Radiographic Graft Surveillance in Lung Transplantation: Prognostic Role of Parametric Response Mapping

Abstract

Rationale: Chronic lung allograft dysfunction (CLAD) results in significant morbidity after lung transplantation. Potential CLAD occurs when lung function declines to 80–90% of baseline. Better noninvasive tools to prognosticate at potential CLAD are needed.

Objectives: To determine whether parametric response mapping (PRM), a computed tomography (CT) voxel-wise methodology applied to high-resolution CT scans, can identify patients at risk of progression to CLAD or death.

Methods: Radiographic features and PRM-based CT metrics quantifying functional small airway disease (PRM^fSAD) and parenchymal disease (PRM^PD) were studied at potential CLAD (n = 61). High PRM^fSAD and high PRM^PD were defined as ⩾30%. Restricted mean modeling was performed to compare CLAD-free survival among groups.

Measurements and Main Results: PRM metrics identified the following three unique signatures: high PRM^fSAD (11.5%), high PRM^PD (41%), and neither (PRM^Normal; 47.5%). Patients with high PRM^fSAD or PRM^PD had shorter CLAD-free median survival times (0.46 yr and 0.50 yr) compared with patients with predominantly PRM^Normal (2.03 yr; P = 0.004 and P = 0.007 compared with PRM^fSAD and PRM^PD groups, respectively). In multivariate modeling adjusting for single- versus double-lung transplant, age at transplant, body mass index at potential CLAD, and time from transplant to CT scan, PRM^fSAD ⩾30% or PRM^PD ⩾30% continue to be statistically significant predictors of shorter CLAD-free survival. Air trapping by radiologist interpretation was common (66%), was similar across PRM groups, and was not predictive of CLAD-free survival. Ground-glass opacities by radiologist read occurred in 16% of cases and were associated with decreased CLAD-free survival (P < 0.001).

Conclusions: PRM analysis offers valuable prognostic information at potential CLAD, identifying patients most at risk of developing CLAD or death.

At a Glance Commentary

Scientific Knowledge on the Subject

The onset of potential chronic lung allograft dysfunction (CLAD), defined as a 10–20% decline in lung function after lung transplantation, should alert the provider to investigate for possible causes for decline. Radiographic or other biomarkers that predict which patients with potential CLAD will progress more quickly to definite CLAD or death are lacking.

What this Study Adds to the Field

This study analyzes the application of parametric response mapping (PRM), an imaging modality applied to high-resolution computed tomography images. PRM provides quantification of functional small airway disease and parenchymal disease in the transplanted lung. In this cohort, the presence of elevated functional small airway disease or parenchymal disease measured by PRM at the time of potential CLAD was associated with increased risk of CLAD or death. This easily applicable imaging biomarker can assist with prognostication at early spirometric decline and has the capability to add to longitudinal graft monitoring strategies.

Chronic lung allograft dysfunction (CLAD) is the predominant cause of long-term graft failure and death after lung transplantation (1). CLAD is diagnosed when FEV₁ declines by ⩾20% and remains persistent for 3 months and treatable causes or complications have been excluded. The main subtypes of CLAD are 1) bronchiolitis obliterans syndrome (BOS), which is characterized by persistent spirometric obstruction, and 2) restrictive allograft syndrome (RAS), which is defined by low TLC (⩽90% of baseline) and persistent radiographic opacities (1–3). Some authors have used FVC as a surrogate for TLC (4, 5). The underlying pathology of both phenotypes of CLAD is fibrotic remodeling of the lung that is preceded by immune-mediated graft injury. Once persistent lung function loss occurs, prognosis is poor, with most patients demonstrating ongoing decline (6). Early recognition of ongoing graft changes before fulminant irreversible fibroproliferation is key in improving outcomes in this disease. In an attempt to diagnose CLAD in its early stages, BOS 0-p was initially proposed in 2002 as a decline in pulmonary function to <90% of the post-transplant baseline with concomitant decline in forced expiratory flow at 25–75% of lung volume (7). In the new International Society for Heart and Lung Transplantation (ISHLT) guidelines, this early spirometric decline has been renamed “potential CLAD” (1).

Radiographic monitoring of the lung allograft by high-resolution computed tomography (HRCT) scan can offer insight into etiologies of spirometric decline and is a commonly used modality in lung transplant recipients. However, radiographic features at the time of potential CLAD have not previously been closely studied, and their utility in identifying underlying pathogenic features or predicting CLAD onset and survival is not known. Parametric response mapping (PRM) is a novel CT voxel-based method that, when applied to paired inspiratory and expiratory HRCT scans, is capable of detecting functional small airway disease (fSAD) and parenchymal disease (PD). This technique has been used widely in the field of pulmonary disease and has offered insight into bone marrow transplant–related pulmonary complications and chronic obstructive lung disease (8, 9). We have previously demonstrated the applicability of this methodology in the lung transplant population (10). However, PRM has not been previously applied to patients at early spirometric decline after transplant. PRM provides objective, quantifiable measurements of lung physiologic parameters. This technique removes the interobserver variability common in radiographic interpretation of chest imaging (11–13). PRM may also allow the detection of abnormalities in the allograft that are not seen with the naked eye.

In this article, we evaluate the radiographic findings in lung transplant recipients at the time of potential CLAD. We assess the PRM findings at potential CLAD and study the prognostic utility of PRM variables in predicting future CLAD and survival.

Methods

Patient Population

Lung transplant recipients at the University of Michigan who had an HRCT scan available within 90 days before or after potential CLAD were included in the study. Transplants occurred from January 2004 to January 2016; CT scans were performed from December 2005 to September 2016. All CT scans were performed before CLAD onset. Baseline FEV₁ and FVC were defined as the average of the two highest FEV₁ and FVC values, respectively, obtained 3 or more weeks apart after transplant. Potential CLAD was defined per ISHLT guidelines as a persistent fall in FEV₁ to <90% but >80% of the post-transplant baseline FEV₁ (1). The potential CLAD date was defined as the date of spirometric decline onset. CLAD was defined per ISHLT guidelines as persistent decline in FEV₁ (two FEV₁ values obtained ⩾3 wk apart that are both ⩽80% of the post-transplant FEV₁ baseline) and/or FVC to ⩽80% of the post-transplant baseline (1). FEV₁ first decline and concurrent decline patterns were defined per previous manuscripts (5, 10). Briefly, FEV₁ first decline was defined as a persistent fall in FEV₁ with a preserved FVC (FVC >80% of the post-transplant FVC baseline) at the time of spirometric decline. Concurrent decline was defined as a persistent fall in both FEV₁ and FVC to ⩽80% of the post-transplant baseline on the same date or isolated FVC fall to ⩽80% of the post-transplant baseline. Clinical infection at potential CLAD was defined as a combination of symptoms, infiltrates, and microbiologic testing consistent with pulmonary infection. Bronchoscopic findings, including clinical data supportive of infection or acute rejection at the time of potential CLAD, were tallied.

HRCT and PRM

The HRCT images were retrospectively reviewed independently by two dedicated chest radiologists, and key radiographic findings were scored systematically, including the presence of ground-glass opacities (GGOs), nodularity, consolidation, and interstitial changes. The radiologists were asked to offer an impression of a diagnosis of BOS, RAS, and/or infection on the basis of the CT scan. The lead radiologist’s interpretations were used for the main analyses. PRM was applied to all paired CT scans as previously described (10). Briefly, lungs from both paired CT scans were segmented from the thoracic cavity using an in-house algorithm written in Matlab (MathWorks, Inc.). The whole lung inspiratory CT scan was spatially aligned to the incremental expiratory CT scan using Elastix, an open-source image registration algorithm (14, 15). This process allows the paired images to share the same geometric space, where each voxel, the smallest unit of volume in a three-dimensional image dataset, consists of Hounsfield unit (HU) values at inspiration and expiration. Each voxel was classified on the basis of a scheme of three predetermined thresholds as previously described (8). In brief, voxels with values greater than or equal to 950 HU and less than 2,810 HU at inspiration and greater than or equal to 2,856 HU at expiration were classified normal (PRM^Normal; green voxels), voxels with values greater than or equal to 950 HU and less than 2,810 HU at inspiration and less than 2,856 at expiration were classified as fSAD (PRM^fSAD; yellow voxels), and voxels greater than or equal to 2,810 HU at inspiration were PD (PRM^PD; purple voxels). The relative lung volumes, calculated as the sum of all voxels within a class normalized to the sum of all voxels within the expiratory lungs multiplied by 100, were used as global measures. Single-lung transplant recipient PRM values reflect only the transplanted lung.

Statistical Analysis

Analyses were conducted using R version 3.6.1. Continuous variables were summarized using means with SDs. Categorical variables were summarized using counts and percentages. Fisher exact test was used to compare differences in characteristics across the three PRM groups (16). Distributions of PRM variables are displayed via density plots (density function in R). The observed patterns of CLAD subtype development over time are displayed via cumulative incidence curves. The κ coefficient was used to measure inter-rater reliability between radiologists (17).

Time-to-event outcomes are displayed using Kaplan-Meier curves (18). Two-sample comparisons were done using both the log-rank test (19–21) and the restricted mean survival test (RMST) (22) for two sample tests. Multivariable restricted mean modeling was used to adjust for potential confounders (23). The RMST method and the corresponding restricted mean model methodology are useful in situations in which the area between survival curves is large, in scenarios in which hazards are not proportional, and in scenarios in which one group of interest has a survival curve that drops to zero early in follow-up. Data were censored on February 19, 2018.

Results

Patient Population

The study cohort included 61 lung transplant recipients with potential CLAD and an HRCT scan within 90 days of their potential CLAD date. Average age at transplant was 49.4 (SD, 12.7) years. Sixty-six percent of patients were male. Seventy-five percent of patients received a double-lung transplant; the remainder received single-lung transplants. The average lung allocation score at transplant was 42.7 (SD, 14.3). The group on average achieved good pulmonary function after transplant with an average FEV₁ of 81.6% predicted (SD, 22.3% predicted) and FVC of 77.1% predicted (SD, 19.1% predicted). The median time from potential CLAD date to the HRCT scan was 3 days (interquartile range, −17 to 32).

PRM Quantitative Evaluation and Radiographic Features at Potential CLAD

Quantification of fSAD (yellow), PD (purple), and normal parenchyma (green) was obtained by PRM on potential CLAD HRCT scans. Distribution of PRM variables at the time of potential CLAD is shown in Figure 1. A wide distribution of both fSAD and PD was noted, with more patients having elevated PD as compared with elevated fSAD. Using a previously identified cutoff criteria of ⩾30%, 41% (25/61) of the patients at potential CLAD had high PD compared with 11.5% (7/61) of patients with high fSAD (10). There were no patients who were found to have both high fSAD and PD. Hence, on the basis of the predominant HRCT PRM phenotype, patients could be categorized into three distinct signatures of PRM^Normal, PRM^fSAD, and PRM^PD (Figure 1B).

Figure 1.

(A) Distribution of parametric response mapping (PRM) of fSAD and PD at time of potential chronic lung allograft dysfunction. (B) Distribution of distinct PRM signatures among the cohort (n = 61). fSAD = functional small airway disease; PD = parenchymal disease.

The three PRM signature groups were evaluated for differences in clinical variables or HRCT findings observed by a radiologist (Table 1). Radiologist read scoring of various CT characteristics stratified by PRM signature groups are also presented in Figure 2. Air trapping was a common radiographic finding that was seen in 66% of patients at potential CLAD. However, there was no difference in the distribution of radiographic air trapping in patients with high PRM^fSAD, PRM^PD, or PRM^Normal (P = 0.27). GGOs were more common in the PRM^PD group (32%) compared with the PRM^fSAD (0%) and PRM^Normal (7%) groups (P = 0.03). The lead chest radiologist was asked to offer an impression of a diagnosis of BOS, RAS, or infection for each patient’s CT results. The radiologist interpreted HRCT scans to suggest BOS in 44% of patients at the time of potential CLAD. Only 5% of patients were believed to have RAS. More patients in the PRM^PD group as compared with the PRM^Normal group were clinically believed to have infection by the radiologist (P = 0.002).

Table 1.

Clinical and Radiologic Variables Stratified by PRM Indices

	All (N = 61)	PRMNormal (n = 29)	PRMfSAD ⩾30% (n = 7)	PRMPD ⩾30% (n = 25)	P Value*
HRCT variables, n (%)
Air trapping	40 (66)	18 (62)	4 (57)	18 (72)	0.27
Mild	19 (31)	8 (28)	2 (29)	9 (36)	0.86
Moderate	19 (31)	9 (31)	2 (29)	8 (32)	1.0
Severe	2 (3)	1 (3)	0 (0)	1 (4)	1.0
Ground-glass opacities	10 (16)	2 (7)	0 (0)	8 (32)	0.03
Nodularity	22 (36)	7 (24)	4 (57)	11 (44)	0.15
Consolidation	7 (11)	1 (3)	1 (14)	5 (20)	0.11
Interstitial changes	3 (5)	0 (0)	0 (0)	3 (12)	0.15
Radiologist impression, n (%)
BOS	27 (44)	11 (38)	5 (71)	11 (44)	0.36
RAS	3 (5)	1 (3)	0 (0)	2 (8)	0.72
Infection	26 (43)	6 (21)	4 (57)	16 (64)	0.003
Bronchoscopy at potential CLAD, n (%)	42 (69)	18 (62)	5 (71)	19 (76)	0.46
Diagnostic transbronchial biopsy at potential CLAD, n (%)	29 (48)	13 (45)	3 (43)	13 (52)	0.94
Acute rejection at potential CLAD†, n (%)	10 (34)	4 (31)	1 (33)	5 (38)	0.98
Clinical infection at potential CLAD, n (%)	24 (39)	8 (28)	4 (57)	12 (48)	0.18
BMI at potential CLAD, kg/m2, mean (SD)	27.4 (5.35)	28.4 (4.06)	23.1 (7.14)	27.4 (5.74)	0.06

Definition of abbreviations: BMI = body mass index; BOS = bronchiolitis obliterans syndrome; CLAD = chronic lung allograft dysfunction; fSAD = functional small airway disease; HRCT = high-resolution computed tomography; PD = parenchymal disease; PRM = parametric response mapping; RAS = restrictive allograft syndrome.

^*Fisher exact test.

^†Acute rejection at potential CLAD indicates patient had a transbronchial biopsy that showed any grade of type A or B acute rejection. The denominator is number of diagnostic transbronchial biopsies in that group.

Figure 2.

Distribution of radiologic features and radiologist diagnosis stratified by PRM group. fSAD = functional small airway disease; HRCT = high-resolution computed tomography; PD = parenchymal disease; PRM = parametric response mapping.

Agreement between the chest radiologists was assessed with the κ coefficient. There was only slight to fair agreement between radiologists for the key radiographic variables (24). Presence of air trapping yielded a κ coefficient of 0.189 (P = 0.131). Presence of GGOs yielded a κ coefficient of 0.311 (P = 0.004).

Of the patients who had bronchoscopy (69%) or diagnostic transbronchial biopsy (48%) performed at the time of potential CLAD, no difference was noted in the presence of clinical infection or acute rejection between PRM groups (Table 1). Bronchoscopy findings were also compared between the three PRM groups, and no difference between the incidence of acute rejection or infection was noted (Table 1). Among other clinical variables examined, the only difference among groups that approached statistical significance was body mass index (BMI) at the time of potential CLAD.

Role of CT and PRM Metrics in Predicting Survival and Progression to CLAD

Next, we investigated whether PRM metrics at the time of potential CLAD can predict progression to CLAD and death. Figure 3 demonstrates CLAD-free survival in patients with potential CLAD stratified by HRCT PRM groups. Patients with high PRM^fSAD or PRM^PD were significantly more likely to progress to CLAD or death compared with patients without elevated PRM values (P ⩽ 0.002). Patients with either high PRM^fSAD or high PRM^PD had similar CLAD-free median survival times (0.46 years and 0.50 years, respectively; P = 0.6). Patients with PRM^Normal had a significantly longer median CLAD-free survival of 2.03 years (P = 0.004 and 0.007 compared with PRM^fSAD and PRM^PD groups, respectively). Kaplan-Meier curves using different thresholds for elevation in PRM^fSAD and PRM^PD are shown in the online supplement (Figure E1). We performed a similar analysis with exclusion of the 10 patients with acute rejection at the time of potential CLAD and found similar results (Figure E2).

Figure 3.

Chronic lung allograft dysfunction (CLAD)–free survival from time of potential CLAD stratified by PRM indices. Each circle on the Kaplan-Meier curves represents a subject who died before the development of CLAD. fSAD = functional small airway disease; PD = parenchymal disease; PRM = parametric response mapping; RMST = restricted mean survival test.

All-cause mortality as the primary outcome was also examined (Figure E3). Results demonstrated a similar trend; patients with an HRCT signature of PRM^Normal at the time of potential CLAD had a median survival time of 6.51 years compared with patients with either elevated PRM^fSAD or PRM^PD, who had a median survival time of only 3.23 years (PRM^Normal vs. PRM^fSAD RMST P = 0.022; PRM^Normal vs. PRM^PD RMST P < 0.001).

The predictive ability of PRM values at the time of potential CLAD was further evaluated in a multivariate model (Table 2). In this model adjusting for single- versus double-lung transplant, age at transplant, BMI at the time of potential CLAD, and time from transplant to CT scan, PRM^fSAD ⩾30% and PRM^PD ⩾30% are both statistically significant predictors of shorter CLAD-free survival. A 49-year-old double-lung transplant recipient with potential CLAD and a BMI of 22 kg/m² who is 1 year after transplant and has PRM^fSAD ⩾30% has an estimated CLAD-free survival time over 5 follow-up years of 0.55 years (485.2 × 0.41 = 198.93 d) compared with an identical patient with PRM^Normal signature, who would live 1.3 years (485.2 d) CLAD free over 5 follow-up years. A similar patient as described above with elevated PRM^PD would live CLAD free only 0.72 years (485.2 × 0.54 = 262 d).

Table 2.

Multivariable Restricted Mean Analysis Evaluating Impact of Clinical and PRM Variables on Survival

	Estimate*	Lower Bound	Upper Bound	P Value
Intercept	485.2	282.6	833.3	<0.001
PRMfSAD ⩾30%	0.410	0.224	0.751	0.004
PRMPD ⩾30%	0.542	0.295	0.998	0.049
Single-lung transplant	1.133	0.677	1.896	0.634
Age at transplant (10-yr unit, centered at 49 yr old)	0.996	0.984	1.008	0.512
Transplant to CT scan (yr, centered at 1 yr)	1.133	1.055	1.217	<0.001
BMI at potential CLAD (centered at 22 kg/m2)	1.049	1.007	1.092	0.022

Definition of abbreviations: BMI = body mass index; CT = computed tomography; CLAD = chronic lung allograft dysfunction; fSAD = functional small airway disease; PD = parenchymal disease; PRM = parametric response mapping.

^*Estimate for intercept reflects the estimated CLAD-free survival over 5 follow-up years for a 49-year-old lung transplant patient with BMI of 22 who is 1 year after double-lung transplant with neither elevated PRM measure (485.2 d; 1.3 yr).

We have previously demonstrated that at the time of CLAD onset, patients presenting with an isolated decline in FEV₁ <80% of baseline (FEV₁ first decline) have significantly higher PRM^fSAD, whereas patients with concurrent decline in FEV₁ and FVC below their respective baselines (concurrent decline) have significantly higher PRM^PD than control subjects (10). Figure 4 shows the cumulative incidence of the CLAD subtypes (FEV₁ first decline vs. concurrent decline) eventually seen in patients with potential CLAD as stratified by PRM pattern. Each curve represents the probability of a specific event occurring at each time point. At every time point, the cumulative probability of development of FEV₁ first decline, concurrent decline, death, and CLAD-free survival add up to 100%. In our observed data, more patients with high PRM^fSAD developed FEV₁ first decline as compared with concurrent decline or death before CLAD. At 1 year, the estimated likelihood of FEV₁ first decline was 42.9% in the high PRM^fSAD group, 32% in the high PRM^PD group, and 25% in the PRM^Normal group. However, the cumulative incidence of developing FEV₁ first decline was higher than developing concurrent decline at all time points after transplant, even in the high PRM^PD group.

Figure 4.

Cumulative incidence curves demonstrate the physiologic subtypes of chronic lung allograft dysfunction (CLAD) that patients develop over time after the onset of potential CLAD, stratified by PRM metric. fSAD = functional small airway disease; PD = parenchymal disease; PRM = parametric response mapping.

The predominant CT features of air trapping, GGO, and nodularity, as characterized by a chest radiologist at the time of potential CLAD, were also evaluated in terms of their ability to predict CLAD-free survival. The presence of radiologic air trapping at potential CLAD did not predict CLAD-free survival (Figure 5A). Similarly, nodularity on CT scan did not predict CLAD-free survival (P = 0.3; data not shown). However, the presence of GGOs at the time of potential CLAD was predictive of CLAD-free survival, as shown in Figure 5B. Patients with GGOs at the time of potential CLAD had a median CLAD-free survival of 0.42 years compared with patients without GGOs, whose median CLAD-free survival was 1.29 years (RMST P < 0.001). Eight of 10 patients with GGOs by visual read were in the high PRM^PD group. Figure 6 shows a Kaplan-Meier curve with the PRM^PD group stratified by the presence or absence of GGOs. Patients with high PRM^PD in absence of GGOs by visual read had significantly worse CLAD-free survival compared with patients in the PRM^Normal group (P = 0.03). Survival was further worse in the group with both high PRM^PD and GGOs.

Figure 5.

Chronic lung allograft dysfunction (CLAD)-free survival stratified by air trapping (AT) (A) and ground-glass opacities (GGOs) (B) as interpreted by a dedicated chest radiologist. Presence or absence of AT on high-resolution computed tomography (HRCT) scan at potential CLAD did not predict CLAD-free survival (A). Patients found to have GGOs at potential CLAD had shorter CLAD-free survival (0.42 yr) compared with patients without GGOs (1.29 yr) (RMST P < 0.001) (B). RMST = restricted mean survival test.

Figure 6.

CLAD-free survival stratified by parametric response mapping (PRM) for parenchymal disease (PD) versus no elevation in PRM metrics. The PRM^PD group is further stratified by the presence or absence of ground-glass opacities (GGOs). Patients with high PRM^PD with GGOs fared worse. CLAD = chronic lung allograft dysfunction; RMST = restricted mean survival test.

Discussion

In this study, we evaluate the utility of the novel radiographic methodology of PRM applied to HRCT scans at the time of early spirometric decline. We demonstrate that PRM-based quantitative evaluation of parenchymal and functional small airway disease on HRCT scans at the time of potential CLAD (FEV₁ of 80–90% of baseline) can inform us regarding graft pathogenic changes and prognosis. The following three unique categories were identified on the basis of the predominant PRM signature: high PRM^PD (41%), high PRM^fSAD (11%), and PRM^Normal (48%). Significant patterns emerged on evaluation of CLAD-free survival, with 66% of the patients in either high PRM^fSAD or high PRM^PD groups demonstrating progression to CLAD or death within 1 year, increasing to 84% by 2 years. In contrast, only 31% of patients with a predominantly normal PRM pattern demonstrated progression to CLAD or death at 1 year. This significant predictive ability of abnormal PRM signature was maintained in multivariate analyses after accounting for other significant clinical variables such as time from transplant. Visual evaluation of HRCT scans at the time of potential CLAD revealed high prevalence of air trapping across all PRM groups, but no prognostic significance of this finding was noted. In contrast, GGOs were identified in only 16% of cases and were associated with decreased CLAD-free survival; PRM^PD signature together with visual GGO identification portended particularly poor prognosis. In summary, we demonstrate that a combined radiographic and spirometric early graft surveillance strategy can allow for the identification of patients at high risk of CLAD onset or death.

It is recognized that improving allograft survival and long-term outcomes after lung transplant requires that CLAD be identified in its early stages of pathogenesis. As spirometry has been the primary method of graft surveillance, a potential BOS stage (BOS 0-p), defined by a 10–19% decrease in FEV₁ from baseline, was added to the original BOS staging system in 2002 (7). We have previously published on the sensitivity, specificity, and positive predictive value of BOS 0-p and demonstrated that 56% of patients develop BOS within 1 year of meeting BOS 0-p criteria (6). Similar results were noted by Hachem and colleagues in a different cohort (25). Together, these studies suggested that persistent mild decline in lung function should trigger diagnostic workup and increased vigilance. The new ISHLT guidelines agree that the ⩾10% threshold falls outside the normal day-to-day variability of FEV₁ and should trigger closer monitoring and term it as “potential” CLAD (1). Although HRCT scans are increasingly used clinically at this stage, very few studies have investigated the prevalence of radiographic abnormalities and their prognostic significance. Furthermore, the advent of quantitative postprocessing techniques has further improved on the diagnostic and prognostic ability of HRCT scans in lung diseases (26–29), providing additional tools to evaluate graft health.

Our study provides the first evaluation of small airway disease at the time of early persistent spirometric decline or potential CLAD and demonstrates that PRM can help identify a subgroup of patients with already established fSAD and high risk of rapid progression. PRM methodology is unique in its ability to provide quantitative assessment of fSAD and has been extensively used and validated in patients with chronic obstructive pulmonary disease (9, 30, 31). Our previous work has established the application of PRM to the lung transplant population and found fSAD as assessed by PRM to be a robust predictor of survival in CLAD (10). Our novel evaluation of fSAD at the time of potential CLAD in the present study revealed that 11% (7/61) of patients already had established significant small airway disease with PRM^fSAD of ⩾30%, with all of them progressing to definitive CLAD or death by 2.1 years. The significance of this finding by machine-based analyses is augmented by the lack of prognostic ability of air trapping by radiologist assessment. Air trapping was common, noted in 66% of all HRCT scans, and showed no difference across the three PRM groups. Agreement regarding the presence or absence of air trapping between two experienced chest radiologists was poor. Furthermore, no difference was noted in CLAD-free survival in patients with and without air trapping.

An intriguing find was the high number of patients with evidence of infiltrative or PD, with 41% of HRCT scans at potential CLAD demonstrating PRM^PD ⩾30%. PRM^PD identifies the increase in Hounsfield density, which is not delineated by human eye. An example is shown in the CT and PRM images of one such patient (Figure 7). Significantly, patients with high PRM^PD demonstrated equally poor prognosis as patients with high fSAD, with 84% of patients developing CLAD or dying within 2 years. Although surprising, this finding has precedence in a recent investigation by Horie M and colleagues, in which quantitative lung density on low-dose CT scans in a similar population with early spirometric decline was investigated. Higher quantitative lung density was noted in patients who developed CLAD within 3 years compared with those patients who remained stable for at least 3.5 years (32). It is known that immune- and non–immune-mediated lung injury and tissue damage precedes the development of fibrosis in CLAD. Detection of inflammatory and early fibrogenic phases of CLAD that herald established progressive irreversible graft dysfunction is key so that treatment modalities can be aimed at these early stages. The predominant CLAD phenotype seen in patients with high PRM^PD was still obstructive, suggesting that PRM^PD could potentially be identifying such an infiltrative profibrotic process. The significance of pathological changes in the lung parenchyma was also noted in the visual CT analysis. GGO was noted to be more common in the high PRM^PD group and was the only visual radiographic feature that predicted CLAD-free survival. Furthermore, both in our study and in the study by Horie and colleagues, the presence of visually noted GGO was additive in improving prognostic ability of the quantitative measures. Future investigations into the underlying pathologic features that are associated with these radiographic findings can shed novel light on CLAD pathogenesis.

Figure 7.

Inspiratory and expiratory high-resolution computed tomography (CT) and parametric response map images at the time of potential CLAD in a 27-year-old patient who had undergone bilateral lung transplant 4 years prior. High-resolution CT results showed no significant findings, but parametric response mapping (PRM) analysis demonstrated high PRM^PD of 39%. CLAD = chronic lung allograft dysfunction; PD = parenchymal disease.

Our study is novel and impactful in that the ability to identify radiographic signatures and prognosticate, when mild spirometric changes are noted, can have significant impact on clinical care and outcomes of lung transplant recipients. The recent ISHLT guidelines recommend HRCT assessment at the time of potential CLAD, and PRM methodology can be applied easily to these HRCT inspiratory and expiratory images (1). PRM technology is also unique in that it has been used and validated in other pulmonary conditions and is commercially available (8, 9, 31). However, our present study, although it lays a foundation for the use of PRM in graft surveillance, suffers from limitations of being a single-center investigation and a retrospective study design. The thresholds identified by our study can be considered as markers of a window that should alert the clinician to perform further evaluation; thresholds need to be reevaluated and validated in a separate cohort. Future multicenter prospective studies using the time points and radiographic signatures identified by this present study will be important in further assessing the applicability of this radiographic biomarker in lung allograft monitoring. Another important future direction will be combining this PRM methodology with other physiological (oscillometry and exhaled nitrogen) and biological (cell-free DNA and mesenchymal colony-forming units) graft monitoring methodologies (33–39).

In summary, we demonstrate that PRM-based small airway and PD measurements in lung allografts allow for early CLAD detection and prognostication. Our study provides evidence for using combined radiographic and spirometric monitoring of lung transplant patients with the use of quantitative analytic methodologies at the time of early decline in lung function or potential CLAD. Future multicenter prospective studies are recommended to further assess the applicability of this radiographic biomarker in lung allograft monitoring.