Immune checkpoint inhibitors (ICI), namely antibodies against programmed death receptor-1 (PD-1), have considerably improved the outcome of patients with advanced melanoma and are capable to induce long-lasting responses in melanoma patients.1 2 However, primary or acquired resistance against ICI is common and occurs in 50%–60% of the patients.3 Therefore, prognostic biomarkers are urgently needed that identify patients who might benefit from anti-PD-1 antibodies more than others.

Clinical experience at our institution suggests that patients displaying extensive tumor burden and fast-growing tumors tend be non-responders to ICI.4 This clinical experience is supported by data from Ribas et al indicating that high total tumor burden of more or equal than 102 mm according to Response Evaluation Criteria in Solid Tumors (RECIST) V.1.1 correlates with lower response rates in patients treated with pembrolizumab.5 Indirect markers for tumor growth or tumor cell turnover like lactate dehydrogenase (LDH) have been studied extensively in the setting of immunotherapy with anti-PD-1 antibodies and were associated with survival.4 6–9 Thus, the direct investigation of tumor growth as a prognostic marker seems obvious. As early as in the 1960s to 1990s, tumor growth rate (TGR) by means of tumor doubling time (TDT) has been studied in patients with cancer with pulmonary metastases undergoing surgical resection.10 11 Only patients with slow-growing pulmonary metastases benefited from surgery and achieved long-term overall survival (OS).10 However, only little is known about the impact of TGR in the context of systemic therapy. In 2014, a French study demonstrated the superiority of initial metastatic kinetics compared with LDH and American joint committee on cancer (AJCC) stage of disease in patients treated with chemotherapy.12 A recent study reported on fast growing metastases with an intraindividual broad range of TGR being associated with impaired survival in patients treated with BRAF inhibitors (BRAFi).13 The group around Hartung et al determined pretreatment disease kinetics by measuring every metastasis in each patient. The discovery of hyperprogressive disease in patients receiving ICI brought pretreatment TGR again into a broader focus. However, the results in respect of the prognostic impact of pretreatment TGR were conflicting.14–17 Champiat and colleagues even found an inverse correlation of TGR with objective response in a single-center study including 131 patients with 21 distinct cancer entities treated with antibodies directed against PD-1 or programmed cell death 1 ligand 1 (PD-L1).14

The aim of this study was to analyze the prognostic impact of pretreatment TGR and total tumor burden on OS in melanoma patients receiving anti-PD-1 antibodies. Moreover, we aimed at developing a feasible method of approximating pretreatment TGR that could replace the ineffective and time-consuming measurement of each metastasis. We hypothesized that high metastatic growth rate (MGR) correlates with inferior survival and lower response rates to anti-PD-1 therapy in patients with metastatic melanoma.

Indirect serum biomarkers of tumor burden and tumor cell turnover like LDH are commonly used for prediction of response and monitoring of the course of disease.5 6 8 22 However, they are prone to failure due to limited specificity and their surrogate nature.23 24 Therefore, direct approaches for measuring tumor burden and disease kinetics are needed. In the present study, pretreatment MGR was found to represent a reliable prognostic marker strongly correlating with OS and PFS of melanoma patients treated with anti-PD-1 antibodies. In multivariable analysis, pretreatment MGR was clearly superior to the established prognostic factors tumor burden, LDH, liver metastasis and CNS metastasis.

In the last decade, TGR has been of increasing interest due to limitations of the established criteria to evaluate response to anticancer therapy. Several studies could show that the variation of on-treatment compared with pretreatment TGR is superior to the determination of objective response based on RECIST and that a significant number of patients classified with PD showed decreasing TGR under therapy.15 25 26 Moreover, with the introduction of ICI, TGR was discovered as a measure to identify patients showing hyperprogressive disease.14 16 17 27 28 Interestingly, the results of our study are in sharp contrast to findings of Champiat and colleagues who found an inverse correlation between pretreatment TGR and the percentual change of the sum of RECIST target lesions under therapy with anti-PD-1 or anti-PD-L1 antibodies in a single-cohort study on a heterogeneous set of 131 patients with 21 distinct tumor entities.14 The authors drew the conclusion that unlike in targeted therapy, ICI showed improved efficacy in patients exhibiting faster TGR. Importantly, the authors missed to notice the confounding effect of their heterogeneous cohort composed of fast-growing cancers like high-grade glioma (median TDT: 63.4 days)29 or melanoma (median TDT: 61 days)30 and slow-growing cancers like adenocarcinoma of the lung (median TDT: 258 days)19 or clear cell renal cell carcinoma (median TDT: 826 days).31 However, anti-PD-1 therapy induced response rates and PFS largely differ between these entities (melanoma: objective response rate (ORR) 32%, median PFS 6.9 months; renal cell carcinoma: ORR 25%, median PFS 4.6 months; esophagogastric cancer: ORR 12%, median PFS 1.4 months).22 32 33

The question whether pretreatment MGR is prognostic in general or specifically predictive for outcome in anti-PD-1-treated patients cannot be answered by our study. Although some studies implicated a strong general prognostic impact on survival of cancer patients,10–12 the extent of MGR’s discriminatory power was unexpectedly impressive in our data. Moreover, besides OS, high MGR was also clearly associated with unfavorable PFS and failure to achieve disease control or an objective response. In multivariable analysis of all three independent cohorts, MGR was most clearly associated with OS compared with the diameter of the largest target lesion, sum of RECIST target lesions, LDH, presence of liver metastasis, and presence of brain metastasis. Superiority of MGR over LDH was confirmed by the combined Kaplan-Meier analysis of both biomarkers that demonstrated the pronounced prognostic impact of MGR compared with LDH.

In comparison to the studies published so far, the main strength of our study is the inclusion of three independent cohorts comprizing a multicenter external validation cohort and 337 patients in total. Therefore, we can conclude with certainty that MGR constitutes a powerful and valuable novel prognostic marker for patients treated with anti-PD-1 antibodies.

Our data suggest that anti-PD-1 antibodies are not capable to inhibit rapidly growing metastases. Recently, Huang et al delineated the ratio of T-cell invigoration to tumor burden as being closely associated with response to anti-PD-1 checkpoint blockade.34 Patients with low ratios of Ki67 positive PD-1 positive T-cells to tumor burden exhibited low ORR and impaired survival. In line are recent findings of our group that showed that the interruption of interferon-induced senescence leads to an uncontrolled growth of melanoma metastases.35 These results indicate that a disequilibrium between unleashed tumor growth and T-cells leads to fatal outcomes.

In the recent past, several efforts have been made to identify novel prognostic markers in the setting of ICI with PD-1 antibodies. Most promising, but also controversial is the utilization of PD-L1 status as a predictor of response.36 37 Although several studies have highlighted the predictive impact of PD-L1 expression in tumor tissue, several limitations deserve cautious appraisal of this molecular biomarker.2 37–40 A major disadvantage of PD-L1 status concerns the prerequisite of surgically accessible metastases and its high effort. Moreover, technical issues like PD-L1 expression heterogeneity within the microenvironment hampers the evaluation of immunohistochemistry.36

In previous studies, TGR was either computed based on all 1–5 RECIST target lesions, 1–10 iRECIST target lesions, or by measuring the total number of each measurable metastasis.12–17 25–27 The approach used by Gaudy-Marqueste et al as well as Hartung et al implicated the measurement of all metastases with a diameter of at least 10 mm in three dimensions.12 13 Especially in patients harboring hundreds of metastases, this approach is very time-consuming and limits sample size.13 Data derived from chest roentgenograms had shown little intraindividual variation in TDT of multiple metastatic lesions suggesting that measuring only one target lesion could be sufficient to reflect disease kinetics.41 42 This justifies our rationale for measuring only one target lesion which seems suitable to reflect tumor growth dynamics. Different models of tumor growth have been developed, comprizing exponential, exponential-linear, Gompertz function, or logistic growth, and there is a long-lasting discussion on what is the best to describe this dynamic process.43–45 To take the classical exponential model into account, our study includes the TDT used by many authors.19 43 46 47 In addition, we introduce a novel approach that, like the logistic or Gompertz models, considers the declining growth rate of larger metastases when angiogenesis, nutrient and oxygen depletion, as well as tumor cell necrosis increasingly play a role.44 45 48 While TDT is constant at 30 days per doubling for two lesions that increase from 10 to 20 mm (diameter) within 90 days, and from 40 to 80 mm within 90 days, respectively, MGR reflects this more dramatic absolute growth of the larger lesion with calculated growth rates of 3.4 mm/month and 13.5 mm/month, respectively. Murphy et al presented in detail with impressive examples that all models of tumor growth encounter their limits under certain conditions.48 Despite these considerations, the herewith introduced MGR, although not claiming to represent a theoretical mathematical model of tumor growth, was capable to discriminate the patients more efficiently than TDT.

We are aware of several limitations of our study. The retrospective design makes it susceptible for a patient selection bias. However, we included all consecutive patients receiving nivolumab or pembrolizumab who had at least one prebaseline staging at our center. Moreover, a third independent cohort of patients enrolled at 12 participating centers and assessed by 12 independent and experienced dermato-oncologists and radiologists confirmed the results observed in the two monocentric cohorts. Thereby, site-specific treatment procedures and patient selection bias could be minimized. To the best of our knowledge, this is by far the largest set of patients analyzed concerning pretreatment tumor growth kinetics.

Another putative limitation constitutes the impact of prior treatment regimens on MGR. In the discovery cohort 36 of the 53 patients had been previously treated with ipilimumab. It seems suggestive that patients showing stable or even slightly decreasing target lesions on treatment with ipilimumab might benefit from an anti-PD-1 antibody or tend to benefit from any ICI. However, MGR remained a strong prognostic factor in subgroup analysis of patients with any prior systemic therapy, prior BRAFi, prior ipilimumab, and treatment-naïve patients.

The probably most important limitation of the concept of pretreatment tumor growth dynamics as baseline prognostic factor is the prerequisite of at least one prebaseline staging. In a personal statement by Jean Jacques Grob, Georgina Long, Dirk Schadendorf and Keith Flaherty published in 2015, the expert authors had discussed the option of postponing the start of therapy to achieve this premise.11 12 31 However, ethical reservations should be discussed carefully when deciding about treatment delays owed to diagnostic procedures. Given a median MGR of approximately 3 mm/month and a median doubling time of 41 days, consecutive staging examinations should be separated by at least 1 month to account for measuring inaccuracy. This recommendation for a minimum and an ideal time interval between the two successive measures of target lesions is in accordance with previous suggestions.12 49 In our validation cohort, the median interval between the pretreatment staging examinations was 86 days (IQR: 61–116 days) indicating a low risk for measuring inaccuracy.

Apart from intentional treatment delays, there can be several other reasons that normally lead to consecutive staging examinations prior to anti-PD-1 therapy. In our study, the most common cause for the existence of two staging examinations before initiation of ICI was the requirement to complete staging information, for example, in cases where only a CT of the abdomen was available or in cases where a PET scan was needed to improve diagnostic sensitivity. The second most common cause were regular on-treatment staging examinations during prior therapies like ipilimumab or BRAFi. Delays between initial staging and start of therapy also led to the necessity to perform an additional staging scan immediately before anti-PD-1 therapy. However, this reason was less common.

In conclusion, high pretreatment TGR is associated with unfavorable outcome and non-response in advanced melanoma patients treated with anti-PD-1 antibodies. With MGR, we propose a novel measure of tumor growth kinetics that independently predicts survival, superiorly compared with total tumor burden, LDH, site of metastasis and other known factors. As a time-efficient method, determination of MGR can be easily implemented in routine clinical settings and should be explicitly considered prior to therapeutic decisions. Investigation of the predictive impact of MGR in distinct therapeutic regimens like PD-1 blockade, combined ICI with anti-CTLA-4 and anti-PD-1, and small molecule inhibitors targeting the MAPK pathway, is warranted.