Risk prediction of knee osteoarthritis using artificial neural network

Abstract

Purpose: Knee osteoarthritis (OA) is the most common joint disease of adults worldwide. Since there has been no effective therapy, treatments of advanced radiographic knee OA are limited. Studies have shown that the early diagnosis and treatment of OA could help preventing aggravation of symptoms. Late diagnosis results in the socio-economic burden of illness associated with OA. Therefore, clinicians face a significant challenge of identifying patients who are at high risk of radiographic knee OA in a timely and appropriate way. This study aimed at the first development and validation of artificial neural network (ANN) model for radiographic knee OA risk prediction. Few reports have investigated the ability of ANN for knee OA risk prediction in a clinical manner. Logistic regression (LR), which is the gold standard method for analyzing binary medical data, is also used to compare its performance with that of ANN. Methods: ANN, one of the widely used approaches in machine learning, is an area of artificial intelligence technology and a mathematical system which mimic biological neural networks. The networks can be trained to recognize underlying patterns of diseases. ANN has been successfully used in medical decision support system. The 5th Korea National Health and Nutrition Examination Surveys (KNHANES V-1) data were used to develop ANN and LR models for radiographic knee OA. A logistic regression analysis was used to determine predictors for the models. The dataset were separated randomly into two independent groups, training and internal validation groups. The training group, comprised of two thirds (1777 participants) of the entire dataset, was used to construct the ANN and LR models. The internal validation group, comprised of one third (888 participants) of the entire dataset, was used to assess the ability to predict radiographic knee OA. In the KNHANES V-1, bilateral antero-posterior, lateral, and weight-bearing antero-posterior plain radiographs of knees were taken. Radiographic changes relating to OA were assessed using the Kellgren/Lawrence (KL) grade. The radiographic images were graded by trained two radiologists with concordant grades accepted. We defined radiographic knee OA as having KL grade ≥2 in one or both knees. Area under the curve (AUC) of the receiver operating characteristic (ROC), accuracy, sensitivity, and specificity were calculated to compare the performance of the two prediction models. Results: The ANN and LR were built using the independent predictors including sex, age, body mass index, educational status, hypertension, moderate physical activity, and knee pain. In the internal validation, ANN and LR predicted radiographic knee OA risk with AUC of 0.81 and 0.76, accuracy of 73.6% and 68.0%, sensitivity of 73.2% and 73.8%, and specificity of 73.9% and 65.1%, respectively. Figure 1 shows ROC curves of ANN and LR for radiographic knee OA in the internal validation group. ANN had significantly better AUC than LR. Conclusions: To our knowledge, we were the first to develop an ANN model for radiographic knee OA risk prediction using large population-based data. The ANN model might be a cost-effective screening tool identifying patients with untreated knee OA. These patients can then be received further evaluation such as knee radiograph and physical examination. The selected predictors including sex, age, body mass index, educational status, hypertension, moderate physical activity, and knee pain can be self-assessed or easily identified by the public health center. Therefore, the ANN can be easily used and might contribute to the advancement of clinical decision tools. Further studies should be targeted at constructing an extended prediction model for progressive knee OA through the collection of prospective data.