Predicting rapid decline in kidney function among type 2 diabetes patients: A machine learning approach
Background: Diabetic kidney disease (DKD) is one of the typical complications of type 2 diabetes (T2D), with approximately 10 % of DKD patients experiencing a Rapid decline (RD) in kidney function. RD leads to an increased risk of poor outcomes such as the need for dialysis. Albuminuria is a known k...
Saved in:
| Main Authors: | , , , , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Elsevier
2025-01-01
|
| Series: | Heliyon |
| Subjects: | |
| Online Access: | http://www.sciencedirect.com/science/article/pii/S2405844024165978 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1841526250588340224 |
|---|---|
| author | Eri Nakahara Kayo Waki Hisashi Kurasawa Imari Mimura Tomohisa Seki Akinori Fujino Nagisa Shiomi Masaomi Nangaku Kazuhiko Ohe |
| author_facet | Eri Nakahara Kayo Waki Hisashi Kurasawa Imari Mimura Tomohisa Seki Akinori Fujino Nagisa Shiomi Masaomi Nangaku Kazuhiko Ohe |
| author_sort | Eri Nakahara |
| collection | DOAJ |
| description | Background: Diabetic kidney disease (DKD) is one of the typical complications of type 2 diabetes (T2D), with approximately 10 % of DKD patients experiencing a Rapid decline (RD) in kidney function. RD leads to an increased risk of poor outcomes such as the need for dialysis. Albuminuria is a known kidney damage biomarker for DKD, yet RD cases do not always show changes in albuminuria, and the exact mechanism of RD remains unclear. Previous studies focused on a limited number of laboratory tests, no comprehensive study targeting a wide range of laboratory tests has been done. We target to develop a model that predicts RD among T2D and points to key laboratory tests of interest in understanding RD from various laboratory tests. Methods: Our machine learning model predicts whether RD, as represented via eGFR, will happen within 1 year. Additionally, the model uses Recursive feature elimination with cross-validation (RFECV) to eliminate the features that do not contribute to the prediction. We trained and assessed the model using 1202 types of laboratory tests from 3438 diabetes patients at the University of Tokyo Hospital. Result: The means (95 % confidence interval) of the receiver operating characteristic area under the curve (ROC-AUC), precision-recall area under the curve, accuracy rate, and F1-score of an 8-feature-model were 0.820 (0.811, 0.829), 0.430 (0.410, 0.451), 0.754 (0.747, 0.761), and 0.500 (0.485, 0.515), respectively. The RFECV revealed that 7 test types (MCH, γ-GTP, Cre, HbA1c, HDL-C, eGFR, and Hct) contributed to RD prediction. The model's ROC-AUC of 0.820 improves on the ROC-AUC of 0.775 seen in previous studies. Conclusion: The proposed model accurately predicts RD among diabetes patients and helps physicians focus on inhibiting progression of kidney damage. The contributing laboratory tests may serve as alternative biomarkers for DKD. |
| format | Article |
| id | doaj-art-1674c3ce6b7f4cd98e28bad25a6205e0 |
| institution | Kabale University |
| issn | 2405-8440 |
| language | English |
| publishDate | 2025-01-01 |
| publisher | Elsevier |
| record_format | Article |
| series | Heliyon |
| spelling | doaj-art-1674c3ce6b7f4cd98e28bad25a6205e02025-01-17T04:49:41ZengElsevierHeliyon2405-84402025-01-01111e40566Predicting rapid decline in kidney function among type 2 diabetes patients: A machine learning approachEri Nakahara0Kayo Waki1Hisashi Kurasawa2Imari Mimura3Tomohisa Seki4Akinori Fujino5Nagisa Shiomi6Masaomi Nangaku7Kazuhiko Ohe8Nippon Telegraph and Telephone Corporation, Japan; The University of Tokyo, JapanThe University of Tokyo, Japan; Corresponding author. 7-3-1 Hongo, Bunkyo-ku, Tokyo, 113-8655, Japan.Nippon Telegraph and Telephone Corporation, Japan; The University of Tokyo, JapanThe University of Tokyo, JapanThe University of Tokyo, JapanNippon Telegraph and Telephone Corporation, JapanNippon Telegraph and Telephone Corporation, JapanThe University of Tokyo, JapanThe University of Tokyo, JapanBackground: Diabetic kidney disease (DKD) is one of the typical complications of type 2 diabetes (T2D), with approximately 10 % of DKD patients experiencing a Rapid decline (RD) in kidney function. RD leads to an increased risk of poor outcomes such as the need for dialysis. Albuminuria is a known kidney damage biomarker for DKD, yet RD cases do not always show changes in albuminuria, and the exact mechanism of RD remains unclear. Previous studies focused on a limited number of laboratory tests, no comprehensive study targeting a wide range of laboratory tests has been done. We target to develop a model that predicts RD among T2D and points to key laboratory tests of interest in understanding RD from various laboratory tests. Methods: Our machine learning model predicts whether RD, as represented via eGFR, will happen within 1 year. Additionally, the model uses Recursive feature elimination with cross-validation (RFECV) to eliminate the features that do not contribute to the prediction. We trained and assessed the model using 1202 types of laboratory tests from 3438 diabetes patients at the University of Tokyo Hospital. Result: The means (95 % confidence interval) of the receiver operating characteristic area under the curve (ROC-AUC), precision-recall area under the curve, accuracy rate, and F1-score of an 8-feature-model were 0.820 (0.811, 0.829), 0.430 (0.410, 0.451), 0.754 (0.747, 0.761), and 0.500 (0.485, 0.515), respectively. The RFECV revealed that 7 test types (MCH, γ-GTP, Cre, HbA1c, HDL-C, eGFR, and Hct) contributed to RD prediction. The model's ROC-AUC of 0.820 improves on the ROC-AUC of 0.775 seen in previous studies. Conclusion: The proposed model accurately predicts RD among diabetes patients and helps physicians focus on inhibiting progression of kidney damage. The contributing laboratory tests may serve as alternative biomarkers for DKD.http://www.sciencedirect.com/science/article/pii/S2405844024165978Artificial intelligenceDiabetic kidney diseaseMachine learningRapid declineRecursive feature elimination |
| spellingShingle | Eri Nakahara Kayo Waki Hisashi Kurasawa Imari Mimura Tomohisa Seki Akinori Fujino Nagisa Shiomi Masaomi Nangaku Kazuhiko Ohe Predicting rapid decline in kidney function among type 2 diabetes patients: A machine learning approach Heliyon Artificial intelligence Diabetic kidney disease Machine learning Rapid decline Recursive feature elimination |
| title | Predicting rapid decline in kidney function among type 2 diabetes patients: A machine learning approach |
| title_full | Predicting rapid decline in kidney function among type 2 diabetes patients: A machine learning approach |
| title_fullStr | Predicting rapid decline in kidney function among type 2 diabetes patients: A machine learning approach |
| title_full_unstemmed | Predicting rapid decline in kidney function among type 2 diabetes patients: A machine learning approach |
| title_short | Predicting rapid decline in kidney function among type 2 diabetes patients: A machine learning approach |
| title_sort | predicting rapid decline in kidney function among type 2 diabetes patients a machine learning approach |
| topic | Artificial intelligence Diabetic kidney disease Machine learning Rapid decline Recursive feature elimination |
| url | http://www.sciencedirect.com/science/article/pii/S2405844024165978 |
| work_keys_str_mv | AT erinakahara predictingrapiddeclineinkidneyfunctionamongtype2diabetespatientsamachinelearningapproach AT kayowaki predictingrapiddeclineinkidneyfunctionamongtype2diabetespatientsamachinelearningapproach AT hisashikurasawa predictingrapiddeclineinkidneyfunctionamongtype2diabetespatientsamachinelearningapproach AT imarimimura predictingrapiddeclineinkidneyfunctionamongtype2diabetespatientsamachinelearningapproach AT tomohisaseki predictingrapiddeclineinkidneyfunctionamongtype2diabetespatientsamachinelearningapproach AT akinorifujino predictingrapiddeclineinkidneyfunctionamongtype2diabetespatientsamachinelearningapproach AT nagisashiomi predictingrapiddeclineinkidneyfunctionamongtype2diabetespatientsamachinelearningapproach AT masaominangaku predictingrapiddeclineinkidneyfunctionamongtype2diabetespatientsamachinelearningapproach AT kazuhikoohe predictingrapiddeclineinkidneyfunctionamongtype2diabetespatientsamachinelearningapproach |