Intelligent objects will speed the transformation of medicine in the information age
Craig Feied MD, Mark Smith MD, Jon Handler MD, Michael Gillam MD, Meera Kanhouwa MD
From
the
As medicine enters the information age, physicians must struggle with a mind-numbing variety of information resources. As medical informaticists, our early efforts to cope with this embarrassment of riches led us to develop "information agents" – small programs that search multiple resources to find focused answers to specific questions. Information agents have proven to be extremely useful in clinical practice, and such agents are increasingly available on the Internet at sites such as www.ncemi.org and www.emedicine.com.
Depending on the results returned by an information agent, a physician may take actions such as ordering a test, prescribing a medication, or performing a procedure. Often the action needed is directly linked to the results of an informational query. For example, consider the query "is this drug safe for my patient, a woman who could potentially be pregnant?" If the answer is "no," then a pregnancy test will be ordered before the drug can be prescribed. This type of question-response-action sequence often is highly stereotyped, and could conceivably lend itself to further automation. Today, the information agent speeds the query time, but the requirement for human attention and human volition remains as a bottleneck in the overall process.
Information agents will continue to be an important aid to clinical decision making for the near future, but a much more fundamental change is coming in the form of "intelligent objects" that may help to close the gap between information and action. We chose the term "intelligent objects" to describe a process whereby ordinary everyday things seem to take on a life of their own and reach out to perform independent actions. Intelligent objects are inanimate physical objects whose deployment and use can be governed or affected by transactions that occur within a computer or across a network.
Intelligent objects evolve in an environment of interconnected networks that provide a rich environment in "cyberspace," a non-dimensional space reflecting the "there" associated with a network connecting two or more real dimensional locations. Within this environment, the intelligence that will be attached to inanimate objects is created by a computer programming technique in which many small software modules ("objects") interact with each other by passing information and requests for action ("messages ") back and forth. Upon receipt of a message, an object exercises one or more of its "methods" (the things it "knows" how to do). In this "object-oriented" approach, many independent software objects interact to achieve the kinds of results formerly produced by a single monolithic computer program. Software objects can be very simple or very complex, and can imitate or model the behaviors of real-world objects that have a physical existence. Objects written by different people at different times can work together even though not tightly coupled: to invoke the behaviors of an object, one need only know how to send it a message over some network. By sending and receiving messages, objects can interact between widely separated computers at unknown locations. With ubiquitous connectivity, these software objects become universally accessible, and seem to exist within cyberspace.
Intelligent objects come about through the marriage of cyberspace objects to the real physical objects they represent. In effect, the combination of ubiquitous machine-to-machine connectivity and the use of object orientation in systems design allows us to add narrowly focused "intelligence" to what are now real-life inanimate objects. This intelligence resides in a portion of cyberspace that is closely associated with the physical expression of an object or a class of objects. An example illustrates the concept.
Today, a medication is just a powder in a bottle. Before it can be administered, a clinician unfamiliar with the drug must take active steps to seek out and acquire information about the correct dosing and administration of the drug. Those steps must be repeated by every clinician every time the drug is given. Before an unfamiliar drug is administered, similar information may have been looked up by a student, an intern, a senior resident, an attending, a nurse, and a pharmacist. Each clinician seeks and finds information about the drug from one or more of a variety of different sources, because subsets of information about the medication exist in a variety of locations.
Finding information about an unfamiliar drug often requires secret knowledge of where to look and what questions to ask. The hospital formulary tells whether the drug is available and in what form. Pharmacy guidelines tell how to mix it and with what it is compatible. The resident's survival guide tells how much to prescribe. For usage in pregnancy, a consultant may be called. Dosing in renal failure is found in the Physician's Desk Reference (PDR). Updated manufacturer’s information is found in PDR supplements, although these are a rarely-used resource. Guidelines for off-label use may require a literature search. Potential interactions may be identified if an explicit effort is made to review all of a patient's medications using a drug-drug interaction reference.
Each clinician believes his or her portion of the information is complete and correct, and after a number of tedious look-ups, each clinician believes he or she "knows" all important information about the drug. At this point, the information becomes frozen in human memory, and subsequent changes in published recommendations often fail to change the practice of experienced clinicians. Even the original recommendations may not be followed, as human memory becomes more vague over time. Many iatrogenic injuries and deaths are attributed to incorrectly remembered medication information.
To clinicians of the future, this present state of practice will seem primitive. In the future, all information about a drug will be uniformly available everywhere. No drug will be marketed without a centrally-administered data resource that provides a publicly available up-to-date version of all known information related to that drug. But it will not be physicians, nurses, and pharmacists who use the data resource: in the future, the data resource will principally be used by the drug itself.
This is not a fanciful notion. Even today, many hospitals use locked, computer-controlled pharmacy supply carts from which each dose of medication must be requested for a specific patient. These computerized carts are part of a network-based pharmacy system that tracks medication use in order to manage pharmacy charting, billing, and inventory. Today the pharmacy computer knows a lot about the patient's billing charges and the other drugs that have been ordered, but nothing about the clinical conditions for which the drug is being ordered. Even so, today's pharmacy systems are capable of flagging drug-drug interactions and of recommending less expensive alternatives in selected situations. In the future, a medication and its cyberspace counterpart will be even more tightly linked through an object-oriented pharmacy system, and objects in that system will have access to objects in other clinical systems throughout the enterprise. A medication object will "know" to whom it is being administered, and each medication will "know" its own list of indications, contraindications, interactions, and side effects. A dose of medication will become a "real-world instance" of a class of "cybermedication objects" that possess a wide variety of innate behaviors ("object methods").
In the future, when a nurse requests a dose of medication for a patient, the cybermedication object will examine its new environment (the patient data record) and will send back messages containing the correct dose for that patient, the necessary adjustments in any other medications for compatibility, the method of mixing, infusate compatibility, and any other desirable information. If a medication knows it shouldn't be given in pregnancy, it won't allow itself to be given to a pregnant patient unless the physician explicitly forces the issue with an override. If the medication can't tell from its environment whether the patient might be pregnant, then if there is already urine in the lab, it will order and check a pregnancy test. If there is no urine available, it will ask for some.
Medications will not be the only intelligent objects in the emergency department of the future. Lab tests will also be intelligent, and will order companion tests if needed for meaningful interpretation. An order for a CSF glucose, for example, might look for a recent serum glucose, and if none was found, might proceed to order one. It is entirely possible that an ECG might read itself, compare itself to its younger self, order a set of cardiac enzymes, request that the pharmacy cart release a thrombolytic agent, and put in a page for the CCU resident, notifying the clinician of these actions as they are performed.
Across the country and around the world, teams of developers are even now working to develop the intelligent objects we have described here. As standards for information exchange continue to evolve, so too will the number and complexity of the actions that can be taken by semi-autonomous intelligent objects. The combination promises an exciting era for medicine in the next few decades.
Correspondence:
Craig F. Feied, MD, FACEP, FAAEM
Director,
Director of Informatics, EMC medical group, Medstar Health System
cfeied@ncemi.org