Aims: To consider temporomandibular joint (TMJ) anatomic interactions
in order to refine hard tissue models differentiating (1)
joints diagnosed with disc displacement with reduction (DDwR) or
without reduction (DDw/oR) from asymptomatic joints
(Normals), and (2) DDwR joints from DDw/oR joints.
Methods:TMJ tomograms of 84 women with unilateral DDwR
and 78 with unilateral DDw/oR were compared against each other
and against those of 42 female Normal joints through the use of
14 linear and angular measurements, 8 ratios, and 34 interactions.
A classification tree model for each comparison was tested for fit
with sensitivity, specificity, accuracy, and log likelihood and compared
to logistic regression models. Results: In the classification
tree model comparison, the DDwR model versus the Normal
model realized 35.9% log likelihood (88.0% sensitivity, 66.7%
specificity); the DDw/oR model versus the Normal model realized
38.8% log likelihood (69.6% sensitivity, 85.7% specificity). The
DDwR model versus the DDw/oR model realized 33.3% log likelihood
(76.0% sensitivity, 73.1% specificity). In the logistic regression
model comparison, the DDwR model versus the Normal
model realized 40.8% log likelihood (82.1% sensitivity, 78.6%
specificity) and the DDw/oR model versus the Normal model realized
61.1% log likelihood (85.9% sensitivity, 90.5% specificity).
The DDwR model versus the DDw/oR model realized 21.5% log
likelihood (60.3% sensitivity, 79.8% specificity). The addition of
interactions to the logistic regression models improved the previously
published log likelihood from 99% to 149%. Conclusion:
The interactions improved logistic regression models and the data
suggest that anatomic characteristics influence joint functional status.
Because the models incorporated nearly all considered
anatomic measurements, no anatomic factor is redundant in the
closed TMJ biological system. J OROFAC PAIN 2004;18:192–202