Views: 1,386 | Downloads:
28
| Responses: 0
| XML | Respond to this article | Alert & updates | Request permissions | Email to a friend |
Original ArticleOpen Access
Prediction of Low Bone Mineral Density in Postmenopausal Women by Artificial Neural Network Model Compared to Logistic Regression Model
Ongphiphadhanakul B 
,
Rajatanavin R ,
Chailurkit L ,
Piaseu N ,
Teerarungsikul K ,
Sirisriro R ,
Komindr S ,
Pauvilai G
,
Rajatanavin R ,
Chailurkit L ,
Piaseu N ,
Teerarungsikul K ,
Sirisriro R ,
Komindr S ,
Pauvilai G Measuring bone mineral density (BMD) is currently the best modality to diagnose osteoporosis
and predict future fractures. The use of risk factors to predict BMD and fracture risk has
been considered to be inadequate for precise diagnostic purpose, but it may be helpful as a screening
tool to determine who actually needs BMD assessment. Recently, artificial neural network
(ANN), a nonlinear computational model, has been used in clinical diagnosis and classification.
In the present study, we evaluated the risk factors associated with low BMD in Thai postmenopausal
women and assessed the prediction of low BMD using an ANN model compared to
a logistic regression model. The subjects consisted of 129 Thai postmenopausal women divided
into 2 groups, 100 subjects in the training set and the remaining 29 subjects in the validation
set. The subjects were classified as having either low BMD or normal BMD by using BMD value
1 SD lower than the mean value of young adults as the cutoff point. Decreased body weight,
decreased hip circumference and increased years since menopause were found to be associated
with low BMD at the lumbar spine by logistic regression. For the femoral neck, increased age
and decreased urinary calcium were associated with low BMD. The models had a sensitivity of
85.0 per cent, a specificity of 11.1 per cent and an accuracy of 62.0 per cent for the diagnosis of
low BMD at the lumbar spine when tested in the validation group. For the femoral neck, the
sensitivity, specificity and accuracy were 90.5 per cent, 12.5 per cent, and 69.0 per cent, respectively.
Models based on ANN correctly classified 65.5 per cent of the subjects in the validation
group according to BMD at the lumbar spine with a sensitivity of 80.0 per cent and a specificity
of 33.3 per cent while it correctly classified 58.6 per cent of the subjects at the femoral
neck with a sensitivity of 76.2 per cent and a specificity of 12.5 per cent. There was no significant
difference in terms of accuracy, sensitivity and specificity in the prediction of low BMD at
the lumbar spine or the femoral neck between ANN model and logistic regression model. We
concluded that ANN does not perform better than convention statistical methods in the prediction
of low BMD. The less than perfect performance of the prediction rules used in the prediction of
low BMD may be due to the lack of adequate association between the commonly used risk
factors and BMD rather than the nature of the computational models.
and predict future fractures. The use of risk factors to predict BMD and fracture risk has
been considered to be inadequate for precise diagnostic purpose, but it may be helpful as a screening
tool to determine who actually needs BMD assessment. Recently, artificial neural network
(ANN), a nonlinear computational model, has been used in clinical diagnosis and classification.
In the present study, we evaluated the risk factors associated with low BMD in Thai postmenopausal
women and assessed the prediction of low BMD using an ANN model compared to
a logistic regression model. The subjects consisted of 129 Thai postmenopausal women divided
into 2 groups, 100 subjects in the training set and the remaining 29 subjects in the validation
set. The subjects were classified as having either low BMD or normal BMD by using BMD value
1 SD lower than the mean value of young adults as the cutoff point. Decreased body weight,
decreased hip circumference and increased years since menopause were found to be associated
with low BMD at the lumbar spine by logistic regression. For the femoral neck, increased age
and decreased urinary calcium were associated with low BMD. The models had a sensitivity of
85.0 per cent, a specificity of 11.1 per cent and an accuracy of 62.0 per cent for the diagnosis of
low BMD at the lumbar spine when tested in the validation group. For the femoral neck, the
sensitivity, specificity and accuracy were 90.5 per cent, 12.5 per cent, and 69.0 per cent, respectively.
Models based on ANN correctly classified 65.5 per cent of the subjects in the validation
group according to BMD at the lumbar spine with a sensitivity of 80.0 per cent and a specificity
of 33.3 per cent while it correctly classified 58.6 per cent of the subjects at the femoral
neck with a sensitivity of 76.2 per cent and a specificity of 12.5 per cent. There was no significant
difference in terms of accuracy, sensitivity and specificity in the prediction of low BMD at
the lumbar spine or the femoral neck between ANN model and logistic regression model. We
concluded that ANN does not perform better than convention statistical methods in the prediction
of low BMD. The less than perfect performance of the prediction rules used in the prediction of
low BMD may be due to the lack of adequate association between the commonly used risk
factors and BMD rather than the nature of the computational models.
Download:
PDF