Abstract: FR-PO795

Use of a Validated Algorithm to Estimate Potentially Avoidable Readmissions in Hemodialysis Patients

Session Information

Category: Dialysis

  • 601 Standard Hemodialysis for ESRD

Authors

  • Sinclair, Matthew R., Northwell Health, Commack, New York, United States
  • Mathew, Anna, Northwell Health, Commack, New York, United States
  • Rosen, Lisa M., The Feinstein Institute for Medical Research, Manhasset, New York, United States
  • McLeggon, Jody-Ann, Northwell Health, Commack, New York, United States
  • Fishbane, Steven, Northwell Health, Commack, New York, United States
Background

Hemodialysis (HD) patients face up to 35% 30-day readmission, with associated anxiety and financial costs. There is no validated measure to identify potentially avoidable readmissions (PAR) in HD patients. In the general medicine population, the SqLape computer algorithm identifies PAR with 96% sensitivity and specificity. The aim of this study is to validate SqLape in the HD population.

Methods

We used chart review as the validation gold standard. We randomly selected 100 charts of HD patients with a 30-day readmission in Northwell Health, ending 9/30/2015. SqLape identified patients with a PAR and chart reviewers, blinded to algorithm results, conducted a simultaneous chart review. PAR chart review criteria was developed a. priori. To assess reliability of PAR classification by chart review, a second reviewer adjudicated a random 20% of charts. A 2x2 table cross classified algorithm and chart review identified PAR. Sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) were calculated. All analysis used SAS version 9.4.

Results

The kappa statistic for inter-rater reliability of readmission classification was 0.667 (substantial agreement). 19% and 72% of readmissions were classified as potentially avoidable by chart review and algorithm, respectively. The sensitivity and specificity of SqLape was 60.0% (95% CI: 26.2, 87.8) and 25.6 (95% CI: 13.5, 41.2), respectively. The PPV and NPV was 15.8% (95% CI: 6.0, 31.3) and 73.3% (95% CI: 44.9, 92.2), respectively (Table 1).

Conclusion

We assessed the use of an algorithm to identify PAR in HD patients. The algorithm performed well to identify the proportion of patients without a potentially avoidable readmission, with acceptable sensitivity and NPV. Specificity and PPV of the algorithm were poor, perhaps related to unique causes of PAR in HD patients(i.e fluid and electrolyte abnormalities, vascular access issues). Future studies should focus on development of an accurate measure of potentially avoidable readmissions in HD patients, as a foundational starting point for further study of this topic.