Temporomandibular disorder (TMD) ranks second only to headache as the clinical condition most likely to cause craniofacial pain and dysfunction in the U.S. population. Yet in clinical practice, TMD arguably is the least understood and least effectively managed form of craniofacial pain and dysfunction, with treatment for many patients based on little more than symptomatic care. This discrepancy between the scope of suffering and paucity of effective, evidence-based care can be attributed in part to the fact that TMD is a highly heterogeneous disorder. Numerous different biological mechanisms may contribute to orofacial pain, and the most efficacious treatment is likely to depend on the mechanism that is causing the pain. Moreover, patients who do not meet the clinical criteria for TMD may nevertheless experience subclinical symptoms caused by these same biological mechanisms. Such patients may have elevated risk of developing first-onset TMD, and it may be possible to prevent the development of TMD in these patients by providing them with appropriate preventative therapy. Thus, it would be desirable to identify more homogeneous subgroups among TMD patients and TMD-free controls with comparable symptoms. Fortuitously, this goal can be accomplished without an expensive large-scale study. We have generated a large data set during the course of the Orofacial Pain: Prospective Evaluation and Risk Assessment (OPPERA;U01-DE017018-08) study, which is a large-scale prospective study designed to identify psychological, physiological, and genetic factors contributing to the onset and persistence of TMD. Our proposed study is to reanalyze the data collected in OPPERA using cluster analysis and other machine learning methods to identify clinically relevant subtypes of TMD and TMD-like symptoms. To achieve these goals, we will focus on the following specific aims: 1. Using cluster analysis, we will identify and validate latent constructs underlying pain symptoms that define subgroups of OPPERA participants. We hypothesize that we will identify novel subgroups of patients with TMD that differ in clinically and etiologically meaningful ways. We further hypothesize that TMD-free controls can be grouped into similar clusters. 2. Using discriminant analysis, logistic regression, and more advanced machine learning methods, we will develop rules for classifying TMD patients into subgroups. We hypothesize that we will develop simple classification criteria for the different subgroups of TMD that we identify in Specific Aim 1. 3. Using genetic association analysis, we will identify SNP's that are associated with subtypes of TMD. We hypothesize that we will identify SNP's associated with each subtype of TMD, offering additional biological insight into the mechanisms underlying different forms of TMD. !
Temporomandibular disorder (TMD) ranks second only to headache as the clinical condition most likely to cause craniofacial pain and dysfunction in the U.S. population, yet TMD arguably is the least understood and least effectively managed form of craniofacial pain. This discrepancy between the scope of suffering and paucity of effective, evidence-based care can be attributed in part to the fact that TMD is a highly heterogeneous disorder. The objective of this proposal is to identify subgroups of TMD and corresponding subgroups among TMD-free controls with similar symptoms, which may ultimately lead to personalized (and more effective) treatment for TMD and other chronic pain disorders.
|Gaynor, Sheila; Bair, Eric (2017) Identification of relevant subtypes via preweighted sparse clustering. Comput Stat Data Anal 116:139-154|
|Slade, G D; Ohrbach, R; Greenspan, J D et al. (2016) Painful Temporomandibular Disorder: Decade of Discovery from OPPERA Studies. J Dent Res 95:1084-92|
|Bair, Eric; Gaynor, Sheila; Slade, Gary D et al. (2016) Identification of clusters of individuals relevant to temporomandibular disorders and other chronic pain conditions: the OPPERA study. Pain 157:1266-78|
|Brownstein, Naomi C; Cai, Jianwen; Slade, Gary D et al. (2015) Parameter estimation in Cox models with missing failure indicators and the OPPERA study. Stat Med 34:3984-96|