Features selection in a predictive model for cardiac surgery- associated acute kidney injury

doi:10.21203/rs.3.rs-3103913/v1

Download PDF

Research Article

Features selection in a predictive model for cardiac surgery- associated acute kidney injury

https://doi.org/10.21203/rs.3.rs-3103913/v1

This work is licensed under a CC BY 4.0 License

Version 1

posted

You are reading this latest preprint version

Background

Cardiac surgery-associated acute kidney injury (CSA-AKI) is related to increased mobility and mortality. A large number of studies have explored the risk factors of AKI using traditional logistic regression (LR), which requires a generalized linear relationship between covariates and outcome. However, limited studies have explored the influence of different feature selection (FS) methods on the predictive performance of CSA-AKI. Therefore, we aimed to compare the impact of different FS methods for CSA-AKI.

Methods

CSA-AKI is defined according to the kidney disease: Improving Global Outcomes (KDIGO) criteria. Both traditional LR and machine learning methods were used to select the potential risk factors for CSA-AKI. The area under the receiver operating characteristic curve (AUC) was used to evaluate the performance of the models. In addition, the importance matrix plot by random forest was used to rank the features' importance.

Results

A total of 1977 patients undergoing cardiac surgery at Fuwai Hospital from December 2018 to April 2021 were enrolled. The incidence of CSA-AKI during the first postoperative week was 27.8%. We concluded that different enrolled numbers of features impact the final selected feature number. The more you input, the more likely its output with all FS methods. In terms of performance, all selected features by various FS methods demonstrated excellent AUCs. Meanwhile, the embedded method demonstrated the highest accuracy compared with the LR method, while the filter method showed the lowest accuracy. Furthermore, NT-proBNP was found to be strongly associated with AKI. Our results confirmed some features that previous studies have reported and found some novel clinical parameters.

Conclusions

In our study, FS was as suitable as LR for predicting CSA-AKI. For FS, the embedded method demonstrated better efficacy than the other methods. Furthermore, NT-proBNP was found to be strongly associated with AKI.

Feature selection

Machine learning

Logistic regression

Cardiac surgical procedure

Acute kidney injury

The incidence of cardiac surgery-associated acute kidney injury (CSA-AKI) varies between 7% and 40%, depending on the definition, patient characteristics, and the type of cardiac surgery^[1]. This common and serious complication poses a great threat to patient outcomes [1–7]as well as the health care system ^[7], and early prediction and detection of AKI can bring major benefits. Currently, the Kidney Disease Improving Global Outcome (KDIGO) criteria are recommended for the diagnosis of AKI. However, these criteria are not applicable for timely prediction[5, 8]. Several risk scores have been tested for helpfulness in risk stratification^{[3, 9–12]}. However, most of these potential risk factors in these models were analyzed with the traditional logistic regression method, which could only handle the linear relationship between variable and outcome[2, 4]

With the booming data in electronic medical records (EMR), it is difficult to build precise models with high-dimensional data. The irrelevant and redundant features negatively affect the accuracy and add difficulties when searching for useful knowledge[13]. Moreover, the relationship between intraoperative variables and AKI has not been as widely investigated as preoperative variables [4]. Machine learning, equipped with the capacity to remove redundant attributes, can improve task learning efficiency and optimize the performance of predictive models [14]. Although an increasing number of studies have explored the predictive model of CSA-AKI by machine learning [15] there is no consensus on the appropriate feature selection (FS) method for a given domain and problem[5]. As the data from the electronic database are high dimensional, different enrolled features may influence the performance of the models.

This study aimed to compare the different FS methods based on machine learning with logistic regression (LR) from three aspects: predictive performance, outcome accuracy, and feature importance. Moreover, whether the number of enrolled features impacts the performance was also explored.

Study population

This was a retrospective observational study, and the study protocol was approved by the Ethics Committee of Fuwai Hospital in Beijing, China (2022-ZX063) on 3rd November 2022. We performed this retrospective study in compliance with the Declarations of Helsinki and the guidelines for Good Clinical Practice. Written informed consent was exempted by the Ethics Committee of Fuwai Hospital due to the retrospective nature of this study. Patients who underwent cardiac surgery with cardiopulmonary bypass (CPB) at our institution from December 2018 to April 2021 were enrolled.

Data collection

Preoperative variables, including demographic characteristics, laboratory values, and medical and medication histories, were extracted. In addition, surgery time, CPB time, aortic clamp time, and surgery type were also extracted.

Study design

Two groups enrolled in different features were used to explore the impact of various FS methods on CSA-AKI. For group I, we included a maximum number of features (i.e., all features extracted, a more inclusive group). As for group II, we retained the reported features related to CSA-AKI[3, 6, 16–19]. LR and ML engineering were performed simultaneously in the two groups to determine suitable variables. The whole process showed in Fig. 1.

AKI definition

The primary outcome was postoperative AKI, defined according to Kidney Disease Improving Global Outcomes (KDIGO) criteria, which was determined as the maximal change in serum creatinine level during the first seven postoperative days [20]. The baseline serum creatinine (SCr) level was defined as the most recent measurement before surgery, and an increase in serum creatinine by 1.5–1.9 times baseline or by ≥ 26.5 µmol/l (0.3 mg/dl) from baseline was identified as KDIGO-AKI. We did not choose decreasing urine output as our objective indicator because it is not specific and can be affected by various clinical variables; meanwhile, the definition of oliguria is not constant with changeable cutoff values for different situations [21].

Statistical analysis

Continuous variables of patient characteristics were expressed as the mean ± standard deviation or median with an interquartile range. The categorical variables of patient characteristics were reported as the frequency number and percentage. Student’s t-test or the Mann–Whitney U test was used for continuous variables, and the χ2 test or Fischer’s exact test was applied for categorical variables (p < 0.05 indicates statistical significance).

Feature selection and model evaluation

LR was the traditional method for detecting the association between features and AKI (p < 0.1 indicates statistical significance). In contrast, ML was used as a novel method to select relevant features. Generally, the feature selection based on ML can be divided into three main categories: filter, wrapper, and embedded methods[13](Fig. 2).In this study, we aimed to compare maximal information coefficient (MIC), recursive feature elimination (RFE), and random forest (RF) with LR.

1)Filter: Based on the univariate relationship, features are selected with the endpoint (i.e., AKI). We set p < 0.05 as a threshold to decide whether a feature should be included [22]. (Fig. 2a)

2)Wrapper

The wrapper method trains the estimator on an initial set of features and obtains the importance of each feature through the coef_ attribute or the feature_importances attribute. (Fig. 2b)

3)Embedded

This includes two main steps. First, certain machine learning algorithms and models are trained to obtain the weight coefficients of each feature, and then the features are selected according to the weight coefficients from the largest to the smallest. (Fig. 2c)

The importance matrix plot for the RF method was used to rank the features selected. Furthermore, we used AUC to evaluate the model's performance, which was compared with the LR by the Delong test. All analyses were developed in open-source software libraries (Python version 3.9).

Characteristics of the analyzed cohort

A total of 1977 patients were enrolled in the analysis, in which 550 (27.8%) patients were identified with postoperative AKI. (Figure 1) Those with AKI were more likely to be older (age: 55 vs 58, p<0.001), higher BMI (23.59 vs 24.06, p<0.001), and female (36.3% vs 41.5%, p=0.03) with lower blood pressure(mmHg) (SBP:129.78 vs 123.09, p<0.001; DBP:74.38 vs 71.28, p<0.001). Patients with the poor condition were more vulnerable to suffering from AKI. In terms of intraoperative parameters, longer surgical time (min) (244.52 vs 262.59, p<0.001), longer CPB and aortic clamp time (min) (108 vs 124, p<0.001; 75 vs 89.5, p<0.001) were detected to have negative effects on patient outcome. In addition, the surgical procedure (including the CPB procedure) was also found to impact on the incidence of AKI. Table 1 presents a detailed summary of this information.

Outcomes of different selection methods

In Group I, 66 available features were included for analysis. Finally, 46 features were identified by LR methods; 47 were selected by the MIC method. Regarding the wrapper (RFE), the features that negatively influenced the cross-validation score were removed. Hence, 36 features were selected. As for the embedded (RF), when the threshold of the model was set as 0.01, 18 features were included. (Table 2)

In Group II, 41 features reported to be related to CSA-AKI were enrolled. Eventually, there were 35 features selected by the LR method, 30 features selected by MIC, 31 by the wrapper (RFE), and nine features selected by the embedded (RF) method. (Table 2)

Comparison of different FS methods

In terms of AUCs, ML techniques and traditional statistical approaches demonstrated excellent predictive performance in both groups. In Group I, the features selected by LR, and MIC methods all achieved an AUC of 0.97. RFE and RF performed the AUC with 0.98 and exhibited a significant difference compared with LR (p <0.01). In Group II, the LR method performed an AUC of 0.98. The MIC and RFE method achieved an AUC of 0.97. Furthermore, the RF method performed an AUC of 0.99 and showed a significant difference from the LR method (p <0.01). (Figure 3)

Regarding the selection accuracy of ML methods compared with the traditional LR method, we concluded that for Group I, the accuracy of the MIC method was 72.34% compared with LR. Meanwhile, the accuracy was 91.66% for the RFE and 100% for the RF method. Furthermore, in Group II, the accuracy of the MIC method was 96.67%, 95.54% for the RFE, and 100% for the embedded method.

Identification of Significant Features

An importance matrix plot was conducted by random forest model to identify potential risk factors for CSA-AKI. The plot demonstrated that NT-proBNP was the most influential factor. Furthermore, in both groups, the top 4 most important features for all FS methods were almost the same, including prothrombin time (PT), blood urea nitrogen (BUN), left ventricular ejection fraction (LVEF), and NT-proBNP. (Table 3 & Figure 4)

In this retrospective cohort study, we compared LR and ML techniques to select potential risk factors for AKI. Firstly, we concluded that different enrolled numbers of features impact the final selected feature number. The more inputs, the more outputs with all FS methods. Secondly, as for the performance, all selected features by various FS methods demonstrated excellent AUCs. Meanwhile, the embedded method demonstrated the highest accuracy compared with the LR method, while the filter method showed the lowest accuracy. Thirdly, regarding the importance ranking of features, our results confirmed some features that previous studies have reported and found some novel clinical parameters.

Tremendous research has shown that ML outperformed LR for the prediction of AKI [2, 4, 23, 24]. However, a recent meta-analysis reported that ML algorithms were comparable to regression models in developed models [25]. Particularly ML techniques have become prevalent in medical fields in recent years, where it can reduce dimensionality and help us understand the causes of disease.[26]The performance of different FS methods in predicting CSA-AKI remains unknown. Researchers used traditional logistic regression or Cox regression to explore the potential risk factors for AKI[1, 6, 16, 27–29]. Previous studies mainly used these features to build models and predict AKI[1, 16, 27]. Recently, some studies found other features might influence the progression of AKI [17–19, 28, 30, 31], yet there is a lack of sufficient multicenter evidence to support their findings. Hence, we included as many features as possible in Group I, and 25 more were not commonly recognized before compared with Group II. We used LR and ML approaches to select the relevant features in the two groups. We identified that the more you input, the more likely its output with all FS methods. In addition, the number of features selected by ML methods was less than LR methods, which might promote establishing an easier and more effective predictive model. Meanwhile, the number of selected features by the embedded method is smaller than the filter and wrapper method, which may be attributed to the filter method mainly used to screen data and reduce the size. In contrast, the wrapper and the embedded method are based on the filter method and are better at handling and processing data[30].

Our study showed the AUCs between FS and LR methods were significant statistically. Still, all performances demonstrated excellent, which implies that FS and LR methods can be performed comparably well in selecting predictors of CSA-AKI. Furthermore, no matter dealing with a lower or higher dimensional dataset, the embedded method performed better than other methods, which might be attributed to the feature subset search process of the embedded method incorporated into the classifier training process. Moreover, the LR method performed better in a lower dimension than in a higher dimension dataset. [30] This might be due to the LR method being more proficient in dealing with liner relationship problems. Notably, the AUCs of all FS methods in our study were beyond 0.9. Perhaps this is because we included serum creatinine (SCr) in our analysis, which is an important indicator of the definition of KDIGO-AKI. However, Koyner Carey et al. found that the algorithm for predicting severe AKI did not change significantly after excluding the SCr variable. Future studies should investigate the complex connection between baseline SCr and postoperative AKI. Additionally, we included many intraoperative variables that might improve the performance of our models[2].

In terms of the selection accuracy of the ML methods compared with the LR method, our analysis indicated that the embedded method achieved the highest accuracy despite the number of the features inputted. In contrast, the filter method achieved less accuracy and was influenced by the number of features inputted. It may be due to the wrapper, and embedded methods have a built-in algorithm, in which the feature selection process and algorithm training are performed simultaneously. Therefore, their results are more accurate and reliable[30]. Nevertheless, the number of selected features in the embedded method was related to the threshold value, yet the problem defining the threshold used to discriminate a truly informative feature from the other noninformative ones[31].

According to the importance matrix plot, the most influential factor was NT-proBNP in all FS methods. It is still unclear about the relationship between NT-proBNP and AKI. One possible mechanism is that NT-proBNP acts as a mediator between cardiovascular and renal dysfunction in that its production is stimulated by elevated venous pressure, which contributes to renal impairment[32]. Another potential hypothesis is that elevated NT-proBNP will cause systematic vasodilation and antagonize the renin-angiotensin-aldosterone system, which might directly prompt renal hypotension[33]. Eventually, it’s a vicious circle between cause and effect. In addition, the laboratory values such as PT, BUN, total and direct bilirubin, and features to evaluate left ventricular function were found to have a higher potential impact on the progression of CSA-AKI. Consistent with previous studies, our study also demonstrated decreased LVEF as a risk factor of AKI[34, 35]. one classical theory explains it might relate to renal hypotension, associated ischemia, and insufficient oxygen supply[36]. As for prolonged PT, an increasing number of evidence has demonstrated that AKI might correlate with an interaction between inflammation cascade activation and deranged coagulation pathways, leading to endothelial cell dysfunction, microvascular damage, and extensive microthrombi[37]. Urea nitrogen is a protein terminal metabolite and is excreted through the kidneys, though its level can be influenced by various factors, it can still be used as a biomarker to some extent. Further prospective studies are warranted to explore the relationship and underlying mechanism. In addition, preoperative hemodynamic variables such as heart rate and systolic blood pressure were ranked in the top 10 in our cohort, suggesting better managing these features before surgery might benefit the patient. Abundant evidence denotes that intraoperative variable, including surgery and the CPB procedure, are closely associated with postoperative AKI [3, 4]. In one single-center cohort, patients at low risk of AKI were reclassified as high risk after including intraoperative variables[38]. We also confirmed that CSA-AKI was associated with the aortic clamp time, CPB, and surgery time. In addition, we found that the preoperative use of β-blockers might help mitigate AKI. However, it is still controversial whether pharmacological interventions are beneficial for high-risk patients [39, 40]. Additionally, the top 10 features included were almost the same between two groups in different methods. However, the importance ranking of these features were different which may attribute to the interaction between features and thus, when more features are enrolled, their importance percentages will change. Furthermore, some features such as ALT, and the volume of intraoperative blood salvage, which were statistically important, were only detected by ML methods. Additionally, Lee, Hofer et al. found that a hybrid of FS and LR could perform comparably with deep neural network. This implies that FS and LR could be combined in future research, as FS could reduce the number of parameters, decrease the learning time, and avoid the problems of dimensionality; meanwhile LR could output explainable variables with low computational cost[41],which need more investigation when faced with booming data and high dimensional statistics..

Our study also has several limitations. First, this is a retrospective analysis with single-center data and a relatively small number of cases. The performance of machine learning algorithms might be different for a larger dataset with a different distribution of patient characteristics in different institutions. Second, the most important variables are not clinically modifiable, and whether our results could benefit high-risk patients is unknown. Nevertheless, further prospective trials are imperative to evaluate whether the adjustment of modifiable predictors could yield beneficial results[2]. But we have confirmed the importance of the intraoperative variables. Third, we did not determine the time of the latest measurement before surgery, which might impact our results. Fourth, we only took the drug usage into our analysis, the effects of drug changes before and after surgery are still unknown. Fifth, another potential limitation is we did not use urine volume as a criterion to diagnose AKI, and different diagnostic criteria might lead to different results. Sixth, we did not include some biomarkers, such as cystatin C, tissue inhibitor of metalloproteinases 2 (TIMP-2), and insulin-like growth factor-binding protein 7 (IGFBP7)[42], which have been reported to have high specificity and sensitivity, irrespective of potentially interfering conditions in our analysis. Further investigation could explore whether these biomarkers could help better detect postoperative AKI.

In conclusion, ML was as suitable as LR for selecting potential risk factors for CSA-AKI. For ML, the embedded method demonstrated better efficacy than the other methods. Furthermore, NT-proBNP was found to be strongly associated with AKI.

BMI

body mass index

NYHA

Classification of NYHA heart function

smoke1m

smoke within a latest month

group

high or low dosage tranexamic acid

ACEI

angiotensin converting enzyme inhibitors

Temp

body temperature before surgery

heart rate

SBP

systolic blood pressure

DBP

diastolic blood pressure

SBP-DBP

LVEF

left ventricular ejection fraction

LVEDD

Left ventricular end diastolic dimension

Lab1

white blood cell count

Lab2

neutrophil count

Lab3

hemoglobin count

Lab4

platelet count

Lab5

ALT, alanine aminotransferase

Lab6

AST, alkaline phosphatase

Lab7

GGT, gamma-glutamyl transferase

Lab8

ALP, alkaline phosphatase

Lab9

direct bilirubin

Lab10

total bilirubin

Lab11

serum creatinine

Lab12

BUN, blood urea nitrogen

Lab13

TP, total protein

Lab14

ALB, albumin

Lab15

PT, prothrombin time

Lab16

D-dimer

Lab17

NT-proBNP, N-terminal brain sodium peptide

Lab18

Hs-CRP, hypersensitive C-reactive protein

MED1

allergy

MED2

diabetes mellitus

MED3

hypertension

MED4

hyperlipidemia

MED5

chronic kidney disease

MED6

COPD, chronic obstructive pulmonary disease

MED7

peripheral vascular disease

MED8

cerebrovascular accident

MED9

infective endocarditis

MED10

fully recovered ischemic central nervous system injury

MED11

lacunar cerebral infarction

MED12

non-invasive tests suggesting carotid artery stenosis > 79%

MED13

previous cardiac valvular disease

MED14

previous coronary heart disease

MED15

previous congenital heart disease

MED16

previous aortic disease

MED17

previous carotid surgery

MED18

previous peripheral vascular disease

Intra1

volume of intraoperative blood salvage

Intra2

lowest temperature during CPB

Intra3

hemoglobin at end of CPB

Intra4

CPB again

Intra5

surgery duration

Intra6

CPB duration

Intra7

aortic clamping duration

Sur1

emergency surgery

Sur2

CABG, coronary Artery Bypass Graft

Sur3

valve Surgery

Sur4

congenital heart disease surgery

Sur5

aortic surgery

CPB

cardiopulmonary bypass

IABP

Intra-aortic balloon pump

ECMO

Extracorporeal Membrane Oxygenation

Acknowledgements

None

Authors’ contributions

QL and JJS contributed to the manuscript preparation. QL and YYC participated in the analysis and interpretation of the data, CHZ and HL contributed to reviewing and editing, and visualization. JS and HL contributed to the supervision and administration of the project. HL takes the responsibility for communication with the journal during the manuscript submission, peer-review, and publication process.

Funding

None declared.

Availability of data and materials

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

Ethics approval and consent to participate

This retrospective observational study was approved by the Ethics Committee of Fuwai Hospital in Beijing, China (2022-ZX063), and written informed consent was waived.

Consent for publication

Not applicable

Competing interests

The authors declare that they have no competing interests. All authors declare that the research was conducted without any commercial or financial relationships construed as a potential conflict of interest.

Thongprayoon C et al. Diagnostics, Risk Factors, Treatment and Outcomes of Acute Kidney Injury in a New Paradigm. J Clin Med, 2020. 9(4).
Lee HC et al. Derivation and Validation of Machine Learning Approaches to Predict Acute Kidney Injury after Cardiac Surgery. J Clin Med, 2018. 7(10).
Thongprayoon C et al. Predicting Acute Kidney Injury after Cardiac Surgery by Machine Learning Approaches. J Clin Med, 2020. 9(6).
Tseng PY, et al. Prediction of the development of acute kidney injury following cardiac surgery by machine learning. Crit Care. 2020;24(1):478.
Wu L, et al. Feature Ranking in Predictive Models for Hospital-Acquired Acute Kidney Injury. Sci Rep. 2018;8(1):17298.
Amini S, et al. Risk Factors and Outcome of Acute Kidney Injury after Isolated CABG Surgery: a Prospective Cohort Study. Braz J Cardiovasc Surg. 2019;34(1):70–5.
Zaouter C, et al. A combined approach for the early recognition of acute kidney injury after adult cardiac surgery. Anaesth Crit Care Pain Med. 2018;37(4):335–41.
Ostermann M, Joannidis M. Acute kidney injury 2016: diagnosis and diagnostic workup. Crit Care. 2016;20(1):1–13.
Chertow GM, et al. Preoperative renal risk stratification. Circulation. 1997;95(4):878–84.
Thakar CV, et al. A clinical score to predict acute renal failure after cardiac surgery. J Am Soc Nephrol. 2005;16(1):162–8.
Mehta RH, et al. Bedside tool for predicting the risk of postoperative dialysis in patients undergoing cardiac surgery. Circulation. 2006;114(21):2208–16.
Wijeysundera DN, et al. Derivation and validation of a simplified predictive index for renal replacement therapy after cardiac surgery. JAMA. 2007;297(16):1801–9.
Saeys Y, Inza I, Larranaga P. Rev feature selection techniques Bioinf Bioinf. 2007;23(19):2507–17.
Guyon I, Elisseeff A. An introduction to variable and feature selection. J Mach Learn Res. 2003;3(Mar):1157–82.
Lee CK, et al. Development and Validation of a Deep Neural Network Model for Prediction of Postoperative In-hospital Mortality. Anesthesiology. 2018;129(4):649–62.
Kristovic D, et al. Cardiac surgery-associated acute kidney injury: risk factors analysis and comparison of prediction models. Interact Cardiovasc Thorac Surg. 2015;21(3):366–73.
Neugarten J, et al. Sex and the Risk of AKI Following Cardio-thoracic Surgery: A Meta-Analysis. Clin J Am Soc Nephrol. 2016;11(12):2113–22.
Yi Q, et al. Risk Factors for Acute Kidney Injury after Cardiovascular Surgery: Evidence from 2,157 Cases and 49,777 Controls - A Meta-Analysis. Cardiorenal Med. 2016;6(3):237–50.
Wang J, et al. Independent Risk Factors Contributing to Acute Kidney Injury According to Updated Valve Academic Research Consortium-2 Criteria After Transcatheter Aortic Valve Implantation: A Meta-analysis and Meta-regression of 13 Studies. J Cardiothorac Vasc Anesth. 2017;31(3):816–26.
Thomas ME, et al. The definition of acute kidney injury and its use in practice. Kidney Int. 2015;87(1):62–73.
Hori D, et al. Defining oliguria during cardiopulmonary bypass and its relationship with cardiac surgery-associated acute kidney injury. Br J Anaesth. 2016;117(6):733–40.
Li X, et al. Diagnostic Value of Procalcitonin on Early Postoperative Infection After Pediatric Cardiac Surgery. Pediatr Crit Care Med. 2017;18(5):420–8.
Koyner JL, et al. The Development of a Machine Learning Inpatient Acute Kidney Injury Prediction Model. Crit Care Med. 2018;46(7):1070–7.
Penny-Dimri JC, et al. Machine Learning Algorithms for Predicting and Risk Profiling of Cardiac Surgery-Associated Acute Kidney Injury. Semin Thorac Cardiovasc Surg. 2021;33(3):735–45.
Christodoulou E, et al. A systematic review shows no performance benefit of machine learning over logistic regression for clinical prediction models. J Clin Epidemiol. 2019;110:12–22.
Remeseiro B, Bolon-Canedo V. A review of feature selection methods in medical applications. Comput Biol Med. 2019;112:103375.
Hu P, et al. Derivation and validation of a model to predict acute kidney injury following cardiac surgery in patients with normal renal function. Ren Fail. 2021;43(1):1205–13.
Kim WH, et al. Can we really predict postoperative acute kidney injury after aortic surgery? Diagnostic accuracy of risk scores using gray zone approach. Thorac Cardiovasc Surg. 2016;64(04):281–9.
Brown JR et al. Multivariable prediction of renal insufficiency developing after cardiac surgery. Circulation, 2007. 116(11_supplement): p. I-139-I-143.
Hameed SS, et al. Filter-wrapper combination and embedded feature selection for gene expression data. Int J Advance Soft Compu Appl. 2018;10(1):90–105.
Paja W, Pancerz K, Grochowalski P. Generational Feature Elimination and Some Other Ranking Feature Selection Methods, in Advances in Feature Selection for Data and Pattern Recognition. 2018. p. 97–112.
Wang C, et al. Prediction of acute kidney injury after cardiac surgery from preoperative N-terminal pro-B-type natriuretic peptide. Br J Anaesth. 2021;127(6):862–70.
Chen HH, et al. Low-dose dopamine or low-dose nesiritide in acute heart failure with renal dysfunction: the ROSE acute heart failure randomized trial. JAMA. 2013;310(23):2533–43.
Li Y, et al. A novel machine learning algorithm, Bayesian networks model, to predict the high-risk patients with cardiac surgery-associated acute kidney injury. Clin Cardiol. 2020;43(7):752–61.
Chen Z, et al. Analysis of the Impact of Medical Features and Risk Prediction of Acute Kidney Injury for Critical Patients Using Temporal Electronic Health Record Data With Attention-Based Neural Network. Front Med (Lausanne). 2021;8:658665.
Xin Q, et al. A Predictive Model Based on Inflammatory and Coagulation Indicators for Sepsis-Induced Acute Kidney Injury. J Inflamm Res. 2022;15:4561–71.
Fani F, et al. Recent advances in the pathogenetic mechanisms of sepsis-associated acute kidney injury. J Nephrol. 2018;31(3):351–9.
Adhikari L, et al. Improved predictive models for acute kidney injury with IDEA: Intraoperative Data Embedded Analytics. PLoS ONE. 2019;14(4):e0214904.
Ostermann M et al. Cardiac Surgery Associated AKI Prevention Strategies and Medical Treatment for CSA-AKI. J Clin Med, 2021. 10(22).
Meersch M, Zarbock A. Prevention of cardiac surgery-associated acute kidney injury. Curr Opin Anaesthesiol. 2017;30(1):76–83.
Aatila M et al. Keratoconus Severity Classification Using Features Selection and Machine Learning Algorithms. Comput Math Methods Med, 2021. 2021: p. 9979560.
Massoth C, Zarbock A, Meersch M. Acute Kidney Injury in Cardiac Surgery. Crit Care Clin. 2021;37(2):267–78.

Table 1: Patient characteristics and postoperative outcomes

Variables	All patients (n=1977)	Post-operative AKI (n=550)	No post-operative AKI (n=1427)	*P value*
Demographic data
Age, years, median, (IQR)	3656(47,64)	58(51,65)	55(44,63)	<0.001
Gender, female, n (%)	746(37.7)	228(41.5)	518(36.3)	0.03
Body-mass index (kg/m2) mean, (SD)	24.31(3.54)	23.59(3.47)	24.60(3.54)	<0.001
Race (n, %)
(1)	1859(94)	514(93.5)	1345(94.3)
(2)	26(1.3)	11(2.0)	15(1.1)
(3)	35(1.8)	13(2.4)	22(1.5)
(4)	33(1.7)	3(0.5)	30(2.1)
(5)	24(1.2)	9(1.6)	15(1.1)
Preoperative variables
Heart rate, mean, (SD)	78.69(14.16)	82.06(17.01)	77.39(12.66)	<0.001
SBP, mean, (SD)	127.92(18.52)	123.09(18.61)	129.78(18.15)	<0.001
DBP, mean, (SD)	73.52(12.47)	71.28(13.44)	74.38(11.97)	<0.001
Body temperature mean, (SD)	36.38(0.26)	36.37(0.27)	36.39(0.27)	0.11
Smoke, (n, %)	655(33.1)	191(34.7)	464(32.5)	0.34
Smoke within a latest month, (n, %)	270(13.7)	64(11.6)	206(14.4)	0.10
Allergy, (n, %)	192(9.7)	46(8.4)	146(10.2)	0.21
NYHA class (n, %)
Class I	236(11.9)	25(4.5)	211(14.8)
Class II	1266(64)	317(57.6)	949(66.5)
Class III	441(22.3)	191(34.7)	250(17.5)
Class IV	34(1.7)	17(3.1)	17(1.2)
Diabetes mellitus (n, %)	219(11.1)	55(10)	164(11.5)	0.34
Hypertension (n, %)	669(33.8)	172(31.3)	497(34.8)	0.13
Hyperlipidemia (n, %)	698(35.3)	176(32)	522(36.6)	0.13
Chronic kidney disease (n, %)	3(0.2)	3(0.5)	0(0)	0.03
COPD (n, %)	3(0.2)	1(0.2)	2(0.1)	1
Peripheral vascular disease (n, %)	12(0.6)	4(0.7)	8(0.6)	0.75
Cerebrovascular accident (n, %)	31(1.6)	14(2.5)	17(1.2)	0.03
Infective endocarditis (n, %)	16(0.8)	6(1.1)	10(0.7)
Fully recovered ischemic central nervous system injury (n, %)	42(2.1)	6(1.1)	10(0.7)	0.38
Lacunar cerebral infarction (n, %)	16(0.8)	19(3.5)	23(1.6)	0.01
Non-invasive tests suggesting carotid artery stenosis >79%, (n, %)	14(0.7)	6(1.1)	8(0.6)	0.21
Previous cardiac valvular disease, (n, %)	1466(74.2)	489(88.9)	977(68.5)	<0.001
Previous coronary heart disease, (n, %)	472(23.9)	106(19.3)	366(25.6)	0.003
Previous congenital heart disease, (n, %)	351(17.8)	48(8.7)	303(21.2)	<0.001
Previous aortic disease, (n, %)	236(11.9)	47(8.5)	189(13.2)	0.004
Previous peripheral vascular disease (n, %)	340(17.2)	159(28.9)	181(12.7)	<0.001
Previous carotid surgery, (n, %)	123(6.2)	47(8.5)	76(5.3)	0.008
LVEDD, mean, (SD)	52.01(10.21)	54.31(11.18)	51.12(9.67)	<0.001
LVEF (%), mean, (SD)	61.33(6.24)	58.96(7.71)	62.24(5.29)	<0.001
WBC count, mean, (SD)	6.24(1.61)	6.23(1.74)	6.25(1.56)	0.85
Neutrophil count, mean, (SD)	69.35(8.79)	71.10(8.85)	68.67(8.68)	<0.001
Hemoglobin count mean, (SD)	138.05(17.61)	135.86(19.42)	138.90(16.80)	0.001
Platelet count, mean, (SD)	205.58(57.32)	190.50(57.24)	211.40(56.30)	<0.001
ALT, median, (IQR)	19(13,29)	20(13,31)	18(12,28)	0.03
AST, median, (IQR)	25(21,31)	27(22,34)	25(21,30)	<0.001
ALP, mean, (SD)	67.97(21.71)	69.94(26.99)	67.21(19.25)	0.03
GGT, median, (IQR)	25(17,41)	46.74(61.81)	24(16,37)	<0.001
Direct bilirubin, median, (IQR)	3.36(2.44,5.12)	4.15(2.83,6.88)	3.12(2.36,4.55)	<0.001
Total bilirubin, median, (IQR)	12.12(9.10,16.58)	13.95(9.97,20.02)	11.53(8.83,15.51)	<0.001
Serum creatinine, mean, (SD)	84.55(18.07)	90.18(21.40)	82.39(16.10)	<0.001
BUN, median, (IQR)	5.98(4.88,7.41)	6.89(5.68,8.47)	5.71(4.68,6.94)	<0.001
TP, mean, (SD)	68.30(5.62)	67.66(6.00)	68.54(5.44)	0.003
ALB, mean, (SD)	39.99(3.35)	39.27(3.60)	40.27(3.21)	<0.001
PT, mean, (SD)	13.47(1.70)	14.15(2.44)	13.21(1.21)	<0.001
D-dimer, median, (IQR)	0.23(0.17,0.35)	0.27(0.19,0.47)	0.22(0.17,0.32)	<0.001
NT-proBNP (LN), median, (IQR)	252.40(79.25,875.00)	7.31(6.95,7.81)	4.96(4.00,5.75)	<0.001
Hs-CRP, median, (IQR)	0.83(0.35,2.17)	1.11(0.49,3.15)	0.73(0.32,1.83)	<0.001
Statins (n, %)	109(5.5)	25(4.5)	84(5.9)	0.24
β-blocker (n, %)	735(37.2)	267(48.5)	469(32.8)	<0.001
ACEI (n, %)	145(7.3)	39(7.1)	106(7.4)	0.79
Intraoperative variable
High dosage Tranexamic acid, n (%)	980(49.6)	284(51.6)	696(48.8)	0.25
Emergency surgery, n (%)	18(0.9)	8(1.5)	10(0.7)	0.11
Volume of intraoperative blood salvage, median, (IQR)	200(200,210)	200(200,220)	200(200,210)	0.04
Lowest temperature during CPB, median, (IQR)	32.0(31.0,31.9)	32.0(31.0,32.7)	32.0(31.1,33.0)	0.002
Hemoglobin at end of CPB, mean, (SD)	103.8(14.91)	101.58(14.79)	103.66(14.93)	0.005
CPB again, (n, %)	47(2.4)	13(2.4)	34(2.4)	0.98
IABP, (n, %)	12(0.6)	5(0.9)	7(0.5)	0.28
ECMO, (n, %)	3(0.2)	1(0.2)	2(0.1)	1
Surgery duration, mean, (SD)	249.55(79.70)	262.59(86.52)	244.52(76.35)	<0.001
CPB duration, median, (IQR)	113(85,148)	124(95,162)	108(81,142)	<0.001
Aortic clamping duration， median, (IQR)	80(57,109)	89.50(66.75,120.00)	75(54,105)	<0.001
CABG (n, %)	531(26.9)	128(23.3)	403(28.2)	0.02
Valve surgery (n, %)	1292(65.4)	444(80.7)	848(59.4)	<0.001
Congenital heart disease surgery (n, %)	290(14.7)	42(7.6)	248(17.4)	<0.001
Aortic surgery (n, %)	166(8.4)	34(6.2)	132(9.3)	0.02

Table 2: Features selected by ML and LR methods.

Group I

Group II

All features

group, gender, age, race, BMI, NYHA, smoke, smoke1m,

β-blocker, ACEI, statins,

Temp, HR, SBP, DBP, LVEF,

LVEDD, Lab1-18, MED1-18, Intra17, IABP, ECMO, Sur1-5

gender, age, race, BMI,

NYHA, smoke, sd, LVEF, L VEDD,

Lab1-18, MED2-7, MED13-14,

MED16, MED18

Intra5- 7, Sur1-3, Sur5

Filter

gender, race, NYHA, smoke,

smoke1m, ACEI, HR, SBP, DBP, LVEF, LVEDD,

Lab1-2,

Lab4-12, Lab14-17,

MED1, MED3, MED5, MED7, MED9, MED10-13, MED15,MED17-18

ntra1-2, Intra5, Intra7, ECMO, Sur1, Sur3-5

gender, age, race, NYHA, sd,

LVEF, LVEDD, Lab2, Lab4,

Lab6-14, Lab17-18,

MED2, MED4, MED7,

MED14, MED18, Intra5-7,

Sur2-3, Sur5

Wrapper

age, BMI, NYHA, β-blocker,

Temp, HR, SBP, DBP, LVEF,

LVEDD, Lab1-17, MED7,

MED13, MED15, MED18

Intra2, Intra3, Intra5-7,

Sur3

gender, age, BMI, NYHA, smoke, sd, LVEF, LVEDD, Lab2-14, Lab17, Lab18, MED4, MED13, MED7,

MED18, Intra5-7, Sur2-3

Embedded

age, BMI, HR, SBP, LVEF, LVEDD,

Lab2, Lab4, Lab7, Lab9-12,

Lab14, Lab15, Lab17,

Intra5-6

BMI, NYHA, LVEF, Lab9-12,

Lab17

Logistic Regression

gender, age, ace, BMI, NYHA, β-blocker, HR, DBP, SBP, LVEF,

LVEDD, Lab2-4,

Lab6-18, MED4, MED5,

MED7, MED10-11,

MED13-18, Intra2-3,

Intra5-7, Sur2-5

gender, age, race, BMI, NYHA, SD, LVEF, LVEDD, Lab2-4,

Lab6-14, Lab17, Lab18,

MED4-7, MED13-14,

MED16, MED18, Intra5-7,

Sur2, Sur3, Sur5

Abbreviations: BMI=body mass index; NYHA=Classification of NYHA heart function; smoke1m=smoke within a latest month; group=high or low dosage tranexamic acid; ACEI=angiotensin converting enzyme inhibitors; Temp=body temperature before surgery; HR=heart rate; SBP=systolic blood pressure; DBP=diastolic blood pressure; sd=SBP-DBP; LVEF=left ventricular ejection fraction; LVEDD=Left ventricular end diastolic dimension; Lab1=white blood cell count; Lab2=neutrophil count; Lab3=hemoglobin count; Lab4= platelet count; Lab5=ALT, alanine aminotransferase; Lab6=AST, alkaline phosphatase; Lab7=GGT, gamma-glutamyl transferase; Lab8=ALP, alkaline phosphatase; Lab9=direct bilirubin; Lab10=total bilirubin; Lab11=serum creatinine; Lab12=BUN, blood urea nitrogen; Lab13=TP, total protein; Lab14=ALB, albumin; Lab15=PT, prothrombin time; Lab16=D-dimer; Lab17=NT-proBNP, N-terminal brain sodium peptide; Lab18=Hs-CRP, hypersensitive C-reactive protein; MED1=allergy; MED2= diabetes mellitus; MED3=hypertension; MED4=hyperlipidemia; MED5=chronic kidney disease; MED6=COPD, chronic obstructive pulmonary disease; MED7=peripheral vascular disease; MED8=cerebrovascular accident; MED9=infective endocarditis; MED10=fully recovered ischemic central nervous system injury; MED11=lacunar cerebral infarction; MED12=non-invasive tests suggesting carotid artery stenosis >79%; MED13=previous cardiac valvular disease; MED14=previous coronary heart disease; MED15=previous congenital heart disease; MED16=previous aortic disease; MED17=previous carotid surgery; MED18=previous peripheral vascular disease; Intra1=volume of intraoperative blood salvage; Intra2=lowest temperature during CPB; Intra3=hemoglobin at end of CPB; Intra4=CPB again; Intra5=surgery duration; Intra6=CPB duration; Intra7=aortic clamping duration; Sur1=emergency surgery; Sur2=CABG, coronary Artery Bypass Graft ; Sur3=valve Surgery; Sur4=congenital heart disease surgery; Sur5=aortic surgery; CPB=cardiopulmonary bypass; IABP=Intra-aortic balloon pump; ECMO=Extracorporeal Membrane Oxygenation

Table 3: Top 10 features of different FS methods

	Filter	Wrapper	Embedded
Group 1	Lab17	Lab17	Lab17
	Lab15	Lab15	Lab15
	Lab12	LVEF	Lab12
	LVEF	Lab12	LVEF
	Lab4	Lab9	Lab4
	Lab9	Lab4	Lab9
	HR	Lab11	HR
	Lab2	HR	Lab10
	Lab10	Lab10	Lab7
	Lab11	SBP	Lab14
Group 2	Lab17	Lab17	Lab17
	Lab15	Lab15	Lab15
	LVEF	Lab12	LVEF
	Lab12	LVEF	Lab12
	Lab10	Lab9	Lab10
	Lab9	Lab4	Lab11
	Lab4	Lab10	Lab4
	Lab11	HR	Lab9
	SBP	Lab11	HR
	Lab2	SBP	SBP

Abbreviations:

HR = heart rate; SBP = systolic blood pressure; DBP = diastolic blood pressure; LVEF= left ventricular ejection fraction; Lab2 = neutrophil count; Lab4 = platelet count; Lab5= ALT, alanine aminotransferase; Lab7= GGT, gamma-glutamyl transferase. Lab8 = ALP, alkaline phosphatase; Lab9=direct bilirubin; Lab10 = total bilirubin; Lab11 = serum creatinine; Lab1 2= BUN, blood urea nitrogen; Lab15 = PT, prothrombin time; Lab16=D-dimer; Lab17= NT-proBNP, N-terminal brain sodium peptide; Lab18 = Hs-CRP, hypersensitive C-reactive protein.

No competing interests reported.

Download PDF

Version 1

posted

You are reading this latest preprint version

Features selection in a predictive model for cardiac surgery- associated acute kidney injury

Status:

Version 1

Abstract

Background

Methods

Results

Conclusions

Figures

Introduction

Materials & Methods

Study population

Data collection

Study design

AKI definition

Statistical analysis

Feature selection and model evaluation

Results

Discussion

Conclusions

Abbreviations

Declarations

References

Tables

Additional Declarations

Status:

Version 1