Comparative efficacy and safety of monotherapy and combination pharmacotherapies for idiopathic pulmonary fibrosis: a network meta-analysis of randomized controlled trials

Abstract

Background

Idiopathic pulmonary fibrosis (IPF) is a fatal, progressive fibrosing lung disease. While novel agents and combinations are emerging, head-to-head trials are scarce, underscoring the need for robust comparative evidence to guide treatment prioritization.

Methods

We conducted a systematic review and network meta-analysis (NMA) of IPF RCTs in adults, searching PubMed, Embase, Cochrane Library, and Web of Science through 1 December 2025. Eligible studies compared pharmacotherapies (mono/combination) with placebo or active comparators and reported prespecified outcomes. Primary outcomes were forced vital capacity (FVC) change and disease progression; secondary outcomes included all-cause mortality, adverse events (AEs), serious AEs (SAEs), and DLCO change. Risk of bias was assessed via RoB 2, with frequentist random-effects NMA for continuous/binary outcomes and a random-effects model for disease progression. Treatment rankings used SUCRA, and evidence certainty was evaluated via GRADE with CINeMA.

Results

Thirty-five reports (8,983 participants, 21 strategies) were included. RoB 2 ratings were low (19 studies) or raised some concerns (16). For FVC preservation, recombinant human pentraxin-2 (RHP) (SMD = 0.72, 95% CI: 0.23–1.20), rentosertib (SMD = 0.63, 95% CI:0.04–1.22), and nintedanib (SMD = 0.50, 95% CI:0.27–0.73) showed the highest statistical probability of benefit versus placebo, though these rankings are based primarily on low-certainty indirect comparisons. For disease progression, pirfenidone, pamrevlumab, and nerandomilast showed favorable but imprecise effects. No regimen significantly reduced all-cause mortality or showed clear AE/SAE differences versus placebo, with wide uncertainty in comparisons. Evidence certainty was mostly low to very low, driven by imprecision and indirect comparisons in star-shaped networks.

Conclusions

RHP, rentosertib, and nintedanib showed consistent signals for preserving FVC versus placebo. However, comparative effects on progression, mortality, and safety remain uncertain due to sparse data and limited head-to-head evidence. Large, well-designed trials with harmonized endpoints and longer follow-up are needed to validate these exploratory rankings and define optimal treatment sequencing and combinations.

Registration

PROSPERO CRD420261289653.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12890-026-04342-8.

Background

Idiopathic pulmonary fibrosis (IPF) is a chronic, progressive fibrosing interstitial pneumonia of unknown etiology, characterized by irreversible pulmonary remodeling and a gradual decline in lung function. It predominantly affects the elderly, with a higher incidence and increased risk in males [1–3]. The clinical course of IPF is highly variable; while some patients experience prolonged periods of stability, others may face rapid progression or acute respiratory exacerbations [4]. Despite ongoing advancements in therapeutic strategies, IPF remains a progressive lung disease with a poor prognosis. Previous studies have indicated that patients who do not receive antifibrotic therapy have a median survival time of only 3 to 5 years following diagnosis [5–7]. Additionally, a 5-year follow-up of a prospective multicenter registry cohort reported a 5-year survival rate of approximately 53.7% in IPF patients [8].

Antifibrotic therapy remains the cornerstone of pharmacological treatment for idiopathic pulmonary fibrosis (IPF). Pirfenidone and nintedanib are the primary agents recommended in the current Evidence-based Guidelines for the Diagnosis and Treatment of Idiopathic Pulmonary Fibrosis, both of which received approval from the U.S. Food and Drug Administration in 2014 [9]. However, the clinical benefits of these treatments are primarily limited to slowing disease progression, as neither has been shown to reverse fibrosis or provide a cure [10, 11]. Furthermore, long-term treatment is often hindered by adverse reactions, with gastrointestinal side effects, particularly diarrhea, being the most common for nintedanib [12]. Pirfenidone is also frequently associated with gastrointestinal and dermatologic side effects [13]. These adverse events may necessitate dose adjustments or discontinuation, which can impact patient adherence and real-world treatment efficacy. Therefore, the development of more effective and better-tolerated therapies, along with the optimization of treatment regimens, remains a critical focus in this field.

In recent years, novel therapies, including multi-target small molecules, monoclonal antibodies, and pathway-specific inhibitors, have increasingly advanced to the randomized controlled trial (RCT) phase for validation [14, 15]. Concurrently, combination strategies—where new drugs are added to existing antifibrotic treatments—have become more prevalent, further complicating clinical decision-making. Notably, in October 2025, the U.S. Food and Drug Administration (FDA) approved the PDE4 inhibitor nerandomilast for the treatment of IPF, marking a shift in treatment paradigms [16]. However, existing RCTs predominantly utilize placebo as the standard control, with a lack of direct head-to-head comparisons between different novel therapies, as well as between monotherapy and combination strategies. Consequently, the relative efficacy and safety of these treatments cannot be definitively established based on direct evidence alone. To address this gap, the present study aims to systematically search for and integrate RCTs on IPF drug treatments, employing network meta-analysis to compare monotherapies with combination therapies. The primary focus will be on outcomes such as lung function (e.g., FVC, DLCO), disease progression, all-cause mortality, and adverse events (AEs) and serious adverse events (SAEs), with the goal of providing more robust and interpretable evidence to inform individualized treatment strategies for IPF.

Materials and methods

This network meta-analysis (NMA) was conducted and reported in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses extension for Network Meta-Analyses (PRISMA-NMA) guidelines [17]. To ensure transparency and reproducibility, the study protocol was prospectively registered with PROSPERO (the International Prospective Register of Systematic Reviews; CRD420261289653).

Data sources and search strategy

A systematic search was performed in PubMed/MEDLINE, Embase, the Cochrane Library, and Web of Science from database inception through December 1, 2025. The search strategy combined controlled vocabulary terms (MeSH/Emtree) with free-text keywords. Searches were organized around three core concepts: disease-related terms (e.g., idiopathic pulmonary fibrosis, cryptogenic fibrosing alveolitis), study design–related terms (e.g., randomized controlled trial, randomized clinical trial), and intervention-related terms (e.g., nintedanib, pirfenidone, N-acetylcysteine, imatinib, bosentan, ambrisentan, macitentan, ziritaxestat, pamrevlumab, simtuzumab, cyclophosphamide, among others). No language restrictions were applied. The full search strategies for all databases are provided in Supplementary Table 2.

Additionally, the reference lists of eligible studies and relevant systematic reviews were manually screened, and citation tracking of key articles was conducted to identify potentially overlooked studies. An updated search was performed prior to final inclusion to ensure the currency of the evidence. When key outcome data were missing or inadequately reported, efforts were made to contact the original study authors to obtain further information.

Selection criteria

Eligibility criteria

1. Population: Adult patients with idiopathic pulmonary fibrosis (IPF), diagnosed after exclusion of known causes and exhibiting a usual interstitial pneumonia (UIP) pattern. UIP was confirmed by high-resolution computed tomography (HRCT) of the chest or by histopathological evaluation of surgical lung biopsy specimens [18–20].

2. Any pharmacological therapy for IPF, administered as monotherapy or in combination. This includes approved antifibrotic agents (e.g., nintedanib, pirfenidone) as well as other targeted, immunomodulatory, or antifibrotic candidate therapies (e.g., endothelin receptor antagonists).

3. Comparators: Placebo or alternative pharmacological treatment regimens.

4. Outcomes: Randomized controlled trials were required to report at least one of the following: Serious adverse events (SAEs), defined as severe adverse events occurring during the trial.Any adverse events (AEs), defined as all adverse events reported during the trial; All-cause mortality (DEATH); Change in forced vital capacity (FVC) from randomization to the end of follow-up; Change in diffusing capacity of the lung for carbon monoxide (DLCO) from randomization to the end of follow-up; Disease progression, defined as the first occurrence of a > 10% decline in FVC or a > 15% decline in DLCO from baseline.

5. Combination therapy: In this network meta-analysis, the term ‘combination therapy’ predominantly refers to ‘add-on’ strategies—wherein a novel experimental agent was administered to patients already receiving stable background therapy (e.g., nintedanib or pirfenidone). We explicitly extracted and categorized these add-on regimens as distinct intervention nodes to evaluate their synergistic potential compared to monotherapy.

Exclusion criteria

1. Studies reporting multiple publications from the same trial or patient cohort; in such cases, only the most complete and most recent report was included.

2. Randomized controlled trials in which anticoagulant therapy was the primary intervention.

3. Studies that did not clearly define idiopathic pulmonary fibrosis or that included mixed interstitial lung diseases without extractable IPF-specific subgroup data, or in which the diagnostic criteria did not meet established definitions for IPF.

4. Non-randomized studies, including observational studies, case–control studies, cohort studies, case reports, and review articles, as well as animal or in vitro experiments.

Study selection

Two investigators independently screened the literature by first reviewing titles and abstracts, followed by full-text assessment of studies potentially meeting the inclusion criteria. Any disagreements were resolved through discussion with a third investigator. To ensure data accuracy and timeliness, when multiple publications originated from the same trial, priority was given to the most recently published report with the most comprehensive outcome data. All included studies were cross-checked to confirm consistency and completeness.

Data extraction

Two investigators independently extracted data from the randomized controlled trials following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) framework. Any discrepancies were resolved through discussion with a third author. The following information was collected from each study: first author and year of publication, trial registration details, study name, sample size, baseline demographic characteristics (including age, sex, and country), duration of follow-up, and detailed intervention protocols for both experimental and control groups (including dosage, route of administration, and dosing frequency).

For continuous outcomes (FVC and DLCO), mean changes from baseline with their corresponding standard deviations (SDs) were preferentially extracted. When studies reported only baseline and endpoint means and SDs, the mean change was calculated as the difference between these time points. The SD of the change was estimated using the baseline SD, endpoint SD, and the assumed correlation coefficient (R), according to the following formula:

For our primary analysis, we imputed a robust correlation coefficient of R = 0.94. This value was calculated directly from the precise variance estimates and 95% confidence intervals reported in the large-scale phase 3 FIBRONEER-IPF trial [21]. Given the frequent underreporting of variance data for secondary continuous outcomes (such as DLCO) across trials—where data are often reported only as medians/ranges or derived from complex mixed models making algebraic back-calculation unfeasible—we universally applied this FVC-derived correlation coefficient to impute missing variance for all continuous outcomes to maintain methodological consistency. To rigorously test the robustness of this imputation, we prespecified sensitivity analyses utilizing alternative, more conservative correlation coefficients (R = 0.80 and R = 0.50), in accordance with Cochrane Handbook recommendations.”

Quality assessment

The quality of the included randomized controlled trials was assessed using the Cochrane Risk of Bias 2 (ROB 2) tool. ROB 2 systematically evaluates potential sources of bias across five domains: the randomization process, deviations from intended interventions, missing outcome data, measurement of outcomes, and selective reporting. Each domain, as well as the overall risk of bias, was classified as low risk, some concerns, or high risk [22].

Statistical analysis

Network meta-analyses were performed using Stata MP version 17.0 and R software. Continuous and dichotomous outcomes were analyzed using a frequentist framework in Stata, while time-to-event outcomes (e.g., disease progression) were analyzed using a Bayesian approach in R. For continuous outcomes, effect estimates were expressed as mean differences (MDs) with 95% confidence intervals (CIs) when identical scales and units were used across studies; otherwise, standardized mean differences (SMDs) with 95% CIs were calculated. Dichotomous outcomes were summarized using odds ratios (ORs) with 95% CIs.

Multi-arm trials were analyzed using an augmented data structure to generate all pairwise comparisons, while maintaining the correlation structure among multiple comparisons within the same study. This approach was employed to prevent the underestimation of standard errors arising from the repeated use of a shared comparator. For dichotomous outcomes, a continuity correction was applied across all treatment arms within the study when zero events or zero non-events occurred in any arm. Specifically, 0.5 was added to each cell of the 2 × 2 contingency table to avoid infinite odds ratio (OR) estimates.

As most interventions were directly compared with placebo, the overall evidence network exhibited a star-shaped structure without closed loops. Consequently, global inconsistency tests based on closed loops, as well as local inconsistency assessments such as node-splitting, were not conducted. The primary analyses were performed under the assumptions of consistency and transitivity, using random-effects consistency models. Between-study variance (τ²) was estimated using the restricted maximum likelihood (REML) method.

For time-to-event outcomes (e.g., disease progression), Bayesian network meta-analyses were performed in R using the gemtc and rjags packages. Treatment effects were modeled on the logarithmic hazard ratio scale (lnHR), along with their corresponding standard errors. Given the potential clinical heterogeneity across studies, such as the widely varying follow-up durations (ranging from 3 to 96 weeks), a random-effects model was selected for the primary analysis of disease progression to appropriately account for between-study variance.

Three independent Markov chains were run, each with a burn-in period of 20,000 iterations, followed by 100,000 sampling iterations (thinning = 1). Model convergence was assessed using trace plots, the Gelman–Rubin diagnostic, and the evaluation of effective sample sizes.

Network plots were constructed to illustrate the relationships among interventions, with node sizes proportional to the cumulative sample size for each treatment and edge thickness proportional to the number of studies providing direct comparisons. To rank the interventions, multiple ranking metrics were calculated and reported, including the surface under the cumulative ranking curve (SUCRA), the probability of being the best treatment (PreBest), and mean ranks. Rank probabilities were also presented to reflect ranking uncertainty, thereby enhancing the robustness and interpretability of the results.

Publication bias and small-study effects were assessed using comparison-adjusted funnel plots when more than 10 studies were available. Sensitivity analyses were conducted using a leave-one-out approach, in which each study was sequentially removed and the random-effects consistency model refitted. The direction and magnitude of pooled effect estimates were then compared. Furthermore, to address potential clinical heterogeneity and violations of the transitivity assumption, an additional sensitivity analysis was prespecified to exclude trials conducted specifically in patients with acute exacerbations of IPF (e.g. studies evaluating cyclophosphamide added to glucocorticoids), restricting the network strictly to patients with chronic stable IPF. Additionally, univariable network meta-regression analyses were performed to explore the influence of study-level covariates on treatment effects. Regression coefficients, 95% confidence intervals, and Wald test P-values were reported, with P < 0.05 considered indicative of statistical evidence suggesting a potential effect-modifying role of the covariate.

GRADE assessment

The certainty of evidence derived from the network meta-analysis was systematically evaluated using the GRADE framework in conjunction with the CINeMA (Confidence in Network Meta-Analysis) tool. For randomized controlled trials, the initial level of evidence was rated as “high” and then assessed across six domains: within-study risk of bias, indirectness, imprecision, heterogeneity, inconsistency, and between-study bias (including publication bias and small-study effects).

Each domain was categorized as “no concerns,” “some concerns,” or “major concerns,” in accordance with the GRADE downgrading rules (evidence was downgraded by one level for some concerns and by two levels for major concerns). The overall certainty of evidence was ultimately classified as high, moderate, low, or very low.For continuous outcomes evaluated using standardized mean differences (SMDs), an SMD of 0.5 was predefined as the proxy threshold for the minimum clinically important difference (MCID) to evaluate clinical significance and assess imprecision, in accordance with established statistical criteria and GRADE guidance [23].

Results

Characteristics of the included studies

The initial literature search identified a total of 1,632 records. After removing duplicates and screening titles and abstracts, 140 studies underwent full-text assessment. Ultimately, 35 publications were included [14–16, 21, 24–54], comprising 42 randomized controlled trials (RCTs) (Fig. 1) with a total of 8,983 participants. The included studies evaluated 21 treatment strategies, including nintedanib, imatinib, N-acetylcysteine (acetylcysteine), sildenafil, pirfenidone, bosentan, recombinant human pentraxin-2 (RHP), ambrisentan, macitentan, pamrevlumab, simtuzumab (SV), pirfenidone plus N-acetylcysteine (Pirfen-Ace), cyclophosphamide added to glucocorticoids (CTX), nerandomilast, bexotegrast, GLPG1205, CC-90,001, ziritaxestat, anti-αvβ6 monoclonal antibody (AMA), rentosertib, and senolytics dasatinib and quercetin (SD-Q). The publication years ranged from 2008 to 2025 (Fig. 2).

Fig. 1.

Flowchart according to the Preferred Reporting Items for Systematic Reviews and Meta analyses (PRISMA) guideline

Fig. 2.

Network of eligible comparisons. A Any adverse events (AEs); B Serious adverse events (SAEs); C All-cause mortality (DEATH); D Change in forced vital capacity (FVC); E Change in diffusing capacity of the lungs for carbon monoxide (DLCO); F Disease progression (DP). Abbreviations: AEs, adverse events; SAEs, serious adverse events; FVC, forced vital capacity; DLCO, diffusing capacity of the lungs for carbon monoxide; DP, disease progression.

Regarding geographic distribution, most trials were conducted as global multicenter studies (32 RCTs), while the remaining studies were carried out in the United States (3), Germany (2), the United Kingdom (1), France (1), and Japan (3). Sample sizes ranged from 12 to 785 participants (median, 154), with generally balanced allocation between intervention and control groups. The study populations predominantly consisted of older adults, with mean baseline ages ranging from 63.2 to 72.6 years. Follow-up durations varied substantially, from 3 to 96 weeks, with the most common follow-up periods being approximately 52 weeks (7 trials) and 12 weeks (6 trials). Placebo was the most frequently used comparator. Detailed information on the intervention and control regimens, including dosage, route of administration, and dosing frequency, is provided in Table 1.

Table 1.

Table of detailed information on interventions in the included studies

First author	Year of Publication	Intervention regimen	Comparator regimen
Richeldi L	2014	Nintedanib 150 mg, oral, twice daily	150 mg of placebo twice daily
Richeldi L	2018	Nintedanib 150 mg, oral, twice daily	151 mg of placebo twice daily
Daniels C.E	2010	Imatinib mesylate 600 mg, oral, once daily	matching placebo
Homma S	2012	N-acetylcysteine (Acetylcysteine) 352.4 mg, inhalation, twice daily	no treatment (or placebo)
Jackson R.M	2010	Sildenafil citrate 20 mg, oral, once daily for 3 days, then twice daily for 3 days, followed by three times daily	matching placebo
King T.E. Jr	2014	Pirfenidone 2403 mg/day, oral	matching placebo
King T.E. Jr	2008	Bosentan 62.5 mg, oral, twice daily; titrated to 125 mg twice daily	matching placebo
King T.E. Jr	2011	Bosentan initial dose: 62.5 mg, oral, twice daily; titrated to 125 mg twice daily after 4 weeks (maintain 62.5 mg twice daily if body weight ≤ 40 kg)	matching placebo
Lancaster L	2020	Nintedanib 150 mg, oral, twice daily	matching placebo
Maher T.M.	2018	Ziritaxestat 600 mg, oral, once daily	matching placebo
Maher T.M.	2019	Nintedanib 150 mg, oral, twice daily	matching placebo
Martinez F.J	2014	Acetylcysteine 600 mg, oral, three times daily	matching placebo
Noble P.W	2011	Pirfenidone 2403 mg/day, oral	matching placebo
Noble P.W	2011	Pirfenidone 1197 mg/day, oral	matching placebo
Raghu G	2018	Recombinant human pentraxin-2 10 mg/kg, intravenous injection, every 4 weeks	matching placebo
Raghu G	2013	Ambrisentan 5 mg/day, oral; titrated to 10 mg/day after 2 weeks	matching placebo
Raghu G	2013	Macitentan 10 mg, oral, once daily (morning)	matching placebo
Raghu G	2017	Simtuzumab 125 mg (1 mL, 125 mg/mL), intravenous injection	matching placebo
Richeldi L	2020	Pamrevlumab 30 mg/kg, intravenous infusion, every 3 weeks	matching placebo
Taniguchi	2010	Pirfenidone 1800 mg/day, oral	matching placebo
B. van den Blink	2016	Recombinant human pentraxin-2 10 mg/kg, intravenous injection	matching placebo
Zisman D.A	2010	Sildenafil 20 mg, oral, three times daily	matching placebo
Sakamoto	2021	Pirfenidone 1200–1800 mg/day + N-acetylcysteine 352.4 mg twice daily, both oral	1200–1800 mg pirfenidone per day
Behr J	2016	N-acetylcysteine dispersible tablets 600 mg, oral, three times daily (fasting); combined with pirfenidone 1602–2403 mg/day, oral	matching placebo and 1602–2403 mg pirfenidone per day
Naccache et al.	2017	Cyclophosphamide added to glucocorticoids 200 mg/m², intravenous injection, on Days 0, 15, 30, 60	matching placebo(250 ml of 0·9% NaCl)
Richeldi L	2025	Nerandomilast 18 mg, oral, twice daily	matching placebo
Maher T.M	2025	Nerandomilast 18 mg, oral, twice daily	matching placebo
Lancaster L	2024	Bexotegrast 320 mg, oral	matching placebo
Raghu G	2024	Pamrevlumab 30 mg/kg, intravenous infusion, every 3 weeks	matching placebo
Strambu I.R	2023	GLPG1205 100 mg, oral, once daily	matching placebo
Mattos W.L.L.D	2024	CC-90,001 400 mg, oral, once daily	matching placebo
Maher T.M	2023	Ziritaxestat 600 mg, oral, once daily; combined with pirfenidone/nintedanib/no background therapy	matching placebo
Maher T.M	2023	Ziritaxestat 200 mg, oral, once daily; combined with pirfenidone/nintedanib/no background therapy	matching placebo
Xu Z	2025	Rentosertib 60 mg, oral, once daily	matching placebo
Raghu G	2022	Anti-avb6 Monoclonal Antibody 56 mg, subcutaneous injection, once weekly	matching placebo
Xu Z	2025	Rentosertib 60 mg, oral, once daily	matching placebo
Nambiar A	2023	Senolytics dasatinib 100 mg + quercetin 1250 mg, both oral, once daily; intermittent dosing (3 consecutive days for 3 consecutive weeks, total 9 doses)	matching placebo

Risk of bias assessment (ROB 2)

Risk of bias was assessed using the Cochrane ROB 2 tool. Among the 35 included studies, 19 were judged to be at overall low risk of bias, while 16 were rated as having some concerns; detailed assessments are presented in Supplementary Fig. S1. Most trials provided adequate reporting of the randomization process and blinding procedures, and the primary outcomes (including FVC, DLCO, all-cause mortality, and SAEs/AEs) were largely based on objective measurements or clearly defined event outcomes, thereby minimizing the risk of bias related to outcome measurement.

Network meta-analyses

Consistency and inconsistency

Disease progression and change in FVC were designated as the primary outcomes, while all-cause mortality, any adverse events (AEs), serious adverse events (SAEs), and diffusing capacity of the lung for carbon monoxide (DLCO) were considered secondary outcomes. As the evidence networks for all outcomes lacked closed loops, global inconsistency testing, local inconsistency assessments, and loop-specific inconsistency analyses could not be performed.

Given these network characteristics, network meta-analyses were conducted using consistency models under the assumption of consistency. The plausibility of the transitivity assumption was evaluated by comparing key clinical characteristics and potential effect modifiers across the included studies, including follow-up duration, mean participant age, study country, and comparator or background therapy. Based on the baseline characteristics summarized in Table 2, the included studies demonstrated reasonable comparability in terms of overall population features (e.g., mean age predominantly within an older population and a predominance of multicenter designs). However, follow-up duration varied across trials, and thus, follow-up time was further explored through meta-regression analyses. Furthermore, because the evidence network is predominantly star-shaped and the overall certainty of evidence is mostly low to very low, all SUCRA rankings presented in the subsequent analyses should be interpreted cautiously as probabilistic estimations rather than definitive clinical hierarchies.

Table 2.

Table of basic information for the included studies

First author	Year of Publication	Clinical trial registration number	Experiment Name	Average age(year)	Country	Intervention (n)	Comparator (n)	Follow-up time
Richeldi L	2014	01335464、01335477	INPULSIS-1	66.9/66.9	Multi-centre	309	204	52w
Richeldi L	2018	00514683、01170065	TOMORROW	70.1	Germany	85	85	52w
Daniels C.E	2010	00131274、00477269	-	66/67.8	Multi-centre	59	60	96w
Homma S	2012	-	-	67.6/68.2	Japan	38	38	48w
Jackson R.M	2010	359,736	-	70/71	United Kingdom	14	15	6 m
King T.E. Jr	2014	1,366,209	ASCEND	68.4/67.8	Multi-centre	278	277	52w
King T.E. Jr	2008	71,461	BUILD-1	65.3/65.1	Multi-centre	71	83	48w
King T.E. Jr	2011	391,443	BUILD-3	63.5/63.2	The United States	407	209	12 m
Lancaster L	2020	1,979,952		68.8/66.2	The United States	56	57	6 m
Maher T.M.	2018	2,738,801	FLORA	64/67	Multi-centre	15	5	12w
Maher T.M.	2019	2,788,474	INMARK	70.5/70.2	United Kingdom	116	230	12w
Martinez F.J	2014	650,091		68.3/67.2	The United States	133	131	60w
Noble P.W	2011	287,729	CAPACITY 004	65.7/66.3	Multi-centre	174	174	72w
Noble P.W	2011	287,716	CAPACITY 006	66.8/67.0	Multi-centre	171	173	72w
Raghu G	2018	2,550,873	-	69/67.6	Multi-centre	77	39	28w
Raghu G	2013	768,300	ARTEMIS-IPF	66.1/65.8	Multi-centre	329	163	34.7w
Raghu G	2013	903,331	MUSIC	66/64	Multi-centre	119	59	12 m
Raghu G	2017	1,769,196	RAINIER	67.7/68.5	Multi-centre	272	272	82w
Richeldi L	2020	1,890,265	PRAISE	68.3/68.4(68.3)	Multi-centre	50	53	48w
Taniguchi	2010	-	-	65.4/65.7	Japan	108	104	52w
B. van den Blink	2016	1,254,409	-	66.7/65.5	Multi-centre	6	14	57d
Zisman D.A	2010	517,933	STEP-IPF	69.76/68.20	Multi-centre	89	91	12w
Sakamoto	2021	-	-	73.3/71.0	Japan	34	36	48w
Behr J	2016	2012-000564-14	-	66.7/67.5	Multi-centre	60	62	8w
Naccache et al.	2017	2,460,588	-	71.1/71.2	France	60	59	6 m
Richeldi L	2025	5,321,069	-	70.3/69.9	Multi-centre	392	393	52w
Maher T.M	2025	05321082、06238622	-	66/66.6	Multi-centre	391	392	52w
Lancaster L	2024	04396756、06097260	INTEGRIS-IPF	71.4/72.1	Multi-centre	22	31	12w
Raghu G	2024	3,955,146	ZEPHYRUS-1	72.6	Multi-centre	181	175	48w
Strambu I.R	2023	3,725,852	PINTA	70.5/68.3	Multi-centre	45	23	72w
Mattos W.L.L.D	2024	3,142,191	-	69.7/71.6	Multi-centre	37	36	24w
Maher T.M	2023	3,711,162	ISABELA 1	69.4/70.6	Multi-centre	174	174	52w
Maher T.M	2023	3,733,444	ISABELA 2	69.2/70.6	Multi-centre	259	258	53w
Xu Z	2025	5,938,920	-	69.5	Multi-centre	10	14	12w
Raghu G	2022	3,573,505	-	69.5/68.5	Multi-centre	54	52	26w
Xu Z	2025	5,154,240	-	65.7/68.3	Multi-centre	18	17	12w
Nambiar A	2023	2,874,989	-	69.5/64.8	Multi-centre	6	6	3w

Change in FVC

A total of 30 studies involving 6,655 participants and 19 interventions were included in the analysis of change in FVC. For consistency of interpretation, a higher FVC change value was defined as indicating a smaller decline in FVC (i.e., better preservation of lung function); thus, an SMD > 0 was interpreted as favoring the intervention over the comparator.

Compared with placebo, recombinant human pentraxin-2 (RHP; SMD = 0.72, 95% CI: 0.23–1.20), rentosertib (SMD = 0.63, 95% CI: 0.04–1.22), and nintedanib (SMD = 0.50, 95% CI: 0.27–0.73) were associated with significantly smaller declines in FVC, suggesting potential benefits in slowing lung function deterioration (Fig. 4A). Based on SUCRA rankings, the interventions associated with the smallest FVC declines were RHP (SUCRA = 90.9%), rentosertib (85.3%), and nintedanib (82.7%).

Fig. 4.

Network meta-analysis results for different outcomes.Network of eligible comparisons. A Change in FVC(FVC); B All-cause mortality(DEATH).GRADE quality of evidence assessment summary for primary/secondary outcomes. *:High quality of evidence. $:Moderate quality of evidence. &:Low quality of evidence. ^:Very low quality of evidence

Serious adverse events

Thirty studies involving 7,932 participants and 21 interventions were included in the analysis of serious adverse events (SAEs). Compared with placebo, pamrevlumab (OR = 0.71, 95% CI: 0.46–1.09), bexotegrast (OR = 0.42, 95% CI: 0.04–4.36), and cyclophosphamide added to glucocorticoids (CTX; OR = 0.79, 95% CI: 0.38–1.63) showed a trend toward a lower risk of SAEs, although none of these differences reached statistical significance (Fig. 3A).Conversely, rentosertib exhibited a non-significant trend toward a higher risk of SAEs compared with placebo (OR = 1.42, 95% CI: 0.72–2.77). Notably, the effect estimate for bexotegrast was associated with substantial uncertainty, reflecting the limited precision of the available evidence.

Fig. 3.

Network meta-analysis results for different outcomes.Network of eligible comparisons. A Serious adverse event(SAES); B Any adverse event(AES).GRADE quality of evidence assessment summary for primary/secondary outcomes. *:High quality of evidence. $:Moderate quality of evidence. &:Low quality of evidence. ^:Very low quality of evidence

Based on SUCRA rankings, the interventions associated with the lowest risk of SAEs were pamrevlumab (SUCRA = 81.5%), bexotegrast (79.5%), CTX (71.5%), and RHP (69.2%).

Any adverse events

A total of 29 studies involving 7,095 participants and 21 interventions were included in the analysis of any adverse events (AEs). Compared with placebo, cyclophosphamide added to glucocorticoids (CTX; OR = 0.69, 95% CI: 0.26–1.85), simtuzumab (SV; OR = 0.68, 95% CI: 0.23–2.03), and macitentan (OR = 0.67, 95% CI: 0.06–7.20) showed a nonsignificant trend toward a lower risk of AEs (Fig. 3B). However, the wide confidence interval for macitentan indicates limited precision of the available evidence.

In contrast, compared with placebo, imatinib (OR = 11.60, 95% CI: 2.76–48.80), N-acetylcysteine (OR = 13.57, 95% CI: 1.48–124.36), sildenafil (OR = 14.00, 95% CI: 1.30–151.03), and nintedanib (OR = 2.90, 95% CI: 1.58–5.32) were associated with a significantly higher incidence of AEs. Based on SUCRA rankings, the interventions associated with the lowest risk of AEs were CTX (SUCRA = 81.0%), SV (80.2%), and macitentan (73.3%).

All-cause mortality

Twenty-four studies involving 7,294 participants and 17 interventions were included in the analysis of all-cause mortality. Compared with placebo, pirfenidone plus N-acetylcysteine (Pirfen-Ace; OR = 0.17, 95% CI: 0.01–2.75), nintedanib (OR = 0.54, 95% CI: 0.17–1.74), nerandomilast (OR = 0.57, 95% CI: 0.16–2.00), and pirfenidone (OR = 0.52, 95% CI: 0.08–3.45) showed a trend toward reduced all-cause mortality; however, none of these differences reached statistical significance (Fig. 4B). Notably, the confidence intervals for Pirfen-Ace and pirfenidone were wide, indicating substantial uncertainty and limited precision in the effect estimates.

Based on SUCRA rankings, the interventions most favorably ranked for all-cause mortality were Pirfen-Ace (SUCRA = 84.5%), nintedanib (68.3%), nerandomilast (66.0%), and pirfenidone (65.6%).

Change in DLCO

Eight studies involving 1,595 participants and nine interventions were included in the analysis of change in diffusing capacity of the lung for carbon monoxide (DLCO). For consistency of interpretation, higher DLCO change values were defined as indicating smaller declines (i.e., better preservation of diffusing capacity); thus, an SMD > 0 was interpreted as favoring the intervention over the comparator.

Compared with placebo, imatinib (SMD = 0.41, 95% CI: 0.04–0.77) was associated with a significantly smaller decline in DLCO. Rentosertib (SMD = 0.50, 95% CI: −0.17 to 1.18) and ambrisentan (SMD = 0.10, 95% CI: −0.09 to 0.29) also showed trends toward reduced DLCO decline relative to placebo, although these differences were not statistically significant (Fig. 5B).

Fig. 5.

Network meta-analysis results for different outcomes.Network of eligible comparisons. A Disease progression (DP); B Change in DLCO (DLCO)

Based on SUCRA rankings, the interventions associated with the smallest DLCO declines were imatinib (SUCRA = 86.1%), rentosertib (85.8%), ambrisentan (60.3%), and RHP (59.0%).

Disease progression

Ten studies involving 3,771 participants and nine interventions were included in the analysis of disease progression. Compared with placebo, pirfenidone (HR = 0.57, 95% CI: 0.27–1.21), pamrevlumab (HR = 0.78, 95% CI: 0.35–1.74), and nerandomilast (HR = 0.84, 95% CI: 0.39–1.82) showed trends toward a reduced risk of disease progression; however, these differences did not reach statistical significance (Fig. 5A).

Based on SUCRA rankings, pirfenidone (SUCRA = 85.1%), nerandomilast (72.3%), and pamrevlumab (64.7%) were the highest-ranked interventions for disease progression outcomes.

Sensitivity analyses, meta-regression, and publication bias

Leave-one-out sensitivity analyses were conducted across all six outcomes to assess the robustness of the findings. In each iteration, one study was sequentially excluded, and the remaining studies were reanalyzed using a random-effects consistency network meta-analysis model. The results demonstrated that, regardless of which study was removed, the direction of the pooled effects for each intervention relative to placebo remained unchanged. Variations in point estimates were generally small, with substantial overlap of the 95% confidence intervals. Furthermore, the statistical significance of the primary comparisons was not materially altered, indicating good robustness of the main analyses (Supplementary Table S3).

Furthermore, a sensitivity analysis was conducted to assess the impact of the assumed correlation coefficient (R) used for imputing missing SDs in continuous outcomes (FVC and DLCO). The network meta-analysis was recalculated using alternative conservative coefficients of R = 0.80 and R = 0.50 (Supplementary Figures S6–S8). As expected, applying lower correlation coefficients slightly widened the 95% confidence intervals; however, it did not substantially alter the pooled effect sizes, the treatment rankings, or the ultimate statistical significance of our primary comparisons, confirming the robustness of our findings for continuous outcomes.

In addition, univariable network meta-regression analyses were performed for all six outcomes to explore whether study-level covariates acted as potential effect modifiers of relative treatment effects. The covariates examined included treatment duration, follow-up duration, study country, and mean participant age. No statistically significant linear associations were observed between these covariates and the treatment effects of any intervention relative to the control (all P > 0.05), suggesting a limited modifying role of these factors at the study level (Supplementary Table S4). Given the absence of closed loops in the evidence network and the limited sample size for some comparisons, these meta-regression findings should be interpreted with caution.

Finally, funnel plots were generated for all outcomes to assess small-study effects and potential publication bias. The plots were largely symmetrical, with no clear evidence of systematic asymmetry or extreme outliers, indicating a low likelihood of publication bias or small-study effects (Supplementary Fig. S3).

Crucially, we performed an additional sensitivity analysis excluding the trial evaluating cyclophosphamide (CTX), as this study exclusively enrolled patients with acute exacerbations of IPF, representing a clinically distinct population from chronic stable IPF. After excluding this study, the network meta-analysis was re-run for relevant outcomes (AEs, SAEs, all-cause mortality, and FVC). The overall direction of the pooled effects and the statistical significance for the remaining interventions versus placebo remained robust and consistent with the primary analysis (Supplementary Table S7). This confirms that the inclusion of the acute exacerbation trial did not disproportionately skew the comparative efficacy and safety estimates of treatments for stable IPF.

GRADE assessment

After evaluating the certainty of evidence using the CINeMA framework, substantial variability in evidence certainty was observed across outcomes, with the overall distribution predominantly classified as low or very low.

For the outcome of any adverse events, 231 pairwise comparisons were assessed, of which 11 were rated as moderate certainty, 146 as low certainty, and 74 as very low certainty. For serious adverse events, the same 231 comparisons were evaluated, resulting in 8 comparisons with moderate certainty, 152 with low certainty, and 71 with very low certainty.

For all-cause mortality, 153 pairwise comparisons were assessed; one comparison was rated as high certainty, 106 as low certainty, and 46 as very low certainty. For change in FVC, 190 pairwise comparisons were evaluated, with 3 comparisons rated as moderate certainty, 79 as low certainty, and 108 as very low certainty.

Discussion

This study systematically synthesized evidence from randomized controlled trials published up to December 2025, comprising 35 publications, 42 RCTs, and a total of 8,983 patients with idiopathic pulmonary fibrosis (IPF). Using a network meta-analysis framework, we compared the relative efficacy and safety of 21 pharmacological treatment strategies. As most trials used placebo as a common comparator, the resulting evidence network exhibited a star-shaped structure without closed loops. Consequently, our findings are primarily based on indirect comparisons within a consistency model and should be interpreted in conjunction with sensitivity analyses, meta-regression, and CINeMA-based assessments of evidence certainty.

Our network meta-analysis suggests that recombinant human pentraxin-2 (RHP), rentosertib, and nintedanib are associated with smaller declines in FVC, indicating more consistent evidence of lung function preservation. For other outcomes, including DLCO and disease progression, several interventions showed trends toward potential benefit; however, most between-group comparisons were characterized by wide confidence intervals and predominantly low or very low certainty of evidence. Given that the evidence network largely relied on placebo-controlled trials, lacked closed loops, and was primarily based on indirect comparisons, these findings should be interpreted with caution and require validation in head-to-head trials with longer follow-up durations.

Rentosertib is an AI-designed small-molecule inhibitor targeting TRAF2 and NCK-interacting kinase (TNIK), and has been shown to attenuate IPF progression through TNIK pathway inhibition [55]. Recombinant human pentraxin-2 (RHP) exerts antifibrotic effects via multiple mechanisms, including direct inhibition of the differentiation of peripheral blood mononuclear cells into profibrotic fibrocytes [56, 57], suppression of monocyte-to-proinflammatory macrophage polarization, and reduction of transforming growth factor β1 (TGF-β1), a key mediator of pulmonary fibrosis [58]. Nintedanib inhibits several receptor tyrosine kinases, including platelet-derived growth factor receptors (PDGFRs), fibroblast growth factor receptors (FGFRs), and vascular endothelial growth factor receptors (VEGFRs), thereby blocking downstream profibrotic signaling pathways [59–61]. These receptors are aberrantly activated in IPF, promoting fibroblast proliferation, migration, and extracellular matrix (ECM) deposition; nintedanib’s inhibition of these receptors mitigates these pathogenic processes [60, 62].

In terms of safety, imatinib, acetylcysteine, sildenafil, and nintedanib were all associated with a higher incidence of any adverse events compared with placebo. Among these agents, nintedanib—despite being recommended as a first-line therapy for IPF in current clinical guidelines—has well-documented adverse effects, most notably gastrointestinal symptoms (e.g., diarrhea and nausea) and hepatotoxicity [63]. These toxicities are likely a direct consequence of its multitarget inhibition of receptor tyrosine kinases [60], which may limit tolerability and treatment adherence in real-world clinical practice, highlighting the need for careful patient selection and close monitoring during therapy.

With regard to serious adverse events, our network meta-analysis indicated that rentosertib showed a non-significant trend toward a higher risk compared with placebo (OR = 1.42, 95% CI: 0.72–2.77). In the primary trials, the most frequently reported serious adverse events leading to treatment discontinuation were related to hepatic toxicity [53]. Given the low certainty of the indirect evidence in our network, this observation should not be interpreted as a definitive safety conclusion. Rather, it represents a potential safety signal suggesting that, despite promising efficacy signals, AI-designed novel agents require careful monitoring and further rigorous clinical evaluation in larger, direct head-to-head trials to comprehensively characterize their safety profiles. For all-cause mortality, both pirfenidone acetate (Pirfen-Ace) and nintedanib demonstrated a trend toward reduced mortality [64], which is consistent with their established roles as first-line pharmacological treatments for IPF in clinical practice. The survival benefit of nintedanib may be partially explained by its ability to inhibit the formation of circulating C-reactive protein monomer (CRPM), a biomarker implicated in fibrotic processes, potentially improving long-term outcomes in patients with IPF [65].

This network meta-analysis offers several notable strengths in evaluating pharmacological treatments for idiopathic pulmonary fibrosis (IPF), particularly in terms of the comprehensiveness of the included evidence, the diversity of interventions, and the timeliness of the literature. Compared to previous network meta-analyses that focused on a limited number of agents—such as the study by Wu et al. (2024), which evaluated 13 drugs with a search cutoff in November 2022 [66], and the real-world analysis by Kou et al. (2024), which was restricted to the two conventional agents, pirfenidone and nintedanib, with evidence up to March 2023 [67] —the present study incorporates a substantially broader evidence base. Specifically, we included 42 randomized controlled trials (RCTs) involving 8,983 patients and evaluated 21 distinct interventions.

In addition to widely used antifibrotic agents such as pirfenidone and nintedanib, the scope of interventions in this analysis extended to several emerging therapies, including recombinant human pentraxin-2 (RHP), rentosertib, and ziritaxestat, as well as combination strategies such as ziritaxestat plus nintedanib and pirfenidone plus acetylcysteine. By systematically incorporating both novel monotherapies and combination regimens, this study addresses an important gap in the existing literature, which has largely overlooked these treatment approaches.

Regarding outcome assessment, this study comprehensively evaluated both efficacy and safety endpoints, including any adverse events (AAEs), serious adverse events (SAEs), changes in forced vital capacity (FVC), changes in diffusing capacity of the lung for carbon monoxide (DLCO), and disease progression. This multidimensional outcome framework allows for a more holistic and clinically relevant comparison of competing interventions, moving beyond isolated efficacy or safety considerations.

Importantly, the literature search was extended through December 2025, enabling the inclusion of the most recently published clinical trials and mitigating the risk of evidence obsolescence that may have affected earlier syntheses. Furthermore, the included studies predominantly comprised multicenter trials (32 studies) conducted across multiple countries, providing a larger cumulative sample size and broader geographic representation. These features enhance the robustness, generalizability, and clinical relevance of the indirect comparisons within the network, ultimately offering a more comprehensive and up-to-date evidence base to inform individualized treatment decision-making for patients with IPF.

Furthermore, it is imperative to contextualize the SUCRA rankings generated in this NMA within the broader realities of clinical practice and trial history. A high statistical ranking in a specific node does not supersede the primary outcomes of large-scale clinical trials. For instance, our analysis identified imatinib and rentosertib as having high SUCRA rankings for the preservation of DLCO. However, these mathematical estimations must be interpreted with extreme caution. Imatinib definitively failed to demonstrate clinical benefit in primary endpoints such as disease progression and survival in large-scale randomized controlled trials, and rentosertib currently lacks definitive phase 3 validation. High SUCRA probabilities for secondary endpoints—often driven by smaller sample sizes or specific indirect network pathways—cannot overrule the negative primary outcomes of these established trials. Therefore, SUCRA rankings in this study should be viewed strictly as exploratory hypothesis-generating signals, rather than standalone endorsements of clinical viability.

Although this network meta-analysis provides valuable evidence for clinical decision-making regarding novel and combination therapies for idiopathic pulmonary fibrosis (IPF), several limitations must be considered when interpreting the results. First, most of the outcome networks are star-shaped and lack closed loops, preventing the full application of classical inconsistency tests. As a result, the indirect comparisons heavily rely on the transitivity assumption. While we assessed robustness through baseline feature comparisons, network regression, and sensitivity analyses, the potential bias introduced by unmeasured effect modifiers cannot be entirely excluded.

Second, some comparisons are based on a limited number of studies or individual trials, leading to imprecise effect estimates and unstable SUCRA rankings. This issue is particularly evident in sparse endpoint data, such as changes in DLCO and disease progression.

Besides, another limitation relates to the methodological quality of the included trials. As illustrated in our Risk of Bias assessment (Supplementary Figure S1), while the network is primarily anchored by high-quality pivotal trials, several included studies were rated as having ‘some concerns’ or a ‘high risk’ of bias, often due to issues in the randomization process, deviations from intended interventions, or missing outcome data. The inclusion of these trials introduces a potential source of bias that could subtly influence the overall network stability and the precision of the pooled effect estimates. Although the robustness of our core findings is supported by the heavy statistical weighting of large, low-risk phase 3 trials, the presence of these high-risk studies dictates that the network’s comparative efficacy estimates—and particularly the SUCRA rankings—should be interpreted with appropriate methodological caution.

Lastly, although the use of standardized mean differences (SMDs) for continuous outcomes enhances comparability across studies, it compromises some clinical interpretability. Moreover, variability in follow-up durations, background treatments, and outcome measurement protocols across trials may have contributed to heterogeneity. Furthermore, because our analysis relied on aggregate study-level data rather than individual patient data (IPD), we were inherently constrained by how the original trials handled censoring, competing risks (such as death), and missing follow-up data. Variations in these primary statistical approaches may introduce additional methodological heterogeneity into our pooled estimates. These limitations do not undermine the comparative value of this study, but given the lack of closed loops to assess inconsistency and the reliance on low-certainty indirect evidence, the SUCRA rankings should be viewed strictly as exploratory probabilistic preferences rather than definitive measures of clinical superiority. The identified limitations highlight the need for future research to prioritize head-to-head or additive randomized controlled trials (RCTs) of mechanism-based complementary drugs and to explore heterogeneity at the individual data level. Such studies will help shift IPF treatment from merely delaying disease progression to enabling precise stratification and achieving long-term clinical benefits.

Abbreviations

AE

Adverse Event

AMA

Anti-αvβ6 Monoclonal Antibody

CI

Confidence Interval

CTX

Cyclophosphamide added to glucocorticoids

DLCO

Diffusing Capacity of the Lung for Carbon Monoxide

FVC

Forced Vital Capacity

HR

Hazard Ratio

IPF

Idiopathic Pulmonary Fibrosis

MD

Mean Difference

NMA

Network Meta-analysis

Pirfen-Ace

Pirfenidone plus N-acetylcysteine

PRISMA

Preferred Reporting Items for Systematic Reviews and Meta-Analyses

RCT

Randomized Controlled Trial

RHP

Recombinant Human Pentraxin-2

RoB 2

Cochrane Risk of Bias 2

SAE

Serious Adverse Event

SD

Standard Deviation

SD-Q

Senolytics dasatinib and quercetin

SMD

Standardized Mean Difference

SUCRA

Surface Under the Cumulative Ranking Curve

SV

Simtuzumab

UIP

Usual Interstitial Pneumonia

Authors’ contributions

Jitao Xu conceived and designed the study and drafted the manuscript; Xinyu Liu and Xitong Liang performed the literature search, data extraction, and statistical analysis; Xinrui Cai and Weibin Qian drafted the manuscript; all authors critically revised the manuscript for important intellectual content and approved the final version.

Funding

This work was supported by the National Natural Science Foundation of China (Grant No. 82274478).

Data availability

All data analyzed in this study are included in this published article and its supplementary information files. The datasets used or analyzed during the current study are available from the corresponding author on reasonable request.

Declarations

Ethics approval and consent to participate

This study is a network meta-analysis of published randomized controlled trials (RCTs). Ethical approval and informed consent were obtained by the original studies included in this analysis. The authors confirm that all methods were performed in accordance with the relevant guidelines and regulations.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.