Randomized Trials Outcomes and Adverse Events

1
Randomized TrialsOutcomes and Adverse Events

• Steven R. Cummings, MD
• Director, UCSF Coordinating Center
• Assistant Dean for Clinical Research

2
Fracture Avoidance Trial (FAT)

• Potential outcomes
• All diagnosed fractures
• Symptomatic vertebral fractures
• New vertebral fractures detected by x-ray
• Osteoporosis Quality of Life scale
• Number of days of disability due to fracture
• Height loss

3
How to start?

• Measure all of these outcomes
• Call one primary and assess all the others as
  secondary outcomes

4
Why one primary outcome?

• To calculate sample size
• Gives that outcome more credibility
• In general, the FDA requires that an outcome be
  primary in order to approve a drug for that
  indication
• Primary vs. secondary is artificial, but useful.
  Credibility should derive from plausibility.

5
Which primary for FAT?

• Potential outcomes
• All diagnosed fractures
• Symptomatic vertebral fractures
• New vertebral fractures detected by x-ray
• Osteoporosis Quality of Life scale
• Number of days of disability due to fracture
• Height loss

6
Which Primary Outcome?

• Alternatives
• All fractures
• Symptomatic vert fx
• New vert fx on x-ray
• Quality of Life scale
• Days of disability
• Height loss

• Considerations
• Most clinically important
• Inexpensive measure
• Smallest shortest study
• Can be used for FDA approval

7
Which Primary Outcome?

• Alternatives
• All fractures
• Symptomatic vert fx
• New vert fx on x-ray
• Quality of Life scale
• Days of disability
• Height loss

• Considerations
• Most clinically important
• Inexpensive measure
• Smallest shortest study
• Can be used for FDA approval
• Other assessments included as secondary

8
Why not make bone density the primary outcome?

• Vertebral fracture on x-ray requires 2,000
• BMD as primary requires lt 200 women

9
Why not make bone density the primary outcome?

• Vertebral fracture on x-ray requires 2,000
• BMD as primary requires lt 200 women
• Not yet accepted as a valid surrogate
• What makes a surrogate valid?

10
Why not use BMD as a surrogate?

• A valid surrogate is
• Strongly associated with the outcome
• Treatment induced changes in the surrogate
  consistently predict changes in the clinical
  outcomes
• Believed and accepted

• BMD
• Low BMD strongly predicts fractures
• Treatment induced changes in BMD underestimate
  reduction in fractures. Inconsistently.
• FDA does not accept it

11
How about combining events?

• Composite endpoints
• Increase number of events
• Improve power unless they dilute effect
• Must reflect the same (or very similar)
  underlying biology

• Combining all fractures
• Doubles the number of events
• Txs decrease vert fxs 50, other types 0-25
• Fractures have different relationships to bone
  density and trauma

12
One more thing...

• Women can suffer recurrent fractures
• Alternatives
• Number of fractures
• Number of women who suffer at least one fracture
• Time to first fracture

13
Count subjects or events?

• Difficult issue
• Counting multiple events increases power
• Conservative approach count subjects
• Because events cluster in subjects, are not
  statistically independent. Counting events tends
  to overestimate the effect.

14
On the other hand...

• Counting subjects (or time to first event)
• Ignores effect of treatment on recurrent events
• Can underestimate the long-term effect of
  treatment by depletion of susceptibles.

15
Depletion of susceptibles

• Assume a randomized trial of a treatment to
  prevent fractures 100 pbo vs. 100 treatment
• 50 subjects susceptible 50 would NOT fx
• No treatment (on placebo) 20 of susceptibles
  fracture/year
• Treatment reduces risk of fracture 50 in
  susceptibles, year after year

16
Depletion of susceptibles

• 100 pbo vs. 100 treatment
• No treatment (on placebo) 20 fracture/year
• Treatment reduces fracture 50 year after year

17
Depletion of susceptibles underestimates
long-term effects

• PBO TX
• Fx N (susc) Fx N (susc) RR
• Baseline 100 (50) 100 (50)
• Year 1 10 90 (40) 5 95 (45) 0.5
• Year 2 8 82 (32) 5 90 (40)
• Year 3 6 76 (26) 4 86 (36)
• Year 4 5 71 (21) 4 82 (32) 0.7

18
The lessons

• Keep subjects in treatment and follow-up to the
  degree it is ethical
• Dont stop after 1st event assess recurrent
  events
• Be careful about estimating long-term effects of
  treatment
• Analyze effect on recurrent outcomes
• Consider frailty models (time between events
  rather than time to the first event)

19
Adverse Events

• Alternative approaches
• Elicited vs. volunteered
• Simple counts vs. severity
• At the end vs. along the way
• The FDA system
• Serious AEs
• Attribution to the study treatment

20
Approaches to AEsVolunteered vs. elicited

• Pro elicited
• Standardized
• More sensitive
• Easier to code
• Con
• Miss unexpected AEs
• More positives
• Milder, less certain cases

• Pro volunteered
• Catch unexpected AEs
• Fewer data to code
• Finds serious cases
• Con
• Unstandardized
• Less sensitive misses cases
• Hard to code

21
Which approach is most likely to find real AEs?

• Evidence is mixed
• Sensible approach standard questions to elicit
  uncommon AEs known to be related to drug.
• Additional open ended questions to capture
  unexpected AEs.

22
The Bunion Problem

• FIT Trial of alendronate in 6,400 women for 4
  years
• Recorded over 20,000 episodes of URIs (and
  thousands of reports of bunions!)
• Enormous data management effort and cost
• How could this be avoided?

23
How to minimize nuisanceAEs

• Elicit uncommon, plausible and important AEs
• Limit collection of minor AEs to samples of
  subjects

24
FDA AE classifications

• Serious AEs
• Deaths
• Hospitalized overnight
• Cancer (except skin cancer)
• Birth defects
• SAEs definitely or probably due to study drug
  must be reported to company and by the company to
  FDA in 24

25
Attribution

• Serious AEs must be classified as
• Definitely
• Probably
• Possibly, or
• Not...
• ...related to the study drug
• This is only required of SAEs

26
Attribution

• Attributions to drug as generally as likely with
  placebo as with active drug