Abstract
Over the last decade, the use of computers and robotics in medicine has increased commensurate with emergent advances in technology. This article largely focuses on the challenges that the U.S. Food and Drug Administration faces when evaluating new technologies for entry into the market. How different categories of devices are categorized and what types of data have been used for regulatory approval or clearance are described. These are compared with expectations that the clinical community may have for these devices. A brief discussion of current regulatory thinking about these types of devices is also included.
The use of computers and robotic surgical tools is emerging in orthopaedic surgery as an aid to surgeons; however, as with any new technology, there are challenges to discovering the true value of the new devices to the patient and the surgeon. Clinical and regulatory validation of these devices adds another dimension to the questions surrounding these types of devices. The sources of the challenges that regulators face include nomenclature, regulatory precedent, clinical validation parameters, consensus standards for testing and characterizing devices, software and adaptation issues, hardware compatibility, and indications for the use of various devices.
The first hurdle to overcome is understanding what is being discussed. Computer-assisted orthopaedic surgery is known by many names in different medical arenas. The terms used to refer to the instruments and machines—computer-assisted orthopaedic surgery, computer-guided surgery, image-guided surgery, surgical navigation, robots, and stereotactic devices—are used interchangeably and mean different things to professionals, industry, and regulators. At the U.S. Food and Drug Administration (FDA), many computer-assisted devices are termed stereotactic devices1. However, in the practice of medicine, the term stereotactic often refers to the mechanical frames that are used during neurosurgery for precisely locating neural points of interest. To further confuse things, similar devices, such as breast biopsy2 and radiation therapy devices3 that use stereotactic guidance technology, are in a different regulatory category because of different indications for use, although they may share similar technological characteristics to stereotactic devices. The computer-assisted orthopaedic navigation systems, which are infrared or electromagnetic field-based, are the systems more appropriately termed as computer-assisted surgery or computer navigation systems; however, they are also currently termed as stereotactic instruments by regulation1. These stereotactic instruments also include devices that are used to assist surgeons in a variety of medical specialties including, but not limited to, ear, nose, and throat; neurology; and trauma applications. These types of devices can be used for both preoperative planning and real-time intraoperative feedback for guiding a surgical procedure.
A second group of computer-assisted surgery devices are fly-by-wire systems that mimic a surgeon's movements much like an airplane's control system mimics the pilot's movements to move the control planes of the aircraft, which is how they acquired their label. At the FDA, these fly-by-wire systems are mostly grouped together and referred to as diagnostic devices, including endoscopes and accessories4. These may include robotic arms that require a surgeon's input to perform a directed portion of a surgical procedure.
A third group of computer-assisted surgery devices are designed to perform autonomous clinical actions under the indirect supervision of the user. Such robotic systems or robots contrast with the other computer-assisted surgery systems that only aid or mimic the user's clinical actions. Robotic systems are not currently classified by regulation.
Another type of device that may be confused with medical devices used to treat patients is the surgery simulator. These types of computers, which may involve "virtual surgery games" for the purposes of training or preparing surgeons for patient care and to develop research, also serve to further confuse the issue. However, computer simulations are beyond the scope of this discussion, which is limited to devices that are used in direct patient care.
The majority of these types of devices are regulated through the premarket notification pathway to market—the 510(k) process5. Most are classified as class-II devices, with some exceptions. The class-II devices include those that fit under Section 876.15004—endoscope and accessories, such as the fly-by-wire systems that usually require clinical data—and those devices that are classified under Section 882.45601. Stereotactic instruments, which include preoperative planning devices, stereotactic mechanical frames, and computer-assisted or navigation systems, require preclinical data and can also require clinical data depending on the indication and the technology. Currently, these types of robotic devices may be marketed through either the 510(k) or premarket approval pathway and would require clinical data in both cases for clearance or approval.
In summary, most of these stereotactic and navigation devices, as well as endoscopes and accessories, fit into a class-II designation, requiring a 510(k) application. Devices that collect diagnostic information, make a prognosis, and initiate a clinical action under the direct supervision of the physician may require clinical data. Devices that autonomously initiate clinical action on the basis of information gathered without direct physician intervention could require a premarket approval application and will require clinical data. We suggest contacting the FDA regarding each particular device and its intended use. Table I summarizes the regulatory classes for different types of computer-assisted devices and the types of general data categories that may be requested to support a marketing application.
Challenges
Orthopaedic surgeons are finding ever-increasing utility for computer-assisted devices for patient outcomes research, for limb realignment procedures, for improving the accuracy and reducing the variability of prosthetic placement, for taking into account sophisticated soft-tissue balancing factors, for fracture fixation realignment, and for a growing variety of reconstructive procedures and challenges that confront them every day.
The dilemma is how to determine whether these machines do any better than a conventionally trained surgeon, what prosthetic alignment parameters are optimal, what outcomes are important, and what kinds of questions we need to answer to show safety, effectiveness, and clinically meaningful benefit.
The biggest challenge is deciding what types of data are needed to validate each of the systems clinically in a meaningful way. This is due to the fact that, in each specific device, the complexity of the technology is such that multiple variables, definitions, and functions are possible. This makes it difficult to come to a consensus about what standards can be developed logistically to characterize even the different device subtypes into logical groups by their characteristic functions.
For the devices that are intended to improve the accuracy and deviation of prosthetic placement, for example, several factors come into play: individual patient anatomy; differing imaging and/or palpation processes to acquire data; differing definitions of reference coordinates; differing surgical approaches and procedures used to implant different devices; different prosthetic types, anatomic locations, and variations of devices within an anatomic location; and varying technologies and/or functions of each navigation device.
In image-guided surgery, preoperative medical data are used to plan, simulate, guide by means of real-time feedback, or otherwise assist a surgeon in performing a medical procedure. In preoperative planning, data sources may include computed tomography, magnetic resonance imaging, or plain radiographic images. Each of these modalities provides differing types and degrees of accurate input, which require different software validation and coordination with the guidance devices. In addition, each modality made by different manufacturers may have slightly different basic software validation and content, further complicating the ability to consolidate standards for evaluating the accuracy and efficacy of the input information. To improve the use of medical imaging, standard formatting that allows the computer-assisted systems to input and recognize data sets regardless of manufacturer or modality potentially can be used. However, the accuracy and quality of the different images will still vary. In addition, the relative orientation of the image to the patient's coordinates requires registration with the patient's anatomy. If real-time registration is desired, such as in the case of fluoroscopy, the computer-assisted system must instantly identify the orientation of the fluoroscopy unit relative to the patient, which may affect the accuracy of the incoming data and ultimately the accuracy of the navigation. One advance to mitigate this risk includes the development of modality and manufacturer-specific tracking accessories for the imaging unit that allow the navigation system to track the imaging system's position relative to the patient in real time.
Using the preoperative patient and imaging data, a preoperative plan that specifies how one or more tasks are to be performed during surgery, and even a simulated procedure, may be constructed. In some cases, the preoperative plan is calculated or "optimized" on the basis of the system's algorithms. There are a variety of established elective and trauma procedures and multiple surgical approaches that a surgeon may choose from to perform a procedure6. The surgical procedure is then performed within a coordinate system relative to the patient in the operating room. In some devices, this surgical registration is the process of establishing a transformation between the preoperative data and plan and the actual patient. In other devices, no preoperative imagery is required, and generalized, built-in anatomic software models are transformed to match the data acquired during the surgical registration process. There are so many complexities of patient anatomic structure, and variability in the definitions of the planes of the body, that a standard is difficult to define. One way to accommodate individual patient anatomic variations is intraoperative registration, which allows the three-dimensional coordinate system of the preoperative plan to be correlated to the patient's anatomy in real time. In imageless systems, the generalized anatomic model surfaces are altered to correspond to the patient's anatomic surfaces as they are palpated intraoperatively with a navigated instrument. Surgical execution is performed with use of either passive methods, in which the surgeon is guided by information from the preoperative plan, or active methods, in which a semiautonomous device, such as a robotic arm, performs surgical tasks under the supervision and direction of a surgeon and can be adjusted as the procedure warrants. The vast majority of these systems allow for the user to convert from a navigated procedure into a traditional procedure at any point in time6-9.
Finally, there are hundreds of implantable devices, each with its own implantation and dimensional specifications that are based on technology that may become obsolete or incompatible with certain navigation software. This makes it difficult to determine whether the accuracy and safety of the navigation device are altered by the different nuances in the specifications for each implant, especially in systems that support the navigation of a generic "universal" implant. Couple these difficulties as described, and one can see the roadblocks to defining and easily characterizing navigation devices into discrete categories for the purpose of developing consensus standards or evaluating device safety and effectiveness.
The professional community has high expectations for computer-assisted surgical and navigational devices; these expectations have been fueled by some successful anecdotal outcomes as measured by surgeons for their individual patients. However, there have been relatively few randomized, controlled clinical trials that have compared outcomes between navigation and conventional surgical technique7,8. Although there are some promising data, the cost, utility, long-term effectiveness, and learning curve skew the risk-benefit curve against the use of a navigation device for many surgeons.
What is expected by the clinical community may not be answered easily by bench data; however, in order to compare devices within a regulatory framework, there have to be ways to acquire and use bench data as well as clinical data to show substantial equivalence. Often the challenge for regulators and industry is finding enough similarities to compare "apples to apples." As has been described, no consensus standard has been developed for comparing these types of devices. Bench data that may be used to characterize a device include computer software verification and validation, general positional accuracy and repeatability, specific accuracy and repeatability as it relates to a certain function or calculation, and compatibility verification for a specific prosthetic device and navigated instruments. In general, bench testing can be used to demonstrate within an idealized environment that the system can accurately and reliably perform the claimed functions. What bench testing lacks is the ability to document the robustness of the system when subjected to the variability of a clinical environment. Ideally, a particular implant system would have established alignment criteria that are strongly correlated to a positive clinical result. In such a case, clinical data may not be needed as a correlation between bench testing and clinical end points could be suggested. Often, cadaver studies are useful to simulate the anatomic variability that can be anticipated in real-world clinical settings. Depending on the type of device, electrical safety testing, electromagnetic interference testing, sterilization validation, and biocompatibility testing may also be used. For certain indications and claims, clinical data, including human factor (user) evaluations, feasibility of use (surgical workflow), precision of prosthetic and limb alignment measurement, and prosthetic fit-and-fill information, may be helpful. In cases where devices are designated as having a moderate-to-severe risk, clinical trials that collect data related to pain and function, as well as to the surgical accuracy of placement and limb or spinal alignment in comparison with a control group, are of particular interest to the clinical community.
In the absence of standards, additional data that may be considered as supportive of a marketing application may include, but are not limited to, the length of a surgical procedure, a description of potential and known anticipated and unanticipated adverse events, accuracy of performance (linear and angular accuracy or fit-and-fill accuracy), conversion to conventional surgery rates, surgical outcomes, training requirements, an estimated learning curve, human factors that influence the devices' performance, and clinical outcomes with regard to pain and function.
The data we do have are mostly based on cadaver feasibility studies. While there is merit in these studies, there are economic, clinical, and cultural limitations to relying on these types of studies for clinical validation of computer-assisted device software programs. The use of cadaver studies to validate the precision of alignment is limited by the tissue available, and the concept of precision itself is controversial. For example, is precision within 2 or 3 mm acceptable or do we need a precision of 1 mm? How do we know this with certainty, except by theoretical convention without longer-term data? Even the more economical testing with use of "phantoms" or mechanical jigs to validate systems has a limitation in that each has to be tailored to its own navigation system and may not be applicable to other navigation systems or even to all of the features contained within one system. Other types of data that may be used to support marketing applications may depend on the intended use. Case studies, cadaver studies, and bench data may be included. Supplemental collection of intraoperative data in real time and then the analysis of the data postoperatively with comparison to the manual surgical results represent data that, although observational, could be used to verify the planning portion of a computer-assisted device.
Objective measurements of performance that can be used to compare different systems are difficult to establish. Some of the questions that are important to both surgeon operators and technicians that should be considered when developing these measurements include the following:How accurately do the system's actions reflect the operator's intent?Does the device enhance surgical capabilities?How does the computer "intermediary" affect operative flow?How easily are glitches in the technology handled?Are there procedures and/or anatomic sites that are more or less amenable to this technology?How steep is the learning curve?What criteria are used to "optimize" a preoperative plan?How and when should the user override the recommendations of the system?How does the system define anatomic references and axes?How does the system dictate the surgical workflow, and how does this differ from traditional techniques?
How accurately do the system's actions reflect the operator's intent?
Does the device enhance surgical capabilities?
How does the computer "intermediary" affect operative flow?
How easily are glitches in the technology handled?
Are there procedures and/or anatomic sites that are more or less amenable to this technology?
How steep is the learning curve?
What criteria are used to "optimize" a preoperative plan?
How and when should the user override the recommendations of the system?
How does the system define anatomic references and axes?
How does the system dictate the surgical workflow, and how does this differ from traditional techniques?
However, while these are important questions to ask, they are more difficult to answer objectively. Technical characteristics are more easily provided for comparison. These may include, for example, software verification and validation; performance specifications (mechanical, electrical, force feedback, hysteresis, and reliability); system design (fly-by-wire, infrared-based navigation, electromagnetic field-based navigation, robotic arm, or robot); human factors known to affect performance (the learning curve and usability); known technical failures (software anomalies and physical limitations); known additional safety issues and/or adverse events; and additional operative time needed.
A desire to improve patient outcomes has always made surgeons take a careful look at their own patient outcomes and has stimulated interest in honing experience and improving technique. In judging the clinical utility of computer-assisted navigation and surgical preparation, important considerations include the ability of the device to locate anatomic structures and to evaluate alignment, fit and fill, soft-tissue balance, and appropriate sizing of implants. Improved pain relief and function, functional stability through a full range of motion, increased risk of fractures or infections, additional incisions required, and increased surgical time are clinical outcomes. Long-term outcomes that are expected and desired, but not yet proven, include reduced rates of revision surgery with less prosthetic wear and concomitant prosthetic durability and longevity as a result of better, more precise alignment and placement of the implant. Although randomized, concurrently controlled clinical trials may provide this information6-9, particularly as a way to compare patient outcomes with standards of care, to date there have been few level-I clinical trials10 to investigate these short and longer-term end points.
As is the case for any new technology, there still remain several unresolved issues. The most obvious is the long-term clinical importance, tempered by the risks and cost involved. These are the challenges we will face as this technology is modified and developed to meet the needs of the professional community and the problems patients present to them. The FDA continues to work with standards organizations, manufacturers, and physicians to develop guidance and standards for testing to ensure that safe and effective products are consistently made available for patient care. 
Note: The authors thank Ms. Christy Foreman, Mr. Mark Melkerson, Mr. Neil Ogden, Dr. Long Chen, and Mr. Dwight Yen for their review of this manuscript.
Rauh MA, Munjal S, Phillips MJ, Krackow KA. Surgical navigation in adult reconstruction surgery: techniques and clinical experience. Instr Course Lect.2008;57:699-706.57699
2008
[PubMed]
Ulrich SD, Mont MA, Bonutti PM, Seyler TM, Marker DR, Jones LC. Scientific evidence supporting computer-assisted surgery and minimally invasive surgery for total knee arthroplasty. Expert Rev Med Devices.2007;4:497-505.4497
2007
[CrossRef]
Ulrich SD, Bonutti PM, Seyler TM, Marker DR, Jones LC, Mont MA. Outcomes-based evaluations supporting computer-assisted surgery and minimally invasive surgery for total hip arthroplasty. Expert Rev Med Devices.2007;4:873-83.4873
2007
[CrossRef]
Ulrich SD, Bonutti PM, Marker DR, Dethmers DA, Reinhold JH, Robinson Y, Jones LC, Seyler TM, Mont MA. Computer-assisted surgery in total joint arthroplasty: complications, pitfalls and solutions. Exhibit at the Annual Meeting of the American Academy of Orthopaedic Surgeons; 2008 Feb 25-28; San Francisco, CA.
2008
Wright JG, Swiontkowski MF, Heckman JD. Introducing levels of evidence to the journal. J Bone Joint Surg Am.2003;85:1-3.851
2003
[CrossRef]