Decision Support Systems in Oncology: Are we there yet?

December 12, 2008
Mathieu d?Aquin, Jean Lieber

,
Amedeo Napoli

ONCNG Oncology, November 2008, Volume 9, Issue 11

The push for clinical decision support technology in medicine is a logical consequence of our experiences as consumers and the need for intelligent support at the bedside.

The push for clinical decision support technology in medicine is a logical consequence of our experiences as consumers and the need for intelligent support at the bedside. However, the reality is that clinical decision support technology has yet to demonstrate sufficient practical value in clinical practice.

Decision support system applications have a long history in the field of oncology. Almost from the birth of the so-called “expert systems,” there have been programs dedicated to helping practitioners diagnose, treat, or monitor cancers. Among the first to explore this area was ONCOCIN in the mid-’80s. Like many other systems, ONCOCIN was based on a set of rules, encoding actions to trigger in specific situations. More generally, a clinical decision support system such as ONCOCIN is a computer program that takes, as input, the description of a medical situation and provides, as a result, information supporting the practitioner in making the appropriate decisions concerning this situation. Such a program intends to capture—to encode—medical knowledge and to apply problem-solving methods to automatically “reason” based upon this knowledge.

Although there has been a lot of research, resulting in several prototypes and experiments in applying decision support systems in medicine, none have ever been able to support the daily practice of oncology. Expert systems, after generating so much enthusiasm, have been disappointing and acquired a bad reputation. They were blamed for not being adaptive enough to particular situations, not being helpful enough—as they were not able to explain their results—and also not being dynamic enough, as they were hard to modify to keep pace with the advances in medicine.

However, the underlying technologies for decision support systems have evolved immensely since the early expert systems. Research in knowledge management has produced adequate methods for capturing and maintaining computerized expertise, and provided the appropriate level of interaction between the system and the practitioner. Therefore, this might be the time to consider once again the question of decision support systems in oncology: Is their bad reputation still justified, and what are the obstacles to widespread adoption?

The Kasimir project

The beginnings of an answer to these questions may come from our experience in building a modern decision support system in oncology: the Kasimir project.

Born in 1997, the Kasimir project gathers specialists in oncology, as well as researchers in psycho-ergonomics and in computer science, with the goal of supporting practitioners in managing and using decision knowledge in the Lorraine region of France. This decision knowledge can be found in decision protocols: documents summarizing the action to be applied in common medical situations according to the state of the art in medicine and following the principle of evidence-based medicine. For example, the breast cancer treatment protocol provides the necessary information to help a physician select, depending on the patient and on her cancer, the standard treatment to apply. Other decision protocols considered in this project concern the treatment of prostate cancers, the surveillance of breast cancers, and the inclusion of patients in clinical trials.

The Kasmir Project focuses on two levels of decision support: protocol application and protocol adaptation. The study of protocol application has led to computer programs within the Kasimir system that are stable and are based on standard methods and tools used in knowledge-based systems. Such a program has a user-friendly interface for describing a patient and her cancer, and for displaying the associated standard decision. Some studies carried on by physicians have shown a statistically significant improvement in decision protocol observance when the physicians use the Kasimir system, compared to their observance when they use paper-based versions of the same protocol.

For a majority of patients, the standard decision given by protocol application can be applied as such. The specialists in psychoergonomics involved in the project have shown that for the other patients (about 40%), the oncologists adapt the decision protocol (ie, actually use it, but with a critical eye and not in a straightforward way). There are many reasons why a protocol has to be adapted, including drug contraindications, closeness to a decision threshold, and interfering conditions such as pregnancy. The complex and rich domain of decision protocol adaptation has been deeply studied in the Kasimir project () but still requires additional research in order to develop tools that could be used by physicians in their daily practice. Therefore, we have concentrated our effort for transferring this research to the medical community mainly on protocol application.

The key to effective protocol application is implementing a computerized protocol (ie, the translation of the decision protocol document into a machine-understandable file expressed in a standard knowledge representation formalism). The translation process is not straightforward; it requires the knowledge engineer in charge of this task to interact with physicians in order to better understand the medical concepts used in the protocol and to point out some implicit pieces of knowledge (that are to be made explicit for computers). It must also be tested by an oncology specialist before being introduced into practice. This entails important efforts from oncology experts and knowledge engineers, who must not only build the first version of the computerized protocol, but also continue to maintain and update it. This is indeed critical, as some protocols, such as the breast cancer treatment protocol, are reconsidered every six months to take into account the progress of research in oncology.

The end-users of the Kasimir system are physicians who are not highly specialized in oncology. Those we have met have tested the system for several weeks and are willing to use it in their daily practice. However, as previously mentioned, implementing and maintaining protocols is a time-intensive process. Unfortunately, because there are relatively few oncology specialists, at least in Western Europe, and because of the growing number of cancers, the time specialists can devote to non-clinical endeavors is decreasing. Moreover, they do not directly benefit from this work; as specialists in oncology, they do not need the help of such a decision support system.

What does the future hold?

What is the future of decision support systems like Kasimir? At first glance, it seems that, even considering advances in technology, they are still confronted by the so-called knowledge acquisition and maintenance bottleneck. Oncology experts spend a lot of time on the acquisition and maintenance of paper-based decision protocols. Once these documents are published, it is understandable that the authors would be reluctant to spend additional time and effort validating the computerized protocol implemented by knowledge engineers. But must this effort be additional? The job of knowledge engineers is the acquisition, management, and maintenance of knowledge. Hence, it might be possible to create an efficient synergy between oncologists and knowledge engineers and produce directly computerized protocols. From our experience in the Kasimir project, we believe that such an approach is feasible and that it would lead to:

• A benefit in time for the oncologists (for the acquisition and the maintenance of the protocols and because simple cases—that do not require adaptation—can be managed by less-specialized physicians).

• A benefit in competence for less specialized physicians (ability to solve at least simple cases).

• A benefit in healthcare quality for patients (having the level of speciality in oncology fitting the complexity of his or her case).

Such a shift in organization requires a political decision and acceptance by the medical community of the need to implement knowledge management practices. Our experience as computer scientists can only show that this shift is technologically possible, while it seems to be, from the viewpoint of the users of the medical system, more and more desirable.

Mathieu d’Aquin is from the Knowledge Media Institute, the Open University, Walton Hall, Milton Keynes, UK. Jean Lieber and Amedeo Napoli are on the Orpailleur team, LORIA (CNRS, INRIA), Nancy University, Vandoeuvre-lès-Nancy, France.