The protocols of this study were approved by the institutional review board of Samsung Medical Center. Waivers of informed consent were obtained. This is a retrospective study of patients with a biopsy-confirmed fibrotic IIP who had honeycombing of less than 5% of lung volume on CT at the time of surgical lung biopsy. From 1996 to 2008, a total of 438 patients who were suspected of having an ILD underwent surgical lung biopsy at Samsung Medical Center, a tertiary referral center in Seoul, Korea. Patients who had received histologic diagnoses of cellular NSIP or other ILD were excluded; patients with a history of collagen vascular disease were also excluded. Data were extracted for those patients for whom both HRCT and pulmonary function test (PFT) data within 6 months of surgical lung biopsy were available. We enrolled patients who had received histologic diagnosis of fibrotic IIP with neither overt (i.e., ≥ 5% of whole lung volume) honeycombing nor acute illness (e.g., pneumonia or acute exacerbation of pulmonary fibrosis on HRCT) for data analysis; patients were selected by two radiologists, both of whom had more than 5 years of ILD HRCT interpretation experience and one of whom was guarantor of this entire study. Finally, patients were excluded if they met any of the following three reasons affecting prognosis: lost to follow-up evaluation during the follow-up period of less than 2 years, cancer concurrently associated with ILD, or previous lung transplantation. We included a total of 154 patients who did not meet any of these exclusion criteria in our study (Fig. 1).

To select fibrotic IIP cases in which honeycombing is less than 5% of whole lung volume, the two radiologists estimated independently the extent of honeycombing in each case of fibrotic IIP at five levels: aortic arch, azygos arch, bronchus intermedius, right inferior pulmonary vein, and liver dome. We included cases in which none of the five levels scanned contained honeycombing of 5% or more of the scanned lung area; in other words, we rejected cases in which either observer called honeycombing 5% at any scan level of lung.

Surgical lung biopsy was performed in consideration of HRCT findings via either open thoracotomy (n = 8) or video-assisted thoracoscopic surgery (n = 146). In all patients, biopsy specimens were obtained from at least two sites (two different lobes). Biopsy slides were reviewed by a lung pathologist who had 18 years of ILD interpretation experience and who read the slides according to the American Thoracic Society/European Respiratory Society consensus classification of interstitial pneumonia [8]. When the diagnosis was not evident, two additional pathologists at other hospitals were consulted and a final decision was made by consensus.

A chest physician (with 15 years of ILD management experience) reviewed the clinical data by referring to patient hospital records. The physician was blinded to the CT or pathologic results. Data included initial (n = 154) and follow-up (n = 117) PFT results, treatment regimens, and causes of death. Forced spirometry volume and single-breath DL_CO were determined by using pulmonary function units. Forced vital capacity (FVC) and DL_CO were expressed as a percentage of the predicted value based on height, age, sex, and ethnicity. Changes in PFT results were presented as the percentage change from the initial values. Improvements and deteriorations were defined as 10% or greater changes in FVC and 15% or greater changes in DL_CO [9]. Patients who did not belong to either the improvement or the deterioration group in their PFTs were considered to have stable disease. Survival period and cause of death were identified from medical records. If necessary, direct contact was made with patients’ families or national death registry data were accessed.

HRCT was performed using various CT scanners. Scans were obtained at end of inspiration and in supine position from the lung apices to the bases. Contrast medium was not injected in any patient. Data were reconstructed by using a high-spatial-frequency (bone) algorithm, with 1.0- to 1.5-mm section thickness and at 10-mm intervals for transverse images. All images were viewed with a PACS (PathSpeed, GE Healthcare) workstation at window settings appropriate for viewing the lung parenchyma (window level, -700 HU; window width, 1500 HU) and the mediastinum (window level, 50 HU; window width, 400 HU).

Three independent chest radiologists (with 19, 12, and 10 years of HRCT interpretation experience), who were unaware of clinical, PFT, or pathologic data, analyzed the CT images. First, these three reviewers analyzed initial HRCT scans and were blinded to the results of follow-up scans. Next, the reviewers viewed the initial and follow-up HRCT images simultaneously and in a side-by-side manner. When there were more than two follow-up CT examinations of a patient, the last follow-up CT examination was used for comparison. The interval between the initial and the last follow-up CT studies was at least 6 months. On both the initial and the follow-up CT scans, the observers made subjective assessments of the overall extent of lung parenchymal abnormalities as well as the extent of reticulation, honeycombing, and ground-glass opacification (GGO) by browsing and observing all HRCT images from the thoracic inlet to the lung base. A score, which was estimated to the nearest 5% of parenchymal involvement compared with whole lung volume (100%), was assigned to each parenchymal abnormality and was based on the percentage of lung parenchyma that exhibited evidence of lung abnormality. The total score was calculated by adding the scores of each lung parenchymal abnormality. Because three reviewers took part in the reading session, the final score for the total extent and for the extent of each parenchymal abnormality was calculated by the use of the mean of the scores from the three reviewers. GGO was considered to be present when images showed hazy increased lung attenuation but with preservation of bronchial and vascular margins. Reticulation was defined as innumerable interlacing lines, suggestive of mesh. Honeycombing was considered to be present when clustered cystic airspaces of 3-10 mm in diameter with shared well-defined walls were identified, usually located in the subpleural areas of the lungs [10].

The craniocaudal distribution of lung lesions was classified as upper-zone (above hilum) or lower-zone (below hilum) predominant or as random (both upper and lower zones). The transverse distribution was classified as subpleural (2 cm from the pleural surface), random (both subpleural and central lung involvement), or peribronchovascular (along the bronchovascular bundles). Inconsistencies among the individual reviewers were resolved by majority opinion.

Interobserver agreement among the three radiologists on the presence of each parenchymal abnormality, the total extent of lung lesions, and the distribution of lung abnormalities was determined via intraclass correlation coefficients. Group comparisons were made by using either the Mann-Whitney test or the Fisher exact test.

The interrelationship between the extent of initial CT abnormalities and PFT results (extent of each initial CT finding with the initial FVC or DL_CO) was evaluated by the use of the Spearman correlation coefficient test. Furthermore, the extent of initial CT abnormalities was correlated with the changes in PFT results using the same method.

A univariate logistic regression model was used to determine which variables contribute to the prediction of patient survival in terms of initial CT findings and PFT results and changes in CT findings and PFT results on follow-up studies. Univariate and multivariate Cox proportional hazards regression analysis was used to identify significant variables predictive of survival status. On multivariate analysis, all variables included in the univariate analysis were considered, and the analysis was performed using a forward stepwise procedure. Survival analysis was also conducted using the Kaplan-Meier method. Survival curves were compared via the log-rank test. The cutoff value of the overall extent of parenchymal abnormalities at baseline CT evaluation for different survival durations was determined via receiver operating characteristic curve analysis. A p value of less than 0.05 was regarded as statistically significant. Statistical analyses were conducted by using SPSS for Windows, version 15.0.

Among 154 patients, 101 had UIP and 53 had fibrotic NSIP. Significant differences were noted in the demographic features of age, sex, PFT results, and overall survival. Patients with UIP were significantly older than those with fibrotic NSIP (59 ± 7 vs 55 ± 9 years, respectively; p = 0.001) and were more likely to be male (male-to-female ratio, 52:49 vs 12:41, respectively; p < 0.001). The median duration of clinical follow-up for mortality was 5.2 years (range, 0.6-13.5 years). Overall survival was 5.1 ± 2.8 and 6.8 ± 3.4 years in UIP and fibrotic NSIP groups, respectively (p = 0.001). Serial CT scans obtained after a minimum follow-up period of 6 months were available for 132 patients (85 with UIP and 47 with fibrotic NSIP). The median interval between baseline and follow-up CT studies was 2.8 years (range, 0.5-13.1 years). No significant differences in the presence of follow-up CT studies were detected between UIP (85 of 101 patients [84%]) and fibrotic NSIP (47 of 53 patients [89%]) groups (p = 0.09). Of the 154 patients in both groups, only four received treatment before undergoing the initial CT study. During the follow-up period, 125 patients (81%) were treated with corticosteroid alone in 54 patients (35%), corticosteroid plus cytotoxic therapy in 44 patients (29%), azathioprine alone in 21 patients (14%), and cyclophosphamide alone in six patients (4%).

The CT findings of two groups of patients with IIP are shown in Tables 1, 2, 3. On the baseline CT study (Table 1), the patients with UIP had more fibrosis (reticulation plus honeycombing) and less GGO on HRCT (more reticulation and less GGO in UIP; p < 0.001) (Figs. 2A, 2B, 3A, and 3B). The lower lung zones were predominantly involved in all subgroups. On the axial plane, 62% of patients with UIP showed subpleural predominance, whereas 55% of patients with fibrotic NSIP exhibited random distribution. However, the transverse distribution of lung lesions was not statistically different between the UIP and fibrotic NSIP groups (p = 0.06).

Follow-up CT studies (Table 2 and Figs. 4A, 4B, and 4C) showed increases in the extent of reticulation (3% and 8% increase in extent in UIP and fibrotic NSIP, respectively, compared with baseline studies; p = 0.03), reductions in the extent of GGO (2% and 10% decrease in extent in UIP and fibrotic NSIP, respectively; p = 0.009), and progression of honeycombing (5% and 3% increase in extent in UIP and fibrotic NSIP, respectively; p = 0.08). Overall extent of lung lesions showed an increase (6%) in UIP and a decrease (-4%) in NSIP (p = 0.04). On serial CT, lung lesions showed distributional changes from subpleural to both subpleural and central (n = 13 [15%] in UIP, and n = 4 [9%] in fibrotic NSIP; p = 0.96) and from lower to both lower and upper zones (n = 14 [16%] in UIP, and n = 2 [4%] in fibrotic NSIP; p = 0.60).

Interreviewer agreement for most of the ILD-related CT findings was very good (intraclass correlation coefficients, 0.92 [0.90-0.94], 0.85 [0.81-0.88], 0.81 [0.76-0.85], 0.93 [0.91-0.94], 0.76 [0.70-0.81], and 0.74 [0.68-0.80] for total extent, reticulation, GGO, emphysema, axial direction, and craniocaudal direction, respectively), whereas the interreviewer agreement for honeycombing was good (intraclass correlation coefficient, 0.65 [0.55-0.73]).

At the initial PFTs, FVC values (percentage predicted) were 82% ± 17% and 74% ± 18% for UIP and fibrotic NSIP, respectively (p = 0.03), whereas DL_CO values (percentage predicted) were 77% ± 20% and 81% ± 21%, respectively (p = 0.02). In 117 patients, follow-up PFTs were available within 1 month of follow-up CT. Among 76 patients with UIP, 11 patients (14%) had improved PFT results, 36 patients (47%) had stable PFT results, and 29 patients (38%) had worsened PFT results. In contrast, 17 of 41 patients (41%) with fibrotic NSIP had improved PFT results, particularly in FVC values (Table 3).

In patients with UIP, moderate negative correlation was found between the extent of initial CT abnormalities and PFT results (overall extent on CT with initial FVC, and overall extent and reticulation on CT with initial DL_CO). In addition, GGO extent on the initial CT was significantly correlated with changes in FVC, and the initial overall extent was also significantly correlated with changes in DL_CO. Changes in overall CT extent showed moderate positive correlation with changes in FVC (Table 4).

In patients with fibrotic NSIP, overall extent on CT showed moderate negative correlation with the initial FVC value. Changes in overall CT extent showed moderate positive correlation with changes in FVC, and changes in reticulation extent also showed moderate positive correlation with changes in FVC (Table 4).

All 154 patients were followed until either the patient’s death or August 2010. A total of 26 patients died during a median follow-up period of 5.2 years, and the 2-year survival rate was 83%. On the basis of results of univariate logistic regression analysis for survival prediction (Table 5), the overall extent of lung parenchymal abnormalities (odds ratio, 1.1; p = 0.01) and reticulation (odds ratio, 1.0004; p = 0.01) on baseline CT and changes in overall extent (odds ratio, 1.08; p = 0.03) and the development of honeycombing (odds ratio, 1.002; p = 0.03) on serial CT were associated with patient survival prediction. No significant difference was identified in PFT results between survival and nonsurvival groups; however, there was greater reduction in FVC (odds ratio, 1.05; p < 0.01) in the nonsurvival group. On univariate and multivariate Cox proportional hazards analysis (Table 6), the only prognostic factor was the overall extent of parenchymal abnormalities on baseline CT. This turned out to be predictive of poor prognosis for survival.

Kaplan-Meier survival curves plotted for survival in terms of the overall extent of parenchymal abnormalities on baseline CT are shown in Figure 5. Patients with an overall extent of pulmonary fibrosis of up to 35% of the whole lung volume had significantly better survival rates than those with an overall pulmonary fibrosis extent of more than 35% of lung volume (median survival, 3.9 vs 0.8 years, respectively; p = 0.029).

The current study had several limitations. First, our study was retrospectively designed at a single tertiary referral center. Therefore, some selection bias may have existed. Additionally, the intervals between serial CT examinations and the total follow-up interval period were variable, and the treatment regimen and duration were not standardized. However, the median 5.2-year follow-up period in a cohort including 132 patients was sufficiently long for the analysis of survival data. Moreover, 81% (125 of 154) of patients were treated with a corticosteroid and a cytotoxic agent (cyclophosphamide).

Second, we focused principally on CT determinants of patient prognosis. There may have been other clinical variables, such as the 6-minute-walk test [15], that would have helped to predict patient survival. Despite these limitations, the current study included a large number of patients with clinicoradiologically and pathologically diagnosed fibrotic IIP. Further studies, particularly prospective ones, including a larger cohort and greater number of clinical predictors will be necessary for a broader generalization of our study results.

Third, in our study, CT scans were obtained with patients in the supine position. Therefore, accurate estimation of honeycombing may have been difficult because dependent density or a subpleural line may have closely simulated the appearance of early lung fibrosis (i.e., honeycombing).

Even in cases of fibrotic IIP with little honeycombing on baseline CT, honeycombing and reticulation progress and GGO shows decrease in their extent on follow-up CT. In univariate logistic regression analysis, the overall extent of lung parenchymal abnormalities and reticulation on baseline CT and changes in overall extent and the development of honeycombing on serial CT were associated with shorter patient survival. However, in multivariate analysis, only the overall extent of lung fibrosis on the baseline CT appears to help predict survival in fibrotic IIPs with little honeycombing.