Using electronic health records data for mortality predictive model development : study concept development and data preparation process /
Publications on mortality predictive models usually present statistical model development and validation in considerable details. The behind the scene data preparation process prior to fit a predictive model tends to be described sparsely. Electronic health records data are complex; hence, appropria...
| Main Author: | |
|---|---|
| Format: | eBook |
| Language: | English |
| Published: |
London :
SAGE Publications Ltd.,
2020.
|
| Series: | SAGE Research Methods Cases : Medicine and Health.
|
| Subjects: | |
| Online Access: | Connect to the full text of this electronic book |
| Summary: | Publications on mortality predictive models usually present statistical model development and validation in considerable details. The behind the scene data preparation process prior to fit a predictive model tends to be described sparsely. Electronic health records data are complex; hence, appropriate methods must be employed to ensure data accuracy and integrity, which are the foundations of any statistical model. This case study focuses on the clinical and empiric rationales and practical methods used to evaluate electronic numeric laboratory test results data that were used for mortality model development in our previous publication. Numeric laboratory data are used for mortality predictive models because they are objective, quantitative, and widely automated in the United States. The objective nature ensures the data reliability and reproducibility. The quantitative nature enables precise estimation of the graded relationship of the degree of physiologic derangement and the risk of mortality. However, raw laboratory data from multiple sites may need unit conversion, valid range, and data completeness checking by examining data distribution with a multidisciplinary team including clinicians in conjunction with literature review. Patients with a particular missing laboratory result may be considered as not clinically indicated, hence being grouped with the reference group. This treatment of missing data is consistent with common clinical practice and supported by the comparable mortality rate of patients with missing data and those in the reference group. Finally, converting the results from a logistic regression model to a risk scoring system makes it easier to implement in risk-adjusted outcome studies. |
|---|---|
| Physical Description: | 1 online resource. |
| Bibliography: | Includes bibliographical references and index. |
| ISBN: | 9781529743067 1529743060 |