Introduction

Heart failure (HF) is the leading cause of hospital admission in people aged 65 years or older in developed countries [1]. The chest X-ray (CXR) is one of the core investigations of breathless patients [2, 3]. In 1917, Danzer first identified cardiomegaly as a possible indicator of left ventricular dilation [4]. As hydrostatic pressure increases in the lungs, signs of pulmonary congestion start to appear—pulmonary venous congestion, interstitial oedema, alveolar oedema and pleural effusions—that are often used as entry criteria in trials of therapies for patients presenting with acute HF [5, 6].

The CXR is not a diagnostic test for HF. There is little relation between cardiothoracic ratio (CTR) and left ventricular systolic function [7]. Not all patients with acute HF have pulmonary congestion on their CXR [8]. However, amongst patients presenting to hospital with acute myocardial infarction, worsening features of HF on the CXR are related to increasing risk of death [9,10,11,12,13].

Fifty years ago, when rheumatic heart disease was common, acute pulmonary oedema was dramatic and life threatening [14]. Today, most patients admitted to hospital with HF no longer have new acute pulmonary oedema but instead present with decompensation of chronic underlying ventricular dysfunction, as a consequence of gradual but progressive increases in cardiac filling pressures [15, 16]. We, therefore, described the appearance of the initial CXR in a modern cohort of patients hospitalised for HF and assessed the relation between CXR appearance and outcome. We constructed a CXR scoring system and assessed the relation between the score and outcome in addition to clinical variables.

Methods

Study setting

The observational study to predict readmission for heart failure patients (OPERA-HF) is a prospective observational study which enrolled patients hospitalised for HF in the Hull University Hospitals NHS Trust, UK. Patients had to fulfil the following criteria to be included: age > 18 years; usual residence in the region served by the Hull University Hospitals NHS Trust; hospitalisation with HF; treatment with loop diuretics; and at least one of the following; left ventricular ejection fraction ≤ 40%; left atrial dimension ≥ 4.0 cm [17], or N-terminal pro-B-type natriuretic peptide (NT-proBNP) > 400 pg/mL (if in sinus rhythm) or > 1200 pg/mL (if in atrial fibrillation—AF) [18]. Patients who were unable to understand and comply with the protocol or unable or unwilling to give informed consent were not included in the study.

Chest radiography

The flow of patients through the study is shown in Fig. 1. The position (supine or erect) and projection of each film is written physically on the X-ray. In the present analysis, we included all patients who had an erect chest radiograph. The patient’s first radiograph on admission was reviewed. We excluded patients who had had only a supine film (as fluid in the lungs and pleura might have been distributed differently) or in whom the film was too rotated for interpretation. One investigator (Daniel Pan, DP) retrospectively reviewed all CXRs, grading the presence and severity of potential features of HF, blind to all other clinical data.

Fig. 1
figure 1

Flowchart illustrating the flow of patients through the study. CXR chest X-ray, CT computed tomography

Full size image

The approach to grading the CXR is illustrated in Fig. 2. The following were recorded for each patient:

  • The film projection—whether posterior–anterior (PA) or anterior–posterior (AP);

  • Alveolar oedema—graded as absent, present (oedema covering any of the lungs, but not all zones in both lungs) or severe (oedema covering all zones in both lungs).

  • Kerley B lines—graded as the presence or absence of prominent interstitial fissure lines in and between the lung lobes.

  • Pleural effusions—graded as present or not present. Presence is when there are both costophrenic and cardiophrenic blunting of one lung field.

  • Pulmonary venous congestion—graded as present or absent. Presence is when there is an increase in the proportion of vessels in the upper lung zones compared to the lower lung zones.

  • Heart size—measured as CTR.

Fig. 2
figure 2

Illustration demonstrating how chest X-ray features of heart failure was identified

Full size image

The CTR was calculated using the original method described by Danzer [4]. A single line is drawn dividing the cardiac silhouette vertically through the midline. Measurement ‘A’ is widest point from the right heart border to the line; measurement ‘B’ is the widest point from the left heart border to the line. Measurement ‘C’ is the widest measurement from the midline to the inner borders of the ribs. Thus, the CTR is defined as (A + B)/2C. An unmeasurable CTR was recorded where the cardiac borders were too obscure for accurate measurement. Although modern dictum suggests that accurate CTR interpretation is not possible in AP films, we considered a CTR of > 0.55 to be cardiomegaly regardless of projection, in keeping with the previous literature [19].

Signs related to other pathology as routinely reported by the radiologist were also recorded. To assess repeatability of findings, 25 CXRs were randomly selected and scored by a second investigator (Shirley Sze, SS).

Patients were followed until 1st of May 2017; the main outcome of interest was all-cause mortality.

The study had ethical approval from the South Yorkshire Research Ethics Committee (REC ref: 12/YH/0344) and is conducted in accordance with ICH-GCP, Declaration of Helsinki, the Data Protection Act 1998 and NHS Act 2006.

Statistical analysis

Continuous variables are expressed as median and interquartile range (IQR). Variables that were not normally distributed were log-transformed before analysis in further models. Categorical variables are expressed as numbers and percentages (%). Missing data were excluded from statistical analysis. Pearson’s Chi-squared test and Fisher’s exact row test were used to compare categorical variables between groups. Student’s t test and the Kruskal–Wallis test were used to compare continuous variables between the groups depending on the normality of the distribution.

Several Cox proportional hazard regression models were used to investigate the relation between variables and all-cause mortality. A first model was used to assign a weighting to the CXR variables based on their relation to prognosis. Only CXR variables were included in this model. Those significantly related to mortality on univariable analysis (a liberal threshold of P < 0.10 was used) were chosen for the multivariable model. The assigned points of the CXR score were generated by the beta coefficient (log hazard ratio) for each variable from the multivariable analysis, multiplied by ten and rounded to the nearest whole number [20].

A base model was constructed to assess the incremental value of the new CXR score in addition to other important clinical variables in predicting mortality. The base model included variables which had few missing data (< 1%) and which were also significantly related to mortality on univariable Cox regression analysis (again, a liberal threshold of P < 0.10 was used).

The individual CXR variables and the CXR score were then added to the base model, separately and in combination, to the find the best model to predict mortality. The new models’ cumulative discrimination was measured using Harrell’s C statistic; its statistical significance in comparison to the base model was assessed with the likelihood ratio test. A 2-sided P value ≤ 0.05 was considered statistically significant.

Venn diagrams are used to illustrate the relation between different features of pulmonary congestion. Kaplan–Meier curves, censored at 4 years, with the log-rank statistic were used to illustrate outcome.

Repeatability of CXR findings was tested using weighted Cohen’s Kappa statistic and Bland–Altman plots [21].

All analyses were performed using STATA software (version 14.2, StataCorp, Texas, United States) and Excel version 2016 (Microsoft, Redmond, United States).

Results

Participants

Between October 2012 and November 2016, 1145 patients were enrolled into OPERA-HF. After excluding 70 patients because they did not have a CXR during admission and another 100 patients because the CXRs were inadequate for interpretation, 975 remained (Fig. 1).

Table 1 shows the clinical and laboratory findings of patients recruited into the study on admission to hospital. Most patients had been hospitalised for HF in the previous year (60%). A quarter (26%) of patients did not have NT-proBNP measured on admission. Half of patients (54%) presented with breathlessness on exertion; 26% with breathlessness at rest and in the remainder, a symptom other than breathlessness was the reason for admission. 11% of patients presented with HF associated with an acute coronary syndrome.

Table 1 Clinical and laboratory findings of patients recruited into the study on admission to hospital
Full size table

Chest radiography appearances

Table 2 shows the CXR appearances of patients in the study. Most patients had AP CXRs (71%,). There was a low prevalence of features on the CXR unrelated to HF, including: consolidation (N = 91); pulmonary fibrosis (N = 16); hiatus hernia (N = 3); tumours/nodules (N = 7); pulmonary infarction (N = 5); pulmonary plaques (N = 13) and pneumonectomy (N = 1).

Table 2 Radiological findings of patients recruited into the study on admission, and in-hospital mortality, 30-day mortality and readmission to hospital within 30 days of discharge
Full size table

Cardiomegaly was present in 67% and un-measureable in 6%. The median CTR (excluding those in whom a measurement was not possible, N = 55) was 0.57 (IQR 0.53–0.61) in those with PA films and 0.60 (IQR 0.55–0.64) in those with AP films. Pulmonary venous congestion was present in 78%, and a degree of alveolar oedema was present in 64% (although only 10% had severe alveolar oedema). Only one film had a pleural effusion that covered most of one lung field.

Figure 3 illustrates the percentage of patients with different features of pulmonary congestion. A third (33%) of patients had all four features of pulmonary congestion (Fig. 3); only 5% had none and 10% had only one. Features of pulmonary congestion were generally more common with increasing CTR and on AP compared to PA films as shown by the bar charts illustrating the proportion of patients with each feature of pulmonary congestion, by cardiothoracic ratio and chest X-ray projection in Fig. 4.

Fig. 3
figure 3

Venn diagram illustrating percentage of patients with different features of pulmonary congestion

Full size image
Fig. 4
figure 4

Bar charts illustrating proportion of patients with each feature of pulmonary congestion, by cardiothoracic ratio and chest X-ray projection. Pearson’s Chi-squared test was used

Full size image

Generation of the CXR score, relation to clinical variables and relation to outcome

After a median follow-up of 671 days, 440 (45%) patients had died of which 41 died during the index admission. A fifth (19%) of the patients was readmitted with worsening HF within 30 days of discharge from their index hospitalisation.

Table 3 shows Cox regression analyses of chest X-ray variables and formation of the chest X-ray score; Fig. 5 shows Kaplan–Meier curves of all-cause mortality for chest X-ray projection, cardiothoracic ratio, pleural effusion, alveolar oedema. In univariable Cox regression analysis, each radiological abnormality, except pulmonary venous congestion, was associated with all-cause mortality. Consequently, a score was constructed using all CXR signs apart from pulmonary venous congestion.

Table 3 Univariable and multivariable Cox regression analyses of chest X-ray variables and formation of the chest X-ray score
Full size table
Fig. 5
figure 5

Kaplan–Meier curves of all-cause mortality for chest X-ray projection, cardiothoracic ratio, pleural effusion, alveolar oedema, Kerley B lines and the chest X-ray score. AP anterior–posterior, PA posterior–anterior, HF heart failure

Full size image

Table 4 shows clinical and laboratory findings of patients recruited into the study and in-hospital mortality, 30-day mortality and readmission to hospital within 30 days of discharge by increasing chest X-ray score. There was no relation between the CXR score and the presence (or absence) of left ventricular systolic dysfunction. Although associated with all-cause mortality at 4 years, the CXR score was neither associated with in-hospital mortality, nor mortality or readmission within 30 days of discharge. A worsening CXR score was associated with older age, higher serum concentrations of potassium and urea and higher plasma NT-proBNP (especially if the patient was in AF), lower values for diastolic blood pressure, haemoglobin, and albumin, being a woman, prior CABG and presenting with breathlessness as the main symptom.

Table 4 Clinical and laboratory findings of patients recruited into the study and in-hospital mortality, 30-day mortality and readmission to hospital within 30 days of discharge by increasing chest X-ray score
Full size table

CXR scores were lower in the 5% of patients (n = 49) whose first radiograph was taken more than 24 h after admission than in those who had their CXR earlier [median score 3 (IQR 3–7) v 7 (5–9), P < 0.001].

Relation of the CXR score to prognosis in addition to other variables

Tables 5 and 6 show univariable and multivariable Cox regression analyses of clinical and laboratory variables and the chest X-ray score On univariable Cox regression analysis, age, gender, hospitalisation for heart failure in the previous year, prior MI or CABG, malignancy, lower systolic and diastolic blood pressures, lower heart rate, lower sodium, haemoglobin, chloride and albumin and higher potassium, renal function, troponin, and NT-proBNP, and a higher CXR score was related to mortality. In a multivariable model, older age and higher serum concentrations of urea and sodium, lower systolic blood pressure, lower heart rate and lower serum concentrations of chloride and albumin were independent predictors of mortality. The CXR score was also independently associated with mortality (HR 1.10, 95% CI 1.07–1.13 per point increase, P < 0.001). However, CXR score did not improve the model’s c-statistic for predicting mortality at different time-points as shown in Table 7.

Table 5 Univariable Cox regression analyses of clinical and laboratory variables and the chest X-ray score
Full size table
Table 6 Multivariable Cox regression analysis
Full size table
Table 7 Reproducibility statistics (Cohen’s kappa for categorical variables and Bland Altman limits of agreement for continuous variables) got the chest X-ray variables and the chest X-ray score as well as the Harrell’s c-statistic and likelihood ratio tests of the multivariable Cox regression models
Full size table

Reproducibility

CXR variables varied in reliability, with alveolar oedema being the most reliable and CTR being the least, as shown in Table 7.

Discussion

There are three main findings from this study. First, radiological evidence of congestion is very common in patients presenting to hospital with HF. Second, patients presenting with breathlessness as their dominant symptom have a higher CXR score. Third, increasing pulmonary congestion on the CXR score is related to worsening HF as assessed by other clinical measures such as age, potassium, urea, NT-proBNP, haemoglobin and albumin, and is associated with increasing all-cause mortality, but not related to overall prognosis when these other variables are taken into account.

We found that the commonest abnormal feature on the chest X-ray was venous congestion, presumably representing a rise in left atrial pressure. As heart failure worsens, fluid can escape from pulmonary capillaries into the interstitial and pleural spaces resulting in Kerley B lines and pleural effusions, the next two most common features we found. Frank alveolar oedema was the least common finding, and represents the patient passing a ‘tipping point’ when the capacity of the lymphatics to remove fluid is exceeded and fluid starts to accumulate in the airspaces of the lungs [16].

Previous older studies have graded congestion in an incremental fashion based on the aforementioned sequence of events, with stage one being pulmonary venous congestion, stage two interstitial oedema and stage three alveolar oedema [11,12,13, 22]. We chose to grade all features separately and found that although many had pulmonary venous congestion and few had severe alveolar oedema, patients commonly have features of pulmonary congestion not in sequence. It may be that if pulmonary congestion happens very rapidly, the radiographic features might not occur in order.

Our study findings confirm that the CXR can identify pulmonary congestion as a potential cause of breathlessness in patients presenting to hospital. Since congestion on the CXR is still related to mortality in HF patients on univariable analysis, this should prompt clinicians to adequately diurese the patient when necessary.

Only a small proportion of patients (91/975, 9%) in our cohort had evidence of pneumonia on the CXR. Future studies should investigate whether the CXR can help to differentiate pulmonary congestion from other causes of breathlessness, and influence therapeutic decision making beyond HF, such as the initiation of antibiotics. The CXR remains a source of important clinical information and may be used within artificial intelligence algorithms, to help assist clinicians improve the diagnosis of AHF and guide management [23].

To our knowledge, this is the first study which has found a relation between CXR projection and all-cause mortality. If a patient is too unwell to stand, inhale and hold their breath, then an AP film in the sitting upright position is performed [24], and thus projection is a marker of a patient’s fitness during their acute illness.

Our findings agree with previous work, which found that cardiomegaly on the CXR, regardless of projection, was related to a worse prognosis in patients with acute and chronic HF [9, 10]. However, despite a high level of readmission to hospital within 30 days of discharge, in-hospital and 30-day mortality was low. None of these outcomes strongly related to the degree of congestion, consistent with the observation that early readmissions are often not for HF [25,26,27]. Our CXR score was strongly related to clinical variables that are related to the prognosis of patients with acute HF. Thus, it is not surprising that CXR features do not add much to the predictive power of models that included these variables.

One surprising finding was that a lower initial heart rate was related to higher mortality. Lourenco and colleagues had similar findings amongst 564 patients presenting with acute HF to a single centre [28]. In both studies, the first heart rate measurement was recorded on presentation to hospital. Although a higher resting heart rate predicts higher mortality in chronic HF, a faster initial heart rate in acute HF may be a marker of a preserved autonomic nervous response.

Limitations

This is a single centre study. Most patients were of Caucasian ethnicity. Notably, Kobayashi and colleagues recently examined pulmonary congestion on the CXR in 117 patients and found that a worsening congestion score index was associated with a composite of all-cause mortality or rehospitalisation for HF at 90 days [29]. The index was generated from dividing the chest X-ray into six zones, and grading the severity of congestion in each lung zone. Our method of grading the CXR is far more likely to be the way CXRs are routinely examined in the UK. We have also taken into account radiograph projection or CTR, both of which we have shown to relate to worse prognosis.

Our version of the CXR score need to be validated in another population. Although reports from radiologists were used, no radiologists were involved in the retrospective review of the films. This may account for some of the variation in reproducibility of CXR variables and possibly misinterpretation of films, such as pneumonia being mistaken for localised alveolar oedema. However, diagnosis and treatment of HF is started as soon as the patient is admitted to hospital and not usually after radiology consult. The Heart Failure Association of the European Society of Cardiology has published a position paper which recommends proper assessment of cardiomegaly, pulmonary venous congestion, pleural effusions, interstitial and alveolar oedema by all admitting physicians [30]. This study, therefore, reflects how CXRs would be interpreted by the majority of clinicians treating patients presenting with breathlessness.

Most CXRs were done within the first 24 h of admission but the CXR scores of those who had their first film after 24 h were lower. It is unknown which of these CXR appearances were influenced by diuretic administration. In another study, Kobayashi and colleagues found that residual pulmonary congestion on the CXR prior to discharge predicted a composite outcome of all-cause mortality or rehospitalisation for heart failure at 1 year [31].

Conclusions

In patients admitted to hospital for HF, there is a high prevalence of radiographic features of congestion. Worsening features of congestion are associated with variables related to worse prognosis. An increase in the CXR score is related to increasing all-cause mortality but is not an independent predictor of outcome when other variables are taken into account.