Validation, Multivariate Modeling, and the Construction of Heat-Map Prediction Matrices for Survival in the Context of Missing Data
| Autor: | Li Hua Yue, Shankar Srinivasan, Albert Elion-Mboussa |
|---|---|
| Rok vydání: | 2018 |
| Předmět: |
Multivariate statistics
Multivariate analysis Computer science Context (language use) Feature selection Missing data 03 medical and health sciences Variable (computer science) 0302 clinical medicine 030220 oncology & carcinogenesis Statistics Representation (mathematics) Categorical variable 030215 immunology |
| Zdroj: | Biopharmaceutical Applied Statistics Symposium ISBN: 9789811078194 |
| DOI: | 10.1007/978-981-10-7820-0_17 |
| Popis: | In non-interventional trials, there may be a large degree of missing data as investigators are not obliged to collect data on procedures and assessments which are not part of standard practice. When only 10–15% of the data is missing per variable, it can lead to 40% or higher of the patient records missing on at least one variable when there are many variables. This can present a problem for conventional multivariate analysis as modeling, variable selection and validation using only complete patient records leads to considerable loss of information and a strong bias in results toward those for whom we have 100% of the data. We will describe a series of steps involving multivariate modeling, variable selection, and internal and external validation leading finally to a heat-map representation of prediction based on the data from a non-interventional registry. Emphasis will be on the explication of this analysis route in the context of missing data. The construction of the heat-map prediction matrix will be demonstrated for the categorical endpoint of deaths within 180 days and 3-year survival post-enrollment for patients with newly diagnosed multiple myeloma. Relevant R and SAS codes are provided in the text. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|