Long-Term Survival in Patients Responding to Anti–PD-1/PD-L1 Therapy and Disease Outcome upon Treatment Discontinuation

Since 2010, antagonistic monoclonal antibodies, such as CTLA-4, PD-1, and PD-L1, which target immune checkpoints, have revolutionized the field of oncology. Their superiority over standards of care was first demonstrated in melanoma (1, 2). Anti–PD-1 and anti–PD-L1 antibodies [anti–PD-(L)1] showed a higher level of efficacy and an improved safety profile over anti–CTLA-4 and were then rapidly developed in many cancer types (3–7). Although response rates vary across the different tumor types, the durability of responses appears to be a common feature shared by responder patients across all cancer types. This feature, together with the observation that some patients with stable disease (SD) have a clear clinical benefit, explains why anti–PD-(L)1-treated patients have longer overall survival (OS) compared with conventional therapies.

Although the prognosis of responding versus nonresponding patients is expected to be very different, the released anti–PD-(L)1 clinical trials have not yet reported survival data for complete responders versus partial responders, and there are no available data in the literature on the long-term survival of patients who respond to these immunotherapies.

This study aimed to describe the progression-free survival (PFS) and OS of patients who respond to anti–PD-(L)1 therapies in early-phase clinical trials, with a focus on long responders and the duration of response after therapy discontinuation. Clinical and biological factors associated with long response duration and relapse after therapy discontinuation were also investigated.

A retrospective analysis of patients treated with anti–PD-(L)1 antibodies was performed in the cohort of patients who had been treated at the Gustave Roussy Early Drug Development Department. The inclusion criteria were patients who were at least 18 years old, with a performance status of 0 or 1, and treated in a phase I trial with anti–PD-1 (pembrolizumab or nivolumab) or anti–PD-L1 (durvalumab or atezolizumab) monotherapy (at least 1 infusion). The patients' disease had to be measurable per RECIST 1.1 criteria and/or IrRC (Cheson criteria for patients treated for lymphoma; ref. 8). Multiple myeloma and myelodysplastic syndromes were excluded from our analysis because of their own specific evaluation criteria. Patients with brain metastasis were generally eligible if they were treated and controlled by surgery or radiotherapy. Immunotherapy durations depended on the therapeutic protocol. For the MEDI4736 (durvalumab) and MPDL3280A (atezolizumab) protocols, therapy could be provided for a maximum of 12 months if no progression occurred. Patients could be retreated in case of secondary progression. The therapy duration was 24 months for MK3475 (pembrolizumab) in the absence of progression. Patients were treated until disease progression in the nivolumab trials. Clinical and biological data were collected through a dedicated patient file database. The study was conducted in accordance with the principles of the Helsinki Declaration. All patients included in this retrospective study were treated in clinical trials approved by the French health agency ANSM and an ethical committee. Patients signed an informed consent prior to their inclusion in the trial. The redaction of the manuscript followed the STROBE guidelines for observational studies. Tumor responses were assessed by RECIST 1.1 or IrRC criteria (Cheson criteria for patients treated for lymphoma) according to the requirements of the respective protocols and the tumor types. Radiologic response assessments were centrally reviewed at Gustave Roussy Center by two trained radiologists.

The objective response rate (ORR; investigator assessed) was defined as the number of patients with either a complete response (CR) or a partial response (PR) divided by the number of patients who were treated in the study. The date of first response and the date of best response were recorded for every patient to define CR, PR, and the percentage of the best response. For patients with stable or progressive disease (PD) as the best response, the date retained was the date of the first evaluation. For all patients, the date of progression was recorded.

For the overall population, PFS was defined as the time from the first infusion of anti–PD-(L)1 (cycle 1, day 1) to the first documented tumor progression, death from any cause or last tumor evaluation. OS was defined as the time from the first infusion of anti–PD-(L)1 (cycle 1, day 1) to death or last news. For the responder patient subgroup analysis, PFS and OS were calculated for patients who responded and were alive at 3 months. For the responder patients, PFS and OS were also calculated from the date of the first objective response to progression/death or last news (rPFS and rOS). For complete responders, PFS and OS were also calculated from the first CR assessed (crPFS and crOS). Time from therapy initiation to the objective response was assessed. For secondary refractory disease, survival after relapse was also calculated from the date of observed relapse to death or last news. A long response was defined as patients with responses that lasted at least 2 years. The date of last news was collected for patients who were still alive. The PFS and OS analyses were performed for patients who had responded to immunotherapy and for the entire treated cohort using the Kaplan–Meier method, and then, PFS and OS analyses by tumor type were performed with adjusted Cox models for all variables to obtain corresponding hazard ratios.

Log-rank analyses and an adjusted Cox model were generated to differentiate responses as follows:

- Clinical parameters, such as the tumor type, number of treatment lines received before anti–PD-1 therapy, Royal Marsden Hospital (RMH; ref. 9) score (correlated with a metastatic site number greater than or equal to 3, an increased LDH level, and a decreased albumin level), metastatic site number, sex, age, type of immunotherapy (anti–PD-1 vs. anti–PD-L1), stage of disease, best response, and first response.

- Biological parameters, such as the LDH level and neutrophil-to-lymphocyte ratio (NLR). Unfortunately, PD-L1 tumor expression data were not available retrospectively.

The distributions of factors between the two groups, “responders” (Rs) versus “nonresponders” (NRs), were compared using Student and Fisher exact tests to identify associations between the response and these clinical and biological parameters. Hereafter, the analysis was focused on the group of responder patients, and the distribution of factors between long responder (LR) and short responder patients was also compared.

The distributions of factors between the two groups defined by a prolonged response after immunotherapy discontinuation and relapse were compared using the nonparametric Mann–Whitney test.

The response after therapy discontinuation was examined for patients for whom therapy was stopped without progression.

Two distinct landmark analyses have been used. First, the landmark survival approach has been used to identify the therapy duration, which was associated with a risk of relapse lower than 5% after therapy discontinuation. In this landmark analysis, quartile (q)1, q2, and q3 therapy durations were chosen as different landmark baselines. Additionally, a landmark analysis has been used to estimate an association between response and OS from a specific landmark starting point. These landmark starting points were the time points of the first disease assessment and of the best response assessment (10). Because we could not choose a fixed time point for all the patients (retrospective study with data from different prospective clinical trials), adaptive time points have been chosen to perform these landmark analyses. One time point was the first disease assessment. This time point is approximately the same for every phase I protocol (median of 2 months after therapy initiation). Because late responders could be missed with such early time point, we have performed the same analyses with the time point “best response” (PR or CR) which is more appropriate but varies for each patient.

A P value of ≤0.05 was retained as the cutoff for statistical significance.

From December 2011 to January 2017, 306 patients were enrolled at Gustave Roussy Cancer Center in a phase I trial testing an anti–PD-(L)1 therapy. Of these patients, 44 were subsequently excluded: 23 did not receive treatment, 4 had no available data, 16 had multiple myeloma, and 1 had myelodysplastic syndrome. In the final analysis, 262 patients were included. Among these patients, 156 were treated with anti–PD-1 antibody (nivolumab, n = 27; pembrolizumab, n = 129), and 106 were treated with anti–PD-L1 antibody (atezolizumab, n = 47; durvalumab, n = 59; Fig. 1). At the time of treatment, the patients presented with stage II disease in 1% of patients (n = 2), stage III in 5% (n = 14), and stage IV in 93% (n = 243); the stage data were missing for 1% (n = 3) of the patients. The majority of patients (93%) had solid tumors of all types (n = 243) and a hematologic malignancy such as Hodgkin lymphoma (HL; n = 9) and non-Hodgkin lymphoma (NHL; n = 10). The patient characteristics are summarized in Table 1.

The median follow-up duration was 34 (range, 6–62) months for responder patients. Among the 262 patients who were evaluable, 76 (29%; 95% CI, 24–35) developed an objective response. Among these patients, 48 (18%; 95% CI, 14–24) presented an objective response at 3 months. The median PFS of these 48 responding patients was 30 months [interquartile range (IQR) = 16–not reached], but the median OS was not reached [IQR = 53–not reached; Fig. 2A). The 3-year PFS was 47% (95% CI, 31–61; 17 subjects at risk), whereas the 3-year OS was 84% (95% CI, 68–93; 23 subjects at risk). For the 76 RP, the median rPFS was 19 months (IQR = 9–not reached), and the rOS was also not reached (IQR = 31–not reached; Supplementary Fig. S1a). The median time from therapy initiation to response was 2 months (IQR = 2–3; range, 1–18).

Partial responses (PR) were observed in 55 patients (21%; 95% CI, 16–26). The PFS for PRs (rPFS) was 15 months (IQR = 6–30), and the median OS for PRs (rOS) was 49 months (IQR = 22–not reached; Supplementary Fig. S1b). The 3-year rPFS was 21% (95% CI, 10–35; 5 subjects at risk). The median rOS was 49 months (IQR = 22–not reached), the median 3-year rOS was 56% (95% CI, 38–71; at risk subjects: 14), and the median 5-year rOS was not reached.

A CR was observed in 21 (8%; 95% CI, 5–12) patients. No deaths occurred in this subgroup during a median follow-up of 45 months (range, 13–62). The median crPFS was not reached (IQR = 17–not reached) after a median time from therapy initiation to CR of 11.5 months (IQR = 7–20.5; range, 2–35; Supplementary Fig. S1c). Only 3 patients presented a CR at 3 months.

In 39 (15%) cases, immunotherapy was discontinued. These patients did not present disease progression. In 9 of the 39 discontinuations, the patients stopped therapy because of an adverse event (n = 9, 3 CR and 3 PR and 3 SD). For the other 30 patients, the treatment was stopped per protocol or because of a prolonged response, and the response at discontinuation was PR (n = 14), CR (n = 14), and SD (n = 2). Relapse was observed in 15 cases [38%; urothelial carcinomas (3), breast cancer (1), non–small cell lung cancer (NSCLC) (1), choroid melanoma (1), melanoma (3), colorectal cancer (1), Merkel cell carcinoma (1), HL (1), renal cell carcinoma (2) and ovarian cancer (1)], including 2 CR (12% of CR), 9 PR (53% of PR), and 4 SD (80% of SD) after a median PFS of 19 months (IQR = 5–23) and a median time off therapy of 4 months (IQR = 2–12.5). However, 24 (62%) patients [melanoma (14), MSI colorectal cancer (4), HL (2), NHL (1), NSCLC (1), urothelial cancer (1), and head and neck cancer (1)], of whom 14 presented a CR (88% of CR), were still responding. Thus, the median progression-free survival from therapy discontinuation was not reached [IQR (9-NR); Fig. 3A and B].

The median therapy duration was longer in the group that did not relapse (n = 24): median of 15.5 (IQR = 11–23) months versus 11 (IQR = 8.5–11.5) months (n = 15) in relapsing patients (P = 0.017).

Therapy discontinuation within the first 12 months of anti–PD-(L)1 was associated with a higher risk of relapse than after 12 months (P = 0.002; Supplementary Fig. S2).

After treatment reintroduction in 8 of 15 patients in the relapsing group, 4 presented SD before reprogression. Two presented a PR, one had a CR, and one had PD as the best response.

We could not find any association of quantitative variables, such as the number of metastases, LDH level, NLR, age, and number of prior lines of treatment, or of qualitative variables, such as sex, tumor type, RMH score, toxicity occurrence, and stage of the disease, with the incidence of relapse after immunotherapy discontinuation. However, the type of response seemed to have an impact on relapse (CR vs. PR vs. SD; P = 0.006), especially for the CR response versus PR (P = 0.02). In contrast, no significant difference was found for PR versus SD (P = 0.36). Because the statistical power to identify factors associated with relapse after discontinuation is low (small cohort), these negative results should be interpreted more as inconclusive rather than as negative.

Among the 76 responders, 31 (40%; 95% CI, 30–53) were long-term responders. Their clinical and biological features are listed in Supplementary Table S1. The tumor response assessment was based on IrRc in 77% of the patients and RECIST 1.1 in 23% of the patients. The most represented tumors were melanoma (58%), HL (13%), NSCLC (10%), and urothelial cancer (10%). We selected one year (2013) as the time of transition for melanoma and the other tumor types included in anti–PD-(L)1 protocols to decrease historical bias in the LR proportion analysis function of the tumor type due to the high proportion of melanoma patients selected during initial years of immunotherapy use. This could be more representative of the percentage of overall LRs independent of the tumor type. Among the patients who started treatment in 2013, a long-term response was observed in 23% (95% CI, 10–42) of melanoma (n = 30), 22% (95% CI, 3–60) of urothelial carcinoma (n = 9), and 14% (95% CI, 0–58) of NSCLC (n = 7) patients treated in 2013.

Overall, 76 (29%; 95% CI, 24–35) patients were objective responders (21 CRs and 55 PRs), and 186 (71%; 95% CI, 65–76) patients were nonresponders. The ORR was 38% (95% CI, 30–48) for IrRC evaluation versus 21% (95% CI, 14–29) for RECIST 1.1. The tumor assessment was based on IrRC more often in responders (69%) than in nonresponders (47%, P = 0.0016) and in long-term responders (80%) than in non–long-term responders (62%, P = 0.045). The median number of metastatic sites was higher in nonresponders than in responders [median 2 (IQR = 1–3) vs. 2 (IQR = 2–3), P = 0.0335]. No association was found between the response and NLR, LDH level, age, number of metastases per organ, sex, RMH score, immune adverse event occurrence, and disease stage (all P > 0.05).

Progression after objective response was observed in 48 patients (64%; 95% CI, 51–74) of responders): 42 (76%; 95% CI, 63–87) progressed after a PR, whereas only 6 (29%; 95% CI, 11–52) progressed after a CR. Characteristic features are listed in Supplementary Table S2. The most frequent tumor types were melanoma (42%), HL (15%), urothelial carcinoma (6%), and NSCLC (6%). The percentage of relapse after an initial response was 53% (95% CI, 38–71) for nonmelanoma and non-HL cancer types, 60% (95% CI, 42–77) for melanoma and 88% (95% CI, 47–100) for HL. All relapses occurred before 5 years from therapy initiation. The median rPFS in secondary refractory patients was 12 months (IQR = 6–16), and the 3-year PFS in secondary refractory patients was 2% (95% CI, 1–11). The median survival from relapse was 26 months (IQR = 9–NR), the 3-year survival from relapse was 43% (95% CI, 23–62), and 5-year survival was not reached (Supplementary Fig. S3). The median PFS after reprogression was 15 months in melanoma, 17 months in NSCLC, 10 months in lymphoma, and 16 months in urothelial carcinoma and did not statistically differ between these different tumor types (P = 0.84) (Supplementary Fig. S4). Upon progression, 9 patients who had previously stopped the treatment were allowed to be retreated per protocol with the same anti–PD-(L)1, 5 were treated with another immunotherapy, 7 were treated with a targeted therapy, and 17 received chemotherapy. The study was continued for 5 patients who progressed under treatment (so-called “treatment beyond progression”), and PD was controlled with local therapy (surgery or radiotherapy). Four patients received no further therapy and died. No data were available for one patient.

At the time of data collection, 52 patients (20%; 95% CI, 15–25) of the overall population were still alive 2 years after immunotherapy initiation. Twenty-two patients were still in response (38.5%; 95% CI, 29–57), while 30 (61.5%; 95% CI, 43–71) discontinued the treatment because of PD. Of the 30 patients who presented PD, 63% were evaluated by RECIST 1.1 and 37% by IrRC. Their characteristics are listed in Supplementary Table S3.

The ORR was 29% (95% CI, 24–35) across all tumor types: 89% (95% CI, 52–100) for HL (n = 9), 55% (95% CI, 21–86) for NHL (n = 9), 50% (95% CI, 16–84) for MSI-high colorectal cancer (n = 8), 43% (95% CI, 32–55) for melanoma (n = 76), 33% (95% CI, 0–91) for Merkel cell carcinoma (n = 3), 33% (95% CI, 0–91) for MSS colorectal cancer (n = 3), 33% (95% CI, 0–91) for endometrial adenocarcinoma (n = 3), 33% (95% CI, 0–91) for SCLC (n = 3), 28% (95% CI, 12–49) for urothelial carcinoma (n = 25), 25% (95% CI, 0–81) for gastric adenocarcinoma (n = 4), 20% (95% CI, 7–41) for NSCLC (n = 25), 20% 20 (95% CI, 3–56) for cervical epidermoid carcinoma (n = 10), 20% (95% CI, 0–72) for glioblastoma (n = 5), 14% (95% CI, 0–58) for prostatic adenocarcinoma (n = 7), 12.5% (95% CI, 0–53) for ovarian adenocarcinoma (n = 8), 7.7% (95% CI, 0–36) for renal cell cancer (n = 13), and 9.5% (95% CI, 12–30) for head and neck cancer (n = 21; Supplementary Table S4). No objective response was observed for the other tumor types. CRs among responders were 38% (95% CI, 22–58) for melanoma (n = 31) and (95% CI, 9–76) HL (n = 8), 29% (95% CI, 4–71) for urothelial carcinoma (n = 7), 25% (95% CI, 1–81) for MSI-high colorectal cancer, and 20% (95% CI, 1–72) each for NHL (n = 5) and NSCLC (n = 5). Only one Merkel cell carcinoma reached an objective response that presented CR. No CR was achieved for the other responder tumor types.

The median PFS was 3 months (IQR = 1–14); the 3-year and 5-year PFS percentages were 12% (95% CI, 8–16) and 10% (95% CI, 6–15). The median OS was 13 months (IQR = 5–55); the 3-year and 5-year OS percentages were 29% (95% CI, 23–35) and 24% (95% CI, 17–31; Fig. 4).

At the first disease assessment (median of 2 months; IQR = 1–2), 122 of these patients presented a PD (47%; 95% CI, 40–53); 91 patients, a SD (35%; 95% CI, 29–41); 45 patients, a PR (17%; 95% CI, 13–22); and only 3 patients, a CR (1.1%; 95% CI, 0–3) (no data were available for one patient). We found an association between the response evaluated at the first disease assessment and OS by landmark analysis from this time point. The comparison of OS from PD versus non-PD patients at the first disease evaluation showed a significant impact of being in PD on OS (median OS: 5 vs. 37 months; HR PD/non-PD: 7.4; 95% CI, 5.40–10.14; P < 0.0001). Similar results were found upon stratification per the first type of response (5 months for PD vs. 18 months for SD vs. not reached for PR and CR; P < 0.0001; Fig. 2B and C). In addition, log-rank analysis highlighted the predictive value for OS of the best response type assessed from the landmark point “best response” (Fig. 2D). The median time from the start of treatment to best response was 1 month (IQR = 1–2) for PD, 2 months (IQR = 1–2) for SD, 9 months (IQR = 5–17.5) for PR and 11.5 months (7–22) for CR. Pseudoprogression occurred in 2.6% (95% CI, 0–9) of the responding cohort.

In 75 patients (29%; 95% CI, 23–35), immunotherapy was continued beyond disease progression. The median PFS was 2 months (IQR = 1–7.5), and the median therapy duration was 4 months (IQR = 2–11). The median OS was 13 months (IQR = 6–24.5). Thirty-seven patients (49%; 95% CI, 38–61) presented PD at the first disease evaluation; 19 (25%; 95% CI, 16–35), SD; 16 (36%; 95% CI, 13–32), RP; and 3 (95% CI, 1–11), RC. Forty-one patients (55%) were evaluated with IrRC, and 33 (45%) were evaluated with RECIST 1.1.

A lower RMH score was associated with a longer PFS: the median PFS (months) was 5 (IQR = 2–19) for RMH0, 2 (IQR = 1–16) for RMH1, 2 (IQR = 1–6) for RMH2, and 1 (IQR = 1–2) for RMH3. HR for progression associated with RMH3 versus RMH0 was 2.40 (95% CI, 1.04–5.51; P = 0.001; Fig. 5A; Supplementary Fig. S5a).

No associations between PFS and age, sex, prior line number, metastatic site number, metastatic site, stage of disease, NLR, immune-related adverse event (irAE) occurrence, and immunotherapy type (anti–PD-L1 or anti–PD-1) were found (Fig. 5A).

OS was found to be affected by the RMH score: [median OS: months (IQR)] RMH 0: 19 (IQR = 10–NR); RMH 1: 13 (IQR = 6–44); RMH 2: 8 (IQR = 3–23); RMH 3: 3 (IQR = 2–13); P < 0.0001 and adjusted HR: 1.54 (95% CI, 1.21–1.96); P = 0.0005; Fig. 5B; Supplementary Fig. S5b).

The overall median NLR was 4 and was chosen as a cutoff for performing the survival analysis. Longer OS was observed in NLR < 4: the median OS (months) was 19 (IQR = 9–NR) in NLR < 4 and 9 (IQR = 3–25) in NLR≥4 (HR: 0.62; 95% CI, 0.45–0.84; P < 0.0001; Supplementary Fig. S6a). After adjusting for confounding factors, compared with NLR < 4, the HR of death associated with NLR ≥ 4 was 1.04 (95% CI, 1.02–1.08; P = 0.002; Fig. 5B). A subgroup analysis revealed an increase in survival in the PD subgroup for NLR < 4, with a median survival increase to 8 months versus 4 months for the NLR ≥ 4 group (P = 0.017; HR, 0.64; 95% CI, 0.43–0.95; Supplementary Fig. S6b). Similarly, an impact of NLR was also found in patients with SD, with a median OS of 19 months versus 9 months for NLR < 4 versus NLR ≥ 4, respectively (P < 0.0001; HR: 0.35; 95% CI, 0.20–0.65; Supplementary Fig. S6c). However, there was no impact of NLR in the case of an objective response (P = 0.627; Supplementary Fig. S6d).

Lymphocyte and neutrophil counts were assessed in nonresponders and responding patients. No significant difference was found between the two groups (P = 0.74 and P = 0.62, respectively). However, the median neutrophil counts tended to be higher and the median lymphocyte counts tended to be lower in nonresponders versus responders (Supplementary Fig. S7a–S7c).

Furthermore, after adjusting for all variables, we found a significant association between a longer OS and a lower number or prior lines of treatments, which was not found with the log-rank analysis. The median OS was 15 months for patients with <2 prior lines of therapy (IQR = 5–53) and was of 12 months for patients with ≥2 prior lines of therapy (IQR, 5–43; P = 0.353). But the adjusted HR was 1.15 (95% CI, 1.03–1.29; P = 0.014).

No influence on OS was found for other variables, including anti–PD-1 and anti–PD-L1 therapy, after Cox analysis was adjusted for all variables (including tumor type; Fig. 5B).

An incidence of 29% of all grades of CTCAE irAEs was found. Within these adverse events, thyroiditis represented 21% (grade 1: 20%; grade 2: 1%); colitis, 14% (grade 1: 6%, grade 2: 4%, grade 3: 4%); and other irAEs, such as grade 1 purpura, grade 1 and 3 psoriasis, fatigue, grade 1 dry syndrome, grade 3 lichen planus, grade 4 autoimmune neutropenia and grade 1 lipasemia increase, 14%; followed by pneumonitis (6%, grade 1: 4%; grades 2 and 3: 1%), grade 1 vitiligo (10%), arthritis (9%, grade 1: 5%, grade 2: 2%, grade 3: 1%), grade 1 pruritus (8%), nephritis (4%, grades 2, 3, and 4: 1%), grade 3 Guillain Barre (3%), grade 3 diabetes (3%), hepatitis (3%, grade 1: 1.5%; grade 3: 1.5%), and grade 1 autoimmune hemolytic anemia (AIHA; 1%). Notably, there was no reported case of hypophysitis. The median time from the start of treatment to the occurrence of irAEs was 3 months (IQR = 1–7.25; vitiligo: 3 months, rash: 9 months, pruritus: 3 months, autoimmune diabetes: 21 months, pneumonitis: 4 months, hepatitis: 0.5 months, thyroiditis: 4 months, arthritis: 7 months, Guillain–Barre: 0.75 months, AHAI: 10 months and other: 3 months). Management of these adverse events was based on protocol recommendations with the following steps: therapy was continued for grade 1 toxicity, infusion was maintained in the case of grade 2 toxicity, therapy was definitively stopped in the case of grade 3 and 4 adverse events, and steroids were initiated as a function of the toxicity.

This study presents a 5-year follow-up of patients with solid tumors who were treated with anti–PD-(L)1 monotherapy in phase I clinical trials. Monotherapies with antagonistic monoclonal antibodies that lock PD-1 [nivolumab (11), pembrolizumab (12)] or its ligand PD-L1 [atezolizumab (13) and durvalumab (3)] can mediate rapid and durable responses across multiple cancer types, including in the population of phase I patients with advanced malignancies. This study provides new insights into the long-term outcomes of patients treated with anti–PD-(L)1, highlighting the impact of clinical and biological features at baseline at the first and best response, and the potential impact of therapy duration to achieve a prolonged response after therapy discontinuation. The ORR of the whole cohort of our patients does not reflect the high variability of response rates seen across tumor histologies. Ongoing disease-specific prospective clinical trials shall illustrate the impact on survival of the differences in ORRs. However, a common feature seen across tumor histologies is the long duration of tumor responses. Durable responses were found even after treatment discontinuation due to irAEs or per clinical study arrest since relapse occurred in only 38% of this cohort. This durability of the response is currently explained by the theoretical rationale that anti–PD-(L)1 therapy can generate a polyclonal and memory adaptive antitumor immunity that is able to control the heterogeneity of the disease and to reset the tumor–host immune interaction toward cancer rejection (14, 15). Additionally, CR before therapy discontinuation is shown to be a positive factor for a prolonged response upon discontinuation. Interestingly, patients in CR appeared to maintain their response after therapy discontinuation, whereas relapse occurred in 87% of patients with SD or PR.

However, we found that an anti–PD-(L)1 therapy duration of less than 12 months in responding patients (or SD) was associated with a higher rate of relapse and that these relapses occurred early after immunotherapy discontinuation. This finding could provide guidance regarding the current question of the optimal duration of anti–PD-(L)1 treatments. In the case of relapse, responses upon immunotherapy reintroduction were observed in only 25% of cases, suggesting that an adaptive mechanism was responsible for immune escape (16, 17). Such adaptive escape mechanisms seemed to also be encountered in patients who developed secondary refractory disease upon anti–PD-(L)1 without therapy discontinuation. Several acquired resistance mechanisms have been recently proposed: loss-of-mutation–associated antigens during immunotherapy (12, 14), JAK 1/2 loss-of-function mutations, truncating mutations in the gene encoding β-2-microglobulin that impair MHC class I surface expression (19–21), or T-cell exhaustion and the upregulation of alternate coinhibitory immune checkpoints (22–24). These escape mechanisms must be better understood to either circumvent (e.g., CAR-T cells or bispecific T-cell engagement of antibodies against MHC-I-negative cancers) or prevent them (e.g., upfront combinations that might prevent the selection and emergence of resistant subclones).

MSI-high colorectal cancer was associated with a higher response rate than other MSI tumors, although the difference might only be due to the low number of patients analyzed. One MSS colorectal cancer patient developed an objective response, which might be explained either by an MSI-like phenotype or by false negativity of IHC staining (25–27). Moreover, we confirmed the impact of the RMH score, including the negative impact of the number of metastatic sites (≥3) on the response and on survival (OS) compared with the response duration (PFS) in the context of immunotherapy, as described in prior studies and with tumor-targeted therapies (28, 29).

In agreement with Templeton and colleagues (30), we found a median NLR at 4. By using this value as a cutoff, we found a significant association between NLR > 4 and a shorter OS, which was marked in the PD and SD subgroups but was not significant for patients with PR or CR. This association was not found for PFS. It was an arbitrary cutoff chosen for NLR stratification that is clinically relevant. As we have previously shown (31), NLR might be a simple and meaningful routine marker for evaluating the systemic imbalance toward myeloid immune suppression. Neutrophils are important innate immune cells that can play key roles in tumor development by favoring chronic inflammation, which is a driving force for tumor initiation and progression (N2 profile; refs. 32, 33). Indeed, chronic inflammation induces N2 neutrophils, which promote angiogenesis and tissue remodeling while suppressing the cytolytic activity of immune cells, such as lymphocytes, activated T cells, and natural killer cells, and therefore constitute innate resistance components to anti–PD-(L)1 antibody (32, 34).

The long-term follow-up in this retrospective study offers the opportunity to assess anti–PD-(L)1 OS and PFS rates at 3 and 5 years. First, this analysis confirmed the survival benefit provided by immunotherapy, with a median OS not reached in this group of patients and a landmark 5-year OS of 63%. However, the PFS at 5 years was much lower than the OS (32%), suggesting that early disease progression in some patients receiving immune-checkpoint blockade could evolve into durable disease stabilization or regression. PFS may underestimate the efficacy of immune-checkpoint–targeted therapy when it is evaluated using conventional response criteria (RECIST 1.1). This methodological bias is supported by the imbalance observed herein between patients with clinical benefits and the imaging criteria that were used for their disease assessment. Indeed, 80% of the responding patients were evaluated with IrRc versus 69% of progressors with RECIST 1.1. Hodi and colleagues reported a probable underestimation of anti–PD-1 therapy benefit by RECIST 1.1 criteria in approximately 15% of patients (35). This phenomenon could explain why 11% of patients presenting PD in our cohort were still alive 2 years after immunotherapy initiation, as a consequence of atypical responses such as pseudoprogression, mixed responses or the appearance of new lesions during the overall disease regression. Besides the impact of the cancer histologies treated, the higher response rate found with IrRC criteria raises the question of the impact of the radiologic method used for response evaluation on the actual disease status of a patient treated with anti–PD-(L)1. This question could not be answered by our retrospective study, but ongoing prospective clinical trials comparing RECIST 1.1 to irRC and iRECIST shall answer that question in the near future.

Beyond the atypical responses and imaging criteria, another aspect might explain the discrepancy between PFS and OS. One hypothesis is that immunotherapy might cause a sensitization to chemotherapy and that the next line of palliative chemotherapy becomes more efficacious than expected in a subset of patients. This hypothesis is supported by the rationale that some chemotherapies can generate immunogenic cell death (36) and the synergy observed in first-line nonsquamous lung cancers when combining chemotherapy with pembrolizumab (37). Because anti–PD-(L)1 antibodies have a long half-life, the next line of palliative chemotherapy is possibly testing a sequential immunotherapy/chemotherapy combination. Cases of exceptional responses to chemotherapy after anti–PD-(L)1 failures have been reported and support such a hypothesis (38).

Overall, the long-term survival to anti–PD-(L)1 monotherapies shall be confirmed in further studies, with stronger statistical power (prospective trials, with more cases, disease specific, and longer median follow-up).

The major discovery of this retrospective analysis is the impact of the type of radiologic response obtained upon anti–PD-(L)1 monotherapy and the OS of patients. We found that an objective response at first disease and best response assessment could be associated with the long-term outcome of the patient and might act as an early marker of a favorable outcome (10). Moreover, 100% of the patients in CR survived beyond 3 and 5 years. However, more than 50% of patients with PR will eventually relapse through acquired resistance to immunotherapy, which raises practical clinical questions in terms of patient management. There are currently no differences in patient care regardless of whether a patient develops a complete or a PR: we continue to infusing anti–PD-(L)1 antibodies on a biweekly or triweekly basis until progression or the development of an irAE. These results could help clinicians to validate therapy discontinuation for patients who achieve a CR after long-lasting therapy, whereas PRs who stop improving at some point should potentially receive additional treatments to prevent acquired resistance or relapse under or after immunotherapy. Stratification strategies based on the dynamics of the response under anti–PD-(L)1 might therefore be needed to increase the OS of patients treated with immunotherapy.

This retrospective study reports one of the longest disease follow-ups ever reported for anti–PD-(L)1 monotherapy across various tumor types and provides valuable insights into the long-term outcomes of patients treated with immunotherapy. In contrast to tumor cell–targeted therapies, we demonstrate that PFS can no longer be used as a surrogate for OS. Atypical radiologic responses and unexpected responses to subsequent lines of treatment seem to contribute to a significant survival benefit in a subset of patients. These results underline the observation that RECIST 1.1 should no longer be used to evaluate patients treated with immunotherapy and that novel imaging assessment criteria such as iRECIST (39) should help to capture the actual clinical activity of immunotherapies. Moreover, the long-term follow-up of anti–PD-(L)1 reveals a striking outcome difference between complete and partial responders. Most partial responders eventually develop acquired resistance, which should incite clinicians to build add-on therapeutic strategies for tumor lesions without further regression to prevent the development of secondary resistance.

S. Champiat reports receiving speakers bureau honoraria from AstraZeneca, Bristol-Myers Squibb, Janssen, MSD, Novartis, and Roche. J.-M. Michot is a consultant/advisory board member for Bristol-Myers Squibb. S. Postel-Vinay is an employee of and reports receiving commercial research grants from Merck KGaA, and reports receiving speakers bureau honoraria from AstraZeneca. E. Angevin is a consultant/advisory board member for Abbvie, Bristol-Myers Squibb, Celgene, GlaxoSmithKline, MSD, and Roche. V. Ribrag is a consultant/advisory board member for Bristol-Myers Squibb, Epizyme, Gilead, Incyte, Infinity, MSD, Nanostring, Pharmamar, Roche, and Servier. A. Hollebecque reports receiving speakers bureau honoraria from Merck Serono, and is a consultant/advisory board member for Amgen, Gritstine Oncology, Servier, and Spectrum Pharmaceuticals. J.-C. Soria is an employee of Medimmune, is a consultant/advisory board member for Astex, AstraZeneca, Clovis, Gamamabs, GlaxoSmithKline, Lilly, Merus, Mission Therapeutics, MSD, Pfizer, Pharmamar, Pierre Fabre, Roche-Genentech, Sanofi, Servier, and Takeda. C. Robert is a consultant/advisory board member for Amgen, Bristol-Myers Squibb, Merck, Novartis, Pierre Fabre, and Roche. C. Massard is a consultant/advisory board member for Amgen, Astellas, Bayer, Beigene, Bristol-Myers Squibb, Celgene, Debiopharm, Genentech, Ipsen, Jansen, Lilly, Medimmune, Novartis, Orion, Pfizer, Roche, and Sanofi. A. Marabelle reports receiving commercial research grants from Boehringer Ingelheim, Bristol-Myers Squibb, Merus, and Transgene, speakers bureau honoraria from Astra Zeneca/Medimmune, Bristol-Myers Squibb, Merck Serono, MSD, and Roche/Genentech, holds ownership interest (including patents) in Anti-CD81 mAb and Pegascy SAS, and is a consultant/advisory board member for Bristol-Myers Squibb, Medimmune, MSD, Pfizer, Roche/Genentech, Sanofi, and Symphogen. No potential conflicts of interest were disclosed by the other authors.

Conception and design: M.-L. Gauci, S. Champiat, S. Ammari, R. Bahleda, A. Gazzah, C. Massard, A. Marabelle

Development of methodology: M.-L. Gauci, E. Lanoy, S. Champiat, R. Bahleda, A. Gazzah, J.-M. Michot, A. Marabelle

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): M.-L. Gauci, S. Champiat, S. Aspeslagh, R. Bahleda, A. Gazzah, J.-M. Michot, E. Angevin, V. Ribrag, A. Hollebecque, J.C. Soria, C. Robert, C. Massard

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): M.-L. Gauci, E. Lanoy, S. Champiat, C. Caramella, S. Ammari, R. Bahleda, J.C. Soria, C. Robert, A. Marabelle

Writing, review, and/or revision of the manuscript: M.-L. Gauci, E. Lanoy, S. Champiat, C. Caramella, S. Aspeslagh, A. Varga, C. Baldini, R. Bahleda, A. Gazzah, J.-M. Michot, S. Postel-Vinay, E. Angevin, V. Ribrag, A. Hollebecque, J.C. Soria, C. Robert, A. Marabelle

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): M.-L. Gauci, S. Ammari

Study supervision: A. Gazzah, A. Marabelle

We thank all the patients who participated in these phase I clinical trials.

This retrospective research did not receive any specific grant from any funding agencies in the public, commercial, or not-for-profit sectors.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.