BIO-STATISTICS
-AN IMPORTANT TOOL IN PHARMACUETICAL
PROGRAMMING
Soma Sekhar Sriadibhatla, GStat
09/25/2018
Presented to Syneos Health & Bristol-Myers Squibb Employees at Hopewell, NJ
The information in this presentation is based on presenter’s expertise, experience and
literature survey from several journals and research articles. The presenter has right to use
Gstat logo based on current accreditation from American Statistical Association.
DISCLAIMER
DRUG DEVELOPMENT
Phase 0 Phase I
Phase
IIB/III
Phase IV
Dosing/Toxicity
levels Safety/Dosage
Efficacy/Monitoring
of Adverse Events
Safety/Efficacy
Marketing
IND NDA
Drug
Discovery
70%
25-
30%
*Information and statistics from FDA
sBLA
Patients in
oncology
Results in
Lesser pt.
SCOPE
Design of Clinical trials
Sample Size determination,
Sampling, Study Population,
Selection Bias
Treatment Arms,
Randomization
Test of Hypothesis,
Descriptive Statistics
Endpoints like ORR using
International Myeloma Working
Group
Analysis : Baseline characteristics,
Safety analysis : G3/4 infusions
reactions
RANUNI
*Information from a study protocol
ENDPOINT
S
OS
Rand. To Death
from any cause
Precisely
measured;
Universally
Accepted
Larger
Studies; Non-
cancerous
Deaths
PFS/TTP
Rand. To Prog. Of
disease/Death
Smaller
Sample Size;
objective &
quantitative
Tumor
Assessment,
Deaths rand. Rel.
to tumor
progression
Survival, ITT
population
Not precise as
subjected to
selection bias/open
labelled; definitions
vary, Sensitive
https://www.fda.gov/downloads/drugsGuidanceComplianceRegulatoyInformation/Guidance/UCM071590.pdf
ORR
Single
arm/Rand.
Studies
ENDPOINTS …
Pt. tumor size
reduction/
predefined
amount/min. time
Sum of Partial
responses plus
complete responses
Can be
assessed in
single arm,
drug-
antitumor
Not
comprehensive
measure of drug
activity
DFS
Rand. To recurrence
of tumor or death
*https://www.fda.gov/downloads/drugsGuidanceComplianceRegulatoyInformation/Guidance/UCM071590.pdf
CENSOR =0;
EVENT DeathDt. – RANDt. + 1
CENSOR = 1 LassDt. – RANDt. + 1
Prevent loss of info. / Retain Org. Sample
Size
SURVIVAL
Estimation of Survivorship
Prob. Of survival
to a given time
point
Proc lifetest data =adpfs method = km;
Time pfstime*censor (0);
Strata trt;
Run;
No. of Events/KM PFS
https://www.pharmasug.org/proceedings/2016/DG/PharmaSUG-2016-DG03.pdf
TUMOR
ASSESSMENT
% change in
tumor size from
baseline at a visit
Water Fall; ORR
SURVIVAL…
Prob. of Events in
TRTA vs TRTB
1 vs 3 vs 0.33/ HAZARD’S
RATIO
Proc Phreg data = ;
Model sruvtime*censor(1);
Strata ;
Run;
Decrease/Increase in rate for parameter of
interest
https://www.pharmasug.org/proceedings/2016/DG/PharmaSUG-2016-DG03.pdf
Not valid when
both treatments
cross-interact(KM
plot)
Power
Analysis
80-90%
Ho: µa = µb
Ha: µa ne µb(one
sided/two sided)
STATS 101
=0.05;
µa-b/SD
Proc Power;
Twosamplemeans test = diff;
Meandiff =;
Alpha=;
Ntotal=;
Power=.;
Run;
Test of
Hypothesis
Ho: Ma = Mb
Ha: Ma > Mb(one
sided/two sided)
PROC
ANOVA/
GLM
Pvalue < 0.05 reject H0
Balanced vs Unbalanced/ Proc Mixed
PROC
ttest for
Means
http://support.sas.com/documentation/cdl/en/statugpssintroduction/61767/PDF/default/statugpssintroduction.pdf
One tailed vs
two tailed
Z vs T vs
F
SD??
Size??
Multiple
hypothesis; gate
keeping
procedure
STATS 101…
Pearson's
Coefficient
Correlation between two
variables
Proc freq;
tables /chisq (or CHM);
exact pcorr (or fisher) (OR);
Run;
Independence
Fisher’s
exact test
Association between two
cat.variables
Pvalue (when sample size is small)
Chi square (when sample size is large)
Two sample
vs One
Sample ttest
Two variables/one variable
When two variables are not independent/
Paired ttest
http://www2.sas.com/proceedings/sugi25/25/btu/25p069.pdf
CHM: Smoking vs
Lung Cancer;
stratified cat.data
Cell has zero??
KEEP LEARNING BIOSTATISTICS!!!
Design of
Experiments
Sampling
techniques Descriptive
Statistics
Significance
Inferential
Statistics
Survival
analysis,
Hazards Ratio
Count more;
Proc Freq;
Run;

Biostatistics

  • 1.
    BIO-STATISTICS -AN IMPORTANT TOOLIN PHARMACUETICAL PROGRAMMING Soma Sekhar Sriadibhatla, GStat 09/25/2018 Presented to Syneos Health & Bristol-Myers Squibb Employees at Hopewell, NJ
  • 2.
    The information inthis presentation is based on presenter’s expertise, experience and literature survey from several journals and research articles. The presenter has right to use Gstat logo based on current accreditation from American Statistical Association. DISCLAIMER
  • 3.
    DRUG DEVELOPMENT Phase 0Phase I Phase IIB/III Phase IV Dosing/Toxicity levels Safety/Dosage Efficacy/Monitoring of Adverse Events Safety/Efficacy Marketing IND NDA Drug Discovery 70% 25- 30% *Information and statistics from FDA sBLA Patients in oncology Results in Lesser pt.
  • 4.
    SCOPE Design of Clinicaltrials Sample Size determination, Sampling, Study Population, Selection Bias Treatment Arms, Randomization Test of Hypothesis, Descriptive Statistics Endpoints like ORR using International Myeloma Working Group Analysis : Baseline characteristics, Safety analysis : G3/4 infusions reactions RANUNI *Information from a study protocol
  • 5.
    ENDPOINT S OS Rand. To Death fromany cause Precisely measured; Universally Accepted Larger Studies; Non- cancerous Deaths PFS/TTP Rand. To Prog. Of disease/Death Smaller Sample Size; objective & quantitative Tumor Assessment, Deaths rand. Rel. to tumor progression Survival, ITT population Not precise as subjected to selection bias/open labelled; definitions vary, Sensitive https://www.fda.gov/downloads/drugsGuidanceComplianceRegulatoyInformation/Guidance/UCM071590.pdf
  • 6.
    ORR Single arm/Rand. Studies ENDPOINTS … Pt. tumorsize reduction/ predefined amount/min. time Sum of Partial responses plus complete responses Can be assessed in single arm, drug- antitumor Not comprehensive measure of drug activity DFS Rand. To recurrence of tumor or death *https://www.fda.gov/downloads/drugsGuidanceComplianceRegulatoyInformation/Guidance/UCM071590.pdf
  • 7.
    CENSOR =0; EVENT DeathDt.– RANDt. + 1 CENSOR = 1 LassDt. – RANDt. + 1 Prevent loss of info. / Retain Org. Sample Size SURVIVAL Estimation of Survivorship Prob. Of survival to a given time point Proc lifetest data =adpfs method = km; Time pfstime*censor (0); Strata trt; Run; No. of Events/KM PFS https://www.pharmasug.org/proceedings/2016/DG/PharmaSUG-2016-DG03.pdf
  • 8.
    TUMOR ASSESSMENT % change in tumorsize from baseline at a visit Water Fall; ORR SURVIVAL… Prob. of Events in TRTA vs TRTB 1 vs 3 vs 0.33/ HAZARD’S RATIO Proc Phreg data = ; Model sruvtime*censor(1); Strata ; Run; Decrease/Increase in rate for parameter of interest https://www.pharmasug.org/proceedings/2016/DG/PharmaSUG-2016-DG03.pdf Not valid when both treatments cross-interact(KM plot)
  • 9.
    Power Analysis 80-90% Ho: µa =µb Ha: µa ne µb(one sided/two sided) STATS 101 =0.05; µa-b/SD Proc Power; Twosamplemeans test = diff; Meandiff =; Alpha=; Ntotal=; Power=.; Run; Test of Hypothesis Ho: Ma = Mb Ha: Ma > Mb(one sided/two sided) PROC ANOVA/ GLM Pvalue < 0.05 reject H0 Balanced vs Unbalanced/ Proc Mixed PROC ttest for Means http://support.sas.com/documentation/cdl/en/statugpssintroduction/61767/PDF/default/statugpssintroduction.pdf One tailed vs two tailed Z vs T vs F SD?? Size?? Multiple hypothesis; gate keeping procedure
  • 10.
    STATS 101… Pearson's Coefficient Correlation betweentwo variables Proc freq; tables /chisq (or CHM); exact pcorr (or fisher) (OR); Run; Independence Fisher’s exact test Association between two cat.variables Pvalue (when sample size is small) Chi square (when sample size is large) Two sample vs One Sample ttest Two variables/one variable When two variables are not independent/ Paired ttest http://www2.sas.com/proceedings/sugi25/25/btu/25p069.pdf CHM: Smoking vs Lung Cancer; stratified cat.data Cell has zero??
  • 11.
    KEEP LEARNING BIOSTATISTICS!!! Designof Experiments Sampling techniques Descriptive Statistics Significance Inferential Statistics Survival analysis, Hazards Ratio Count more; Proc Freq; Run;