Abstract

Pulmonary hypertension (PH) is highly prevalent in patients with interstitial lung disease (ILD) and is associated with increased morbidity and mortality. Widely available noninvasive screening tools are warranted to identify patients at risk for PH, especially severe PH, that could be managed at expert centres. This review summarises current evidence on noninvasive diagnostic modalities and prediction models for the timely detection of PH in patients with ILD. It critically evaluates these approaches and discusses future perspectives in the field. A comprehensive literature search was carried out in PubMed and Scopus, identifying 39 articles that fulfilled inclusion criteria. There is currently no single noninvasive test capable of accurately detecting and diagnosing PH in ILD patients. Estimated right ventricular pressure (RVSP) on Doppler echocardiography remains the single most predictive factor of PH, with other indirect echocardiographic markers increasing its diagnostic accuracy. However, RVSP can be difficult to estimate in patients due to suboptimal views from extensive lung disease. The majority of existing composite scores, including variables obtained from chest computed tomography, pulmonary function tests and cardiopulmonary exercise tests, were derived from retrospective studies, whilst lacking validation in external cohorts. Only two available scores, one based on a stepwise echocardiographic approach and the other on functional parameters, predicted the presence of PH with sufficient accuracy and used a validation cohort. Although several methodological limitations prohibit their generalisability, their use may help physicians to detect PH earlier. Further research on the potential of artificial intelligence may guide a more tailored approach, for timely PH diagnosis.

Shareable abstract

There is no single noninvasive test to accurately detect pulmonary hypertension in interstitial lung disease. Currently, a combination of echocardiography, chest computed tomography, functional modalities and biomarkers is needed. AI may facilitate this. https://bit.ly/4ewPI4T

Introduction

Pulmonary hypertension (PH) is a frequent complication in patients with interstitial lung disease (ILD) associated with poor functional status, a need for supplemental oxygen, reduced quality of life and worse prognosis [1]. Thus, early recognition of PH in ILD is important in planning diagnostic tests, initiating targeted vascular therapy where appropriate and available, and considering referral for lung transplantation. Right heart catheterisation (RHC) remains the gold standard for diagnosing PH. However, it is not performed on a routine basis, but, as per European (European Respiratory Society (ERS) and the European Society of Cardiology (ESC)) guidelines [2], it is considered when severe PH is suspected, when the patient's management will be influenced by RHC results, including consideration of pulmonary vasodilators on an individual case basis in expert centres, referral for transplantation, exclusion of left heart dysfunction and inclusion in clinical trials [3].

Widely available, noninvasive tools for screening for PH are ideal in this population, where PH onset can be difficult to detect as symptoms and signs can be nonspecific. In most centres, Doppler transthoracic echocardiography (TTE) is the mainstay for initial PH screening in this setting. This may be performed routinely or following clinical assessment where PH is suspected. Although commonly available, the usefulness of echocardiography in detecting PH relies on the presence of sufficient tricuspid regurgitation to assess right ventricular (RV) systolic pressure (RVSP). Tricuspid regurgitation velocity (TRV) is often difficult to measure in patients with ILD and may result in over- or underestimation of RVSP [4–6]. This information is usually supplemented by indirect echocardiographic markers of PH, that slightly increase the diagnostic accuracy, mainly in detecting severe PH (figure 1) [7]. According to the proceedings of the Sixth World Symposium on PH [1], severe PH was defined as a mean pulmonary artery pressure (mPAP) ≥35 mmHg or an mPAP ≥25 mmHg with a low cardiac index (<2.0 L·min−1·m−2). The 2022 ERS/ESC guidelines on PH proposed a new definition for severe PH, based on the pulmonary vascular resistance (PVR) >5 Wood units) only [2].

FIGURE 1

Echocardiographic markers used to detect pulmonary hypertension in a patient with idiopathic pulmonary fibrosis. a) Continuous-wave Doppler echocardiography on a modified right ventricular (RV) focused four-chamber view depicts an increased tricuspid valve regurgitation velocity of 3.3 m·s−1. b) M-mode echocardiography applied on the RV free wall at the level of the tricuspid annulus reveals a reduced tricuspid annular plane systolic excursion. c) Right heart focused four-chamber view shows a severely dilated right atrium and a normal size of left atrium. d) Inferior vena cava distension (>21 mm) with diminished inspiratory collapsibility as depicted on a subcostal view.

Other, noninvasive modalities that might raise suspicion for the presence of PH in ILD include circulating biomarkers, such as brain natriuretic peptide (BNP) and N-terminal pro-brain natriuretic peptide (NT-proBNP), ECG, pulmonary function tests (PFTs), exercise testing, chest computed tomography (CT) and cardiac magnetic resonance imaging (CMR).

There is currently no single noninvasive test capable of accurately detecting and diagnosing PH in ILD patients. The use of different noninvasive parameters may improve diagnostic accuracy, but guidelines still mandate the use of RHC to confirm the diagnosis [2]. Therefore, there is a strong need for a multiparametric score that accurately predicts PH and, potentially, its severity in this patient group.

The scope of this review is to summarise the current evidence about noninvasive, diagnostic modalities and prediction models to detect PH in a timely manner in patients with ILD, critically evaluate these models and present future perspectives in this field.

Search strategy and study selection

A comprehensive literature search was carried out in PubMed and Scopus according to published guidance on narrative reviews [8] using the following terms: “(interstitial lung disease) AND (pulmonary hypertension) AND (screening)”, “(interstitial lung disease) AND (pulmonary hypertension) AND ((non-invasive parameters) OR (pulmonary hypertension predictors) OR (diagnostic markers))”, “(idiopathic pulmonary fibrosis) AND (pulmonary hypertension) AND (screening)”, “(hypersensitivity pneumonitis) AND (pulmonary hypertension) AND (screening)”, “(systemic sclerosis) AND (lung fibrosis) AND (pulmonary hypertension) AND (screening)”. Original research articles focusing on the identification of either univariate noninvasive diagnostic markers or multivariate diagnostic models and/or scores to detect PH in patients with ILD, registered until the end of March 2024, were selected to be discussed in this review. Review articles, letters to the editor, communications, conference abstracts, publications not in the English language, animal studies and data from ongoing research were excluded. All references cited in the retrieved articles were also reviewed to identify additional published work.

The PRISMA flowchart summarises the study selection process (figure S1) [9]. Out of 2687 articles initially identified through Scopus, PubMed and additional sources search, 38 original research articles, all of which were observational studies, either prospective or retrospective, were selected for discussion in this review.

In nine studies, PH was estimated using echocardiography only and not by performing RHC. Therefore, these studies are considered to have lower diagnostic quality and are presented in table S1 [10–18]. In the rest of the studies, RHC was performed in patients with moderate or high echocardiographic probability of PH. All studies published before 2022 used the haemodynamic definition of the Sixth World Symposium [1] to define severe PH.

To date, various diagnostic models have been proposed to be utilised in everyday practice in order to detect PH in patients with ILD (table 1). Furthermore, several noninvasive markers demonstrated at least a moderate correlation with mean or systolic pulmonary artery pressure (table 2).

TABLE 1

Overview of observational studies presenting multi-parametric prediction models of pulmonary hypertension (PH) in patients with interstitial lung disease (ILD)

TABLE 2

Overview of observational studies exploring predictors of pulmonary hypertension (PH) not included in a multi-parametric model in patients interstitial lung disease (ILD)

Imaging modalitiesTTE

TTE is currently the most popular and widely available noninvasive screening tool to identify patients with ILD and suspected PH. The most commonly used variables are the peak TRV, the derived tricuspid gradient and the estimated RVSP, derived from continuous wave Doppler echocardiography. The diagnostic accuracy of TRV to detect PH in patients with ILD is generally moderate (area under the curve (AUC) 0.73, 95% CI 0.68–0.79; p<0.01), as confirmed by a prospective study in 265 patients with ILD, including patients with idiopathic pulmonary fibrosis (IPF), connective tissue disease and lung fibrosis, sarcoidosis, and other idiopathic interstitial pneumonias (IIPs) [6]. This may be attributed to the difficulty in visualising the tricuspid regurgitant jet in many patients with ILD due to challenges in obtaining satisfactory acoustic windows and, therefore, may lead to over- or underestimation of the RV systolic pressure when compared with diagnostic RHC. In addition, in another study, only a moderate correlation between estimated RVSP and systolic PAP (R=0.66, p<0.0001) was detected in patients with IPF that underwent RHC [19]. A similar correlation was found by Heiden et al. [20], between TRV and mPAP measured with RHC in patients with pulmonary Langerhans cell histiocytosis (R=0.7, p=0.004).

On the other hand, Bax et al. [7] demonstrated that combining RVSP with other secondary echocardiographic markers, increases diagnostic accuracy in patients with ILD. RVSP, in addition to five indirect echocardiographic markers of PH (early diastolic pulmonary regurgitation velocity, right atrial area, RV fractional area change (FAC), RV to left ventricle (LV) basal diameter ratio and LV eccentricity index) were utilised to build a simple stepwise echocardiographic multivariable model that accurately detected severe PH among 210 ILD patients. Overall, 85 (40%) patients had severe PH (mPAP ≥35 mmHg). A score ≥7 predicted severe PH with a high sensitivity of 89% and specificity of 71% (AUC 84.8%). Importantly, RVSP could be estimated in 92% of studies. However, even when reducing this percentage to 60%, a fair diagnostic accuracy was maintained (AUC 74.4%). This score was validated in a separate group of 61 patients of the same consecutive cohort with similar diagnostic accuracy.

The diagnostic accuracy of Doppler echocardiography can be improved, when combined with other noninvasive markers. Ruocco et al. [21] built a multiparametric score by combining RVSP ≥40 mmHg with other echocardiographic indices (estimated mPAP ≥25 mmHg derived by peak diastolic pulmonary regurgitation velocity and tricuspid annular systolic plain excursion (TAPSE) ≤16 mm), an elevated BNP >50 pg·mL−1 and a low diffusing capacity for carbon monoxide (DLCO <40%) estimated with spirometry. A score ≥3 demonstrated an excellent concordance with invasive measurements (concordance: 0.964; Cohen's K index: 0.825). Major limitations of this study were that RHC was only performed in patients with an RVSP ≥40 mmHg and that patients with PH potentially attributed to left heart disease were not excluded. Further studies have revealed an improved diagnostic accuracy of estimated RVSP when included in multivariable models containing parameters obtained from either PFTs (gas exchange derived pulmonary vascular capacitance and forced vital capacity (FVC)/DLCO ratio) [22] or high-resolution CT (HRCT) (pulmonary artery (PA) area and ratio of segmental artery to adjacent bronchus diameter located in the apicoposterior segment of the left upper lobe) (R2=0.53, p=0.0009) [23] or from both noninvasive modalities (PA to ascending aorta (AA) diameter ratio and FVC/DLCO) [24].

In general, RV dysfunction as assessed with common echocardiographic markers, such as TAPSE, FAC, RV pulsed tissue Doppler S wave velocity appears in the later stages of PH and is therefore not a suitable parameter for early detection of PH. However, RV global longitudinal strain (RV GLS) may play a role in early diagnoses of PH in patients with IPF. D’Andrea et al. [13] demonstrated a good linear correlation between RV GLS and noninvasively estimated mPAP (R=0.72, p<0.0001) in 52 patients with IPF, while RV GLS was impaired even among patients with IPF without increased pulmonary artery pressures, indicating an early impact on RV function. Further studies with haemodynamic assessment of IPF patients are needed to test the diagnostic accuracy of RV speckle tracking parameters in the early detection of PH in this population [25].

In summary, utilisation of RVSP, when technically feasible, along with secondary echocardiographic markers of PH, can reliably detect severe PH in the majority of patients with ILD [7]. Combining echocardiographic with other noninvasive markers, such as DLCO or FVC/DLCO or CT-derived PA/AA diameter ratio, may further increase diagnostic accuracy, but needs to be further investigated and validated in additional cohorts.

Chest CT

HRCT is routinely performed to diagnose the subtype of ILD and to evaluate the lung parenchyma for the presence and severity of fibrosis and coexistent emphysema (figure 2) [26]. CT pulmonary angiography (CTPA) is generally performed when pulmonary embolism, that may co-exist in patients with ILD and/or chronic thromboembolic PH (CTEPH) is suspected, or to detect other causes of PH, such as anomalous pulmonary venous drainage [27]. Recently, CTPA was used to assess pulmonary vessel volume among patients with chronic lung disease (CLD) of multiple aetiologies and PH. A lower small pulmonary vessel volume was detected among patients with severe PH (mPAP ≥35 mmHg) and CLD compared to patients with mild–moderate PH-CLD (mPAP 21–34 mmHg), whereas patients with PH and ILD had smaller pulmonary vessel volumes compared to those with PH and chronic obstructive lung disease [28]. Moreover, a small retrospective study on smokers who had undergone lung resection demonstrated that CT signs of small vessel volume loss was associated with histological findings [29].

FIGURE 2

Chest imaging signs suggestive of the presence of pulmonary hypertension in a patient with idiopathic pulmonary fibrosis. a) Chest radiograph showing a dilated main pulmonary artery with peripheral vascular “pruning” and cardiomegaly. b) High-resolution computed tomography depicting a diffuse fibrotic lung disease characterised by striking traction bronchiectasis, ground-glass lesions and fine reticulation. c) Computed tomographic pulmonary angiography (CTPA) showing a dilated pulmonary artery (PA) at the bifurcation level (37.3 mm) and an increased PA to ascending aorta diameter ratio of 1.05. d) CTPA also demonstrates an increased basal right ventricular diameter to left ventricular diameter ratio of 1.6.

HRCT

Whether the extent of interstitial fibrosis as assessed with HRCT is a predictor of precapillary PH in ILD remains controversial. Identification of a fibrotic pattern in patients with hypersensitivity pneumonitis was associated with a high probability of PH [18, 30]. Whereas the extent of lung fibrosis was an independent predictor of PH in patients with systemic sclerosis (SSc) [10], it did not correlate with RHC-derived mPAP in patients with advanced IPF [31, 32].

Although the measurement of overall cardiac size is possible in a noncontrast CT, accurate measurement of cardiac chambers and vascular structures is considered challenging and needs expertise. Furukawa et al. [33] managed to reliably measure the diameter of the PA at its bifurcation and the largest diameter of the AA on HRCT images and developed a simple clinical score for predicting PH in 273 patients with IPF consisting of % predicted DLCO <50%, PA/AA ratio ≥0.9 and arterial oxygen tension <80 kPa with a good diagnostic accuracy (AUC 0.76, 95% CI 0.68–0.83). In addition, Ratanawatkul et al. [34] achieved a moderate diagnostic accuracy to detect PH among ILD patients, using mean PA diameter >32.5 mm and PA/AA diameter ratio ≥0.92 as measured on HRCT (AUC 0.66 and 0.69).

CTPA

In general, CTPA is the method of choice for accurate measurement of PA and AA diameter and allows to determine the size of ventricles and atria and the position of interventricular septum (for instance, bowing to the left ventricle in the presence of increased RV pressures) (figure 2). PA dilatation (more than 25 or 30 mm) was found to be related to a higher probability of PH in ILD patients [32, 35, 36], but it may occur even in the absence of PH in patients with pulmonary fibrosis [37]. In patients with advanced IPF, the main PA diameter did not correlate with RHC-derived mPAP [31]. Similarly, Devaraj et al. [37] evaluated CT findings in 30 patients with all subtypes of ILD and PH and found that absolute PA diameter did not correlate with mPAP or pulmonary vascular resistance index (PVRi) measured by RHC, whereas correcting for the size of AA, i.e. main PA diameter to ascending aorta diameter ratio, significantly strengthened the correlation to mPAP and PVRi in this group of patients. On the other hand, Chin et al. [38] found that the main PA diameter had a much better diagnostic accuracy for PH detection (AUC 0.87) than PA/AA diameter ratio or right or left pulmonary artery diameter (table 2) and correlated well with mPAP in 110 patients with ILD. Therefore, given the existing evidence, PA diameter per se cannot be considered a reliable noninvasive predictor of PH among patients with pulmonary fibrosis.

In contrast, a ratio of >0.9 between the diameters of PA to AA was found to be an independent predictor of RHC-derived mPAP >20 mmHg with a good diagnostic accuracy (AUC 0.75) among 177 consecutive patients with IPF and was associated with a worse prognosis [39]. This ratio has also been implemented in a validated multiparametric eight-variable score consisting of clinical (syncope or presyncope, clinical signs suggestive of PH, need for supplemental oxygen, presence of connective tissue disease or sarcoidosis), functional parameters (6-min walking distance (6MWD) <350 m, DLCO <40%) and biomarkers (BNP >50 pg·mL−1 or NT-proBNP >300 pg·mL−1) achieving an excellent diagnostic accuracy in early detection of PH among 154 ILD patients with an AUC of 0.92 [36]. A score ≥6 demonstrated high probability of PH and a referral to a PH expert centre for further evaluation with an echocardiogram and RHC is recommended by the authors. On the other hand, a retrospective analysis in 235 patients with IPF evaluated for lung transplant revealed that a PA/AA diameter ratio <1.1 along with a QRS axis <90° on ECG and a normal RV function accurately ruled out precapillary PH with a negative predictive value of 85% [40]. In a recent meta-analysis of 12 studies and 1959 patients, a PA/AA diameter ratio >1 predicted PH of any type with a sensitivity of 0.65 and specificity of 0.83 [41].

Another potential marker is a ratio >1 of the axial diameter of the RV to the LV, which had a moderate diagnostic accuracy (AUC 0.69) in identifying PH, was associated with increased PVR and also predicted mortality or transplantation in patients with ILD [42]. However, its diagnostic accuracy has not been tested in the context of a multivariable model and this method needs a contrast lung CT to enable definition of right heart chambers.

Finally, a CT-derived diagnostic model including PA diameter, RV outflow hypertrophy, interventricular septal angle and left ventricular area was derived among 247 patients in different clinical settings with great diagnostic accuracy in detecting all types of PH (mPAP >20 mmHg) in the derivation (AUC 0.92) and validation cohort (AUC 0.94) [43]. A second model, proposed by the same authors [43], incorporated PA diameter, RV outflow tract thickness and RV/LV diameter ratio with similar diagnostic accuracy (AUC 0.89) in the derivation cohort, but lower accuracy in the validation cohort (AUC 0.86). However, this study was not restricted to patients with chronic lung disease, but included consecutive patients suspected of having PH of any cause.

Overall, PA/AA ratio >0.9 seems to be a reliable CT-derived independent predictor of PH in patients with ILD [33, 36, 39]. Other CT markers, such as RV/LV diameter ratio >1, RV outflow hypertrophy, flattened ventricular septum appear clinically useful, but should further be tested in multivariable models focusing on patients with ILD.

CMR

CMR is the gold standard imaging modality for accurate quantification of RV volumes and function. Whereas it has been widely used in patients with pulmonary arterial hypertension (PAH), both as a diagnostic and a prognostic tool, in ILD there is a misconception that CMR may be technically impractical to perform, as patients with lung disease cannot tolerate lying flat for a long period of time. Indeed, several CMR studies among patients with ILD have proved its feasibility and showed that it is useful for accurately measuring RV function, especially in those patients with low-quality echocardiographic views [44]. The RV longitudinal strain as assessed with CMR was significantly impaired in ILD patients with PH than ILD patients without PH and predicted short-term mortality [44]. In addition, an RV ejection fraction ≤45% strongly predicted adverse outcomes among ILD patients [45]. Studies investigating the diagnostic accuracy of several CMR parameters to detect PH are still lacking. Chin et al. [38] found systolic and diastolic PA area calculated with CMR to accurately predict PH (AUC 0.88 and 0.89, respectively), similar to the CT-derived PA diameter and correlated well with RHC-derived mPAP. In addition, a CMR model, using ventricular mass index, PA relative area change and systolic septal angle diagnosed severe PH, using both definitions, with excellent accuracy (AUC 0.91, p<0.0001) in 167 patients with various types of CLD [46].

Overall, further studies are needed to identify if and which CMR parameters can be implemented into a validated diagnostic model, to guide detection of severe PH in patients with ILD. Currently, its use should be best limited for risk stratification of ILD patients with established PH.

Functional markersPulmonary function assessment

In patients with ILD, PFTs are routinely performed to evaluate lung function impairment, with FVC and DLCO being the most commonly assessed parameters. Both parameters are decreased in patients with pulmonary fibrosis, irrespective of the presence of PH, and correlate with disease progression [37, 47]. DLCO reduction is nonspecific for ILD and/or related PH as it can be reduced both in vascular and fibrotic disease as well as in patients with chronic pulmonary embolism, emphysema, cigarette smoking, pulmonary oedema and anaemia (all possible comorbidities) [47]. However, a severely reduced or decreasing DLCO, disproportionate to lung volume parameters, has been found to be significantly associated with the development of PH among ILD patients [15, 39, 48]. In addition, the majority of available multiparametric PH prediction scores include DLCO as an independent predictor of PH, along with other functional or imaging parameters, which significantly improves their diagnostic accuracy [21, 33, 36]. Decreased DLCO also holds prognostic significance in established PH-ILD, as it was the only variable, together with the diagnosis of IPF, that independently predicted mortality in ILD patients with severe PH [49]. In the absence of emphysema, transfer coefficient of the lung for carbon monoxide (KCO) may be more specific for the pulmonary vasculature, as it is adjusted for alveolar volume. Corte et al. [11] proposed a vascular index for patients with IIP with either a baseline KCO ≤50% and/or a 15% decline in KCO at 6 months, which was strongly associated with a high probability of PH, as estimated with TTE at follow-up assessment, even after adjustment for age, gender, emphysema and baseline KCO.

In patients with SSc, the presence of a markedly reduced DLCO also predicts the presence of precapillary PH, which can be attributed either to PAH in those without significant lung fibrosis or may be associated with ILD [10, 50–52]. DLCO can be partitioned into two transfer components, namely membrane conductance for CO (DmCO) and CO loading on haemoglobin (Hb), which is the product of the CO-Hb chemical reaction rate (θCO) and the mass of Hb in the alveolar capillary blood volume (Vcap) [53]. A decreased DmCO is interpreted as a “thickening’ of the alveolo–capillary network or as a decrease in lung area. Sivova et al. [51] demonstrated that a Vcap <19 mL had a better AUC than DLCO, DmCO and FVC/DLCO ratio to detect PH in patients with SSc and ILD.

An increased FVC/DLCO ratio seems to be a reliable, but also nonspecific, lung function marker that predicts PH in patients with ILD [18] and has been recently included in several clinical scores with a good diagnostic accuracy [22, 24, 54]. Zisman et al. [55] developed a mathematical formula that predicts mPAP using the predicted FVC/DLCO ratio and the oxygen saturation with very good accuracy (table 1). However, this formula was validated in two distinct small cohorts of patients with IPF [56] and its validity remains to be assessed in a larger cohort of patients. Finally, Sobiecka et al. [16] presented total lung capacity (TLC) % predicted/DLCO predicted as a more accurate lung function marker than FVC/DLCO ratio or DLCO to recognise ILD patients with a high echocardiographic probability of PH; however, without investigating its accuracy to detect PH confirmed with RHC.

In summary, it seems that the majority of the available multiparametric models propose an increased FVC/DLCO ratio [22, 24, 31, 54] or a reduced DLCO [21, 33, 36] to be used as independent predictors of PH. Discrepancies among studies in terms of the optimal cutoff values prohibit the selection of a specific cutoff value for each parameter. Disproportionate, longitudinal reduction in DLCO in comparison to FVC appears a useful clinical sign for increasing vascular involvement. Further research is therefore required towards this direction.

Exercise capacity assessment

The 6-min walk test (6MWT) is a useful tool for the assessment of exercise capacity in patients with heart or lung disease. 6MWD is the most widely used marker to assess exercise tolerance among patients with PH and has also a prognostic significance [2]. However, it may be affected by age, sex and comorbidities [57]. Reduced exercise capacity and hypoxaemia are nonspecific markers to predict PH. Chronic hypoxia is important in the aetiology of PH due to lung disease, but it cannot explain the development of disproportionate PH. In patients with IPF, the early development of PH may be associated with increased fibrotic cell mediators, abnormal vasculature or response to hypoxia [58]. Nocturnal and exercise desaturation are common and may precede and contribute to the development PH. A significant decrease in 6MWD of less than 350 m, oxygen saturation at rest below 90% and a worse oxygen desaturation during the 6MWT below 88% were found to be independent predictors of PH [16, 18, 36, 39, 48, 54] and have been adopted by few recent multiparametric scores to detect PH [16, 18, 54]. Reduced heart rate recovery at the end of 6MWT (maximal heart rate achieved minus heart rate at 1 min after exercise <13 beats per minute) has also been shown to predict PH in patients with IPF [59].

The cardiopulmonary exercise test (CPET) is another noninvasive test that can objectively assess exercise tolerance and, compared to 6MWD, provides several additional prognostic markers relevant to PH. In a multicentre retrospective study, the presence of PH was best predicted by gas exchange efficiency for carbon dioxide (minute ventilation (VʹE)/carbon dioxide production (VʹCO2) slope (VʹE/VʹCO2)) (cutoff ≥152% pred; AUC 0.94) and peak oxygen uptake (≤56% pred; AUC 0.83), followed by diffusing capacity in 135 IPF patients [60], with resting lung volumes failing to predict PH. ILD patients with PH undergoing evaluation for lung transplantation demonstrated significantly lower mixed-expired carbon dioxide pressure (PECO2) and end-tidal carbon dioxide pressure (PETCO2) during certain levels of exercise with a distinctive activity pattern for PECO2/PETCO2 [48]. A follow-up study by the same group showed that patients with severe PH, defined by mPAP ≥40 mmHg, had significantly higher VʹE/VʹCO2 and lower PETCO2 at maximal exercise compared to patients with mild to moderate PH or no PH reflecting disproportionate ventilation with worsening mPAP. PETCO2 had the best diagnostic accuracy (AUC 0.98) for PH followed by DLCO (AUC 0.86) [61].

Among 35 patients with Langerhans cell histiocytosis, those with PH had lower values for peak oxygen consumption (VʹO2) (% predicted) and oxygen pulse with greater oxygen desaturation during CPET compared to patients without PH [20]. A moderate correlation was detected between resting mPAP values and both the exercise oxygen saturation nadir (R=−0.73, p=0.001) and VʹO2 at anaerobic threshold (R=−0.61, p=0.01). Although CPET provides greater diagnostic capability compared to the 6MWT in assessing exercise tolerance, its robust interpretation in patients with ILD and PH presents specific technical challenges around the frequent requirement for oxygen supplementation which must be administered within a closed system, if maximal patient effort is to be achieved. In contrast, the 6MWT is less technically challenging and more easily repeatable in patients with a high level of long-term/ambulatory oxygen use. Therefore, it has been used as a validated end-point in multiple ILD studies.

Based on available literature, a 6MWD <350 m [36, 54] in combination with oxygen saturation <88% during the 6MWT [54] could be used as independent predictors of PH. Further studies are required on CPET to identify the variable with best diagnostic accuracy.

Biomarkers

BNP and NT-proBNP are produced mainly from ventricular myocytes as a response to ventricular and pressure overload and have been shown to correlate with parameters of pulmonary circulation obtained by RHC or TTE, but their predictive value is limited by confounding factors such as renal or left-heart disease [62]. A BNP value of more than 50 pg·mL−1 or an NT-proBNP >300 pg·mL−1 have been incorporated in several composite scores for PH detection in ILD patients [21, 36]. In addition, an NT-proBNP <95 ng·L−1 had a high negative predictive value for the presence of PH among 212 individuals with ILD, indicating that it may be used as a rule-out test for PH in those patients [63]. Besides diagnostic purposes, BNPs may be used as prognostic markers and for risk stratification purposes in patients with an established diagnosis of PH [62]. So far, no other biomarkers have been shown to predict PH among ILD patients.

In general, BNPs cannot reliably predict the presence of PH as individual variables, but significantly increase the diagnostic accuracy of other noninvasive parameters. On the other hand, the presence of PH is rather unlikely when BNP or NT-proBNP values are within normal ranges.

ECG

Right-axis deviation (RAD) and RV strain have been identified as independent predictors of precapillary PH in mixed cohorts of patients including patients with various CLD [64, 65]. In ILD, RAD was an independent predictor of precapillary PH together with other noninvasive parameters among 235 patients with IPF screened for lung transplantation [40].

Overall, the presence of RAD and RV strain may suggest the presence of PH, but are not disease-specific markers. On the other hand, a normal ECG does not exclude the presence of PH in patients with ILD, but when combined with a normal RV function and a normal PA/AA ratio, the negative predictive value rises up to 85% [40].

Overview of overall approach to detection of PH in patients with ILD including use of current multi-parametric models

To date, there is no consensus or a validated tool to guide us on who will eventually benefit from RHC, as the prevalence of PH among ILD patients is high, but only a proportion of them will receive targeted pulmonary vascular therapies or have a different management approach, if PH is confirmed. In general, there is a variation in the availability of targeted pulmonary vascular therapies among countries and clinical management may differ among centres. Currently, there is a tendency to proceed to RHC when noninvasive screening tests indicate severe PH as highly likely, when co-existent PAH or CTEPH are suspected, when pulmonary vasoactive therapies are considered or during work-up for lung volume reduction surgery or lung transplantation [3], while RHC should not be performed in patients for whom the diagnosis of PH will not change their management plan.

As discussed, there is currently no single noninvasive test that can accurately detect PH in patients with ILD and thus there is no proposed “standard” approach to assess a patient's risk for developing PH in ILD. Recently, the modified Delphi study, conducted by a panel of experts in ILD and PH, concluded that an early screening for PH with a low threshold of suspicion should be applied [66].

RVSP remains the single most predictive factor of PH, with other readily available variables obtained by TTE, chest CT, PFTs or exercise tests able to increase its diagnostic accuracy [7, 22–24]. However, in which way these variables should be used in combination and be weighted in risk-stratifying ILD patients remains largely unknown. The modified Delphi consensus proposed TTE and BNP or NT-proBNP as initial screening tests, when PH is clinically suspected [66]. Additionally, a number of triggers for suspicion of PH, including symptoms and signs disproportionate to the disease severity, abnormal findings on chest CT scan, PFTs and the 6MWT could be used together to evaluate ILD stability or progression and further risk stratify for PH in patients with ILD [36].

The majority of existing composite scores to early detect PH in patients with ILD are derived from retrospective studies and lack validation in external cohorts (table 1). Only two available noninvasive scores, by Bax et al. [7] and Nathan et al. [54], detecting the presence of haemodynamically diagnosed PH among ILD patients have been validated in a second cohort.

Nathan et al. [54] recently developed a clinical score (the FORD model) based on the easily available functional parameters: FVC/DLCO ratio, 6MWD, lowest oxygen saturation recorded during 6MWT and race, with a moderate diagnostic accuracy to detect PH in a large cohort of patients with mild to moderate IPF (AUC 0.75), while it achieved a lower accuracy in the validation cohort (AUC 0.69). This may be attributed to the fact that the validation cohort included also patients with severe IPF, who were being evaluated for lung transplantation, in whom most noninvasive functional parameters were already impaired, as most markers are not PH-specific and, therefore, their diagnostic accuracy to detect PH was limited.

On the other hand, Bax et al. [7] produced a more robust stepwise score, including echocardiographic variables only, with RVSP being the first parameter evaluated, and presented a better diagnostic accuracy (AUC >0.8) in both cohorts [7]. However, the aim of this score was to identify the presence of severe PH among ILD patients and, therefore, it is not so sensitive in detecting mild or moderate PH.

In summary, the score by Nathan et al. [54] could be widely used by pulmonologists to screen for the presence of even mild PH in all patients with IPF, as it comprises functional and easily obtainable parameters. The echocardiographic based score by Bax et al. [7] could be used in expert centres with experience in echocardiography when severe PH is suspected in patients with various types of ILD.

Future perspectives

Given the nonspecific medical symptoms and the limitations of the conventional diagnostic modalities and available multi-parametric models, it is not surprising that time to definite diagnosis of PH among patients with ILD is quite prolonged, potentially contributing to delayed diagnosis with very poor outcomes [67]. After the careful evaluation of different noninvasive diagnostic tools, many of them need further exploration. Therefore, there is a strong need for improving methods to detect patients with ILD who are at risk for severe PH and refer them in a timely fashion for RHC in expert centres. Confirmation of PH diagnosis will enable timely initiation of inhaled treprostinil, as per the INCREASE trial [68], or enrolment in ongoing randomised controlled trials on targeted vascular therapies [69]. In addition, performing further studies using simple statistics in carefully phenotyped patients may have a role in better defining which variables or combination of variables can aid detecting PH in ILD. The ongoing PHINDER trial, a large multicentre prospective study, aims to assess the diagnostic value of noninvasive tests and how they can be best combined and weighted to gain the maximal diagnostic accuracy in identification of patients with PH (NCT05776225) [70].

Over the last few years, there has been a rapid progress in the field of machine learning and deep learning (DL) neuronal networks for improving the detection of cardiovascular diseases, otherwise known under the banner of artificial intelligence (AI) [71, 72]. Regarding the field of PH, there have been a few studies with promising results. DL algorithms discriminated patients with PH, diagnosed with RHC, using TTE with high accuracy and good generalisability [73–75]. A transparent AI framework has been recently developed to map the three-dimensional anatomical CT models of 75 patients with PH and evidence of lung disease, aiding in a more detailed phenotyping. It predicted high performance and robustness for the presence of emphysema and honeycombing on CT, but lacked accurate prediction of ground glass reticulation [76]. Furthermore, an automated RV/LV ratio analysis (≥1.12), through segmentation of RV and LV with the use of an 18 convolutional layer neural network model on CTPAs of 202 patients with suspected PH, was more sensitive for the detection of PH of any type when compared with manual RV/LV (≥1.16), PA diameter (≥29 mm) and PA/AA diameter ratio (≥1) [77]. Moreover, Diller et al. [78] developed a DL algorithm that achieved a great accuracy and sensitivity of 97.6% and 100%, respectively, on a per patient basis to detect PAH on routine echocardiograms irrespective of RV dilatation. The algorithm outperformed conventional echocardiographic evaluation and provided prognostic information at an expert level.

Therefore, we may hypothesise that tailored DL algorithms, incorporating noninvasive data from echocardiography, chest CT, functional and other noninvasive tests, may provide an improved, widely accepted screening tool for PH in the setting of ILD (figure 3). Research should focus on the development and validation of such algorithms in order to test their diagnostic accuracy, external validity and generalisability in the clinical setting, aiming for a widely available screening tool that can be used by nonexperts as well.

FIGURE 3

Noninvasive independent predictors of pulmonary hypertension (PH) in patients with interstitial lung disease and the potential use of artificial intelligence to improve diagnostic accuracy. Variables in bold have been included in multiparametric models. Variables in italics are univariate predictors of PH. #: Transthoracic echocardiography (TTE) variables can be used to predict severe PH as proposed in the stepwise composite echocardiographic score by Bax et al. [7]. The other non-TTE variables in bold predict the presence of any PH. ¶: Right ventricle (RV) and right atrium dilatation, RV hypertrophy, RV dysfunction (fractional area change ≤34%), early diastolic pulmonary regurgitation velocity ≥2.2 m·s−1, left ventricle (LV) eccentricity index ≥1.1, interventricular septal flattening, dilated inferior vena cava [7]. AA: ascending aorta; AI: artificial intelligence; CTPA: computed pulmonary angiography; BNP: brain natriuretic peptide; DLCO: diffusion capacity for carbon monoxide; FVC: forced vital capacity; 6MWD: 6-min walk distance; NT-proBNP: N terminal pro-brain natriuretic peptide; PA: pulmonary artery; PETCO2: end-tidal carbon dioxide pressure; RAD: right axis deviation; RBBB: right bundle branch block; SpO2: peripheral oxygen saturation; TRV: tricuspid regurgitation velocity; VʹE/VʹCO2: ventilatory efficiency; V