Project Progress Report
Lin Yu
2022-08-08
Background
- 研究背景/假设：重医三院2017-2021年数据显示，因哮喘急性加重住院逐年减少；假设AECOPD住院率的下降与哮喘使用控制性药物增多，以及规范治疗增加有关；
研究背景/假设：重医三院2017-2021年数据显示，因哮喘急性加重住院逐年减少；假设AECOPD住院率的下降与哮喘使用控制性药物增多，以及规范治疗增加有关；
研究目的：
描述门诊哮喘患者数量自2017年至2021年的变化趋势；
探究使用哮喘药物与哮喘急性加重住院率之间的相关性。
Timeline
Background
Background
Title: Dynamic model updating (DMU) approach for statistical learning model building with missing data
Download DMU for statistical learning model building with missing data.pdf
Background
Title: Dynamic model updating (DMU) approach for statistical learning model building with missing data
Download DMU for statistical learning model building with missing data.pdf
Publication info:
Year: 2021
Author(s): Rahi Jain and Wei Xu, Dalla Lana School of Public Health, University of Toronto
Journal name: BMC Bioinformatics ISSN 1471-2105 (IF: 3.169)
Article type: methodology article
Some Techniques
Complete Cases Analysis (CCA)
- Easy to conduct, but provides biased estimates in cases when data is not missing completely at random (MCAR)
Imputation approach
Single imputation:
mean, median, etc.:underestimates the variance of the predictor and ignores the relationship between the predictors
regression: y = combination of linear other predictors,relies on the linear relationship
Multiple imputation: PMM handles MCAR and MAR types of missing data. It considers uncertainty (variance) (note: MICE package in R, for performing MI), in cases when missing values are in the tail of a distribution, predictive mean matching (PMM) have biased imputation.
Rationale
Rationale
Data collection is not simultaneous
Rationale
Data collection is not simultaneous
Instead of replacing the missing values with predicted value to compete the dataset
Rationale
Data collection is not simultaneous
Instead of replacing the missing values with predicted value to compete the dataset
Dynamic model updating(DMU) approach focuses on building the model on incomplete info rather than on imputed info.
How
Hierarchical clustering: it partitions the samples based on the distance of a sample with other samples.
The similar samples have lower distance among them as compared to dissimilar samples
Hierarchical clustering
