This blog post is the second in a series where we’ll be sharing some statistics around medical device recalls, adverse events, and reported product problems to explore what kinds of analysis one can do by linking different medical device data sources.
Part I: Top 20 Adverse Events
Are the top most-frequently recalled medical devices the same as those with the most adverse events?
To continue exploring the topic of reported product problems we took medical device recalls over 18 years from 2002 to 2019 counted the number of recall instances for the type of device.
“The FDA uses the term recall when a manufacturer takes a correction or removal action to address a problem with a medical device that violates FDA law.”- more information.
Similar data preparation challenges applied in mapping the medical device recalls to distributed device types as they did for the adverse events mappings from Part I, namely:
The main difference in analysis occurs in the mapping step between DIs and recalls. The recalls mapping was not as straightforward as initially expected.
Historically recalls are split into two files by the FDA:
The Recalls database has 59k recalls dating back to 2002, and will list the associated premarket submission number if available.
The Recall Enforcement Reports database has 19k recalls dating back to 2012, and will list issuing company, and product code info in a text blob field which may contain a mixture of Unique Device Identifiers, brand names and descriptions, part numbers, and serial/lot numbers.
In an ideal world all recalls would have Unique Device Identifiers submitted in a separate data field with a one-to-many relationship between recalls and the list of affected devices, making it easy for supply chain analysts to map device data.
That said, given the recency of UDI legislation starting around 2014 and the text blob field, we used the submitted premarket submission numbers instead. These are available on both GUDID device identifiers and Recalls dataset records.
This process allowed us to map 19k (~33% of total) of historically issued recalls to active device identifier records within GUDID.
One caveat to note is that premarket submission numbers have one-to-many relationships to device identifiers. These relationships could result in the same recall mapping to multiple devices. To address the possible repetitive mappings, we normalize the list of recalls by the number of devices within the device type category.
To answer our initial question, we see that 10% of the top 20 device types overlap between historical device recalls and reported adverse events.
Four areas of further analysis which would add more clarity to the overall picture would be to:
This blog post is the first in a series where we’ll be sharing some statistics around medical device recalls, adverse events, and reported product problems to explore what kinds of analysis one can do by linking different medical device data sources.
Are there medical devices with inherent design characteristics which make them a higher risk to patients than others? If so, which criteria define these types of devices?
To start exploring this topic we decided to take adverse events reported by manufacturers, providers, and volunteers over a 28 year time period from 1992 to 2019 to the FDA's MAUDE database and normalized the number of events based on the total number of currently available devices within a specific device category.
There are various challenges in preparing this kind of analysis, namely:
We addressed these by first defining a device as a device identifier or DI submitted to GUDID. These represent unique codes for a device make and model, so, for example, an iPhone 6s would be one device, and iPhone 7 would count as a second device.
We then categorized the device identifiers using the Global Medical Device Nomenclature (GMDN) terms submitted for each device identifier by manufacturers to GUDID. These terms group devices based on similar characteristics, and in the iPhone example both the 6s and 7 models would fit in the Smartphones category.
The last piece required reasonably mapping device identifiers to reported adverse events. We base these mappings on a combination of the company name, brand name, catalog number, device identifier, premarket number, product code, and device description. We considered events meaningful where the reporter indicated that a product malfunction, patient injury or death occurred.
The mapping process was easiest when reported events included device identifiers, and the hope is that with increased adoption of Unique Device Identifiers (UDIs) provider and manufacturer usage, more reports would contain device identifiers.
Taking a step back, we recognize the crucial missing context around the usage of these devices over the same time period. The usage would be one of the necessary data pieces to determine whether an event matters. For example, it wouldn't be surprising to find surgical staplers and glucose monitors having higher usage with patients than artificial hearts.
One way to track the usage would be for health insurance payers like Medicare encouraging the inclusion of UDI’s (or even DI's) on submitted charges and having public access to these usage datasets.
With these disclaimers in mind, we can see three tentative groupings within the top 20 device categories:
An area of further analysis would be exploring comparable categories of lower-ranked devices for similar relative event activity.
The data sources used in this series of analysis include the FDA’s Global Unique Device Identification Database GUDID, Manufacturer and User Facility Device Experience MAUDE, Medical Device Recalls, and Recall Enterprise System RES.
If you have feedback, are interested in receiving the full list of events by category, or would like to be notified when we post the next part of the series, please feel free to reach out to us below.Join the journey to analyze healthcare supply data, follow-us on LinkedIn.