or Contact Us.