Skip to main content
Download PDF

Short-term prognostic models for severe acute kidney injury patients receiving prolonged intermittent renal replacement therapy based on machine learning

BMC Medical Informatics and Decision Making volume 23, Article number: 133 (2023) Cite this article

Abstract

Background

As an effective measurement for severe acute kidney injury (AKI), the prolonged intermittent renal replacement therapy (PIRRT) received attention. Also, machine learning has advanced and been applied to medicine. This study aimed to establish short-term prognosis prediction models for severe AKI patients who received PIRRT by machine learning.

Methods

The hospitalized AKI patients who received PIRRT were assigned to this retrospective case-control study. They were grouped based on survival situation and renal recovery status. To screen the correlation, Pearson’s correlation coefficient, partial ETA square, and chi-square test were applied, eight machine learning models were used for training.

Results

Among 493 subjects, the mortality rate was 51.93% and the kidney recovery rate was 30.43% at 30 days post-discharge, respectively. The indices related to survival were Sodium, Total protein, Lactate dehydrogenase (LDH), Phosphorus, Thrombin time, Liver cirrhosis, chronic kidney disease stage, number of vital organ injuries, and AKI stage, while Sodium, Total protein, LDH, Phosphorus, Thrombin time, Diabetes, peripherally inserted central catheter and AKI stage were selected to predict the 30-day renal recovery. Naive Bayes has a good performance in the prediction model for survival, Random Forest has a good performance in 30-day renal recovery prediction model, while for 90-day renal recovery prediction model, it’s K-Nearest Neighbor.

Conclusions

Machine learning can not only screen out indicators influencing prognosis of AKI patients receiving PIRRT, but also establish prediction models to optimize the risk assessment of these people. Moreover, attention should be paid to serum electrolytes to improve prognosis.

Peer Review reports

What’s already known about this topic?

  • PIRRT is gradually applied in AKI patients due to its combination of different hemodialysis modes.

  • As a result of artificial intelligence advancement, machine learning has been applied in AKI patients to establish a warning model.

Research gap

  • Studies establishing prognosis prediction models of AKI patients receiving PIRRT by machine learning are scarce. Our research filled a research gap of prognosis prediction models for these patients.

What does this article add?

  • Sodium, Total protein, Lactate dehydrogenase (LDH), Phosphorus, Thrombin time, Liver cirrhosis, chronic kidney disease stage, number of vital organ injuries, and AKI stage were associated with mortality, while Sodium, Total protein, LDH, Phosphorus, Thrombin time, Diabetes, peripherally inserted central catheter and AKI stage were selected to predict the 30-day renal recovery.

  • Naive Bayes has a good performance in the prediction model for survival, Random Forest has a good performance in 30-day renal recovery prediction model, while for 90-day renal recovery prediction model, it’s K-Nearest Neighbor.

Introduction

Acute kidney injury (AKI) is one of the most common clinical issues worldwide. AKI occurs in approximately 10–15% of patients admitted to hospital, while its incidence in intensive care has been reported in more than 50% of patients [1, 2].Severe AKI is a common fatal disease that seriously affects patients’ prognosis. Several multinational and multicenter international cross-sectional studies have confirmed that the increase in AKI severity is related to hospital mortality [3, 4]. Renal replacement therapy (RRT) is an essential and highly efficient action for severe AKI. As a hybrid RRT mode, prolonged intermittent renal replacement therapy (PIRRT) has been gradually applied in clinical practice because of its combination of continuous renal replacement therapy (CRRT) and intermittent hemodialysis (IHD) [5, 6]. However, only a few studies have evaluated the prognosis prediction model of patients receiving PIRRT.

The advancement of machine learning has led to the emergence of risk prediction models based on artificial intelligence in medicine area. Some studies have attempted to establish a multidisciplinary and cross-ward unified early warning model for AKI [7, 8]. However, the use of machine learning to establish prognosis prediction models for PIRRT is rare.

Thus, the research aimed to concentrated on the indicators before PIRRT onset, build short-term prognosis prediction models for severe AKI patients receiving PIRRT by machine learning, provide new ideas for risk assessment and ensure timely treatment of high-risk patients, which is of great significance to improve prognosis.

Methods

Population

AKI patients who received PIRRT from January 2012 to October 2018 at the Shanghai General Hospital, Shanghai Jiao Tong University School of Medicine, Shanghai, China, were included. They were grouped based on 30-day survival situation. The surviving patients were divided into kidney recovery and non-recovery groups at 30- and 90-day post-discharge, respectively. The recovery of renal function was defined as decreased Scr combined with discontinuation of any form of RRT or urine volume > 800 ml/24 h [9]. The follow-up was carried out by phone and outpatient visits. The Ethics Committee of Shanghai General Hospital, Shanghai Jiao Tong University School of Medicine, approved the study. The need for informed individual consent was waived by this committee.

Inclusion criteria

① AKI patients (according to the 2012 Kidney Disease: Improving Global Outcomes (KDIGO) definition of AKI [1]); ② PIRRT was treated. We implemented common PIRRT scheme [10] with a Prismaflex M60 CRRT system and a 0.6 m2 AN69 dialysis membrane of Baxter, Deerfield, IL, USA. Patients underwent PIRRT 3–7 times a week with a duration of 4–6 h per time.

Exclusion criteria

① Chronic kidney disease (CKD) stage 4/5 patients or maintenance dialysis patients; ② Other RRT modes; ③ Out of contact; ④ Data missing, including cases of low data quality and significant data(creatinine, urine volume) loss. For cases of low data quality, we defined it as cases with the absence of ≥ 20% of all research indicators, and patients who died within 48 h after the first PIRRT.

Observed variables

As we reported before, the indicators we focused on include [10]: ① Basic information: gender, age, AKI etiology and stage; ② History of important chronic diseases: Hypertension, Diabetes, Coronary heart disease(CHD), CKD, Chronic obstructive pulmonary disease (COPD) and so on; ③ Laboratory tests(before first PIRRT): routine test of blood, liver function, renal function(including serum creatinine, estimated Glomerular filtration rate), serum electrolytes, coagulation indicators and so on; ④ Medication : diuretics, antihypertensive agents, lipid-lowering drugs, etc.; ⑤Other factors: including peripherally inserted central catheter (PICC), the number of vital organ injuries, etc.; ⑥PIRRT related parameters: PIRRT times, PIRRT frequency; ⑦ Follow-up indices: serum creatinine (Scr) and urine volume.

Statistical analysis and data processing

Baseline analysis

Quantitative variables were described by mean ± standard deviation or median and quartile, while a t-test or non-parametric test was used according to the distribution of data. Qualitative variables were described by the number of cases and percentages of each category, while the chi-square and Fisher’s tests were used according to the applicable conditions of the test. P < 0.05 indicated a statistically significant difference. The data were analyzed using SPSS 25 (IBM, Armonk, NY, USA) and python 3.7.4.

Prognostic analysis

Firstly, research indicators were divided into two categories: categorical data and measurement data. For the measurement data, the partial ETA square was selected for correlation measurement; for categorical variables, the chi-square test was selected. Then, the indices related to survival outcome and 30-day and 90-day renal recovery were selected, respectively. The value of the partial ETA test > 0.16 indicated a strong correlation [11]. A P-value < 0.05 indicated a correlation between variables for the chi-square test.

Secondly, the correlation among selected indices was evaluated. For measurement data-measurement data, Pearson’s correlation coefficient was selected; the absolute value 0.8–1.0 indicated a very strong correlation, 0.6–0.8 indicated a strong correlation, and 0.4–0.6 was a medium correlation [12]. For the categorical data-categorical data, the chi-square test was selected. For the measurement data-categorical data, the partial ETA square was applied.

Finally, through correlation screening, the obtained uncorrelated indicators and prognostic outcomes were input into machine learning models, including Logistic Regression, Support Vector Machines (SVM), K-Nearest Neighbor (KNN), Naive Bayes, Perceptron, Stochastic Gradient Descent (SGD), Decision Tree, Perceptron and Random Forest for training, and the prognosis prediction models of patients with severe AKI who received PIRRT were obtained.

This study predicted the prognosis of patients according to the sequence of survival outcomes and 30- and 90-day renal recovery (Fig. 1).

Fig. 1
figure 1

Flowchart of the construction of the prognosis prediction model

Full size image

Results

Baseline information of the subjects

A total of 493 subjects, including 340 (68.97%) males and 153(31.03%) females, had an average age of 61.6 ± 16.6 years. Eventually, 256 (51.93%) patients died, and 237 (48.07%) patients survived during the follow-up. Among surviving patients, 150 (30.43%) patients recovered at 30 days and 163(33.06%) at 90 days, respectively. The basic information of the subjects were shown in Table 1.

Table 1 Baseline information of subjects
Full size table

Experimental process

The short-term prognosis prediction models for severe AKI patients receiving PIRRT were built according to the steps shown in Fig. 2. The algorithm in steps format was shown in Table 2.

Fig. 2
figure 2

Logistic regression flowchart

Full size image
Table 2 Algorithm: Framework of the survival prediction process
Full size table

Screening of correlation indicators

  1. (1)

    Screening of indicators related to survival outcomes.

A correlation between patient survival outcome and the above observation indices was established using a partial ETA square for measurement data and a chi-square test for categorical variables. The results were shown in Table 3. Indices not included in the table were not related to survival outcomes.

Table 3 Correlation between indices and survival outcome
Full size table

Pearson’s correlation coefficient/chi-square test was used to measure the pairwise correlation between measurement data-measurement data and categorical variables-categorical variables retrospectively, and five measurement indices and the four categorical variables were selected. Subsequently, the correlation between those five measurement indices and four categorical variables was analyzed again (Table 4). After screening, all 9 indices were used to predict the survival outcomes: Sodium, Total protein, LDH, Phosphorus, Thrombin time, liver cirrhosis, CKD stage, number of vital organ injuries, and AKI stage.

Table 4 Correlation between measurement indicators and categorical variables (partial ETA square)
Full size table
  1. (2)

    Screening of indicators related to 30-day renal recovery.

The same method was used to screen out the indicators related to 30-day renal recovery. The correlation between indices and 30-day renal recovery was shown in Additional file 1. After measuring the pairwise correlation between measurement data-measurement data and categorical variables-categorical variables, 5 measurement indices and 3 categorical variables were identified. Eventually, all these 8 indices, including Sodium, Total protein, LDH, Phosphorus, Thrombin time, Diabetes, PICC, and AKI stage were selected to predict the 30-day renal recovery (Table 5).

Table 5 Correlation between measurement and categorical indicators (partial ETA square)
Full size table
  1. (3)

    Screening of indicators related to 90-day renal survival outcome.

For 90-day renal function condition, the same analysis was utilized. The correlation between indices and 90-day renal recovery was shown in Additional file 1. After measuring the pairwise correlation between data, six measurement indices and two categorical variables were identified. All eight indices, including Hematocrit, Chlorine, Total protein, Uric acid, blood Phosphorus, Prothrombin time, CKD stage and diabetes, were selected (Table 6).

Table 6 Correlation between measurement and categorical indicators (partial ETA square)
Full size table

Establishment and verification of the prognosis prediction models

Based on the above-selected indicators, different machine learnings, including Logistic Regression, SVM, KNN, Naive Bayes, Perceptron, SGD, Decision Tree, Perceptron and Random Forest were used to build the prediction models.

In order to verify the model performance, we used the Accuracy (ACC), Precision, Recall, F1 and Area Under Curve (AUC) indicators. These indicators can be calculated by the confusion matrix. Meanwhile, True Positive (TP)、True Negative (TN)、 False Positive (FP)、False Negative (FN) were introduced to describe data. The accuracy rate represents the accuracy of the model’s predictions, and higher its values represent more accurate the model predictions. F1 indicator is obtained through the combination of precision and recall, which represents the balance between precision and recall.

$$Accuracy=\frac{{TP+TN}}{{TP+TN+FP+FN}}$$
$$Precision=\frac{{TP}}{{TP+FP}}$$
$$Recall=\frac{{TP}}{{TP+FN}}$$
$$F1=\frac{{2 \times Precision \times Recall}}{{Precision+Recall}}$$

Survival outcome prediction model

Nine indicators selected above were used as dependent variables of the survival outcome prediction model. In order to obtain a reliable and stable model, we used the 10-fold cross-validations to test the model performances, enabling the models to match the training dataset as closely as possible. The results obtained by the eight machine learning methods were shown in the Table 7. The P-R curves is shown in the Fig. 3.

Table 7 prognosis prediction model
Full size table
Fig. 3
figure 3

P-R curve of survival outcome

Full size image

In the prediction model for survival, Naive Bayes and Random Forest have better performance on the accuracy index, with a 1.21–13.00% increase over the rest of the methods. As for the F1 index, SGD has better performance, increasing by 1.04–22.00% compared to the rest of the methods. Considering the AUC index, Naive Bayes performs well, with a 2.11–14.89% increase over the rest of the methods. Overall, Naive Bayes has a good performance in the prediction model for survival.

30-day renal recovery prediction model

Eight indicators selected above were used as dependent variables of the 30-day renal recovery prediction model. We used the 10-fold cross-validations to test the model performances. The results obtained by the eight machine learning methods were shown in the Table 7. The P-R curves is shown in the Fig. 4.

Fig. 4
figure 4

P-R curve of 30-day renal recovery

Full size image

In the prediction model for 30-day renal recovery, Random Forest has better performance on the accuracy index, with a 11.30–24.60% increase over the rest of the methods. As for the F1 index, Naive Bayes has better performance, increasing by 8.45–150.21% compared to the rest of the methods, which was obtained after excluding SVM method. SMV method performs poorly on Precision, Recall and F1 indices, this may be due to difference in positive and negative sample sizes, especially when the negative sample size is significantly greater than the positive sample size. Considering the AUC index, Random Forest performs well, with a 5.22–33.00% increase over the rest of the methods. Overall, Random Forest has a good performance in the prediction model for 30-day renal recovery.

90-day renal recovery prediction model

Hematocrit, Chlorine, Total protein, Uric acid, blood Phosphorus, Prothrombin time, CKD stage, and Diabetes were used as dependent variables of the 90-day renal recovery prediction model. Because of the lesser data, we used the 10-fold cross-validations to test the model performances. The results obtained by the eight machine learning methods were shown in the Table 7. The P-R curves is shown in the Fig. 5.

Fig. 5
figure 5

P-R curve of 90-day renal recovery

Full size image

In the prediction model for 90-day renal recovery, Logistic Regression, SVM, KNN and Random Forest have better performance on the accuracy index, with a 4.16–32.30% increase over the rest of the methods. As for the F1 index, Logistic Regression and SVM have better performance, increasing by 0.44–15.87% compared to the rest of the methods. Considering the AUC index, KNN performs well, with a 0.65–28.84% increase over the rest of the methods. Overall, KNN has a good performance in the prediction model for 90-day renal recovery.

Discussion

This is the first attempt to use machine learning to establish prognosis prediction models for severe AKI patients who received PIRRT, and Naive Bayes has a good performance in the prediction model for survival, Random Forest has a good performance in 30-day renal recovery prediction model, while for 90-day renal recovery prediction model, it’s KNN.

Specifically, Sodium, Total protein, LDH, Phosphorus, Thrombin time, Liver cirrhosis, CKD stage, number of vital organ injuries, and AKI stage entered the prognostic prediction model of survival outcome. The indices used to predict 30-day renal recovery include Sodium, Total protein, LDH, blood Phosphorus, Thrombin time, Diabetes, PICC, AKI stage. The indices that could predict 90-day renal recovery include Hematocrit, Chlorine, Total protein, Uric acid, blood Phosphorus, Prothrombin time, CKD stage, and Diabetes.

As mentioned above, AKI severity is related to the prognosis, and the same conclusion is reached in this study, meanwhile, attention should be paid to serum electrolytes. Previous studies on serum Sodium were mainly focused on CKD patients since they are prone to abnormal serum Sodium levels due to their weakened ability to maintain water homeostasis, while a sharp decline in renal function can lead to acute changes in serum Sodium levels, which may result in higher mortality and often require immediate treatment to avoid severe neurological complications [13]. Therefore, AKI and abnormal serum Sodium have been under intensive focus recently. In a large retrospective study, a Cox proportional hazards model of 13,621 ICU patients with AKI showed a U-shaped association between Sodium levels and AKI survival [14]. Lee et al. and Edward et al. confirmed similar conclusions [15, 16]. Although Calcium and Phosphorus Metabolism is one of the prognostic factors of CKD, we observed that serum Phosphorus is related to the prognosis of the severe AKI population [17]. Serum phosphorus can be an early biomarker for AKI prediction during pediatric cardiac surgery [18]. Moreover, this study indicates that serum Chlorine is related to the prognosis of AKI patients received PIRRT, which is consistent with the conclusion of Sadan et al. [19].

In the present study, Diabetes is associated with the prognosis of the subjects. Previously, diabetes was associated with CKD, and diabetic nephropathy is the most common cause of end-stage renal disease worldwide. However, some studies have confirmed that diabetic patients are more likely to develop AKI than non-diabetic patients and that diabetes is a major risk factor for AKI [16, 20].

Nevertheless, the present study has several limitations: ① patients may have errors in judging urine volume after discharge. ② Although scientific algorithms have been used for modeling and model verification, this study still has shortcomings in the retrospective design.③ This is a single-center observational study requiring a multicenter external cohort with a large sample size to verify and improve the model. ④ The influence of the specific PIRRT scheme on prognosis was ignored. ⑤ Due to the uneven sample size of positive and negative samples, the accuracy and other indices of the prediction models were influenced.

Conclusions

In this study, for patients with severe AKI who received PIRRT, the short-term mortality rate was 51.93%, the 30-day renal recovery rate was 30.43%, and the 90-day renal recovery rate was 33.06%. Machine learning can not only screen out factors influencing the short-term prognosis, but also establish prediction models to optimize the risk assessment of these people. These predictors are easy to obtain in the early stage of AKI and attention should be paid to serum electrolytes. Naive Bayes has a good performance in the prediction model for survival, Random Forest has a good performance in 30-day renal recovery prediction model, while for 90-day renal recovery prediction model, it’s K-Nearest Neighbor.

Data Availability

The datasets used and/or analyzed during the current study are available from the corresponding author upon reasonable request.

Abbreviations

ACC:

Accuracy

AKI:

Acute kidney injury

AUC:

Area Under Curve

CHD:

Coronary heart disease

CKD:

Chronic kidney disease

COPD:

Chronic obstructive pulmonary disease

CRRT:

Continuous renal replacement therapy

GFR:

Glomerular filtration rate

IHD:

Intermittent hemodialysis

KNN:

K-Nearest Neighbor

LDH:

Lactate dehydrogenase

PICC:

Peripherally inserted central catheter

PIRRT:

Prolonged intermittent renal replacement therapy

RRT:

Renal replacement therapy

Scr:

Serum creatinine

SGD:

Stochastic Gradient Descent

SVM:

Support Vector Machines

References

  1. Ronco C, Bellomo R, Kellum JA. Acute kidney injury. Lancet. 2019;394:1949–64.

    Article  CAS  PubMed  Google Scholar 

  2. Susantitaphong P, Cruz DN, Cerda J, et al. World incidence of AKI: a meta-analysis. Clin J Am Soc Nephrol. 2013;8:1482–93.

    Article  PubMed  PubMed Central  Google Scholar 

  3. Hoste EA, Bagshaw SM, Bellomo R, et al. Epidemiology of acute kidney injury in critically ill patients: the multinational AKI-EPI study. Intensive Care Med. 2015;41:1411–23.

    Article  PubMed  Google Scholar 

  4. Kaddourah A, Basu RK, Bagshaw SM, et al. Epidemiology of Acute kidney Injury in critically ill children and young adults. N Engl J Med. 2017;376:11–20.

    Article  PubMed  Google Scholar 

  5. Wang AY, Bellomo R. Renal replacement therapy in the ICU: intermittent hemodialysis, sustained low-efficiency dialysis or continuous renal replacement therapy? Curr Opin Crit Care. 2018;24:437–42.

    Article  CAS  PubMed  Google Scholar 

  6. Neves JB, Rodrigues FB, Castelão M, et al. Extended daily dialysis versus intermittent hemodialysis for acute kidney injury: a systematic review. J Crit Care. 2016;33:271–3.

    Article  PubMed  Google Scholar 

  7. Cheng P, Waitman LR, Hu Y, et al. Predicting Inpatient Acute kidney Injury over different Time Horizons: how early and Accurate? AMIA Annu Symp Proc. 2017;2017:565–74.

    PubMed  Google Scholar 

  8. Tomašev N, Glorot X, Rae JW, et al. A clinically applicable approach to continuous prediction of future acute kidney injury. Nature. 2019;572:116–9.

    Article  PubMed  PubMed Central  Google Scholar 

  9. Chen X, Ma T. Sustained low-efficiency daily diafiltration for diabetic nephropathy patients with acute kidney injury. Med Princ Pract. 2014;23:119–24.

    Article  PubMed  PubMed Central  Google Scholar 

  10. Wei W, Rong S, Li X, et al. Short-term prognosis and influencing factors of patients with acute kidney injury treated with prolonged intermittent renal replacement therapy. Int J Clin Pract. 2021;75(5):e14020.

    Article  CAS  PubMed  Google Scholar 

  11. Richardson JTE. Eta squared and partial eta squared as measures of effect size in educational research[J]. Educational Res Rev. 2011;6(2):135–47.

    Article  Google Scholar 

  12. Benesty J, Chen J, Huang Y, et al. Pearson correlation coefficient[M]//Noise reduction in speech processing. Berlin, Heidelberg: Springer; 2009. pp. 1–4.

    Google Scholar 

  13. Joergensen D, Tazmini K, Jacobsen D. Acute Dysnatremias - a dangerous and overlooked clinical problem. Scand J Trauma Resusc Emerg Med. 2019;27:58.

    Article  PubMed  PubMed Central  Google Scholar 

  14. Gao XP, Zheng CF, Liao MQ, et al. Admission serum sodium and potassium levels predict survival among critically ill patients with acute kidney injury: a cohort study. BMC Nephrol. 2019;20:311.

    Article  PubMed  PubMed Central  Google Scholar 

  15. Lee SW, Baek SH, Ahn SY, et al. The Effects of Pre-Existing Hyponatremia and subsequent-developing acute kidney Injury on In-Hospital mortality: a retrospective cohort study. PLoS ONE. 2016;11:e0162990.

    Article  PubMed  PubMed Central  Google Scholar 

  16. Mezones-Holguin E, Niño-Garcia R, Herrera-Añazco P, et al. Possible association between dysnatremias and mortality during hospitalization in patients undergoing acute hemodialysis: analysis from a peruvian retrospective cohort. J Bras Nefrol. 2019;41:501–8.

    Article  PubMed  PubMed Central  Google Scholar 

  17. Moon H, Chin HJ, Na KY, et al. Hyperphosphatemia and risks of acute kidney injury, end-stage renal disease, and mortality in hospitalized patients. BMC Nephrol. 2019;20:362.

    Article  PubMed  PubMed Central  Google Scholar 

  18. Burra V, Nagaraja PS, Singh NG, et al. Early prediction of acute kidney injury using serum phosphorus as a biomarker in pediatric cardiac surgical patients. Ann Card Anaesth. 2018;21:455–9.

    Article  PubMed  PubMed Central  Google Scholar 

  19. Sadan O, Singbartl K, Kandiah PA, et al. Hyperchloremia is Associated with Acute kidney Injury in patients with subarachnoid hemorrhage. Crit Care Med. 2017;45:1382–8.

    Article  CAS  PubMed  Google Scholar 

  20. Saran R, Robinson B, Abbott KC, et al. Epidemiology of Kidney Disease in the United States. Am J Kidney Dis. 2017;69:A7–a8. US Renal Data System 2016 Annual Data Report:.

Download references

Acknowledgements

None.

Funding

This study was supported by the Clinical Research Innovation Plan of the Shanghai General Hospital, Shanghai Jiao Tong University School of Medicine (No. CTCCR-2018C09), the National Natural Science Foundation of China (No. 81970636) and the Clinical Characteristic Trial Research (2022) of the Shanghai General Hospital, Shanghai Jiao Tong University School of Medicine (No. CCTR-2022C09). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Author information

Author notes

  1. Wenqian Wei, Zhefei Cai and Lei Chen have contributed equally to this work.

Authors and Affiliations

  1. Department of Nephrology, Shanghai General Hospital, Shanghai Jiao Tong University School of Medicine, Shanghai, China

    Wenqian Wei, Lei Chen, Weijie Yuan & Shu Rong

  2. Hangzhou Dianzi University, Hangzhou, China

    Zhefei Cai & Yingle Fan

Authors
  1. Wenqian Wei
    View author publications

    You can also search for this author in PubMed Google Scholar

  2. Zhefei Cai
    View author publications

    You can also search for this author in PubMed Google Scholar

  3. Lei Chen
    View author publications

    You can also search for this author in PubMed Google Scholar

  4. Weijie Yuan
    View author publications

    You can also search for this author in PubMed Google Scholar

  5. Yingle Fan
    View author publications

    You can also search for this author in PubMed Google Scholar

  6. Shu Rong
    View author publications

    You can also search for this author in PubMed Google Scholar

Contributions

WW and SR carried out the studies, and participated in data collection. LC and WY helped with patient follow-up. YF and ZC designed and conducted the statistical analysis. WW, ZC and LC helped to draft the manuscript. All authors read and approved the final manuscript.

Corresponding authors

Correspondence to Yingle Fan or Shu Rong.

Ethics declarations

Ethics approval and consent to participate

The study was approved by the Ethics Committee of Shanghai General Hospital, Shanghai Jiao Tong University School of Medicine (2018KY156). The need for informed individual consent was waived by this committee. All methods were performed in accordance with the relevant guidelines and regulations.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.

Additional information

Publisher’s Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Electronic supplementary material

Below is the link to the electronic supplementary material.

Supplementary Material 1

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated in a credit line to the data.

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Wei, W., Cai, Z., Chen, L. et al. Short-term prognostic models for severe acute kidney injury patients receiving prolonged intermittent renal replacement therapy based on machine learning. BMC Med Inform Decis Mak 23, 133 (2023). https://doi.org/10.1186/s12911-023-02231-2

Download citation

  • Received:

  • Accepted:

  • Published:

  • DOI: https://doi.org/10.1186/s12911-023-02231-2

Keywords

BMC Medical Informatics and Decision Making

ISSN: 1472-6947

Contact us