UIP is the hallmark radiologic pattern of IPF. Honeycombing is a distinguishing feature of UIP and must be present for a definite HRCT diagnosis of UIP to be made. It can be seen with or without peripheral traction bronchiectasis or bronchiolectasis. The typical distribution of UIP is subpleural with basal predominance, although some upper lobe involvement is common; in some cases, the craniocaudal distribution of UIP may be relatively uniform (64, 65). Asymmetric disease may occur in up to 25% of cases (66). Several studies have demonstrated that the positive predictive value of a radiologic diagnosis of UIP on HRCT for a pathologic diagnosis of UIP is between 90% and 100% (67–71); however, a significant minority of patients with histopathologic UIP do not fulfill HRCT criteria for UIP (68, 70–72).

Mediastinal lymphadenopathy may be present in patients with UIP (73). Ground-glass opacification may be present, but it is not a dominant feature and is usually accompanied by a superimposed reticular pattern. Rarely, small ossified nodules within areas of fibrosis may be present, and these are more common (29%) in patients with UIP when compared with other fibrotic lung diseases (74). Patients with UIP may have features of pleuroparenchymal fibroelastosis at the lung apices (75, 76); however, there is no clear cut-off of the proportions of each pattern, and these cases should be regarded as UIP/IPF, if consistent with that diagnosis after MDD. UIP may present as an acute exacerbation (Figure 6) or coexist in patients with emphysema (Figure E1).

Figure 6. Acute exacerbation of idiopathic pulmonary fibrosis. (A and B) Transverse computed tomography sections obtained in the upper and mid lung zones and (C and D) during acute exacerbation showing newly developed, bilateral ground-glass opacification in both lungs on a background of usual interstitial pneumonia pattern.

Download Figure | Download Powerpoint

[More] [Minimize]

In the 2011 guideline, an HRCT pattern consisting of subpleural, basal-predominant reticular abnormalities without honeycombing was assigned the HRCT diagnosis category of “possible UIP” (2). Since 2011, several studies have reported that selected patients with a “possible UIP” pattern on HRCT according to the 2011 guidelines are highly likely to have histopathologic UIP, despite the absence of radiologic honeycombing (77). Specifically, an HRCT pattern of possible UIP with peripheral traction bronchiectasis or bronchiolectasis in the correct clinical setting likely represents histopathologic UIP on biopsy (65, 78–80). Therefore, subpleural, basal-predominant reticular abnormalities with peripheral traction bronchiectasis or bronchiolectasis should be regarded as “probable UIP.” As with a UIP pattern, ground-glass opacification may be present in probable UIP, but it is not a dominant feature. Many patients with an HRCT pattern of probable UIP will be determined to have IPF once other factors such as histopathology are considered.

It is now recognized that atypical HRCT features frequently (i.e., about 30%) accompany a histopathologic pattern of UIP/IPF (81). Therefore, the category “indeterminate for UIP pattern” should be assigned when HRCT demonstrates features of fibrosis but does not meet UIP or probable UIP criteria and does not explicitly suggest an alternative diagnosis. This category includes a subset of patients with very limited subpleural ground-glass opacification or reticulation without obvious CT features of fibrosis, for whom there is a suspicion that early UIP or probable UIP is present. In such cases, it should be confirmed with prone inspiratory views that the subpleural opacities do not represent dependent atelectasis (Figure E2).

In some cases of fibrotic lung disease, there is clinical suspicion of IPF, but the HRCT pattern suggests an alternative diagnosis. Examples include bronchocentric fibrosis in the upper lobes or profuse mosaic attenuation that suggest hypersensitivity pneumonitis, posterior fibrotic retraction of the hila in sarcoidosis, or extensive ground-glass opacification with subpleural sparing in fibrotic nonspecific interstitial pneumonia (NSIP). Occasionally, the HRCT presentation may be that of a UIP, probable UIP, or indeterminate for UIP pattern, but ancillary findings suggest an alternative diagnosis. In such situations, an alternative diagnosis to IPF should be reconsidered.

Patients with an acute exacerbation of IPF have bilateral ground-glass opacification with or without consolidation on a background of lung fibrosis (Figure 6). In the absence of a previous HRCT study, bilateral ground-glass opacity and/or consolidation on a background of a UIP pattern is highly suggestive of an acute exacerbation and can be used to confirm an underlying IPF diagnosis in the appropriate clinical context.

The guideline panel recognized there is no reasonable alternative to the proposed course of action, so a motherhood statement was made to take a detailed history of medication use and environmental exposures at home, work, and other places that the patient frequently visits, to identify or exclude potential causes of ILD (e.g., hypersensitivity pneumonitis, pneumoconiosis, drug toxicity). This is supported by an observational study that enrolled 1,084 patients with new-onset ILD of unknown cause reporting that 47% of the patients were identified as having hypersensitivity pneumonitis on detailed assessment, suggesting that a cause can be found in many patients who present with ILD (84). The panel’s clinical experience is that identification and removal of potential causative environmental factors may result in improved clinical outcomes.

Many panelists use published questionnaires in their clinical practices to consider environmental exposures at home, work, and frequently visited places (84–86). Such questionnaires may be tailored to cultural habits and geographical differences. Examples of pertinent exposures include mold, birds, down feathers, animals, metal dusts (e.g., brass, lead, steel), wood dust (e.g., pine), vegetable dust, exposure to livestock, stone polishing and cutting, medications taken, current or recent occupations (e.g., hair dressing), and current or recent hobbies (27, 87–92). Although some panelists use the presence of antibody in serum against specific antigen to prompt further evaluation for hypersensitivity pneumonitis, the test is not standardized and the specificity and sensitivity for the diagnosis of hypersensitivity pneumonitis is unknown. The panelists who use serum antibody testing believe that such tests may identify an antigen that was not suspected by clinical history and, therefore, may prompt further investigations for the suspected etiology; also, if serum antibody testing is negative, the results reinforce the conclusion that the patient does not have hypersensitivity pneumonitis.

Diagnosis of IPF mandates exclusion of other causes of ILD, including CTD-ILD (Table E2). The guideline panel concluded that foregoing serological testing was not a reasonable alternative. Therefore, a motherhood statement was made to perform routine serological testing in all patients with newly identified ILD. Although there was overwhelming agreement to perform serological testing, there was far less agreement about which serological tests to perform.

The majority of panelists acknowledged routinely testing for CRP (C-reactive protein), erythrocyte sedimentation rate, antinuclear antibodies (by immunofluorescence), rheumatoid factor, myositis panel, and anti–cyclic citrullinated peptide. Other detailed tests are performed on a case-by-case basis according to associated symptoms and signs. These include muscle enzymes (creatinine phosphokinase, myoglobin, and aldolase), antisynthetase antibodies (Jo-1 and others if available), anti–MDA5 (melanoma differentiation-associated protein 5), anti–Mi-2, anti-NXP2 (nuclear matrix protein 2), anti–TIF1-γ (transcriptional intermediary factor 1-γ), anti-SRP (signal recognition particle), anti-HMGCR (3-hydroxy-3-methylglutaryl-CoA reductase), anti-SAE (small ubiquitin-related modifier–activating enzyme), anti-U1RNP (U1 ribonucleoprotein), anti-PM/Scl75 (polymyositis/scleroderma 75), anti-PM/Scl100, and anti-Ku (93). If systemic sclerosis (i.e., scleroderma) is suspected, additional tests include: anti–Scl-70/topoisomerase-1, anti-centromere, anti-RNA polymerase III, anti-U1RNP, anti-Th/To, anti-PMScl, U3 RNP (fibrillarin), and anti-Ku. If Sjögren syndrome is suspected, additional tests include: anti-SSA/Ro (Sjögren-specific antibody A) and anti-SSB/La. If vasculitis is suspected, an additional test includes anti-cytoplasmic antibodies. A small minority of the panelists include all of the detailed tests listed above as an “ILD panel” at initial screening/baseline evaluation.

The guideline panelists do not refer all patients with new ILD to a rheumatologist; rather, referring only those with positive clinical manifestations, serologies, or other characteristics atypical for IPF (e.g., female, age <60 yr). In many CTD-ILDs, the lung disease is the first, dominant, or only feature of the CTD and, therefore, some patients will not fit standard rheumatologic diagnostic criteria at presentation. The term “interstitial pneumonia with autoimmune features” has been suggested to describe such patients; however, this is a research definition that requires validation (94).

Our systematic literature search yielded 2,492 titles but did not identify any studies that 1) compared clinical outcomes among patients who underwent BAL cellular analysis to those who did not undergo BAL cellular analysis, or 2) reported the test characteristics of BAL cellular analysis for distinguishing IPF from other ILDs. Therefore, we sought studies that compared BAL cell type proportions across different ILDs. The full text of 14 articles was reviewed, and eight were selected for analysis (95–102) (Table E7a).

Studies enrolled patients with IPF, performed BAL, and measured components of the BAL fluid including the percentage of neutrophils (95–100, 102), macrophages (95–99, 102), lymphocytes (95–102), and eosinophils (95, 97–100, 102), as well as the CD4/CD8 ratio (95, 97, 99, 100). The measurements were then compared with those from patients with other types of ILD, including hypersensitivity pneumonitis (95, 96, 100), sarcoidosis (95, 99, 100), idiopathic NSIP (95, 97, 100–102), cryptogenic organizing pneumonia (COP, previously called bronchiolitis obliterans organizing pneumonia) (95–97, 100), eosinophilic pneumonia (95), respiratory bronchiolitis-associated ILD (RB-ILD) (96), and lymphocytic interstitial pneumonia (LIP) (96). Most studies reported mean cell type proportions, but some reported medians. Because the question was about using BAL to distinguish IPF from other types of ILD, we compared cell type measurements among patients with IPF to those among patients with other types of ILD using the mean difference (MD):

The guideline panel had very low confidence in the estimated differences in the BAL fluid cellular composition of patients with IPF compared with patients with other ILDs. Its confidence was particularly diminished by the small number of studies and patients per study, as well as the wide range of proportions for each cell type across studies. In addition, the evidence was indirect (the question is about patients with ILD of unknown cause, but the patients studied all had confirmed causes of ILD), and there was a risk of selection bias due to lack of consecutive enrollment and detection bias due to measurements being performed in different laboratories and samples being obtained from different bronchopulmonary segments. The panel noted that many of the statistically significant differences were small and probably not clinically important.

Some BAL cell type proportions were markedly (>10%) different in patients with IPF compared with patients with other ILDs (Figure E3). Patients with IPF had a slightly increased proportion of eosinophils compared with healthy patients but a markedly lower proportion of eosinophils than patients with eosinophilic pneumonia; thus, patients with a markedly elevated eosinophil count are more likely to have eosinophilic pneumonia than IPF. Patients with IPF had a similar to slightly higher proportion of lymphocytes and CD4/CD8 ratio in their BAL than healthy patients but a markedly lower proportion of lymphocytes and CD4/CD8 ratio in their BAL than patients with sarcoidosis; thus, patients with a markedly elevated proportion of lymphocytes and CD4/CD8 ratio are more likely to have sarcoidosis than IPF.

Cellular analysis of the BAL fluid may help distinguish IPF from some alternative ILDs, most notably eosinophilic pneumonia and sarcoidosis.

Although none of the studies reported any complication from the BAL procedure, bronchoscopy is an invasive procedure that requires time and effort, has some risk of complications, and is uncomfortable to some patients.

Despite having very low confidence in the estimated effects described above, the panel believed that some of the differences were large enough that they were likely true, and those differences were consistent with the panel’s collective clinical experience managing thousands of such patients. When the panel weighed the desirable consequences of BAL cellular analysis in patients who have an HRCT pattern of probable UIP, indeterminate for UIP, or an alternative diagnosis (i.e., identifying or excluding eosinophilic pneumonia, sarcoidosis, infection, and malignancy) versus the undesirable consequences (i.e., risk of a complication, burden, cost), the majority of the panel concluded that the upsides of the procedure outweigh the downsides in such patients. There were some strong dissenting opinions, but there was general agreement that BAL is appropriate when the radiologic differential diagnosis includes eosinophilic pneumonia, sarcoidosis, or infection. In contrast, the panel concluded that alternative diagnoses that can be excluded by BAL cellular analysis are sufficiently rare in patients who have an HRCT pattern of UIP that the downsides of the procedure typically outweigh the upsides in these patients.

Our systematic literature search yielded 945 titles but identified no studies that compared clinical outcomes among patients who underwent SLB to those who did not. Thus, we selected studies that measured diagnostic yield of SLB using an MDD as the diagnostic decision-maker. The full text of 54 articles was reviewed, and 26 were selected for analysis (103–128) (Table E8). All studies enrolled patients with ILD of unknown cause and did not exclude those with an HRCT pattern of UIP.

Pooling studies (unweighted) indicated that SLB obtained an adequate sample in all patients (11 studies; 918 of 918, 100%; 95% CI, 99–100%), although the panel acknowledged that this is not always the case in clinical practice. The proportion of SLB that resulted in a specific diagnosis (i.e., the diagnostic yield) was high (26 studies; 2,338 of 2,651, 88.2%; 95% CI, 86.9–89.4%), with a minority being deemed unclassifiable (26 studies; 313 of 2,651, 11.8%; 95% CI, 10.6–13.1%). Among final diagnoses, approximately one-third were IPF (24 studies; 752 of 2,360, 31.9%; 95% CI, 30.0–33.8%), and many others were potentially treatable etiologies like infection, sarcoidosis, hypersensitivity pneumonitis, eosinophilic pneumonia, lymphangioleiomyomatosis, COP, and vasculitis.

Overall mortality was low (23 studies; 79 of 2,268, 3.5%; 95% CI, 2.8–4.3%), but some of the deaths were probably disease related, because procedure-related mortality was lower (6 studies; 7 of 410, 1.7%; 95% CI, 0.8–3.5%). Many series reported no mortality, suggesting that lower procedural mortality is possible depending on center-specific variables such as patient selection. Additional complications included exacerbations (15 studies; 116 of 1,891, 6.1%; 95% CI, 5.1–7.3%), bleeding (7 studies; 6 of 756, 0.8%; 95% CI, 0.4–1.7%), severe bleeding (4 studies; 1 of 461, 0.2%; 95% CI, 0.04–1.2%), prolonged air leak (13 studies; 90 of 1,527, 5.9%; 95% CI, 4.8–7.2%), respiratory infection (9 studies; 32 of 496, 6.5%; 95% CI, 4.6–9.0%), neuropathic pain (1 study; 3 of 66, 4.5%; 95% CI, 1.6–12.5%), and delayed wound healing (4 studies; 14 of 430, 3.3%; 95% CI, 2.0–5.4%).

The guideline panel had very low confidence in the estimated effects of SLB on patient-important outcomes because of the study designs, inconsistent magnitudes of effect, and risk of selection bias due to lack of consecutive enrollment.

Treatment of IPF was estimated in a prior iteration of this guideline to reduce 1-year mortality from roughly 8% to 5.5% and to increase the likelihood of slowed disease progression from 60.1% to 68% (129). These estimates, combined with those described above, suggest that for every 1,000 SLBs that are performed, 1,000 adequate specimens will be obtained, 882 specific diagnoses will be made, and 319 patients will be determined to have IPF. Assuming that all receive therapy, then for every 1,000 patients who undergo SLB, 1-year mortality will be reduced from 26 to 18 patients, and disease progression will be slowed in 217 patients instead of 192 patients. In addition, many patients will be determined to have an alternative cause of ILD that is potentially treatable. However, 17 patients will die as a result of the procedure, 61 patients will experience a perioperative exacerbation, and 65 patients will acquire a perioperative respiratory infection.

SLB obtains adequate specimens from 100% (95% CI, 99–100%) of patients, from which a definitive diagnosis can be made in 89% (95% CI, 88–90%). Potentially effective therapies exist for the roughly 30% of patients who will be diagnosed with IPF, as well as for many of the patients who will be determined to have an alternative diagnosis.

SLB has potential complications, the most important of which is procedural mortality (1.7%; 95% CI, 0.8–3.5%) and the most common of which is respiratory infection (6.5%; 95% CI, 4.6–9.0%). Others include exacerbations, bleeding, prolonged air leak, neuropathic pain, and delayed wound healing, the most common of which occur in 6% of patients.

When the desirable consequences were weighed against the undesirable consequences, the guideline panel concluded that the upsides of SLB outweigh the downsides for most patients with newly detected ILD of uncertain etiology whose HRCT pattern is probable UIP, indeterminate for UIP, or an alternative diagnosis. The conclusion was strengthened by the panel’s opinion that making a diagnosis provides additional unquantified benefits, such as more accurate estimates of prognosis, cessation of additional diagnostic testing, and the initiation of more specific treatment. However, it was mitigated by the panel’s low confidence in the estimated effects of SLB. The panel emphasized that the decision to perform SLB should be made in the context of an MDD by experienced clinicians.

The opposite was true among patients whose HRCT pattern is UIP, for whom the panel was certain that the downsides of SLB outweigh the upsides. Because the likelihood of finding an etiology other than UIP is small in such patients, SLB is best considered confirmatory and, therefore, was judged by the panel to not be worth the risk of complications.

Our systematic literature search yielded 945 titles but identified no studies that compared clinical outcomes among patients who underwent TBBx to those who did not. Thus, we selected studies that measured diagnostic yield of TBBx using an MDD as the diagnostic decision-maker. The full text of 16 articles was reviewed, and 7 were selected for analysis (128, 130–135) (Table E9). The studies enrolled patients with ILD of unknown cause and did not exclude those with an HRCT pattern of UIP.

Pooling studies (unweighted) indicated that TBBx obtained an adequate sample in roughly three-fourths of cases (five studies; 640 of 825, 77.6%; 95% CI, 74.6–80.3%). Among the adequate samples, a specific diagnosis was obtained from roughly half (seven studies; 409 of 948, 43.1%; 95% CI, 40.0–46.3%), with a slight majority deemed unclassifiable (seven studies; 539 of 948, 56.9%; 95% CI, 53.7–60.0%). Among all TBBx, only one-third yielded a specific diagnosis (i.e., the diagnostic yield) (seven studies; 409 of 1,133, 36.1%; 95% CI, 33.4–38.9%); however, it should be noted that there is uncertainty whether these specific diagnoses were actually correct, because the small samples are susceptible to sampling error and reduced ability to detect scattered histological features such as granulomas. There were no procedure-related deaths (one study; 0 of 49, 0%; 95% CI, 0–7.3%), with other complications including pneumothorax (one study; 5 of 49, 10.2%; 95% CI, 4.4–21.8%) and prolonged air leak (one study; 3 of 49, 6.1%; 95% CI, 2.1–16.5%).

The guideline panel had very low confidence in the estimated effects of TBBx on patient-important outcomes because TBBx was not compared with SLB within the same population, there was inconsistency in the magnitude of effect, only one study reported complications with a small event rate in that study, and there was a risk of selection bias due to lack of consecutive enrollment.

For every 1,000 TBBx performed, 780 adequate specimens will be obtained and 360 diagnoses will be made (i.e., SLB avoided). The remaining 640 patients will be undiagnosed after TBBx, many of whom will proceed to undergo SLB. No patients will die, but 102 will obtain a pneumothorax, with 61 having a prolonged air leak.

TBBx obtains adequate specimens from 77.6% (95% CI, 74.6–80.3%) of patients, from which a definitive diagnosis can be made and SLB avoided in 36.1% (95% CI, 33.4–38.9%).

Approximately 64% (95% CI, 61–67%) of patients will remain undiagnosed after TBBx.

The panel believed that a major limitation of the evidence was that the studies did not stratify patients according to HRCT pattern. It was argued that patients whose HRCT pattern is probable UIP, indeterminate for UIP, or an alternative diagnosis are significantly more likely to have an etiology detectable by TBBx (e.g., sarcoidosis) than patients with an HRCT pattern of UIP. Thus, if patients had been stratified according to their HRCT pattern, the diagnostic yield and number of SLBs avoided would probably have been higher among those with an HRCT pattern of probable UIP, indeterminate for UIP, or an alternative diagnosis and lower among those with an HRCT pattern of UIP.

There was no consensus on whether avoiding 360 SLBs outweighed 640 patients remaining undiagnosed and having to undergo a second diagnostic procedure. As a result, there was no agreement about whether patients with an HRCT pattern of probable UIP, indeterminate for UIP, or an alternative diagnosis should routinely undergo TBBx. The panel made no recommend for or against TBBx as an alternative to SLB, meaning that until additional evidence becomes available, TBBx should be considered on a case-by-case basis. There was strong agreement that patients with an HRCT pattern of UIP should not undergo TBBx, because the likelihood of finding an etiology other than UIP is small and not worth the risk of complications in such patients.

Machine learning using molecular signatures is being developed to make a molecular diagnosis of UIP in TBBx specimens but is not yet available in routine clinical practice. The guideline panel acknowledges that recent studies about the utility of molecular diagnostic tools that involve machine learning using TBBx samples are promising (136, 137); further studies to validate this are pending. This recommendation will be revisited in future iterations of this guideline as related evidence accumulates.

Our systematic literature search yielded 945 titles but identified no studies that compared clinical outcomes among patients who underwent lung cryobiopsy to those who did not. Thus, we selected studies that measured diagnostic yield of lung cryobiopsy using an MDD as the diagnostic decision-maker. The full text of 25 articles was reviewed, and 13 were selected for analysis (126, 127, 132, 135–146) (Table E10). The studies enrolled patients with ILD of unknown cause and did not exclude those with an HRCT pattern of UIP.

Pooling studies (unweighted) indicated that lung cryobiopsy obtained an adequate sample in the vast majority of cases (10 studies; 720 of 749, 96%; 95% CI, 94–97%). Among the adequate samples, a specific diagnosis was obtained in more than four-fifths of cases (13 studies; 692 of 833, 83%; 95% CI, 80–85%), with the remaining deemed unclassifiable (13 studies; 141 of 833, 17%; 95% CI, 15–20%). Among lung cryobiopsy procedures, the majority yielded a specific diagnosis (i.e., the diagnostic yield) (13 studies; 692 of 862, 80%; 95% CI, 77–83%).

Overall mortality was low (seven studies; 15 of 597, 2.7%; 95% CI, 1.7–4.3%), but some deaths were likely disease related, because procedure-related mortality was even lower (three studies; 1 of 427, 0.2%; 95% CI, 0.04–1.3%). Additional complications included exacerbations (three studies; 1 of 82, 1.2%; 95% CI, 0.2–6.6%), bleeding (six studies; 28 of 541, 5.2%; 95% CI, 3.6–7.4%), severe bleeding (eight studies; 5 of 674, 0.7%; 95% CI, 0.3–1.7%), prolonged air leak (two studies; 47 of 352, 13.4%; 95% CI, 10.2–17.3%), and respiratory infection (three studies; 3 of 409, 0.7%; 95% CI, 0.2–2.1%).

The guideline panel had very low confidence in the estimated effects of lung cryobiopsy on patient-important outcomes because lung cryobiopsy was not compared with SLB within the same population, there was inconsistency in the magnitude of effect, the complication event rate was low, and there was a risk of selection bias due to lack of consecutive enrollment.

For every 1,000 lung cryobiopsies performed, 950 adequate specimens are obtained, and 790 diagnoses are made (i.e., SLB avoided). This means that 210 patients will remain undiagnosed after lung cryobiopsy, many of whom will proceed to SLB. Two patients will die from the procedure, and 12 patients will experience an exacerbation.

Lung cryobiopsy obtains adequate specimens from 96% (95% CI, 94–97%) of patients, from which a definitive diagnosis can be made and SLB avoided in 80% (95% CI, 77–83%). Compared with SLB, lung cryobiopsy is associated with fewer respiratory infections and a trend toward less procedural mortality.

Roughly 20% (95% CI, 17–23%) of patients will remain undiagnosed after lung cryobiopsy. Compared with SLB, patients who undergo lung cryobiopsy are more likely to have bleeding or a prolonged air leak.

Although the panel was enthusiastic about the desirable consequences of lung cryobiopsy, this was offset by concern about the lack of standardized procedure and approach and the heterogeneous rates of adverse events noted in previous studies (147–149). The panel identified many questions that need to be answered before recommending widespread use of cryobiopsy, including: How many specimens should be obtained to optimize diagnostic yield while minimizing complications? From which portion of the lung should they be obtained in relation to the microanatomy of the lung and diseased lung tissues? For how long should the probe be cooled?

The panel concluded that it is reasonable for experienced centers and experts with a track record of performing the procedure safely to continue performing lung cryobiopsy in patients whose HRCT pattern is probable UIP, indeterminate for UIP, or an alternative diagnosis. However, the panel believed very strongly and recommends that such experts work toward developing a standardized procedure that optimizes the balance between diagnostic yield and complications. Those who have not yet begun to perform cryobiopsy should wait until the procedure has been standardized before implementing this into clinical practice. In patients whose HRCT pattern is UIP, the panel believed that the downsides of lung cryobiopsy outweigh the upsides. Because the likelihood of finding an etiology other than UIP is small, lung cryobiopsy is best considered a confirmatory test and, therefore, was judged by the panel to not be worth the risk of complications.

Our systematic literature search yielded 189 titles but identified no studies that 1) compared clinical outcomes among patients who underwent single-discipline decision-making (SDD; either a single clinician or a group of clinicians from the same discipline) to those who underwent MDD, or 2) reported the test characteristics of SDD using MDD as the reference standard. Therefore, we sought studies that measured agreement between SDD and MDD. The full text of 17 articles was reviewed, and 5 were selected for analysis (84, 150–153) (Table E11). Numerous studies measured agreement among individuals, but they were not selected for our analysis because they did not specifically compare SDD to MDD.

One study enrolled patients with an SDD diagnosis of IPF (84), and four studies enrolled patients with an SDD diagnosis of various types of ILD, including IPF (150–153). The studies subjected the patients to MDD and then compared the SDD diagnosis to the MDD diagnosis. In three studies, the SDD consisted of decision-making by a single respiratory clinician (84, 150, 151), in one study it consisted of either a single respiratory clinician or a single internist (152), and in one study it consisted of a group of pathologists (153). In three studies, the MDD consisted of decision-making by a respiratory clinician, radiologist, and pathologist (84, 150, 152); in one study it consisted of a radiologist and pathologist (84); and in one study it consisted of a respiratory clinician and a pathologist (153).

When measured as a proportion, median agreement between SDD and MDD was 70%, with a range from 47% to 87%. When measured using a Cohen’s kappa score, agreement was moderate (κ = 0.331; 95% CI, 0.269–0.392). The guideline panel had very low confidence in this estimated agreement. Its confidence was diminished by the risk of bias conferred by not enrolling patients with true diagnostic uncertainty, not consecutively enrolling patients, the inconsistency of the estimates, the small study sizes, and possible indirectness (the question pertains to those for whom there is a suspicion of IPF, but this was not reported in the studies).

For every 1,000 patients who undergo diagnostic decision-making, SDD and MDD will derive the same diagnosis in 700 patients and different diagnoses in 300 patients. If one accepts MDD as the reference standard, then as many as 300 patients will be potentially subject to incorrect therapy, delayed therapy, or unnecessary additional diagnostic testing.

SDD is more efficient for decision-making, given the increased time and effort required to obtain the opinion of colleagues in an MDD.

If one accepts MDD as the reference standard for diagnostic decision-making, SDD demonstrated suboptimal agreement (median, 70%; range, 47–87%).

The guideline panel agreed that MDD is preferred, because the notion that as many as 300 patients may be subject to incorrect therapy, delayed therapy, or unnecessary additional diagnostic testing was deemed unacceptable. The panel believes the benefit of MDD is greatest when the HRCT pattern is probable UIP, indeterminate for UIP, or an alternative diagnosis, or when there exist discordant clinical, radiologic, and/or histologic data. There was substantial discussion on what MDD entails. Until further research is done to optimize MDD, the panel concluded that it consists of an interaction between a pulmonologist (and rheumatologist on a case-by-case basis), radiologist, and pathologist. The modus operandi of the interaction is deferred to the confronted clinicians and could be face-to-face, by telephone, Internet/e-mail, text, and/or reading interpreted reports by experts via copies (printed, scanned, faxed). Face-to-face or voice-to-voice MDDs are encouraged when the formal clinical reports of the interpretation by experts in different disciplines are in discordance.

Our systematic literature search yielded 429 articles but identified no studies that compared clinical outcomes among patients who underwent specific serum biomarker measurements to those who did not. The literature review on diagnostic accuracy studies was limited to four specific serum biomarkers on the basis of input from the committee.

MMP-7. We selected studies that measured the diagnostic accuracy of MMP-7 for distinguishing IPF from other types of ILD. We reviewed the full text of 12 articles and selected two studies (154, 155). One study evaluated the ability of serum MMP-7 to distinguish IPF from a heterogeneous mixture of alternative ILDs (154), and the other looked at the ability of serum MMP-7 to distinguish IPF from sarcoidosis, idiopathic NSIP, hypersensitivity pneumonitis, CTD-ILD, and drug-induced ILD (155). Serum MMP-7 levels distinguished IPF from other ILDs with a median sensitivity, specificity, accuracy, and diagnostic odds ratio of 71.7% (range, 71–72.3%), 64.4% (63–66.3%), 68.4% (68.3–68.5%), and 4.7 (4.2–5.1), respectively (Table E12a).

SPD. We selected studies that measured the diagnostic accuracy of SPD for distinguishing IPF from other types of ILD. We reviewed the full text of 16 articles and selected one study (154). The study evaluated the ability of serum SPD to distinguish IPF from a heterogeneous mixture of alternative ILDs. Serum SPD levels distinguished IPF from other ILDs with a sensitivity, specificity, accuracy, and diagnostic odds ratio of 70.0%, 65.0%, 68.5%, and 3.1, respectively (Table E12b).

CCL-18. We sought studies that reported the diagnostic accuracy of CCL-18 for distinguishing IPF from other types of ILD. We reviewed the full text of six articles and selected no studies.

KL-6. We sought studies that reported the diagnostic accuracy of KL-6 for distinguishing IPF from other types of ILD. We reviewed the full text of 55 articles and selected no studies.

The guideline panel had very low confidence in the estimated effects. Its confidence was diminished by the risk of bias conferred by not describing the reference standard, not stating whether the enrolled patients had true diagnostic uncertainty, and not consecutively enrolling patients. Moreover, the studies were small, and none of the studies were performed using Clinical Laboratory Improvement Amendments–approved assays.

Assuming that 30% of patients with ILD have IPF, then for every 1,000 patients who undergo serum MMP-7 measurement for the purpose of distinguishing IPF from other ILDs, 672 patients will get a true-positive or true-negative result, whereas 338 patients will get a false result potentially leading to inappropriate therapy, delayed therapy, or unnecessary additional diagnostic testing. Similarly, for every 1,000 patients who undergo serum SPD measurement for the purpose of distinguishing IPF from other ILDs, 665 patients will get a true-positive or true-negative result, whereas 335 patients will get a false result potentially leading to inappropriate therapy, delayed therapy, or unnecessary additional diagnostic testing.

More than one-half of patients with ILD who undergo serum MMP-7 or SPD measurement will be correctly distinguished as having IPF or an alternative ILD. Samples for serum testing are easily obtained with few complications.

More than one-third of results will be incorrect, leading to inappropriate therapy, delayed therapy, or unnecessary additional diagnostic testing, all of which may be associated with complications. In addition, testing for these biomarkers is costly and not widely available.

For the time being, the guideline panel dismissed serum biomarker measurement as an approach to distinguishing IPF from other ILDs because of the high false-positive and false-negative result rates.