Addressing Survivorship Bias in Neurocognitive Outcomes After Early Complex Cardiac Surgery Using Clustering and Propensity Scores

Abstract

Background

Although advances in cardiac surgery have increased survival rates from congenital heart disease, neurocognitive and functional outcomes have not significantly improved. We hypothesized that the absence of change in outcome scores may be due to survivorship bias. Our study aimed to address this by adjusting neurocognitive and functional outcome trend lines using k-mean clustering and propensity score (PS) methods.

Methods

Prospective follow-up was conducted on 266 children with single ventricle congenital heart disease who underwent the Norwood procedure at age ≤6 weeks at Stollery Children's Hospital, Edmonton, Alberta, between 1997 and 2016. PS and k-mean clustering methods were used to adjust outcomes for children with more complex conditions. Crude and adjusted trend lines for neurocognitive and functional outcomes were analyzed using multiple linear regression models.

Results

Multiple logistic regression determined age at surgery, total ventilation days, deep hypothermic circulatory arrest time, and total days chest open were significant in PS calculation. The adjusted linear time-trend analysis of neurocognitive and functional outcomes showed no change in Full Scale Intelligence Quotient and Visual Motor Integration scores. Although not robust to using the different PS adjustment methods, General Adaptive Composite scores may have decreased over time. Models with PS adjustment were not different from models without PS adjustment.

Conclusions

PS adjustment with clustering did not confirm survivorship bias. PS-adjusted models demonstrated stable Full Scale Intelligence Quotient and Visual Motor Integration scores over time, and inconsistent findings for General Adaptive Composite scores. Improved survival after more complex cardiac surgery was not associated with worse long-term neurocognitive outcomes.