1 / 20

An overview of risk-assessment algorithms for predicting stroke

An overview of risk-assessment algorithms for predicting stroke. Presented by Ke-Shiuan Lynn Ph.D. Definition of Stroke.

jada
Download Presentation

An overview of risk-assessment algorithms for predicting stroke

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. An overview of risk-assessment algorithms for predicting stroke Presented by Ke-Shiuan Lynn Ph.D.

  2. Definition of Stroke • The National Stroke Association defines stroke as "a cerebrovascular injury that occurs when blood flow to the brain is interrupted by a clogged or burst artery. The interruption deprives the brain of blood and oxygen, and causes brain cells to die." • Stroke is also called a "cerebrovascular accident" or CVA. 中研院生醫所潘文涵老師實驗室

  3. Statistics of Stroke (American) • Stroke is the thirdleading cause of death in the United States. • Stroke costs the United States $30 billion annually in health care costs. • Four out of five American families will be affected by stroke during their lifetime. • Every year 550,000 Americans, 50,000 Canadians and 32,000 Mexicans experience a new or recurrent stroke. • Stroke is the number one cause of adult disability. Three million Americans are currently permanently disabled because of stroke. 中研院生醫所潘文涵老師實驗室

  4. Statistics of Stroke (Taiwan) • Stroke is the secondleading cause (10.71%) of death in Taiwan (1992). ----- www.doh.gov.tw • Approximately 1.5% of the Taiwanese population over the age of 35 are diagnosed with stroke. ---- http://ntuh.mc.ntu.edu.tw/edu/health/649.htm • Approximately 0.3% of the Taiwanese population over the age of 35 are diagnosed with new onset of stroke annually. ---- http://ntuh.mc.ntu.edu.tw/edu/health/649.htm 中研院生醫所潘文涵老師實驗室

  5. Importance of Developing Risk Assessment Algorithm for Stroke • Reduce the risk of sudden death caused by stroke • Help physicians to choose antithrombotic therapy (e.g. warfarin or aspirin) more judiciously. • The by-product, crucial factors for stroke, may help medical professionals to obtain better insight about the cause of stroke. 中研院生醫所潘文涵老師實驗室

  6. Factors Associated with Stroke Framingham Study 中研院生醫所潘文涵老師實驗室

  7. Conventional Approach I– Deriving Risk Score Step 1: Use an univariate (e.g. logistic) model to assess the significance of each of the potential factors to stroke. Step 2: Use either a forward stepwise method or a backward stepwise method to select crucial factors into a multivariate model, e.g. logistic or Cox model. Step 3: Assign a suitable point to each of the factors according to the computed coefficients. Step 4: Sum up the points of the factors as the total score Step 5: Calibrate the computed score with actual event rate. 中研院生醫所潘文涵老師實驗室

  8. Conventional Approach I – Result • The computed score had Hosmer-Lemeshow statistics of 7.6 (20 or less indicate good calibration) • The c statistics was 0.66 中研院生醫所潘文涵老師實驗室

  9. Conventional Approach II– Deriving Prediction Rule Step 1: Choose candidate factors. Step 2: Select a factor randomly and calculate event rates for patients with and without the factor. Step 3: If the rates differ significantly, keep the factor as a criterion and use the patient group corresponding to higher event rate as the population of interest at the next level. Otherwise, discard the factor. Step 4: Repeat Step 1 to Step 3 until no remaining factors had statistically significant split. 中研院生醫所潘文涵老師實驗室

  10. Conventional Approach II– Result 中研院生醫所潘文涵老師實驗室

  11. Conventional Approach II – Result (cont.) 中研院生醫所潘文涵老師實驗室

  12. Possible Alternatives Factor selection and weighing – a supervised learning approach • Definition: Supervised learning is a technique of learning relationships between inputs and outputs from a set of (inputs, output) pairs when the corresponding outputs of the inputs are known. • Use an artificial neural network as classifier and apply either network growing (analog to forward stepwise) method or network pruning (analog to backward stepwise) to perform factor selection and weighting 中研院生醫所潘文涵老師實驗室

  13. How Does It Work? Architecture of a two-layer feedforward neural network Possible Factors outcome 中研院生醫所潘文涵老師實驗室

  14. v3  v2 v1 How Does It Work? (cont.) • In plain English: We tend to select minimal number of factors such that, in the space spanned by the selected factors, the factors computed from stroke patients can be separated from those computed from non-stroke subjects. v3 = diag (w) v1 中研院生醫所潘文涵老師實驗室

  15. Possible Alternatives (cont.) Factor selection and weighing – an unsupervised learning approach • Definition: unsupervised-learning techniques attempt to extract certain knowledge from a set of inputs without knowing their corresponding outputs. • The Principal Component Analysis (PCA) is to identify an m-dimensional subspace of the n-dimension input space that are most significant and then project the input data onto this space. In doing so, the original n variables are reduced to m variables that are mutually orthogonal (uncorrelated) in the resultant subspace. 中研院生醫所潘文涵老師實驗室

  16. Pros and Cons • Conventional Approach I: Pros: Significance of a factor can be individually identified. Cons: • The assessed risk is only statistically correct. • The selected factors may be correlated. • The selection of proper distributional form and the accuracy of measured onset time can affect the accuracy of the result. 中研院生醫所潘文涵老師實驗室

  17. Pros and Cons (cont.) • Conventional Approach II: Pros: • Significance of a factor can be individually identified. • Easy to implement. Cons: • The assessed risk is only statistically correct. • The selected factors may be correlated. • This approach may not suitable for multi-cluster data. 中研院生醫所潘文涵老師實驗室

  18. Pros and Cons (cont.) • Artificial Neural Network Pros: • Its capability of approximating complicated input-output relationship may lead to high accuracy. • The outcome can be either discrete or continuous. • No calibration process is needed. Cons: • The selected factors may be correlated. • Individual Significance of the factors can not be explicitly obtained. 中研院生醫所潘文涵老師實驗室

  19. Pros and Cons (cont.) • PCA Pros: the selected factors are uncorrelated Cons: • It requires considerable computational effort for eigenvectors and eigenvalues when the sample number is large. • The projected new factors may not be easy to interpret. • Need to design a classifier for the projected new factors 中研院生醫所潘文涵老師實驗室

  20. Questions 中研院生醫所潘文涵老師實驗室

More Related