Prolonged hospital length of stay after pediatric heart transplantation: A machine learning and logistic regression predictive model from the Pediatric Heart Transplant Society

https://doi.org/10.1016/j.healun.2022.05.016Get rights and content

Background

Heart transplantation (HT) is the gold standard for managing end-stage heart failure. Multiple quality metrics, including length of stay (LOS), have been used in solid organ transplantation. However, limited data are available regarding trends and factors influencing LOS after pediatric HT. We hypothesized that various donor, peri-transplant and recipient factors affect LOS after pediatric HT.

Methods

We analyzed patients <18years at time of HT from January 2005 to December 2018 in the Pediatric Heart Transplant Society database, and examined LOS trends, defined prolonged LOS (PLOS = LOS>30days after HT), identified factors associated with PLOS and assessed outcomes.

Results

Of 4827 patients undergoing HT, 4414 patients were discharged and included for analysis. Overall median LOS was 19days[13,34]. Median LOS was longer in patients with congenital heart disease(CHD = 25days[15,43] than with cardiomyopathy(CM = 17days[12,27] across all ages. Median LOS in age <1year was 26-days[16,45.5] and in age >10year was 16days[11,26]. PLOS was seen in 1313 patients(30%). Patients with PLOS were younger, smaller and had longer CPB times. There was no difference in utilization of VAD at HT between groups, however, ECMO use at listing(8.45% vs 2.93%,p < 0.05) and HT was higher in the PLOS group(9.22% vs 1.58%,p < 0.05). PLOS was more common in patients with previous surgery, CHD, single ventricle physiology, recipient history of cardiac arrest or CPR, end organ dysfunction, lower GFR, use of mechanical ventilation at HT and Status 1A at HT.

Conclusion

We present novel findings of LOS distribution and define PLOS after pediatric HT, providing a quality metric for individual programs to utilize and study in their practice.

Section snippets

Materials and methods

PHTS maintains a multicenter, prospective, event‐driven database for pediatric patients (<18 years) listed for HT. Collection of date of discharge began in 2005, therefore data query spanned January 2005-December 2018, and included patients from multiple institutions (Table S1). Institutional Review Board approval was obtained at each institution. Information was collected on demographics and other patient data at listing, patient and procedural data at HT, and event data surrounding listing,

Results

A total of 4827 patients received a HT during the study period. Of these, 413 were excluded as described in Figure 2 leaving 4414 patients for analysis.

Discussion

Our study is the first to provide insight into LOS trends after pediatric HT. The use of LOS as a quality metric is clearly important as it encompasses various domains of quality.1,24,25 Furthermore, an improved understanding of factors affecting LOS can allow transplant centers to estimate their expected LOS, identify patients at risk for PLOS and counsel families accurately. Due to a paucity of LOS data in pediatric HT, heterogeneity of diagnoses and unique characteristics of pediatric

Limitations

Limitations of our study include use of 30 days as the cutoff for PLOS. This was chosen after a careful analysis of the LOS distribution of the overall cohort. Knowing these limitations, we chose this cutoff as it also aligns well with other clinically important parameters including 30-day survival and readmission data. This study is additionally subject to limitations inherent to analyses of registry data, as they are bound by preexisting variables. However, PHTS provides a comprehensive,

Conclusion

In summary, we present novel findings of LOS distribution and define PLOS after pediatric HT. This study is the first to inform about factors influencing LOS and develop a prediction model for PLOS. Furthermore, this provides a quality metric for individual programs to utilize and further study trends in their own practice. This tool will allow better risk stratification and counseling about expected LOS. Future studies are needed to assess the accuracy of this prediction model and guide

Author contributions

All authors contributed to study design and data analysis. DG drafted the manuscript and all authors assisted with writing and critical revision of the manuscript.

Foundation

This study was supported by the Pediatric Heart Transplant Society.

Disclosure statement

None

Acknowledgments

The authors thank the other investigators, staff, and statisticians at the University of Alabama at Birmingham, Alabama, and the Pediatric Heart Transplant Society

References (28)

  • Tianqi C, Carlos G. XGBoost: a scalable tree boosting system %@ 9781450342322 %U...
  • L. Breiman

    Random forests

    Machine Learning

    (2001)
  • L. Breiman

    Bagging predictors

    Machine Learning

    (1996)
  • EW Steyerberg et al.

    Assessing the performance of prediction models: a framework for traditional and novel measures

    Epidemiology (Cambridge, Mass)

    (2010)
  • Cited by (4)

    View full text