Abstract: TH-PO148
Detection of Medial Vascular Calcification in Patients with CKD Using Cost-Effective Classifiers
Session Information
- CKD-MBD: Clinical
October 24, 2024 | Location: Exhibit Hall, Convention Center
Abstract Time: 10:00 AM - 12:00 PM
Category: Bone and Mineral Metabolism
- 502 Bone and Mineral Metabolism: Clinical
Authors
- Lindholm, Bengt, Renal Medicine and Baxter Novum, Department of Clinical Science, Intervention and Technology, Karolinska Institutet, Stockholm, Sweden
- Bialonczyk, Urszula, Nalecz Institute of Biocybernetics and Biomedical Engineering, Polish Academy of Sciences, Warsaw, Poland
- Debowska, Malgorzata, Nalecz Institute of Biocybernetics and Biomedical Engineering, Polish Academy of Sciences, Warsaw, Poland
- Qureshi, Abdul Rashid Tony, Renal Medicine and Baxter Novum, Department of Clinical Science, Intervention and Technology, Karolinska Institutet, Stockholm, Sweden
- Bobrowski, Leon, Nalecz Institute of Biocybernetics and Biomedical Engineering, Polish Academy of Sciences, Warsaw, Poland
- Soderberg, Magnus, Pathology, Clinical Pharmacology and Safety Sciences, AstraZeneca R&D, Gothenburg, Sweden
- Stenvinkel, Peter, Renal Medicine and Baxter Novum, Department of Clinical Science, Intervention and Technology, Karolinska Institutet, Stockholm, Sweden
- Lukaszuk, Tomasz, Faculty of Computer Science, Bialystok University of Technology, Bialystok, Poland
- Poleszczuk, Jan T., Nalecz Institute of Biocybernetics and Biomedical Engineering, Polish Academy of Sciences, Warsaw, Poland
Background
Despite medial vascular calcification (mVC) being linked to high cardiovascular morbidity and mortality in chronic kidney disease (CKD) patients, identifying mVC remains challenging. This study evaluates the cost-effectiveness of various machine learning (ML) frameworks for mVC assessment using circulating biomarkers and traditional risk factors.
Methods
A dataset of 152 CKD patients without (n=93) or with (n=59) mVC was used as an input to classifiers combined with feature selection algorithms: 1) logistic regression (LR), 2) support vector machine (SVM), 3) random forest (RF), 4) elastic net (EN) and 5) relaxed linear separability (RLS). These methods selected features that inform about the mVC probability. Incremental cost-effectiveness ratio (ICER) was calculated for each framework, considering both performance and procedure costs. Prevalence rate and possible years gained were treated as unknown parameters.
Results
On the analyzed dataset, the number of features selected by the methods varied between 5 (LR) and 21 (SVM), but all classifiers offered similar predictive performance, Tab. 1. LR emerged as the most cost-effective option, Fig. 1.
Conclusion
The results provide novel insights into biomarkers related to mVC and support the use of ML algorithms as a complementary tool to imaging techniques. The study emphasizes considering cost-effectiveness when selecting classifiers, as minor performance improvements may not justify the additional costs of required inputs.
Tab. 1. The results.
| Logistic regression | Support vector machines | Random forest | Elastic net | Relaxed linear separability | |
| Accuracy | 0.74 | 0.71 | 0.74 | 0.76 | 0.77 |
| Selected features | age, copeptin, diabetes mellitus, sex, fat body mass index | age, copeptin, diabetes mellitus, choline, osteoprotegerin, sex, body mass index, fat body mass index, sclerostin, carboxy-terminal collagen crosslinks, desphospho-uncarboxylated MGP, homocysteine, IgM antibodies against phosphorylcholine, advanced glycation end products (skin autofluorescence), angiopoietin 2, undercarboxylated osteocalcin, high sensitivity C-reactive protein, insulin-like growth factor 1, IgM antibodies against malondialdehyde, lean body mass index, troponin T | age, copeptin, choline, osteoprotegerin, body mass index, sclerostin | age, copeptin, diabetes mellitus, choline, osteoprotegerin, sex, body mass index, sclerostin, desphospho-uncarboxylated MGP, homocysteine, IgM antibodies against phosphorylcholine | age, copeptin, diabetes mellitus, choline, osteoprotegerin, sex, fat body mass index, carboxy-terminal collagen crosslinks, apolipoprotein B1, free triiodothyronine, triglycerides, carboxylated osteocalcin, pentraxin-related protein, trimethylamine N-oxide, thyroid-stimulating hormone, uric acid |
Fig. 1. ICER with the respect to mVC prevalence.