Paper
Learning a Consistent PRO-Outcomes Metric through KCCA for an Efficacy Assessing Model of Acupuncture
- Authors:
- Zhaohui Liang; Gang Zhang; Li Jiang; Wenbin Fu
- Abstract
- The efficacy of acupuncture for various disorders is assessed by patient-reported outcome (PRO) technique, thus questionnaires based on PRO were developed and applied in clinical practice and trials. The outcomes of many studies found that different PRO outcomes were inconsistent with each other on evaluating the clinical effect to the same patient. This phenomenon leads to great challenges for both real doctor’s diagnosis and computer aid studies on clinical evaluation. In this paper, we propose a machine learning model based on use Kernel Canonical Correlation Analysis (KCCA), which has been successfully applied in many statistical learning tasks. Through applying transformation on assessing outcomes and diagnostic information, the model can work as a potential approach to improve the correlation between effective outcomes and neck pain disorders with various diagnostic sub-types in clinical practice. To express non-linear relationship between different input variables, linear transformation in some feature space induced by kernel function is applied. We also extend the traditional KCCA algorithm to a multiple variables version aiming at analysis more than two measurements simultaneously. To show the effectiveness of the proposed algorithm, we construct a learning model to evaluate the acupuncture treatment effect for neck pain through both learnt metric and original PROs for comparison. The experimental data originate from a multi-center randomised control trial on acupuncture for cervical spondylosis neck pain, which was registered on Chinese Clinical Trial Registry (No: ChiCTR-TRC-00000184, http://www.chictr.org). The result shows the model is effective and practical.
- Keywords
- Metric Learning; Kernel Canonical Correlation Analysis; Feature Space Transformation; Therapeutic Effectiveness Evaluation; Cervical Spondylosis
- StartPage
- 79
- EndPage
- 88
- Doi