Many complex issues must be considered and incorporated when planning exactly how all the extensive information collected in the peri-anesthetic period will be input into whatever system will capture and store the data.
During the preanesthetic evaluation and intra- and postoperatively, the human user has to enter data into the system. Available user interface technologies include the familiar keyboard, and the more interesting options of pen computers, voice recognition, and the "point-and-click" entry mechanism that starts with user-configurable lists of likely entries. The latter is arguably the most likely to result in codified data, which is clearly very desirable for outcome studies. What are some of the criteria to decide among the available options?
Computerized Pen Entry
Particularly in the intra-operative environment, pen based computers are attractive because the act of "writing" an anesthetic record is familiar and can be used where space is at a premium. Handwriting recognition can work nicely, particularly for users already familiar with Graffiti via Palm Operating System devices. More ambitious systems that attempt to recognize unconstrained writing usually require some training of the device, and also of the user. However, pen computers can also enable the user to annotate with digital ink (bitmap), which is useable with basically no training; however, the resulting record has the disadvantage of being just as illegible as current paper records. The disadvantage of free-form annotation is keenly felt by a clinician taking over a complex case in progress. On the other hand, point-and-click structured entry of drugs and fluids allows the record keeper system to present legible entries and current totals.
Voice recognition technology is constantly improving and is becoming quite useable, at least when the computer can be given easily distinguishable choices and some assurance that the user will be speaking one of the choices. Intra-operative use of voice entry poses challenges, including entries being made by multiple, sometimes very stressed, persons and considerable background noise that must be ignored. Even with the challenges properly addressed, there remains a possible problem if the surgeon and anesthesiologist do not agree on a point to be recorded; a quick selection from a list of choices or some quiet typing would be less controversial than a spoken entry.
Of course, voice can be recorded without transforming it into printed text. This has considerable advantages in cases of emergencies where a record of rapidly evolving events is required.
Drugs and fluids lend themselves very well to point-and-click entry, resulting in legible documentation (including timing) and highly structured data for future study. It also has the advantage of providing current totals, which can be very helpful to a clinician taking over a case.
The capture of intra-operative procedures and events presents a problem. At least routine events can be anticipated, and context-sensitive text fragments can be provided for convenient entry and editing, if necessary. Any combination of the above technologies can be made available and mixed by the user.
Automated Data Collection
Many physiologic monitors have one or more ports to send at least current (and sometimes past) data to a computer for automated data collection. Vendors vary in the quality and accessibility of their documentation, and in the complexity of their communication protocols. More complex protocols typically are the result of greater functionality offered by the device.
An automated record-keeping system must use a computer to collect data from the physiologic monitors, requiring many physical connections. Should the clinician use this dedicated computer to enter drugs, fluids, and notes? While that arrangement is common today, an additional machine might offer significant advantages. A second computer can be wireless and follow the patient to a postoperative care unit.
One disadvantage of automated data collection is that machines in general may record artifacts along with valid data. How bad is the problem?
Some monitors provide measures of signal quality, but there are arguments for and against advertising that a record contains poor data. Some clinicians want to have the quality measures noted so that their action or inaction in response to an artifact can be seen in the appropriate light.
There are many sources of artifacts, well recognized by clinicians, who must be able to mark artifacts when they are recorded. One classic example is a noninvasive blood pressure cuff reading that is wildly incorrect because a member of the surgical team is leaning on the cuff through the drapes. Technology to recognize and suppress artifacts has improved greatly, but, much remains to be done.
The sharing of data generated in the anesthesia environment between other stations in the hospital or other hospitals requires special considerations, one of which is confidentiality. All patient-identifying communications should be encrypted and users must be adequately authenticated to gain access to confidential information. Any system that allows an individual to run interactive queries should be considered very carefully before deployment, because overlapping results of seemingly unrelated queries could ultimately identify an individual patient.
Distributed vs. Centralized Systems
Whether data are copied into a central repository (a centralized system) or reside at or near the point of care (a distributed system), some kind of networking is likely to be utilized. How would one decide between the two types of systems?
A distributed system offers advantages can adapt to diverse components produced by different vendors; health care providers would then be able to keep their existing information systems and still be part of a larger information system that helps individual patients as well as the community as a whole. The resulting competition among software vendors brings economic benefits, both in the software market itself and in healthcare in general; providers with good information management can offer excellent care efficiently. A distributed system can be more reliable than a centralized equivalent, because while there are admittedly more things to break (and in very different places), they can offer redundancy. Finally, such a system can distribute data such that the bulk of any patient’s data is geographically close to the patient. Long distance accesses will be required, but, not nearly to the extent that they would be if all data were stored in one place.
The next logical question to ask is how might one glue diverse information systems into a unified whole? Should the system be based entirely on exporting from one system and importing (copying) to the next? What other options are available? Copying demographic data for a patient transferred from one hospital to another can reduce data entry costs, time, and reduce errors. The same advantage can accrue to, for example, storing a list of allergies, which each provider caring for a patient should have on hand. Health Level Seven (HL7, http://www.hl7.org/) is a widely used standard for exchanging data in these and other situations.
An alternative to copying data is to create software components that not only hold data but also contain the logic required to manipulate the data. The Common Object Request Broker Architecture (CORBA, http://www.omg.org) is a readily available set of standards for defining components that are distributed across a network, and potentially running on different kinds of computers. Components can be used both to create new kinds of software and to economically make pre-existing systems compatible with the new systems.
Successful outcome studies will require a chain of systems and practices. Widely accepted and used electronic information systems will result in large amounts of codified data. A most-likely choice of a distributed database will encourage exchange of data while respecting the needs of autonomous care providers and institutions. Queries of this database will need to cope with differing data dictionaries and systems, but, distributed object-oriented programming techniques and associated standards should be able to provide the necessary links between systems. Patient confidentiality will be a significant concern, and the system will have to cope with attacks that could include collections of seemingly innocent queries whose combined results could be used to uniquely identify an individual patient.
Dr. Schwab is from the Department of Anesthesiology, University of Florida, Gainesville.