Medical devices that utilize computer software are becoming common place in today's health care, ranging from insulin pumps, to devices that dispense drugs, and those that monitor heart rhythms. Many medical devices have been used successfully to provide better patient care, often with costs savings, and in most situations actually address some areas of human driven errors (Berman, 2004).
At the same time software to program and/or control medical devices can, and have, introduced errors that affect patient outcomes. The errors are in part due to the software or a failure of the interface between the hardware and software in the medical device, and the person using the device. Errors in medical devices raise questions concerning the strength of the design and testing of human computer interactions when developing medical devices. Sufficient human computer interface testing should ensure that most errors are identified before products are released.
The FDA has an approval process for medical devices with differences depending on the criticality level of the application of the medical devices. It appears that human computer interaction testing is not a clearly required component of the FDA medical device approval process. We believe that human computer interaction testing should be required before medical devices are approved by the FDA. The challenge is in specifying criteria to help a manufacturer determine what is "sufficient" human computer interaction testing.
Medical Device Failures
A few examples of problems with the human computer interface of medical devices will provide some background for this paper.
The Therac-25 device was developed to provide radiation to cancer patients. The presence of numerous flaws in the software led to massive radiation overdoses, resulting in the deaths of three people (Leveson, 1993). It was poor user interface controls that caused prescription and dose rate information to be entered improperly (McQuaid, 2009).
A problem with a flow control knob is cited by the FDA (Sawyer, 1996): "A physician treating a patient with oxygen set the flow control knob between 1 and 2 liters per minute, not realizing that the scale numbers represented discrete rather than continuous, settings. There was no oxygen flow between the settings, yet the knob rotated smoothly, suggesting that intermediate settings were possible. The patient, an infant, became hypoxic before the error was discovered. Human computer interaction testing of this device should have identified this problem."
A volumetric infusion pump is a medical device that delivers intravenous fluids and medicine to patients. The Baxter Colleague triple channel infusion pump generates fault codes under the condition of changing a fluid supply at the same time the supply goes to zero (an understandable and appropriate behavior). Unfortunately this fault also stopped the other two channels from continuing to operate, causing a life threatening situation for patients. At least 9 patients' deaths have been attributed to miscommunication between the care giver and the software that runs these medical devices (Infusion Pump Recall, 2009).
The Journal of the American Medical Association (Koppel, Metlax, Cohen, Aboluck, Localio, Kimmell, & Strom, 2005) reported on an examination of a hospital computer system and suggested that computers increased the error risk. They said, "... the problem is that hospital computer systems are not designed for the way real-life hospitals work." It appears that their recommendation is for more human computer interaction testing.
Further examples can be found in an article published by Dick Sawyer (1996). And the complexity of user-computer interaction is growing: The deployment of medical devices to the field such that "Caregivers and clinical engineers ... are becoming lost in a swirl of technology, and [we] face unanticipated interference between devices" (Lee & Pappas, 2006). …