Xie, Yunfei
(2019)
Assessing risk factors and predicting sepsis mortality using logistic and survival methods.
Master's Thesis, University of Pittsburgh.
(Unpublished)
Abstract
Identifying sepsis patients with high risk of death is crucial for making treatment decisions and has a public health significance. Sepsis mortality can be predicted by including clinical features and biomarkers in a predictive model. Hypotheses: (1) Clinical features combined with biomarkers would significantly enhance prediction power over clinical features alone; (2) time-trends of measurements contribute to prediction; (3) Cox proportional hazards model is more informative than logistic model.
Sepsis patients with complete data were identified from the Protocol-based Care in Early Septic Shock (ProCESS) trial. The trial obtained measurements at baseline (0 hours), 6 hours, and 24 hours of hospital admission, as well as patients’ within-60-day-of-admission death time. To evaluate biomarkers, logistic regressions with biomarkers and clinical features were compared to logistic regressions with clinical features only. To assess trends, at each time point, trends variables were evaluated in logistic regressions. To compare statistical models, landmark mortality within 3-day, 7-day, 14-day, and 60-day of admission were modeled using logistic regressions; a Cox model was developed to predict mortality over the same period. Areas under the Receiver Operating Characteristic curve (AUC) with bootstrap confidence intervals (CI) were used to evaluate model performance.
There were 528 patients included in baseline cohort (60-day mortality: 25%, mean age: 60 years, mean baseline lactate: 2.41 mmol/L), 534 patients in 6 hours cohort (24%, 60, 2.35), and 432 patients in 24 hours cohort (21%, 60, 2.26). At baseline, the AUC increased significantly from 0.766 [95% CI] = [0.710, 0.826] to 0.812 [0.749, 0.868] when biomarkers were added. In all models, trends were nonsignificant. For logistic models, 3-day model has AUC 0.888 [0.836, 0.939]; 7-day model has AUC 0.827 [0.776, 0.879]; 14-day model has AUC 0.858 [0.820, 0.895]; and 60-day model has AUC 0.795 [0.716, 0.835]. For the Cox model, the time-dependent AUC ranges between (0.859, 0.884).
Biomarkers provided incremental discrimination ability over clinical features alone to predict 60-day mortality at baseline. Trends of time-dependent variables did not increase predictive power. Logistic models and Cox models have similar predictive power in predicting short-term mortality but a Cox model is better in predicting long-term mortality.
Share
| Citation/Export: |
|
| Social Networking: |
|
Details
| Item Type: |
University of Pittsburgh ETD
|
| Status: |
Unpublished |
| Creators/Authors: |
|
| ETD Committee: |
|
| Date: |
26 July 2019 |
| Date Type: |
Publication |
| Defense Date: |
24 July 2019 |
| Approval Date: |
26 July 2019 |
| Submission Date: |
20 July 2019 |
| Access Restriction: |
No restriction; Release the ETD for access worldwide immediately. |
| Number of Pages: |
44 |
| Institution: |
University of Pittsburgh |
| Schools and Programs: |
School of Public Health > Biostatistics |
| Degree: |
MS - Master of Science |
| Thesis Type: |
Master's Thesis |
| Refereed: |
Yes |
| Uncontrolled Keywords: |
Sepsis, Mortality, Predictive Model |
| Date Deposited: |
26 Jul 2019 14:43 |
| Last Modified: |
26 Sep 2019 16:51 |
| URI: |
http://d-scholarship.pitt.edu/id/eprint/37107 |
Metrics
Monthly Views for the past 3 years
Plum Analytics
Actions (login required)
 |
View Item |
![[feed]](/images/feed-icon-32x32.png){kind=icon}