More than 3 million (mostly senior) consumers own medical-alert devices — panic buttons worn on the wrist or a pendant — that can call for help if you fall or suffer another emergency and can't reach a phone and 911.

But if you are relying on, or thinking about buying one of these devices, you need to know what happened when Consumers' Checkbook tested 11 of them with the help of the Arlington County, Va., Emergency Communication Center, which acted as a neutral testing site and doesn't endorse any products or companies.

Checkbook found disturbing deficiencies. Most devices delayed getting us to 911. Some sent false alarms. And when the researchers called for help, one device "found" them … in China!

Not good if there were true life-or-death emergency. And getting 911 help is a major selling point in ads for these gadgets.

In most cases, pressing the alert button doesn't call 911. Instead, that connects you to a company call center. Checkbook tested how long it took those monitoring services to answer by pressing the button on nine devices a total of 290 times at various locations.

"Response time is absolutely critical for effective emergency response, and it's more critical the more serious the emergency," said Christopher Carver, director of Public Safety Answering Point/9-1-1 Operations for the National Emergency Number Association.

That's why, at peak traffic, government 911 centers aim to answer 90% of calls within 10 seconds. Call centers for only three devices answered in 30 seconds or less, on average. Three others took more than a minute. In individual trials, some took as much as 2 to 3 minutes.

A 1-minute wait during a medical emergency is a long time. Two or more minutes is an eternity.

Real 911 dispatchers are trained to quickly and directly interrogate the caller to assess the situation, determine the victim's location, and decide who to send to help. They are also trained to instruct callers to begin CPR or stem bleeding until medics arrive. Most of the alert devices studied delay your ambulance by inserting typically lesser-trained agents between you and the top guns of emergency dispatch.

"Any addition to the time of response has a negative impact on the outcome," said Carver.

In addition to slow response times, in Checkbook's testing, medical-alert companies often had trouble determining where the calls were coming from — exactly or even within a reasonable margin of error. That jibes with the complaints of 911 managers, who say they often "were not able to get a good location, or the location was from an hour before, or there was no phone number, so we couldn't call them back," said Angelina Candelas-Reese, 911 systems manager for Arlington County, Va.

In testing, the researchers asked each operator where the device pegged their location. Some were usually spot-on or reasonably close, while others were blocks away or had no clue. On one call, Bay Alarm Medical's call center "found" the caller in China. In general, these devices also have trouble locating callers from high-rise buildings in urban areas: To which of 10 or more floors should first-responders go?

Among the 911 community, medical-alert devices are also notorious for false alarms, which can needlessly delay responses to real emergencies.

Some of the devices tested sent false alarms. On a few occasions, paramedics showed up at Checkbook's office, ready with gurney, oxygen, defibrillator, the works.

Because of the problems found, Checkbook could not recommend most of the devices tested. But if you insist …

• Consider the GreatCall Lively Mobile, the only device tested that can call local 911 directly, bypassing the company call center. It had the fastest average response time. It also uses recent advances in cellular 911 location technology, which can deliver greater precision and promises to dramatically improve emergency response times.

• Shop first on the considerations raised here, secondarily on price, which ranged from about $130 to $625 the first year for device and monitored service.

• Test response time and location accuracy during a 30-day free trial by hitting that button a dozen or more times at different hours, days and locations. (Device makers recommend such testing.) Unacceptable performance? Return it.

• When you get your device, complete your customer profile with as much detail as possible about medical conditions and prescriptions, including dosage and times per day. The more call center operators know, the better they and emergency responders can help.

• Just say no to the fall detection option.

Twin Cities Consumers' Checkbook magazine and is a nonprofit organization with a mission to help consumers get the best service and lowest prices. We are supported by consumers and take no money from the service providers we evaluate. You can access all of Checkbook's ratings for free until July 7 at