Medical device companies, like the hundreds that dot Minnesota, do not know everything about the safety of their products.

The Food and Drug Administration requires device makers to prove new devices are safe to use as directed, but there's only so much testing the agency can require before approving or denying a new product for market. That's why the FDA requires "postmarket" surveillance, which includes investigating all complaints about their devices and reporting confirmed problems publicly.

But researchers and regulators say the device-surveillance system is deeply flawed, exposing patients and doctors to greater risks.

Many device problems are never logged in the FDA's public system for adverse-event reporting known as MAUDE. And a study published last month by University of Minnesota researchers found that even when device companies do have the relevant information at hand, they are often biased in favor of waiting too long to issue product recalls.

"The reality is, if you look at the results, clearly there is under-reaction" to events that spur device recalls, said Kingshuk "KK" Sinha, a department chair with the U's Carlson School of Management.

Sinha's paper, published Jan. 29 in the journal Production and Operations Management, applied digital analytics to millions of medical device product reports and recall records and found a high "signal to noise" ratio correlated with delays in reacting to safety events.

In other words, the more adverse event reports there are for a ­particular device, the more likely a company will exhibit "under-reaction bias" in deciding whether to issue a recall, Sinha's research found. Older and widely used devices tend to have scores of reports in MAUDE, making safety-signal detection difficult.

"There is always something. But sometimes there is something more fundamental going on … the distribution [of reported problems] has shifted, and is no longer noise. It is actually a signal," Sinha said. "We should train our executives and decisionmakers just like fighter pilots, to make sure their reaction-time responses are optimal."

The Food and Drug Administration knows there's a problem, and has long sought to build a better surveillance system.

In 2012, the FDA announced plans to develop a national evaluation system for health technology, or NEST, to generate better evidence for regulation and decisionmaking throughout the total life of a device. The FDA recently awarded a $3 million grant to create the "coordinating center" that will build the public-private system.

"We have acknowledged that medical device reporting is an imperfect science, for a number of reasons," said Dr. William Maisel, deputy science director in the FDA's medical device division. With NEST, "the goal would be to create access to information that could then be tapped into either for routine surveillance or when there was a particular issue that arises."

But will analyzing more granular data allow for better analysis or generate more "noise" that can conceal a ­signal? A Jan. 25 report in the New England Journal of Medicine suggests such a system can indeed detect signals from noise.

That study, supported by grants from the FDA, used a data-mining program called DELTA to analyze a privately managed registry of outcomes from 146,000 people who had a minimally invasive procedures to unblock coronary arteries.

The study found that a device sold by Cardinal Health called the Mynx was associated with "a significantly greater risk of vascular complication" compared to alternative vascular-closure devices like Angio-Seal, formerly sold by St. Jude Medical.

The absolute risk of having any complication was 1.21 percent with the Mynx and 0.76 percent with the other devices. Such differences might appear small, but they apply to a large population — more than 1.8 million people were listed in the registry as having the type of minimally invasive heart surgery that uses a vascular closure device between January 2011 and ­September 2013.

Cardinal Health defended the Mynx' safety. "Patient safety is our number one priority. We have data that supports the safety and effectiveness of our vascular-closure devices. We stand behind our products," Cardinal Health spokeswoman Ellen Barry said in an e-mail.

A NEJM editorial recommended the FDA should issue an alert to doctors, but not place a "black box" warning on the device or remove it from market.

So far the FDA has not called for a recall or any change to the Mynx label.

The editorial writers noted that the Mynx has properties that make it a better choice for some patients, despite the difference in risks. They also listed standard defenses when a device is linked to lower clinical performance, including that doctors may be using it on more complex patients, and that there is a "learning curve" associated with doctors using the device correctly.

Such explanations make it even harder to use adverse-event reports in MAUDE as a source of data about whether to recall a device.

"Unfortunately it's going to be difficult to differentiate signal from noise so long as we have such a poor-quality data system. MAUDE, in my opinion, is at best a canary in a coal mine," said Dr. Joseph Ross, a Yale School of Medicine internal medicine doctor and associate editor with JAMA Internal Medicine.

But using registry data to find safety signals among vascular-closure devices may be low-hanging fruit.

Dr. Zian Tseng, an electrophysiologist in San Francisco, has led a six-year project to examine each sudden cardiac death in San Francisco. The study team has found that 5 percent of the people who died of sudden cardiac death in the city between 2011 and 2013 had pacemakers or defibrillators, which should have worked to prevent their deaths. Half of the devices (11 total) showed problems like prematurely depleted batteries, busted lead wires, and failure to detect unusual heart rhythms.

None of the devices with problems in the study has been recalled, Tseng said. Because full autopsies are rarely performed for suspected cardiac-related deaths outside a hospital, such reports typically would not be entered into MAUDE or electronic databases.

Tseng found it surprising that some of the companies didn't seem interested in his findings until they were published in 2015 in JAMA Internal Medicine. "Some were very dismissive. Others were very appreciative," Tseng said. "It was startling, the range of responses we got."

Joe Carlson • 612-673-4779