How to Talk to Your Statistician

1
How to Talk to Your Statistician Reaching a
Common Understanding
George B. Stoms Vital Systems,
Inc.George.Stoms_at_VTLsys.com [phone]
x222
2
Before we begin
KeepingRoles Responsibilitiesin Perspective
3
Before we begin
The Statisticians Nightmare
4
Before we begin
Every Non-Statisticians Dream?
5
UMC User Group Survey
A recent survey of WHO Drug Dictionary Users
provided some insight into the state of use
within companies. As we go through this, please
think about what is done at your
company. Source WHO Drug Dictionary - European
User Group Presentation by Petra Barth
6
Preliminary result from survey

• Generally happy with the quality and quantity of
  the communication between coders analysis but
• involve coders in meetings
• Coders should be included in meetings
• Coders and analysis team not always invited or
  present at all internal team meetings
• The coders (dictionary experts) should be brought
  in more often for consultation.
• An actual meeting with statistics discussing the
  analysis plan, dictionary version being used,
  conventions,
• Additional training would be helpful so that we
  each understand one another's jobs. The industry
  tends to work in silos and not see the big
  picture.

7
The expert
In your organization who is the dictionary expert? In your organization who is the dictionary expert? In your organization who is the dictionary expert? In your organization who is the dictionary expert?
Senior coder   57
Clinical Data Manager   24
Statistician   5
Other...   39
8
The expert
Is the dictionary expert and/or coder involved in Yes No
The development of the Study Protocol
The development of the Study Protocol 18 70
The definition of Medications of Interest
The definition of Medications of Interest 47 40
The Statistical Analysis Plan
The Statistical Analysis Plan 20 70
The definition of tables that include concomitant medication
The definition of tables that include concomitant medication 40 47
The actual analysis of concomitant medication data at study end
The actual analysis of concomitant medication data at study end 28 62
9
ATC assigment
Does your company require the coders to select an ATC class that corresponds to the indication? Does your company require the coders to select an ATC class that corresponds to the indication? Does your company require the coders to select an ATC class that corresponds to the indication? Does your company require the coders to select an ATC class that corresponds to the indication?
Yes   86
No   9
Don't know   5
10
ATC - process
What is the process used for manual ATC coding? What is the process used for manual ATC coding? What is the process used for manual ATC coding? What is the process used for manual ATC coding?
Select an ATC that best captures the indication for each time a medication is taken. (One per incidence.)   82
Select an ATC that best captures the "typical" indication for a medication in this trial.  (One per medication per trial.)   5
Select an ATC that best captures the "typical" indication for a medication in all trials conducted by the company.  (One per medication per company.)   5
11
ATC if it is not available

• ATC selection is sometimes difficult or
  impossible for example if none of the available
  ATC classes fit the indication or if no
  indication is given. Does your organization have
  a principle for how to handle this?
• Select an ATC that captures the "typical"
  indication
• query the site for more specifics

12
ConMed lists
When a list of concomitant medication is sorted by ATC class how is it done? When a list of concomitant medication is sorted by ATC class how is it done? When a list of concomitant medication is sorted by ATC class how is it done?
A programmatic merge is done with all applicable ATC codes based on a collected Drug Code         0
The table is based on an ATC class that is selected by a coder based on the indication -         67
We dont sort lists by ATC         22
Dont know         11
13
Presentation and Analysis of ConMeds

• There are important distinctions between

DRUG CLASS
INDICATION
Purpose for taking this particular instance of
dosing
Similar Anatomic Target / Action Similar
Therapeutic PurposesSimilar Pharmacologic
Form Similar Chemical Structures
14
CM Coding to ATC
CRFCM
WHO Drug(version)
DrugCode
CMDECOD
CMCLAS
ATC 1?
ATC 2?
ATC 3?
Medications in a given Drug Class
15
CDISC SDTM ATC codes
By knowing the dictionary and version used, the
reviewer will be able to obtain intermediate
levels in a hierarchy (as in MedDRA), or a drugs
ATC codes (as in WHO Drug). The dictionary
version should be listed in the Comments column
of the Define data Definition document.

• Implementation Guide 3.1.1
• CMDECOD with the drugs generic name, and CMCLAS
  with the drug class only if the dictionary used
  codes drugs to a single class. When using
  WHODRUG, for example, CMCLAS would not be filled
  since a drug may have multiple classes.
• Use only when the dictionary used codes to a
  single class. If using a dictionary that allows
  links to multiple classes, then omit CMCLAS from
  the dataset. For example, sponsors who use WHO
  Drug, which allows links from a medication to
  multiple ATC codes, would not include CMCLAS.
• Implementation Guide 3.1.2
• CMDECOD with the drug's generic name, and CMCLAS
  with the drug class used for the sponsors
  analysis and summary tables. If coding to
  multiple classes, follow assumption 4.1.2.8.1 or
  omit CMCLAS.

16
CM Coding to ATC
CRFCM
WHO Drug(version)
DrugCode
CMDECOD
CMCLAS
CRFMH
Related Record
MeaningfulGrouping
MedDRA(version)
LLT
CRFAE
PT
Related Record
SOC
Medications taken for
For more information, please refer toOct 28,
2011 MedDRA User Group presentation
17

• An elephant is a mouse built to government
  specifications.- Lazarus Long

But what should we call it when we build
something to industry mythology about
government specifications??? Where do
regulations specify coding indication to ATC?
18
Impediments to change

• Some reluctance is due to
• Change (Weve always done it this way)
• Stakeholders, Pilot Studies, Timing/Planning
• Lack of tools for SAS programmers to import and
  relate relevant datasets within the dictionary
• Lack of tools for programmers to join the
  dictionary to the clinical datasets (CM, AE, MH)
• Lack of clarity on what analyses are needed by
  Regulatory Authority / Company

19
WHO Drug Dictionary (Version C)
20
WHO Drug Dictionary (Version B2)
DD
DDA
INA
21
SAS 1 - Configuration Program
SETUP DRIVE AND ROOT MAPPING
let drivec let root\data\WHOdd
let SasData\SAS\sd2

UPDATE SOURCE
let UMC\UMC\2009-03 let
DictFiles\9145_who_dd_mar_1_2009\B2
UPDATE VERSION AND FORMAT let
DictProductDD DD or DDE or
DDE-HERBAL let DictVersion2009-MAR
YYYY-MMM let DictFormatB2
B2 or C
INPUT
FOLDER filename UMC "drive.root.UMC.D
ictFiles." OUTPUT FOLDER
libname SASdata "drive.root.UMC.SasData."
DD
INA
DDA
include Your File Location Here\WHOdd_config.s
as"/nosource2
22
SAS 2a Import DD (drug names)
data SASdata.dd infile umc("dd.txt")
missover length WHOdrug 30. DrugName
45. input DrugID 1 - 6
Seq1 7 - 8 Seq2
9 - 11 ChkDgt 12
Desig 13 SourceYr 14
- 15 SourceCd 16 - 19
CompnyCd 20 - 24 NoIngred 25
- 26 SaltEstr 27 CodeYr
28 - 29 CodeQtr 30
UMCName 31 75
label DrugID "Drug Record Number"
Seq1 "Sequence 1" Seq2
"Sequence 2" ChkDgt
"Check digit" Desig
"Designation" SourceYr "Source
Year" SourceCd "Source Code"
CompnyCd "Company Code" NoIngred
"Number of ingredients" SaltEstr
"Salt/ester code" CodeYr "Year
Coded" CodeQtr "Quarter Coded"
DrugName "Drug name" UMCName
"UMC Full Drug Name" WHOdrug
"Drug Dict Version"
STRIP OFF UMC SLASHED MODIFIERS
DrugNameUMCName if ( substr(UMCName,36,1)"/"
substr(UMCName,45,1)"/" ) then
DrugNamesubstr(UMCName,1,35)
WHOdrug"DictProduct."" ""DictVersion.""
""DictFormat." run
DD
INA
DDA
23
SAS 2b Import DD (drug names)
Create Indexes proc datasets
librarySASdata modify dd
index create DrugCode(DrugID Seq1 Seq2)
DrugName Compressed quit
DD
INA
DDA
24
SAS 3 Import DDA (drug to ATC)
data dda infile umc("dda.txt")
missover input
DrugID 1 - 6 Seq1
7 - 8 Seq2 9 - 11
ChkDgt 12
ATCcode 13 - 19 CodeYr
20 - 21 CodeQtr 22
ATCflag 23
label DrugID "Drug Record
Number" Seq1 "Sequence
1" Seq2 "Sequence 2"
ChkDgt "Check Digit"
ATCcode "ATC Code"
CodeYr "Year Coded"
CodeQtr "Quarter Coded"
ATCflag "Official ATC Code"
run
DD
INA
DDA
25
SAS 4 Import INA
data atc length ATCcode 7
ATCLevel 1 ATCText 50 infile
umc("ina.txt") missover input _at_1
ATCcode 7. _at_8 ATCLevel 1. _at_9 ATCText 50.
run
Create Indexes proc datasets
libraryWork modify atc
index create ATCcode quit
DD
INA
DDA
26
SAS 5 Import / Join ATC Levels
proc sql
Create the ATC_levels table create table
atc_levels as select a.ATCcode as ATCcode,
b.ATCcode as ATC1code, b.ATCtext as ATC1text,
case when c.ATCcode b.ATCcode then "" else
c.ATCcode end as ATC2code, case when c.ATCcode
b.ATCcode then "" else c.ATCtext end as
ATC2text, case when d.ATCcode c.ATCcode then
"" else d.ATCcode end as ATC3code, case when
d.ATCcode c.ATCcode then "" else d.ATCtext end
as ATC3text, case when e.ATCcode d.ATCcode
then "" else e.ATCcode end as ATC4code, case
when e.ATCcode d.ATCcode then "" else e.ATCtext
end as ATC4text
from (select from atc) a left join
atc b on substr(a.ATCcode,1,1)
b.ATCcode left join atc c on
substr(a.ATCcode,1,3) c.ATCcode
left join atc d on
substr(a.ATCcode,1,4) d.ATCcode left join
atc e on substr(a.ATCcode,1)
e.ATCcode quit
Create index on atc_levels proc
datasets libwork modify
atc_levels index create ATCcode
quit
DD
INA
DDA
27
SAS 6 Import / Link Drug with ATC levels
proc sql create table SASdata.drug_atc
as select a.DrugID, a.ATCcode,
b.ATC1code, b.ATC1text,
b.ATC2code,
b.ATC2text,
b.ATC3code, b.ATC3text,

b.ATC4code, b.ATC4text from dda a
left join atc_levels b on
a.ATCcodeb.ATCcode where a.seq1lt001 and
a.seq2lt001 order by
a.drugid quit
Create index proc datasets
libSASdata modify drug_atc
index create DrugID ATCcode
quit
DD
INA
DDA
28
SAS 7 Analysis Many to Many Merge
libname whodd Your Drive Mapping Here
where WHOdd is" libname sasout Your
Drive Mapping Here where study data is"
START SAS CODE - MANY TO MANY MERGE (CM ATC)

proc sql
create table sasout.cm_ATC as select a.,
b. from sasout.cm_coded_DrugID a
LEFT JOIN whodd.drug_atc(rename(drugidwho_dr
ugid)) b on a.drugidb.who_drugid
quit run END SAS CODE - MANY TO MANY
MERGE

CM
DD
INA
DDA
29
The Result?

• Tables Grouping by True (all) Drug Class
• Tables Grouping by True (actual) Indication from
  MH/AE
• Analyses Avoids missed Drug Class if taken
  for another indication or off-label
• Less work for Coders
• Less work for SAS Programmers
• Less work for Medical Writers

30
How to Talk to Your Statistician Reaching a
Common Understanding
George B. Stoms Vital Systems,
Inc.George.Stoms_at_VTLsys.com [phone]
x222