ASN's Mission

To create a world without kidney diseases, the ASN Alliance for Kidney Health elevates care by educating and informing, driving breakthroughs and innovation, and advocating for policies that create transformative changes in kidney medicine throughout the world.

learn more

Contact ASN

1401 H St, NW, Ste 900, Washington, DC 20005

email@asn-online.org

202-640-4660

The Latest on X

Kidney Week

Abstract: FR-PO457

Use of a Novel Machine Learning Tool ("DoWhy") to Compare Mortality Risk Between High Volume Hemodiafiltration (HV-HDF) and Hemodialysis (HD) Patients in a Large Cohort from Latin America (LA)

Session Information

Category: Dialysis

  • 801 Dialysis: Hemodialysis and Frequent Dialysis

Authors

  • Guinsburg, Adrian M., Fresenius Medical Care Holdings Inc, Waltham, Massachusetts, United States
  • Diaz Bessone, Maria Ines, Fresenius Medical Care Holdings Inc, Waltham, Massachusetts, United States
  • Caseiro, Jorge M., Fresenius Medical Care Holdings Inc, Waltham, Massachusetts, United States
  • Hymes, Jeffrey L., Fresenius Medical Care Holdings Inc, Waltham, Massachusetts, United States
Background

Since hemodiafiltration (HDF) provides higher clearance of middle molecules compared with hemodialysis (HD), several studies showed HDF may result in better outcomes than HD. However, these findings are prone to residual confounding (C) and selection bias (SB). In this study we focused on a methodology that allows to analyze differences in clinical outcomes accounting C and SB.

Methods

We included incident patients (pts) to dialysis in Fresenius Medical Care units in LA between Jan 12-Dec 22, after 90 days of treatment. Pts were classified by modality: HD or HDF (>90% in any modality). High volume HDF (HV-HDF) was considered if infusion volume >21 liters. Values were compared using t-test or Chi-square accordingly. To compare risk of death and odds to reach targets, we use a novel machine learning tool (Python library “DoWhy”), which provides interface for causal inference analysis. A model using “modality” as treatment and baseline variables as covariates was built to estimate “average treatment effect” (ATE) on outcomes of interest using propensity score stratification (PSS), which estimates the risk/odd % change between treatment and control accounting for baseline differences and SB.

Results

99,496 pts were included (97.1% HD, 2.9% HDF, Table 1). HDF pts were younger, higher vintage and higher prevalence of male gender, diabetes, hypertension and heart disease. After PSS, ATE for mortality showed risk reduction for HDF vs HD of -4.04%, and -5.1% for HV-HDF vs HD. Odds to reach targets were increased on HV-HDF or HDF as compared to HD (Table 2).

Conclusion

Machine learning tools are an alternative when interventional studies are not available and there are unbalanced control/treated cohorts. In our study, after controlling for possible C and modelling using PSS, HDF and HV-HDF showed a reduced risk of dead and increased odds of reaching targets vs HD.

Tables 1&2