Nomograms for Predicting the Risk of Acute Heart Failure in ICU Patients with Atrial Fibrillation

doi:10.21203/rs.3.rs-3792463/v1

Download PDF

Research Article

Nomograms for Predicting the Risk of Acute Heart Failure in ICU Patients with Atrial Fibrillation

https://doi.org/10.21203/rs.3.rs-3792463/v1

This work is licensed under a CC BY 4.0 License

Version 1

posted

You are reading this latest preprint version

Background

This study developed two nomograms to predict the incidence of acute heart failure (AHF) in patients of atrial fibrillation (AF) and evaluated the predictive value of the nomograms.

Methods

18179 patients of AF from the Medical Information Mart for Intensive Care IV (MIMIC-IV) database were randomly divided into two groups in a seven-to-three ratio to form a training cohort (n = 12725) and a validation cohort (n = 5454). Least absolute shrinkage and selection operator (LASSO) regression analyses were used to identify associated risk variables. Two nomograms were established. The new models were assessed in terms of the concordance index (C-index), the area under the curve (AUC) of receiver operating characteristic (ROC) analysis, calibration curve, and decision curve analysis (DCA).

Results

Least absolute shrinkage and selection operator (LASSO) regression analysis identified seven potential predictors of acute heart failure in nomogram 1 and three potential predictors in nomogram 2. Multivariate logistic regression analysis was used to evaluate the effects of these predictors and create final models. The concordance index values were 0.768 and 0.696, respectively. The areas under the curves for the training and validation sets in nomogram 1 were 0.768 and 0.763, respectively. The areas under the curves for the training and validation sets in nomogram 2 were 0.696 and 0.692, respectively.

Conclusion

Seven variables were identified as predictors in nomogram 1 and three variables were identified in nomogram 2. Our nomograms are reliable convenient approaches for predicting acute heart failure in patients with atrial fibrillation.

nomogram

Medical Information Mart for Intensive Care

acute heart failure

atrial fibrillation

Atrial fibrillation (AF) is an atrial ectopic tachyarrhythmia [1]. The loss of conduction and refractory period synchronisation in various parts of the atrium leads to uncoordinated AF in muscle fibres [2]. Possible mechanisms for AF include ectopic pacing points, circular motion, and multiple micro re-entrances in the atrium [3]. The Global Burden of Disease study has estimated that there are 33.5 million patients with AF [4]. AF leads to 2-fold and 1.5-fold increases in the all-cause mortality rate in women and men, respectively, due to progressive heart failure, cardiac arrest, and embolism; thus, it is a large burden on society [4, 5].

The relationship between heart failure and AF has received increasing attention in recent years [8]. AF and acute heart failure (AHF) are increasing in prevalence, exacerbate each other, and share clinical features [6]. Animal and human studies have confirmed the connection between AHF and AF [7]. The mechanisms include activation of the renin angiotensin aldosterone system (RAAS), mechanical traction on the left atrium, and atrial electrophysiological remodelling [7, 8]. AF also plays a role in the occurrence and progression of AHF [9]. Thus far, research has mainly focused on the incidence of AF in patients with AHF; few studies have examined the incidence of AHF in patients with AF.

In oncology and pharmacy, nomograms are widely used to determine prognosis and evaluate the incidence of an event [10]. This study used data from the MIMIC-IV database to establish and validate two nomograms that can predict the incidence of AHF in patients with AF; these nomograms can be used by patients and clinicians, respectively.

Patient and public involvement

Patients and/or the public were not involved.

Objectives

This study developed two nomograms to predict the incidence of acute heart failure (AHF) in patients of atrial fibrillation (AF) and evaluated the predictive value of the nomograms.

Database

MIMIC-IV is a publicly available database containing data from 76,540 intensive care unit (ICU) stays from 2008 to 2019 in the Beth Israel Deaconess Medical Center [11]. This database contains all baseline data of patients and most laboratory results that can be used for various studies [12]. It aims to eliminate obstacles to scientific research [13]. The version we used is v2.2, and PostgreSQL v16 was used to search and extract data from it (http://www.postgresql.org/) [14].

Study populations

This is a retrospective analysis using data from patients with AF stored in MIMIC-IV database. The inclusion criteria were as follows:

1. Patients with previous history or new onset of AF.

2. Patients aged older than 18 years.

3. Patients admitted to the ICU for the first time.

Data collection and measurement

The characteristics used for our work included demographic data (e.g., age, gender, race and marital status), vital signs (e.g., heart rate, blood pressure and fluid management), laboratory results (e.g., routine blood tests, chemistry and coagulation) and common score systems in the ICU (e.g., LODS score, CHARLSON score and OASIS score). All data were collected six hours before admission to the ICU to 24 hours after admission to the hospital. Unmarried, married, divorced and widowed status was marked as 0, 1, 2 and 3, respectively. White people were marked as 0, and people of other races were marked as 1. Women were marked as 0, and men were marked as 1.

Statistical analysis

Statistical analysis was achieved by using R software (Version 4.1.3; R Foundation for Statistical Computing, Vienna, Austria; https://www.r-project.org) [15]. The distribution of continuous variables was assessed using the Shapiro–Wilk test. Mean and standard deviation (SD) are reported for continuous variables having a normal distribution; median and interquartile range (IQR), for continuous variables having a skewed distribution; and numbers and percentages, for categorical variables. The Kruskal–Wallis H test was used to compare continuous data. The Fisher’s exact test was used to compare categorical variables. Seventy percent of the data were allocated to the training set for developing the model, and the remaining thirty percent were allocated to the validation set for validation. Records with missing values were deleted or replaced by using chained equations as appropriate. The LASSO method, which accounts for multicollinearity and avoids model overfit, was used to select the best candidate variable for establishing the nomograms [16]. To avoid making the clinical prediction models too complex, the lambda value was set to ‘1se’ so that the error was within one standard error of the minimum. Sensitivity and specificity were used to evaluate the models’ performance. The calibration C index (bootstrap resampling performed 1000 times), the calibration curve and ROC curve were used to evaluate the discrimination and calibration of the models [17]. DCA was used to identify the clinical benefit of the models [18]. The area under the receiver operating characteristic curve values were drawn to assess the stability of the models [19]. All significance tests were two-sided, and a P value < 0.05 was considered statistically significant.

TRIPOD reporting guidelines

We used the TRIPOD reporting guidelines [20].

Clinical characteristics

Supplemental Fig. 1 is a flowchart showing the cohort selection process. This study included 18179 patients. Of these, 70% (12725) were randomly assigned to the training set and the remaining 30% (5454) were assigned to the validation set.

Supplemental table 1 summarizes the data of the enrolled variables. Of the 18179 patients with AF, 4582 were admitted to the ICU for an AHF. The age, weight, body mass index (BMI), heart rate, respiratory rate, fluid management, red blood cells (RBC), white blood cells (WBC), platelet, mean corpuscular volume (MCV), red blood cell distribution width (RDW), anion gap, bicarbonate, BUN, creatinine, Ca, blood glucose, Na, K, international normalized ratio (INR), prothrombin time (PT), partial thromboplastin time (PTT), apsⅢ score, charlson score, lods score and sapsⅡ score were higher in AHF patients with AF than in AHF patients without AF, the height and gcs score were the same, while the mean blood pressure, temperature, hemoglobin, hematocrit, mean corpsular hemoglobin (MCH), mean corpuscular hemoglobin concentration (MCHC), chloride, and RDW were lower. Widowed female patients of AF were more likely to develop AHF.

Variable selection

Initially, 18189 patients and 46 variables were screened. After removing variables with missing values exceeding 5% of the total, 39 variables were retained in the study. Supplemental Fig. 2 shows ROC curves of all screened variables. Supplemental Fig. 3 illustrates AUC values of all variables. Supplemental Fig. 4 is a heatmap involving all screened variables. Variables with missing values not exceeding 5% were replaced by using chained equations. The values of troponin-T and N-terminal pro-b-type natriuretic peptide were missing in more than 50% and 70% of patients, respectively. Although they are significant predictors of cardiac function, once these indicators are tested, it is possible to determine whether AHF has occurred, so these variables were excluded. Twentyfold cross-validated LASSO regression and manual screening identified ten variables: age, respiratory rate, fluid management, MCHC, bicarbonate, BUN and chloride for nomogram 1, and charlson score, lods score and sapsⅡ score for nomogram 2. Figure 1 illustrates the variable selection process. These ten variables were included in the multivariate logistic regression model for predicting the occurrence of AHF, and their odd ratios and 95% confidence intervals (CIs) were 1.32 (1.20–1.46), 1.16 (1.10–1.22), 1.34 (1.25–1.43), 0.72 (0.67–0.79), 1.78 (1.66–1.92), 1.89 (1.77–2.02), 0.69 (0.64–0.74), 2.61 (2.36–2.88), 2.34 (2.07–2.66) and 0.55 (0.49–0.63) respectively

Nomograms construction

A nomogram was established for predicting the occurrence of AHF among AF patients for use by patients (Fig. 2) and the other nomogram was established for use by clinicians (Fig. 3). The ten variables identified above were assigned different initial scores and a total score was calculated. The probability of developing AHF increased with the total score.

Assessment and validation

The calibration curves demonstrated high conformity between the predicted and observed probabilities (Fig. 4). ROC curves for the training and test sets were used to evaluate the discriminability of the models. The area under the ROC curve was 0.768 in the training set and 0.763 in the validation set in nomogram 1. The area under the ROC curve was 0.696 in the training set and 0.692 in the validation set in nomogram 2. The results indicated that our prediction models have good stability and accuracy. Figure 5 shows that the higher the threshold values are, the greater the net benefit of the nomograms decision curves. This indicates the good clinical values of our nomograms.

This population-based study retrospectively investigated 18,179 ICU patients with AF. Two nomograms were developed and validated to gauge the risk of AHF in those patients. The nomograms contained 10 variables: age; respiratory rate; fluid management; mean corpuscular haemoglobin concentration (MCHC); bicarbonate; blood urea nitrogen (BUN); chloride; and the Charlson, LODS, and SAPS II scores. Both nomograms had good clinical performance. ROC analysis of the training and validation cohorts revealed AUCs of 0.768 and 0.763, respectively, for nomogram 1; these AUCs were 0.696 and 0.692 for nomogram 2. The findings suggest that our models are sufficiently predictive.

The relationship between AF and heart failure remains an area of study [21]. Ischemia, blood pressure, blood glucose, age, and atherosclerosis are generally regarded as risk factors for both diseases [21, 22]. The incidences of heart failure and AF both gradually increase with age [23], consistent with our findings.

When heart failure occurs, the respiratory rate increases. Breathing adjustments affect the treatment of heart failure [24]. This study recognised respiratory rate as a risk factor for AHF in patients with AF, similar to the results in some previous studies [25]. Input and output volumes are important for evaluating the degree of heart failure [26]; reductions in output and input are termed fluid management. Our model 1 showed that greater fluid management increases the risk of AHF. Relative erythrocyte hypochromia, which manifests as a low MCHC, affects the onset of AHF [27]; a decrease in the MCHC corresponds to an increase in the risk of AHF. The serum bicarbonate level also significantly increases in patients with AHF, and it can serve as a predictive indicator. BUN is related to renal function and is not an independent risk factor for AHF, although it can serve as a warning signal. Chloride is associated with cardiorenal and neuroendocrine systems [28], and many studies regard the chloride level as a prognostic marker [29]. The present study identified it as a predictive factor, such that a decrease in chlorine was accompanied by an increase in the incidence of AHF. Additionally, the present study utilised three scoring systems to establish the second nomogram. The use of model 2 requires some medical expertise; thus, it is designed for clinicians. Although nomogram 2 is less accurate and predictive compared with nomogram 1, it can provide validation for clinical decision-making.

This study had some limitations. First, all data was obtained from the MIMIC-IV database. Incomplete records and possible data errors may have reduced model accuracy. Second, this was a single-centre retrospective study with internal validation alone; external validation is needed to confirm the reliability of the results. Third, blood lipids were not included in the study due to the lack of corresponding records in the MIMIC-IV database. Blood lipids are important risk factors for cardiovascular diseases [30], and further research is needed concerning their implications. Fourth, the types and severity of AF were not considered; thus, we did not determine the risk of AHF at each AF severity level. Additionally, all types of AHF were pooled in our analyses. Finally, our nomograms may predict the possibility of AHF in patients with AF, but they cannot determine whether interventions should be promptly applied.

Two nomograms to predict the risk of AHF in AF patients were developed using data from the MIMIC-IV database. The models had good predictive ability. The nomograms can help to screen AF patients for the risk of AHF and could facilitate good clinical decisions.

Strengths and limitations of this study

This study utilized a large amount of data to establish two models with high accuracy.
This study uses multiple methods to verify the reliability of the models.
This study did not have external validation.
This study only included some common data. Data with incomplete records were excluded.

AHF acute heart failure

AF atrial fibrillation

MIMIC Medical Information Mart for Intensive Care

LASSO least absolute shrinkage and selection operator

AUC area under the curve

ROC receiver operating characteristic

DCA decision curve analysis

C-index Concordance index

SD standard deviation

IQR interquartile range

RBC red blood cells

WBC white blood cells

MCV mean corpuscular volume

RDW red blood cell distribution width

BUN blood urea nitrogen

INR international normalized ratio

PT prothrombin time

PTT partial thromboplastin time

MCH mean corpsular hemoglobin

MCHC mean corpuscular hemoglobin concentration

BMI body mass index

RAAS renin angiotensin aldosterone system

Acknowledgements

We would like to thank all MIMIC project team members.

Declarations

Ethics approval and consent to participate

This study is a retrospective analysis of data stored in a public database. All information is available and free for the public, so the agreement of the medical ethics committee board and the need for informed consent were not necessary.

Consent for publication

Not applicable.

Availability of data and materials

The datasets used/or analyzed during the current study are available from the corresponding authors on reasonable request.

Competing interests

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Funding

No funding.

Author contributions

Ziyang Wu conceived, designed the study and drafted the manuscript. Yong Qiao, Gaoliang Yan, Yuhan Qin, Huihong Tang and Shiqi Liu drafted the manuscript. Chengchun Tang and Dong Wang revised the manuscript and approved the final version to be published. All authors read and approved the final manuscript.

Baman JR, Passman RS, Atrial Fibrillation. JAMA. 2021;325(21):2218. 10.1001/jama.2020.23700. PMID: 34061143.
Zimetbaum P. Atrial Fibrillation. Ann Intern Med. 2017;166(5):ITC33-ITC48. doi: 10.7326/AITC201703070. Erratum in: Ann Intern Med. 2017;166(12):920. PMID: 28265666.
Brundel BJJM, Ai X, Hills MT, Kuipers MF, Lip GYH, de Groot NMS. Atrial fibrillation. Nat Rev Dis Primers. 2022;8(1):21. 10.1038/s41572-022-00347-9. PMID: 35393446.
Wijesurendra RS, Casadei B. Mechanisms of atrial fibrillation. Heart. 2019;105(24):1860–7. 10.1136/heartjnl-2018-314267. Epub 2019 Aug 23. PMID: 31444267.
Bosch NA, Cimini J, Walkey AJ. Atrial Fibrillation in the ICU. Chest. 2018;154(6):1424–34. 10.1016/j.chest.2018.03.040. Epub 2018 Apr 6. PMID: 29627355; PMCID: PMC6335260.
Kotecha D, Lam CS, Van Veldhuisen DJ, Van Gelder IC, Voors AA, Rienstra M. Heart Failure With Preserved Ejection Fraction and Atrial Fibrillation: Vicious Twins. J Am Coll Cardiol. 2016;68(20):2217–2228. doi: 10.1016/j.jacc.2016.08.048. PMID: 27855811.
Gorenek B, Halvorsen S, Kudaiberdieva G, Bueno H, Van Gelder IC, Lettino M, Marin F, Masip J, Mueller C, Okutucu S, Poess J, Potpara TS, Price S, Lip GYH. Atrial fibrillation in acute heart failure: A position statement from the Acute Cardiovascular Care Association and European Heart Rhythm Association of the European Society of Cardiology. Eur Heart J Acute Cardiovasc Care. 2020;9(4):348–57. Epub 2020 Jan 24. PMID: 31976747.
DiMarco JP. Atrial fibrillation and acute decompensated heart failure. Circ Heart Fail. 2009;2(1):72 – 3. 10.1161/CIRCHEARTFAILURE.108.830349. PMID: 19808318.
Diaz J, Martinez F, Calderon JM, Fernandez A, Sauri I, Uso R, Trillo JL, Redon J, Forner MJ. Incidence and impact of atrial fibrillation in heart failure patients: real-world data in a large community. ESC Heart Fail. 2022;9(6):4230–9. Epub 2022 Sep 16. PMID: 36111519; PMCID: PMC9773729.
Zhang L, Tang L, Chen S, Chen C, Peng B. A nomogram for predicting the 4-year risk of chronic kidney disease among Chinese elderly adults. Int Urol Nephrol. 2023;55(6):1609–17. 10.1007/s11255-023-03470-y. Epub 2023 Jan 31. PMID: 36720744.
Johnson AEW, Bulgarelli L, Shen L, Gayles A, Shammout A, Horng S, Pollard TJ, Hao S, Moody B, Gow B, Lehman LH, Celi LA, Mark RG. MIMIC-IV, a freely accessible electronic health record dataset. Sci Data. 2023;10(1):1. 10.1038/s41597-022-01899-x. Erratum in: Sci Data. 2023 Jan 16;10(1):31. Erratum in: Sci Data. 2023 Apr 18;10(1):219. PMID: 36596836; PMCID: PMC9810617.
Liu T, Zhao Q, Du B. Effects of high-flow oxygen therapy on patients with hypoxemia after extubation and predictors of reintubation: a retrospective study based on the MIMIC-IV database. BMC Pulm Med. 2021;21(1):160. 10.1186/s12890-021-01526-2. Erratum in: BMC Pulm Med. 2021 Jun 9;21(1):194. PMID: 33985472; PMCID: PMC8118109.
Wu WT, Li YJ, Feng AZ, Li L, Huang T, Xu AD, Lyu J. Data mining in clinical big data: the frequently used databases, steps, and methodological models. Mil Med Res. 2021;8(1):44. 10.1186/s40779-021-00338-z. PMID: 34380547; PMCID: PMC8356424.
Stothers JAM, Nguyen A. Can Neo4j Replace PostgreSQL in Healthcare? AMIA Jt Summits Transl Sci Proc. 2020;2020:646–653. PMID: 32477687; PMCID: PMC7233060.
Chan BKC, Data Analysis Using R, Programming. Adv Exp Med Biol. 2018;1082:47–122. 10.1007/978-3-319-93791-5_2. PMID: 30357717.
Tibshirani R, Journal of the Royal Statistical Society. Regression Shrinkage and Selection Via the Lasso, : Series B (Methodological), Volume 58, Issue 1, January 1996, Pages 267–288, https://doi.org/10.1111/j.2517-6161.1996.tb02080.x.
Mandrekar JN. Receiver operating characteristic curve in diagnostic test assessment. J Thorac Oncol. 2010;5(9):1315-6. 10.1097/JTO.0b013e3181ec173d. PMID: 20736804.
Van Calster B, Wynants L, Verbeek JFM, Verbakel JY, Christodoulou E, Vickers AJ, Roobol MJ, Steyerberg EW. Reporting and Interpreting Decision Curve Analysis: A Guide for Investigators. Eur Urol. 2018;74(6):796–804. 10.1016/j.eururo.2018.08.038. Epub 2018 Sep 19. PMID: 30241973; PMCID: PMC6261531.
Janssens ACJW, Martens FK. Reflection on modern methods: Revisiting the area under the ROC Curve. Int J Epidemiol. 2020;49(4):1397–1403. 10.1093/ije/dyz274. PMID: 31967640.
Collins GS, Reitsma JB, Altman DG, Moons KG. Transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD): The TRIPOD statement.
Hu CY, Wang CY, Li JY, Ma J, Li ZQ. Relationship between atrial fibrillation and heart failure. Eur Rev Med Pharmacol Sci. 2016;20(21):4593–600. PMID: 27874936.
Rosano G, Savarese G. Inter-twinned relationship between heart failure and atrial fibrillation. Heart. 2020;106(15):1125–1126. 10.1136/heartjnl-2020-316886. Epub 2020 Jun 1. PMID: 32482671.
Bessissow A, Khan J, Devereaux PJ, Alvarez-Garcia J, Alonso-Coello P. Postoperative atrial fibrillation in non-cardiac and cardiac surgery: an overview. J Thromb Haemost. 2015;13 Suppl 1:S304-12. doi: 10.1111/jth.12974. Erratum in: J Thromb Haemost. 2015;13(11):2134. PMID: 26149040.
Masip J, Peacock WF, Price S, Cullen L, Martin-Sanchez FJ, Seferovic P, Maisel AS, Miro O, Filippatos G, Vrints C, Christ M, Cowie M, Platz E, McMurray J, DiSomma S, Zeymer U, Bueno H, Gale CP, Lettino M, Tavares M, Ruschitzka F, Mebazaa A, Harjola VP, Mueller C, Acute Heart Failure Study Group of the Acute Cardiovascular Care. Association and the Committee on Acute Heart Failure of the Heart Failure Association of the European Society of Cardiology. Indications and practical approach to non-invasive ventilation in acute heart failure. Eur Heart J. 2018;39(1):17–25. 10.1093/eurheartj/ehx580. PMID: 29186485; PMCID: PMC6251669.
Siniorakis E, Arvanitakis S, Tsitsimpikou C, Tsarouhas K, Tzevelekos P, Panta S, Aivalioti F, Zampelis C, Triposkiadis F, Limberi S. Acute Heart Failure in the Emergency Department: Respiratory Rate as a Risk Predictor. In Vivo. 2018 Jul-Aug;32(4):921–925. 10.21873/invivo.11330. PMID: 29936481; PMCID: PMC6117786.
Miller WL. Fluid Volume Overload and Congestion in Heart Failure: Time to Reconsider Pathophysiology and How Volume Is Assessed. Circ Heart Fail. 2016;9(8):e002922. 10.1161/CIRCHEARTFAILURE.115.002922. PMID: 27436837.
Kleber M, Kozhuharov N, Sabti Z, Glatz B, Isenreich R, Wussler D, Nowak A, Twerenbold R, Badertscher P, Puelacher C, du, Fay de Lavallaz J, Nestelberger T, Boeddinghaus J, Wildi K, Flores D, Walter J, Rentsch K, von Eckardstein A, Goudev A, Breidthardt T, Mueller C. Relative hypochromia and mortality in acute heart failure. Int J Cardiol. 2019;286:104–110. doi: 10.1016/j.ijcard.2019.02.060. Epub 2019 Feb 27. PMID: 30853296.
Cooper LB, Mentz RJ, Gallup D, Lala A, DeVore AD, Vader JM, AbouEzzeddine OF, Bart BA, Anstrom KJ, Hernandez AF, Felker GM. Serum Bicarbonate in Acute Heart Failure: Relationship to Treatment Strategies and Clinical Outcomes. J Card Fail. 2016;22(9):738–42. 10.1016/j.cardfail.2016.01.007. Epub 2016 Jan 14. PMID: 26777758; PMCID: PMC4945480.
Zandijk AJL, van Norel MR, Julius FEC, Sepehrvand N, Pannu N, McAlister FA, Voors AA, Ezekowitz JA. Chloride in Heart Failure: The Neglected Electrolyte. JACC Heart Fail. 2021;9(12):904–15. Epub 2021 Oct 6. PMID: 34857174.
Siri-Tarino PW, Sun Q, Hu FB, Krauss RM. Saturated fat, carbohydrate, and cardiovascular disease. Am J Clin Nutr. 2010;91(3):502–9. 10.3945/ajcn.2008.26285. Epub 2010 Jan 20. PMID: 20089734; PMCID: PMC2824150.

No competing interests reported.

Download PDF

Version 1

posted

You are reading this latest preprint version

Nomograms for Predicting the Risk of Acute Heart Failure in ICU Patients with Atrial Fibrillation

Status:

Version 1

Abstract

Background

Methods

Results

Conclusion

Figures

Introduction

Methods

Patient and public involvement

Objectives

Database

Study populations

Data collection and measurement

Statistical analysis

TRIPOD reporting guidelines

Results

Clinical characteristics

Variable selection

Nomograms construction

Assessment and validation

Discussion

Conclusion

Abbreviations

Declarations

References

Additional Declarations

Supplementary Files

Status:

Version 1