Introduction
Nintedanib is a tyrosine kinase inhibitor that inhibits key pathways involved in lung fibrosis in interstitial lung diseases (ILDs) and is approved for the treatment of idiopathic pulmonary fibrosis (IPF), other chronic fibrosing interstitial lung diseases with a progressive phenotype (progressive fibrosing ILDs) and for the treatment of systemic sclerosis-associated ILD (SSc-ILD) [
1,
2]. In the pivotal INPULSIS-1
® and INPULSIS-2
® trials and a supportive Phase II dose finding trial (TOMORROW
®) in patients with IPF, a twice daily (BID) 150 mg dose of nintedanib significantly reduced the annual rate of decline in forced vital capacity (FVC) suggesting slowing in disease progression [
3‐
6]. Thereafter, clinical efficacy of the 150 mg dose has additionally been shown in randomized trials in SSc-ILD [
7] and progressive fibrosing ILDs other than IPF [
8]. In clinical trials, side effects of nintedanib were generally manageable, while common adverse events (AEs) in the nintedanib arms were gastrointestinal (mainly diarrhea, nausea and vomiting) and hepatic enzyme elevations. Recommendations for monitoring and dose modification for management of AEs are included in the nintedanib prescribing information [
2].
The pharmacokinetics of nintedanib and factors affecting its plasma concentration (exposure) have been described previously [
9‐
12]. Gender and renal function have no influence on nintedanib pharmacokinetics, whereas hepatic impairment, Asian race, body weight and age affect nintedanib exposure [
10,
12]. Other than hepatic impairment, these covariates have a relatively small effect on plasma exposure, whereas unexplained interpatient variability in nintedanib exposure is high [
10]. For patients with mild and moderate hepatic impairment, an approximately twofold and eightfold increase in nintedanib exposure has been shown respectively [
12]. In this analysis, we explore the association between nintedanib plasma exposure and safety—specifically diarrhea and liver enzyme elevations—and evaluate the impact of selected intrinsic and extrinsic factors on these outcomes. The models were first developed in IPF with different dose levels available; subsequently, data from patients with SSc-ILD and progressive fibrosing ILDs other than IPF randomized to the therapeutic dose of 150 mg BID or placebo were used to further explore associations between plasma exposure and safety.
Materials and methods
Studies included
Data were analyzed from 1403 subjects with IPF treated with 50–150 mg nintedanib BID (n = 895) or placebo (n = 508) in three clinical studies: one Phase II trial (TOMORROW) evaluating 50–150 mg nintedanib BID or placebo (n = 342) [
6], and two identical Phase III trials (INPULSIS-1 and INPULSIS-2) evaluating 150 mg nintedanib BID or placebo (n = 1061) [
3].
In addition, data from 576 subjects with SSc-ILD randomized to either 150 mg nintedanib BID (n = 288) or placebo (n = 288) in the SENSCIS
® Phase III trial [
7] and from 663 subjects with chronic fibrosing ILDs with a progressive phenotype other than IPF randomized to either 150 mg nintedanib BID (n = 332) or placebo (n = 331) in the INBUILD
® Phase III trial [
8] were analyzed. Study designs and results have been published previously [
3,
6‐
8,
13‐
15]. In all studies, dose interruption or reduction for the management of AEs was allowed. In the INBUILD study, specific efforts were made to exclude patients with IPF (as efficacy and safety had already been established for IPF).
The primary efficacy endpoint in all studies was the annual rate of decline in FVC, as assessed over a 52-week treatment period (despite a variable treatment period beyond week 52 in the SENSCIS and INBUILD trials).
Safety was assessed based on the occurrence of AEs, laboratory tests, physical examination, vital sign recordings, and 12-lead electrocardiogram.
For the assessment of nintedanib plasma exposure, at least two pre-dose blood samples were scheduled (Table
1) and analyzed by validated liquid chromatography mass spectrometry.
Table 1
Summary of the trials contributing data to the analyses
TOMORROW Randomized Phase II, 52 weeks | IPF (n = 342)b | Placebo (n = 85) | Pre-dose, Days 1, 29, 169, 365 and end of treatment | Baseline, Days 1, 15, 29, 43, 85, 127, 169, 211, 253, 309, 365 |
Nintedanib 50 mg BID (n = 86) |
Nintedanib 100 mg BID (n = 86) |
Nintedanib 150 mg BID (n = 85) |
INPULSIS-1 Randomized Phase III, 52 weeks | IPF (n = 513) | Placebo (n = 204) | Days 29 and 169 | Baseline, Days 1, 15, 29, 43, 85, 127, 169, 211, 253, 309, 365 |
Nintedanib 150 mg BID (n = 309) |
INPULSIS-2 Randomized Phase III, 52 weeks | IPF (n = 548) | Placebo (n = 219) | Days 29 and 169 | Baseline, Days 1, 15, 29, 43, 85, 127, 169, 211, 253, 309, 365 |
Nintedanib 150 mg BID (n = 329) |
SENSCIS Randomized Phase III, 52 weeks | SSc-ILD (n = 576) | Placebo (n = 288) | Days 29 and 169 | Baseline, Days 1, 15, 29, 43, 85, 127, 169, 211, 253, 309, 365 |
Nintedanib 150 mg BID |
(n = 288) |
INBUILD Randomized Phase III, 52 weeks | Progressive fibrosing ILD (n = 663) | Placebo (n = 331) | Days 29 and 169 | Baseline, Days 1, 15, 29, 43, 85, 127 (optional), 169, 211 (optional), 253, 309 (optional), 365 |
Nintedanib 150 mg BID (n = 332) |
The studies were conducted in accordance with the Declaration of Helsinki and approved by the ethics committee of the co-ordinating centre: Comitado Etico Provinciale, Modena, Italy (TOMORROW and INPULSIS-1); Committee on Human Research, UCSF, San Francisco CA, USA (INPULSIS-2); Kantonale Ethikkommision (SENSCIS); Pulmonary and Critical Care Medicine Chesapeake Institutional Review Board, Columbia, USA (INBUILD). Written informed consent was obtained from all subjects before study entry. A summary of the trials contributing data to the analyses is shown in Table
1.
Safety endpoints and exposure metrics
For the derivation of the safety endpoints, treatment-emergent AEs or laboratory values with an onset/worsening date between the first drug intake and a residual effect period after the discontinuation of study medication (14 days for the Phase II TOMORROW trial and 28 days for Phase III trials) within a 52-week treatment period were considered. The following safety endpoints were derived: (1) time of first alanine transaminase (ALT) and/or aspartate transaminase (AST) elevation to at least three times the upper limit of normal (ULN) over 52 weeks; and (2) time of first onset of diarrhea of any grade over 52 weeks.
Observed and population pharmacokinetic (PopPK) model-predicted pre-dose drug concentrations at steady state (C
pre,ss) were used as exposure metrics for IPF. For the derivation of observed C
pre,ss, the available pre-dose plasma measurements from each patient were collapsed into one value (geometric mean of all dose-normalized pre-dose concentrations per patient; in case of one value, this value was taken) and multiplied by the starting dose. To obtain predicted C
pre,ss values, empirical Bayes estimates were generated using the PopPK model in IPF [
10,
16] and observed nintedanib concentrations as well as baseline patient characteristics of relevant covariates in the respective trials. If no valid nintedanib concentration was available for a particular patient, only the baseline patient characteristics were considered for the prediction. The dose-normalized predicted C
pre,ss concentrations were multiplied by the actual single dose taken by a patient on a specific day to account for dose reductions and treatment interruptions (use of time-matched exposure for predicted C
pre,ss). As the two exposure measures provided comparable results in the analyses of IPF, only predicted C
pre,ss concentrations were used for subsequent exposure-safety analyses in patients with SSc-ILD and progressive fibrosing ILDs other than IPF (see “
Discussion” section). The Phase II TOMORROW trial also included an arm with patients treated at 50 mg nintedanib once daily (n = 86), which was excluded from the current analysis. This was due to the exposure measure of C
pre,ss referring to one specific time point to represent the overall plasma exposure of a patient in the exposure-safety analysis. Depending on the influence of other pharmacokinetic parameters on response (e.g. maximum plasma levels were expected to be higher for the once-daily schedule than for the BID schedule despite providing the same C
pre,ss), steady-state trough concentrations from daily administration were not considered subject to the same interpretation as those from BID administration.
Exposure–liver enzyme elevations analyses
An exposure–liver enzyme elevation model was initially developed using data from combined IPF studies. Parametric time-to-first-event (survival) modelling [
17] was applied to investigate the relationship between nintedanib exposure and the probability of developing a liver enzyme elevation. The model was first fit to placebo-treated patients, where survival times were assumed to follow a Weibull distribution [
17]. This model was compared with a model assuming a constant baseline hazard. Subsequently, the nintedanib drug effect was included by simultaneously analyzing all patients (both placebo- and nintedanib-treated). Linear, log-linear, maximum effect (E
max) and sigmoidal E
max relationships on the log of the hazard were evaluated as drug effect functions, and both PopPK-predicted and observed C
pre,ss values were used as exposure measures.
Finally, a stepwise covariate analysis consisting of univariate analysis (
p = 0.05), forward inclusion (
p = 0.05) and backward elimination (
p = 0.001) was used to explore factors potentially influencing the exposure–safety relationship. Stricter criteria were used in the backward elimination step, as commonly done to reduce selection bias of the multiple covariate testing procedure [
18]. The covariates tested for IPF included age, height, gender, body weight, body surface area, Asian subpopulations, smoking status and study.
The exposure–liver enzyme elevation model developed for IPF was subsequently applied to combined data from the IPF Phase II/III studies (TOMORROW, INPULSIS-1, INPULSIS-2) and the SENSCIS Phase III trial. Finally, data from the INBUILD trial were added (pooled analysis of the TOMORROW, INPULSIS-1, INPULSIS-2, SENSCIS and INBUILD trials).
The drug effect was re-evaluated by exploring linear, log-linear and Emax functions on the log of the hazard. Covariates were assessed using a stepwise modelling approach as described for IPF. SSc subtype (diffuse cutaneous SSc vs. limited cutaneous SSc), anti-topoisomerase antibody status (positive vs. negative) and mycophenolate (mofetil/sodium/acid) use at baseline were explored as additional SSc-ILD-specific covariates besides the demographics typically investigated for IPF. In addition, methotrexate use at baseline (yes vs. no), use of disease-modifying antirheumatic drugs with known hepatotoxic effects at baseline (yes vs. no) and FVC % predicted at baseline (as a surrogate of disease severity) were explored as covariates based on the combined data across all indications. Differences between study populations (IPF studies vs. SENSCIS vs. INBUILD) were also explored.
For the definition of liver enzyme elevation events in the INBUILD trial, two different reference ranges of liver enzyme (ALT and AST) measurements were considered. The reference ranges for ALT and AST as defined in 2014 by the central laboratory provider were used for the primary analysis of the INBUILD trial and are hereinafter referred to as “2014 reference ranges used for INBUILD primary analysis”. Independently from the INBUILD trial, the central laboratory provider updated the reference ranges for the applied ALT and AST assays in 2019, in parallel to the study conduct, to better align ranges with usual and customary practice. These are hereinafter referred to as “2019 updated reference ranges” (see Additional file
1: Table S1). The update was triggered by consultation with different peer reference laboratories and recent literature [
19] and was based on a reference population of 256 male or female volunteers. A sensitivity analysis was conducted with the 2019 updated reference ranges for ALT and AST to assess the potential influence of this update.
Exposure–diarrhea analyses
Congruent to the analyses focusing on liver enzyme elevations, exposure–diarrhea analyses were initially developed based on data from IPF studies. Parametric time-to-first-event (survival) modelling was applied to investigate the relationship between nintedanib exposure and the probability of developing an episode of diarrhea (any severity grade) over 52 weeks. As with the exposure–liver enzyme elevation analyses, the analysis consisted of three steps: (1) time-to-first-event analysis based on placebo-treated patients; (2) addition of the effect of nintedanib exposure on the risk of experiencing diarrhea (using observed and predicted C
pre,ss) by simultaneously fitting placebo- and nintedanib-treated patients; and (3) a stepwise covariate analysis [
18] consisting of univariate analysis (
p = 0.05), forward inclusion (
p = 0.05) and backward elimination (
p = 0.001). Linear, log-linear, E
max and sigmoidal E
max relationships were evaluated as drug effect functions. The same covariates as for the liver enzyme elevation variable were evaluated. However, as no exposure–diarrhea model could be established to describe the data, the covariate analysis was additionally performed for a model using categorical dose (instead of plasma exposure) as a predictor of the diarrhea risk.
Exploratory analyses were performed to further differentiate between exposure and dose as predictors of the diarrhea risk. Therefore, patients from the 100 mg BID treatment group were optimally matched by nintedanib exposure to patients from the 150 mg BID treatment group (1:2 matching) using a SAS
® macro developed by Bergstralh and Kosanke [
20]. Afterwards, the number of diarrhea events was compared between the exposure-matched treatment groups (including patients with comparable plasma exposure despite having received different nintedanib doses). Exposure–diarrhea modelling was not further pursued for data from trials in SSc-ILD (SENSCIS) and chronic fibrosing ILDs with a progressive phenotype other than IPF (INBUILD). Instead, exploratory analyses across trials evaluating the number and proportion of diarrhea events by exposure tertile and severity grade were performed.
Model selection and evaluation
Model selection was guided by numerical change in objective function values; identifiability of parameters and precision of parameter estimates; the correlation between the estimates of fixed-effect parameters; numerical stability; ability to obtain a successful covariance step; and visual inspection of basic goodness-of-fit plots. Adequacy of the base and final models was confirmed using simulation-based diagnostics and bootstrap analysis [
21].
Software
The exposure–safety analyses were performed using NONMEM® (version 7.3 or higher, ICON Development Solutions, Hanover, MD, USA). The maximal-likelihood estimation method ($ESTIMATION METHOD = 0 LIKE) was used for model fitting and parameter estimation in the time-to-first-event modelling.
Visual predictive checks, non-parametric bootstrap analysis and covariate analysis were performed using Perl-speaks-NONMEM (version 4.6.0 or higher) [
22,
23]. Post-processing and descriptive statistics were performed using R (version 3.2.2 or higher) and SAS
® (version 9.2 or higher; SAS Institute Inc, Cary, NC, USA).
Discussion
The exposure–safety analyses reported here were conducted to understand the relationship between nintedanib exposure and safety in terms of liver enzyme elevations and diarrhea, and support dose selection for patients with IPF, other chronic fibrosing ILDs with a progressive phenotype and for SSc-ILD. Data from several BID doses in IPF trials (50–150 mg) provided a relatively wide range of nintedanib exposure, enabling a good exploration of the exposure–safety relationship. Additional data from patients with SSc-ILD and progressive fibrosing ILDs other than IPF from the SENSCIS and INBUILD trials allowed comprehensive analyses across indications.
As the assessment of liver enzyme elevation events in IPF was based on a limited number of events (~ 1% in the placebo group and ~ 5% in the 150 mg nintedanib BID group) and no difference between patient populations was to be expected (i.e. the mechanism was considered indication-independent), safety data on liver enzyme elevations from studies in IPF were combined with data from studies in SSc-ILD and progressive fibrosing ILDs other than IPF (SENSCIS and INBUILD) for analysis. As such, a higher power for the detection of potential covariate effects was obtained than by using data from single studies only.
Both observed and PopPK model-predicted Cpre,ss values were selected as exposure measures for the analyses in IPF. For the scaling of observed Cpre,ss, the starting dose was used. The predicted Cpre,ss values were implemented time-dependently thus also taking into account dose reductions and treatment interruptions during trials. The two exposure variables were highly correlated, as both were derived using actual plasma concentration measurements from the trials. However, for the model-predicted values, patient demographics, pharmacokinetic variability and actual dosing history was additionally considered. This was assumed to further minimize bias, as it enabled derivation of exposure variables also for patients without a measured nintedanib plasma concentration and dose changes during the trial were taken into account. As analyses in IPF suggested consistent results for the two exposure measures, analyses in SSc-ILD and progressive fibrosing ILDs other than IPF were performed by using time-matched predicted Cpre,ss values only. Of note, the relationship between exposure and safety risks tended to be steeper when using predicted Cpre,ss than by using observed Cpre,ss. Hence, use of predicted Cpre,ss led to larger changes in the safety risks for subgroups with altered nintedanib exposure being predicted than through use of observed Cpre,ss (conservative approach).
With respect to liver enzyme elevations, a positive correlation between nintedanib plasma exposure and ALT or AST elevations ≥ 3 × ULN was found based on initial analyses in patients with IPF (using combined data from the TOMORROW and INPULSIS trials). This was confirmed by analyzing combined data from trials in IPF, SSc-ILD (SENSCIS) and chronic fibrosing ILDs with a progressive phenotype other than IPF (INBUILD). On top of the exposure-related risk (covering known factors leading to exposure increase such as Asian race, low body weight or high age), females were estimated on average to have a 3.7-fold higher risk of experiencing ALT or AST elevations ≥ 3 × ULN than males, and data in SSc-ILD and chronic fibrosing ILDs with a progressive phenotype other than IPF were again in line with findings from initial analyses in IPF.
During covariate assessment, no difference in the exposure–liver enzyme elevation relationship was found between patients included in the IPF trials and patients with SSc-ILD from the SENSCIS trial. Likewise, no clear difference of this relationship between patients from the INBUILD trial as compared to IPF trials or the SENSCIS trial was detected. Although the main analysis, using reference ranges for ALT and AST defined in 2014 by the central laboratory, suggested an approximately twofold higher (exposure- and gender-adjusted) probability of transaminase elevations for patients in the INBUILD trial than for patients in the IPF or SENSCIS trials, no significant difference between trials was present in the sensitivity analysis using 2019 updated reference ranges for ALT and AST. These reference ranges were established by the central laboratory provider independently from the INBUILD trial in 2019 (see “
Materials and methods” section) and were more closely aligned to those used in previous nintedanib studies (see Additional file
1: Table S1). Overall, it needs to be taken into account that due to the lack of standardized reference ranges for ALT and AST measurements, differences between laboratories can affect the comparability of assessments on drug-induced liver disease based on ALT or AST [
24‐
26]. Characteristics of the local reference population used for the determination of reference ranges or differences in the methodology by manufacturers to establish recommended reference intervals have been identified as relevant factors contributing to this variability [
27,
28]. With this in mind, the 2019 updated reference ranges for the INBUILD trial (more closely aligned to the reference ranges from previous nintedanib trials) might be more appropriate for comparison of liver enzyme elevation events between the different nintedanib studies than the 2014 reference ranges used for the INBUILD primary analysis. The 2014 reference ranges for INBUILD (with lower ULNs than in previous nintedanib trials) are considered to provide a more conservative estimate of the incidence of liver enzyme elevation events. As such, the use of these values led to a higher number of observed events (by a factor of ~ 2) in the INBUILD trial than in IPF trials or in the SENSCIS trial, and trigger a study effect in the exposure–liver enzyme elevation model. However, the sensitivity analysis indicates that numerical differences in liver enzyme elevations between trials can be explained by assay differences, data variability and patient demographics (e.g. distribution of females or factors influencing exposure such as low/high body weight, age or Asian ethnicity) such that a clear population effect cannot be determined. The relationship between nintedanib plasma exposure and ALT or AST elevations ≥ 3 × ULN was weak to moderate across all models and indications (taking into account the steepness of the exposure–safety curve), and was therefore comparable between studies in IPF, SSc-ILD and chronic fibrosing ILDs with a progressive phenotype other than IPF. Liver enzyme elevations normalized in the majority of patients in nintedanib trials either spontaneously or with dose reduction, treatment interruption or discontinuation.
The comprehensive analyses with regard to diarrhea presented here indicate that there is no association between exposure and the risk of diarrhea based on data from IPF, SSc-ILD or chronic fibrosing ILDs with a progressive phenotype other than IPF. However, a clear relationship between the dose administered and diarrhea was observed. This suggests that local gut concentrations might be more relevant than plasma exposure for the occurrence of diarrhea.
The exposure–safety analyses described here, in combination with recently published exposure–efficacy analyses for nintedanib [
29] support the therapeutic dose of 150 mg nintedanib BID in patients with chronic progressive fibrosing ILDs overall and in subgroups of patients, where nintedanib plasma exposure may be altered (e.g. due to known factors leading to an exposure change such as Asian race, low or high body weight and low or high age). Previous PopPK analyses indicated that intrinsic or extrinsic factors have only a small to moderate influence on nintedanib plasma exposure [
10]. Thus, single covariate effects of Asian race, body weight or age did not change the plasma exposure by more than 50% and were well within the variability range of nintedanib. In addition, recently published exposure-efficacy analyses [
29] indicate that an increase in exposure might still be beneficial in terms of efficacy. At the same time, for an exposure increase by up to 50%, a limited safety impact is expected, as based on the current analysis, plasma exposure has no influence on diarrhea occurrence and only a moderate influence on transaminase elevations (Fig.
1). Adverse events were manageable by dose reductions and treatment interruptions. Based on this, altered nintedanib exposure does not warrant a priori dose adjustment (excluding patients with hepatic impairment having specific labelling recommendations [
1,
2]). Due to a potentially higher frequency of liver enzyme elevations, patients with elevated nintedanib exposure (e.g. due to Asian race, low body weight, high age or combinations of these risk factors) should, however, be closely monitored for tolerability. Additional assessments on tolerability and safety and the appropriate use of nintedanib have been published previously [
30].
Conclusions
In summary, results of exposure–safety analyses for nintedanib were consistent across nintedanib studies including patients with IPF, other chronic fibrosing ILDs with a progressive phenotype and SSc-ILD. A positive correlation between nintedanib exposure and ALT or AST elevations in general and female gender as an exposure-independent risk factor was found. This relationship was considered weak to moderate across different indications.
With regard to diarrhea, the actual dose administered was found to be a better predictor of the risk of experiencing diarrhea than plasma exposure, suggesting that local gut concentrations may be more relevant than plasma exposure. Therefore, a change in diarrhea risk is not expected for patients with altered nintedanib exposure (e.g. due to low/high age, body weight or Asian race). Based on this, no a priori dose adjustment is recommended in patients with altered nintedanib exposure (except for patients with hepatic impairment substantially affecting nintedanib plasma exposure). However, due to a potentially higher frequency of AEs, close monitoring for tolerability is warranted for patients with elevated nintedanib exposure (e.g. due to Asian race, low body weight, high age or combinations of these risk factors).
The presented exposure–safety analyses, in combination with recently published exposure–efficacy analyses for nintedanib [
29], provide a platform to assess the risk–benefit profile of nintedanib in IPF, other chronic fibrosing ILDs with a progressive phenotype and SSc-ILD, and support the therapeutic dose of 150 mg nintedanib BID across different indications of chronic fibrosing ILDs.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.