36
Borko Jovanovic, MS, PhD Biostatistician [email protected] 14 June 2013

Borko Jovanovic, MS, PhD Biostatistician [email protected]

  • Upload
    others

  • View
    2

  • Download
    0

Embed Size (px)

Citation preview

Borko Jovanovic, MS, PhD Biostatistician

[email protected]

14 June 2013

Topics Celebration of statistics

Randomization and Randomness Dropouts and substitution Why stop early? Accounting for dropouts in analysis

The year of Statistics http://www.youtube.com/watch?v=nTBZuQR7dRc&fe

ature=youtu.be

2013 is the Year of Statistics as per United Nations proclamation

“Statistician” = most looked after profession since 2010

www.statistics2013.org www.amstat.org www.chicagoasa.org

Who created some of these …? p-value

2x2 table, Chi-square test

t-test

Mantel-Haenszel stratification for 2x2 tables?

R.A. Fisher (1890-1962) “Statistical Methods for Research Workers” (1923) p-value

Suspected 2x2 table creator Karl Pearson circa 1904

Known creator of the t-test William Sealy Gosset circa 1908

William Haenszel took my class in 1990’s: The Proof

What the IRB has to say …..As a subject in this study you will be randomly (like

a flip of a coin) assigned to one of the following treatment groups:

Heads or tails?

Randomize me

Randomization and randomness Randomization and Randomness

Why stop early?

Dropouts and substitution

Collecting Data

Randomization and Randomness The Oxford English Dictionary defines 'random'

as: "Having no definite aim or purpose; not sent or guided in a particular direction; made, done, occurring, etc., without method or conscious choice; haphazard."

Randomization in clinical trials Can not safely guess treatment assignment of next

patient

Fairness in Randomization Takes away a bias in assignment

Equalizes all observable and unobservable variables so

that proper comparison of treatments can be made

To see how risks depend on treatment arms

“I do not want that placebo” “We need you in the study so we can decide if treatment works”

A two arm clinical trial

Clinical trials Randomization ascertains “equal (unbiased)

treatment” Patients arrive at times t = 1, 2, 3, 4, 5, 6, 7, 8, 9, 10 Possible assignments of 5 per each of 2

groups group are a, b, b, a, a, b, a, a, b, b a, a, a, a, b, b, b, a, b, b a, a, a, a, a, b, b, b, b, b a, b, a, b, a, b, a, b, a, b

Block Randomization Blocks of 2: ab, ba (two) Blocks of 4: aabb, abab, baab, baba, abba,

bbaa (six) Blocks of 6: aaabbb, aababb, … bbbaaa

(fifteen) Useful when trial is to be stopped early Superiority of one treatment stop Excessive toxicity of one treatment stop Lack of funding stop

Final assignment aabb ab abab bbaaba baba 1234 56 78910 111213141516 17181920

Comes out as an Excel sheet

In Excel patient treatment

1 a 2 a 3 b 4 b 5 a 6 b 7 a 8 b 9 a 10 b 11 b 12 b 13 a 14 a 15 b 16 a 17 b 18 a 19 b 20 a

Randomized blocks Provide balance between

assignments whenever the trial is stopped

IRB about dropping out If you choose to be in this study, you have

the right to be treated with respect, including respect for your decision whether or not you wish to continue or stop being in the study. You are free to choose to stop being in the study at any time.

Dropping out is bad for the system

What happens if a patient drops out from the study? Need a replacement

Two cases: A) blinded study B) un-blinded study

Consider a blinded study, drug and placebo look alike,

treatment duration and final evaluation = 30 days

Replacement in blinded study patient treatment date reg date drop date repl new num order repl

1 drug 2 18 40 1a 22 drug 33 placebo 104 placebo 12 15 30 4a 15 drug 226 placebo 247 drug 348 placebo 359 drug 35

10 placebo 3611 placebo 4512 placebo 5013 drug 51 60 65 13a 314 drug 5515 placebo 5616 drug 5717 placebo18 drug19 placebo20 drug

Do not encourage dropping out At the onset of consenting

Replacement in blinded study Since no one knows the assignment(other than the

pharmacy or un-blinded drug preparer) just add patients to vacated slots in order of them dropping out, and before adding new patients

The balance of randomization will remain exactly as it was

By all means preserve retention by explaining the importance of their study participation and adherence to the protocol

Un-blinded study Simple replacement can not work as we do know the

treatment assignment ahead of time

We must use extra random assignments

Un-blinded study randomization list

Replace 4, 1, 13 By reserve 21, 22, 23

patient treatment1 drug2 drug3 placebo4 placebo5 drug6 placebo7 drug8 placebo9 drug

10 placebo reserve treat11 placebo 21 placebo12 placebo 22 drug13 drug 23 placebo14 drug 24 drug15 placebo 25 drug16 drug 26 placebo17 placebo18 drug19 placebo20 drug

Replacements will provide an approximate adherence to randomization

patient treatment1 drug2 drug3 placebo4 placebo5 drug6 placebo7 drug8 placebo9 drug

10 placebo reserve treat replaces11 placebo 21 placebo 412 placebo 22 drug 113 drug 23 placebo 1314 drug 24 drug15 placebo 25 drug16 drug 26 placebo17 placebo18 drug19 placebo20 drug

Stopping study for toxicity, efficacy, lack of funds There is always an imaginary boundary which we

should not cross to stay in the game Too toxic – look into in Phase 1 trials Not efficacious enough – Phase 2, 3 trials Super efficacious – Phase 2, 3 trials Funds run out – all phases

Stopping for efficacy – must cross boundary

Do not feel bad for stopped studies Although you have put so much time and energy into your

study, it can be stopped for many reasons.

Ensure that the study terminates properly and safely.

“Any new findings developed during the course of this research that may affect your willingness to continue in this study will be shared with you. Your participation in this study may be stopped by the investigator without your consent if he thinks it is in your best interest not to participate.”

Stopping for efficacy – various boundaries control for Type 1 error

Data collection After much time and money, data are collected and

everyone's work is summarized in one Excel file

Research coordinators can make sure all data are entered in a way understandable to ‘everyone’

Bad files

id age Any drop drop 5-10 drop > 202 48 0 0 03 54 1 1 05 63 0 0 19 63 1 1 010 56 1 1 012 64 1 1 113 62 1 1 014 60 1 1 018 53 0 0 n/a20 60 1 0 11 40 0 0 04 56 1 0 .6 63 1 1 17 68 1 1 08 54 1 011 62 1 0 115 50 0 0 016 33 0 0 na17 51 1 1 NA19 50 1 0 0

Better files

id gender age Any dropdrop 5-10 drop > 20 possible2 f 48 0 0 03 f 54 1 1 05 f 63 0 0 19 f 63 1 1 010 f 56 1 1 012 f 64 1 1 113 f 62 1 1 014 f 60 1 1 018 f 53 0 0 n/a 020 f 60 1 0 11 m 40 0 0 04 m 56 1 0 . 16 m 63 1 1 17 m 68 1 1 08 m 54 1 0 111 m 62 1 0 115 m 50 0 0 016 m 33 0 0 na 017 m 51 1 1 NA .19 m 50 1 0 0

Between a rock and a hard place Enter num=0.00000567 Or 10,000xnum =0.567 Or 100,000xnum = 5.67 Or 1,000,000xnum=56.7 Etc…And remember to tell 3 years later

Enter results for Applicable Clinical Trials into

Clinicaltrials.gov within ONE YEAR of study completion for your primary outcome measure and I can assist you with the process.

Clinical Research Coordinators

Fundamental in the process of research

Essential in recruitment, registration, randomization, retention (the 4R’s)

And the 5th R would be a round of applause for all of you attending this conference today!!!!

Statisticians never sleep