Skip to main content
Download PDF

Decrease of haemoconcentration reliably detects hydrostatic pulmonary oedema in dyspnoeic patients in the emergency department – a machine learning approach

International Journal of Emergency Medicine volume 17, Article number: 114 (2024) Cite this article

Abstract

Background

Haemoglobin variation (ΔHb) induced by fluid transfer through the intestitium has been proposed as a useful tool for detecting hydrostatic pulmonary oedema (HPO). However, its use in the emergency department (ED) setting still needs to be determined.

Methods

In this observational retrospective monocentric study, ED patients admitted for acute dyspnoea were enrolled. Hb values were recorded both at ED presentation (T0) and after 4 to 8 h (T1). ΔHb between T1 and T0 (ΔHbT1-T0) was calculated as absolute and relative value. Two investigators, unaware of Hb values, defined the cause of dyspnoea as HPO and non-HPO. ΔHbT1-T0 ability to detect HPO was evaluated. A machine learning approach was used to develop a predictive tool for HPO, by considering the ability of ΔHb as covariate, together with baseline patient characteristics.

Results

Seven-hundred-and-six dyspnoeic patients (203 HPO and 503 non-HPO) were enrolled over 19 months. Hb levels were significantly different between HPO and non-HPO patients both at T0 and T1 (p < 0.001). ΔHbT1-T0 were more pronounced in HPO than non-HPO patients, both as relative (-8.2 [-11.2 to -5.6] vs. 0.6 [-2.1 to 3.3] %) and absolute (-1.0 [-1.4 to -0.8] vs. 0.1 [-0.3 to 0.4] g/dL) values (p < 0.001). A relative ΔHbT1-T0 of -5% detected HPO with an area under the receiver operating characteristic curve (AUROC) of 0.901 [0.896–0.906]. Among the considered models, Gradient Boosting Machine showed excellent predictive ability in identifying HPO patients and was used to create a web-based application. ΔHbT1-T0 was confirmed as the most important covariate for HPO prediction.

Conclusions

ΔHbT1-T0 in patients admitted for acute dyspnoea reliably identifies HPO in the ED setting. The machine learning predictive tool may represent a performing and clinically handy tool for confirming HPO.

Background

Acute-onset dyspnoea is one of the leading symptoms for patients admission at the emergency department (ED) [1,2,3]. It is estimated to be responsible for up to five million ED visits each year in the United States [4] with 30-day mortality ranging from 8 to 13% [1, 5]. Thus, a rapid and precise identification of the underlying mechanisms is of utmost importance.

However, diagnosing the cause of dyspnoea and acute respiratory failure is often challenging in the setting of emergency. In particular, differential diagnosis between hydrostatic pulmonary oedema (HPO) and other causes of dyspnoea may be complicated, as symptoms and signs are non-specific and physical examination is often not decisive [2, 6, 7]. To discriminate them, emergency physicians need to integrate their clinical assessment with rapidly available investigations, such as chest radiograph, electrocardiogram, bedside lung and cardiac ultrasound and blood tests [3, 8,9,10,11]. Today, the diagnosis of HPO is mainly based on the dosage of the plasma level of B-type natriuretic peptide (BNP), which is costly and is affected by many factors such as inflammation, obesity and suffers from some false positives, especially in case of renal failure [12,13,14]. Moreover, there is a grey zone in which its plasmatic values are not informative [15].

Pathophysiologically, HPO is characterised by the filtration of protein-poor plasma through the pulmonary capillary barrier toward the interstitium [2, 16]. It has been demonstrated that this volume is large enough for inducing haemoconcentration, resulting in an increase of haematocrit, haemoglobin (Hb) or plasma protein concentration, reflecting the decrease in plasma volume and the volume of lung oedema [17, 18]. Such a haemoconcentration has been demonstrated to reliably diagnose weaning-induced pulmonary oedema, i.e. pulmonary oedema induced by switching patients from positive to negative pressure ventilation during a spontaneous breathing trial (SBT) in the intensive care unit (ICU) [19,20,21]. Nevertheless, the diagnostic accuracy of Hb changes in the ED setting for detecting HPO in a population of dyspnoeic patients still needs to be determined.

Methods

Patients

This retrospective, observational, monocentric study was conducted in the emergency department of a university hospital. Patients were included if (i) they were admitted to the ED for dyspnoea as leading problem, (ii) at least two Hb measurements were retrievable, one at ED admission before any treatment except oxygen was started, and one after 4 to 8 h. Patients were excluded if (i) no data regarding therapeutic management, radiological and biochemical investigations were retrievable and (ii) they were not subsequently admitted to an hospital ward. The study was approved by the Local Ethics Committee (Comitato Etico Territoriale Interaziendale AOU Maggiore della Carità di Novara, CE 93/20). Due to the retrospective nature of the study and the indications of the Ethic Committee, the need for an informed consent was waived.

Collected data and study design

At ED admission (T0), we collected patients’ demographic, past medical history with special attention to cardiac and respiratory comorbidities, and home therapy information. Data regarding physical examination and vital signs were also collected, including arterial blood pressure, heart rate, respiratory rate, peripheral oxygen saturation and body temperature.

Still at T0, we collected blood samples for cellular blood count, biochemical variables including C-reactive protein (CRP), and arterial blood gas analysis (ABL90 FLEX, Radiometer Medical ApS, Copenhagen, Denmark). The value of Hb was collected from the arterial blood gas sample.

Other variables collected at T0 included radiological investigations decided by clinicians in charge (chest X-ray, lung ultrasound or CT-scan) and, when available, ECG and transthoracic echocardiography performed by a cardiologist.

At 4 to 8 h from ED presentation (T1), we again collected vital signs and blood samples data, including Hb. The difference between Hb values between T1 and T0 (ΔHbT1-T0) was calculated, both as absolute ((HbT1-HbT0) and relative ((HbT1-HbT0)/HbT0*100) change. At T1, the treatment received between two timepoints, the destination of the patient and the retained diagnosis were also recorded. Finally, we collected the last value of Hb recorded before hospital discharge.

Data were retrieved from electronic medical record used in the ED (PsNet, Hi.Tech SpA, Firenze, Italy). All data were anonymised and uploaded on a Research Electronic Data Capture platform (REDCap consortium, Vanderbilt University, TN, USA), hosted at the Università del Piemonte Orientale.

Diagnosis of HPO

Two expert investigators (FG and LMC), unaware of changes in Hb values, defined a posteriori the cause of dyspnoea as HPO and non-HPO, by reviewing all patients’ medical records. They particularly considered the past medical history (e.g., acute onset of dyspnoea), typical clinical signs of HPO (e.g., bilateral rales, orthopnoea, low-extremity pitting oedema), typical findings on chest X-ray, lung ultrasound (B-lines) and echocardiography - when performed - and absence of an obvious alternative diagnosis (e.g., pulmonary embolism at CT-scan with no infiltrates). The response to treatment of the clinical, radiological and gas exchange abnormalities was also considered. For patients defined as non-HPO, the expert investigators made an alternative diagnosis. We planned to exclude patients with non-consensual diagnosis from analysis.

Statistical analysis

Sample size

In a previous study, ΔHb for diagnosing HPO provided an area under the receiver operating curve (AUROC) of 0.79 [19]. The sample size was determined for obtaining such an AUROC with a margin of error in the sample estimates of 0.026, as suggested in the literature [22]. A sample of 706 patients was necessary. This calculation was performed using R3.4.2 software [23].

Descriptive statistics

Data are reported as median [interquartile range] for continuous variables and percentages (absolute values) for qualitative variables. The Wilcoxon test was performed for comparing continuous variables and the Pearson’s chi-square test (or Fisher’s exact test where appropriate) for categorical ones. The diagnostic ability of both absolute and relative ΔHbT1−T0 for detecting HPO was evaluated by calculating the AUROC with the relative confidence intervals. A p-value of 0.05 was considered as the limit for significant effects. This statistical analysis was performed with MedCalc (Mariekerke, Belgium) and R3.4.2 software.

Machine learning models

A machine-learning approach was then considered to develop a predictive tool for HPO, by considering as a covariate the ability of ΔHbT1−T0, together with baseline patient characteristics, such as age and sex, the presence of coronary artery disease, arterial hypertension, chronic obstructive pulmonary disease, cardiac valvular diseases or chronic renal failure. The algorithms used to train the model were the Random Forest (RF), Recursive Partitioning and Regression Trees (RPART), Gradient Boosting Machine (GBM), Support Vector Machine (SVM), and a Generalized Linear Model via Elastic Net penalized maximum likelihood (GLMNET). The characteristics of each model, the evaluation of their performance as well as the impact of each predictor on the model’s performance are detailed in Supplementary Material 1.

Results

Patient characteristics

Between 1st June 2018 and 31st December 2019, 706 patients admitted to the ED with dyspnoea as leading problem were enrolled in the study. Among them, 203 were assigned to the HPO group and 503 to the non-HPO group. No differences were observed between the two groups in terms of gender distribution and age (Table 1). HPO patients showed a higher prevalence of cardiovascular diseases, such as arterial hypertension, dilated cardiopathy and coronary artery disease compared to the non-HPO ones. Also, for patients in whom it could be retrieved, left ventricular ejection fraction (LVEF) at T0 was lower in the HPO group (n = 113) compared to the non-HPO one (n = 211) (49[38–57]% vs. 55[40–60]%, respectively; p < 0.001). The remaining patient characteristics are detailed in Table 1.

Table 1 Patient characteristics
Full size table

Clinical, biological and radiological characteristics of HPO and non-HPO patients

HPO patients showed a significantly higher value of both systolic and diastolic arterial blood pressure at ED presentation compared to non-HPO patients (Table 2), whereas no differences were observed in terms of heart rate (p > 0.05). The severity of respiratory impairment was similar between the two groups, both in terms of respiratory rate and arterial oxygenation, according to the arterial oxygen partial pressure over inspired oxygen fraction ratio (PaO2/FiO2) level (Table 2). One-hundred twenty-two patients underwent echocardiography at the ED. LVEF was significantly lower in the HPO (n = 61) compared to the non-HPO patients (n = 61) (36 [30,31,32,33,34,35,36,37,38,39,40,41,42,43,44,45,46]% vs. 50 [40–57]%, respectively; p < 0.001).

Table 2 Haemodynamic, biological and ultrasound findings
Full size table

Laboratory findings and first ED therapy

Non-HPO and HPO patients showed similar levels of plasma creatinine, sodium, and potassium values at baseline (Supplementary Material 2). However, HPO patients showed lower CRP levels compared to the non-HPO ones (1.64 [0.74–5.44] vs. 3.40 [0.99–9.83]mg/dL, respectively; p < 0.001) and higher levels of BNP (2150 [1424–2675] vs. 273 [135–660]pg/mL, respectively; p < 0.001). Of note, BNP values were available for only 135 (58 HPO and 77 non-HPO) patients (Supplementary Material 2). As for the initial therapy in the ED, patients with HPO were more likely to receive diuretics, nitrates, and continuous positive airway pressure (CPAP) over the first hour compared to non-HPO patients. On the other hand, non-HPO patients were more frequently treated with steroids, bronchodilators and antibiotics (Supplementary Material 3). The final diagnoses are reported in Supplementary material 4.

Haemoglobin values and changes over time

Hb level was significantly higher in HPO than in non-HPO patients at T0 (Table 3). It significantly decreased from T0 to T1 in HPO patients, while it did not change significantly in non-HPO patients (Table 3). The time interval between T0 and T1 was similar among HPO and non-HPO patients (Table 3). ΔHbT1−T0 were more pronounced in HPO than non-HPO patients, both as relative (-8.2 [-11.2 to -5.6] vs. 0.6 [-2.1–3.3]%, respectively) and absolute (-1.0 [-1.4 to -0.8] vs. 0.1 [-0.3–0.4]g/dL, respectively) values (p < 0.001 for both) (Fig. 1; Table 3). The last value of Hb recorded before hospital discharge was not different between HPO and non-HPO patients (11.6 [10.3–13.1] vs. 11.4 [10.2–13.0]g/dL, respectively; p = 0.371).

Fig. 1
figure 1

Percentage haemoglobin changes in HPO and non-HPO patients

Full size image
Table 3 Haemoglobin changes between HPO and non-HPO patients
Full size table

Diagnostic ability of ΔHbT1-T0 to detect HPO

Hb value at T0 detected HPO with an AUROC of 0.591 [0.545–0.636], with a sensitivity of 57 [50–64]% and a sensibility of 58 [54–67]% (Supplementary material 5). A relative ΔHbT1−T0 of -5% could detect HPO with a sensitivity of 88 [82–92]% and a specificity of 87 [84–90]%, with an AUROC of 0.901 [0.896–0.906] (Fig. 2, Supplementary material 5). An absolute ΔHbT1−T0 of -0.6 g/dL detected HPO with a sensitivity of 85 [83–86]% and a specificity of 89 [88–90]%, with an AUROC of 0.903 [0.897–0.908] (Supplementary material 5). No significant differences emerged when relative and absolute ΔHbT1−T0 AUROC curves were compared (p = 0.765), whereas both curves were superior to the one of Hb at T0 (p < 0.001).

Fig. 2
figure 2

Area under the receiver operating characteristic curve for the ability of relative ΔHbT1-T0 to detect HPO. AUROC: are under the ROC curve; HPO: hydrostatic pulmonary oedema; ROC: receiver operating characteristics; ΔHbT1−T0: haemoglobin difference between T1 and baseline

Full size image

Predictive model

With the machine learning approach, the performance of the considered models was similar except for the RPART algorithm, which demonstrated a lower performance, especially in comparison with GBM, which emerged as the best performing algorithm (Supplementary Material 6). The GBM algorithm detected HPO with an AUROC of 0.911 [0.909–0.914]) (Supplementary Material 7). According to the variable importance plot, ΔHbT1−T0 was confirmed as the most important covariate for detecting HPO. The predictive tool has been made freely available in a Shiny-web application [24] (Supplementary Material 8).

Discussion

Our study shows that a decrease in blood Hb concentration ≥ 5% 4 to 8 h after ED admission compared to the measurement made at the first blood sample reliably detects HPO in patients with dyspnoea. Through a machine learning approach, we could integrate such an information to create a predictive model for improving the role of ΔHb in confirming HPO at 4 to 8 h.

Acute hydrostatic pulmonary oedema results from the increase in pulmonary capillary pressure promoting the filtration of hypo-oncotic fluid from the vessels to the interstitial and alveolar spaces of the lung parenchyma [16]. The most important symptom, which usually brings the patient to seek medical attention at the ED is dyspnoea and is often associated with clinical signs of venous congestion. As for the radiological clues, chest X-ray, CT-scan and lung ultrasounds provide specific findings that are suggestive for HPO [9, 25,26,27]. Nevertheless, none of these findings is sufficient to diagnose pulmonary oedema [6, 28], alone or combined, to create diagnostic scores [29,30,31], the overall accuracy is only intermediate and not tailored for the emergency department setting [32,33,34]. In theory, the direct demonstration of an increased pulmonary artery occlusion pressure would be the reference method for defining HPO as the cause of dyspnoea [19], but right heart catheterisation is most often impossible in the ED [35]. Bedside echocardiography may provide significant information regarding the ventricular function and surrogate estimates of LV filling pressure [36]. Nevertheless, it requires specific skills, it is time consuming and often difficult to perform in patients with respiratory distress [21].

BNP and NT-proBNP, cardiac neurohormones secreted by the ventricles and atria in response to pressure overload, have been shown to have good diagnostic accuracy for HPO. They became the most adopted methods for establishing HPO diagnosis in clinical practice [37,38,39]. Nevertheless, these variables suffer from a large grey zone of values in which their diagnostic ability is reduced [40, 41], and they are influenced by many factors such as age, sex, and renal impairment [42, 43]. In addition, their determination is costly.

Figueras and Weil [17] firstly showed that changes in haematocrit could be used as estimates of plasma volume changes. Subsequently, they observed a significant increase in total plasma protein concentration and plasma colloid osmotic pressure in patients with acute HPO, compared to the reference group [17]. After treatment of HPO, they observed a significant decrease in haematocrit and total protein concentration, as well as a significant increase in plasma volume, despite the fact that urine output had exceeded fluid intake. The authors attributed these findings to the fact that treatment and resolution of pulmonary oedema lead to a re-expansion of the intravascular volume due to reabsorption of hypo-oncotic fluid from the lungs [17]. Henning and Weil subsequently evaluated the changes in Hb, plasma colloid osmotic pressure and plasma volume in 14 patients treated for acute cardiogenic pulmonary oedema with phentolamine and furosemide [18]. They observed that 4 h after phentolamine administration, Hb levels and plasma colloid osmotic pressure significantly decreased compared to baseline, whereas the mean plasma volume increased. When furosemide was subsequently administered, Hb further decreased, as well as plasma colloid osmotic pressure. Interestingly, even though urinary output exceeded fluid intake, plasma volume further increased, confirming the plausible explanation that it was related to redistribution of the hypo-oncotic fluid that had migrated to the lungs [18].

These seminal findings have been used to identify HPO in different settings. Among them, the increase in plasma protein and haematocrit levels during spontaneous breathing trials in mechanically ventilated patients in the ICU has been shown to reliably identify “weaning-induced pulmonary oedema” (WIPO) as the reason for SBT failure [19,20,21, 44]. Also, in patients with chronic heart failure, it has been recently shown that changes in Hb are significantly correlated with changes in diastolic pulmonary artery pressure, making haemoconcentration a plausible marker for ambulatory evaluation of clinical stability of such patients [45]. Other studies investigated Hb changes over days specifically in patients admitted for heart failure [46, 47], but none in a general population of dyspnoeic patients at the ED.

Our study demonstrates the solidity of Hb changes in detecting HPO in dyspnoeic patients in the emergency department. At ED presentation, the Hb level were slightly but significantly higher in HPO compared to non-HPO patients. It decreased in HPO patients at T1, which is in line with the findings of Henning and Weil in HPO patients [18], while it remained unchanged in patients with no HPO. Even more, the relative and absolute ∆HbT1−T0 reliably detected HPO, with a large AUROC and high positive predictive values (Supplementary Material 5). This diagnostic accuracy was better than the Hb value measured at T0. No differences in Hb levels between HPO and non-HPO patients were observed before hospital discharge. Interestingly, the diagnostic threshold of Hb changes we found for HPO detection is similar to the one observed by Dres et al. for the diagnosis of WIPO [20].

As a clinical application, our results suggest that considering the changes in Hb should allow a reliable diagnosis of HPO. We measured Hb level changes between admission and H-4 to H-8, so that the diagnosis could only be retrospective. Even more, the decrease in haemoconcentration would likely not have been observed without an efficient treatment of HPO. Then, the method could not be used to make HPO diagnosis on admission and decide for a therapeutic strategy. However, confirming the diagnosis of HPO a posteriori may still have clinical benefit. Moreover, the machine learning approach allowed us to create a predictive model to evaluate how Hb changes influence the likelihood of having HPO after such a time interval, according to other significant elements that a priori could influence the final diagnosis, notably chronic health conditions. In this regard, the GBM model showed an excellent ability for HPO detection through ΔHbT1−T0, which emerged as the most significant covariate that could determine the patient’s status. Through the web-based application, such model could be used in clinical practice to confirm the diagnosis. Nevertheless, further studies should investigate whether a shorter-term assessment of the decrease in haemoconcentration would perform with a similar diagnostic accuracy e.g., after only one hour.

Limitations

This study has several limitations. First, this is a retrospective single-centre analysis, and our results should be confirmed in a prospective cohort, on a shorter interval for the second Hb measurement. Second, the diagnosis of HPO was established by two experts, based on several arguments. Indeed, no other solution could have been chosen as the reference, provided that the pulmonary artery catheter cannot be used in the context where we conducted the study. Third, BNP was not available for most of the patients, which may have added a substantial clue for the determination of HPO. Fourth, we did not specifically compare Hb levels variation among patients with an intravascular congestion phenotype and patients with a tissue congestion phenotype (both pulmonary and systemic). Finally, Hb was assessed through the blood gas analyser, and haematocrit and plasma protein concentration were not used to estimate haemoconcentration as done by other authors [19]. However, this is no obvious argument why these variables should indicate haemoconcentration better than Hb.

Conclusion

Decrease in Hb concentration over the first hours of treatment reliably identifies hydrostatic pulmonary oedema as the main cause of dyspnoea in patients admitted to the emergency department.

Data availability

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

Abbreviations

BNP:

B-type natriuretic peptide

CRP:

C-reactive protein

ECG:

Electrocardiogram

ED:

Emergency department

GBM:

Gradient boosting machine

GLMNET:

Generalized linear model via elastic net penalized maximum likelihood

Hb:

Haemoglobin

HPO:

Hydrostatic pulmonary oedema

ICU:

Intensive care unit

LVEF:

Left ventricular ejection fraction

RF:

Random forest

RPART:

Recursive partitioning and regression trees

SBT:

Spontaneous breathing trial

SVM:

Support vector machine

WIPO:

Weaning-induced pulmonary oedema

T0 :

ED admission

T1 :

4 to 8 h from ED presentation

ΔHbT1−T0 :

The difference between Hb values between T1 and T0

References

  1. Lindskou TA, Pilgaard L, Søvsø MB, Kløjgård TA, Larsen TM, Jensen FB, et al. Symptom, diagnosis and mortality among respiratory emergency medical service patients. PLoS ONE. 2019;14:e0213145.

    Article  CAS  PubMed  PubMed Central  Google Scholar 

  2. Ware LB, Matthay MA. Clinical practice. Acute pulmonary edema. N Engl J Med. 2005;353:2788–96.

    Article  CAS  PubMed  Google Scholar 

  3. Wang CS, FitzGerald JM, Schulzer M, Mak E, Ayas NT. Does this dyspneic patient in the Emergency Department have congestive heart failure? JAMA. 2005;294:1944–56.

    Article  CAS  PubMed  Google Scholar 

  4. Cairns C, Kang K, National Hospital Ambulatory Medical Care Survey. : 2019 Emergency Department Summary Tables [Internet]. National Center for Health Statistics (U.S.); 2022 Apr. https://stacks.cdc.gov/view/cdc/115748

  5. Sørensen SF, Ovesen SH, Lisby M, Mandau MH, Thomsen IK, Kirkegaard H. Predicting mortality and readmission based on chief complaint in emergency department patients: a cohort study. Trauma Surg Acute Care Open. 2021;6:e000604.

    Article  PubMed  PubMed Central  Google Scholar 

  6. Stevenson LW, Perloff JK. The limited reliability of physical signs for estimating hemodynamics in chronic heart failure. JAMA. 1989;261:884–8.

    Article  CAS  PubMed  Google Scholar 

  7. Murray JF. Pulmonary edema: pathophysiology and diagnosis. Int J Tuberc Lung Dis. 2011;15:155–60.

    CAS  PubMed  Google Scholar 

  8. Russell FM, Ehrman RR, Cosby K, Ansari A, Tseeng S, Christain E, et al. Diagnosing Acute Heart failure in patients with undifferentiated Dyspnea: a lung and Cardiac Ultrasound (LuCUS) Protocol. Acad Emerg Med. 2015;22:182–91.

    Article  PubMed  Google Scholar 

  9. Lichtenstein DA, Mezière GA. Relevance of lung ultrasound in the diagnosis of acute respiratory failure: the BLUE protocol. Chest. 2008;134:117–25.

    Article  PubMed  PubMed Central  Google Scholar 

  10. Chetran A, Costache AD, Ciongradi CI, Duca ST, Mitu O, Sorodoc V, et al. ECG and Biomarker Profile in patients with Acute Heart failure: a pilot study. Diagnostics (Basel). 2022;12:3037.

    Article  CAS  PubMed  Google Scholar 

  11. Szabó GV, Szigetváry C, Szabó L, Dembrovszky F, Rottler M, Ocskay K et al. Point-of-care ultrasound improves clinical outcomes in patients with acute onset dyspnea: a systematic review and meta-analysis. Intern Emerg Med. 2022.

  12. Maisel A. B-type natriuretic peptide measurements in diagnosing congestive heart failure in the dyspneic emergency department patient. Rev Cardiovasc Med. 2002;3(Suppl 4):S10–17.

    PubMed  Google Scholar 

  13. van Kimmenade RRJ, Januzzi JL, Bakker JA, Houben AJ, Rennenberg R, Kroon AA, et al. Renal clearance of B-Type natriuretic peptide and amino terminal Pro-B-Type natriuretic peptide: a mechanistic study in hypertensive subjects. J Am Coll Cardiol. 2009;53:884–90.

    Article  PubMed  Google Scholar 

  14. Levitt JE, Vinayak AG, Gehlbach BK, Pohlman A, Van Cleve W, Hall JB, et al. Diagnostic utility of B-type natriuretic peptide in critically ill patients with pulmonary edema: a prospective cohort study. Crit Care. 2008;12:R3.

    Article  PubMed  PubMed Central  Google Scholar 

  15. Logeart D, Saudubray C, Beyne P, Thabut G, Ennezat P-V, Chavelas C, et al. Comparative value of Doppler echocardiography and B-type natriuretic peptide assay in the etiologic diagnosis of acute dyspnea. J Am Coll Cardiol. 2002;40:1794–800.

    Article  PubMed  Google Scholar 

  16. Ingbar DH. Cardiogenic pulmonary edema: mechanisms and treatment - an intensivist’s view. Curr Opin Crit Care. 2019;25:371.

    Article  PubMed  Google Scholar 

  17. Figueras J, Weil MH. Blood volume prior to and following treatment of acute cardiogenic pulmonary edema. Circulation. 1978;57:349–55.

    Article  CAS  PubMed  Google Scholar 

  18. Henning RJ, Weil MH. Effect of afterload reduction on plasma volume during acute heart failure. Am J Cardiol. 1978;42:823–7.

    Article  CAS  PubMed  Google Scholar 

  19. Anguel N, Monnet X, Osman D, Castelain V, Richard C, Teboul J-L. Increase in plasma protein concentration for diagnosing weaning-induced pulmonary oedema. Intensive Care Med. 2008;34:1231–8.

    Article  PubMed  Google Scholar 

  20. Dres M, Teboul J-L, Anguel N, Guerin L, Richard C, Monnet X. Extravascular lung water, B-type natriuretic peptide, and blood volume contraction enable diagnosis of weaning-induced pulmonary edema. Crit Care Med. 2014;42:1882–9.

    Article  CAS  PubMed  Google Scholar 

  21. Ferré A, Guillot M, Lichtenstein D, Mezière G, Richard C, Teboul J-L et al. Lung ultrasound allows the diagnosis of weaning-induced pulmonary oedema. Intensive Care Med. 2019.

  22. Hajian-Tilaki K. Sample size estimation in diagnostic test studies of biomedical informatics. J Biomed Inf. 2014;48:193–204.

    Article  Google Scholar 

  23. R Core Team. R: A Language and Environment for Statistical Computing [Internet]. Vienna, Austria: R Foundation for Statistical Computing. 2015. https://www.R-project.org/

  24. Chang W, Cheng J, Allaire JJ, Xie Y, McPherson J. shiny: Web Application Framework for R [Internet]. 2020. https://CRAN.R-project.org/package=shiny

  25. Komiya K, Ishii H, Murakami J, Yamamoto H, Okada F, Satoh K, et al. Comparison of chest computed tomography features in the acute phase of cardiogenic pulmonary edema and acute respiratory distress syndrome on arrival at the emergency department. J Thorac Imaging. 2013;28:322–8.

    Article  PubMed  Google Scholar 

  26. Copetti R, Soldati G, Copetti P. Chest sonography: a useful tool to differentiate acute cardiogenic pulmonary edema from acute respiratory distress syndrome. Cardiovasc Ultrasound. 2008;6:16.

    Article  PubMed  PubMed Central  Google Scholar 

  27. Mahdyoon H, Klein R, Eyler W, Lakier JB, Chakko SC, Gheorghiade M. Radiographic pulmonary congestion in end-stage congestive heart failure. Am J Cardiol. 1989;63:625–7.

    Article  CAS  PubMed  Google Scholar 

  28. Collins SP, Lindsell CJ, Storrow AB, Abraham WT, ADHERE Scientific Advisory Committee, Investigators and Study Group. Prevalence of negative chest radiography results in the emergency department patient with decompensated heart failure. Ann Emerg Med. 2006;47:13–8.

    Article  PubMed  Google Scholar 

  29. McKee PA, Castelli WP, McNamara PM, Kannel WB. The natural history of congestive heart failure: the Framingham study. N Engl J Med. 1971;285:1441–6.

    Article  CAS  PubMed  Google Scholar 

  30. Marantz PR, Tobin JN, Wassertheil-Smoller S, Steingart RM, Wexler JP, Budner N, et al. The relationship between left ventricular systolic function and congestive heart failure diagnosed by clinical criteria. Circulation. 1988;77:607–12.

    Article  CAS  PubMed  Google Scholar 

  31. Schocken DD, Arrieta MI, Leaverton PE, Ross EA. Prevalence and mortality rate of congestive heart failure in the United States. J Am Coll Cardiol. 1992;20:301–6.

    Article  CAS  PubMed  Google Scholar 

  32. Remes J, Miettinen H, Reunanen A, Pyörälä K. Validity of clinical diagnosis of heart failure in primary health care. Eur Heart J. 1991;12:315–21.

    Article  CAS  PubMed  Google Scholar 

  33. Fonseca C, Oliveira AG, Mota T, Matias F, Morais H, Costa C, et al. Evaluation of the performance and concordance of clinical questionnaires for the diagnosis of heart failure in primary care. Eur J Heart Fail. 2004;6:813–20.

    Article  PubMed  Google Scholar 

  34. Collins S, Storrow AB, Kirk JD, Pang PS, Diercks DB, Gheorghiade M. Beyond pulmonary edema: diagnostic, risk stratification, and treatment challenges of acute heart failure management in the emergency department. Ann Emerg Med. 2008;51:45–57.

    Article  PubMed  Google Scholar 

  35. Laher AE, Watermeyer MJ, Buchanan SK, Dippenaar N, Simo NCT, Motara F, et al. A review of hemodynamic monitoring techniques, methods and devices for the emergency physician. Am J Emerg Med. 2017;35:1335–47.

    Article  PubMed  Google Scholar 

  36. Lamia B, Maizel J, Ochagavia A, Chemla D, Osman D, Richard C, et al. Echocardiographic diagnosis of pulmonary artery occlusion pressure elevation during weaning from mechanical ventilation. Crit Care Med. 2009;37:1696–701.

    Article  PubMed  Google Scholar 

  37. Maisel AS, Krishnaswamy P, Nowak RM, McCord J, Hollander JE, Duc P, et al. Rapid measurement of B-type natriuretic peptide in the emergency diagnosis of heart failure. N Engl J Med. 2002;347:161–7.

    Article  CAS  PubMed  Google Scholar 

  38. Mueller C, Scholer A, Laule-Kilian K, Martina B, Schindler C, Buser P, et al. Use of B-type natriuretic peptide in the evaluation and management of acute dyspnea. N Engl J Med. 2004;350:647–54.

    Article  CAS  PubMed  Google Scholar 

  39. van der Burg-de Graauw N, Cobbaert CM, Middelhoff CJFM, Bantje TA, van Guldener C. The additive value of N-terminal pro-B-type natriuretic peptide testing at the emergency department in patients with acute dyspnoea. Eur J Intern Med. 2009;20:301–6.

    Article  PubMed  Google Scholar 

  40. Schwam E. B-type natriuretic peptide for diagnosis of heart failure in emergency department patients: a critical appraisal. Acad Emerg Med. 2004;11:686–91.

    Article  PubMed  Google Scholar 

  41. McCullough PA, Nowak RM, McCord J, Hollander JE, Herrmann HC, Steg PG, et al. B-type natriuretic peptide and clinical judgment in emergency diagnosis of heart failure: analysis from Breathing not properly (BNP) multinational study. Circulation. 2002;106:416–22.

    Article  PubMed  Google Scholar 

  42. Chircop R, Jelinek GA. B-type natriuretic peptide in the diagnosis of heart failure in the emergency department. Emerg Med Australas. 2006;18:170–9.

    Article  PubMed  Google Scholar 

  43. Chenevier-Gobeaux C, Claessens Y-E, Voyer S, Desmoulins D, Ekindjian OGJ-C. Influence of renal function on N-terminal pro-brain natriuretic peptide (NT-proBNP) in patients admitted for dyspnoea in the Emergency Department: comparison with brain natriuretic peptide (BNP). Clin Chim Acta. 2005;361:167–75.

    Article  CAS  PubMed  Google Scholar 

  44. Liu J, Shen F, Teboul J-L, Anguel N, Beurton A, Bezaz N, et al. Cardiac dysfunction induced by weaning from mechanical ventilation: incidence, risk factors, and effects of fluid removal. Crit Care. 2016;20:369.

    Article  PubMed  PubMed Central  Google Scholar 

  45. Almufleh A, Desai AS, Fay R, Ferreira JP, Buckley LF, Mehra MR, et al. Correlation of laboratory haemoconcentration measures with filling pressures obtained via pulmonary arterial pressure sensors in ambulatory heart failure patients. Eur J Heart Fail. 2020;22:1907–11.

    Article  CAS  PubMed  Google Scholar 

  46. Konishi M, Matsuzawa Y, Suzuki H, Akiyama E, Iwahashi N, Maejima N, et al. Higher level at admission and subsequent decline in hemoglobin in patients with acute pulmonary edema. Circ J. 2014;78:896–902.

    Article  PubMed  Google Scholar 

  47. Fujita T, Inomata T, Yazaki M, Iida Y, Kaida T, Ikeda Y, et al. Hemodilution after initial treatment in patients with Acute Decompensated Heart failure. Int Heart J. 2018;59:573–9.

    Article  CAS  PubMed  Google Scholar 

Download references

Acknowledgements

This study was presented at the 36th annual congress of the European Society of Intensive Care Medicine (21– 25 October 2023, Milan, Italy) and awarded as one of the ten best abstracts of the congress.

Funding

Not applicable.

Author information

Authors and Affiliations

  1. Department of Translational Medicine, Università degli Studi del Piemonte Orientale, Via Solaroli 17, Novara, 28100, Italy

    Francesco Gavelli, Luigi Mario Castello, Ilaria Nerici, Simona Priora, Valentina Giai Via, Matteo Bertoli, Claudia Foieni, Michela Beltrame, Mattia Bellan, Pier Paolo Sainaghi, Nello De Vita, Filippo Patrucco & Gian Carlo Avanzi

  2. Emergency Medicine Department, AOU Maggiore della Carità di Novara, C.so Mazzini 18, Novara, 28100, Italy

    Francesco Gavelli, Luigi Mario Castello, Ilaria Nerici, Simona Priora, Valentina Giai Via, Matteo Bertoli, Claudia Foieni, Michela Beltrame & Gian Carlo Avanzi

  3. Service de médecine intensive-réanimation, Hôpital de Bicêtre, Le Kremlin-Bicêtre, Hôpitaux universitaires Paris- Saclay, APHP, rue du Général Leclerc, Paris, France

    Francesco Gavelli, Xavier Monnet & Jean-Louis Teboul

  4. Department of Environmental and Preventive Science, University of Ferrara, Ferrara, Italy

    Danila Azzolina

Authors
  1. Francesco Gavelli
    View author publications

    You can also search for this author in PubMed Google Scholar

  2. Luigi Mario Castello
    View author publications

    You can also search for this author in PubMed Google Scholar

  3. Xavier Monnet
    View author publications

    You can also search for this author in PubMed Google Scholar

  4. Danila Azzolina
    View author publications

    You can also search for this author in PubMed Google Scholar

  5. Ilaria Nerici
    View author publications

    You can also search for this author in PubMed Google Scholar

  6. Simona Priora
    View author publications

    You can also search for this author in PubMed Google Scholar

  7. Valentina Giai Via
    View author publications

    You can also search for this author in PubMed Google Scholar

  8. Matteo Bertoli
    View author publications

    You can also search for this author in PubMed Google Scholar

  9. Claudia Foieni
    View author publications

    You can also search for this author in PubMed Google Scholar

  10. Michela Beltrame
    View author publications

    You can also search for this author in PubMed Google Scholar

  11. Mattia Bellan
    View author publications

    You can also search for this author in PubMed Google Scholar

  12. Pier Paolo Sainaghi
    View author publications

    You can also search for this author in PubMed Google Scholar

  13. Nello De Vita
    View author publications

    You can also search for this author in PubMed Google Scholar

  14. Filippo Patrucco
    View author publications

    You can also search for this author in PubMed Google Scholar

  15. Jean-Louis Teboul
    View author publications

    You can also search for this author in PubMed Google Scholar

  16. Gian Carlo Avanzi
    View author publications

    You can also search for this author in PubMed Google Scholar

Contributions

FG conceived and designed the study, performed data analysis and interpretation, and wrote the manuscript. LMC and DA performed data analysis and interpretation and contributed to writing the manuscript. XM, JLT and GCA participated in data analysis and interpretation and contributed to writing the manuscript. IN, SP, VGV, MaBer, CF, MiBel acquired the data and contributed to data analysis. MaBel, PPS, NDV, FP participated in data interpretation and contributed to writing the manuscript. All authors reviewed and approved the manuscript.

Corresponding author

Correspondence to Francesco Gavelli.

Ethics declarations

Ethics approval and consent to participate

The study was approved by the Local Ethic Committee (Comitato Etico Territoriale Interaziendale AOU Maggiore della Carità di Novara, CE 93/20). Due to the retrospective nature of the study, the Ethic Committee considered the need for an informed consent as not necessary. The authors certify that the study was performed in accordance with the ethical standards as laid down in the 1964 Declaration of Helsinki and its later amendments or comparable ethical standards.

Consent for publication

Not applicable.

Competing interests

Xavier MONNET and Jean-Louis TEBOUL are members of the Medical Advisory Board of Pulsion Medical Systems (Getinge). Xavier MONNET gave lectures for Baxter Healthcare and AOP. The other authors have no conflicts of interest to declare.

Additional information

Publisher’s note

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

Electronic supplementary material

Below is the link to the electronic supplementary material.

Supplementary Material 1

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/.

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Gavelli, F., Castello, L.M., Monnet, X. et al. Decrease of haemoconcentration reliably detects hydrostatic pulmonary oedema in dyspnoeic patients in the emergency department – a machine learning approach. Int J Emerg Med 17, 114 (2024). https://doi.org/10.1186/s12245-024-00698-y

Download citation

  • Received:

  • Accepted:

  • Published:

  • DOI: https://doi.org/10.1186/s12245-024-00698-y

Keywords

International Journal of Emergency Medicine

ISSN: 1865-1380

Contact us