Content uploaded by Edin Begic
Author content
All content in this area was uploaded by Edin Begic on Jan 12, 2024
Content may be subject to copyright.
Integrating Machine Learning in Clinical
Decision Support for Heart Failure Diagnosis:
Case Study
Lemana Spahi´c1,2,3,4,5,6,7,8,9,AdnaSofti´c1,2,3,4,5,6,7,8,9(B),
Azra Durak-Nalbanti´c1,2,3,4,5,6,7,8,9,EdinBegi´c1,2,3,4,5,6,7,8,9,
Bojan Staneti´c1,2,3,4,5,6,7,8,9,andHarisVrani´c1,2,3,4,5,6,7,8,9
1Verlab Research Institute for Biomedical Engineering, Medical Devices and Artificial
Intelligence, Sarajevo, Bosnia and Herzegovina
adna@verlabinstitute.com
2International Burch University Sarajevo, Sarajevo, Bosnia and Herzegovina
3Bioengineering Research and Development Center BioIRC, Kragujevac, Serbia
4Clinic for Heart, Blood Vessels and Rheumatism, Clinical Center University of Sarajevo,
Sarajevo, Bosnia and Herzegovina
5Sarajevo School of Science and Technology, Sarajevo, Bosnia and Herzegovina
6General Hospital Abdulah Nakaš, Sarajevo, Bosnia and Herzegovina
7University Clinical Centre of the Republic of Srpska, Banja Luka, Bosnia and Herzegovina
8Medical Faculty, University of Banja Luka, Banja Luka, Bosnia and Herzegovina
9Sarajevo School of Science and Technology, 71 000 Sarajevo, Bosnia and Herzegovina
Abstract. Heart failure is the leading cause of hospitalization in people older than
65. Accurate referrals can reduce the devastating impact of heart failure. Timely
diagnosis of heart failure from other cardiovascular conditions based only on
symptoms is a major challenge. Machine learning has demonstrated potential for
overcoming the diagnostic challenges of cardiovascular diseases. Many research
papers are now focusing on application of artificial intelligence methods applied
to diagnosis of heart failure, where databases continue to be a limitation. The
current study used a dataset of 368 patients (297 patients with diagnosed heart
failure, 71 control subjects) from an upper middle-income country, containing
information on subject population characteristics, symptoms and laboratory test
results. Manual feature selection was performed, focusing on clinical symptoms
that are easily measurable. Four common machine learning methods were tested
and compared: Decision Tree (DT) algorithm, Random Forest (RF) algorithm,
Support Vector Machine (SVM) and Naïve Bayes (NB) algorithm. Models were
developed through a holdout process of training-validation and testing. Our final
model was a Decision Tree, achieving an AUC of 94.3%, with the advantage of
being fully intelligible and easily interpreted. The performance achieved suggested
that intelligible machine learning models can enhance symptom-based referral of
heart failure.
Keywords: Artificial intelligence ·Heart failure ·Prediction ·Machine learning
© The Author(s), under exclusive license to Springer Nature Switzerland AG 2024
A. Badnjevi´c and L. Gurbeta Pokvi´c (Eds.): MEDICON 2023/CMBEBIH 2023, IFMBE Proceedings 93, pp. 696–705, 2024.
https://doi.org/10.1007/978-3-031-49062-0_73
Integrating Machine Learning in Clinical Decision Support 697
1 Introduction
Heart failure remains a significant health issue for the general public, and it is anticipated
that its prevalence will rise by 46% between 2012 and 2030 [1]. Heart failure (HF) is
along-lastingandgradualconditioncharacterizedbyabnormalitiesinthestructureor
function of the heart that lead to either a decrease in the left ventricular ejection fraction
(LVEF) known as HF with reduced ejection fraction (HFrEF), or a preserved LVEF
known as HF with preserved ejection fraction (HFpEF) [2]. According to the World
Health Organization’s research, more than 5% of adults encounter errors in diagnosis.
The most frequent diagnostic errors were identified by analyzing 190 cases in a study,
which found that congestive heart failure is misdiagnosed in 7% of cases, a higher rate
than the 5% misdiagnosis rate for cancer [3]. Late diagnosis of cardiovascular conditions
is dangerous because the clinical outcomes and chances of recovery deteriorate quickly
in such situations. Diagnostic errors may occur when healthcare providers overlook mild
early symptoms and fail to conduct laboratory tests or additional medical exams to assess
essential cardiac parameters, such as hypertension or high blood sugar and cholesterol
levels.
The problem of defining the prognosis in HF is complex and represents a challenge
due to multiple etiologies, different age patients, frequent different comorbidities, lim-
ited possibilities of examining paracrine pathophysiological systems, distinguishing the
entities systolic versus diastolic relaxation, diverse individual progression of the disease
and outcome (sudden death compared to death in progressive heart failure), effectiveness
of treatment and therapeutic adherence. Due to the influence of all the above, as well as
their interaction, which is still incompletely examined, it is difficult to predict how each
individual will respond to a specific treatment, that is, it complicates the prediction of
the clinical outcome. Risk stratification based on existing traditional predictors is often
not adequate in daily clinical practice because it leads to failure to recognize high-risk
patients or inadequate recognition of low-risk as high-risk patients. The reason for this
may be that these conventional predictors do not reflect the existing, newly discovered
pathophysiological mechanisms in heart failure, namely inflammation, oxidative stress,
neurohumoral activation, vascular remodeling, myocyte injury, renal comorbidity. Thus,
biomarkers that represent their reflection can be potentially new predictors of survival
that will redefine risk stratification.
The artificial intelligence (AI) and Machine learning (ML) algorithms have the poten-
tial to improve the accuracy of heart failure diagnosis by identifying patterns in patient
data that may be difficult for humans to detect. The advantages of implementing AI in the
healthcare sector stem from the ability to store and manage vast amounts of data, enhance
the reliability of clinical communication, minimize the incidence of human error, and
enhance disease detection, leading to lower mortality rates. Additionally, employing AI
technology in healthcare can result in a reduction of costs, with reports suggesting that
patient outcomes may be improved by as much as 30–40% and treatment costs cut by
up to 50% [4].
This paper presents how machine learning can be used to support clinical decision-
making and improve heart failure diagnosis accuracy.
698 L. Spahi´c et al.
2 Methods
The dataset obtained from a local hospital consisted of 245 samples of patients that were
admitted to hospital with diagnosis heart failure and 255 samples of patients that did not
have any indication of heart failure. Due to the fact that almost half of the patients did not
have key parameters recorded and several others have had parameters that represented
significant outliers, they have been removed from the dataset. In the end, the dataset
used for the development of machine learning algorithms consisted of 356 samples.
The data was divided into training and testing subset with an approximately equal
representation of healthy and heart failure instances in an 80–20 (%) splitting ratio.
This splitting ratio is the most common for application of machine learning algorithms
[5]. The training subset consisted of 297 samples while the validation dataset consisted
of 71 samples. Overall, the training dataset consisted of 143 heart failure patients and
154 healthy individuals. The parameters were grouped into 4 groups: demographic fea-
tures, medical history, clinical image and biochemical parameters. The basic descriptive
statistics of all parameters included as inputs and their correlation with heart failure are
presented in Tables 1and 2.
Table 1. Statistical analysis of continuous variables in the dataset
Attribute Min Max Mean STD
Age [year] 23 90 67 14.34
Trop onin [ pg/mL] 4.7 1946 71.59 176.187
Cystatin [mg/L] 0.005 5.01 1.003 0.707
CA125 [units/mL] 0.326 1760.8 89.27 163.67
LAV volume [mL/m2]0.079 330 57.87 57.77
Right ventricular systolic
pressure (RVSP) [mmHg]
1125 35.65 20.34
E/e′′ 124 11.68 6.07
TAPSE [mm] 724 17.39 4.32
Creatinine [µmol/L] 41 266 94.72 30.73
EGFR [mL/min/1.73 m2]16 140 79.14 26.09
HGB [g/L] 76.4 178 140.49 18.19
Na [mEq/L] 121 148 139.84 3.49
Cholesterol [mmol/L] 1.128 7.4 3.48 1.19
Albumins [g/L] 20 53 39.27 6.71
Other variables that were introduced to the machine learning algorithms were
descriptive such as: Sex, presence or absence of COPD, tricuspid regurgitation, mitral
regurgitation and diastolic function. In order to select the parameters expected to have
the most pronounced influence on heart failure, Pearson correlation was determined for
all parameters in relation to heart failure diagnosis (Table 2).
Integrating Machine Learning in Clinical Decision Support 699
Table 2. Parameters and their correlation with HF
Parameter Correlation (%)
Diastolic function 69
Left atrial volume 63
Chronic obstructive pulmonary disease 63
Age 51
Cystatin 50
Pulmonary artery pressure 49
Cholesterol 44
Tricuspid Annular Plane Systolic Excursion (TAPSE) 39
Estimated Glomerular Filtration Rate (eGFR) 34
Tricuspid regurgitation (TR) 34
CA125 21
Natrium (Na) 19
Creatinine 13
Troponin 13
Sex 11
E/e′′ 3
Hemoglobin (HGB) 2
Mitral Regurgitation (MR) 0
Albumines 0
For the purpose of development of the expert system, several machine learning
algorithms were tested, but the overall flow of decision making in this clinical decision
support system can be represented graphically as in Fig. 1.
Fig. 1. Block diagram of the classification system
As this is a classification task, the output of the expert system is patient status
expressed in the form of “at risk for heart failure”and“no risk of heart failure”. A range
of different parameters represented as both categorical and continuous variables are used
700 L. Spahi´c et al.
with respect to their pronounced diagnostic significance for heart failure. The architecture
of the expert system in the processing unit is dependent upon the choice of the machine
learning algorithm Following machine learning algorithms were implemented and tested:
Decision Tree (DT) algorithm, Random Forest (RF) algorithm, Support Vector Machine
(SVM), Naïve Bayes (NB).
Performance of the proposed approaches was initially evaluated through comparisons
of accuracy and computational time, while the best performing algorithm was further
evaluated for accuracy, sensitivity and specificity.
Accuracy =TP +TN
TP +FP +TN +FN ;Sensitivity =TP
TP +FN ;Specificity =TN
TN +FP ;
where: TP means true positive, TN means true negative, FP means false positive and FN
means false negative. The confusion matrix was used to measure classifier’s accuracy
by comparing actual and predicted values.
3 Results and Discussion
Artificial intelligence (AI) has become increasingly important in medicine [6,7], espe-
cially in cardiology due to advancements in information and communication technology
that enable the easy storage, acquisition, and retrieval of large amounts of data [8,9].
Proper diagnosis of heart failure (HF) is critical because patients with HF have a poor
prognosis similar to that of oncologic malignancies, and medication can increase sur-
vival [10,11]. Assessment of left ventricular systolic and diastolic function is essential
in determining the patient’s therapeutic modality, with brain natriuretic peptide (BNP)
and N-terminal proBNP (NT-proBNP) being strong predictors of clinical outcomes [12,
13]. The left atrial volume index (LAVI) is a key determinant of diastolic function and
apredictorofnatriureticpeptideincreaseandpatientprognosis.Severalotherfactors
outside of left ventricular function, such as an increase in the left atrium diameter, mitral
regurgitation, and right ventricular dysfunction, are potential predictors of survival in
HF [14–16]. BNP and NT-proBNP levels have prognostic value in patients with acute
HF, but serial measurements may be necessary for therapeutic guidance. BNP and NT-
proBNP values are affected by age, gender, and body mass index, and renal failure can
increase NT-proBNP levels. However, some patients with acute HF may not have ele-
vated BNP or NT-proBNP levels, and their values should be correlated with clinical
data. Acute regional dysfunction of the left ventricle is a sign of myocardial ischemia,
leading to cell necrosis [17–20].
The presence of mitral and tricuspid regurgitation, renal injury or failure, elevated
levels of cystatin C, troponin, and CA125, and comorbidities such as chronic obstructive
pulmonary disease (COPD) are all predictors of poor prognosis and patient outcomes in
heart failure. Mitral and tricuspid regurgitation are associated with diastolic dysfunction
and may require interventional surgery. Renal injury or failure can occur even in the
absence of intrinsic renal disease and is related to high renal venous pressure, which
reduces renal perfusion pressure. Cystatin C is a superior marker of glomerular filtra-
tion compared to creatinine and predicts the degree of glomerular filtration accurately.
Elevated levels of troponin in heart failure are associated with an unfavorable outcome,
Integrating Machine Learning in Clinical Decision Support 701
and inflammation may play a role in the secretion of CA125 in heart failure. COPD
exacerbates the progression of atherosclerosis and the development of ischemic heart
disease, leads to pulmonary hypertension and right-sided HF, and is associated with a
higher prevalence of comorbidities and lower usage of beta-blockers [21–28].
This paper presents the work on developing an expert system based on machine
learning algorithms for prediction of HF. During this research 5 different expert systems
were built. Figure 2.Representstheinitialresultsandcomparisonbetweenthetested
machine learning algorithms.
Fig. 2. Expert system performance comparison based on different ML algorithms
As it can be deduced from Fig. 2.TheRFandDTclassifiershadthebesttraining
accuracy, however, it was important to determine what was the most common point of
failure for other algorithms, so their validation performance was observed (Tables 3–6).
Table 3. Performance evaluation for expert system based on Decision Tree algorithm
HealthyExpert system Heart failureExpert system
HealthyREAL 31 0
Heart failureREAL 436
Specificity
88.6%
Sensitivity
100%
Accuracy 94.3%
As it can be seen from Tables 3–6,theunderperformingalgorithmssuchasNB,
DT and SVM had faced a bottleneck with classification of healthy instances. In order
to analyze this occurrence, the structure of the dataset was taken into account. As there
were 19 variables in the dataset in total, some of them non-specific to heart failure such
as the age, sex and presence of COPD, it can be discussed that these classifiers have
performed misclassification due to this.
Patient treatment and likelihood of survival is significantly impacted by early HF
diagnosis since it allows for effective and potentially successful therapy prior to any major
702 L. Spahi´c et al.
Table 4. Performance evaluation for expert system based on Random Forest algorithm
HealthyExpert system Heart failureExpert system
HealthyREAL 35 0
Heart failureREAL 036
Specificity
100%
Sensitivity
100%
Accuracy 100%
Table 5. Performance evaluation for expert system based on Support Vector Machine algorithm
HealthyExpert system Heart failureExpert system
HealthyREAL 0 0
Heart failureREAL 35 36
Specificity
0%
Sensitivity
100%
Accuracy 50.7%
Table 6. Performance evaluation for expert system based on Naive Bayes algorithm
HealthyExpert system Heart failureExpert system
HealthyREAL 0 0
Heart failureREAL 35 36
Specificity
0%
Sensitivity
100%
Accuracy 50.7%
decline in cardiac output. On the other hand, a late-stage diagnosis of HF dramatically
reduces the therapeutic advantages of therapies and the likelihood of survival [29]. Early
HF diagnosis is possible with the use of machine learning and data mining techniques. By
extracting useful information from the patient data sets, these approaches can efficiently
identify heart failure at an early stage [30].
AstudybyGuoetal.assessedresearchthatmadeuseofelectronichealthrecords,
including studies that looked at demographic data, medical history, laboratory and imag-
ing results, and genetic profiles. They make claims about the great accuracy of these
prediction tools while also acknowledging the difficulties that cutting-edge machine-
learning algorithms still have to face. One of the most prevalent issues in this area is that
the electronic health record cannot be fully integrated (medical reports, wide variety of
imaging results). However, since these algorithms are based on machine learning, they
are not suitable for patients with rare disorders or unusual profiles [31,32]. Pan˘aetal.
Integrating Machine Learning in Clinical Decision Support 703
(2021) implemented a contactless system that can predict the worsening of heart failure
through voice analysis. The dataset of this study consisted of 16 samples. They used
algorithms such as Artificial Neural Networks (ANN), Support Vector Machine (SVM)
and K Nearest Neighbors (KNN). The study sample was small, and consequently the
obtained results are preliminary [33].
The dataset used for our study consisted of 356 samples and machine learning
algorithms that were implemented and tested are: Decision Tree (DT), Random Forest
(RF), Support Vector Machine (SVM), Naïve Bayes (NB) and Artificial Neural Network
(ANN).
The Artificial Intelligence-Clinical Decision Support System (AI-CDSS) is a hybrid
tool that combines expert and machine learning input to help doctors diagnose heart
failure. In their published study, Dong-Ju Choi, Jin Joo Park, and colleagues assessed the
diagnostic efficacy of the AI-CDSS on a group of 97 dyspnea patients. They measured the
degree of agreement between the algorithm’s findings and those of experts in heart failure.
With a concordance rate of 98% between AI-CDSS and heart failure specialists, 44
percent of the 97 patients had heart failure. The concordance rate between the AI-CDSS
and non-heart failure specialists was 76%, on the other hand. Finally, they emphasized
the value of AI-CDSS in diagnosing heart failure, particularly in cases where a heart
failure expert is not available [11].
4 Conclusion
The utilization of machine learning in heart failure is rapidly expanding. ML algorithms
have the potential to aid in HF diagnosis, classification, and prognosis. The availability of
validated ML tools in clinical practice will undoubtedly have a positive impact on HF care
and outcomes. In our study, we evaluated four machine learning techniques, including
the Decision Tree (DT), Random Forest (RF), Support Vector Machine (SVM), and
Naïve Bayes (NB) algorithms. These models were developed using a holdout process of
training-validation and testing. Our final model was the Decision Tree algorithm, which
achieved an AUC of 94.3%. Furthermore, this model is fully interpretable and easily
understood, providing an advantage over other methods. The results of our study suggest
that interpretable machine learning models have the potential to enhance the referral of
HF patients based on their symptoms.
References
1. Di Palo, K.E., Barone, N.J.: Hypertension and heart failure: Prevention, targets, and treatment.
Cardiol. Clin. 40(2), 237–244 (2022). https://doi.org/10.1016/j.ccl.2021.12.011
2. Arrigo, M., et al.: Acute heart failure. Nat. Rev. Dis. Primers. 6(1), 16 (2020). https://doi.org/
10.1038/s41572-020-0151-7
3. Še´ckanovi´c, A., Šehovac, M., Spahi´c, L., Rami´c, I., Mamatnazarova, N., Gurbeta Pokvi´c,
L., Badnjevi´c, A., Kacila, M.: Review of artificial intelligence application in cardiology. In:
2020 9th Mediterranean Conference on Embedded Computing (MECO), Budva, Montenegro,
pp. 1–5 (2020) doi:https://doi.org/10.1109/MECO49872.2020.9134333
4. WHO: Retrieved from: https://apps.who.int/iris/bitstream/handle/10665/252410/978924151
1636-eng.pdf,January10th(2020)
704 L. Spahi´c et al.
5. Visalakshi, S., Radha, V.: A literature review of feature selection techniques and applica-
tions: Review of feature selection in Data Mining. In: 2014 IEEE International Conference
on Computational Intelligence and Computing Research (2014) doi:https://doi.org/10.1109/
iccic.2014.7238499
6. Sari´c, R., Joki ´c, D., Beganovi´c, N., Pokvi´c, L.G., Badnjevi´c, A.: FPGA-based real-time epilep-
tic seizure classification using Artificial Neural Network. Biomed. Signal Process. Control
62,102106(2020).https://doi.org/10.1016/j.bspc.2020.102106
7. Stokes, K., et al.: A machine learning model for supporting symptom-based referral and
diagnosis of bronchitis and pneumonia in limited resource settings. Biocybern. Biomed. Eng.
41(4), 1288–1302 (2021). https://doi.org/10.1016/j.bbe.2021.09.002
8. Johnson, K.W., et al.: Artificial intelligence in cardiology. J. Am. Coll. Cardiol. 71(23),
2668–2679 (2018). https://doi.org/10.1016/j.jacc.2018.03.521
9. Begic, E., et al.: From heart murmur to echocardiography—congenital heart defects
diagnostics using machine-learning algorithms. Psychiatr. Danub. 33(Suppl 13), 236–246
(2021)
10. Ponikowski, P., Voors, A. A., Anker, S. D., Bueno, H., Cleland, J. G. F., Coats, A. J. S.,
Falk, V., González-Juanatey, J. R., Harjola, V. P., Jankowska, E. A., Jessup, M., Linde, C.,
Nihoyannopoulos, P., Parissis, J. T., Pieske, B., Riley, J. P., Rosano, G. M. C., Ruilope, L. M.,
Ruschitzka, F., Rutten, F. H., ESC Scientific Document Group: 2016 ESC Guidelines for the
diagnosis and treatment of acute and chronic heart failure: The Task Force for the diagnosis
and treatment of acute and chronic heart failure of the European Society of Cardiology (ESC)
Developed with the special contribution of the Heart Failure Association (HFA) of the ESC.
Eur. Hear. J. 37(27), 2129–2200 (2016)https://doi.org/10.1093/eurheartj/ehw128
11. Choi, D.J., Park, J.J., Ali, T., Lee, S.: Artificial intelligence for the diagnosis of heart failure.
NPJ Digit. Med. 3, 54 (2020). https://doi.org/10.1038/s41746-020-0261-3
12. Cao, Z., Jia, Y., Zhu, B.: BNP and NT-proBNP as diagnostic biomarkers for cardiac dysfunc-
tion in both clinical and forensic medicine. Int. J. Mol. Sci. 20(8), 1820 (2019). https://doi.
org/10.3390/ijms20081820
13. Alcidi, G., Goffredo, G., Correale, M., Brunetti, N.D., Iacoviello, M.: Brain natriuretic peptide
biomarkers in current clinical and therapeutic scenarios of heart failure. J. Clin. Med. 11(11),
3192 (2022). https://doi.org/10.3390/jcm11113192
14. Lim, T.K., Ashrafian, H., Dwivedi, G., Collinson, P.O., Senior, R.: Increased left atrial volume
index is an independent predictor of raised serum natriuretic peptide in patients with suspected
heart failure but normal left ventricular ejection fraction: Implication for diagnosis of diastolic
heart failure. Eur. J. Heart Fail. 8(1), 38–45 (2006). https://doi.org/10.1016/j.ejheart.2005.
05.008
15. Kim, M.K., et al.: Tissue Doppler-derived E/e’ ratio as a parameter for assessing diastolic
heart failure and as a predictor of mortality in patients with chronic kidney disease. Korean
J. Intern. Med. 28(1), 35–44 (2013). https://doi.org/10.3904/kjim.2013.28.1.35
16. Strachinaru, M., et al.: Relation between E/e’ ratio and NT-proBNP levels in elderly patients
with symptomatic severe aortic stenosis. Cardiovasc. Ultrasound 13, 29 (2015). https://doi.
org/10.1186/s12947-015-0021-8
17. Park, S.J., et al.: N-terminal pro-B-type natriuretic Peptide in overweight and obese patients
with and without diabetes: An analysis based on body mass index and left ventricu-
lar geometry. Korean Circ. J. 39(12), 538–544 (2009). https://doi.org/10.4070/kcj.2009.39.
12.538
18. Gupta, D.K., Wang, T.J.: Natriuretic peptides and cardiometabolic health. Circ. J. 79(8),
1647–1655 (2015). https://doi.org/10.1253/circj.CJ-15-0589
19. Sobhani, K., et al.: Sex differences in ischemic heart disease and heart failure biomarkers.
Biol. Sex Differ. 9(1), 43 (2018). https://doi.org/10.1186/s13293-018-0201-y
Integrating Machine Learning in Clinical Decision Support 705
20. Omar, H.R., Guglin, M.: Acute systolic heart failure with normal admission BNP: Clinical
features and outcomes. Int. J. Cardiol. 232,324–329(2017).https://doi.org/10.1016/j.ijcard.
2016.12.069
21. Mutlak, D., Lessick, J., Khalil, S., Yalonetsky, S., Agmon, Y., Aronson, D.: Tricuspid regurgi-
tation in acute heart failure: Is there any incremental risk? Eur. Heart J. Cardiovasc. Imaging
19(9), 993–1001 (2018). https://doi.org/10.1093/ehjci/jex343
22. Qiu, X., Liu, C., Ye, Y., Li, H., Chen, Y., Fu, Y., Liu, Z., Huang, X., Zhang, Y., Liao, X., Liu,
H., Zhao, W., Liu, X.: The diagnostic value of serum creatinine and cystatin c in evaluating
glomerular filtration rate in patients with chronic kidney disease: A systematic literature
review and meta-analysis. Oncotarget 8(42), 72985–72999 (2017) https://doi.org/10.18632/
oncotarget.20271
23. Einwoegerer, C.F., Domingueti, C.P.: Association between increased levels of cystatin c and
the development of cardiovascular events or mortality: A systematic review and meta-analysis.
Arq. Bras. Cardiol. 111(6), 796–807 (2018). https://doi.org/10.5935/abc.20180171
24. Falcão, F. J. A., Oliveira, F. R. A., Cantarelli, F., Cantarelli, R., Brito-Júnior, P., Lemos, H.,
Silva, P., Camboim, I., Freire, M. C., Carvalho, O., & Sobral-Filho, D. C.: Carbohydrate
antigen 125 predicts pulmonary congestion in patients with ST-segment elevation myocardial
infarction. Braz. J. Med. Biol. Res. =Rev. Bras. Pesqui. Medicas E Biol. 52(12), e9124
(2019) doi:https://doi.org/10.1590/1414-431X20199124
25. Durak-Nalbantic, A., et al.: Predictors of hospitalization for heart failure decompensation in
18-months follow-up after index hospitalization for acute heart failure. Med. Arch. (Sarajev.,
Bosnia Herzeg.) 72(4), 257–261 (2018). https://doi.org/10.5455/medarh.2018.72.257-261
26. Durak-Nalbanti´c, A., Džubur, A., Dili´c, M., Pozderac, Z., Mujanovi´c-Naranˇci´c, A., Kuli´c,
M., Hodži´c, E., Resi´c, N., Brdjanovi´c, S., Zvizdi´c, F.: Brain natriuretic peptide release in
acute myocardial infarction. Bosn. J. Basic Med. Sci. 12(3), 164–168 (2012) doi:https://doi.
org/10.17305/bjbms.2012.2470
27. Durak-Nalbantic, A., et al.: Serum level of tumor marker carbohydrate antigen-CA125 in
heart failure. Med. Arch. (Sarajev., Bosnia Herzeg.) 67(4), 241–244 (2013). https://doi.org/
10.5455/medarh.2013.67.241-244
28. Donal, E., et al.: New echocardiographic predictors of clinical outcome in patients presenting
with heart failure and a preserved left ventricular ejection fraction: A subanalysis of the Ka
(Karolinska) Ren (Rennes) Study. Eur. J. Heart Fail. 17(7), 680–688 (2015). https://doi.org/
10.1002/ejhf.291
29. Ponikowski, P., et al.: Heart failure: preventing disease and death worldwide. ESC Hear. Fail.
1(1), 4–25 (2014). https://doi.org/10.1002/ehf2.12005
30. Aljaaf, A. J., Al-Jumeily, D., Hussain, A. J., Dawson, T., Fergus, P., Al-Jumaily, M.: Predicting
the likelihood of heart failure with a multi-level risk assessment using decision tree. In: 2015
Third International Conference on Technological Advances in Electrical, Electronics and
Computer Engineering (TAEECE) (2015) doi:https://doi.org/10.1109/taeece.2015.7113608
31. Guo, A., Pasque, M., Loh, F., Mann, D.L., Payne, P.R.: Heart failure diagnosis, readmission,
and mortality prediction using machine learning and Artificial Intelligence Models. Curr.
Epidemiol. Rep. 7(4), 212–219 (2020). https://doi.org/10.1007/s40471-020-00259-w
32. Shameer, K., et al.: Predictive modeling of hospital readmission rates using electronic medical
record-wide machine learning: A case-study using mount Sinai heart failure cohort. Pac.
Symp. Biocomput. 22,276–287(2017).https://doi.org/10.1142/9789813207813_0027
33. Pan˘a, M.-A., et al.: Reducing the heart failure burden in Romania by predicting congestive
heart failure using Artificial Intelligence: Proof of concept. Appl. Sci. 11(24), 11728 (2021).
https://doi.org/10.3390/app112411728
