C t ib tiC t ib ti ff Bi t ti tiBi t ti ti ttContributionContribution of of BiostatisticsBiostatistics to to Designing Clinical TrialsDesigning Clinical TrialsDesigning Clinical Trials Designing Clinical Trials

Satoshi Morita, PhDSatoshi Morita, PhDDept. of Biostatistics and Epidemiology, Dept. of Biostatistics and Epidemiology,

Yokohama City UniversityYokohama City UniversityYokohama City UniversityYokohama City University

An elephant in the living roomAn elephant in the living roomA li i l i l i diA li i l i l i diA clinical trial in an uncommon diseaseA clinical trial in an uncommon disease

Low incidence Slow accrual

El#1

Low incidence Slow accrual

Conducting a clinical trial inConducting a clinical trial in such a disease often infeasible

In addition, if low mortality and morbidity

Impractical to achieve adequate power

Another elephant in the living roomAnother elephant in the living roomN i ll d d iN i ll d d iNo universally accepted endpointNo universally accepted endpoint

Malignant ascites secondary to gastric Cancer: severe end-stage manifestation

of the disease

El#2

of the disease,

poses particular problems to cliniciansEl 2 poses particular problems to clinicians in terms of providing suitable treatment.

HoweverHowever NONO universally accepted measureuniversally accepted measure forforHowever, However, NONO universally accepted measure universally accepted measure for for assessing response to therapy.assessing response to therapy.

Why? No measurable lesions!!Why? No measurable lesions!!Why? No measurable lesions!!Why? No measurable lesions!!Unable to be assessed by the RECIST criteria.Unable to be assessed by the RECIST criteria.

Today I would like to talk aboutToday I would like to talk aboutToday, I would like to talk about…Today, I would like to talk about…

How biostatistics can contribute to designing and analyzing clinical trials in uncommonand analyzing clinical trials in uncommon diseases?

What biostatisticians have done and are doing for it?doing for it?

Examples: Cancer clinical trialsExamples: Cancer clinical trialspediatric cancer, sarcomasmalignant ascites (gastric cancer)

Uncommon diseaseUncommon diseaseEx. Pediatric cancerEx. Pediatric cancer

Pediatric doctors often rely on evidence from adultPediatric doctors often rely on evidence from adult clinical trials.

It is natural to consider “borrowing strength” from previous or simultaneous adult studies.previous or simultaneous adult studies.

Clinical trial designs

Prior InformationDose-level(s)Sample size required

External datap q

Trial monitoring (Stop/go decision)(Stop/go decision)

Statistical plan

Ex. Pediatric phase I trial in Ex. Pediatric phase I trial in metastatic sarcomametastatic sarcoma

Patients with stage IV Ewing’s sarcoma rhabdomyosarcomastage IV Ewing s sarcoma, rhabdomyosarcomaage 1 – 30 yrs

Objective:Objective:to establish “pediatric dosage” of topotecan in combination with Cyclophosphamide (fixed dose)in combination with Cyclophosphamide (fixed dose)

and Melphalan (fixed dose)

Using prior information / knowledgeUsing prior information / knowledgeUsing prior information / knowledgeUsing prior information / knowledgePrior information:

Optimal dose in adults: 4 mg/m2/dayOptimal dose in adults: 4 mg/m /dayClinical trials in other cancers (ovarian cancer multiple myeloma )(ovarian cancer, multiple myeloma,…)

In the present study,Two dose levels of topotecan: 3.5 and 3.0p

# of patients: at most 20 (Pt accrual: >2 yrs)What if more than two levels?What if more than two levels?

Much more patients! Study design / statistical methodMuch longer time!! Continual reassessment method

A “Bayesian” approach

A phase II trial for sarcomaA phase II trial for sarcomaA phase II trial for sarcomaA phase II trial for sarcomaThall et al (2003) present a Bayesian study designThall et al. (2003) present a Bayesian study design

for a single-arm phase II trial to examine the efficacy of the targeted drug imatinib for sarcomaefficacy of the targeted drug imatinib for sarcoma with many subtypes.

Sarcoma is uncommon

The goal was to construct a design that allowed the efficacy of imatinib to be evaluatedthe efficacy of imatinib to be evaluated

in the multiple subtypes.

Subgroups?Subgroups?

Borrowing strength!!Borrowing strength!!Borrowing strength!!Borrowing strength!!

Hierarchical Bayesian modelHierarchical Bayesian model

1 2 3 4 5 6 7 8 9 1. Synovia

. Leiomyo

3. Malignahistiocy

. Fibrosar

. Liposarc

. Ewing’s

. Osteosa

8. Rhabdosarcom

9. Periphesheath sa

0. Angiosl sarcoma

osarcoma

nt fibrousytoma

rcoma

coma

sarcoma

arcoma

myoa ral nervearcoma

sarcoma

a

Borrowing strength (cont’d)Borrowing strength (cont’d)Borrowing strength (cont d)Borrowing strength (cont d)G / N d i iGo / No go decision

Conventional design

Hierarchical B i d ldesign

(repeating separate trials)Bayesian model

0/80/5 1/4 3/7 2/8

0/80/5 1/4 3/7 2/8

3/9 2/8 1/8 2/4 1/5

Stop? Continue?

Borrowing strength (cont’d)Borrowing strength (cont’d)Borrowing strength (cont d)Borrowing strength (cont d)The number of patients requiredThe number of patients required.

Larger # of patients

Hierarchical Conventional design Bayesian modeldesign

(repeating separate trials)

More efficient!!

Bayesian approaches!!Bayesian approaches!!Bayesian approaches!!Bayesian approaches!!D it thD it th i i b f ti i b f tDespite the everDespite the ever--increasing number of new agents, increasing number of new agents,

the number of patients available for clinical trials the number of patients available for clinical trials remains limited. remains limited.

There is a growing need for statistical methodologies There is a growing need for statistical methodologies g g gg g gthat can “that can “rapidly and efficientlyrapidly and efficiently” evaluate the ” evaluate the clinical efficacy and safety of new agents.clinical efficacy and safety of new agents.y y gy y g

Bayesian approachBayesian approachBayesian approachBayesian approach

Why Bayesian approaches?Why Bayesian approaches?Why Bayesian approaches?Why Bayesian approaches?

B i h d iB i h d iBayesian methods incorporateBayesian methods incorporateprevious study dataprevious study dataprevious study dataprevious study dataprepre--clinical (animal) dataclinical (animal) data------------i i f tii i f tias prior information.as prior information.

Study design and Data analysisStudy design and Data analysis

Bayesian approach:Bayesian approach:y ppy ppUpdating a prior by observed dataUpdating a prior by observed data

Previous studies

P t iClinicians’ experience

Prior Data Posterior＋

Trial monitoringGo / No go decision

Observed data (+(external) data collected Go / No go decision( (external) data collected

simultaneously)

A pA p value: data from 12 patientsvalue: data from 12 patientsA pA p--value: data from 12 patientsvalue: data from 12 patients

Clinically NOT meaningful: response rate < 25%

4 responses12 ti t

Response rate = 33%12 patients

p

p-value = 0.36

NOT significant? Negative data?

P-valueGOD for clinicians?

Source: Rimm &Bortin, ,

Biomedicine, 1978

P-valueDEVIL for li i i ?clinicians?

ＥＢＭ

Source: Rimm &Bortin, Biomedicine, 1978

Bayesian analysis of Bayesian analysis of Prior distributionPrior distributionresponse rateresponse rate

High posterior probability (Response rate > 0 25)High posterior probability (Response rate > 0.25)

Go to the next phase trial

Prior distribution

Response rate

Prior → PosteriorPrior → PosteriorResponse rateResponse rate

Posterior distribution

Posterior probability (Response rate > 0 25)

Prior + 1 / 3 pts

Posterior probability (Response rate > 0.25)= 60%

Response rate

Prior → PosteriorPrior → PosteriorResponse rateResponse rate

Posterior distributionPrior + 4 / 12 pts

P t i b bilit (R t 0 25)Posterior probability (Response rate > 0.25)= 80%

Response rate

Developing a new endpoint to assess Developing a new endpoint to assess t th f li t itt th f li t itresponse to therapy for malignant ascitesresponse to therapy for malignant ascites

M li t itMalignant ascitesbrings a rapid deterioration(Pain, loss of appetite, obstructive symptoms, dyspenea)

A significant negative impact on performance statusand quality of life of patients

Patient-centered response is thereforePatient centered response is therefore more important than tumor response.

How to measure the response to therapy?How to measure the response to therapy?p pyp py

“G ld t d d”“Gold standard”:Ascitic volume changes by 3D-CT

( i i ll d t ti t th l f(originally used to estimate the volume of organs before and after transplantation)

But, practical limitation…time- and resource-demanding (special equipment, trained radiologists (over 30 mins))

How to measure the response to therapy? How to measure the response to therapy? (cont’d)(cont’d)(cont d)(cont d)

Approximating the ascitic volume:Approximating the ascitic volume:The Five-Point method, b d ti l CT ibased on conventional CT images

BUT, this alternative does not account for “patient centered changes”patient-centered changes

Use the broadly accepted concept of “Clinical Benefit Response”Clinical Benefit Response

in pancreatic cancer as a prototype!!

DiureticFrequency of abdominal Girth of Diuretic

consumptionabdominal paracentesis

Girth of abdomen

Ascites fluid

Non-responder Increase in ascites fluid

StableNo change

Responder Decrease in ascites fluid

PS

Non-responder deterioration in PS

Stableno change

Responder improvement in PS

Clinical Benefit ResponseClinical Benefit Response--Gastric CancerGastric Cancer

Ascites fluidAscites fluidAscites fluidAscites fluid

＋＋ StableStable －－

＋＋ ResponderResponder

PSPS StableStablePSPS StableStableNonNon--responderresponder

2626－－

Still challengingStill challengingStill challenging…Still challenging…Bayesian approaches have the potential ofBayesian approaches have the potential of

making clinical trials in uncommon diseasesfeasiblefeasible.“FDA Guidance for the Use of Bayesian Statistics

in Medical Device Clinical Trials”However, how to incorporate external information , p

is NOT fully established.

New endpoint developmentUseful and appealing, but many steps and longUseful and appealing, but many steps and long

time to establish a new endpoint.