Prognostic implication of lung ultrasound in heart failure: a pooled analysis of international cohorts

Abstract

Aims

Lung ultrasound (LUS) is often used to assess congestion in heart failure (HF). In this study, we assessed the prognostic role of LUS in patients with HF at admission and hospital discharge, and in an outpatient setting, and explored whether clinical factors [age, sex, left ventricular ejection fraction (LVEF), and atrial fibrillation] impact the prognostic value of LUS findings. Further, we assessed the incremental prognostic value of LUS on top of the following two clinical risk scores: (i) the atrial fibrillation, haemoglobin, elderly, abnormal renal parameters, diabetes mellitus (AHEAD) and (ii) the Meta-Analysis Global Group in Chronic Heart Failure (MAGGIC) clinical risk scores.

Methods and results

We pooled data on patients hospitalized for HF or followed up in outpatient clinics from international cohorts. We enrolled 1947 patients at admission (n = 578), discharge (n = 389), and in outpatient clinics (n = 980). The total LUS B-line count was calculated for the eight-zone scanning protocol. The primary outcome was a composite of rehospitalization for HF and all-cause death. Compared with those in the lower tertiles of B lines, patients in the highest tertiles were older, more likely to have signs of HF and had higher N-terminal pro b-type natriuretic peptide (NT-proBNP) levels. A higher number of B lines was associated with increased risk of primary outcome at discharge [Tertile 3 vs. Tertile 1: adjusted hazard ratio (HR): 5.74 (3.26–10.12), P < 0.0001] and in outpatients [Tertile 3 vs. Tertile 1: adjusted HR: 2.66 (1.08–6.54), P = 0.033]. Age and LVEF did not influence the prognostic capacity of LUS in different clinical settings. Adding B-line count to the MAGGIC and AHEAD scores improved net reclassification significantly in all three clinical settings.

Conclusion

A higher number of B lines in patients with HF was associated with an increased risk of morbidity and mortality, regardless of the clinical setting.

Graphical Abstract

Open in new tabDownload slide

Introduction

Congestion is a key pathophysiological feature of heart failure (HF), and one of the main drivers of symptoms, signs, adverse remodelling, and, eventually, hospitalization and premature death.^1–6 An accurate assessment of congestion can provide valuable information on the risk of adverse outcomes and guide HF management.

Assessment of B lines on lung ultrasound (LUS) emerged as a semi-quantitative, quick, and reliable imaging tool to facilitate the diagnosis and quantification of pulmonary congestion in patients with acute HF (AHF) or chronic HF⁷ and improve risk stratification, as recently highlighted by a European Association of Cardiovascular Imaging consensus (EACVI) document.⁸ Serial LUS can also be used for bedside adjustment of diuretic doses in HF. Another advantage of LUS is that it is useful in diagnosing HF correctly in primary care settings where natriuretic peptides may not be immediately available.⁹ Of the different LUS protocols proposed and used (4, 8, and 28 zones), the 8-zone protocol combines good accuracy and reproducibility with a shorter time for image acquisition, and it might, therefore, be preferred to others.¹⁰ However, which clinical factors, such as age, sex, or left ventricular ejection fraction (LVEF), may affect the prognostic interpretation of LUS in patients with HF has not been well studied. Similarly, it is not known whether LUS clearly improves risk stratification, mostly because of the small sample size of studies conducted so far.^11,12 Moreover, previous studies were focused either on acute settings or outpatient settings to assess the prognostic capacity of B lines. However, in this study, we take an integrative approach to assessing the prognostic capacity of B lines in different clinical settings in a pooled cohort of >1900 patients with HF.

Hence, in this research collaboration, we aimed to assess the prognostic relevance of B lines in a broad spectrum of patients with HF in different clinical settings (at admission, discharge, and outpatient clinics) and according to different clinical phenotypes [age, sex, LVEF, and presence/absence of atrial fibrillation (AF)]. Further, we assessed the incremental prognostic value of LUS on top of key biological markers [natriuretic peptides and estimated glomerular filtration rate (eGFR)] and two validated clinical risk scores in patients with HF: (i) atrial fibrillation, haemoglobin, elderly, abnormal renal parameters, diabetes mellitus (AHEAD) and (ii) Meta-Analysis Global Group in Chronic Heart Failure (MAGGIC).

Methods

Patient population

This retrospective study is a collaboration between eight cardiology centres, wherein we pooled the patient-level data from eight cohort studies (see Supplementary data online, Table S1):

• The Nancy cohort study (admission): included patients hospitalized for AHF at ‘Département de Cardiologie Medicale’, Nancy, France, from 3 April 2014 to 4 August 2015.¹³ LUS was performed once, within 3 days of admission.

• The AHF-core study (admission and discharge): an ongoing cohort study (NCT03327532) which includes patients hospitalized for AHF in the two university hospitals of Grand-Est region of France. LUS was conducted within 72 h of admission and again before discharge.

• The Pisa cohort study (admission and discharge): included patients hospitalized for AHF in the Cardiology Department in Pisa from October 2004 to October 2008.^3,14 LUS was conducted within 24 h of admission and before discharge.

• The Siena cohort study (admission and discharge): included patients with new onset or worsening HF in sinus rhythm admitted to the Cardiovascular Diseases Unit in Siena.¹⁵ LUS was performed within 12 h of admission and before discharge.

• The Perugia cohort study (discharge): included patients hospitalized for AHF at the Division of Cardiologia e Fisiopatologia Cardiovascolare in Perugia from April 2014 to October 2014.¹⁶ LUS was conducted at discharge.

• Nis cohort (outpatients): the outpatient cohort enrolled patients with HF with preserved ejection fraction (HFpEF) referred to Cardiology Clinic of the Niska Banja Institute, Nis University, Serbia from 1 June 2016 to 1 February 2018.¹⁷

• The Hull cohort study (outpatient): included patients with HF attending a specialist clinic for a routine follow-up visit from April 2016 to March 2017.¹⁸

• The Barcelona cohort study (outpatient): included patients with HF attending routine clinical appointments during the period from 6 July 2016 to 31 July 2017.¹⁹

The main exclusion criteria in the cohorts were: (i) inability to provide informed consent; (ii) important lung disease potentially affecting LUS findings, such as pulmonary fibrosis, previous pneumectomy or lobectomy, pulmonary cancer, or metastases; and (iii) other severe diseases hampering image acquisition.

Each cohort study was conducted in accordance with the Declaration of Helsinki and was approved by the respective Ethics Committees. Validation of the methodology has been performed and published previously.¹³

LUS and B-line count methods

LUS was performed by experienced cardiologists and resident doctors at each site using a 28-zone scanning protocol in all cohorts except in the Siena and Barcelona cohort studies,^15,19 where the 8-zone method was used, as described below (see Supplementary data online, Figure S1). Subsequently, we used the following method to calculate the sum of B lines for 8 zones from the 28-zone LUS results:

B-line count method (see Supplementary data online, Figure S1): we grouped the 28 zones of LUS⁴ in 8 zones²⁰ as follows:

• Para-sternal and mid-clavicular zones in the second and third intercostal spaces on the right side (four zones grouped into one zone).

• Para-sternal and mid-clavicular zones in the second and third intercostal spaces on the left side (four zones grouped into one zone).

• Anterior and mid-axillary zones in the second and third intercostal spaces on the right side (four zones grouped into one zone).

• Anterior and mid-axillary zones in the second and third intercostal spaces on the left side (four zones grouped into one zone).

• Para-sternal and mid-clavicular zones in the fourth and fifth intercostal spaces on the right side (four zones grouped into one zone).

• Anterior and mid-axillary zones in the fourth and fifth intercostal spaces on the right side (four zones grouped into one zone).

• Para-sternal and mid-clavicular zones in the fourth intercostal space on the left side (two zones grouped into one zone).

• Anterior and mid-axillary zones in the fourth intercostal space on the left side (two zones grouped into one zone).

Thereafter, we summed the highest number of B lines seen in one of the four or two zones of each group to obtain their total count.

Study outcome

The primary outcome of this study was a composite of rehospitalization for HF and all-cause death, and the secondary outcomes were components of the primary outcome. All patients were followed up in each institution, and patients and their relatives were contacted regularly by email or phone to collect clinical events of interest.

Statistical analysis

We combined the patient-level data from the above-mentioned cohorts to constitute the admission cohort (AHF core, Nancy, Siena, and Pisa cohorts), the discharge cohort (AHF core, Pisa, Siena, and Perugia cohorts), and the outpatient cohort (Nis, Hull, and Barcelona cohorts). Categorical variables are presented as percentages, and continuous variables are presented as mean ± standard deviation for normally distributed data and median, 25th and 75th percentiles for skewed data. For each cohort of admission, discharge, and outpatients, patients were divided into three groups, according to the tertiles of B lines.

Time-to-event comparisons were analysed using log-rank test and Cox proportional hazards models. Survival probabilities were estimated using the Kaplan–Meier method and plotted as survival curves with cohort-specific tertiles of B-line counts. Cox proportional hazards models were then used to obtain crude and adjusted hazard ratios (HRs). The Cox models were adjusted for age and sex in the M1 model, for age, sex, eGFR, and systolic blood pressure in the M2 model, and for adjustment variables of the M2 model and natriuretic peptides [Z score of brain natriuretic peptide (BNP) and NT-proBNP] in the M3 model. Further, Cox models were adjusted for cohort as a factor because of differences between the cohorts. Further, we tested the association of outcome with B lines while adjusting for E/e′ on top of M3 model. To illustrate the association of tertiles of B lines with outcomes according to age, sex, LVEF, and presence/absence of AF, forest plots were created for admission, discharge, and outpatient cohorts. In addition, we assessed the association of tertiles of B lines in the upper and lower LUS zones with the primary outcome.

The incremental prognostic value of B-line counts in addition to risk models (clinical/biological model, MAGGIC score, and AHEAD score) was assessed by calculating continuous net reclassification improvement after adding total B-line counts to clinical/biological model, MAGGIC score, and AHEAD score. The clinical/biological model included age, sex, eGFR, systolic blood pressure, and natriuretic peptides (Z score of BNP and NT-proBNP). The MAGGIC score is based on 13 variables and predicts morbidity²¹ and mortality in patients with HF.^21,22 The AHEAD score is a simple risk stratification tool for patients with HF based on five risk factors which are AF, haemoglobin <130 g/L for men and 120 g/L for women, elderly (age > 70 years), abnormal renal function (creatinine > 130 μmol/L), and diabetes mellitus.²³

In addition, we conducted an intra-class correlation analysis to assess the inter-observer variability of the cardiologists involved in B line quantification in each centres (n = 7).²⁴ The resulting intra-class correlation of the assessors of the B lines in this study was 0.80 (95% confidence interval 0.65–0.91), supporting a good reproducibility of the B lines assessment across centres.

The two-tailed significance level was set at P < 0.05. All analyses were performed using SAS version 9.4.6 (SAS Institute Inc., Cary, NC, USA) and R version 3.6.1 (05-07-2019).

Results

Baseline characteristics

In the admission (n = 578) and discharge cohort (n = 389), the median age was 78 years, 41% of the participants were women, and ∼45% participants had LVEF < 40% (Table 1). The number of B lines (maximum number possible = 80) in admission cohort was ≤15 in Tertile 1, 16–29 in Tertile 2, and >29 in Tertile 3; at discharge, B lines were ≤11 in Tertile 1, 12–22 in Tertile 2, and >22 in Tertile 3. In the outpatient cohort (n = 980), the median age was 72 years, 31% of the participants were women, and around 31% of the participants had LVEF < 40%. The number of B lines was ≤1 in Tertile 1, 2–6 in Tertile 2, and >6 in Tertile 3.

Overall, patients in Tertiles 2 and 3 were older and more likely to have clinical signs of congestion and AF than those in Tertile 1; they also had a higher median BNP/NT-proBNP, but LVEF was similar. Both median AHEAD and MAGGIC scores were higher in patients with a higher number of B lines in discharge and outpatient cohorts (all P-values <0.05), whereas AHEAD score was also higher in Tertiles 2 and 3 compared with Tertile 1 in admission cohort.

Prognostic capacity of B lines at admission, discharge, and outpatient setting

Over a median follow-up of 180 (90–589), 126 (75–180), and 829 (527–968) days, the primary outcome occurred in 39% (n = 226), 48% (n = 188), and 30% (n = 296) of patients in admission, discharge, and outpatient cohorts, respectively. A higher B-line count was significantly associated with a higher frequency of primary outcome in all the cohorts (all P-values <0.001; Figure 1).

In the discharge cohort, compared with Tertile 1, Tertiles 2 and 3 were significantly associated with increased risk in the fully adjusted model [HR Tertile 2 vs. Tertile 1: 1.80 (1.01–3.22), P = 0.046 and Tertile 3 vs. Tertile 1: 5.81 (3.30–10.25), P < 0.0001]. When further adjusting on E/e′, the association of a higher number of B-line counts with the primary outcome remained significant in the discharge cohort [Tertile 1 vs. Tertiles 2 and 3 combined, HR: 3.26 (1.24–8.54), P = 0.016].

In the outpatient cohort, Tertile 3 was significantly associated with increased risk in the adjusted clinical model [HR: 3.42 (1.22–9.59)]; however, when the model was further adjusted with natriuretic peptides, no association was observed between number of B lines and outcome.

Further, a higher B-line count was not associated with an increased risk of hospitalization for HF or death at admission and in the outpatient cohort in fully adjusted model, whereas a higher B-line count at discharge (Tertile 3 vs. Tertile 1) was associated with a significantly increased risk of both hospitalization for HF and death (see Supplementary data online, Table S2).

In the admission and discharge cohorts, a higher number of B lines in the upper zone (Tertile 3 vs. Tertile 1) was significantly associated with increased risk of outcomes, whereas a higher number of B lines in the lower zone was less predictive. In contrast, a higher number of B lines in both the upper and lower zones was significantly associated with increased risk in the outpatient cohort, with similar HR: 2.40 (1.12–5.13), P = 0.024 and 2.55 (1.03–6.30), P = 0.043, respectively (see Supplementary data online, Table S3).

Association of B lines with primary outcome according to clinical variables

In the admission cohort, the association of B lines with the primary outcome was affected by the underlying heart rhythm (P-interaction = 0.0016; Figure 2). In the discharge cohort, the association of B lines with primary outcome was influenced by sex (P-interaction = 0.007), while in the outpatient cohort, the prognostic capacity of B lines was not influenced by any clinical variables (all P-interaction > 0.05).

Incremental prognostic value of risk scores by adding number of B-line counts

In the admission and discharge cohorts, the addition of B lines to the clinical, MAGGIC, or AHEAD risk score, significantly improved net reclassification for the primary outcome (all P-values ≤0.01; Figure 3). The net reclassification improvement was >30% in the discharge cohort.

In outpatient cohort, the addition of B lines improved the net reclassification of the clinical model by 13.8% (P = 0.03) and the AHEAD score by 18.9% (P = 0.03); however, the net reclassification improvement was non-significant over MAGGIC score.

Discussion

In this pooled cohort study, we found that higher B-line counts were independently associated with the risk of hospitalization for HF and death in both inpatient and outpatient settings. Further, LUS predicted outcome irrespective of the clinical characteristics. Moreover, the addition of B lines to clinical, MAGGIC, and AHEAD scores improved risk stratification. AF influenced the prognostic capacity of B lines in the admission cohort but not in the discharge and outpatient cohorts.

Prognostic capacity of LUS in hospitalized patients and in outpatients

In a very heterogeneous group of patients with HF, a higher count of B lines predicts the mid-term risk of hospitalization or death in different settings. The association was, however, not significant when adjusting for natriuretic petides in outpatients, possibly because of the limited statistical power of this relatively low-risk subgroup. Similar results were reported by Platz et al.,²⁴ where pre-discharge higher tertiles of B lines were associated with higher risk at short- and mid-term outcomes in patients with HF. Moreover, post hoc analysis of the Evaluation Study of Congestive Heart Failure and Pulmonary Artery Catheterization Effectiveness (ESCAPE) trial demonstrated that persistent pulmonary congestion was associated with a worse prognostic outcome within 6 months of hospitalization for HF, and management of congestion may be more prognostically important than improving cardiac index in these patients.²⁵ However, no significant association between the number of B lines at admission and outcome was identified, despite a sizeable sample size (n = 578) and the high risk of events in this subgroup. Our results are in agreement with other previous reports that assessed pulmonary congestion using chest radiography on admission and found only weak and inconsistent associations with future risk.^26,27 By contrast, discharge congestion, referred to as ‘residual congestion’,¹⁶ is probably the key prognostic variable, capturing the ability to decongest patients during hospitalization, which probably parallels the ability to maintain decongestion thereafter. Moreover, a recent study shows that poor in-hospital improvement in congestion is associated with poor prognosis.²⁸ These findings indicate that a patient may still have good prognosis despite a higher level of congestion if the patients responds well to the given treatment. In addition, previous studies have reported conflicting results on the prediction capacity of B lines measured at admission.^3,14 In a study by Gargani et al.,³ >50 B lines at admission were not associated with increased risk in patients with AHF [adjusted HR: 4.87 (0.88–27.06), P = 0.07], whereas, in another study, a higher number of B lines (>30) was associated with adverse outcomes in the overall population.¹⁴ Taken together, this suggests that residual pulmonary congestion at the time of discharge, rather than congestion at admission, is one of the major factors associated with poor prognosis in HF. Our findings highlight the limited prognostic value of admission congestion rather than the limited value of LUS itself.

Taking advantage of the individual patient data, we found that the prognostic capacity of LUS to identify patients at a higher risk remains mostly unchanged in the subgroup analysis according to age, sex, LVEF, natriuretic peptide levels, and presence/absence of AF. These results suggest that the prognostic value of B lines is unaffected by clinical characteristics, something that is not necessarily true for other echocardiographic indices of congestion.^5,15,29 For instance, in patients with AF, E/A ratio and E/e′ cannot be reliably measured or used to predict future risk.¹⁵

B lines on top of clinical risk scores for predicting the risk of outcome

Another important finding of this study is that adding B-line count to clinical risk scores, such as MAGGIC and AHEAD, improved the net reclassification of patients at high risk regardless of the clinical setting. Our findings concur with the previously published study, which assessed the incremental capacity of B lines on top of 4 different HF risk scores in a cohort of 123 AHF patients at discharge. The study reported that the addition of B-line count significantly improved the predictive value of HF risk scores.³⁰ This improvement is likely related to the detection of subclinical congestion not identified on physical examination. In stable HF patients, up to 50% of patients have subclinical congestion that can go undetected on physical examination at the time of discharge or in an outpatient setting, unless ultrasound is used,^18,31 and that worsening subclinical congestion by ultrasound is associated with higher natriuretic peptides and an increased risk of cardiovascular events. Hence, LUS can improve the ability of AHEAD or MAGGIC score to identify patients at higher risk. In addition, LUS significantly improved net reclassification in a clinical/biological model that included powerful prognosticators, such as natriuretic peptides. This may be related to LUS specifically evaluating pulmonary congestion, which is intrinsically related to dyspnoea, whereas natriuretic peptides are possibly related to less-specific congestion patterns and influenced by renal function and the presence of AF.¹³

Clinical implications

Congestion assessment using LUS can provide important prognostic information in patients with HF irrespective of the phenotype and clinical setting.³² LUS can be implemented systematically along with clinical examination at the time of discharge and during follow-up visits to identify the patients at higher risk, especially those who do not have clinical signs of congestion.¹⁸ Nowadays, it is becoming easy and practical to conduct LUS in any setting with the availability of hand-held devices and transducers that can be easily connected to mobile devices, including smartphones. Moreover, natriuretic peptides assessment is not always available across all hospitals or to many primary care providers; and when it is, it will take time to obtain results.⁹ Findings from LUS can be interpreted in real time and serial examinations can quantify and monitor response to treatment, particularly diuretics.³³ Although results were disappointing for LUS-guided management in hospitalized patients,³⁴ LUS-guided management after discharge was associated with clinical benefits in several clinical trials in patients with HF.^35,36 Recent studies demonstrate that novice residents with limited or no previous experience in LUS can acquire proficiency in reliably quantifying B lines with minimal training.^37,38 Moreover, LUS can be easily performed by nurses and other healthcare providers, which reinforces the incentive to use LUS for assisting with management in this context.³⁹

Limitations

Our study has some limitations. First, although this is a patient-level pooled analysis of eight cohorts, the admission, discharge, and outpatient cohorts were moderate in size, limiting the generalizability of the results. Second, individual cohorts had some differences in available baseline data, methods of variable measurements, and patient selection. Third, experienced practitioners in each institution assessed LUS; however, there could have been inter-observer variability. Further, B-line counts could be different depending on cohort studies, possibly because of the different LUS protocols (patient position, transducer settings) of the cohort studies. Despite these differences, we found a strong association between B lines and outcomes in discharge and outpatient cohorts, adding to the generalizability of our results. LUS findings are pragmatically relevant in a broad array of clinical settings and centres. Fourth, in the Nancy cohort and AHF core, patients underwent LUS assessment within 3 days from their admission, whereas in Siena and Pisa, patients underwent LUS within 12 and 24 h from admission, respectively: it is possible that prior administration of diuretic therapy may have influenced B-line counts in these patients.

Conclusion

In this study, a higher number of B lines in patients with HF was associated with increased risk of morbidity and mortality and improved risk stratification, regardless of the clinical setting or patients’ profile.

Supplementary data

Supplementary data are available at European Heart Journal – Cardiovascular Imaging online.

Data availability

The data that supports the findings of this study are available upon reasonable request from the corresponding author.

References

Author notes