Advances in cancer immunotherapy have dramatically changed the treatment landscape for patients with metastatic melanoma. Durable complete responses have been recorded following high-dose bolus interleukin-2 (IL-2) [1–3], adoptive cell transfer (ACT) [4, 5] and checkpoint inhibition (CPI) using antibodies against CTLA-4, PD-1 or PD-L1 [6–10]. However, a majority of patients treated with immunotherapy are either primary non-responders or eventually develop immunorefractory progressive disease and require additional therapy.

In contrast to up-front metastasectomy, which can be curative in selected patients with oligometastatic disease [11, 12], there are few data describing the use of local therapy (henceforth defined as metastasectomy, ablation or stereotactic body radiotherapy) to manage extra-cranial oligoprogression after immunotherapy. Small series showed that surgery for highly selected patients who progressed after IL-2 or CPI could achieve long PFS [13, 14] and early reports of the Memorial Sloan Kettering Cancer Center (MSKCC) experience quote 60% survival after surgery for a solitary site of progression [15].

Retrospective pattern-of-failure data in patients with non-small cell lung cancer (NSCLC) led to the initiation of two prospective randomized trials of local therapy plus systemic therapy versus local therapy alone [16, 17]. Both trials demonstrated that improved PFS could be achieved by treating established tumors present at baseline, but in neither study was durable PFS achieved by local therapy. However, in a series of 26 patients with metastatic melanoma treated with ACT, investigators in the Surgery Branch, NCI showed that metastasectomy could achieve durable PFS [18], and furthermore showed that patterns of failure after ACT strongly associated with the duration of PFS and OS after surgery. In patients with progression in established tumors present at baseline, resection achieved median PFS of 46 months, and actuarial five-year survival was 73%. In contrast, patients who had resection of new metastases, which had developed after ACT, had median PFS of 3 months and five-year survival of 0%.

For the present study, we reviewed our entire institutional experience of 428 patients with metastatic melanoma treated with CPI. We sought to determine when local therapy was used to control oligoprogression of metastatic disease, and whether patterns of failure after CPI were associated with different outcomes after local therapy.

Four hundred twenty-eight patients with metastatic melanoma were treated with CPI from 2007 to 2018, 187 (44%) under the auspices of a clinical trial and 241 (56%) under standard of care (Fig. 1). With a median follow-up of 45 months, median overall survival for the entire population (n = 428) from the first cycle of CPI was 34 months and actuarial five-year survival was 41%. These data are similar to that reported in the CheckMate 067 trial, but in our series 87 patients (20%) had brain metastases and 35 (8%) had ocular melanomas, both of which were excluded from the Checkmate 067 study [19]. Furthermore, some patients in our experience only received treatment with anti-CTLA-4, and were never treated with anti-PD-1 or anti-PD-L1.Fig1Fig. 1

Consort diagram for patient selection. Abbreviations: CPI = checkpoint inhibition; NED = no evidence of disease; NPRD = non-progressive residual disease

Immunotherapy with CPI has revolutionized the treatment of metastatic melanoma, although most patients will eventually develop disease progression. Here, we have described a single institutional experience using local therapy for selected patients treated with CPI who developed 1–3 sites of progression outside of the CNS. These procedures were technically successful in over 90% of attempts, complications were consistent in incidence with outcomes expected from the procedures performed, and there were no 90-day mortalities.

Local therapy to achieve NPRD makes sense in the context of recent advances in systemic therapy, which are capable of mediating durable regression of widely metastatic disease [6]. Patients rendered NPRD in this study had demonstrated unequivocal evidence of anti-tumor immune responses before developing oligoprogression in 1–3 sites, which appeared to be escape lesions. Other reports have documented that favorable outcomes after metastasectomy are associated with complete resection of all disease [24]. In this report, it should be emphasized that in NPRD patients all progressive disease was eliminated by local therapy; only stable or regressing residual disease was left in situ. It is even possible that some residual disease sites were sterile and no longer had active cancer. Furthermore, NPRD patients were more likely to be continued on CPI therapy after their procedures than those who were rendered NED, which serves to illustrate that the management of these patients was unique. This approach, of using local therapy as an adjuvant to CPI, resulted in PFS that was comparable to the patients who were made NED (Fig. 3 ). Ultimately, more data are needed to determine whether local therapy to achieve NPRD is beneficial for patients. In our opinion it would be ethical to study CPI with or without resection to NPRD in a prospective randomized trial.

In a series of highly selected patients, resection of escape lesions in 20% of patients who had an objective response to IL-2 was curative [13]. In a larger series of patients with progressive disease after ACT, it was shown that long PFS could be achieved by surgery in selected patients and that patterns of failure after ACT were associated with different outcomes after surgery [18]. In this report, we show that patterns of failure after CPI also predict outcomes after local therapy. This is intriguing because, to our knowledge, data from large immunotherapy trials have not shown that patterns of failure carry prognostic weight. For example, in a prospective randomized trial of 101 patients treated with adoptive transfer, 38 patients developed new sites of metastasis while 28 progressed in established sites [4]. Patients with disease progression after ACT had high risk of mortality regardless of how their treatment failed. Furthermore, the significance of a new site of metastasis was downgraded in the immune-related response criteria (irRC), which was derived from CPI response patterns [25]. Whereas in RECIST the appearance of a new site of metastasis automatically denotes progressive disease, in the irRC a new site only contributes to it.

Observations of patients achieving a complete response to immunotherapy support the notion that the disappearance of a metastasis generally implies sterilization. In most patients, an immune-mediated complete response is durable for many years and is potentially curative [5, 6]. When patients do relapse after a complete response, they appear to most frequently do so in new sites of metastasis. For example, a study looking at recurrence patterns after IL-2 showed that relapse after a complete response occurred at a new site of metastasis in 70% of cases [26]. In our series, 7 of the 8 patients (88%) who recurred after a complete response did so in a new site. This implies that in some patients, immune-based treatments can eradicate all clinically evident macrometastases but still fail to clear occult micrometastases, which later progress as new metastases.

While it is likely that the appearance of a new metastasis is a marker of aggressive biology, that explanation does not fully account for why a new site doesn’t appear to carry prognostic weight for all-comers, nor does it explain why patients with other favorable prognostic factors frequently do poorly if they progress in a new site. We hypothesize that new metastases are not, by themselves, more dangerous than progressing established tumors. Rather, new metastases are a marker of a loss of systemic disease control and reflect immunoevasion of occult micrometastatic sites. Therefore, patients with new metastases may benefit from an additional period of observation before intervention with invasive local therapy, in order to allow their disease biology to declare itself. In contrast, progression limited to established tumors leaves open the possibility that systemic disease control was achieved by effective immune clearance of clinically occult micrometastases, even if there are local failures. These patients seem most likely to benefit from local therapy, especially if they can be rendered NED. Thus, treatment failure after immunotherapy may be heterogeneous and require different management strategies.

Local therapy is extremely important for the management of CNS metastases and is supported by level 1 data [27–29]. We did not analyze intra-cranial procedures in this report but we did include some patients who were staged M1d before CPI who later required local therapy in extra-cranial sites (Table 1). All included patients staged M1d had brain metastases that either had been previously treated or had resolved with immunotherapy. Intra-cranial metastases are managed differently and may also be biologically unique; for example we showed that fewer patients with an ongoing complete response were staged M1d (Fig. 1). We did not attempt to explore the relationship between intracranial and extracranial metastases and our findings should not be extrapolated for the management of CNS metastases.

Local therapy, especially major surgery, can carry the risk of morbidity and be extremely costly for the health care system. Patients treated with local therapy in this study were highly selected. Most didn’t develop progression for more than a year after starting CPI, and many experienced impressive shrinkage or resolution of multiple metastases. Our hope is that the selection criteria we have described herein will help clinicians identify patients in whom aggressive local therapy is warranted, while simultaneously helping to spare from the morbidity of invasive procedures those who are unlikely to benefit. However, this study is greatly limited by its retrospective nature and the potential for selection bias. A prospective observational study at a minimum would be required to determine whether patterns of failure are useful for patient selection. Thus, our results should be interpreted with caution.

Up to 18% of patients with oligoprogression after CPI may be candidates for local therapy, which can achieve durable PFS in selected patients. We have shown that long PFS is associated with progression limited to established tumors, while local therapy for new metastases is associated with early relapse and shorter survival. Local therapy to achieve NPRD may be useful to manage patients with discordant responses to CPI, although further study is needed. The curative potential of immunotherapy lies in its ability to completely eliminate both microscopic disease and macroscopic lesions. There may be a significant population of patients treated with CPI who experience microscopic clearance of systemic disease but have local immune failure in established lesions. These patients may be revealed by different patterns of failure after CPI and our experience shows that some of them can be cured by timely intervention with local therapy.