Leveraging External Data in Drug Development for Rare Diseases

Meenakshi Srinivasan1, Navin Goyal2, Daren Austin2, Rashmi Mehta2, Rajendra Singh2

1University of Florida College of Pharmacy, Orlando, Florida, USA 2GlaxoSmithKline plc, Collegeville, PA, USA

Drug development in Rare Diseases is challenging given the prevalence,

limited data and understanding of relevant biomarkers, pharmacodynamics (PD)

and clinical endpoints.

Applying a Model Informed Drug Development (MIDD) framework by

leveraging published data from clinical studies can therefore provide critical

insights into efficient clinical study designs.

METHODS

Reduction in urinary tetranor-prostaglandin D metabolite (t-PGDM) from

baseline can establish desired proof of pharmacology for hPGDS inhibitors.

Data extraction: Mean and standard deviations of change in t-PGDM from

baseline across placebo and four dose levels of active drug were digitized from

the published Phase 1 and Phase 2 studies of TAS-205, a hPGDS inhibitor

under development (Takeshita et al., 2018; Komaki et al.,2020). Key study

features are provided in Table 1.

An exponential placebo model with Imax inhibitory drug effect was chosen with

adequate qualification. (Figure 1 and Table 2)

Placebo model: 𝐸 = 𝐸𝐵𝐴𝑆𝐸 −𝑃𝐸 ∗ (1 − 𝑒−𝐾𝑃∗ 1−𝐸𝐷𝑅𝑈𝐺 ∗𝑇𝐼𝑀𝐸) , where

EBASE is the baseline value, which was fixed to 100, PE describes

magnitude of the placebo effect, KP is the rate constant characterizing the

rate of change in disease severity (placebo response) and TIME refers to

the time in hours.

Drug effect: 𝐸𝐷𝑅𝑈𝐺 =𝐼𝑀𝐴𝑋+𝐷𝑂𝑆𝐸

𝐼𝐶50+𝐷𝑂𝑆𝐸, where IMAX is the maximum inhibitory

drug effect, DOSE is the absolute value of amount of drug normalized by

body weight and IC50 is the value of change in baseline tPGDM at 50% of

IMAX

Rate of change of biomarker decline differed across the two studies

OBJECTIVEThe current work describes considerations when leveraging literature data to

design clinical studies and estimate the probability of pharmacological success

(PoPS) for drugs with similar mechanisms of action.

An example of hematopoietic-prostaglandin D synthase (hPGDS) inhibitors in

Duchenne muscular dystrophy therapy is presented.

NONMEM dataset preparation: The change in the biomarker over time was

the dependent variable (DV) of interest. Within-patient time series are

correlated, hence a multivariate normal distribution was used to generate 100

random samples of the DV per treatment arm using the mvtnorm package in R.

For the DV, the covariance matrix with different correlation coefficients

were used (r = 0, 0.1, 0.5 and 0.9).

Covariates such as age and weight for each treatment group were used

to generate random samples using multivariate normal distribution (r =

0.8).

PD model development: Several PD models were evaluated for placebo

(exponential, power and Weibull) and drug effect (Imax and linear). Potential

difference in population between Phase 1 and Phase 2 studies was estimated

as a fixed effect.

Between subject variability was modeled as log normally distributed

Model assessment included goodness of fit plots, Akaike Information

Criteria (AIC) values, visual predictive checks, bootstrapping.

Phase 1 Phase 2

Sample size (n) 21 36

Age 5-15 years ≥5 years

Treatment groups

(mg/kg/dose)

Step A (n=5): 1.67-3.33

Step B (n=5): 3.33-6.67

Step C (n=5): 6.67-13.33

Low dose (n=11): 6.67-13.33

High dose (n=12): 13.33-26.67

Sample size, placebo

group6 13

Dosing frequency,

periodBID, 7 days BID, 24 weeks

PD assessment time

pointsBaseline, day 1, 4 and 7

Baseline, day 4, week 12 and

week 24

Table 1. Features of clinical studies conducted for TAS-205

INTRODUCTION METHODS

Simulations: Clinical trial simulations for a hypothetical test drug were

conducted using the mrgsolve package in R

Patient enrollment challenges in in rare diseases were evaluated with different

sample sizes per trial were simulated: low (4-9 subjects) and high (10, 20 and 30

subjects) for an average 40 kg subject at two dose levels: low (6.67-13.33) and

high (13.33-26.67) mg/kg/dose.

PoPS was calculated as the proportion of trials with at least 50% (low threshold

criteria) and 90% (high threshold criteria) of participants in a trial achieving

greater than 30% decline in t-PGDM from baseline at 24 hours at various doses

of interest.

RESULTS

RESULTS RESULTS

Parameter Parameter definition Estimate

(%RSE)

Bootstrap median

(95%CI)

𝜽𝑰𝑴𝑨𝑿Maximal inhibitory effect -19.9 (23.7) -20.1 (-34.8 - -13.2)

𝜽𝑰𝑪𝟓𝟎Half-maximal inhibitory

dose25.7 (30.9) 26.3 (15.4 – 50.5)

𝜽𝑷𝑬Magnitude of placebo effect 48.9 (1.1) 49 (47.8 – 49.9)

𝜽𝑲𝑷Rate constant of rate of

change in disease severity0.00645 (4.2) 0.0064 (0.006-0.007)

𝜽𝑺𝑻𝑼𝑫𝒀Fixed effect of study on KP 0.00776 (7.4) 0.0078 (0.0067-0.009)

𝑬𝑩𝑨𝑺𝑬 Baseline effect 100 100

𝝎𝑰𝑴𝑨𝑿𝟐 Interindividual variability on

IMAX (%CV)93.4 (12.8) 93.1 (80.4-106.3)

𝝎𝑲𝑷𝟐 Interindividual variability on

KP (%CV)33.6 (30.1) 33.7 (22.6-43.5)

𝝈𝑷𝑹𝑶𝑷𝟐 Proportional error (%CV) 20.5 (4.6) 20.5 (19.5 – 21.4)

𝝈𝑨𝑫𝑫𝟐

Additive error 10.3 (75.8) 10.3 (0.001-30.8)

CONCLUSIONS

REFERENCES

1. Takeshita, E., Komaki, H., Shimizu-Motohashi, Y., Ishiyama, A., Sasaki, M., & Takeda, S. (2018). A phase I

study of TAS-205 in patients with Duchenne muscular dystrophy. Annals of Clinical and Translational Neurology,

5(11), 1338–1349. https://doi.org/10.1002/acn3.651

2. Komaki, H., Maegaki, Y., Matsumura, T., Shiraishi, K., Awano, H., Nakamura, A., Kinoshita, S., Ogata, K.,

Ishigaki, K., Saitoh, S., Funato, M., Kuru, S., Nakayama, T., Iwata, Y., Yajima, H., & Takeda, S. (2020). Early

phase 2 trial of TAS-205 in patients with Duchenne muscular dystrophy. Annals of Clinical and Translational

Neurology, 7(2), 181–190. https://doi.org/10.1002/acn3.50978

Biomarker inhibition increases with dose. Due to high variability in the data, the

two dose levels lack clear separation in their PD effects.

Higher the sample size, higher is the precision in estimation of the effect.

(Figure 2)

Figure 2. Effect versus time profile for simulations. Solid lines are the lower 2.5th

percentile, median and upper 97.5th percentile at each time point. Shaded bands are the 95%

CI around these percentiles are also plotted. Plots faceted by sample size.

Around 55-60% of subjects from low dose and 66-70% from high dose reach

the response criteria at 24 hours.

At lower sample size, it is not possible to distinguish between the low and high

dose. However, as the sample size increases, the distinction between the dose

levels can be observed, though not clearly separated due to the high variability.

(Figure 3)

Trial simulations conducted with smaller sample sizes may result in potentially

erroneous PoPS conclusions.

Although increasing trial sample size naturally improves power to estimate the

true PoPS, such trials may be impractical in rare diseases.

Trial to trial and population differences between studies, variability in PD

response and sample size have significant impact on PoPS estimation.

Figure 3. Proportion of subjects achieving 10-50% decline of biomarker from baseline at

24 hours. Plots faceted by sample size. Solid line represents the median proportion and the

bands represent the 95% CI around the proportion.

Counterintuitively, using the 90% success criteria, PoPS decreased with

increasing sample size and was <15% at 24 h for both high- and low-dose

groups. This could potentially be a trial design artifact, with a greater number of

possible combinations resulting in trial failures as sample size increases (Figure

5). No orderly trends with 4-9 subjects were observed in PoPS (<20% at 24 h)

(Figure 6).

Larger study sample size with the 50% change from baseline criteria led to a

considerably higher PoPS for the high-dose group, compared to low-dose

(Figure 1). Within each dose group, PoPS increased with higher sample size.

Figure 4. Probability of success: Proportion of simulations with at least 50% of subjects

reaching responder criteria, considering sample sizes of 10, 20 and 30 per trial. Note: In

this case success is defined as probability that 50% subjects achieve greater than or equal to

30% reduction in tPGDM/Cre from baseline. Plots faceted by dose level

Figure 5. Probability of success: Proportion of simulations with at least 90% of subjects

reaching responder criteria, considering sample sizes of 10, 20 and 30 per trial. Note: In

this case success is defined as probability that 90% subjects achieve greater than or equal to

30% reduction in tPGDM/Cre from baseline. Plots faceted by dose level

Figure 6. Probability of success: Proportion of simulations with at least 90% of subjects

reaching responder criteria, considering sample sizes of 4-10 per trial. Note: In this case

success is defined as probability that 90% subjects achieve greater than or equal to 30%

reduction in tPGDM/Cre from baseline. Plots faceted by dose level

Figure 1. VPC plot for

Phase 1 and Phase 2

data by treatment arm.

Step A, B and C, Low

and High dose are the

treatment arm for Phase

1 and 2 data (Table 1).

The blue dots represent

the observations, lines

represent the 90%

percentiles of the

simulated data and the

bands around the lines

represent the 90%

confidence intervals

around these quantiles.

Table 2. Final model parameter estimates

DISCLOSURESM.S is funded through a University of Florida/GlaxoSmithKline Pharmacokinetics/Pharmacodynamics Post-Doctoral

Fellowship. N.G and R.S are former employees of and past/current shareholder in GlaxoSmithKline plc. D.A and R.M

are employees of and shareholders in GlaxoSmithKline plc.