That which we call a pediatrician would by any other name a child treat (or not) That which we call a pediatrician would by any other name a child treat (or not) Amanda Hicks1,* and William R. Hogan1 1 Department of Health Outcomes and Policy, University of Florida, USA ABSTRACT currently represent these different kinds of entities, and the Analyses of workforce data about the number of pediatric specialists in work that remains. the USA and Australia show that data sets representing physicians often represent ontologically different types of things and some do not represent a Section 2 reviews roles in BFO 2.0 (Arp et al., 2015). count of any single type of thing at all. This produces widely variable counts Section 3 reviews the conflicting counts of pediatric of paediatric specialists used in workforce analyses that inform public policy specialists in Australia (Allen, Doherty, Hilton, & Freed, decisions. This paper reviews the different kinds of entities that are counted in these data sets, assesses the extent OBO Library Ontologies based on BFO 2016) and the USA (Freed et al., 2006) based on authoritative can represent these different kinds of entities, and outlines work that remains. data sets. Section 4 reviews the types of things that were This paper provides insight into outstanding issues and difficulties for counted, the ontological entities necessary to represent these modelling health care provider roles. types, and the existing work on each of these areas in existing BFO-based ontologies and outlines work that remains. 1 INTRODUCTION Analyses of workforce data about the number of paediatric 2 ROLES IN BFO-BASED ONTOLOGIES, THE specialists in the USA (Freed, Nahra, Wheeler, & Research CASE OF THE CARDIAC PEDIATRICIAN Advisory Committee of American Board of, 2006) and We can distinguish health care provider roles along two axes: Australia (Freed et al., 2006) show that distinct authoritative the kinds of processes that realize a role and the kind of data sets representing physicians often represent external grounds that are the necessary and sufficient ontologically different types of things from each other, and in conditions for the existence of the role. The former is already some cases represent heterogeneous types of things as the frequently done, and the latter is relatively uncharted territory same type of thing within a single data set. For example, with the exception of the Ontology of Organizational counting pediatric cardiologists based on board certification Structures of Trauma centers and Trauma systems (OOSTT) and hours spent providing care to patients under a particular discussed below. Roles that are grounded in institutional facts age each results in different numbers. Consequently, counts often coexist with rights and obligations on the part of the of pediatric specialists used in workforce analyses that inform bearer of these roles. It may be that these rights and public policy decisions vary widely. We view this as an obligations can also distinguish one type of role from another. ontological problem that can be addressed with robust, formal For example, the nurse practitioner role and the physician representations of physician roles with more rigorous role are both realized in health care encounters. However, a definitions. Despite previous work on roles in BFO (Arp & nurse practitioner does not have the permission or right to Smith, 2008; Arp, Smith, & Spear, 2015) and BFO-based engage in these health care encounters without a supervising ontologies (Hicks, Hanna, Welch, Brochhausen, & Hogan, physician. We do not, however, address rights, permissions, 2016; Hogan, Garimalla, & Tariq, 2011; Peters & and obligations in this paper. Consortium, 2009; Utecht et al., 2016), there is not yet a full account of institutionally grounded roles in these ontologies. 3 CONFLICTING COUNTS This paper reviews the distinct kinds of entities that are counted in authoritative physician data sets, the extent to In this section, we provide a review of two articles that which OBO Library (Smith et al., 2007) ontologies can describe heterogeneous counts of pediatric specialists. Allen et al. review conflicting counts of pediatric specialists in NHWD AHPRA-1 AHPRA-2 RACP MBA Websites 2013 2013 2013 2014 Sept. 2015 Feb. 2016 Number of pediatric cardiologists 21 17 30 31 32 38 listed by data source Table 1. Counts of pediatric cardiologists from various data sources as reported in Allen et al. (2016) * To whom correspondence should be addressed: aehicks@ufl.edu 1 That which we call a pediatrician would by any other name a child treat (or not) AMA ABP Individuals On Survey Survey Survey Received Received 2002 2002 on both AMA, respondents respondents respondents no only adult lists not not pediatric provide some training cardiology ABP cardiology care care training 2512 1846 1558 out of 738 294 119 175 13% 57% 2675 Table 2. Counts of pediatric cardiologists from various data sources as reported in Freed et al. (2006) Australia for eight specialties. The results for pediatric Membership in the Royal Australasian College of cardiologists are reported in Table 1. Physicians (RACP) is voluntary for Australian and New The Australian Health Practitioner Regulation Agency Zealander physicians. Members include full-fledged (AHPRA) administers registration as a medical specialist to specialists and trainee specialists. Table 1 shows that 31 practitioners in Australia who meet certain criteria. The members of the RACP were indicated to be pediatric medical specialties are recognized by The Australian Health cardiologists in the RACP membership database. The criteria Workforce Ministerial Council. ‘Medical specialist’ in this for determining specialties are unclear. That is, it is unclear context has an explicit definition that is tied to institutional whether this is self-identified or whether registration with a regulations and registrations, and likewise, types of governing body is required. It is clear, however, that this specialties are specialties in virtue of being recognized by a count may include pediatric cardiologists in training. legal body. Practitioners in Australia must renew their Registration with the Medical Board of Australia (MBA) registration annually with the AHPRA during which they are is compulsory for practicing doctors and requires the administered a voluntary survey. One of the questions of the following: completion of intern training, having been survey is in which two specialties the practitioner provided awarded a primary degree in medicine and surgery from an the most care in the week prior to taking the survey. For accredited institution in Australia or New Zealand, proof of instance, Dr. Petitcoeur may not be registered as a pediatric identity, completed criminal history check, agreeing to cardiologist, but if she indicates that she spent more time in comply with indemnity insurance registration, and the previous week administering cardiology services (such as competency in English. Table 1 indicates the number of echocardiograms) to children, then she is included in both doctors registered with the MBA who completed intern National Health Workforce Database (NHWD) and AHPRA training in pediatric cardiology or who are currently counts of pediatric cardiologists. participating in such training (and so are “provisionally” The 2013 NHWD count in Table 1 reflects the number registered). of doctors who either were registered as a pediatric Allen et al. conducted an online search for doctors listed cardiologist with the AHPRA or identified as working the as practicing in pediatric specialist fields in Australia. Table most hours in the previous week in pediatric cardiology. 1 lists the number of physicians asserted to provide pediatric AHPRA-1 reflects only the results of the workforce survey. cardiology care on the website of a hospital or clinic that That is, it only reflects the number of practitioners who were provides pediatric care. renewing their registration and indicated pediatric cardiology Allen et al.’s analysis provides insight into the variety of as one of the two specialties in which they spent the most time ontological considerations that ought to be taken into account when modeling medical specialist roles and their realizations. providing specialty care in the previous week. It also illustrates the variety of intensional meanings that may However, one potential data quality issue with AHPRA be meant by a specialist term such ‘pediatric cardiologist’ and survey data is that the specialty described simply as the variety of extensions that result from these (often covert) “cardiology” is intended to be “adult cardiology,” but Allen ambiguities. These lessons are reinforced by Freed et al.’s et al. note that anecdotal evidence suggests that this is not review of counts of pediatric cardiologists in the USA. clear to survey respondents. Consequently, sometimes Freed et al. reviewed counts of pediatric cardiologists in pediatricians such as Dr. Petitcoeur may select “cardiology” the US in 2002 by comparing the individual pediatric if they have already indicated a pediatric specialty elsewhere cardiologists listed in the American Medical Association assuming that pediatric cardiology is a type of cardiology. In (AMA) Masterfile with individual pediatric cardiologists on table 1 AHPRA-2 reflects practitioners who were renewing a roster for the American Board of Pediatrics (ABP). their registration and had either indicated pediatric In table 2 AMA reflects the number of doctors who had cardiology as one of the two specialties in which they had either (a) a primary or secondary specialty listed as pediatric spent the most time providing specialty care in the previous cardiology or (b) whose primary or secondary specialty was week or had indicated cardiology and some other pediatric listed as pediatrics and the other specialty listed as cardiology specialty as the two specialties in which they had spent the or who are listed as board certified in pediatric cardiology. most time providing specialty care in the previous week. This number includes retired and inactive physicians but does 2 That which we call a pediatrician would by any other name a child treat (or not) not include deceased persons. Because the AMA Masterfile complete. In this section we review external grounds of is a compilation of heterogeneous data sources and because physician roles in BFO-based ontologies. different data sources may count pediatric cardiologists according to different criteria, this number does not include 4.2.1. Training the number of physicians who satisfy some explicit set of Both the ABP and the survey data from Freed et al. take ontological criteria and so likely represents a heterogeneous quantity and type of training into account when counting set of persons. pediatric cardiologists. In the Freed Survey, quantity was In Table 2 ABP reflects the number of doctors who were categorized as no training, completed training, and some ever board certified in pediatric cardiology or who had training, which in turn was defined by a fiat boundary (3 completed training in pediatric cardiology but had either not years or less). Type of training includes specific stages of taken or failed the board certification examination. This training such as residency, and training that is pertinent to the includes both retired and deceased persons. specialty, i.e., cardiology training. Freed et al. analyzed the survey results according to how The Ontology for Biomedical Investigations (OBI) and many respondents spend less than 50% of clinical effort on OOSTT represent aspects of training. OOSTT describes pediatric cardiology, how many limit their cardiac care to information content entities that are the specified output of children, and how many have received at least three years of completed training. pediatric cardiac training or at least three years of adult cardiac training. The result is that physicians who are counted successful completion of anesthesiology residency as pediatric cardiologists by the AMA have a variable amount information – An information content entity that is the of training and clinical effort in the area. 40% of survey specified output of a person successfully fulfilling the respondents do not provide any pediatric cardiology care despite being listed as pediatric cardiologists. That is, if these evaluation criteria at the end of an anesthesiology persons do bear a pediatric cardiologist role, it is never residency program. realized. Other individuals are counted as pediatric anesthesiology residency program –medical residency in the cardiologists though not board certified and, of these, some medical speciality that focuses on the administeration have had no training in cardiology, whether pediatric or adult [sic] of medication for the temporary general or local cardiology. suppression of sensory or motor nerve function during some health care encounter or on making decisions 4 ONTOLOGICAL DISTINCTIONS regarding the adminstration [sic] of such medication. Medical Residency - Residency is a stage of graduate medical In this section, we turn to ontological considerations that arise training. from Allen et al.’s and Freed et al.’s work and describe to what extent relevant ontological distinctions are addressed in OOSTT uses VIVO-ISF’s class for medical residency existing BFO-based OBO Library ontologies. (Börner, Conlon, Corson-Rikert, & Ding, 2012). 4.1 Identified as While OOSTT represents the documentation of As in the case of websites described above, individuals can successful evaluation at the end of some stage of training, be identified as pediatric specialists, that is, somebody can which is undoubtedly useful, there is no representation in assert that Dr. Petitcoeur is a pediatric cardiologist. This OOSTT of evaluation criteria or what those criteria measure, assertion by itself does not directly contribute to the i.e., the competencies acquired. Indeed, how to ontologically grounding or the realization of a pediatric specialist role represent these competencies acquired through training is an (although it may lead to the realization of the role by outstanding question. However, this does not hinder our encouraging caretakers to bring their children to Dr. ability to count specialists according to the ontological Petitcoeur for a health care encounter). Unlike identity data criteria outlined here. discussed in Hicks (2016), these data are corrigible. That is, OBI has classes training process and training objective these identity assertions can be verified and corrected by but no subclasses for specific types of training. Training something in the inter-subjective world such as documents process is a subclass of planned process and defined as “a from a licensing body, completion of training, or time spent process that achieves a training objective”, and training providing health care to children. Identification as a pediatric objective is a subclass of objective specification and defined cardiologist does not help discern an ontological analysis of as “An objective specification which is fulfilled by the a role, but it is worth considering here since it has been used provision of some training”. Taken together these definitions to generate counts of pediatric specialists. are circular and the genus of each definition ought to be changed, but these classes do begin to provide a formal 4.2 External grounds framework for describing kinds of training. More work is Some BFO-based ontologies provide a framework for needed to describe the relation between the physician who representing external grounds of roles, but none of them is has completed training and the training process. Simply 3 That which we call a pediatrician would by any other name a child treat (or not) stating that the physician is a participant of the training is not specifically, physicians can gain and lose permission or a sufficient since instructors also participate in the training right to practice. Finally, from the ontological representation process. Here too OBI has a framework that could be further provided by OOSTT, it seems that the same kinds of specified. Training service is a subclass of service and has processes realize both board eligible orthopedic surgeon roles part some training process. Furthermore, service is a kind of and board certified orthopedic surgeon roles. If this is correct, planned process and realizes both a service provider role and it underscores that types of roles are not sufficiently a service consumer role. Given a typology of training distinguished from each other by the kinds of processes that programs for medical specialties, a student of pediatric realize them, but that their origins and persistence conditions cardiology could be distinguished from an instructor of are also distinguishing characteristics that need to be taken pediatric cardiology as the bearer of the service consumer into account for a complete ontological representation. role that is realized in the training service specific to this The Informed Consent Ontology (Marshall et al., 2016) field. has a class authorization which is a subclass of planned In sum, both OOSTT and OBI have some of the process and is defined as “the process of makeing [sic] the necessary representations for specialist training for solving decision of the competent authorities in form of a letter, the counting problems of pediatric specialists, but the document, or verbal or electronic form, that confirms that definitions need work, typologies need to be fully fleshed out, somebody has permission to do something or be somewhere, and the representations integrated. e.g. to realize a given project.” and has specified output some authorization documentation, which is a subclass of 4.2.2. Board certification/registration/permission document and defined as “the documentation that is the Board certification in the USA and registration with the MBA output of the authorizationa [sic] process.” These definitions via the AHRPA in Australia are both mechanisms for taken together are circular, and while they describe the granting permission to persons to practice medicine and are creation of permissions, further work on the nature of the used for generating counts of pediatric cardiologists in permissions and their passing out of being through processes AHRPA-1, AHRPA-2, MBA, and ABP. such as revocation and expiration is needed. OOSTT has a class medical board certification which is the subclass of planned process and is defined as “the process 4.3 Realization by which a healthcare provider (physician, nurse, or other) in Many BFO-based ontologies already distinguish types of the United States demonstrates a mastery of basic knowledge roles based on the conditions of their realization. As we say and skills in a speciality of their occupation through written, in the previous section, this is not sufficient for distinguishing practical, or simulator-based testing.” This provides the role types, but it is necessary. In this section, we describe beginnings of a framework for representing health care various axes along which types of realization processes can provider roles in terms of board certification. For example, be distinguished. orthopedic surgeon role is the superclass of both board 4.3.1. Type of care actually provided eligible orthopedic surgeon role and board certified orthopedic surgeon role where the former is defined as “an Both AHPRA and the Freed survey from Freed et al. consider orthopedic surgeon role that is the outcome of fulfilling all whether a physician actually provides pediatric cardiology obligations to be allowed to take a board certification exam care in their counts. These counts are not of people who bear in orthopedic surgery” and the latter is the specified output of a pediatric cardiologist role, but of the number of people who some medical board certification process. This typology will have realized a pediatric cardiologist role within a specified enable representing, and therefore counting, all specialists period of time. This is an important distinction for who have completed training (as board eligible) and all ontologically representing data items in these data sets and specialists who have passed board exams and achieved achieving semantic integration. However, this is not certification. sufficient since each of these data sets accounts for the type Since board certification is specific to the USA, a of care provided differently. In AHPRA the survey broader account of permissions in general is required. Such respondent is asked to select areas of specialty care that they an account could likely be abstracted from OOSTT with have engaged in with minimal guidance regarding what these careful ontological analysis. More work also is needed to specialist labels mean. In the Freed survey, respondents were describe how these roles cease to exist. Since board asked about whether they engaged in specific diagnostic and certification can expire or be revoked, we need a way to therapeutic procedures such as performing and interpreting represent physicians who have had but no longer have board the results of echocardiograms and cardiac catheterizations. certification. Also, while we presume that a physician who This method of assessing the realization of the specialist role has passed the board is no longer a bearer of a board eligible is less vague and less ambiguous than the AHPRA’s method. role, this is not captured in the current representation in A full ontological analysis of processes that realize specialist OOSTT. Physicians can lose their roles, but more roles will require modeling diagnostic and therapeutic 4 That which we call a pediatrician would by any other name a child treat (or not) procedures as they relate to the realization of specialist roles. population and is realized by providing care to members of This work is outstanding in BFO-based ontologies. that population. 4.3.2. Type of patient actually treated 4.3.4 Quantity of care actually provided Providing cardiology care is not sufficient for the realization In addition to capturing whether the specialist role is realized, of a pediatric cardiologist role. The care also needs to be AHRPA and the Freed survey data take into account the provided to a child. That is, a health care encounter in which quantity of time during which these roles are realized. Again, cardiac care is provided must realize a patient role that this is an important distinction for ontologically representing inheres in a child to be sufficient for realizing a pediatric data items in these data sets and achieving semantic cardiology role. BFO-based ontologies can already represent integration. In AHRPA a pediatric cardiologist is somebody this. However, it is noteworthy that the Freed survey shows for whom providing pediatric cardiology care took up the that not all cardiac procedures performed on children are most or the second most amount of clinical time relative to performed by a physician with pediatric cardiology training. all other types of care. While Freed et al. do not offer a single (Arp & Smith, 2008) distinguishes between bearing a role meaning of ‘pediatric cardiologist’ (since the purpose of their and playing a role where it is possible to play a role without work is to show that different meanings produce different being the bearer of a role. However, clarifying this distinction counts), their survey data capture those who spend more than is required for a complete ontological account of specialist 50% of their clinical effort providing cardiac care and those roles. Such clarification needs to articulate what “play a role” who spend the majority of their time providing such care. means. As a dependent entity, a role cannot exist without There is currently no representation of quantities of clinical inhering in a bearer, so we assume that the locution “play a effort measured in time in BFO-based ontologies. role” is misleading. A general pediatrician Dr. Hari Cotvert 4.3.5 Work status can play a pediatric cardiologist role without being the bearer of such role, so it is unclear what is being “played”. If a Finally, some of these data sources indicate whether a pediatric cardiologist role is being played, it must inhere in physician’s work status is active, on leave, retired, research somebody, and since Dr. Cotvert is not the bearer of this role, not in clinic. The purpose of capturing work status is to Dr. Cotvert would be playing a role that inheres in somebody indicate availability in the work force, i.e., potential or else, which is odd, to say the least. Alternatively, Dr. Cotvert likelihood for a role to be realized. Describing work status might be playing a role that does not inhere in anybody at all, and the conditions necessary for a person to be in to realize a but then Dr. Cotvert is playing a role that could not exist. specialist role is outstanding in BFO-based ontologies. Since neither of these are ontologically coherent, we assume that there is no role that is actually being played, but that “to 5 CONCLUSIONS AND DISCUSSION play a role” means that something fulfills some Specialist terms such as ‘pediatric cardiologist’ are highly counterfactual conditions. Dr. Covert’s actions would realize ambiguous in authoritative data sets. We have provided a a pediatric cardiologist role if Dr. Cotvert were the bearer of review of the different criteria that have been used to count such a role. In the meantime, BFO-based ontologies can pediatric cardiologists in heterogeneous data sources and represent and count the number of people who are the bearer reviewed the extent to which BFO-based ontologies in the of some physician role and participate in the delivery of OBO Library can model these different criteria. While some cardiac care to a child given a typology of cardiac care. of the groundwork has been laid, more work remains to provide a robust and integrated representation of medical 4.3.3 Type of care delivered to type of patient specialist roles suitable for integrating data from We note that pediatric cardiologist role is only realized when heterogeneous data sources. the right kind of care (cardiac care) is delivered to the right kind of patient (a pediatric patient). Each of these criteria and ACKNOWLEDGEMENTS their existing ontological representations have been discussed Work on this paper was supported in part by the NIH/NCATS separately, so here we note that neither criterion alone is Clinical and Translational Science Awards to the University sufficient for describing the realization of a pediatric of Florida UL1 TR000064. The content is solely the cardiologist role. Furthermore, most, if not all, individual responsibility of the authors and does not necessarily specialty roles are realized by delivering health care to a represent the official views of the NIH. Thanks to Selja member of a particular population. A given cardiologist has Seppälä for helpful comments on previous drafts. a training and specialty in either pediatric or adult cardiology (and perhaps both), but does not have training and a specialty REFERENCES in cardiology in general. While ‘cardiologist role’ is a Allen, A. R., Doherty, R., Hilton, A. M., & Freed, G. L. (2016). reasonable superclass, every individual cardiologist role is Inconsistencies in authoritative national paediatric workforce data externally grounded in training with respect to a certain 5 That which we call a pediatrician would by any other name a child treat (or not) sources. Australian health review : a publication of the Australian Hospital Association. doi:10.1071/ah16096 Arp, R., & Smith, B. (2008). Function, role and disposition in basic formal ontology. Arp, R., Smith, B., & Spear, A. D. (2015). Building Ontologies with Basic Formal Ontology: MIT Press. Börner, K., Conlon, M., Corson-Rikert, J., & Ding, Y. (2012). VIVO: A semantic approach to scholarly networking and discovery. Synthesis lectures on the Semantic Web: theory and technology, 7(1), 1-178. Freed, G. L., Nahra, T. A., Wheeler, J. R., & Research Advisory Committee of American Board of, P. (2006). Counting physicians: inconsistencies in a commonly used source for workforce analysis. Acad Med, 81(9), 847-852. Hicks, A., Hanna, J., Welch, D., Brochhausen, M., & Hogan, W. R. (2016). The Ontology of Medically Related Social Entities: recent developments. J Biomed Semantics, 7(1), 1-4. doi:10.1186/s13326-016- 0087-8 Hogan, W. R., Garimalla, S., & Tariq, S. A. (2011). Representing the Reality Underlying Demographic Data. Paper presented at the ICBO-2011 International Conference on Biomedical Ontology: Proceedings of the 2nd International Conference on Biomedical Ontology, Buffalo, NY, USA. Marshall, S. A., Yang, C. C., Ping, Q., Zhao, M., Avis, N. E., & Ip, E. H. (2016). Symptom clusters in women with breast cancer: an analysis of data from social media and a research study. Qual Life Res, 25(3), 547- -557. doi:10.1007/s11136-015-1156-7 Peters, B., & Consortium, O. (2009). Ontology for biomedical investigations. Smith, B., Ashburner, M., Rosse, C., Bard, J., Bug, W., Ceusters, W., . . . Lewis, S. (2007). The OBO Foundry: coordinated evolution of ontologies to support biomedical data integration. Nat Biotechnol, 25(11), 1251. doi:10.1038/nbt1346 Utecht, J., Judkins, J., Brochhausen, M., Colvin Jr., T., Otte, J. N., Rogers, N., . . . Winchell, R. (2016). OOSTT: a Resource for Analyzing the Organizational Structures of Trauma Centers and Trauma Systems. Paper presented at the Joint International Conference on Biological Ontology and BioCreative (ICBO BioCreative 2016), Corvallis, OR. 6