Acute kidney injury (AKI) is a common and critical condition in intensive care unit (ICU) settings, characterized by rapid onset, high rates of underdiagnosis, and poor clinical outcomes. While existing early prediction models predominantly focus on AKI stage 1, the acute dialysis quality initiative (ADQI) emphasizes that early identification of advanced stages (2 and 3) is of greater clinical significance due to their increased severity and substantially higher mortality rates. In this study, we develop and validate a dynamic early prediction model for AKI stages 2 and 3 using multivariate time-series data extracted from the MIMIC-III database. To enhance the model's robustness, we perform comprehensive data preprocessing, including resampling to address temporal irregularities and multiple imputation to handle missing values. The sequence lengths are standardized through truncation, and the baseline feature differences between patient groups are analyzed. First, we evaluate several machine learning models, including logistic regression, for direct prediction within 96 hours after admission to the ICU, with CatBoost demonstrating the highest performance. Subsequently, we propose a four-layer bidirectional long short-term memory (LSTM) network for dynamic risk assessment, which leverages temporal dependencies in clinical data and significantly outperforms static prediction approaches. Furthermore, we incorporate an interpretability analysis to identify key predictive variables, thus supporting clinical decision-making and allowing timely interventions for high-risk patients. Our results demonstrate the potential of deep learning models to improve the early detection of severe AKI, with implications for optimizing ICU management strategies.
Citation: Hui Chang, Qiannan Hou, Yueli Chen. Time-series analysis of acute kidney injury based on machine learning[J]. Big Data and Information Analytics, 2025, 9: 302-327. doi: 10.3934/bdia.2025014
Acute kidney injury (AKI) is a common and critical condition in intensive care unit (ICU) settings, characterized by rapid onset, high rates of underdiagnosis, and poor clinical outcomes. While existing early prediction models predominantly focus on AKI stage 1, the acute dialysis quality initiative (ADQI) emphasizes that early identification of advanced stages (2 and 3) is of greater clinical significance due to their increased severity and substantially higher mortality rates. In this study, we develop and validate a dynamic early prediction model for AKI stages 2 and 3 using multivariate time-series data extracted from the MIMIC-III database. To enhance the model's robustness, we perform comprehensive data preprocessing, including resampling to address temporal irregularities and multiple imputation to handle missing values. The sequence lengths are standardized through truncation, and the baseline feature differences between patient groups are analyzed. First, we evaluate several machine learning models, including logistic regression, for direct prediction within 96 hours after admission to the ICU, with CatBoost demonstrating the highest performance. Subsequently, we propose a four-layer bidirectional long short-term memory (LSTM) network for dynamic risk assessment, which leverages temporal dependencies in clinical data and significantly outperforms static prediction approaches. Furthermore, we incorporate an interpretability analysis to identify key predictive variables, thus supporting clinical decision-making and allowing timely interventions for high-risk patients. Our results demonstrate the potential of deep learning models to improve the early detection of severe AKI, with implications for optimizing ICU management strategies.
| [1] |
Beker BM, Corleto MG, Fieiras C, Musso CG, (2018) Novel acute kidney injury biomarkers: Their characteristics, utility and concerns. Int Urol Nephrol 50: 705–713. https://doi.org/10.1007/s11255-017-1781-x doi: 10.1007/s11255-017-1781-x
|
| [2] |
Perazella MA, Rosner MH, (2022) Drug-induced acute kidney injury. Clin J Am Soc Nephrol 17: 1220–1233. https://doi.org/10.2215/CJN.11290821 doi: 10.2215/CJN.11290821
|
| [3] |
Li JR, (2014) Acute kidney injury in the intensive care unit. Chin J Integr Tradit West Med Intensive Criti Care 3: 238–240. https://doi.org/10.3969/j.issn.1008-9691.2014.03.023 doi: 10.3969/j.issn.1008-9691.2014.03.023
|
| [4] |
Mehta RL, Cerdá J, Burdmann EA, Tonelli M, García-García G, Jha V, et al. (2015) International Society of Nephrology's 0 by 25 initiative for acute kidney injury (zero preventable deaths by 2025): A human rights case for nephrology. Lancet 385: 2616–2643. https://doi.org/10.1016/S0140-6736(15)60126-X doi: 10.1016/S0140-6736(15)60126-X
|
| [5] |
Luo XY, Zhou JC, Xi XM, (2017) Diagnosis of acute kidney injury: Serum creatinine or urine output?. Chin J Crit Care Med 3: 9–13. https://doi.org/10.3877/cma.j.issn.2096-1537.2017.01.004 doi: 10.3877/cma.j.issn.2096-1537.2017.01.004
|
| [6] |
Dong CX, Li Y, (2017) Epidemiology of AKI: Incidence, patient mortality, and renal mortality of AKI. Chin J Blood Purif 16: 8–10. https://doi.org/10.3969/j.issn.1671-4091.2017.01.003 doi: 10.3969/j.issn.1671-4091.2017.01.003
|
| [7] |
Hoste EAJ, Kashani K, Gibney N, Wilson FP, Ronco C, Goldstein SL, et al. (2016) Impact of electronic-alerting of acute kidney injury: Workgroup statements from the 15th ADQI Consensus Conference. Can J Kidney Health Dis 3: 101. https://doi.org/10.1186/s40697-016-0101-1 doi: 10.1186/s40697-016-0101-1
|
| [8] |
Dong GY, Qin JP, An YZ, Kang Y, Yu XY, Zhao MY, et al. (2020) Exploration of the necessity for further refinement of staging and typing in the KDIGO-AKIScr criteria for adult ICU patients: A retrospective analysis of a multicenter prospective study. Chin Crit Care Med 32: 313–318. https://doi.org/10.3760/cma.j.cn121430-20200218-00192 doi: 10.3760/cma.j.cn121430-20200218-00192
|
| [9] |
Li T, Yang Y, Huang J, Chen R, Wu Y, Li Z, et al. (2022) Machine learning to predict post-operative acute kidney injury stage 3 after heart transplantation. BMC Cardiovasc Disord 22: 288. https://doi.org/10.1186/s12872-022-02721-7 doi: 10.1186/s12872-022-02721-7
|
| [10] |
Tomašev N, Glorot X, Rae JW, Zielinski M, Askham H, Saraiva A, et al. (2019) A clinically applicable approach to continuous prediction of future acute kidney injury. Nature 572: 116–119. https://doi.org/10.1038/s41586-019-1390-1 doi: 10.1038/s41586-019-1390-1
|
| [11] |
Song YQ, Cai GY, Guo LX, Pan Q, (2022) Prognostic characteristics and new strategies for prevention and treatment of severe acute kidney injury. J Clin Internal Med 39: 361–364. https://doi.org/10.3969/j.issn.1001-9057.2022.06.001 doi: 10.3969/j.issn.1001-9057.2022.06.001
|
| [12] |
Duff S, Murray PT, (2020) Defining early recovery of acute kidney injury. Clin J Am Soc Nephrol 15: 1358–1360. https://doi.org/10.2215/CJN.13381019 doi: 10.2215/CJN.13381019
|
| [13] |
Jiang L, Zhu Y, Luo X, Wen Y, Du B, Wang M, et al. (2019) Epidemiology of acute kidney injury in intensive care units in Beijing: The multi-center BAKIT study. BMC Nephrol 20: 459–468. https://doi.org/10.1186/s12882-019-1660-z doi: 10.1186/s12882-019-1660-z
|
| [14] |
Sutherland SM, Chawla LS, Kane-Gill SL, Hsu RK, Kramer AA, Goldstein SL, et al. (2016) Utilizing electronic health records to predict acute kidney injury risk and outcomes: Workgroup statements from the 15th ADQI Consensus Conference. Can J Kidney Health Dis 3: 99. https://doi.org/10.1186/s40697-016-0099-4 doi: 10.1186/s40697-016-0099-4
|
| [15] |
Lngkvist M, Karlsson L, Loutfi A, (2014) A review of unsupervised feature learning and deep learning for time-series modeling. Pattern Recognit Lett 42: 11–24. https://doi.org/10.1016/j.patrec.2014.01.008 doi: 10.1016/j.patrec.2014.01.008
|
| [16] |
Bellomo R, Ronco C, Kellum JA, Mehta RL, Palevsky P, the ADQI workgroup, (2004) Acute renal failure—definition, outcome measures, animal models, fluid therapy and information technology needs: The Second International Consensus Conference of the Acute Dialysis Quality Initiative (ADQI) Group. Crit Care 8: R204. https://doi.org/10.1186/cc2872 doi: 10.1186/cc2872
|
| [17] |
Khwaja A, (2012) KDIGO clinical practice guideline for acute kidney injury. Nephron Clin Pract 120: c179–c184. https://doi.org/10.1159/000339789 doi: 10.1159/000339789
|
| [18] | Vagliano I, Lvova O, Schut MC, (2021) Interpretable and continuous prediction of acute kidney injury in the intensive care, In: Public Health and Informatics, Netherlands: IOS Press, 103–107. https://doi.org/10.3233/SHTI210129 |
| [19] |
Le S, Allen A, Calvert J, Palevsky PM, Braden G, Patel S, et al. (2021) Convolutional neural network model for intensive care unit acute kidney injury prediction. Kidney Int Rep 6: 1289–1298. https://doi.org/10.1016/j.ekir.2021.02.031 doi: 10.1016/j.ekir.2021.02.031
|
| [20] | Harrington P, (2013) Machine Learning in Action, Beijing: Posts & Telecommunications Press, 85–86. |
| [21] |
Brown SA, (2016) Patient similarity: Emerging concepts in systems and precision medicine. Front Physiol 7: 561. https://doi.org/10.3389/fphys.2016.00561 doi: 10.3389/fphys.2016.00561
|
Figures(13) / Tables(8)
Hui Chang, Qiannan Hou, Yueli Chen. Time-series analysis of acute kidney injury based on machine learning[J]. Big Data and Information Analytics, 2025, 9: 302-327. doi: 10.3934/bdia.2025014
DownLoad: