Abstract: PO2138
Network Science and Hemodialysis Patients' Kidney Transplant Attitudes
Session Information
- Transplantation: Clinical - Underrecognized Risk Factors, Traditional Considerations, and Outcomes
November 04, 2021 | Location: On-Demand, Virtual Only
Abstract Time: 10:00 AM - 12:00 PM
Category: Transplantation
- 1902 Transplantation: Clinical
Authors
- Gillespie, Avrum, Lewis Katz School of Medicine at Temple University, Philadelphia, Pennsylvania, United States
- Aljurbua, Rafaa, Temple University, Philadelphia, Pennsylvania, United States
- Obradovic, Zoran, Temple University, Philadelphia, Pennsylvania, United States
Background
Hemodialysis patients' attitudes towards kidney transplantation may depend on their local (egocentric) and overall characteristics of their clinic social network. We determine whether these local and overall social network characteristics improve machine learning (ML) logistic regression and neural network models of the patient's transplant attitudes.
Methods
We surveyed hemodialysis patients' social networks and transplant attitifudes in two hemodialysis clinics. We evaluated which ML model (logistic regression vs. neural network) best classified a patient's transplant attitude using survey and egocentric network data. Then, we tested whether multidimensional overall network information represented as a vector (node2vec) improved the model. Models were evaluated for accuracy, precision, recall, and F1-score using Python (version 6.1.4) and Gephi (0.9.2).
Results
The mean age of the 116 surveyed participants was 60 ± 13 years old. Half (55%) identified as male, and 75% identified as Black. Figure 1 shows the 83 participants (circles) who were in a clinic social network. The 33 network isolates are not shown. Network members with postive attitudes (57%) are the red circles. Green circles are those with negative attitudes. Adding egocentric network data improved the accuracy of the ML logistic regression model of transplant attitudes from 58% to 68% and the F1 score from 65% to 74%. The ML logistic regression model outperformed the neural network model in F1 score (73% vs. 66%) when including isolated participants. Addition of the overall network data (node2vec) further improved the F1-score of the ML logistic regression model to 77%.
Conclusion
The participant’s social network characteristics improved ML classification of the participant’s attitude towards kidney transplantation. The ML logistic regression model outperformed the neural network model, testing the limits of ML models on smaller data. Future research will examine how patient social networks disseminate information and affect attitudes and behaviors towards kidney transplantation.
Figure 1.
Funding
- NIDDK Support