Abstract: TH-PO037
Prospective Study of a Predictive Algorithm of Real-Time Fluid Status in Hemodialysis Patients
Session Information
- AI, Digital Health, Data Science - I
November 02, 2023 | Location: Exhibit Hall, Pennsylvania Convention Center
Abstract Time: 10:00 AM - 12:00 PM
Category: Augmented Intelligence, Digital Health, and Data Science
- 300 Augmented Intelligence, Digital Health, and Data Science
Authors
- Sandys, Vicki K., Royal College of Surgeons in Ireland, Dublin, Dublin, Ireland
- Bhat, Lavleen, Royal College of Surgeons in Ireland, Dublin, Dublin, Ireland
- Ninan, Anna, Royal College of Surgeons in Ireland, Dublin, Dublin, Ireland
- Doyle, Kevin, patientMpower, Dublin, Leinster, Ireland
- Conlon, Peter J., Beaumont Hospital Ireland, Dublin, Leinster, Ireland
- Sexton, Donal J., The University of Dublin Trinity College, Dublin, Leinster, Ireland
- O'Seaghdha, Conall M., Royal College of Surgeons in Ireland, Dublin, Dublin, Ireland
Background
An accurate per-session determination of volume status in dialysis patients would have considerable clinical utility.
Methods
We prospectively assessed a predictive algorithm for fluid status in an observational study of hemodialysis (HD) patients. Algorithm performance was compared to clinical volume assessment. Individualised predictions were compared with bioimpedance (BCM) measures at 2-week intervals. Internal validation for linear regression algorithms predicting BCM-based normohydration weight and pre-HD overhydration (OH) has been described (SA-PO351 ASN 2022). MAE, RMSE and Bland-Altman plots were used for continuous outcomes. Precision, recall and F1 score were used for fluid categories: overhydration ≥1.1L, normohydration -1.1L to 1.1L, underhydration ≤-1.1L.
Results
630 HD sessions were assessed. Root mean squared error (RMSE) for the pre-dialysis overhydration index=2.1kg compared to RMSE=1.6kg in the internal validation set. No significant difference was observed between the BCM normohydration weight versus the predicted value [mean difference 0.22 ± 2.25 kg, t (93) = 0.96, p=0.34]. A histogram showed a close alignment in distribution of predicted noromohydration weight and BCM values (Figure). Nursing staff overestimated fluid overload (Table).
Conclusion
The algorithm showed an ability to discriminate fluid categories compared to nursing staff, but overall accuracy was poor. Additional algorithm training in a larger, heterogeneous dataset would be expected to improve precision of fluid status predictions.
Fluid categories
Clinical | BCM | Predictions | |
% | % | % | |
Pre-dialysis OH | |||
≥1.1L | 70.8 | 36 | 65.96 |
-1.1L to 1.1L | 29.2 | 57.5 | 25.53 |
≤-1.1L | 0 | 6.4 | 8.52 |
Post-dialysis OH | |||
≥1.1L | 8.5 | 16 | |
-1.1L to 1.1L | 25.5 | 37.2 | |
≤-1.1L | 65.9 | 46.7 |
Histograms of BCM, predicted and clinical pre-dialysis OH and dry weight
Funding
- Commercial Support – Enteprise Ireland Disruptive Technologies Innovation Fund grant DTIF 2019_86