Transcript slides

Implementing Adaptive
Clinical Trials
4: Drug Supply
Tom Parke
[email protected]
Overview
• There are more treatment arms
• How do we supply more doses?
• Arms may be dropped / introduced
or arms may become more / less
likely to be allocated
• We don’t know how much of each
dose we will need make / package?
• We don’t know which doses to ship?
1. How to make and supply many
treatment arms
2. How much to make?
3. How much to supply?
More treatment arms
• How to manufacture / deliver multiple
treatments?
•
•
•
•
manufacture each one
use combinations
may need multiple placebos
how to ensure patient compliance?
• How to limit overage from additional
treatment packs types?
How many treatment arms?
•
•
•
•
8 doses is probably enough
Often use less (4-6)
Might use more (but only if it was easy)
Might start with few doses and add ‘in between
doses’ only where needed.
Examples
•
•
•
•
•
Stroke: 16 doses (IV)
Migraine: 6 doses
Other A: 3 dose combinations
Other B: 5 doses
Other C: 7 doses and 4 doses
Examples
• IV with 3 concentrations – randomiser
sent ‘recipe’ to centre
• Blister pack with all doses (single shot)
• Take a combination
• take a blue pill and a red pill
• Make all 4 doses
Combinations
• Can try to make as few dose strengths as
possible ...
• Strengths: 0, 1, 3 & 4 – in combinations of 2
• Strengths: 0, 1, 9 & 81 – in combinations of 3
• Strengths: 12.5, 50, 150 – in combinations of 3
Combinations cont’d
• But can be difficult to predict required
quantity of each strength.
• Possibly simpler is say:
• Strengths: 0, 1, 2, 4, 6, ... using the 1 dose to
make intermediate doses. (0,0), (1,0), (2,0),
(2,1), (4,0), (4,1), ...
Combinations
• Assume:
•
•
•
•
•
that 20% get placebo,
20% the best dose,
15% the next two best doses,
10% the two after that and
5% the two after that.
• Consider min & max requirements for tablets for
each treatment dose in turn being ‘best’.
Maximum required tablets
per 100 subjects randomized
Dose
Scheme 1
Scheme 2
Min
Min
Max
Max
0
1
2/3
55
20
35
85
65
60
75
40
15
90
50
35
4
25
75
20
35
10
35
6
Total
285
245
Result
• Need to make 14% fewer tablets per
100 subjects with simpler – more
strengths scheme 2.
• 47% less overrage to supply
combinations.
• Your mileage may vary, but fewer
tablet strengths may not mean less
wastage
Just in time packaging
• Capsules can easily be made different strengths
• If they can be made quickly & to order it is easier:
• to provide adaptive supply as randomisations change
• to prepare new doses to add at interim
• Need prior warning from DMC before dropping or
adding treatment arms.
• DMC need to know lead time for implementation.
• DMC need to monitor accumulating data /
information
• predict interim decision
• predict timing of decision
1. How to make and supply many
treatment arms
2. How much to make?
3. How much to supply?
Example Trial
• Phase 2 trial of a Neuropathic Pain
compound.
• 8 doses plus placebo
• Taken daily for 6 weeks
• Maximum of 250 subjects
Example simulation: fitted
curve
Fitted response
over progressive
weeks
Example simulation: adaptive
dose allocation
How much of each dose?
• How can we determine how much to
manufacture / package?
• When should we schedule new
batches to be manufactured /
packaged?
Simulate the adaptive trial
• Use not just one scenario, but the range of
plausible scenarios
• A max for each treatment arm that covers
90% or 95% of cases should suffice
• Allow more for Placebo
• Propose the limit to the designers – allow
them to include the limit in their simulations.
How many do have to be
able to supply?
Can we reduce the variance
• Look at placebo distribution
• P(allocate to placebo) is fairly uniform
• Length of whisker is just randomness
of allocation.
• Don’t block because ratios are 2 sig fig
(need blocksize of 100) and changed
every week.
• How about partial blocking?
Partial Blocking
Placebo:
Dose1:
Dose2:
Dose3:
Dose4:
Dose5:
Dose6:
25%
6%
9%
15%
26%
13%
6%
1 Placebo
1 Dose4
+ 2 of:
Dose1:
Dose2:
Dose3:
Dose4:
Dose5:
Dose6:
12%
18%
30%
2%
26%
12%
Treatment arms dropped or
introduced
• Trial may not adapt smoothly but only
at interims (1-4 a trial)
• At interim arms may be introduced, or
dropped
• Explore intermediate doses in area of
interest
• Extend range
• Drop ineffective doses
Response
Adding Doses At Interim
Initial Doses
Treatment arms dropped or
introduced
• Can we reduce manufacturing /
packaging and overage for arms that
are dropped?
• Can we avoid unnecessary re-supply
for expired batches?
• Can we avoid manufacturing /
packaging an arm that is not then
introduced?
Randomisation
• Will be central, no site based
• Need to have all possible doses at
each center
• Amount of wastage at centres that
don’t recruit at all will be all the greater.
Simulate the supply
• Model:
•
•
•
•
•
•
Centers
Packs
Subjects
Randomization
Shipments
Depots
Simulate the supply
Central Pharmacy
Depot 1
Depot 2
Center 1
Center 2
Center 3
Center 4
P(R1)=0.05
P(R2)=0.02
P(R3)=0.04
P(R4)=0.02
0.78
0.21
0.08
0.43
Simulate the supply
Central Pharmacy
Depot 1
Depot 2
Center 1
Center 2
Center 3
Center 4
P(R1)=0.05
P(R2)=0.02
P(R3)=0.04
P(R4)=0.02
0.66
0.14
0.97
0.01
Simulate the supply
Central Pharmacy
Depot 1
Depot 2
Center 1
Center 2
Center 3
Center 4
P(R1)=0.05
P(R2)=0.02
P(R3)=0.04
P(R4)=0.02
0.23
0.61
0.40
0.85
Simulate the supply
Central Pharmacy
Depot 1
Depot 2
Center 1
Center 2
Center 3
Center 4
P(R1)=0.05
P(R2)=0.02
P(R3)=0.04
P(R4)=0.02
0.08
0.72
0.78
0.37
Simulating adaptive trials
Central Pharmacy
Depot 1
Depot 2
Center 1
Center 2
Center 3
Center 4
P(R1)=0.05
P(R2)=0.02
P(R3)=0.04
P(R4)=0.02
Simulating adaptive trials
• Need to manage pack types
• Need to include the adaptive
randomisation – use output of
simulation of adaptive trial:
• Run supply simulations with many
different randomisations
Run Simulations before Trial
• Run 1000 simulations of the entire trial, with
no supply cap (2 seconds per simulation for
example of 20 centers x 400 days)
• Get distribution of:
• trial length
• number of lost subjects
• packs shipped from central pharmacy
• If number subject lost unacceptable, check
re-supply parameters & re-simulate
Chart the results of the
simulations
Probability of losing subject
1
0.9
0.8
Probability
0.7
0.6
Re-supply A
0.5
Re-supply B
0.4
Re-supply C
0.3
0.2
0.1
0
0
1
2
3
4
5
Subjects lost
6
7
8
>8
Check Total Required Supply
• Use number of packs shipped from
pharmacy to estimate required supply.
• Check over different scenarios that effect
pack usage (e.g. adaptive randomisation
scenarios)
• Check with simulation
• Estimate likely overage from simulations
Distribution of Required Supply
Number of High Dose Packs shipped over 100 simulations
35
30
% of times
25
20
15
10
5
0
370
380
390
400
410
420
430
Number of packs
440
450
460
470
Simulate what-if scenarios
• Frequency of re-supply to a patient
(pack sizes)
• Adaptive vs non-adaptive
• Manufacture/package all upfront or
make and initial stock and subsequent
batches
• Effect of batch expiry
Using the simulations during the
study
• Simulate forwards
• Do we have enough supply?
• Do have supply we can spare to another
study?
• When do we need that next batch?
• What if we add / remove centres from the
trial?
Treatment arms dropped or
introduced
• Monitor trial data regularly and prime manufacturing
/ packaging
• Extend treatments by using dose combination –
adding 1 dose to 0,2,4,6.
• Manufacture remainder after interim
• Use predictive power report to start manufacturing
before interim.
Initial Supply
Final Supply
Interim
Start of trial
Manufacturing
time
End of trial
Randomisation – Reduce
Wastage
• Use site based randomisation first, then
central randomisation
• Supply just in time – monitor the presence
of subjects in screening
• Pre-randomise subjects during screening
and supply only what is needed.
• Monitor and model recruitment rates during
the trial and auto adjust the re-supply rules
accordingly
• Close non-recruiting centers and re-allocate
supply
1. How to make and supply many
treatment arms
2. How much to make?
3. How much to supply?
Re-supply during the study
• What shipments should be sent today?
• Load current data
• recruitment rates
• shipments
• location of available stock
Treatment arms become
more / less likely to be
allocated
• How do we ensure there is enough
stock at centres for arms that are
becoming more likely to be allocated?
• How do we ensure that we don’t over
supply arms that are less likely to be
allocated?
Adaptive re-supply algorithm
• Re-supply using a Bayesian selftuning scheme.
• At each centre the stock required
is based on:
 Current stock.
 Packs in transit to the centre.
 Subjects likely to be recruited.
 Recruited subjects requiring fresh
supplies..
 Drug in stock which is about to expire
Adaptive re-supply algorithm (2)
• Calculate for a look ahead time plus the
time required to re-supply the centre.
• Use a maximum acceptable probability of
subjects being lost on recruitment - supplies
are dispatched if that (floor) level will be
exceeded.
• Shipment is sized to reduce the probability
of loosing a subject to below a lower
maximum acceptable probability (ceiling)
level.
Re-supply report
Current % randomization
to different arms
Current
supply
state
Adaptive
re-supply
Weekly
report
Current
patient
screening
data
The Wyeth Experience:
•
Working with Adaptive Partner, developed tool to
monitor site inventories:
• Tracked treatment inventories at site.
• Provided predicted requirements based on 99% and 95%
certainty of randomization revision.
• Predictions only based on patients within 4 days of end of
screening period to prevent calculating demand on dropped
patients.
• Provided “pick list” of supplies required by site to
accommodate the updated codes.
• Information was provided to Clinical Supplies one week prior
to having codes loaded into IVRS.
• NO FORCED TREATMENT ALLOCATION occurred in this
study!
And the winner is ...
• Additional supplies were manufactured, packaged
and stored at regional warehouses to accommodate
evolving supply demands
• Overall cost of drug supply for this study:
•
•
•
•
Cost of adaptive design:
Number of patient kits packaged:1440
Cost of traditional design:
Number of patient kits packaged: 686
$422,000
$201,000
• But, savings to Clinical for closing study 2 months
earlier and 180 less patients?
$1.5 million
Summary
• Adaptive trials are a challenge to
supply (but they’re worth it)
• More doses, less certainty
• Use better tools:
• simulator
• adaptive re-supply
• monitoring of patients in screening