Key result
Incorporating postdialysis EHR data into a Random Forests model improved near-term prediction of sudden cardiac death compared to predialysis data alone.
Why the study?
Does the inclusion of acute intra-dialytic changes captured in electronic health records improve near-term prediction of sudden cardiac death in older hemodialysis patients?
Cohort
Yes
Does the inclusion of acute intra-dialytic changes captured in electronic health records improve near-term prediction of sudden cardiac death in older hemodialysis patients?
Machine learning models incorporating acute intra-dialytic changes from electronic health records can modestly improve the near-term prediction of sudden cardiac death in older hemodialysis patients.
Answers are grounded in this paper and cite their sources.
May aid near-term sudden cardiac death risk stratification in older hemodialysis patients; leaves open prospective validation before clinical adoption.
BACKGROUND AND OBJECTIVES: Sudden cardiac death is the most common cause of death among individuals undergoing hemodialysis. The epidemiology of sudden cardiac death has been well studied, and efforts are shifting to risk assessment. This study aimed to test whether assessment of acute changes during hemodialysis that are captured in electronic health records improved risk assessment. DESIGN, SETTING, PARTICIPANTS, & MEASUREMENTS: Data were collected from all hemodialysis sessions of patients 66 years and older receiving hemodialysis from a large national dialysis provider between 2004 and 2008. The primary outcome of interest was sudden cardiac death the day of or day after a dialysis session. This study used data from 2004 to 2006 as the training set and data from 2007 to 2008 as the validation set. The machine learning algorithm, Random Forests, was used to derive the prediction model. RESULTS: In 22 million sessions, 898 people between 2004 and 2006 and 826 people between 2007 and 2008 died on the day of or day after a dialysis session that was serving as a training or test data session, respectively. A reasonably strong predictor was derived using just predialysis information (concordance statistic=0.782), which showed modest but significant improvement after inclusion of postdialysis information (concordance statistic=0.799, P<0.001). However, risk prediction decreased the farther out that it was forecasted (up to 1 year), and postdialytic information became less important. CONCLUSION: Subtle changes in the experience of hemodialysis aid in the assessment of sudden cardiac death and are captured by modern electronic health records. The collected data are better for the assessment of near-term risk as opposed to longer-term risk.
No takes yet. Share an insight, caveat, or question.
Goldstein et al. (2013) conducted a cohort in Hemodialysis. Prediction model using pre- and postdialysis EHR data vs. Predialysis information alone was evaluated on Sudden cardiac death the day of or day after a dialysis session (c-statistic 0.799, p=<0.001). Incorporating postdialysis EHR data into a Random Forests model improved near-term prediction of sudden cardiac death compared to predialysis data alone.
Synapse has enriched 10 related papers. Consider them for comparative context: