Informatics—a computational approach to the complexity of the epilepsies
The review in this issue by Sahoo et al. (2013) recommends that modern informatics techniques in the form of an “ontology” can serve as a means of assisting the current dilemmas regarding epilepsy classification. What is an “ontology,” how does it differ from a “classification,” and what are the perceived advantages?
The common understanding of a “classification” such as we use for seizures and epilepsies is a hierarchical structure about a particular domain of knowledge. Other words that enter into the discussion of organizational schemes include terminologies, taxonomies, and controlled vocabularies. The definitions and implementations of each vary somewhat with the domain being considered and the application for which it is intended. In the realm of biology, we are most familiar with the taxonomy used to describe animal life (i.e., kingdom, phylum, class, order, family, genus, species) based on the principle of who can reproduce with whom. Similarly, the classification of the seizures as described by the International League Against Epilepsy (ILAE) Commission on Classification and Terminology (CTC) in 1981 has as an organizing principle, partial versus generalized ictal onset. Both classifications have great appeal owing to the clarity of the organizational principle, resulting structure, and utility for teaching the topic (domain) being considered. However, what if we wish to consider other aspects of animals such as location, color, feeding behavior, modes of movement, required nutrition, genome, and so on? The complexity of knowledge regarding animals is simply too great to be captured by a taxonomy with only one major axis for distinguishing the members to be considered. Although knowledge about seizures and epilepsies is much more restricted than that of all animals, the concept of complexity pertains. The multiple types of knowledge that are pertinent in this domain defy a simple one or two (or three or four) tiered classification system. The need for a “multidimensional” system was recognized prior to the most recent ILAE CTC Task Force recommendations (Berg et al., 2010) in a 2001 revision (Engel, 2001) that suggested a five-axis diagnostic scheme to characterize seizures and epilepsies. That suggested revision also failed to gain traction in the epilepsy community. Therefore, it would be useful to conduct a “root-cause” analysis of why we have failed to come to consensus despite many hours of work by highly competent individuals with only the best of intentions and potential solutions. How do we move past the current intellectual “gridlock”?
Recognition of Complexity
The first is the need to acknowledge the complexity of knowledge that exists about seizures and epilepsies as well as the multiple uses of that knowledge. These are interrelated concepts. So much of the discussion appropriately focuses on what is needed to provide optimal care for those with seizures. In this context, the information provided by knowledge about seizure onset (focal vs. generalized), with or without impairment of consciousness, and likely etiologies are extremely important. By extension, these “descriptors” are necessary for the development of new therapies based on current strategies. However, the reality exists that this level of knowledge may not be available in resource-challenged regions of the world, in which some degree of diagnosis and management must occur. At the other extreme of clinical care, more precise information is required to perform epilepsy surgery or design new biologic agents based on causative genetic mutations. As soon as we desire to incorporate clinical science (necessary to establish the relationships between individual concepts) and basic science (necessary to understand the mechanisms resulting in seizures), it becomes clear that a relatively simplistic classification system is not up to the task for all stakeholders.
All that is required to deal with this challenge is community acknowledgement of the complexity of seizures and epilepsies, along with a willingness to accept the reality that different “subclassifications” will be needed to address individual applications and contexts in which the knowledge is required. However, these must be harmonized so that one piece of knowledge means the same in each “subclassification.” The ILAE should play a major role in this effort as the internationally recognized organization in this domain.
The second is the need to arrive at common definitions for core terms and concepts. Any attempt to provide an internally consistent system of knowledge with application across the multiple dimensions required for a complete characterization of seizures and epilepsies will be thwarted unless this can be achieved. Of note, this pertains whether the knowledge is discussed by people or computed by machines. “Idiopathic” cannot mean “unknown” and “of presumed genetic etiology,” as these are not synonymous. The situation becomes even more precarious if other modifiers are assumed without clear limits such as pharmacoresponsiveness, age of onset, and spontaneous remission. The same situation applies to the more complicated concept of an “epilepsy syndrome.” Definitions must exist for there to be clarity across users.
A great deal of literature has already been developed to address this need. The National Institute of Neurological Disorders and Stroke (NINDS) has published common data elements for seizures and epilepsies (Loring et al., 2011). Highly regarded glossaries exist for seizures and epilepsies ILAE (Blume et al., 2001) and EEG (Noachter et al. 1999). The rate-limiting step in achieving consistency with regard to definitions of terms is one of consensus among users.
An Adequate Framework
The third is a framework in which this knowledge needs to be “assembled” in a manner that allows use by multiple stakeholders. As described earlier, a tiered system organized around a few defining features is not adequate to achieve this goal. The text document that we have been using for decades (handwritten or electronic) does not have the intrinsic capacity to enable organization. Putting information into a database format (such as those that are on the backend of electronic health records) allows sorting and some re-use of data.
This where the concept of an ontology as a framework of knowledge becomes crucial. Furthermore, there must be a “language” that provides the information that is incorporated into that framework. The nature of modern ontologies allows incorporation of concepts along multiple axes. The structure uses individual pieces of data that are then incorporated into larger concepts that are connected to each other (relationships) by rules based upon knowledge of the domain (e.g., staring + 4–10 years old + 3 Hz spike-wave EEG = Childhood Absence Epilepsy). The manner in which each of these demographic, symptoms, signs, and EEG dimensions can be reassembled into different concepts (syndromes) is illustrated in Fig. 2 of the review by Sahoo et al. (2013). The clinician or laboratory scientist need not be concerned with technology behind the ontology or the language in which it is written (e.g., OWL, Ontology Web Language) any more than we are knowledgeable about how any of the commonly used databases are constructed. These are simply new tools available for us to organize information so as to improve clinical care, teach, and serve as a basis for discovery.
There are additional advantages that flow from the development of a seizure-epilepsy ontology; these include the heuristic value of determining the experiments that needs to be obtained to create the rules that relate one concept to another; the ability to harmonize multiple classification systems (e.g., International League Against Epilepsy, Systematized Nomenclature of Medicine Clinical Terms, and the International Classification of Diseases coding). The latter is of particular importance as it serves as the basis for assessing disease burden, code driven research, and reimbursement in some countries.
How is this sea change to be implemented, as the necessary software and hardware are now readily available? The authors suggest a consortium approach. Ideally this should be informed by the ILAE as the international body that has traditionally guided the seizure and epilepsy classification process. Parallel with the creation of a consortium is the need to educate all stakeholders about the language of clinical informatics. Just as we needed to learn the terminology of molecular genetics a decade ago (allelic heterogeneity), so now we need to understand the basic concepts of clinical bioinformatics (semantic heterogeneity), which has some striking similarities to the language of molecular genetics. Perhaps the most significant challenge is for us as individuals to give a little autonomy with regard to preferred terms and concepts so as to reap the great potential advances that come with a unified framework for the domain of epilepsies facilitated by modern computational methodologies.
The author has no conflict of interest disclosures and confirms that he has read the Journal's position on issues involved with ethical publication and affirms that this report is consistent with those guidelines.