Population Approach For Clinical Drug Safety Assessment in Regulatory Decisions

1
Population Approach For Clinical Drug Safety
Assessment in Regulatory Decisions

• He Sun, Ph.D.
• CDER, FDA

2
Questions to committee

• When should ADR data be treated as continuous
  variable and when it should be treated as
  categorical variable ? What are the bases for
  the preference, if any?
• What are the utility and general limitations of
  using PopPK approach to determine individual /
  special population exposure parameters followed
  by subsequent E-R analyses ?
• What are the general considerations in E-R based
  dose adjustment for special populations ?

3
Exposure-safety (E-R) data types
4
E-R analysis method and process

• Managing/Editing data
• Conducting population PK analysis
• Establishing the base model
• Modeling variabilities
• Searching for significant covariates
• Validating the model, if needed
• Determining individual exposure parameters
• Via post-hoc estimation
• Via simulation
• Deriving secondary exposure parameters
• AUC, Tmic, Ce, and etc.
• Accumulative exposure

5
E-R analysis method and process (cont.)

• Determining E-R relationship for individuals
  and/or special populations
• Classification methods
• Classification and regression tree methods
• Logistic regression methods
• Modeling methods
• Considering accumulative exposure time
• Performing statistical analysis on response for
  significance, per exposure or special populations
• Proposing needed dose adjustment for special
  populations based on E-R analysis results

6
A example background

• Study Design Two phase 3 trials with 1500
  evaluable drug-treated patients receive multiple
  doses, X, 2x and 4X. Patient plasma drug
  concentration measures are sparse and unbalanced.
• End points Safety measures were blood chemistry
  variables (continuous variable) with pre-defined
  values (cut-off point) for claiming No ADR. ADR
  also presented as Yes/No for headache, liver
  toxicity, photo-toxicity and etc. ADRs were
  future classified as mild, moderate and severe.
• PK results Drug-drug interaction increases AUC
  by 310, Cmax by 144 . AUC and Cmax are 132
  and 79 higher in elderly vs. young healthy
  subjects. AUC and Cmax dropped by 50 in Blacks.
• Safety results The total frequency of ADRs
  ranged from 0.1 to 17. Severe cases were rare.
  There were no death.
• Efficacy results Clinically efficacy was well
  demonstrated from the trials. Exposure -
  efficacy relationship was unclear.

7
E-R analysis methods and results

• Classification on frequency
• E data was divided into equally populated
  segments, or
• E data was divided into equally distanced
  segments
• The frequency of ADR was then determined for each
  E segments
• e.g. Total ADR is 18 when AUC gt 1200, 5
  when AUC lt 1200
• Classification on severity
• R data was classified for severity (e.g. Severe,
  moderate and ,mild)
• The average severity of ADR was then calculated
• e.g. Severe ADR occurs when Cmax gt10, mild
  ADR apparent when Cmax lt2 with similar
  frequencies
• Regression and Classification Trees
• R data are successively split along coordinate
  axes of the E variables which maximally
  distinguishes the R variable.
• Classification tree. First split on ADR
  frequency was at AUC 871. Total ADR was 23
  when AUC gt871, 2.5 when AUC lt871
• Regression tree on WBC count.

8
E-R analysis methods and results (cont.)

• Regression models on relationships
• Statistical vs. (semi) physiological based models
• Linear, nonlinear models
• Fixed effect, mixed-effect models
• Logistic regression
• Relationship between E and each R. Odds ratios
  were determined along with 95 CI, in which the
  odds ratio denotes the scalar multiple for the
  likelihood of given ADR associated with unit
  change in E.
• e.g. The odds ratio of acute tissue rejection
  increase 23 when AUC0-24 decrease 10.
• Modeling
• HDL drops below normal on day 95 when Cavg gt10
  ug/ml for patients with high body weight and low
  initial HDL baseline
• Back Pain score (in 0-10) is nonlinearly
  correlated with plasma drug daily total AUC, the
  number of treatment days, and dose regimen (tid
  vs. bid).

9
E-R analysis methods and results (cont.)

• Modeling (cont.)
• QTc prolongation. The relationship between QTc
  and drug C was described by a Emax model.
  Population E0 410 msc., Emax 435 msc. and
  ED50 10 ug/ml. E0 and Emax are gender and age
  dependent.
• Photo toxicity Yes on day 10 when Cavg on day 10
  greater than 8 ug/ml for female patients.
• Liver toxicity frequency linearly correlate with
  Cint and time of exposure.
• Blood chemistry score (in 0-10) is nonlinearly
  correlated with plasma drug AUC and the number of
  treatment days.
• Blood chemistry toxicity nonlinearly increase as
  Cavg increase, and TD50 is correlated with drug
  Cmax.
• Descriptive
• Hypercholesterolemia, defined as gt6.5 mmol/liter,
  ranged from 76 to 87 over the exposure range
  without a significant relation to Cmin (P0.37).

10
E-R analysis methods and results (cont.)

• Descriptive
• Hypertriglyceridemia, defined as gt2.9 mmol/liter,
  rose from 59 to 77 across the exposure groups
  (P0.02).
• Leukocytopenia, defined as lt4x10(9)/liter,
  occurred in 11-19 of patients across the
  exposure quintiles showing no relationship to
  Cmin (P0.76).
• Thrombocytopenia, defined as lt100x10(9)/liter,
  occurred in lt10 of patients in the first 3 Cmin
  quintiles and was 23 and 28 in Cmin quintiles 4
  and 5 (P0.21).
• Because hyperlipidemias responded to
  counter-measure therapies and thrombocytopenia
  had an overall low incidence of 12, drug-related
  ADRs were manageable up to the highest troughs
  observed in this population of 5 ug/ml.
• Dose adjustment in special populations The
  average upper therapeutic limits are likely to be
  10 ug/ml for Cmax and 871 for AUC. Female
  subject more sensitive to phototoxicity and Cavg
  lt 8 ug/ml was considered. bid dose regimen is
  preferred.

11
Questions to committee

• What are the utility and general limitations of
  linking PK obtained from population analysis to
  response endpoints ?
• What are the general considerations in E-R based
  dose adjustment for special populations ?
• When should ADR data be treated as continuous
  variable and when it should be treated as
  categorical variable ? What are the bases for
  the preference, if any?