Abstract
Background and Aims
We aimed to investigate if advanced analytical methods could out-perform existing risk scores for predicting the risk of Stroke Associated Pneumonia (SAP) within the first 7 days after hospital admission.
Methods
Data from the UK Sentinel Stroke National Audit Program (SNAPP) between 2013 to June 2019 were used. XGBoost, Logistic Regression (LR) with elastic net, and LR with elastic net and interaction term models with 30 variables were developed using 80% randomly selected admissions from 2013 to 2018, internally validated on the 20% remaining admissions, and temporally validated on all admissions from 2019. The prestroke Independence [modified Rankin scale], Sex, Age, National Institutes of Health Stroke Scale (ISAN) score was used as reference model. Performance of the risk prediction models was evaluated in terms of accuracy (Brier score), discrimination, calibration, and net reclassification and Decision-curve analyses.
Results
Data from 483, 561 patients were used, with an overall 7-day SAP rate of 8.59%. All developed models outperformed ISAN (area under the ROC curve (AUC) 0.740, 95% confidence interval [CI]: 0.732 to 0.749) on the 2019 temporal validation set with XGBoost achieving the highest AUC (0.782, 95% CI: 0.775 to 0.790). More results will be presented at the conference.
Conclusions
Risk prediction of SAP was improved with state-of-the-art risk prediction methods using more variables compared to the existing ISAN score. The gain in accuracy from complex prediction models may not be enough to justify their use in clinical settings compared to a simple risk score that can be calculated without computers.
We aimed to investigate if advanced analytical methods could out-perform existing risk scores for predicting the risk of Stroke Associated Pneumonia (SAP) within the first 7 days after hospital admission.
Methods
Data from the UK Sentinel Stroke National Audit Program (SNAPP) between 2013 to June 2019 were used. XGBoost, Logistic Regression (LR) with elastic net, and LR with elastic net and interaction term models with 30 variables were developed using 80% randomly selected admissions from 2013 to 2018, internally validated on the 20% remaining admissions, and temporally validated on all admissions from 2019. The prestroke Independence [modified Rankin scale], Sex, Age, National Institutes of Health Stroke Scale (ISAN) score was used as reference model. Performance of the risk prediction models was evaluated in terms of accuracy (Brier score), discrimination, calibration, and net reclassification and Decision-curve analyses.
Results
Data from 483, 561 patients were used, with an overall 7-day SAP rate of 8.59%. All developed models outperformed ISAN (area under the ROC curve (AUC) 0.740, 95% confidence interval [CI]: 0.732 to 0.749) on the 2019 temporal validation set with XGBoost achieving the highest AUC (0.782, 95% CI: 0.775 to 0.790). More results will be presented at the conference.
Conclusions
Risk prediction of SAP was improved with state-of-the-art risk prediction methods using more variables compared to the existing ISAN score. The gain in accuracy from complex prediction models may not be enough to justify their use in clinical settings compared to a simple risk score that can be calculated without computers.
| Original language | English |
|---|---|
| Article number | O0172/#1179 |
| Pages (from-to) | 103-103 |
| Number of pages | 1 |
| Journal | European Stroke Journal |
| Volume | 6 |
| Issue number | Issue 1_suppl |
| DOIs | |
| Publication status | Published - 1 Sept 2021 |
