PharmaSUG 2017 - Paper AD24

SAS Macro for Derivation of Best Overall Response per RECIST 1.1
Bob Zhong, Jiangfan Li, Hong Xie, Peter De Porre, Kenneth Maahs, Kyounghwa Bae
Johnson & Johnson, Spring House, PA

ABSTRACT
Response Evaluation Criteria in Solid Tumors version 1.1 (RECIST 1.1) has been quickly adopted since
its publication in 2009 as an efficacy assessment standard in oncology clinical trials in solid tumors.
However, the rules for determining best overall response (BOR) need to take into considerations of
practical situations and certain clarifications in details are needed to handle real-world data. We
addressed some ambiguities in RECIST 1.1 and provided detailed handlings required to develop a SAS
macro for an accurate derivation of BOR.
BOR is derived from time-point overall responses of CR, PR, SD, PD, NE, and Unknown. To derive BOR,
responses that do not affect derivation of BOR are taken out first. After this step of data manipulation,
only the first three remaining time-point overall responses are needed to derive BOR. Complicated cases
such as PR-SD-PR and unconfirmed CR or PR are calculated in more detail. Our macro can be applied to
clinical trials with/without requirement of CR/PR confirmation, and in both interim analyses (potential
future data available) and final analysis (no future data).
This macro will allow for both accurate and timely determination of best overall responses via computer
programming.

INTRODUCTION
RECIST (Response Evaluation Criteria in Solid Tumors) 1.1 ([1]) provides good concept of best overall
responses (BOR). However, the RECIST rules for determining BOR have ambiguities and inaccuracies.
For example, the RECIST Table 3 ([1] p.235).

Display 1. Best Overall Response when Confirmation of CR and PR Required

The table concludes that the best overall response for two consecutive CRs is CR. While this is correct
most of times, there are cases when this is incorrect. When confirmation is involved, the second CR is
required to be after certain period (defined in the protocol, usually 4 weeks) from the first CR to confirm
the first CR. If the two CRs are too close, then the second CR cannot confirm the first CR, and the
conclusion may not be CR. Another example is a CR followed by PR, the table gives SD, PD, or PR as
possible BOR. However, it is not clear which one should be assigned. In this paper, we clarified these

1
<paper title>, continued

ambiguities, extensively filled in details to bridge gaps in the original RECIST 1.1, set up rules for deriving
BOR, and developed solid SAS macro tools for quick and accurate derivation of BOR. User-guide,
sample programs, and the SAS macro tools (in SAS catalog for both Windows and Linux/Unix) will be
provided upon request.

CR/PR CONFIRMATION
CR CONFIRMATION
To confirm a CR (the first CR), we look for another CR (the confirming CR) at least 4 weeks (study-
dependent) after the first CR. It is required that there is no other overall response, except CR and NE,
between the first CR and the confirming CR.
 If such confirming CR exists, then the subject BOR is confirmed CR.
 If such confirming CR is guaranteed non-existing (e.g. the first CR followed by PD or off-study), then
the subject BOR can be NE or SD, if the requirement of SD is satisfied.
 If there is no confirming CR in the current data, but there is a chance for such confirming CR in the
future data. Then the subject’s BOR is an unconfirmed CR (uCR). This is only applicable to the
interim analysis.

PR CONFIRMATION
To confirm a PR (the first PR), we look for another CR or PR (the confirming CR or PR) at least 4 weeks
(study-dependent) after the first PR. It is required that there is no other overall response, except CR or
PR or NE, between the first PR and the confirming CR or PR.
 If such confirming CR or PR exists, then the subject’s BOR is PR (confirmed).
 If such confirming CR or PR is guaranteed non-existing (e.g. the first PR followed by PD or off-study),
then the subject BOR can be NE or SD, if the requirement of SD is satisfied .
 If there is no confirming CR or PR in the current data but there is a chance for such confirming CR or
PR in the future data. Then the subject’s BOR is an unconfirmed PR (uPR). This is only applicable to
interim analysis.

WHY ENHANCE RECIST 1.1

In some cases, RECIST 1.1 is open to interpretation as it does not give every detail needed to properly
assign BOR. For example, as one can see from Display 1:
 (CR - CR) In a study with CR confirmation of 4 weeks, for a patient with CR-CR-PD on Days 7, 21,
35, RECIST 1.1 concludes the BOR is CR. However, the BOR for this patient should be PD.
 (CR - PR) RECIST 1.1 offers multiple conclusions SD, PD or PR. However, in practice, we need a
one-valued conclusion.
 (PR - CR) In a study with PR confirmation of 4 weeks, for a patient with PR-CR-PD on Days 7, 21, 35,
RECIST 1.1 concludes the BOR is PR. However, the BOR for this patient should be PD.
 (PR - PR) RECIST 1.1 unconditionally concludes BOR is PR for two consecutive PRs. However, for a
patient with PR-PR-PD on Days 7, 21, 35, the BOR should be PD, not PR.
 (PR - SD) RECIST 1.1 unconditionally concludes BOR is SD. However, for a patient with PR-SD-PD
on Days 7, 21, 35, the BOR should be PD, not SD.

SAS MACROS FOR BEST OVERALL RESPONSE PER RECIST 1.1

We developed two SAS macro tools %BESTOVRS_FA (for final analysis) and %BESTOVRS_IA (for
interim analysis) with the same syntax.

2
<paper title>, continued

%bestovrs_fa( dsn = ,
insl = ,
inrs = ,
sdwin = ,
crwin = ,
opt = ,
cleanup = yes );
where
dsn
is the input/output dataset which should contain USUBJID (Unique Subject Identifier).
insl
is the subject level input dataset:
1. should include all patients in &dsn
2. one record per subject
3. sorted by USUBJID
4. contains the following variables:
USUBJID
RFSTDTC - Subject Reference Start Date/Time
For interim analysis, &insl should also contain
FINALFL - Final Status Flag (Y/N)
The value of FINALFL should be exact 'Y' or 'N', case sensitive. If FINALFL = 'Y', then the patient has
no more disease assessment expected in the future. In the final analysis, everyone has FINALFL =
'Y', and so the flag FINALFL is not needed in %BESTOVRS_FA. For interim analysis, caller needs to
derive FINALFL of which "Y" includes (but not limited to)
 lost-to-follow up
 withdrawal of consent
 ended the study
 died
 In some studies, a subsequent anti-cancer therapy also makes FINALFL = 'Y'.
FINALFL does not play the role if the patient has experienced a PD.
inrs
is the input overall response data, containing standard variables: USUBJID RSDTC RSSTRESC, and
sorted by USUBJID. &inrs should contain all overall responses only. &inrs should contain neither
target lesion assessment nor non-target lesion assessment. RSSTRESC should hold standard values
COMPLETE RESPONSE
PARTIAL RESPONSE
STABLE DISEASE
PROGRESSIVE DISEASE
NOT EVALUABLE
or their abbreviations CR/PR/SD/PD/NE. Non-standard values in RSSTRESC will be ignored.
sdwin
is the SD sustaining windows in days. sdwin = 0 or missing means no requirement on SD windows.
crwin
is the CR/PR confirmation windows in days. If missing or 0, then no CR/PR will be confirmed.
opt
is the output dataset option for convenience.
cleanup
is the yes/no option to clean up working data sets.

3
<paper title>, continued

COMPUTATIONAL PRINCIPLES
The computation is trivial if the study does not require CR/PR confirmation. So here we only address the
case when CR and PR need to be confirmed.

NOTATIONS
X1 - number of days from randomization/start of study treatment to 1st time point overall response
evaluation

X2 - number of days from randomization/start of study treatment to 2nd time point of overall response
evaluation

X3 - number of days from randomization/start of study treatment to 3rd time point of overall response
evaluation

XX - minimal number of days required for SD per protocol (i.e. &sdwin). This is the timespan from
RFSTDT to the day of SD.

YY - minimal number of weeks required for CR/PR confirmation (i.e. &crwin)

DATA PREPARATION
Each of the following steps does not affect one's best overall response. The purpose of data preparation
is to enable the best overall response to be determined by one's first 3 time-point assessments.
Step 1. Remove any NE if any other disease evaluation exists except NE.

Step 2. Keep only the last NE from a queue of continuous NEs.

Step 3. Remove any assessment after the first PD, (CR, PR), or (CR, SD).

Step 4. Keep only the last SD from a queue of continuous SDs.

Step 5. Keep only the first PR and the last PR from a queue of continuous PRs

Step 6. Keep only the first CR and the last CR from a queue of continuous CRs.

Step 7. Remove SD before PR or CR for a queue beginning with SD, followed by PR or CR.
That is, (SD, PR, ...) or (SD, CR, ...).

Step 8. Remove the second PR from data queue (PR, PR, CR)

Step 9. Remove the first PR from any two consecutive PRs from a queue of (PR, PR, SD) if the time
interval between the two PRs is less than YY.

Step 10. Remove the first CR from any two consecutive CRs from a queue of (PR, CR, CR) if the time
interval between the two CRs is less than YY.

Step 11. Remove (PR, SD) from a queue beginning with (PR, SD), followed by PR or CR.
That is (PR, SD, PR) or (PR, SD, CR).

Step 12. Remove any assessments except CR if two consecutive CRs exist with the time interval
between two CRs are ≥ YY.

DERIVATION OF BEST OVERALL RESPONSE WITH CR/PR CONFIRMATION

After the above data preparation, we can derive the best overall response using only overall responses at
the first three time-points.

4
<paper title>, continued

Overall Overall response Overall response at 3rd Best overall Response

response at 2nd time point time point (X3)
at 1st (X2)
time
point (X1)
CR Unknown Unknown Final: SD if X1 ≥ XX otherwise, NE
Interim: If no additional disease
assessments are expected* then
SD if X1 ≥ XX otherwise, NE
else uCR
CR with X2 - X1 ≥ YY Any CR
CR with X2 - X1 < YY Unknown Final: SD if X2 ≥ XX otherwise, NE
Interim: If no additional disease
assessments are expected* then
SD if X2 ≥ XX otherwise, NE
else uCR
PR Flag as a data issue with note:
cannot be PR after CR.
SD if X2 ≥ XX otherwise, PD
SD Flag as a data issue with note:
cannot be SD after CR.
SD if X2 ≥ XX otherwise, PD
PD SD if X2 ≥ XX otherwise, PD
PR Any Flag as a data issue with note:
cannot be PR after CR.
SD if X1 ≥ XX, otherwise PD
SD Any Flag as a data issue with note:
cannot be SD after CR.
SD if X1 ≥ XX, otherwise PD
PD Any SD if X1 ≥ XX, otherwise PD
PR CR CR if X3 - X2 ≥ YY, CR;
else it will be changed to leading (PR, CR)
sequence due to data Step 10
Unknown Unknown Final: SD if X1 ≥ XX otherwise, NE
Interim: If no additional disease
assessments are expected* then
SD if X1 ≥ XX otherwise, NE
else uPR
CR with X2 - X1 ≥ YY PR, SD, PD, NE, Unknown PR
Flag as a data issue if PR or SD is the
overall response at the 3rd time point.
CR with X2 - X1 < YY Unknown Final: SD if X2 ≥ XX otherwise, NE
Interim: If no additional disease
assessments are expected* then
SD if X2 ≥ XX otherwise, NE
else uCR
PR Flag as a data issue with note:
cannot be PR after CR.
SD if X2 ≥ XX, otherwise, PD
SD Flag as a data issue with note:

5
<paper title>, continued

Cannot be SD after CR.

SD if X2 ≥ XX otherwise, PD
PD SD if X2 ≥ XX otherwise, PD
PR with X2 - X1 ≥ YY Any PR
PR with X2 - X1 < YY Unknown Final: SD if X2 ≥ XX otherwise, NE
Interim: If no additional disease
assessments are expected* then
SD if X2 ≥ XX otherwise, NE
else uPR
PD SD if X2 ≥ XX, otherwise, PD
SD Unknown SD if X2 ≥ XX otherwise NE
PD SD if X2 ≥ XX otherwise PD
PD Any SD if X1 ≥ XX otherwise PD
SD Unknown Unknown SD if X1 ≥ XX otherwise NE
PD Any SD if X1 ≥ XX otherwise PD
PD Any Any PD
NE Unknown Unknown NE
* Subjects are considered “no additional disease assessments are expected” with the following cases but
not limited to
 lost-to-follow up
 withdrawal of consent
 ended the study
 died
Note that, after data preparation, the above listing exhausts all cases.

SPECIAL CASES
There are a few special cases that RECIST 1.1 offers multiple conclusions in the discussion. However, in
reality, given data, we need to derive a single best overall response. By working with different groups in
clinical trials, we set up the following derivation rules.

CR FOLLOWED BY PR (CR-PR)
In this case, the PR implicitly might be a PD. So if X1 ≥ XX, then the BOR is SD. Otherwise the BOR is
PD.

CR FOLLOWED BY SD (CR-SD)
Similar as in the previous case, the SD implicitly might be a PD. So if X1 ≥ XX, then the BOR is SD.
Otherwise the BOR is PD.

SD APPEARS BETWEEN TWO PRS (PR-SD-PR)

Different clinical studies may handle this situation differently. We pick up the most conservative way of
handling this case, i.e. removing the leading PR-SD (step 11 in the data preparation).
It is noted that this choice here is a conservative one but may not be the best one. The SAS macro will be
updated at the next version with different options so that users can choose the desired one.

SINGLE CR OR PR IN INTERIM ANALYSIS

If a subject (patient) is off the study, then the BOR of the subject follows the conclusion in the final
analysis. That is, if X1 ≥ XX then BOR = SD, otherwise BOR = NE.
If the subject is on the study, then BOR = uCR (unconfirmed CR), or uPR (unconfirmed PR) respectively.

6
<paper title>, continued

SAMPLE PROGRAMS
CREATE SAMPLE DATA
data dat(drop=i);
length USUBJID $1;
do i=1 to 6;
USUBJID = i;
output;
end;
lable USUBJID = Unique Subject Identifier;

%let DATES = 2019-06-16 2018-04-29 2017-07-07 2017-05-14 2016-10-27;

%let DATES = &DATES 2016-10-21 2016-11-01 2015-10-22 2015-02-10;
data insl(drop=i);
length USUBJID $1 RFSTDTC $10 FINALFL $1;
do i=1 to 9;
usubjid = i;
rfstdtc = scan("&DATES", i);
finalfl = ifc(i<'3', 'Y', 'N');
output;
end;
lable USUBJID = Unique Subject Identifier
RFSTDTC = Subject Reference Start Date/Time
FINALFL = Final Status Flag;

data inrs;
infile datalines;
input USUBJID $1. RSSTRESC $21. RSDTC $10.;
datalines;
1 COMPLETE RESPONSE 2019-07-02
1 COMPLETE RESPONSE 2019-09-01
2 PARTIAL RESPONSE 2018-05-10
2 STABLE DISEASE 2018-06-14
2 PARTIAL RESPONSE 2018-07-10
3 STABLE DISEASE 2017-08-07
3 STABLE DISEASE 2017-09-07
3 STABLE DISEASE 2017-10-07
3 PARTIAL RESPONSE 2017-11-07
3 PARTIAL RESPONSE 2017-12-07
4 STABLE DISEASE 2017-06-14
4 PROGRESSIVE DISEASE 2017-07-14
5 STABLE DISEASE 2016-11-27
5 UNKNOWN 2016-12-27
6 PARTIAL RESPONSE 2017-03-10
6 PARTIAL RESPONSE 2017-04-01
;
run;

BEST OVERALL RESPONSES IN FINAL ANALYSIS

libname _TOOLS_ "SAS-library to sasmacr.sas7bcat" access=readonly;
options mstored sasmstore=_TOOLS_;

%bestovrs_fa(dsn = dat,
insl = insl,
inrs = inrs,

7
<paper title>, continued

sdwin = 84, %* 12 weeks;

crwin = 28, %* 4 weeks;
opt = drop=note);
%* Now dat has two variables: USUBJID, BESTOVRS.;

BEST OVERALL RESPONSES IN INTERIM ANALYSIS

libname _TOOLS_ "SAS-library to sasmacr.sas7bcat" access=readonly;
options mstored sasmstore=_TOOLS_;

%bestovrs_ia(dsn = dat,
insl = insl,
inrs = inrs,
sdwin = 84, %* 12 weeks;
crwin = 28, %* 4 weeks;
opt = drop=note);
%* Now dat has two variables: USUBJID, BESTOVRS. In this example,
output from %bestovrs_ia and %bestovrs_fa are different.;

CONCLUSION
We developed this macro tool on Windows SAS 9.2 as well as Linux SAS 9.4. All cases are examined
thoroughly in order to make computation possible. We followed RECIST 1.1, filled in all gaps, covered all
cases, and produced a one-valued macro tool. The macros enable the programming team to compute
Best Overall Response quickly, easily, and reliably.

REFERENCES
[1]. Eisenhauera, E.A., Therasseb, P., Bogaertsc, J., Schwartzd, L.H., Sargente, D., Fordf, R., Danceyg,
J., Arbuckh, S., Gwytheri, S., Mooneyg, M., Rubinsteing, L., Shankarg, L., Doddg, L., Kaplanj, R.,
Lacombec, D., Verweijk, J. 2009. “New response evaluation criteria in solid tumours: Revised
RECIST guideline (version 1.1).” European Journal of Cancer, 45:228 – 247. Also available at
https://ctep.cancer.gov/protocolDevelopment/docs/recist_guideline.pdf

CONTACT INFORMATION
Your comments and questions are valued and encouraged. Contact the author at:
Name: Bob Zhong
Email: [email protected]

Name: Jiangfan Li (James)

Email: [email protected]

Name: Hong Xie

Email: [email protected]

Name: Peter De Porre

Email: [email protected]

Name: Kenneth Maahs

Email: [email protected]

Name: Kyounghwa Bae

Email: [email protected]

8
<paper title>, continued

Janssen Research & Development, LLC.

Johnson & Johnson
1400 McKean Road
PO Box 776
Spring House, PA 19477