Exploratory analysis of a phase III trial of pirfenidone identifies a subpopulation of patients with idiopathic pulmonary fibrosis as benefiting from treatment

Idiopathic pulmonary fibrosis (IPF) is a fatal, progressive fibrotic lung disease with a median survival of 3-5 years and no proven effective therapy to date [1, 2]. Pirfenidone (5-methyl-1-phenyl-2-[1H]-pyridone; Shionogi & Co., Ltd., Osaka, Japan; Marnac Inc., Dallas, TX, USA) [3–7] is an antifibrotic drug for IPF which has combined anti-inflammatory, antioxidant, and antifibrotic effects in experimental models of pulmonary fibrosis [8–12]. A randomized, double-blind, placebo-controlled phase II trial of pirfenidone with 107 Japanese IPF patients demonstrated that pirfenidone significantly reduced the decline in vital capacity (VC) at week 36 compared to placebo (p = 0.037) [7]. These encouraging results prompted us to undertake a phase III clinical trial with 275 Japanese patients. In the Phase III trial, pirfenidone showed significant reduction in the decline of VC at week 52 (p = 0.042) and improved progression-free survival (PFS) time (p = 0.028) [13]. These Phase II and III data led to regulatory approval of pirfenidone in Japan for the treatment of IPF in 2008.

The Phase II trial in Japan also led to two larger, international, randomized trials of pirfenidone for IPF (CAPACITY, study 004 and study 006 with 435 and 344 IPF patients, respectively) [14]. In study 004, the decline in percentage predicted forced vital capacity (FVC) at week 72 was significantly reduced in pirfenidone-treated patients compared to those treated with placebo (p = 0.001). In study 006, the difference in FVC change at week 72 was not significant. However, the decline in %FVC was reduced at all time points during the first year. An analysis of pooled data from the two studies supported the treatment effect of pirfenidone on the %FVC, PFS time, and 6-minute walk test (6MWT) distance. In February 2011, pirfenidone was granted marketing authorization by the European Commission for the treatment of IPF.

Extended analyses of the phase III trial data in Japan revealed a subpopulation of patients who benefited from pirfenidone. Ebina et al. [15, 16] examined the association between pirfenidone efficacy (changes in VC at week 52) and the baseline %VC, and reported that pirfenidone was more effective in patients with relatively mild impairment of lung function (%VC ≥70).

To clarify more precisely which patients specifically benefit from pirfenidone, we examined the association between pirfenidone efficacy and the baseline lung functions including %VC, arterial oxygen partial pressure (PaO₂), and the lowest oxygen saturation in the 6-minute steady-state exercise test (the lowest SpO₂). In each subgroup, the change in VC, PFS time, and subjective symptoms (cough and dyspnea on the Fletcher, Hugh-Jones [F, H-J] Classification scale) were evaluated after high-dose (1800 mg/day) pirfenidone, low-dose (1200 mg/day) pirfenidone, and placebo treatment for 52 weeks.

The phase III trial showed that pirfenidone reduced the decline in VC at week 52 in IPF patients, and significantly prolonged the PFS time, compared to placebo [13]. In this exploratory analysis, patients were grouped by baseline %VC or PaO₂ at rest and the lowest SpO₂ to identify the subpopulations that benefited most from pirfenidone treatment. Specifically, patients were stratified on the basis of %VC (<70 vs ≥70), PaO₂ (<70 vs ≥70), and the lowest SpO₂ (<90 vs ≥90).

The changes in VC and %VC in subpopulations stratified by baseline %VC, PaO₂, and the lowest SpO₂

When patients were stratified by baseline %VC (<70%, ≥70%) and the lowest SpO₂ (<90, ≥90), pirfenidone tended to be more effective in patients with baseline %VC ≥70 and the lowest SpO₂ <90 (Subgroup A) than in the other subgroups. Namely, in Subgroup A, mean declines of VC and %VC at week 52 (all p-values <0.1; Table 1) and the distribution of progression-free survival times (data not shown) were significantly different between those treated with pirfenidone (high-dose, low-dose, and high+low dose) and those treated with placebo. When patients were stratified by PaO₂ (<70, ≥70) and by SpO₂ on exertion (<90, ≥ 90), pirfenidone tended to be more effective in patients with PaO₂ ≥70 and SpO₂ on exertion <90 (Subgroup B) than in the other subgroups. P-values of the comparisons of mean declines in VC (%VC) between the treatment groups (pirfenidone high-dose, low-dose, and high+low dose) and the placebo group were 0.151 (0.198), 0.088 (0.097) and 0.059 (0.074), respectively (Table 2). In Subgroup B, a similar trend was seen in progression-free survival time (data not shown).

Item	Category 1 SpO2	Category 2 Baseline %VC	High-dose Group		Low-dose Group		Placebo Group		P-value
			LS mean (n)	SE	LS mean (n)	SE	LS mean (n)	SE	H vs P	L vs P	H+L vs P
Change in VC	6MWT SpO2≥90	%VC≥70	-0.072 (35)	0.033	-0.004 (16)	0.050	-0.090 (32)	0.035	0.7035	0.1595	0.2581
		%VC<70	-0.263 (9)	0.073	-0.185 (6)	0.090	-0.225 (9)	0.073	0.7181	0.7350	0.9919
	6MWT SpO2<90	%VC≥70	-0.050 (36)	0.049	-0.016 (17)	0.071	-0.199 (36)	0.049	0.0359	0.0372	0.0131
		%VC<70	-0.148 (23)	0.051	-0.181 (15)	0.062	-0.168 (26)	0.048	0.7768	0.8735	0.9539
Change in %VC	6MWT SpO2≥90	%VC≥70	-2.083 (35)	1.070	-0.420 (16)	1.580	-2.801 (32)	1.114	0.6438	0.2209	0.2927
		%VC<70	-9.084 (9)	2.446	-7.340 (6)	2.963	-7.167 (9)	2.449	0.5902	0.9647	0.7427
	6MWT SpO2<90	%VC≥70	-1.737 (36)	1.569	-0.590 (17)	2.264	-6.080 (36)	1.568	0.0555	0.0493	0.0213
		%VC<70	-4.143 (23)	1.637	-5.090 (15)	2.079	-5.669 (26)	1.567	0.5037	0.8280	0.6158

Item	Category 1 SpO2	Category 2 PaO2	High-dose Group		Low-dose Group		Placebo Group		P-value
			LS mean (n)	SE	LS mean (n)	SE	LS mean (n)	SE	H vs P	L vs P	H+L vs P
Change in VC	6MWT SpO2≥90	PaO2≥70 Torr	-0.115 (43)	0.032	-0.060 (22)	0.045	-0.137 (36)	0.035	0.6423	0.1766	0.2667
		PaO2<70 Torr	- (1)	-	- (0)	-	- (5)	-	-	-	-
	6MWT SpO2<90	PaO2≥70 Torr	-0.115 (42)	0.043	-0.087 (28)	0.053	-0.199 (56)	0.038	0.1509	0.0881	0.0585
		PaO2<70 Torr	-0.014 (15)	0.067	-0.102 (4)	0.129	-0.135 (6)	0.105	0.34560	0.8456	0.5556
Change in %VC	6MWT SpO2≥90	PaO2≥70 Torr	-3.486 (43)	1.013	-2.329 (22)	1.412	-4.544 (36)	1.107	0.4833	0.2200	0.2484
		PaO2<70 Torr	- (1)	-	- (0)	-	- (5)	-	-	-	-
	6MWT SpO2<90	PaO2≥70 Torr	-3.755 (42)	1.379	-2.629 (28)	1.698	-6.124 (5)	1.199	0.1975	0.0965	0.0744
		PaO2<70 Torr	-0.204 (15)	2.040	-3.038 (4)	3.951	-3.964 (6)	3.229	0.3364	0.8578	0.5568

(-; not calculated.)

Incidentally, in the phase II trial in Japan, the change in VC was categorized as "improved," "stable," or "deteriorated" in accordance with the ATS/ERS criteria [17] (where separation into three categories was based on 10% change in VC). The categorical change was significantly different between the pirfenidone treatment group and placebo group in the phase II trial but not in the phase III trial as described in the Online Supplementary Materials of our preceding paper [13]. By the way, patients in Subgroup B met the criteria for entry into the phase II trial, and Subgroup B was expected to resemble Subgroup A. Therefore, it was expected that significant differences might be seen by comparing categorical changes in Subgroup A and/or Subgroup B. Thus, the distributions of categorical changes in VC at week 52 were compared between pirfenidone (high- and low-dose) and placebo groups with the Wilcoxon rank sum test in "Subgroup A" and "Subgroup B" of the phase III trial. Then, changes in VC were classified as "improved," "stable," or "deteriorated" using the 10% change criterion employed in the Phase II trial. The differences in the distribution of categorical change in VC were statistically significant between high-dose and placebo groups (p = 0.0691), low-dose and placebo groups (p = 0.0861), and pooled (high + low) dose and placebo groups in Subgroup A (p = 0.0295; Figure 1) but not in "Subgroup B" (data not shown).

Temporal changes of subjective symptoms in subpopulations

To evaluate the changes in subjective symptoms (i.e., cough and dyspnea) during the phase III trial, means of the changes in cough score and dyspnea score (with F, H-J classification) from baseline were calculated at each observation time and are shown in Figures 2 and 3. Further, the results of ANCOVA using the changes in cough and dyspnea scores from baseline to week 52 as responses, also are shown in Figures 2 and 3, respectively. In the full analysis set (FAS), pirfenidone tended to prevent the elevation of these scores more consistently in the high-dose and low-dose groups than in the placebo group, although the differences were not significant. When examined in Subgroups A and B, changes in cough and dyspnea scores showed that pirfenidone prevented increase in cough (Figure 2) and dyspnea (Figure 3) more effectively in Subgroups A and B than in the FAS at week 52. In addition, the differences in dyspnea scores seen in Subgroup A between pirfenidone (high- and low-dose) and placebo groups were significant at week 12 (p-values were 0.0428 and 0.0379, respectively). The significant difference in cough score in Subgroup A was seen between low-dose and placebo group (p = 0.0502).

Our exploratory analyses using changes in VC, categorical changes in VC, PFS time, and scores on subjective symptoms as outcomes suggested that pirfenidone was more effective in patients with mild-to-moderate lung function impairment (baseline %VC ≥ 70 and the lowest SpO₂ < 90; Subgroup A). In addition, pirfenidone had significant effects on some of these outcomes in Subgroup B (baseline PaO₂ ≥ 70 and the lowest SpO₂ < 90). In the population of patients with mild-to-moderate disease, pirfenidone is especially effective in patients with desaturation during exercise which typically corresponds to the lowest baseline SpO₂ < 90.

To evaluate temporal changes in subjective symptoms in the phase III trial, means of the changes in cough and in F, H-J classification scores were calculated. In FAS, pirfenidone tended to prevent the elevation of these scores more consistently in high- and low-dose groups than in the placebo group, but no significant differences were detected. Additional analysis in the present study, however, showed that compared to placebo, pirfenidone significantly prevented elevation of these scores at week 52 and significantly lowered dyspnea score as early as week 12 especially in Subgroup A. These results suggest that pirfenidone can be expected to prevent worsening of subjective symptoms such as dyspnea and exert this effect at a very early stage in the same patient population shown to have reduced impairment of respiratory functions such as VC. We additionally compared the incidence of acute exacerbation between pirfenidone and placebo groups in Subgroup A. The incidence in the pirfenidone group (1.82% [1/55]) was lower than that in the placebo group (8.33% [3/36]), although the difference was not statistically significant (data not shown).

In the 'responsive' subgroup, i.e., patient subgroup with baseline %VC of ≥70 and the lowest SpO₂ of <90, lung function is better preserved though oxygen saturation is deteriorated. These are mutually conflicting characteristics. As pulmonary fibrosis progresses at the early stage of the disease, interstitial septa between the alveoli and capillaries, which are the major sites of gas exchange, become thickened and gas exchange and diffusion become impaired. Hypoxemia is thought to be caused by pulmonary-diffusion impairment and ventilation-perfusion ratio mismatching. The ventilation-perfusion ratio mismatching is more likely during exercise due to dynamic change of pulmonary circulation and increased flow, irrespective of PaO₂ status at rest [20]. In this way, the thickening of the lung interstitial septa leads to desaturation and dyspnea during exercise at an early stage of the disease progression. The results showing that pirfenidone tended to be more effective in patients with %VC ≥70 and the lowest SpO₂ <90 than in patients with %VC <70 and the lowest SpO₂ <90 (Table 1) suggest that the desaturation exhibited by the subgroup with more preserved lung function may have been due to developing fibrosis with inflammatory edema and not to established fibrosis. Pirfenidone could be more effective in treating such 'young' fibrosis and therefore able to inhibit progression of fibrosis during the early stage. Furthermore, pirfenidone, which also has anti-inflammatory activity [3], may be effective in improving dyspnea by reducing inflammatory edema and vascular permeability. Above explanations are speculative, and therefore a future prospective clinical study is necessary to confirm the better response to this drug by this subgroup.

Recent clinical trials of drugs for IPF tend to include mild-to-moderate cases while totally excluding severe cases, as in the clinical trials of IFN-γ [21]. This is probably because the inclusion of severe cases can bias the evaluation of treatment effectiveness. In the phase II trial of pirfenidone for Japanese patients [7], mean decline in VC was 130 mL in the placebo group and 30 mL in pirfenidone-treated groups with a p-value of 0.0356 at week 36. The reduction in decline by pirfenidone was greater than that obtained in the phase III trial in Japan [13]. The entry criteria of the phase II trial included PaO₂ at rest of ≥70 torr and the lowest SpO₂ during 6MET of ≤90%, while criteria of the phase III trial were broader (i.e., 1] oxygen desaturation of >5% difference between resting SpO₂ and the lowest SpO₂, and 2] the lowest SpO₂ >85% while breathing air). These characteristics were similar in patients enrolled in the phase II trial and in the Subgroup B patients of the phase III trial enrolled in this study. Pirfenidone was shown to have a more marked effect on both the patients with the lowest SpO₂ ≤ 90% in Subgroup B as those in the phase II trial (data not shown). Possibly, the phase II trial might have enrolled a population of patients who were more responsive to the drug. The broader criteria for inclusion into the phase III trial might have resulted in a more heterogeneous population and more variable data.

In Japan, the severity of idiopathic interstitial pneumonia (IIPs) is classified on the basis of baseline PaO₂ value at rest, and categories are defined by 10-torr intervals. The grade of IIPs in patients with SpO₂ of < 90 on exertion is increased by one (except for grade I) as described in the online supplemental materials of our previous report of the Phase III trial [13]. In the phase III trial, patients were grouped based on severity to identify the subpopulation that was more responsive to pirfenidone. Pirfenidone was found to be more effective in grade-III patients (data not shown). A more detailed analysis revealed that the population of patients with PaO₂ ≥ 70 and < 80 included many grade-III patients when SpO₂ was < 90% on exertion (data not shown). These findings also may support the efficacy of pirfenidone in patients with desaturation during exercise.

Identifying those patients clinically responsive to pirfenidone is very important. The present analyses revealed that pirfenidone was more effective in populations of patients with relatively favorable baseline %VC and PaO₂, especially in those with desaturation on exertion. Since pirfenidone was more effective in Subgroup A than in Subgroup B, baseline %VC may be a more appropriate index than PaO₂. In addition, patients presenting desaturation during exercise may be comparable to those complaining of dyspnea on exertion. For more beneficial use of pirfenidone, the factors--baseline %VC and the presence/absence of complaints of dyspnea on exertion--may be used to select candidate patients. However, since responsiveness in this study depended on the stage of the disease as determined by respiratory function tests, these factors cannot be regarded as indicative of a responsive phenotype but rather as indicative of a responsive "phenostage" (coinage by the authors). (A 'Phenotype' determining response to therapy, for example to anti-cancer therapy, is generally characterized by expression of a specific gene, whereas the 'responsiveness' of the subgroup identified in this study may be due to the timing of treatment during disease progression rather than a specific gene.) A sub-analysis of data from the CAPACITY trials [14] yielded similar results. The FVC change at week 72 showed that a subpopulation of patients given oxygen during 6MWT at baseline responded favorably to pirfenidone [22]. To determine whether this observation and our findings are equivalent, a detailed sub-analysis of data from the CAPACITY trials or further prospective studies will be needed.

Given the exploratory nature of our study, limitations include post-hoc analysis and small sample size due to stratification. As a consequence, our findings should be considered preliminary and need to be confirmed in future studies. However, the phase II and III trials in Japan as well as the CAPACITY trials identified VC or FVC as a promising endpoint to judge the efficacy of treatment for IPF [14]. The baseline respiratory functions used to select patients for entry into the phase II trial are incidentally similar to those of patients benefiting from drug treatment in the phase III trial. In addition, the following limitations should be considered.

In this study, missing values were imputed with the LOCF method. Collard [23] as well as Swigris and Fairclough [24] pointed out the problems associated with using the LOCF method. Indeed, the method is simple but has its own deficiencies. However, there are no perfect imputation methods, and we explained the issues in detail as a rejoinder to Swigris and Fairclough [25].

In this paper, we analyzed subjective symptom outcomes or "patient-reported outcomes." Though, the results of analyses on these outcomes were not shown in the original paper as pointed by Collard [23]. The reason was that we didn't place much value in cough and dyspnea (F, H-J classification) as patient-reported outcomes, since these outcomes were treated as tertiary end-points in the phase III trial. In addition, these outcomes were observed not retrospectively but prospectively, as was described in section "Efficacy end-points" of our previous paper [13].

Additional clinical studies are desired to confirm the findings obtained in this study.

The results of these explanatory analyses identified IPF patients having baseline %VC ≥ 70 and SpO₂ < 90% during 6MET as the subpopulation that benefited most from pirfenidone treatment in terms of the changes in VC, PFS, and subjective symptoms such as cough and dyspnea. It is suggested that this subpopulation, especially, will benefit from pirfenidone treatment.