Molecular Heterogeneity, Biomarker Discovery, and Targeted Therapy in Asthma - Genomic Biomarkers for Pharmaceutical Development - page 82

with ≥10pg/ml IL13 in sputum supernatants than in tralokinumab-treated patients with
<10 pg/ml sputum IL13 or all placebo patients taken together. There was also a greater mag-
nitude of FEV1 improvement observed in patients with baseline peripheral blood eosinophil
counts ≥0.3×10 9 /L than in patients with <0.3×10 9 /L blood eosinophils [71] . There is no avail-
able published data at this time on the effect of pitrakinra on FEV1. However, while pitrakinra
failed to significantly affect exacerbation rates vs. placebo in all comers, secondary analyses
stratifying the population by baseline FeNO, blood eosinophil counts, or a single nucleotide
polymorphism in IL4RA showed greater exacerbation rate reductions for the treated group
in each defined subset, although there was little overlap between subsets as defined by each
baseline characteristic [69,73] . Overall, each of these studies suggests that IL13 blockade may
provide clinical benefit in terms of lung function and exacerbation reduction in some, but not
all, moderate-severe asthma patients inadequately controlled on ICS.
In the lebrikizumab Phase II proof-of-concept study ('MILLY'), we stratified treatment
assignments at randomization according to a composite of serum IgE and blood eosinophil
counts [72] , which effectively differentiated mild-moderate asthmatics not taking ICS accord-
ing to bronchial epithelial Th2 signature status [74] . However, prior to study unblinding, we
pre-specified a data analysis plan in which we stratified the outcome analyses according to
baseline serum periostin level. This change in stratification was because the prototype peri-
ostin diagnostic assay was not available at the start of the study but was ready by the time
all the patients had been enrolled in the study before unblinding, and we had confidence in
its ability to predict airway eosinophilia in moderate-severe asthma patients based on the
BOBCAT study [46] . As an exploratory analysis, we also evaluated the ability of baseline
FeNO levels to predict clinical benefit. Because we had observed continuous distributions of
these biomarkers, which scaled continuously with airway eosinophil measures in BOBCAT,
there was no obvious biologically definable cutoff to pre-specify. Thus, to maximize statistical
power by creating approximately equally sized subgroups, we pre-specified the median val-
ues of the biomarkers at baseline as cutoffs. It should be noted that the median FeNO level in
this study (21 ppb) was considerably lower than the pre-specified FeNO cutoff in the mepoli-
zumab DREAM study (50ppb) [68] . Considering all comers, lebrikizumab demonstrated a
significant FEV1 benefit vs. placebo at 12 weeks (5.5%, 95% CI 0.8-10.2%). Dichotomizing the
population according to median baseline serum periostin or FeNO levels, we observed com-
parable magnitudes of enrichment for FEV1 improvement in the 'diagnostic-positive' subsets
as compared to the 'diagnostic-negative' subsets (8.2% vs. 1.6% for periostin and 8.6% vs.
1.9% for FeNO, both with wide but non-overlapping confidence intervals). Importantly, this
stratification demonstrated that a significant FEV1 improvement vs. placebo was observed
in the 'diagnostic-positive' subsets but there was no statistically significant FEV1 benefit
in the 'diagnostic-negative' subsets; thus the significant effect observed in all comers was
driven primarily by the effect in the 'diagnostic-positive' subgroups. In assessing serum peri-
ostin and FeNO levels at two visits one week apart during the pre-dose run-in period, we
observed considerably more intrapatient variability in FeNO than in periostin, with a coef-
ficient of variation (CV) of 19.3% (95% CI 17.4-22.2%) for FeNO vs. 5.0% (95% CI 4.4-5.6%)
for serum periostin [72] . This greater level of intrapatient variability in FeNO than in peri-
ostin was also observed across multiple visits in the BOBCAT study [46] . Blood eosinophils,
although not prioritized as predictive biomarkers in the analyses performed, enriched for
responsiveness to mepolizumab and tralokinumab as described above, in the MILLY study