jco.2017.75.6270

JOURNAL OF CLINICAL ONCOLOGY RAPID COMMUNICATION
Durable Complete Response After Discontinuation of
Pembrolizumab in Patients With Metastatic Melanoma
Caroline Robert, Antoni Ribas, Omid Hamid, Adil Daud, Jedd D. Wolchok, Anthony M. Joshua, Wen-Jen Hwu,
Jeffrey S. Weber, Tara C. Gangadhar, Richard W. Joseph, Roxana Dronca, Amita Patnaik, Hassane Zarour, Richard
Kefford, Peter Hersey, Jin Zhang, James Anderson, Scott J. Diede, Scot Ebbinghaus, and F. Stephen Hodi
ABSTRACT
Purpose
Pembrolizumab provides durable antitumor activity in metastatic melanoma, including complete
response (CR) in about 15% of patients. Data are limited on potential predictors of CR and patient
disposition after pembrolizumab discontinuation after CR. We describe baseline characteristics and
long-term follow-up in patients who experienced CR with pembrolizumab in the KEYNOTE-001
study (ClinicalTrials.gov identifier: NCT01295827).
Patients and Methods
Patients with ipilimumab-naive or -treated advanced/metastatic melanoma received one of three
dose regimens of pembrolizumab. Eligible patients who received pembrolizumab for $6 months
and at least two treatments beyond confirmed CR could discontinue therapy. Response was
assessed every 12 weeks by central Response Evaluation Criteria in Solid Tumors version 1.1. For
this analysis, CR was defined per investigator assessment, immune-related response criteria, and
potential predictors of CR were evaluated using univariate and multivariate analyses.
Results
Of 655 treated patients, 105 (16.0%) achieved CR after median follow-up of 43 months. At data
cutoff, 92 patients (87.6%) had CR, with median follow-up of 30 months from first CR. Fourteen
(13.3%) patients continued to receive treatment for a median of $40 months. Pembrolizumab was
discontinued by 91 patients (86.7%), including 67 (63.8%) who proceeded to observation without
additional anticancer therapy. The 24-month disease-free survival rate from time of CR was 90.9% in
all 105 patients with CR and 89.9% in the 67 patients who discontinued pembrolizumab after CR for
observation. Tumor size and programmed death-ligand 1 status were among the baseline factors
independently associated with CR by univariate analysis.
Conclusion
Patients with metastatic melanoma can have durable complete remission after discontinuation of
pembrolizumab, and the low incidence of relapse after median follow-up of approximately 2 years
from discontinuation provides hope for a cure for some patients. The mechanisms underlying
durable CR require further investigation.
J Clin Oncol 36:1668-1674. © 2017 by American Society of Clinical Oncology
INTRODUCTION
Prognosis in patients with metastatic melanoma has
historically been poor, with a median overall survival
(OS) of ,1 year and a 10-year survival rate of ap-
proximately 10%.
1,2
Melanoma treatment evolved
with the introduction of targeted therapy for BRAF
mutant disease and immune checkpoint inhibitors
for patients with advanced disease regardless of
BRAF mutation status. Ipilimumab, a monoclonal
antibody against cytotoxic T-lymphocyte antigen 4
(CTLA-4), was the first checkpoint inhibitor to
demonstrate activity in advanced melanoma, with an
objectiveresponserateof6%to11%andmedianOS
of 9 months to 11 months.
3-5
The anti–programmed
death 1 (PD-1) monoclonal antibodies pem-
brolizumab and nivolumab, which block binding of
PD-1 to its ligands PD-L1 and PD-L2, provided
response rates of 30% to 40% and median OS of
.2 years in patients with metastatic melanoma.
6-9
Pembrolizumab has demonstrated robust antitu-
mor activity and a favorable safety profile in mul-
tiple tumor types, and is currently approved in .60
countries for one or more advanced malignancies,
including unresectable or metastatic melanoma.
Author affiliations and support information
(if applicable) appear at the end of this
article.
Published at jco.org on December 28,
2017.
Processed as a Rapid Communication
manuscript.
Clinical trial registration: NCT01295827.
Corresponding author: Caroline Robert,
MD, PhD, Institut Gustave Roussy,114
rue Edouard Vaillant, 94805, Villejuif Paris
Sud, France; e-mail: caroline.robert@
gustaveroussy.fr.
© 2017 by American Society of Clinical
Oncology
0732-183X/18/3617w-1668w/$20.00
ASSOCIATED CONTENT
See accompanying Oncology
Grand Rounds on page 1649
Data Supplement
DOI: https://doi.org/10.1200/JCO.
2017.75.6270
DOI: https://doi.org/10.1200/JCO.2017.
75.6270
1668 © 2017 by American Society of Clinical Oncology
VOLUME 36 •NUMBER 17 •JUNE 10, 2018
Downloaded from ascopubs.org by 90.44.238.100 on March 3, 2021 from 090.044.238.100
Copyright © 2021 American Society of Clinical Oncology. All rights reserved.

In the multicohort, phase 1b KEYNOTE-001 study
(ClinicalTrials.gov identifier: NCT01295827), pembrolizumab mon-
otherapy provided an objective response rate of 33% per Response
Evaluation Criteria In Solid Tumors version 1.1 (RECIST v1.1) by
independent central review in patients with advanced melanoma.
6
The
two main goals of the current analysis were to (1) characterize
complete response (CR) in terms of time to response, duration of
response, and outcomes for patients from the time of CR and from
the time of discontinuation of treatment after CR; and (2) identify
potential predictors of these optimal treatment outcomes with use
of pembrolizumab in the KEYNOTE-001 study.
METHODS
Study Design and Patient Population
KEYNOTE-001 was an open-label, phase 1b clinical trial that in-
cluded multiple cohorts of patients with advanced solid tumors, including
melanoma and non–small-cell lung cancer. Detailed eligibility criteria for
the patients with ipilimumab-naive or ipilimumab-treated advanced mel-
anoma were published previously by Ribas et al.
6
Briefly, eligible patients
were $18 years of age with confirmed metastatic melanoma, measurable
disease per immune-related response criteria (irRC),
10
Eastern Cooperative
Oncology Group performance status (ECOG PS) 0 or 1, no history of
chemotherapy within 4 weeks of first pembrolizumab dose, and no history
of treatment targeting the PD-1/PD-L1 pathway. Patients were enrolled in
nonrandomized and randomized cohorts.
Treatment and Assessments
Patients received pembrolizumab 2 mg/kg every 3 weeks, 10 mg/kg
every 3 weeks, or 10 mg/kg every 2 weeks until disease progression, in-
tolerable toxicity, or patient or investigator decision to withdraw. After
a protocol amendment, patients who experienced CR, confirmed by two
imaging scans $4 weeks apart, and who received pembrolizumab treatment
of $6 months could, at the discretion of the investigator and if the patient
desired, discontinue treatment after receiving at least two pembrolizumab
doses beyond the initial determination of CR. Patients were not considered
to have achieved CR if they had undergone surgery or radiation for residual
disease. Tumor response was assessed every 12 weeks by independent central
review using RECIST v1.1 (for determining efficacy) and by investigator
review using irRC (for patient management).
For the purpose of this analysis, CR was defined using investigator
response per irRC. Disease-free survival (DFS) was defined as the time
from declaration of CR to documented disease progression or death from
any cause; for patients who did not experience an event, CR duration was
censored at the time of their last disease evaluation. Duration of CR from
time of discontinuation of pembrolizumab was calculated similarly.
Statistical Analysis
All analyses were performed in the treated population, which was
defined as all patients who received at least one dose of pembrolizumab. CR
rates and associated 95% confidence intervals (CIs) were estimated using
the Clopper-Pearson method. DFS was estimated using the Kaplan-Meier
method. SAS software, version 9.4 (SAS Institute, Cary, NC), was used for
all analyses. Several potential prognostic baseline factors (Supplemental
Data) were studied in relation to response. PD-L1 expression was assessed
using a clinical trial immunohistochemistry assay (PD-L1 IHC 22C3
pharmDx; Agilent Technologies, Santa Clara, CA) and the 22C3 antibody
(Merck, Kenilworth, NJ), as previously described by Daud et al
11
; PD-L1
positivity was defined as $1% staining in tumor cells and mononuclear
inflammatory cells. Factors identified using a univariate analysis were
further examined in a multivariate analysis using a step-wise regression for
variable selection. Analyses were performed using a data cutoff date of
September 1, 2016.
RESULTS
Patients
Of the 655 patients with advanced melanoma who received at
least one dose of pembrolizumab, 104 (15.9%) were still receiving
treatment as of the data cutoff date; median follow-up was 43 months
(range, 36 to 57 months). The most common reasons for treatment
discontinuation were progressive disease (41.7%) and adverse events
(AEs; 25.0%). Median duration of pembrolizumab exposure was
6 months (range, 1 day to 55 months).
Clinical Status
Median OS in all 655 treated patients was 23.8 months
(95% CI, 20.2-30.4), with 3-year and 4-year survival estimates of
42% and 37%, respectively. In the 152 treatment-naive patients, the
3-year and 4-year survival estimates were 51% and 48%, respectively.
In the 561 treated patients with an evaluable first clinical response
assessment, CR was the first radiologic response assessment in only
nine patients (1.6%) per irRC, as evaluated by investigator review. In
the 655 overall treated patients, across all assessments, 105 patients
had a best confirmed overall response of CR, as evaluated by in-
vestigator review, for a CR rate of 16.0% (95% CI, 13.3% to 19.1%;
Fig 1). Median time to first objective response was 3 months (range,
0.5 to 11 months); median time to CR was 12 months (range, 3 to
36 months). Notably, 84 of 105 patients (80.0%) with CR as their
best confirmed overall response had partial response (PR) or CR at
their first radiologic response assessment.
Of the 105 patients with confirmed CR as best overall re-
sponse, 92 (87.6%) remained in CR after a median follow-up of
30 months from first declaration of CR; median time receiving
treatment was 24 months (range, 1 to 53 months). At the time of
data cutoff, 14 patients (13.3%) were still electively receiving
pembrolizumab, with a median time on treatment of $40 months
(range, $36 to $53 months; Fig 1). Pembrolizumab was dis-
continued by the other 91 patients (86.7%) who experienced CR,
which included 67 patients (63.8%) who elected to stop treatment
after CR and proceeded to observation without further anticancer
therapy (Fig 1). Among these 67 patients, median time to overall
response was 3 months (range, 0.5 to 11 months), median time to
CR was 13 months (range, 3 to 36 months), and median time
receiving pembrolizumab was 23 months (range, 8 to 44 months;
Fig 2). Median time receiving treatment after achieving CR (CR to
last pembrolizumab dose) was 7 months (range, 0.5 to 41 months).
Of the 67 patients who proceeded to observation without ad-
ditional anticancer therapy, two (3.0%) died (of cardiac failure and
aspiration, considered unrelated to study treatment or progressive
disease), and four (6.0%) had progressive disease 5.6, 8.5, 22.8, and
37.3 months, respectively, after discontinuing pembrolizumab. Three
of these four patients subsequently started a second course of
pembrolizumab treatment. Of these patients, one remained on
pembrolizumab for 4 months before a second occurrence of
progressive disease and that patient is still alive; one is still
receiving pembrolizumab therapy and has been for approximately
jco.org © 2017 by American Society of Clinical Oncology 1669
Complete Responses in Melanoma After Pembrolizumab
Downloaded from ascopubs.org by 90.44.238.100 on March 3, 2021 from 090.044.238.100
Copyright © 2021 American Society of Clinical Oncology. All rights reserved.

15 months; and one was receiving pembrolizumab therapy for
approximately 9 months and subsequently died as a result of ma-
lignant neoplasm progression.
Most patients (n = 61; 91.0%) maintained CR after a median
time off pembrolizumab of 22 months (range, 4 to 46 months). For
the remaining 24 patients (22.9%) who experienced CR but were
0 6 12 18 24 30 36 42 48 54 60
PD
CR
PR
Time to PD or last assessment
Last dose
Time to death
Time Since the Start of Therapy (months)
Fig 2. Time to response and durability of
response from the start of therapy in complete
responders who discontinued pembrolizumab
and proceeded to observation (n = 67). Bar
length is equivalent to the time to the last
imaging assessment by investigator review.
CR, complete response; PD, progressive dis-
ease; PR, partial response.
Had a best confirmed
overall response of CR
per irRC by investigator review*
(n = 105; 16%)
Eligible for a second
course of pembrolizumab, proceeded
to observation without further
anticancer therapy
(n = 67)
Not eligible for a second
course of pembrolizumab,
discontinued pembrolizumab
(n = 24)
Treated with pembrolizumab
(N = 655)
Remained
in CR
(n = 6)
Died
(n = 1)
Remained
in CR
(n = 9)
Died
(n = 3)
Physician
decision
(n = 7)
Withdrew
consent
(n = 3)
PD
(n = 2)
AEs
(n = 12)
Remained
in CR
(n = 61)
Had PD
(n = 4)
Died
(n = 2)
Remained on
pembrolizumab
(n = 14; 13%)
Discontinued
pembrolizumab
(n = 91; 87%)
Fig 1. Patient disposition. (*) 92 patients (87.6%) remained in CR as of the data cutoff date. Of the 13 patients (12.4%) not in CR, seven experienced PD and six died. AE,
adverse event; CR, complete response; irRC, immune-related response criteria; PD, progressive disease.
1670 © 2017 by American Society of Clinical Oncology JOURNAL OF CLINICAL ONCOLOGY
Robert et al
Downloaded from ascopubs.org by 90.44.238.100 on March 3, 2021 from 090.044.238.100
Copyright © 2021 American Society of Clinical Oncology. All rights reserved.

not eligible for a second course of pembrolizumab and who dis-
continued pembrolizumab because of AEs (n = 12), progressive
disease (n = 2), physician decision to withdraw (n = 7), or patient
withdrawal of consent (n = 3; Fig 1), median time receiving
treatment was 21 months (range, 1 to 44 months). Among the 12
patients who discontinued pembrolizumab because of an AE, three
(25.0%) deaths occurred, one each from respiratory failure, ven-
tricular fibrillation, and cardiorespiratory arrest; all were considered
unrelated to study treatment. The remaining nine (75.0%) patients
remained in CR at data cutoff. Of the seven patients who dis-
continued per investigator decision to withdraw because of CR, one
(14.3%) died (of lymphoma) and six (85.7%) remained in CR.
Disease-Free Survival
As of the data cutoff date, only seven of 105 patients with CR
experienced confirmed progressive disease, two while receiving initial
pembrolizumab treatment, four after stopping pembrolizumab and
proceeding to observation, and one still reported as receiving pem-
brolizumab treatment. As of the cutoff date, all seven patients with
progressive disease were alive. For all 105 patients who experienced CR,
the estimated 24-month DFS rate from declaration of CR was 90.9%
(95% CI, 82.5% to 95.4%; Fig 3A). Four of 22 patients who dis-
continued pembrolizumab treatment for reasons other than pro-
gressive disease (AEs [n = 3], protocol violation [n = 1]) and who did
not proceed to observation had died as of the data cutoff date. For the
67 patients who discontinued pembrolizumab after CR for observa-
tion, the estimated 24-month DFS rate from treatment discontinu-
ation was 89.9%. For all 89 patients who discontinued pembrolizumab
after CR for reasons other than progressive disease (regardless of
whether they proceeded to observation), the estimated 24-month DFS
rate from treatment discontinuation was 85.8% (Fig 3B).
Potential Predictors of Complete Response
Among the potential prognostic baseline factors explored and
per univariate analysis, age, ipilimumab status, lactate dehydro-
genase (LDH) level, ECOG PS, BRAF mutation status, prior BRAF
inhibitor therapy, prior lines of therapy, tumor PD-L1 status, site of
metastasis, tumor size, and albumin level were statistically sig-
nificantly associated with CR (Table 1).
Among the 655 patients treated, tumor PD-L1 status was
unknown in 146 patients because of missing samples or insufficient
tissue, and target lesion size per central review was unknown in 72
patients; tumor PD-L1 status and baseline tumor size both were
unknown in 22 patients. Among the 459 patients with assessment
data for baseline tumor PD-L1 status and tumor size, the highest
rates of CR (42.7%) occurred in patients with smaller (1 to 5 cm)
PD-L1–positive tumors at baseline (Table 2). Patients with larger
baseline tumors (5 to 90 cm) had a lower CR rate (,10%), except
for those with baseline tumors 5 to 10 cm and with positive PD-L1
expression who had a CR rate of 20.5%, which was similar to the
CR rate of those patients with smaller tumors (1 to 5 cm) and
negative PD-L1 expression (20.7%).
After adjusting for the association of baseline tumor PD-L1
status and tumor size with CR in a multivariate analysis, only age
(P= .020) was associated with experiencing CR. CR rates were
higher in treatment-naive patients and in those $65 years of age
(Supplemental Data).
Because LDH level was a highly significant predictor of CR rate
in univariate analysis, is a known prognostic factor in melanoma
associated with baseline tumor size, and is easier to measure than
baseline tumor size, CR rates were also examined in subsets of
patients defined by LDH level (less than one times the upper level of
normal [ULN]; more than one times the ULN to no more than twice
the ULN; and more than twice the ULN) and tumor PD-L1 status. A
total of 155 patients could not be classified because of unknown LDH
level or unknown tumor PD-L1 status. Only one CR was observed in
a patient whose baseline LDH level was more than twice the ULN. In
patients with baseline LDH level no more than twice the ULN, CR
rates were lower in those with LDH level more than one times the
ULN to no more than twice the ULN than in those with normal
baseline LDH levels (no more than one times the ULN; Supplemental
Data). In patients with normal baseline LDH levels, CR rates were
higher in those with PD-L1–positive tumors (26.0%) than in those
with PD-L1–negative tumors (16.9%). However, after adjusting for
baseline tumor PD-L1 status and tumor size, the LDH level was no
100
90
80
70
60
Disease-Free Survival (%)
Time Since Achieving CR (months)
50
40
30
20
10
06
97 90 70 63 38 18 9 3 0105
All patients
12 18 24 30 36 42 48 54
No. at risk:
A
100
90
80
70
60
Disease-Free Survival (%)
50
40
30
20
10
06
75 56 42 26 12 51089All patients
12 18 24 30 36 42 48
No. at risk:
Time Since Stopping Therapy (months)
B
Fig 3. Disease-free survival (A) from time of experiencing complete response (CR) in all patients who achieved CR (n = 105) and (B) from time of discontinuation of
pembrolizumab in patients who discontinued after CR for reasons other than progression (n = 89). The hash marks designate patients who were censored at that time point.
jco.org © 2017 by American Society of Clinical Oncology 1671
Complete Responses in Melanoma After Pembrolizumab
Downloaded from ascopubs.org by 90.44.238.100 on March 3, 2021 from 090.044.238.100
Copyright © 2021 American Society of Clinical Oncology. All rights reserved.

longer significantly associated with CR, largely because tumor size
and LDH level were highly related.
DISCUSSION
Although data on antitumor activity and safety of immune
checkpoint inhibitors in oncology are abundant, to our knowledge,
this report is the first in-depth analysis of CR in a large cohort of
patients with metastatic melanoma treated with a PD-1 inhibitor.
Similar results were recently published in a smaller retrospective
study in patients with melanoma.
12
Among patients with advanced
melanoma enrolled in KEYNOTE-001, pembrolizumab provided
a confirmed best overall response based on investigator assessment
per irRC of CR in 16.2% of patients. Risk for disease progression or
death in patients who experienced CR was low; the 24-month DFS
Table 1. Statistically Significant (P,.05) Univariate Comparisons of CR Rates by Baseline Patient and Disease Characteristics
Characteristic No. of Patients With CR/Total No. of Patients Per Group CR Rate (95% CI), % P
Age group, years .003
#64 50/397 12.6 (9.5 to 16.3)
$65 55/258 21.3 (16.5 to 26.8)
Ipilimumab status .012
Naive 62/313 19.8 (15.5 to 24.7)
Exposed 43/342 12.6 (9.3 to 16.6)
LDH level ,.001
.ULN 20/249 8.0 (5.0 to 12.1)
#ULN 85/394 21.6 (17.6 to 26.0)
M stage ,.001
M0 3/8 37.5 (8.5 to 75.5)
M1a 12/50 24.0 (13.1 to 38.2)
M1b 28/89 31.5 (22.0 to 42.2)
M1c 62/508 12.2 (9.5 to 15.4)
ECOG PS .002
085/444 19.1 (15.6 to 23.1)
1 20/210 9.5% (5.9 to 14.3)
BRAF mutation status .028
Wild type 87/493 17.6 (14.4 to 21.3)
Mutant 17/156 10.3 (6.0 to 16.1)
Prior BRAF inhibitor therapy .002
Yes 7/111 6.3 (2.6 to 12.6)
No 98/544 18.0 (14.9 to 21.5)
Lines of prior therapy .001
036/157 22.9 (16.6 to 30.3)
140/207 19.3 (14.2 to 25.4)
2 16/178 9.0 (5.2 to 14.2)
3+ 13/113 11.5 (6.3 to 18.9)
Tumor PD-L1 status .024
Negative 13/123 10.6 (5.7 to 17.4)
Positive 75/386 19.4 (15.6 to 23.7)
Unknown* 17/146 11.6 (6.9 to 18.0)
Site of metastasis ,.001
Lung only 29/84 34.5 (24.5 to 45.7)
Liver, with or without other sites 16/201 8.0 (4.6 to 12.6)
Other sites 46/298 15.4 (11.5 to 20.0)
No baseline metastasis per central review 14/72 19.4 (11.1 to 30.5)
Tumor burden†,.001
$1to,5 cm 57/161 35.4 (28.0 to 43.3)
$5to,10 cm 21/128 16.4 (10.5 to 24.0)
$10 to ,20 cm 13/155 8.4 (4.5 to 13.9)
$20 to #90 cm 0/139 0 (0 to 2.6)
Unknown* 14/72 19.4 (11.1 to 30.5)
Albumin level‡,
quartile (g/dL)
.012
1($2.3 to ,3.7) 14/136 10.3 (5.7 to 16.7)
2($3.7 to ,4.1) 23/190 12.1 (7.8 to 17.6)
3($4.1 to ,4.4) 34/162 21.0 (15.0 to 28.1)
4($4.4 to #5.6) 34/165 20.6 (14.7 to 27.6)
Unknown* 0/2 0 (0 to 84.2)
Abbreviations: CR, complete response; ECOG PS, Eastern Cooperative Oncology Group performance status; LDH, lactate dehydrogenase; PD-L1, programmed death
ligand 1; ULN, upper limit of normal.
*Patients with unknown baseline characteristics were excluded from Pvalue calculation.
†Defined as the sum of the longest dimension of all measurable target lesions assessed by Response Evaluation Criteria in Solid Tumor v1.1.
‡The difference in the number of patients per quartile is a result of many patients having an albumin value of exactly 3.7 g/dL.
1672 © 2017 by American Society of Clinical Oncology JOURNAL OF CLINICAL ONCOLOGY
Robert et al
Downloaded from ascopubs.org by 90.44.238.100 on March 3, 2021 from 090.044.238.100
Copyright © 2021 American Society of Clinical Oncology. All rights reserved.