Eyettention II: A dual-sequence architecture for modeling fixation location, within-word landing position, and fixation duration in reading.
Authors: Deng S, Ding C, Reich DR, Prasse P, Jäger LA
Journal: Behavior research methods
mental health
psychology
open access
Abstract
Temporomandibular disorders (TMD) and orofacial pain (OFP) encompass a
heterogeneous group of musculoskeletal and pain-related conditions that affect
the masticatory system, temporomandibular joints (TMJs), and associated
structures [, ]. These conditions constitute a substantial global health
concern, with a recent meta-analysis estimating a symptom-based prevalence of
approximately 30–40% and an annual incidence of clinically verified first-onset
TMD of nearly 4% [, ]. A substantial subset of affected individuals develops
persistent pain, functional limitations, and psychosocial distress, with
projections indicating that global prevalence may reach nearly 44% by 2050 []. The introduction of standardized diagnostic criteria for TMD (DC/TMD) has
improved the reliability and consistency of diagnosis [, ]; however, diagnostic
classification alone has proven insufficient for guiding personalized treatment
strategies or predicting long-term outcomes [, ]. Consequently, clinicians
managing these patients encounter substantial prognostic uncertainty. Questions
such as “Which patient is at risk of developing chronic pain?”, “Who is likely
to respond to conservative treatment?”, and “When should a more intensive
approach be considered?” remain largely unanswered by current clinical resources
[, ]. Although some longitudinal investigations, most notably the Orofacial
Pain Prospective Evaluation and Risk Assessment (OPPERA) studies, have identified
risk factors and clinical characteristics that may inform the onset or
progression of TMD, these studies have yet to yield concrete, validated
diagnostic or prognostic tools suitable for individual-level clinical
decision-making [, ]. The multifactorial complexity of TMD, driven by numerous
interacting biological, behavioral, and psychosocial factors, further underscores
the challenge of translating population-level risk data into actionable clinical
predictions []. Beyond the identification of individual risk factors, prognosis in TMD/OFP
increasingly requires consideration of the dynamic, multidimensional interactions
among biological, psychological, behavioral, and social domains, as demonstrated
in contemporary biopsychosocial models and longitudinal cohort research [, , ].
Longitudinal evidence has shown that TMD/OFP are characterized by nonlinear
trajectories, symptom fluctuations, and heterogeneous treatment responses, rather
than predictable linear courses [, ]. Therefore, traditional single-variable
prognostic models and clinician intuition alone are often insufficient for
accurate individual-level prediction [, ]. The majority of existing TMD/OFP prognostic studies have relied on regression
models, including logistic and linear regression, to examine the relationships
between baseline predictors and future pain and function at the population level
[, , ]. While these models provide valuable information at the group level,
they are often poorly calibrated, meaning that predicted probabilities do not
closely correspond to observed outcome frequencies. This discrepancy may result
in overconfident or underconfident estimates and generally yield only moderate
performance for individual-level prediction of future pain and function at a
single time point [, ]. Furthermore, many studies focus on a single variable
or domain, such as psychological distress or pain intensity, while neglecting
other variables that may provide unique information from multiple data domains
and could be critical to the pathways underlying TMD/OFP chronicity or recovery.
This limitation constitutes a key motivation for employing machine learning (ML)
models, which have demonstrated improved predictive performance by capturing
interactions across multiple domains of data [, ].
heterogeneous group of musculoskeletal and pain-related conditions that affect
the masticatory system, temporomandibular joints (TMJs), and associated
structures [, ]. These conditions constitute a substantial global health
concern, with a recent meta-analysis estimating a symptom-based prevalence of
approximately 30–40% and an annual incidence of clinically verified first-onset
TMD of nearly 4% [, ]. A substantial subset of affected individuals develops
persistent pain, functional limitations, and psychosocial distress, with
projections indicating that global prevalence may reach nearly 44% by 2050 []. The introduction of standardized diagnostic criteria for TMD (DC/TMD) has
improved the reliability and consistency of diagnosis [, ]; however, diagnostic
classification alone has proven insufficient for guiding personalized treatment
strategies or predicting long-term outcomes [, ]. Consequently, clinicians
managing these patients encounter substantial prognostic uncertainty. Questions
such as “Which patient is at risk of developing chronic pain?”, “Who is likely
to respond to conservative treatment?”, and “When should a more intensive
approach be considered?” remain largely unanswered by current clinical resources
[, ]. Although some longitudinal investigations, most notably the Orofacial
Pain Prospective Evaluation and Risk Assessment (OPPERA) studies, have identified
risk factors and clinical characteristics that may inform the onset or
progression of TMD, these studies have yet to yield concrete, validated
diagnostic or prognostic tools suitable for individual-level clinical
decision-making [, ]. The multifactorial complexity of TMD, driven by numerous
interacting biological, behavioral, and psychosocial factors, further underscores
the challenge of translating population-level risk data into actionable clinical
predictions []. Beyond the identification of individual risk factors, prognosis in TMD/OFP
increasingly requires consideration of the dynamic, multidimensional interactions
among biological, psychological, behavioral, and social domains, as demonstrated
in contemporary biopsychosocial models and longitudinal cohort research [, , ].
Longitudinal evidence has shown that TMD/OFP are characterized by nonlinear
trajectories, symptom fluctuations, and heterogeneous treatment responses, rather
than predictable linear courses [, ]. Therefore, traditional single-variable
prognostic models and clinician intuition alone are often insufficient for
accurate individual-level prediction [, ]. The majority of existing TMD/OFP prognostic studies have relied on regression
models, including logistic and linear regression, to examine the relationships
between baseline predictors and future pain and function at the population level
[, , ]. While these models provide valuable information at the group level,
they are often poorly calibrated, meaning that predicted probabilities do not
closely correspond to observed outcome frequencies. This discrepancy may result
in overconfident or underconfident estimates and generally yield only moderate
performance for individual-level prediction of future pain and function at a
single time point [, ]. Furthermore, many studies focus on a single variable
or domain, such as psychological distress or pain intensity, while neglecting
other variables that may provide unique information from multiple data domains
and could be critical to the pathways underlying TMD/OFP chronicity or recovery.
This limitation constitutes a key motivation for employing machine learning (ML)
models, which have demonstrated improved predictive performance by capturing
interactions across multiple domains of data [, ].