Local ablative treatment for oligoprogression during immunotherapy in metastatic non-small-cell lung cancer: a systematic review and meta-analysis

ABSTRACT

Some patients with metastatic non-small-cell lung cancer (NSCLC) receiving anti-PD-1 or anti-PD-L1 immune checkpoint inhibitors (ICIs) progress at a few sites while disease elsewhere stays controlled, a pattern consistent with focal, oligo-acquired immune escape rather than systemic treatment failure. Ablating the escaping focus may eliminate the resistant clone and preserve an otherwise effective immunotherapy line. We performed a PRISMA 2020-compliant, PROSPERO-registered systematic review and meta-analysis of local ablative treatment (LAT) for oligoprogressive or oligo-resistant metastatic NSCLC during active ICI therapy, with continuation of the same immunotherapy line after LAT. MEDLINE, PubMed, Web of Science, Scopus, and the Cochrane Library were searched from inception to March 8, 2026. From 1270 records, 22 studies were included in the qualitative synthesis. Four studies, one randomized subgroup, and three retrospective cohorts contributed to each primary analysis. LAT-oriented management was associated with improved progression-free survival (PFS; pooled hazard ratio [HR] 0.45, 95% CI 0.33–0.62; 95% prediction interval [PI] 0.28–0.74; I² = 0%) and overall survival (OS; pooled HR 0.41, 95% CI 0.27–0.63; 95% PI 0.24–0.69; I² = 0%). PFS and OS estimates were consistent across sensitivity analyses, including the exclusion of the one PFS estimate measured to a second progression. The evidence is predominantly observational and of low certainty, reflecting the current absence of NSCLC-specific randomized trials. On this evidence, LAT, delivered predominantly as stereotactic body radiotherapy, may be considered a continuation-enabling option for carefully selected patients with focal progression on otherwise effective immunotherapy, pending ongoing randomized trials.

Introduction

Metastatic non-small-cell lung cancer (NSCLC) is not one disease state. Contemporary registries indicate that 30%–40% of newly diagnosed stage IV cases meet anatomical criteria for limited metastatic spread, ¹ and, among patients progressing on anti-PD-1 or anti-PD-L1 immune checkpoint inhibitors (ICIs), between 30% and 60% do so at a limited number of sites. ^{2 , 3} In a lung cancer cohort enriched for acquired resistance to anti-PD-1 or anti-PD-L1 therapy, Schoenfeld and colleagues ⁴ found that oligo-acquired resistance, with disease elsewhere remaining controlled, was present in 56% of patients; these patients had higher baseline tumor mutational burden and deeper, longer initial responses than patients with systemic acquired resistance. A plausible explanation is immune editing: antigen-recognizable clones are eliminated, whereas the clones that persist at focal sites have acquired immune evasion through PD-L1 heterogeneity, loss of antigen presentation, or oncogenic signaling that restores proliferative autonomy independent of checkpoint pressure. ^{4 , 5} Systemic benefit can persist while one or two lesions escape.

Three related terms recur in this literature and are not interchangeable. Oligoprogression, as characterized by the ESTRO and EORTC consensus, denotes progression of a few metastases during active systemic therapy while other sites remain controlled, without a fixed numerical threshold. ⁶ Oligo-acquired resistance is a narrower entity, defined by Schoenfeld and colleagues as progression at three or fewer lesions after a documented complete or partial response to anti-PD-1 or anti-PD-L1 therapy, and is therefore a subset of oligoprogression specific to ICI therapy. ⁴ Focal immune escape describes the underlying biology, namely, spatially restricted loss of immune control through impaired antigen presentation or reduced HLA class I expression, and is a mechanistic explanation rather than a disease-state classification. ⁷ We use oligoprogression throughout as the clinical entity under study.

If most of the disease remains controlled, local ablative treatment (LAT) of the escaping focus may preserve the active treatment line and defer a systemic switch when no clearly superior alternative exists. Stereotactic body radiotherapy (SBRT), also termed stereotactic ablative radiotherapy, achieves 1-y local control rates exceeding 90% at anatomically challenging sites, with grade 3 or higher toxicity below 5%. ⁵ Ablative doses also induce immunogenic cell death, liberating tumor-associated neoantigens and activating antigen-presenting cells via the stimulator of interferon genes (STING) pathway, priming systemic T-cell responses. ^8-10 Chicas-Sett et al. ¹¹ documented abscopal responses in 65% of evaluable patients after SBRT combined with anti-PD-1 therapy in a prospective multicenter series. In a pooled analysis of the PEMBRO-RT and MD Anderson trials, the abscopal response rate was 41.7% versus 19.7% with the addition of radiotherapy to pembrolizumab (odds ratio 2.96; 95% CI 1.42–6.20). ¹⁰ The rationale for LAT, therefore, extends beyond reducing tumor volume.

A classification problem has distorted the evidence base. The ESTRO-EORTC consensus delineated four disease categories: synchronous oligometastatic disease at initial staging, oligoresidual disease persisting after maximum systemic response, oligoprogression during active effective treatment, and oligo-recurrent disease after a treatment-free interval. ⁶ Most existing reviews and meta-analyses have merged at least two of these states. ^12-15 The clinical question, comparator, and progressive-clone biology differ across them. Consolidative ablation of residual disease after maximum response, as studied in recent prospective and comparative series, ^16-20 is not the same intervention as treating a site that has progressed during ongoing treatment. Pooling these states obscures any state-specific effect.

Previous evidence syntheses reflect this problem. Existing reviews either report no pooled estimate for LAT with continuation of the same immunotherapy line, pool across primary tumor types and systemic backbones, or combine oligoprogression with synchronous and oligoresidual disease. ^{13-15 , 21-24}

We therefore restricted this review to oligoprogression during active immunotherapy. Quantitative synthesis was limited to studies meeting the prespecified pooling criteria, and randomized evidence that did not satisfy the assumptions required for hazard-ratio synthesis was retained narratively.

Prospective evidence in this setting is sparse and internally discordant. The prespecified NSCLC subgroup of CURB ²⁵ showed a progression-free survival benefit, while its breast cohort did not, and STOP ²⁶ was null but confounded by 71% crossover. The retrospective literature is directionally consistent, but the decision to offer LAT is not independent of prognosis.

Methods

Search strategy and selection criteria

This review followed the PRISMA 2020 guidelines. ²⁷ The protocol was prospectively registered in PROSPERO (CRD420261281271). We searched MEDLINE, PubMed, Web of Science, Scopus, and the Cochrane Library from database inception to March 8, 2026. Search terms combined NSCLC, ICIs, oligoprogression, oligometastatic disease, LAT, SBRT, surgery, and continuation of systemic therapy. Reference lists, conference proceedings, and prior reviews on treatment after immunotherapy progression were screened. ²⁸ The full search strategies for all four databases are provided in Supplementary Appendix 1.

Eligible studies enrolled adults with metastatic NSCLC receiving anti-PD-1 or anti-PD-L1 therapy and reported outcomes after LAT delivered for oligoprogressive or oligo-resistant disease in which the intended strategy was continuation of the same immunotherapy line after LAT (without interruption or after a temporary hold and resumption). The continuation was not uniform across every patient or comparator arm, and this variation is detailed in the Results and Discussion. We included randomized trials, prospective cohort studies, and retrospective observational studies. Case reports, case series with fewer than five patients receiving LAT, review articles, preclinical studies, and reports without extractable clinical outcome data were excluded. Studies focused exclusively on synchronous oligometastatic disease at first presentation or on post-response consolidation were excluded from the review; the near-eligible reports are listed in Supplementary Appendix 3. Disease-state classification is detailed in Supplementary Appendix 2.

Two reviewers (IM and AH) independently screened titles and abstracts, reviewed eligible full-text reports, and extracted data. Disagreements were resolved by discussion with a third reviewer (OW or NP). Data extracted included study design, sample size, patient characteristics, disease state definition, immunotherapy agent and line, LAT modality and intent, comparator structure, oligoprogression definition, continuation strategy after LAT, outcome definitions, time-to-event estimates, and risk-of-bias variables. Inclusion in the systematic review and in the meta-analysis were adjudicated separately.

Data analysis

The primary outcomes were progression-free survival (PFS) and overall survival (OS). Adjusted hazard ratios (HRs) were preferred when explicitly reported. When only unadjusted HRs were available, we extracted them and recorded that limitation in the evidence profile. If logarithmic HRs and standard errors were not tabulated, we derived them from reported HRs and 95% confidence intervals by standard back-calculation (standard error =  $[\ln ({\text{HR}}_{\text{upper}})-\ln ({\text{HR}}_{\text{lower}})]/3.92$ ). We did not reconstruct time-to-event outcomes from Kaplan‒Meier curves when the reported data could not support a defensible proportional-hazards summary. CURB ²⁵ enrolled both breast cancer and NSCLC patients; we used the prespecified NSCLC subgroup HR rather than the overall trial estimate because the breast cancer arm showed no signal. This remains a subgroup estimate. By ICEMAN criteria, its credibility is moderate, and confirmation still requires an NSCLC-specific intention-to-treat randomized analysis.

PFS was extracted as defined by each study. Brown et al. ²⁹ reported second progression-free survival (PFS2), measured from the start of first-line immunotherapy to the second progression or death; in their oligoprogression cohort, this was the progression that followed the oligoprogression episode. We retained it in the primary PFS analysis because its construction matches the PFS endpoint of Jiang et al. ³⁰ measured from the start of ICI therapy to the first progression after the atypical response, and because every pooled PFS endpoint counts the same event: progression or death after the oligoprogression episode. The time zero differed (start of ICI therapy for Brown et al. and Jiang et al., start of post-progression treatment for Xuzhang et al. ³ and randomization for CURB) and is reported in Table 2. Sensitivity analyses excluded PFS2 and restricted the PFS pool to studies with time zero at or after oligoprogression.

STOP ²⁶ was retained as a qualitative randomized anchor but not pooled for PFS. Its Kaplan‒Meier curves cross before the median, and the reported PFS HR of 0.81 (95% CI 0.48–1.37) does not satisfy a proportional-hazards assumption. We derived an approximate OS HR from reported median survival values and used it only as contextual evidence. All pooled models used HRs as the effect measure; trials with non-proportional hazards were excluded from quantitative synthesis.

Random-effects meta-analysis used restricted maximum likelihood (REML) estimation with Knapp-Hartung adjustment, implemented in the meta package in R. ³¹ We computed 95% prediction intervals alongside confidence intervals and quantified heterogeneity with I² and τ². The primary analysis was restricted to studies in which oligoprogressive or oligo-resistant disease was the dominant clinical setting and a LAT-versus-control comparator had an extractable time-to-event estimate (Tier 1 evidence set). An expanded exploratory sensitivity analysis added comparative studies with broader population definitions or a timing-based comparator (Tier 1 plus Tier 2); because these do not all share the clinically aligned LAT-versus-control structure, they test directional robustness rather than the primary contrast. A sensitivity analysis restricted to randomized studies and subgroup analyses by study design and immunotherapy line was prespecified; neither was reported because only one randomized study met the primary pooling criteria, and the treatment line was not reported uniformly across studies. Leave-one-out analysis assessed whether any single study drove the primary estimates. Funnel plots and Egger’s test were prespecified but not conducted because each model contained fewer than ten studies.

The risk of bias was assessed with Cochrane RoB 2 for randomized trials and the Newcastle‒Ottawa Scale for observational studies. Certainty of evidence was summarized per outcome using the GRADE approach, with domain-by-domain adjudication performed by the senior authors.

Results

The study-selection process is shown in Figure 1. The database searches identified 1270 records. After the removal of duplicates, 745 studies were screened at the title and abstract level, and 193 full-text reports were assessed for eligibility. Of these, 171 were excluded (wrong publication type, n = 68; wrong population, n = 58; wrong study design, n = 20; wrong outcome reporting, n = 14; mixed or non-isolable population, n = 11). Twenty-two studies were retained in the qualitative synthesis. The near-eligible reports excluded at full-text review are listed with reasons in Supplementary Appendix 3.

Figure 1.

PRISMA 2020 study-selection flow diagram.

The study characteristics are shown in Table 1. The 22 included studies were published between 2019 and 2026. Combined parent-cohort enrollment was about 2101 patients. Several studies reported mixed primary histologies, mixed disease states, or pattern-of-failure denominators wider than this review question; the population directly analyzable for oligoprogressive NSCLC on active immunotherapy was therefore approximately 1637 patients. Five studies contributed to the primary analyses: Brown et al. ²⁹ Friedes et al. ³² Jiang et al. ³⁰ Xuzhang et al. ³ and Tsai et al. ²⁵ (the prespecified NSCLC subgroup of CURB). Four studies contributed to the PFS analysis (all except Friedes et al.; Table 1), and four to the OS analysis (all except Jiang et al., who did not report OS). The set comprised four retrospective cohorts and one randomized phase 2 trial; LAT arms ranged from 16 to 42 patients and control arms from 20 to 91, with a median follow-up of about 12–32 months.

Table 1.

Study characteristics, ordered by evidence tier.

Study	Design	N (LAT/Ctrl)	Context	LAT approach	PFS HR (95% CI)	OS HR (95% CI)	Tier
Tier 1 – primary comparative evidence and randomized anchor
Brown et al.* [29]	Retrospective cohort	42/20	Primary comparative*	Radiotherapy	0.51 (0.29–0.89)*	0.4 (0.22–0.73)	Tier 1*
Friedes et al. [32]	Retrospective cohort	39/40	Primary comparative	Radiotherapy	NR ‡	0.35 (0.2–0.62)	Tier 1
Jiang et al. [30]	Retrospective multicenter	16/40	Primary comparative	Mixed LAT	0.25 (0.08–0.84)	NR	Tier 1
Tsai et al. [25]	Randomized Phase 2 trial	31/28	Primary comparative	Radiotherapy	0.41 (0.22–0.75)	0.72 (0.3–1.72)	Tier 1
Xuzhang et al. [3]	Retrospective single-center	17/91	Primary comparative	Mixed LAT	0.48 (0.3–0.78)	0.39 (0.21–0.74)	Tier 1
Schellenberg et al. [26]	Randomized Phase 2 trial	59/31	Randomized anchor	Radiotherapy	NR	NR	Tier 1
Tier 2 – broader comparative/exploratory sensitivity evidence
Belaidi et al. [33]	Retrospective single-center	21/9	Timing comparison	Radiotherapy	0.56 (0.24–1.32)	1.21 (0.3–4.9)	Tier 2
Kothari et al. † [34]	Randomized Phase 2 (NIVORAD; ALTG14/002; TROG 16.01)	34/16	Sensitivity comparative †	Radiotherapy	0.68 (0.36–1.27)	0.86 (0.43–1.75)	Tier 2 †
Xu et al.** [35]	Retrospective single-center	38/77	Exploratory sensitivity**	Radiotherapy	0.67 (0.44–1.01)**	0.45 (0.25–0.81)**	Tier 2**
Tier 3 – narrative, comparative non-pooled, and pattern-of-failure evidence
Ansari et al. [36]	Retrospective single-center	22/NR	Narrative	Radiotherapy	NR	NR	Tier 3
Chicas-Sett et al. [11]	Prospective observational	50/NR	Narrative	Radiotherapy	NR	NR	Tier 3
Deng et al. [2]	Retrospective single-center	NR/NR	Narrative	Mixed LAT	NR	NR	Tier 3
Hirano et al. [37]	Retrospective single-center	10/10	Comparative non-pooled	Mixed LAT	NR	NR	Tier 3
Huet et al. [38]	Retrospective multicenter	61/NR	Narrative	Mixed LAT	NR	NR	Tier 3
Kagawa et al. [39]	Retrospective single-center	10/28	Comparative non-pooled	Mixed LAT	NR	NR	Tier 3
Mahmood et al. [40]	Retrospective single-center	NR/NR	Narrative	Radiotherapy	NR	NR	Tier 3
Mahmood et al. [41]	Retrospective multinational	103/NR	Narrative	NR	NR	NR	Tier 3
Metro et al. [42]	Retrospective multicenter	9/42	Comparative non-pooled	Radiotherapy	NR	NR	Tier 3
Rheinheimer et al. [43]	Retrospective single-center	NR (50% extracranial oligoprogression received LAT)/NR	Narrative	Radiotherapy	NR	NR	Tier 3
Tsur et al. [44]	Retrospective single-center	22/NR	Narrative	Radiotherapy	NR	NR	Tier 3
Wang et al. [45]	Retrospective single-center	24/NR	Narrative	Radiotherapy	NR	NR	Tier 3
Ying et al. [46]	Retrospective single-center	15/NR	Narrative	Radiotherapy	NR	NR	Tier 3

Abbreviations: HR, hazard ratio; ICI, immune checkpoint inhibitor; LAT, local ablative treatment; NR, not reported; OS, overall survival; PFS, progression-free survival; PFS2, second progression-free survival.

^* Brown et al. reported PFS2, measured from the start of first-line immunotherapy to the second progression or death, and continuation of immunotherapy after progression was not uniform; the estimate is included in the primary PFS analysis (rationale in Methods).

^** Xu et al. is an exploratory composite contrast of PFS2 and OS in patients with oligoprogression, oriented so that HRs below 1.0 favor continued ICI therapy plus radiotherapy-based local management.

^† NIVORAD is a broader randomized comparison used in exploratory sensitivity analysis.

^‡ Friedes et al. reported PFS only for oligoprogression versus polyprogression, not by treatment received.

Radiotherapy was the dominant modality. Of the 145 patients in the LAT arms of the five primary studies, 112 were enrolled in studies in which radiotherapy was the only permitted modality (Brown et al., Friedes et al., and CURB). Jiang et al. ³⁰ reported the distribution for their 16 treated patients as radiotherapy in 13, surgery in 2, and another local intervention in 1. At least 125 of 145 patients (86%) therefore received radiotherapy, and surgical resection is documented in only 2. Xuzhang et al. ³ recorded LAT as a binary variable in 17 patients and did not disaggregate radiation from surgery. Modality was not used as a stratification variable. Sensitivity analyses restricted to the radiotherapy-only studies are reported below; those studies permitted both ablative and conventionally fractionated techniques, so these analyses do not isolate ablative dose. Most patients had received first-line or second-line immunotherapy, and progression was limited to 1–5 sites at the time of LAT (per-study thresholds and time zero in Table 2), with one study enrolling atypical mixed responses without a cap. Schellenberg et al. ²⁶ (STOP) was retained as the principal negative randomized anchor (90 patients, 59 irradiated and 31 control, mostly NSCLC). Belaidi et al. ³³ Kothari et al. ³⁴ (NIVORAD), and Xu et al. ³⁵ contributed to qualified sensitivity analyses but not the primary pool because their population definitions were broader, their comparator was a timing contrast, or the continuation strategy was less uniform.

Table 2.

Clinical characteristics of the five studies contributing to the primary analyses.

Study	Median age (y)	Progressive-lesion cap	Immunotherapy line	Continuation after LAT	LAT modality	Time zero, PFS; OS
Brown et al. [29]	68	≤3	First line	20 of 42 continued after progression	Radiotherapy (SBRT, conventional, SRS)	First-line ICI start (PFS2); first-line ICI start
Friedes et al. [32]	66	≤3	First line	Radiotherapy without systemic switch	Radiotherapy (SBRT, hypofractionated, IMRT, SRS, WBRT)	ICI start ‡ ; diagnosis or recurrence
Jiang et al. [30]	NR	None stated (atypical response)	First line and later	All continued the original ICI	Radiotherapy 13, surgery 2, other 1	ICI start; OS not reported
Xuzhang et al. [3]	63*	≤5	First line 49.1%, second or later 50.9%	ICI maintenance in 63 of 108; ICI halt in 45	Radiotherapy or surgery, not disaggregated	Start of post-progression treatment (nPFS); ICI start
Tsai et al. (CURB) [25]	67.5	≤5	First line or later	Same systemic therapy continued	SBRT	Randomization; randomization

Abbreviations: ICI, immune checkpoint inhibitor; IMRT, intensity-modulated radiotherapy; LAT, local ablative treatment; nPFS, PFS measured from the start of post-progression treatment; NR, not reported; OS, overall survival; PFS, progression-free survival; PFS2, second progression-free survival; SBRT, stereotactic body radiotherapy; SRS, stereotactic radiosurgery; WBRT, whole-brain radiotherapy.

^* For Xuzhang et al., the median age is for the overall cohort.

^‡ PFS reported only for oligoprogression versus polyprogression, not by treatment received; Friedes et al. contributed to the OS analysis only.

Tsur et al. ⁴⁴ was retained as a narrative single-arm contribution; its survival data apply to a mixed cohort and were not pooled. Zhou et al. ⁴⁷ (OCEANUS) is discussed as contextual registry evidence and is not counted among the 22, since its dataset cannot separate oligoprogression from polyprogression, and both arms received radiotherapy. The remaining 12 studies were narrative single-arm or pattern-of-failure cohorts, including Ying et al. ⁴⁶ (n = 15; median post-oligoprogression PFS 8 months and OS 12 months).

Four studies contributed to the primary PFS analysis (285 patients, 106 treated with LAT). LAT with continued immunotherapy was associated with an approximately halved hazard of progression or death, pooled HR 0.45 (95% CI 0.33–0.62; 95% prediction interval 0.28–0.74; I² = 0%; Figure 2). Four studies contributed to the primary OS analysis (308 patients, 129 treated with LAT), pooled HR 0.41 (95% CI 0.27–0.63; 95% prediction interval 0.24–0.69; I² = 0%; Figure 3). Neither analysis showed detectable statistical heterogeneity, and both prediction intervals lay below 1.0. Excluding the PFS2 estimate of Brown et al. gave a concordant PFS estimate (HR 0.43, 95% CI 0.25–0.75; prediction interval 0.19–0.95). Restricting the PFS pool to the two studies with time zero at or after oligoprogression (Xuzhang et al. and CURB) left the point estimate unchanged (HR 0.45; 95% CI 0.17–1.19, wide because only two studies contributed). The Knapp‒Hartung small-sample correction widened the intervals relative to the standard random-effects estimator, so these are the conservative figures.

Figure 2.

Primary forest plot for progression-free survival. Random-effects meta-analysis (REML, Knapp‒Hartung adjustment) of the primary PFS analysis. The contributing studies were Brown et al. ²⁹ *, Tsai et al. ²⁵ (CURB, prespecified NSCLC subgroup), Jiang et al. ³⁰ and Xuzhang et al. ³ The overall estimate represents a pooled HR with 95% CI and 95% prediction interval. *Brown et al. reported PFS2, measured from the start of first-line immunotherapy to the second progression or death; time zero for each study is given in Table 2.

Figure 3.

Primary forest plot for overall survival. Random-effects meta-analysis (REML, Knapp‒Hartung adjustment) of the primary OS analysis. The contributing studies were Tsai et al. ²⁵ (CURB, prespecified NSCLC subgroup), Friedes et al. ³² Brown et al. ²⁹ *, and Xuzhang et al. ³ The overall estimate represents a pooled HR with 95% CI and 95% prediction interval. In Friedes et al., the contributing OS estimate was multivariable-adjusted. *Brown et al. measured OS from the start of first-line immunotherapy.

Individual estimates were directionally consistent, spanning PFS HRs of 0.25–0.51 and OS HRs of 0.35–0.72; per-study values are given in Table 1 and Figures 2 and 3.

An expanded exploratory sensitivity analysis adding the two broader comparative studies (Belaidi et al. ³³ † and NIVORAD ³⁴ †) gave PFS HR 0.49 (95% CI 0.39–0.63; I² = 0%; k = 6) and OS HR 0.50 (95% CI 0.32–0.80; I² = 30%; k = 6). Belaidi et al. is a timing comparison in which both arms received radiotherapy, so this analysis probes only the directional robustness of the estimate and does not strengthen the primary LAT-versus-control contrast. A separate exploratory Xu et al. ³⁵ composite contrast, based on PFS2, gave PFS HR 0.52 (95% CI 0.37–0.71; I² = 0%; k = 5) and OS HR 0.42 (95% CI 0.32–0.56; I² = 0%; k = 5). These analyses were stress tests. In leave-one-out analysis, omitting any single study gave pooled HRs of 0.43–0.47 for PFS and 0.38– 0.45 for OS, and every confidence interval excluded 1.0. Restricting the OS pool to the three studies permitting radiotherapy alone (Brown et al., Friedes et al., and CURB) preserved the estimate (HR 0.42, 95% CI 0.19–0.93); the corresponding PFS restriction retained two studies (Brown et al. and CURB) and was imprecise (HR 0.46, 95% CI 0.12–1.84). Omitting Brown et al., whose OS was measured from the start of first-line immunotherapy, gave an OS HR of 0.42 (95% CI 0.18–0.94). To gauge robustness to unmeasured confounding and residual immortal-time bias, we computed E-values: an unmeasured confounder associated with both LAT and outcome by a risk ratio of approximately 2.9 for PFS and 3.1 for OS would be needed to explain away the pooled point estimates, and of approximately 2.1 to move either confidence-interval bound to the null (VanderWeele and Ding approximation for a common outcome ⁴⁸ ; values for pooled estimates are approximate). The full sensitivity and leave-one-out results are presented in Supplementary Appendices 4 and 5.

CURB ²⁵ randomized 143 patients with oligoprogressive breast cancer or NSCLC; the NSCLC subgroup drove the trial signal while the breast cohort showed none. STOP ²⁶ randomized 90 patients across histologies (PFS HR 0.81, 95% CI 0.48–1.37); crossover and crossing Kaplan‒Meier curves at approximately six months make it closer to a comparison of early versus delayed LAT, with an approximate OS HR of 0.88.

The non-pooled narrative and contextual studies described the oligoprogression phenotype and real-world continuation practice. Rheinheimer et al. ⁴³ characterized patterns in 115 patients, most often nodal and adrenal. Deng et al. ² documented a substantial oligoprogressive fraction after first-line immunotherapy. Huet et al. ³⁸ reported 61 patients treated during first-line immunotherapy, most continuing the same regimen. Mahmood et al. ⁴¹ contributed the largest radiotherapy cohort (n = 103; median PFS 6.9 and OS 23.5 months), and Wang et al. ⁴⁵ a median PFS of about 7–9 months. The reported median survival values for the included studies are tabulated in Supplementary Appendix 9 (Table 1).

Toxicity reporting was inconsistent and not amenable to formal pooling. Grade 3 or higher events attributable to LAT were reported in fewer than 5% of patients where recorded, with no attributed treatment-related deaths and no clear excess of immune-related adverse events.

Risk-of-bias summaries are provided in Supplementary Appendix 6 (Figure 1) for the randomized studies and Supplementary Appendix 7 (Figure 2) for the observational studies. CURB ²⁵ was assessed with RoB 2 and judged to have some concerns in the domain of deviations from intended interventions (open-label design, crossover). STOP ²⁶ had a high risk in the same domain. Observational studies were assessed with the Newcastle‒Ottawa Scale; most scored 5–7 out of 9 (moderate risk), driven primarily by selection bias and limited confounder control. No study in the primary pool was judged to be at low risk of bias across all domains.

GRADE certainty was low for both PFS and OS. Because the evidence is predominantly observational, the starting certainty was low. We downgraded once for risk of bias, chiefly retrospective design and confounding by indication. We did not downgrade for imprecision, because the confidence intervals are narrow and the prediction intervals exclude 1.0, nor for inconsistency, because I² was 0%. Against that single downgrade, we rated up once for a large and consistent effect, with pooled HRs of 0.45 for PFS and 0.41 for OS and E-values of approximately 2.9 and 3.1, a magnitude that GRADE recognizes for upgrading observational evidence. Control-arm heterogeneity and mixed modality affect applicability, as discussed below, but the disease-state-aligned primary analysis and the concordant sensitivity analyses argue against a further formal downgrade. Funnel plots were not constructed (k < 10). The net rating is low, and the full GRADE profile is in Supplementary Appendix 8.

The clinical characteristics of the five primary-pool studies are summarized in Table 2.

Discussion

The clinical decision addressed here is specific. A patient receiving an anti-PD-1 or anti-PD-L1 ICI develops progression at one or a few lesions while the remaining disease stays controlled, and the treating team must choose between continuing the current immunotherapy line with LAT to the progressing sites and switching systemic therapy. Previous syntheses have approached this question from wider angles, most often by combining oligoprogression with synchronous oligometastatic disease and with oligoresidual disease after maximal response. These states differ in biology and in the relevant comparator, and we analyzed oligoprogression separately. In this restricted population, an association that was not apparent in the broader syntheses became detectable.

The distinction from earlier syntheses is one of eligibility rather than ambition. Rim and colleagues pooled 20 studies of LAT in oligometastatic NSCLC and reported a clear benefit overall, but their oligoprogression and oligorecurrence subgroup, which included mixed systemic backbones, was not significant for either survival endpoint (OS odds ratio 1.73, p = 0.37; PFS odds ratio 1.78, p = 0.07). ¹² Kim and colleagues reviewed prospective local-therapy trials across four tumor types without pooling survival. ²² Tan and Doyle pooled SBRT across primary tumor types, with NSCLC on active immunotherapy contributing only a small minority of cohorts. ^{13 , 14} Han and colleagues pooled local consolidative therapy in driver-negative oligometastatic NSCLC, combining synchronous, oligoresidual, and metachronous disease. ¹⁵ None restricted eligibility to oligoprogression arising during active anti-PD-1 or anti-PD-L1 therapy with continuation of the same line, and to our knowledge, no previous synthesis has reported pooled survival estimates for that population. The association observed here, absent from the mixed-backbone subgroup of Rim and colleagues, supports analyzing this disease state separately.

Oligoprogression during immunotherapy is thought to reflect focal immune escape, which provides the rationale for LAT. Most of the disease remains under immune control while a minority of clones acquire local resistance, a pattern distinct from systemic treatment failure. Schoenfeld and colleagues ⁴ showed that patients who progress at a limited number of sites have a higher tumor mutational burden and deeper, longer responses than patients with widespread progression, consistent with a largely intact systemic antitumor response. Ablation of the progressing sites removes the resistant clone and may release antigen to a primed immune system through immunogenic cell death and STING-dependent interferon signaling, the mechanism implicated in out-of-field responses when radiotherapy is combined with ICI therapy. ^{8 , 10 , 11} The intended clinical consequence is preservation of an effective immunotherapy line. The largest real-world dataset to date is consistent with this model: after progression and radiotherapy, continuing the same ICI was associated with longer survival than switching, ⁴⁷ and small rechallenge series suggest that LAT may restore sensitivity to subsequent immunotherapy. ³⁷

The main strength of this finding is its consistency across studies rather than the weight of any single study. In four studies per endpoint, LAT with continued immunotherapy was associated with a lower hazard of progression or death (pooled HR 0.45) and of death (pooled HR 0.41). No statistical heterogeneity was detected, and both prediction intervals excluded 1.0, so the effect in a similar population is likely to lie in the same direction. Sensitivity analyses did not materially alter these estimates: excluding the PFS2 estimate, restricting the PFS pool to studies with time zero at or after oligoprogression, restricting the OS pool to radiotherapy-only studies, and adding the broader comparators each left the point estimate between 0.42 and 0.50. Earlier syntheses, pooling across primary tumor types and systemic backbones, were not designed to provide this estimate. ^13-15 The randomized evidence is compatible with this direction. In CURB, the prespecified NSCLC subgroup showed a PFS benefit while the breast cancer cohort did not, ²⁵ and the negative STOP trial is better interpreted as a comparison of early versus late radiotherapy than of treatment versus none, because most control patients subsequently received radiotherapy. ²⁶ Statistical heterogeneity was absent, but clinical heterogeneity was not. Oligoprogression thresholds ranged from three to five progressive lesions, and one study enrolled atypical mixed responses without a numerical cap. The immunotherapy line varied from exclusively first-line to mixed first and later lines. LAT ranged from SBRT alone to mixed ablative approaches, continuation of the index immunotherapy was universal in some cohorts and partial in others, and time zero ranged from the start of ICI therapy to randomization at oligoprogression. An I² of zero indicates that these differences did not produce divergent effect estimates, but with four studies per endpoint, the power to detect heterogeneity is limited, and consistency at this scale should not be mistaken for equivalence of populations. The pooled estimate is best read as an average across a clinically related but not identical group.

Where reported, grade 3 or higher events attributable to LAT occurred in fewer than 5% of patients, with no clear excess of immune-related toxicity, although toxicity reporting was inconsistent across studies.

Patient selection is both the opportunity and the principal caveat. The included studies suggest that any benefit is most likely when progression is genuinely focal and indolent. Brown and colleagues found that brain metastases predicted oligoprogression while liver metastases predicted systemic failure. ²⁹ Xuzhang and colleagues found that LAT was associated with longer post-progression PFS and OS in repeat oligoprogression, but not in induced oligoprogression or in polyprogression. ³ Friedes and colleagues found that median survival declined stepwise as the number of progressing sites increased. ³² Tumor boards already weigh these variables, so they can inform selection in current practice, although none has been validated as a predictor of benefit from LAT. The interval before a change of systemic therapy can be long: in one cohort, the median interval to the next systemic therapy exceeded 3 y in patients with one or two progressing sites. ³⁸ These favorable features are also prognostic; however, so part of the observed association probably reflects selection rather than a treatment effect. This does not argue against the strategy; it is the reason that randomized trials have been designed for this population.

The role of surgery as a local ablative modality warrants separate consideration. In this literature, surgery appears only in mixed-modality cohorts, ^{3 , 30 , 38} and no included study reports a resection-specific comparator from which an HR can be derived. When the OS pool was restricted to studies permitting radiotherapy alone, the estimate was essentially unchanged (HR 0.42), which suggests that the pooled association is not an artifact of the surgical cases; the corresponding PFS restriction retained two studies and was imprecise. Resection and ablative radiotherapy act on the same lesion differently. Surgery removes the resistant clone completely and yields tissue to investigate the mechanism of escape, an advantage that no other modality offers, but it does not release antigen into a primed immune compartment or induce the type I interferon response that underlies the systemic effects attributed to ablative radiotherapy. ^{8 , 10} Selection is also stricter because only the fittest patients with a single accessible site undergo resection, which inflates any apparent benefit. Surgery is therefore a reasonable choice when one site is cleanly resectable or when tissue is needed to guide the next line of therapy, but the continuation-enabling, immune-priming rationale that motivates this review is better supported for radiotherapy than for surgical resection. That rationale rests on immunogenic cell death and cGAS-STING signaling, which have been characterized for ablative radiotherapy but not for resection. ⁴⁹ These are not interchangeable interventions, and the pooled estimate should not be read as modality-agnostic.

The comparator arms were not uniform, and this bounds what the pooled estimate can be said to measure. Only CURB compared LAT against a control that continued the same ICI; elsewhere, control patients variously continued, switched, or escalated systemic therapy. Two observations constrain the resulting uncertainty. A single study with a continuation-only control produced an estimate concordant with the pool, and leave-one-out analysis showed that no individual study drives the result. What the pooled estimate measures is therefore a strategy, local ablation delivered so that an effective immunotherapy line can be preserved, rather than the isolated effect of LAT. That is the decision clinicians face in practice, but any benefit should not be attributed to ablation alone, and separating the two components would require a trial in which both arms continued identical systemic therapy.

Several limitations constrain certainty. The pooled estimates rest on four PFS and four OS studies, mostly retrospective, and the one randomized contribution enters as a prespecified subgroup rather than a dedicated NSCLC trial, and is powered accordingly. ²⁵ We therefore treated it as one of four concordant estimates rather than as the anchor of the analysis; leave-one-out analysis gave pooled HRs of 0.47 for PFS and 0.38 for OS without it. The endpoint definitions and time zero varied. In two of the four PFS studies and three of the four OS studies, time zero preceded the oligoprogression event, so the interval before LAT contributes to both arms and would favor LAT if the patients selected for it had longer initial responses. Confounding by indication and immortal-time bias can also inflate an observational estimate. The mechanistic rationale is best characterized for ablative radiotherapy and should not be assumed for surgery or percutaneous ablation. Dosimetric, biomarker, and patient-reported data were too sparse to model, and publication bias could not be formally assessed, because funnel-plot methods are uninformative below ten studies. Small-study effects, therefore, cannot be excluded, and the direction of any such bias is predictable: retrospective series reporting no benefit from LAT are less likely to be written up and accepted than those reporting a positive association, so unpublished negative experience, if it exists, would shift the pooled estimate toward the null by an unknown amount. These limitations are the reason we rate the certainty of the evidence as low. Within these limits, the clinical implication is narrow. On the basis of predominantly observational, low-certainty evidence, LAT with continuation of the same line is a reasonable option to consider for a fit patient with limited-site oligoprogression whose remaining disease is controlled with immunotherapy, pending the randomized trials described below.

Three randomized trials will determine whether this association is confirmed in prospective testing. CCTG BR.38 (CURB2) is the most directly aligned: a phase III trial randomizing 320 patients with stage IV NSCLC who develop oligoprogression on first-line immunotherapy, with or without chemotherapy, either to switch to second-line systemic therapy or to receive SBRT to all oligoprogressive sites followed by continuation of the same first-line regimen. It carries dual primary endpoints of PFS and OS, stratifies by lesion number and chemotherapy use, and collects patient-reported outcomes, quality of life, and correlative specimens. ⁵⁰ SUPPRESS-NSCLC randomizes 68 patients with one to five progressive extracranial lesions to SBRT with continued systemic therapy versus standard management, with co-primary PFS and OS, time to next systemic treatment, and translational sampling of adaptive immunity in peripheral blood. ⁵¹ HALT addresses the same architecture in the oncogene-addicted setting, randomizing patients progressing at limited sites on a tyrosine kinase inhibitor to radiotherapy with continued targeted therapy or continued targeted therapy alone, indicating whether the strategy generalizes beyond ICI therapy. ⁵²

Several questions will remain open even after these trial reports. None randomizes between LAT modalities, so whether surgery or percutaneous ablation reproduces any benefit of SBRT will remain untested. None compares early with delayed LAT, leaving optimal timing undefined. Dose and fractionation differ across protocols without direct comparison. No trial is powered to identify the lesion-number threshold beyond which the strategy fails, nor to test biomarker-guided selection, although BR.38 stratifies by lesion number, and both trials collect correlative material. Neither addresses the optimal duration of continued immunotherapy after LAT nor the management of a second oligoprogression event. These are the questions the next generation of trials should be designed around.

Disclosure of potential conflicts of interest

IM declares no competing interests. AH declares no competing interests. RZ declares no competing interests. JZ declares no competing interests. OW declares no competing interests. NP declares no competing interests.

Funding

No funding was received for this work.

Data availability statement

Study-level extracted data, the data dictionary, analytic code, and the full search strategy are available from the corresponding author on reasonable request.

AI transparency statement

During the drafting of this manuscript, the authors used Claude to enhance text flow and Grammarly for grammatical refinement. All AI-assisted suggestions were critically reviewed and edited by the authors, who retain sole accountability for the integrity and accuracy of the final content.

Supplementary material

Supplemental data for this article can be accessed at https://doi.org/10.1080/2162402X.2026.2756061.