Efficacy and safety of two-year fixed versus continuous immune checkpoint inhibitor therapy in advanced or metastatic non-small cell lung cancer: a systematic review

Abstract

Background

The optimal duration of immune checkpoint inhibitor (ICI) therapy in advanced or metastatic non-small cell lung cancer (amNSCLC) is unknown. Most trials either continued ICI indefinitely or electively stopped at two years if no progressive disease (PD) or treatment-limiting immune-related adverse events (irAEs) emerged.

Methods

A systematic review of randomized controlled trials (RCTs) and real-world evidence studies (RWEs) was performed for adults with amNSCLC treated with ICI therapy up to August 24, 2024. Patients were divided into two cohorts: a 2 year fixed cohort in which ICI therapy was discontinued after 2 years and a continuous therapy cohort in which ICI therapy was continued beyond 2 years.

Results

Twenty studies and 5027 patients were included. The 5-year overall survival (OS) rates of the two-year fixed cohorts ranged from 69 to 83% across studies and were comparable to continuous therapy cohorts. Four RWEs compared survival outcomes between 2 year fixed and continuous cohorts and found no difference. Patients who completed 2 years of therapy in RCTs tended to have greater rates of irAEs compared to the baseline RCT population. Three RWEs reported higher rates of irAEs in the continuous versus two-year fixed cohorts. Many patients who developed PD after the two-year mark in both cohorts remained alive at the data cutoff. Larger/academic centers favored two-year fixed therapy compared with community centers.

Conclusion

Survival outcomes after ICI discontinuation at 2 years are comparable to continuous therapy in amNSCLC. IrAEs tend to accumulate over time.

Supplementary Information

The online version contains supplementary material available at 10.1007/s00262-025-04143-8.

Introduction

The optimal duration of ICI therapy in NSCLC has long been a point of contention. Phase II/III randomized clinical trials have typically either electively discontinued ICI therapy at the 2 year mark or continued till progression/unacceptable toxicity [1]. Practices in the real world appear to mix of both strategies [2–4]. The FDA approval is for continued therapy until disease progression or toxicity in amNSCLC for nivolumab, cemiplimab, and atezolizumab. For pembrolizumab, approval provides an alternative endpoint of 24 months [5–8].

Studies on the optimal duration of NSCLC are scarce. CheckMate 153 is the only RCT that, as a part of an exploratory analysis, compared outcomes after a fixed 1 year duration of ICI therapy (nivolumab) with continuous therapy and concluded that the optimal duration is > 1 year [9]. There are no randomized trials comparing outcomes at the 2 year mark.

Studies on outcomes of patients where ICI therapy was prematurely discontinued due to irAEs have shown unexpected ongoing responses [10, 11]. Investigators have also shown weak but positive associations between severity of irAEs and duration of responses [12, 13]. These data suggest that with immunotherapy, the relationship between response and dose or duration may not be linear.

Continuing ICI therapy is not without a price. Immune-related adverse events (irAEs) commonly occur during the first few weeks to months of treatment but can have delayed presentation [14, 15]. This approach could potentially lead to the accumulation of irAEs with longer treatment durations. Multiple studies have demonstrated that ICI therapies come with a high price label and pose significant financial risks to involved stakeholders [16]. One systematic review determined the incremental cost-effectiveness ratio (ICER) of nivolumab to range between 39,396 USD and 185,802 USD per quality-adjusted life year (QALY) over docetaxel [17].

Multiple first-line immunotherapy-based regimens have been approved since single-agent pembrolizumab in 2016. Many registration trials have now published long-term follow-up data in the order of 5- or 6-year median follow-up. Several RWEs have also been conducted. We conducted this systematic review to enable shared decision making on the risks and benefits of elective discontinuation of ICI therapy after 2 years in amNSCLC.

Methods

This systematic review was performed according to the recommendations of the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement.

The study protocol is registered in the international prospective register of systematic reviews (PROSPERO ID: CRD42024556987).

Eligibility criteria

Inclusion criteria

1. Adult participants with metastatic or advanced/unresectable NSCLC.

2. ICI monotherapy or in combination.

3. Minimum duration of therapy 2 years or 24 months or 35 cycles. Maximum duration was indefinite (limited by disease progression or toxicity).

4. Minimum follow-up of 3 years after the start of ICI therapy.

5. For multiple publications of the same study, the publication with the longest follow-up data was included in the analysis.

A minimum duration of 2 years was chosen as several registration clinical trials used it for elective discontinuation of therapy. Likely as a result of this body of data, many clinicians opted for this strategy and contemporary mechanisms for reimbursement of ICI therapy were designed for the 2 year mark. Combined, this led to a larger body of real-world evidence.

Exclusion criteria

1. Phase I or dose-finding trials.

2. Did not specifically report survival data for the cohort of patients who completed at least 2 years of ICI therapy (as defined above).

3. ICI therapy in the adjuvant or neo-adjuvant setting.

4. Case reports, case series, systematic reviews, meta-analyses.

5. They evaluated outcomes beyond progression only.

6. Use of small molecule targeted therapy

7. Use of experimental agents (not yet FDA-approved).

Information sources

PubMED, Embase, CENTRAL, and Web of Science were searched for eligible studies from inception up to August 25, 2024.

Retrospective studies and conference abstracts were included. No language restrictions were applied. In cases where full text was not available for an abstract, additional information was obtained either from clinical trial registries or previous publications of the same trial. In cases where the study was in a non-English language, Google translate® was used to translate the abstract into English. If the non-English study was shortlisted for full-text evaluation, the plan was to use translation services offered by TransPerfect®. If the results of the study were published in multiple articles, the publication with the longest follow-up data was considered for the purpose of this review.

Selection process and risk-of-bias assessment

One author (TP) independently extracted the following data items from articles that fulfilled criteria: first author, year of publication, clinical trial identifier, phase of trial, ICI regimen, line of current ICI regimen, number of patients who completed at least 2 years of ICI therapy, median age, percentage of male patients, duration of ICI therapy, duration of follow-up, OS, progression-free survival (PFS), and hazard ratio (HR).

Bias in RCTs was assessed using Version 2 of the Cochrane risk-of-bias tool for randomized trials (RoB 2). Bias in non-randomized trials was assessed using the Risk Of Bias In Non-randomized Studies of Interventions (ROBINS-I).

Data synthesis

The primary outcome was to report the OS and PFS of patients who a) received 2 years of continuous ICI therapy and then discontinued (2 year fixed cohort) and b) continued to receive ICI therapy beyond 2 years (continuation cohort). The secondary outcome was to report on HRs in studies that have compared the survival between 2 year fixed group and continuous groups. Additional outcomes included patterns of tumor recurrence, response to therapy, and rates of immune-related adverse events.

Results

The database search identified 8741 records. In total, 174 full-text articles were evaluated. None of the articles shortlisted for full-text evaluation were in a non-English language. The final analysis included 20 studies that specifically reported data on patients who completed at least 24 months or 2 years of immune checkpoint inhibitor therapy. See Supplementary Fig. 1 for the PRISMA flow chart [18]. The keyword search is described in electronic supplementary material.

The 20 studies included 30 cohorts from 11 RCTs and 9 RWEs for a total of 5027 patients. Of these cohorts, 23 belonged to the 2 year fixed group (N = 2051) and 7 to the continuous (N = 2976). All the 7 continuous ICI therapy groups belonged to RWEs. This is because RCTs with planned treatment beyond 2 years were either phase I studies or did not report separately on outcomes of patients that completed at least 2 years of treatment.

A number of studies, particularly the RWEs, suffered from heterogeneity and significant bias. Hence, a decision was made not to pursue a meta-analysis. Due to differences in the types of reported effects (for example, different time points for overall survival rate), tests for heterogeneity were not performed. However, all of the studies were included in the qualitative synthesis given the value of data of obtained. Risk of bias is summarized in supplementary Figs. 2 and 3.

Study and baseline patient characteristics

Table 1 summarizes the characteristics of the included patient cohorts. In 13 studies (19 cohorts), the ICI therapy studied was the first line of treatment (LoT). The vast majority of patients received either of the following regimens: pembrolizumab, nivolumab (both with or without ipilimumab and/or chemotherapy), or atezolizumab. In all the clinical trials, an ECOG performance status of 0–1 was required per inclusion criteria. In most of the RWEs, ECOG 0–1 was seen in ≥ 80% patients. The exceptions were Sun et al., where ECOG 0–1 ranged from 77 to 80%, and Gemelli et al. with only 64% in this performance status [4, 19]. Geier et al. did not report on performance status [20]. Rousseau et al. used initial inpatient administration of pembrolizumab as an imperfect substitute for poor performance status [3]. This would bring their estimated “good” performance status to the 75–80% range. Regarding discontinuation of ICI therapy at the 2 year mark, most retrospective studies used direct or indirect markers to rule out progressive disease as cause for discontinuation. Sun et al. used a 60 day window for progression and 6 month window for death after discontinuation to label the 2 year fixed cohort [4]. Fantoni et al. and Kim et al. extrapolated from a government scheme for subsidized treatment for up to 2 years [21, 22]. Oligoprogression was not considered PD as long as patient continued to receive ICI therapy. Kobayashi et al. directly excluded patients that discontinued therapy due to PD at the 2 year mark [23]. Rousseau et al. excluded patients that had more than a 6 month gap between ICI doses were considered to have discontinued therapy due to PD and were excluded [3]. In the study by Ardin et al., local practices (no clear reason given or planned discontinuation, 79%) and patient choice (18%) comprised 97% of decisions at the 2 year mark [2].

Table 1.

Characteristics of included studies

Author, year	Name of trial (trial identifier)	Study type	ICI therapy	N	Planned duration of treatment	Line of therapy	Median age (years)	Percentage of males	ECOG 0–1
Ardin et al., 2023		Retrospective	Nivolumab or pembrolizumab monotherapy	30	2 year fixed	Any line	62.7	68	93.30
				61	Continuous				91.8
Awad et al., 2021	KEYNOTE 021 cohort G (NCT02039674)	RCT, phase II	Pembrolizumab + chemotherapy combination	12	2 year fixed	1st	61.8	44.9	100
Brahmer et al., 2022	CheckMate 227 (NCT02477826)	RCT, phase III	Nivolumab monotherapy	149	2 year fixed	1st	64	68.7	100
			Nivolumab/ipilimumab	67	2 year fixed		64	64.4
Cheng et al., 2022	KEYNOTE 407 China extension (NCT03875092); Chinese cohort of global KEYNOTE 407 (NCT02775435)	RCT, phase III	Pembrolizumab + chemotherapy combination	19	2 year fixed	1st	61.2	95.2	100
De Castro et al., 2022	KEYNOTE042 (NCT02220894)	RCT, phase III	Pembrolizumab monotherapy	102	2 year fixed	2nd or later	62.8	70.8	100
Fantoni et al., 2024		Retrospective	Pembrolizumab ± chemotherapy	71	2 year fixed	1st	66	49	90
Garassino et al., 2023	KEYNOTE 189	RCT, phase III	Pembrolizumab + chemotherapy combination	57	2 year fixed	1st	63.1	58.9	100
Geier et al., 2020		Prospective	Nivolumab monotherapy	7	2 year fixed	2nd or later	58	72.7	NA
				11	Continuous
Gemelli, et al., 2023	I-STOP (NCT05418660)	Retrospective	Pembrolizumab or Nivolumab or Atezolizumab monotherapy	27	2 year fixed	Any line	67.7	69	64
				73	Continuous
Herbst et al., 2021	KEYNOTE 010 (NCT01905657)	RCT, phase II/III	Pembrolizumab monotherapy	79	2 year fixed	2nd or later	62	61.4	100
Kim et al., 2022		Retrospective	Multiple	96	2 year fixed	Any line	65	85.4	93.8
Kobayashi et al., 2023		Retrospective	Multiple	12	2 year fixed	Any line	67	65.5	92.7
				29	Continuous
Novello et al., 2023	KEYNOTE 407 (NCT02775435)	RCT, phase III	Pembrolizumab + chemotherapy combination	55	2 year fixed	1st	64.9	81.4	100
Paoli C et al., 2019		Retrospective	Nivolumab	53	Continuous	Any line	68.2	74	100
Reck et al., 2021	KEYNOTE 024 (NCT02142738)	RCT, phase III	Pembrolizumab monotherapy	39	2 year fixed	1st	64.2	61.3	100
Reck et al., 2024	CheckMate 9LA	RCT, phase III	Nivolumab/ipilimumab + chemotherapy	52	2 year fixed	1st	65	70.1	100
Rodriguez Abreu et al., 2022	KEYNOTE 598 (NCT03302234)	RCT, phase III	Pembrolizumab monotherapy	71	2 year fixed	1st	64.1	69.2	100
			Pembrolizumab/ ipilimumab	52	2 year fixed
Rousseau et al., 2024		Retrospective	Pembrolizumab monotherapy	919	2 year fixed	1st	63	65.1	NA
				2156	Continuous
Sun et al., 2023		Retrospective	Multiple regimens	113	2 year fixed	1st	69	45.1	77
				593	Continuous				80
Wu et al., 2020	KEYNOTE 042 China extension (NCT03850444); Chinese cohort of global (NCT02220894)	RCT, phase III	Pembrolizumab monotherapy	22	2 year fixed	1st	61	86	100

RCT, Randomized controlled trial; ICI, Immune checkpoint inhibitor; ECOG, Eastern cooperative oncology group performance status; NA, Not available

Long-term survivorship after 2 years of ICI therapy

The 2 year fixed treatment group

Table 2 summarizes survival outcomes of reported by included studies. In all of these with one exception, the median OS was not reached. The exception was the nivolumab + ipilimumab and PD-L1 tumor proportion score (TPS) of < 1% cohort from the CheckMate 227 trial where the mOS was still in excess of 5 years [24].

Table 2.

Survival data for 2 year fixed and continuous cohorts

Author, year	Study type	N	Planned duration of treatment	Median follow-up duration (months)	Median Overall Survival* (months)	Overall survival rate*	Progression-free survival* (months)	Progression-free survival rate*	Hazard ratio* (if applicable)	p-value
Ardin et al., 2023	Retrospective	30	2 year fixed	50.7	NR	89.2% (78.4–100) at 4 years	NR	64.1% (51.9–79.2) at 4 years	PFS 2 year fixed vs continuous 1.14 (0.54–2.30)	0.77
		61	Continuous		NR	93.1% (85.8–100) at 4 years	NR	68.5% (53.3–88.0) at 4 years
Awad et al., 2021	RCT, phase II	12	2 year fixed	49.4	NR	100% at 4 years	NR	100% at 4 years
Brahmer et al., 2022	RCT, phase III	133	2 year fixed (Nivo/ipi PD-L1 > 1%)	66.7	NR	72.0% at 5 years	43.5 (33.2 to NR)
		44	2 year fixed (Nivolumab PD-L1 > 1%)		NR	72.0% at 5 years	43.5 (19.3 to NR)
		16	2 year fixed (Nivo/ipi PD-L1 < 1%)		66.7 (48.8 to NR)	56.0% at 5 years	38.9 (20.9 to NR)
		23	2 year fixed (Nivolumab PD-L1 < 1%)		NR	64.0% at 5 years	27.7 (16.0 to NR)
Cheng et al., 2022	RCT, phase III	19	2 year fixed	44.9	NR	89.2% at 4 years
De Castro et al., 2022	RCT, phase III	102	2 year fixed	61.1	NR	61.8% at 6 years
Fantoni et al., 2024	Retrospective	71	2 year fixed	38.7	NR	98.2% (88–99.7) at 3 years	46.1 (39.5-NR)
Garassino et al., 2023	RCT, phase III	57	2 year fixed	64	NR	71.9% (58.3–81.8) at 5 years	NR
Geier et al., 2020	Prospective	7	2 year fixed	32	NR	85.7% at 3 years	NR	85.7% at 3 years
		11	Continuous		NR	100% at 3 years	NR	72.0% at 3 years
Gemelli et al., 2023	Retrospective	27	2 year fixed	48.5	NR		NR
		73	Continuous		NR		NR
Herbst et al., 2021	RCT, phase II/III	79	2 year fixed	68.1	NR	83% at 5 years
Kim et al., 2022	Retrospective	96	2 year fixed	33.9	NR	96.4% at 3 years
Kobayashi et al., 2023	Retrospective	12	2 year fixed	43.1	NR	91.7% at 3 years			TTF‡-24 continuous vs 2 year fixed 5.83 (0.76–44.63)	0.054
		29	Continuous		48.3	75.8% at 3 years
Novello et al., 2023	RCT, phase III	55	2 year fixed	57	NR	69.5% (54.8–80.2) at 5 years	NR	58.4% (39.8–73.0) at 5 years
Paoli C et al., 2019	Retrospective	53	Continuous	42.5	31.9 (18.9–42.4)		30.7 (16.8–42.4)
Reck et al., 2021	RCT, phase III	39	2 year fixed	69.7	NR	81.4% at 5 years
Reck et al., 2024	RCT, phase III	52	2 year fixed	64.5	NR	70% (56–81) at 5 years		42% (28–56)
Rodriguez Abreu et al., 2022	RCT, phase III	71	2 year fixed (Pembrolizumab)	33.6	NR	87.6% at 3 years		83.5% at 3 years
		52	2 year fixed (Pembro/ipi)		NR	86.3% at 3 years		72.8% at 3 years
Rousseau et al. (2024)	Retrospective	919	2 year fixed	65.7	NR	77.2% (72.2–82.6) at 6 years			OS continuous vs 2 year fixed HR 0.97 (0.75–1.26)	0.95
		2156	Continuous		NR	77.0% (74.2–80.0) at 6 years
Sun et al., 2023	Retrospective	113	2 year fixed	38	NR	79.0% (66.0–87.0) at 4 years			OS 2 year fixed vs continuous (adjusted) 1.33 (0.78–2.25)	0.29
		593	Continuous		NR	81% (77–85) at 4 years
Wu et al., 2020	RCT, phase III	22	2 year fixed	33	NR	56.6% at 3 years

OS, Overall survival; PFS, Progression-free survival; TTF-24, Time to treatment failure after 24 months; RCT, Randomized controlled trial; NR, Not reached; Pembro, Pembrolizumab; Nivo, Nivolumab; Ipi, Ipilimumab

^*Median Overall Survival, Overall Survival Rate, Median Progression-free Survival, and Progression-free Survival Rate: Values in square brackets indicate the upper and lower limits of the 95% confidence interval

The 6 year OS for KEYNOTE 042 was 61.8% [25]. The 5 year OS for KEYNOTE 189, KEYNOTE 010, KEYNOTE 407, KEYNOTE 024, and CheckMate 9LA ranged from 69.5 to 83% [26–30]. KEYNOTE 021 cohort G, Cheng et al., and KEYNOTE 598 had shorter follow-up durations and also demonstrate high survival rates [31–33]. The RCT with the lowest survival rate was by Wu et al.: The 3 year OS rate stood at 56.6% [34].

Data from RWEs are not dissimilar. The median OS was not reached in any of the studies. Rousseau et al. published the largest study included in this review: 919 patients who received a 2 year fixed course of first-line pembrolizumab had a 6 year OS of 77.2% [3]. Geier et al. published the only prospective RWE in this review. Seven patients received a 2 year fixed course of nivolumab at any LoT of which 6 were alive at 3 years [20]. In Ardin et al.’s retrospective study looking at pembrolizumab or nivolumab monotherapy at any LoT, the 4 year OS rate stood at 89.2% [2]. Fantoni et al. reported a 3 year OS rate of 98.2% after a 2 year course of first-line pembrolizumab in 71 patients [21]. Sun et al. reported a 4 year OS of 79% after a 2 year course of a mix of ICI regimens in 113 patients [4].

Continuous treatment group

Rousseau et al. found a 6 year OS of 77% among 2156 patients in their continuous pembrolizumab cohort [3]. Sun et al. reported 4 year OS of 81% in the continuous cohort [31]. Ardin et al. reported a 4 year OS rate of 93.1% for 61 patients in the continuous cohort [2]. In Gemelli et al.’s retrospective study, the 73 patients in the continuous treatment arm did not reach a median OS due to lack of events [19]. In Geier et al.’s small prospective study, all 11 patients in the continuous treatment arm were alive at the 3 year mark [20]. Paoli C et al. presented results from a retrospective study of 53 patients who received continuous nivolumab at any LoT and reported a mOS of 31.9 months [35].

Comparison between 2 year fixed and continuous treatment groups

Six RWEs compared survival between 2 year fixed and continuous ICI treatment cohorts. No randomized controlled trials compared these two treatment cohorts. Despite attempts to address potential confounders, a serious or critical risk of bias was noted for all RWEs (Supplementary Fig. 3).

Rousseau et al. had 2156 and 919 patients in each group and reported an HR of 0.97 [95% CI, 0.75–1.26] that was statistically not significant [3]. This HR was calculated after weighting on a propensity score based on several co-variables thought to be potential confounders.

Sun et al. compared the OS following 2 year fixed (113) or continuous regimens (593) of multiple first-line ICI regimens [4]. OS rates were similar as noted previously, and the adjusted HR was 1.33 [95% CI, 0.78–2.25], which was also statistically insignificant. Adjustment was performed based on a multivariable Cox regression.

Ardin et al. retrospectively analyzed 30 and 61 patients in 2 year fixed and continuous nivolumab cohorts, respectively [2]. They reported an HR of 1.14 [95% CI, 0.54–2.30] for PFS that was not statistically significant. They also found no significant difference in PFS after calculating the HR on multivariate analysis.

Kobayashi et al. compared time to treatment failure after 24 months of therapy (TTF-24) between 12 patients in the 2 year fixed and 29 in the continuous cohort [23]. TTF was calculated from ICI treatment initiation to death due to any cause or PD. TTF-24 was defined as TTF after the first 24 months. The univariate HR for TTF-24 for continuous vs 2 year fixed regimen was 5.83 [95% CI, 0.76–44.63], indicating a tendency for shorter TTF-24 with a continuous regimen. This was not statistically significant due to the wide CI (p-value 0.054). On multivariate analysis as well, no statistically significant difference was found (p = 0.0811).

Gemelli et al. presented preliminary results from their ongoing retrospective study, I-STOP [19]. They compared 27 patients in the 2 year fixed with 73 in the continuous cohort receiving pembrolizumab at any LoT. At a median follow-up of roughly 4 years, 20.8% patients in the 2 year fixed and 21.9% patients in the continuous cohort had PD. No log-rank test was reported.

Rates of immune-related adverse events (irAEs)

RCTs that reported on cumulative rates of irAEs in the cohort that completed the 2 year fixed course and the starting cohort are included here. In the 2 year fixed cohort from KEYNOTE 189, the rate of irAEs was 40.4% with grade ≥ 3 events in 12.3%. This was higher than the study’s original cohort (27.9% had irAEs) but with a similar severity (13.4% had grade ≥ 3) [26]. KEYNOTE 407 reported irAEs in 38.2% patients and one grade 3 event (1.8%) [27]. This was like their original cohort, but with a lower severity (irAEs in 35.6% and 13.3% grade ≥ 3). The same pattern was seen in KEYNOTE 010 and KEYNOTE 042: The irAE rates were higher in the 2 year fixed cohorts compared to the original cohort (39.2% vs 23% and 40.2% vs 27.5%, respectively), while the severity was the same or lower (grade ≥ 3 in 6.3% vs 6.3% and 5.9% vs 8.2%) [25, 28]. The exception to this rule was KEYNOTE 024: The rate of irAEs was higher in the original cohort than the 2 year fixed cohort (34.4% vs 30.8%), but the severity was greater in the 2 year fixed cohort (7.7% vs 5.3% were grade ≥ 3) [30].

Ardin et al. reported incidence of adverse events after the 2 year mark [2]. In the continuous treatment group, 23% of patients experienced irAEs and 3.3% had a grade 3 event. One patient (3%) in the 2 year fixed group experienced two irAEs.

Kobayashi et al. also reported irAEs after the 2 year mark [23]. One patient in the 2 year fixed group (8.3%) had a grade 3 irAE that was due to a second course of ICI treatment. In contrast, 24% of the continuous treatment group had irAEs.

Paoli et al. reported a high incidence of irAEs: 80%, in their single continuous treatment arm from the initiation of ICI therapy [35]. Grade ≥ 3 irAEs were seen in 5% of patients.

Geier et al. reported on cumulative incidence of irAEs [20]. In their 2 year fixed duration arm, 85.7% of patients had irAEs of any grade, and none had grade ≥ 3. On the other hand, 72.7% of patients in their continuous treatment arm had irAEs of any grade and 9.1% were grade ≥ 3.

It must be noted that comparisons between studies are difficult to perform due to variations in definitions of irAEs, the fact that certain RWEs grouped infusion-related reactions with irAE and the use of different timeframes for reporting (cumulative or after the 2 year mark).

Patterns of progression and response to re-treatment

Insights regarding progressive disease and clinical course were gleaned from swimmer’s plots and re-treatment patterns.

In KEYNOTE 189, 21 of the 57 patients in the 2 year fixed cohort developed PD. Of these, only five died by the end of follow-up [26]. Eight of the 21 patients with PD were given a second course of pembrolizumab, four of which belonged to five that died while the other four were alive at data cutoff.

In KEYNOTE 407, 11 of the 55 patients in the 2 year fixed cohort developed PD. Six of these were alive at data cutoff [27]. Four of those with PD were started on a second course of pembrolizumab, and all four were alive at data cutoff.

The investigators of KEYNOTE 010 provided a more detailed analysis of the 2 year fixed cohort [28]. Thirty-three of the 79 patients developed PD. Of these, 20 remained alive at data cutoff. Twenty-one patients with PD were started on a second course of pembrolizumab. Eleven of these had an objective response (52.3%). At data cutoff, 15 of these 21 patients (77.4%) were alive.

In KEYNOTE 042, 33 patients that developed PD after completing 2 years of therapy were selected to re-treated with pembrolizumab and were given a second course of pembrolizumab [25]. Of these, 12 were alive at data cutoff. They achieved a partial response in 15.2% and stable disease in 60.6% for a disease control rate of 75.8%.

Results from the 39 patients from the 2 year fixed cohort in KEYNOTE 024 showed that 13 patients developed PD [30]. Of these, 11 were alive at data cutoff. Ten of the 13 had received a second course of pembrolizumab.

Moving on to RWEs, Ardin et al. noted that a combined 30 patients out of the 91 from the 2 year fixed group and continuous group experienced relapse [2]. Of these, 23 cases involved a single tumor site, another 2 had bifocal progression, and the remaining 5 had systemic progression.

Sun et al. reported that 11 out of 113 patients in their 2 year fixed duration cohort had PD and were re-treated with ICI ± chemotherapy [4]. In all cases except for one, the same ICI therapy was used. After the re-treatment, a median PFS of 8.1 months was seen. Eight of the 11 patients with PD were alive at data cutoff.

In the retrospective study by Fantoni et al., 20 patients of the 71 in the 2 year fixed cohort of pembrolizumab had PD [21]. Eleven of these (55%) had oligoprogression. They do not provide a definition for what was considered oligoprogression, but it appears to have been disease recurrence amenable to locoregional therapy. One of these was re-treated with pembrolizumab. Fourteen of these 20 patients were alive at data cutoff.

Kobayashi et al. reported outcomes in the form of treatment failure: a composite of death or PD [23]. They did not consider oligo-progression in this definition. Of the 12 patients in the 2 year fixed cohort, 2 had oligo-progression after the 2 year mark. One had treatment failure by data cutoff. Of the 29 patients in the continuous group, 6 had oligo-progression after the 2 year mark. Three of these had treatment failure by data cutoff.

In the retrospective study by Kim et al., 12 of the 96 patients in the 2 year fixed cohort went on to develop PD [22]. Of these, 10 were alive at data cutoff.

Results from I-STOP showed that 16 patients (21.9%) in the continuous cohort and 5 patients (20.8%) in the 2 year fixed cohort developed PD [19]. Two patients in the continuous cohort died, while the rest including those in the 2 year fixed cohort were alive at data cutoff.

University vs community setting

In the study by Ardin et al., 23% of patients at university hospitals continued ICI therapy indefinitely, while 95% did so in community centers [2].

Rousseau et al. performed a multivariate logistic regression and found that the only factor independently associated with pembrolizumab continuation after the 2 year mark was being treated in an intermediate- or low-volume center (OR = 2.33 [1.94–2.82] p < 0.001 and OR = 2.14 [1.64–2.82] p < 0.001, respectively) [3].

Sun et al. found that patients in the 2 year fixed duration group vs continuous group were more likely to be treated in academic center (25 of 113 [22%] vs 65 of 593 [11%]; p = 0.001) [4].

Discussion

The present study is the first systematic review addressing the question of benefits vs risks of continuing ICI therapy beyond 2 years in amNSCLC. Data from RCT cohorts with 2 year fixed treatments showed impressive 5- and 6-year OS rates ranging from 60 to 90%. Data from RWEs had shorter follow-ups but showed very high OS rates in the 2 year fixed treatment groups. Data from continuous cohorts, available only from RWEs, had OS rates similar to the 2 year fixed groups. In all four log-rank tests performed to compare 2 year fixed vs continuous cohorts, the HRs were statistically not significant. All of the RWEs, however, had serious or critical bias stemming from selection and/or confounding factors. This is in part due to the retrospective nature and due to lack of data for why providers chose one treatment strategy over the other.

The investigators of CheckMate 153 presented an exploratory analysis in 2020 on a subset of the main study. The main study was a phase IIIb/IV RCT assessing the safety of nivolumab monotherapy in previously treated advanced NSCLC [9]. They randomized 252 patients who had completed a year of nivolumab to continue therapy or stop at the 1 year mark. A significantly longer survival (mPFS and mOS) was found in the continuous arm (mPFS of 24.7 months vs 9.4 months, HR 0.56 [95% CI, 0.37–0.84] and mOS of not reached v 28.8 months; HR 0.62 [95% CI, 0.42–0.92]). The results strongly favored continuing ICI therapy beyond 1 year. The study was limited by its exploratory nature. The median time on treatment for the continuous group was around 2 years (25.6 months).

Taking the present review and the CheckMate 153 together, it is reasonable to conclude that the optimal duration is more than 1 year, and 2 years seem to be a reasonable discontinuation point. The fact that a patient survives 2 years after initiating ICI therapy for amNSCLC may be a strong predictive factor in itself. This is supported by the long-term survival exhibited by the 2 year fixed treatment groups (RCTs and RWEs) in the present study. The 2 year time point may be close to a hypothetical inflection point beyond which diminishing returns are seen with continued treatment. This can be gleaned from the comparable survival outcomes in 2 year fixed and continuous treatment groups seen in RWEs. It must be noted that while the 5- and 6-year OS rates seen in the 2 year fixed groups of RCTs are remarkably long, these stem from a highly selective group of patients with favorable performance status. Data from RWEs have to mature further to corroborate these findings in a wider population.

Bogani et al. published a meta-analysis to determine the optimal duration of ICI therapy in six solid tumors including amNSCLC [36]. They included 28 RCTs on NSCLC and compared survival between planned 2 year fixed ICI therapy vs continuous therapy and stratified it according to whether standard of care (SoC) was also administered concomitantly. Since no direct comparisons between groups were available, they indirectly calculated the comparisons after extrapolation. They found that 2 year fixed IC monotherapy did not make an impact on PFS or OS compared to continuous therapy. However, 2 year fixed therapy with SoC had significantly better PFS and OS compared to continuous therapy with SoC. A major limitation was that they had no means to determine which patients in the included cohorts had completed a minimum of 2 years of ICI therapy. Since patients that complete 2 years are typically a minority, it implies that only a small portion of the effects observed by Bogani et al. can be attributed to that group.

Hu et al. performed a meta-analysis to report on treatment-free survival (TFS) after ICI discontinuation in metastatic NSCLC [10]. They determined this by extracting patient-level data from published swimmer’s plots from 17 prospective studies. One of the outcomes reported was TFS in patients who completed at least 35 cycles of pembrolizumab. They found that in this cohort of patients, TFS after ICI discontinuation was considerably longer than all other treated patients.

In the present study, rates of irAEs were higher in those who completed 2 years as opposed to those who were treated for shorter durations, although the rates of grade ≥ 3 events were same or lower. It stands to reason that with time, irAEs accumulate, as noted in the literature [37, 38]. Simply tolerating the therapy for a longer duration might explain the cumulative toxicities. However, there is a chance of selection bias for less severe irAEs in the longer-duration groups, as those with more severe events will have ICI therapy discontinued early. Rates of irAEs were greater in continuous groups as opposed to 2 year fixed groups, favoring discontinuation at 2 years.

In our study, patients treated in larger centers tended to have their ICI therapy discontinued at the 2 year mark. Possible reasons include access to a multidisciplinary team at larger centers to aid decision making, more experience with similar cases, and awareness of newer data supporting elective discontinuation.

A significant proportion of patients with PD in the present review were re-treated with ICI therapy in the form of monotherapy or combination therapy with the same or a different checkpoint inhibitor. Literature has shown that ICI re-treatment is a common real-world practice, induces response in a significant proportion of patients, and is not associated with higher rates of irAEs [39, 40]. Many patients had limited disease recurrence amenable to locoregional therapy.

The financial impact of interventions is often reported using ICER. Verma et al. showed in their systematic review that most studies have pegged the ICER for nivolumab over docetaxel above 100,000 USD/QALY, which is a commonly used threshold for considering an intervention as cost-effective [17]. Matter-Walstra et al. showed that reducing the dose or duration of nivolumab for advanced non-squamous NSCLC can reduce the ICER [41]. Specifically, restricting the duration of nivolumab to 3 months brought the ICER down from 177,478 Swiss francs/QALY to 110,349 (46.6% probability that the intervention will be cost-effective compared to docetaxel).

The strengths of this review lie in the focus on NSCLC and data from patients who specifically completed at least 2 years of therapy. Data from over 5000 patients with long median follow-up durations provide a robust picture of the clinical course.

This review has several limitations. Several RCTs were unblinded. Most RWEs suffered from serious biases, and some had small sample sizes. There was significant heterogeneity in treatment regimens including LoT and through the inclusion of both ICI monotherapy and combination therapy. We did not analyze the impact of PD-L1 status on outcomes. No RCTs directly compared survival between 2 year fixed and continuous cohorts. These direct comparisons of the two strategies came only from RWEs, and thus, conclusions must be drawn with caution. Data regarding other standard ICI therapies, such as cemiplimab and durvalumab, could not be included due to a lack of data specific to cohorts that completed 2 years of treatment.

We conclude that survival outcomes following 2 year fixed vs continuous ICI therapy are similar. Two years are a reasonable time point for discontinuation of therapy for those without progressive disease. Immune-related adverse events tend to accumulate over time, while rates of severe events are stable or lower. Progressive disease off therapy can be amenable to ICI re-treatment, with a significant percentage of patients benefitting from a second course of immunotherapy. Continuous ICI therapy increases the overall healthcare costs and financial toxicity.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contribution

T.P. performed the literature search, evidence synthesis and wrote the manuscript. S.B. designed the study concept, contributed to the discussion section and performed critical appraisal of the manuscript.

Data availability

No datasets were generated or analysed during the current study.

Declarations

Conflict of interest

The authors declare no conflict of interest.