Early radiological response pattern predicts long-term response in patients with non-small cell lung cancer treated with immune-checkpoint inhibitors Authors: Markus Kleinberger, Maximilian J. Mair, Julia M. Berger, Lynn Gottmann, Vincent Sunder-Plassmann, Martin Korpan, Isabella Solano Henao, Josef Fuerst, Luzia Berchtold, Ahmed Mohamed, Angelika M. Starzer, Matthias Preusser, Anna S. Berghoff, Julia Furtner Published: 2026 ABSTRACT BackgroundImmune checkpoint inhibitors (ICI) benefit only a subset of patients with non-small cell lung cancer (NSCLC). We evaluated whether response patterns differ by metastatic site and whether early radiological change predicts long-term outcomes.MethodsPatients with advanced NSCLC treated with ICI were prospectively enrolled between 03/2019 - 03/2023. Imaging was assessed blindly by a board-certified radiologist per iRECIST. Early response was defined as change in sum of diameters (SoD) from baseline to first restaging (8–12 weeks). Baseline and early inflammatory laboratory markers were collected. Endpoints were associations of early radiological response with durable clinical benefit (DCB; disease control ≥ 6 months), best overall response (BOR), overall survival (OS), progression-free survival (PFS) and organ-specific tumor burden.ResultsAmong 122 patients (46% female; median age 65) 32.8% were BOR responders and 47.5% achieved DCB. Median OS was 64.6 weeks and PFS 31.6 weeks. Early SoD reduction predicted DCB (p > 0.05, AUC 0.764) and BOR (p < 0.01, AUC 0.701). Organ-specific effects were seen particularly in lung and lymph nodes (all p ≤ 0.021). First-line ICI showed greater early shrinkage (p < 0.001) and higher response (p = 0.024), later-line therapy reduced response odds (BOR: OR 0.34; DCB: OR 0.38). Although selected baseline inflammatory markers were associated with survival in univariable analyses, neither baseline nor dynamic laboratory parameters retained independent significance in multivariable models after correction for multiple testing.ConclusionsSpecific organ sites show differential sensitivity to ICI, with lung and nodal metastases demonstrating the most pronounced early regression. Early radiological dynamics robustly predicted DCB, BOR, PFS, and OS and outperformed laboratory biomarkers, supporting their use as a practical, clinically relevant tool to identify patients most likely to benefit from ICI.Clinical trial numberNot applicable.Supplementary informationThe online version contains supplementary material available at 10.1186/s40644-026-01048-2. Introduction Immune checkpoint inhibitors (ICIs) are an established treatment for non-small cell lung cancer (NSCLC). However, response rates remain limited, and many patients show stable disease or only minimal tumor shrinkage at first restaging, raising uncertainty about continued benefit [1]. Early radiological response patterns have shown predictive value in melanoma. A retrospective study of 56 stage IV melanoma patients treated with ICIs and exhibiting mixed responses per RECIST 1.1 found that progression during further treatment was more frequent when ≥ 3 metastatic organ sites were involved [2]. Another analysis of three pooled prospective melanoma trials demonstrated that changes in tumor burden at 12 and 18 weeks were associated with subsequent response [2, 3]. In NSCLC, smaller studies in chemotherapy-treated patients, not immunotherapy, have suggested that early 18F-FDG-PET imaging may predict long-term response [4, 5], but comparable data for ICI-treated NSCLC are lacking. Mixed responses (MR), where some lesions regress while others progress, represent an additional challenge under ICI therapy [6]. This may reflect heterogeneity of the tumor microenvironment (TME), although mechanisms remain incompletely understood [7]. Furthermore, atypical patterns such as pseudoprogression, initial radiologic worsening followed by tumor control, necessitate adapted criteria like Irecist [6, 8]. This study therefore evaluates radiological response patterns in locally advanced and metastatic NSCLC patients treated with ICIs, with particular focus on mixed and early response dynamics and their association with durable clinical benefit (DCB), progression-free survival (PFS) and overall survival (OS). Material & methods Study design and ethical approval This retrospective imaging analysis was performed in a prospectively enrolled cohort within the Translational Research Unit (TRU) Biobanking Program for Personalized Immunotherapy at the Division of Oncology, Department of Medicine I, Medical University of Vienna, between March 2019 and March 2023. The TRU program (EK Nr: 1164/2019) is an observational study collecting biological samples, clinical data and imaging. The present analysis was approved by the Ethics Committee of the Medical University of Vienna (EK Nr: 2065/2022) and conducted in accordance with the Declaration of Helsinki (2008). All participants provided written informed consent. As an exploratory diagnostic study, no randomization or pre-specified power calculation was performed, and all eligible patients during the study period were included. Eligibility criteria Eligible patients were ≥ 18 years of age, of male or female biological sex, had histologically confirmed NSCLC of locally advanced or metastatic stage, and were treated with an immune-checkpoint inhibitor-based regimen in a palliative setting. Cohort included patients treated with ICI monotherapy and ICI combined with chemotherapy to preserve statistical power and reflect the real-world population of patients receiving ICI-based therapy. Patients were required to have received at least two treatment cycles prior to first radiological restaging. Patients were excluded if they were < 18 years of age, had missing baseline or follow-up imaging, or received fewer than two ICI cycles prior to first restaging. Data collection Patient-specific clinical data were obtained via electronic chart review. Routine laboratory parameters, including white blood cell count, differential counts (neutrophils, lymphocytes), CRP, LDH, and albumin, were collected from clinical records at baseline (prior to treatment initiation) and at first radiological follow-up. Radiological response evaluation Radiological images were retrieved from the institutional PACS and evaluated centrally and blinded to outcomes by a board-certified radiologist. Restaging of patients treated with ICI monotherapy or combination regimens was performed according to iRECIST [9]. All imaging from baseline (treatment initiation) to death or data cut-off (08/2023) was reviewed. Target and non-target lesions were defined per iRECIST, and tumor burden was quantified using the sum of diameters (SoD), including organ-specific analyses. Early radiological response was defined as the relative change in SoD between baseline and first follow-up. Progression was categorized as new lesions or growth of existing target and/or non-target lesions. Best overall response (BOR) and durable clinical benefit (DCB) were assessed at patient level. Responders (BOR) included iCR and iPR; non-responders included iSD and iCPD [10] DCB was defined as disease control ≥ 6 months [11]. Pseudoprogression was defined as initial iUPD not confirmed on subsequent imaging. For exploratory analyses, early response was further stratified by relative SoD change into iCR (100%), iPR (≤−30%), iSD-low (−29% to 0%), iSD-high (+1% to +19%), and i(U)PD (≥+20%) to better capture heterogeneity within iSD. To further assess inter-lesional heterogeneity, we performed an exploratory lesion-level analysis. For each target lesion, relative size change between baseline and first follow-up was calculated. Mixed response (MR) was defined as the concurrent presence of at least one shrinking and one growing lesion within a patient. In addition, heterogeneity was quantified using the variance of relative changes across target lesions. Inflammatory parameter To introduce an additional translational perspective, we evaluated previously published prognostic and predictive laboratory biomarkers at baseline, first follow-up and their relative changes from baseline to first restaging. Here, we analysed parameters included absolute white blood cell count, absolute neutrophils, lymphocytes, monocytes and platelet count (all [G/L]), as well as CRP [mg/dL], albumin [g/L], LDH [U/L]. Additionally, derived inflammatory ratios were analysed: neutrophil-to-lymphocyte ratio (NLR), platelet-to-lymphocyte ratio (PLR), lymphocyte-to-leukocyte ratio (LLR) and CRP-to-albumin ratio (CAR). Named inflammatory markers were selected because they are routinely available in clinical practice and were consistently recorded across the cohort. In addition, parameters such as NLR, PLR, CRP,and LDH have been previously proposed as prognostic or predictive biomarkers in patients receiving immunotherapy. The analysis was therefore restricted to this predefined group of markers to ensure data completeness and reduce heterogeneity in this exploratory setting. Statistical analysis Categorical variables such as demographics and clinical characteristics were descriptively analysed and shown with absolute count and relative frequency. Metric data was presented as mean ± standard deviation or median and 25–75% IQR. Testing for normal distribution was performed using Kolmogorov-Smirnov test. Differences between cancer entities regarding response depth were calculated using the Mann-Whitney U test and the Kruskal-Walli’s test. For correlations between dichotomous variables the Chi-Square test was used. Relevant radiological parameters (e.g., number of lesions or location) were included in a multivariable analysis using the Cox proportional hazard model. Early radiological response (defined as change in SoD between baseline and first follow-up) was evaluated as a predictor of BOR, DCB and survival outcomes (PFS, OS). Analyses were performed in the overall cohort with first follow-up ≤ 14 weeks after treatment initiation and separately in predefined subgroups with follow-up at approximately 8 weeks (±2 weeks) and 12 weeks (±2 weeks). To account for variability in follow-up timing, early response was additionally standardized by calculating relative change per week, assuming a constant rate of change between visits. Results were considered significant at a two-sided p-value < 0.05. As the exploratory studies within this analysis were aimed at the generation of hypotheses, no adjustment for multiple testing was applied [12]. The explorative nature of the p-values was explicitly noted in the methods. Statistical analysis and data visualization were performed using R 4.1.2 (The R Project for Statistical Computing, Vienna, Austria) with RStudio 2021.09.1 (RStudio Inc., Boston, MA, USA). Results Patients characteristics In total, we analysed 122 patients with advanced NSCLC (median age 65 years, range 25–85 years; 56/122 female, 46%). Histologically 90/122 (74%) had adenocarcinoma, 23/122 (19%) squamous cell carcinoma and 9 (7.4%) had other histological types (n < 5). PD-L1 tumor proportion score of ≥50% was present in 37/122 (30.3%) and ≥1% in 49/122 (40.2%) samples. Median time between treatment start and first follow-up was 8.9 weeks (range 2.9–37.1) with 3 administered ICI cycles (range 2–5). Detailed information regarding the patient characteristics is shown in Table 1.Table 1Patients characteristics. ECOG = eastern cooperative Oncology score, TPS = tumor proportion scorePatients’ characteristicsAll NSCLCn=122 (100%)Sexmale66 (54%)female56 (46%)Age at primary diagnosismedian65range25-85Smoking historySmoker108 (88.6%)Never-Smoker12 (9.8%)Not available2 (1.6%)Packyearsmedian (range)40 (10-160)HistopathologyAdenocarcinoma90 (74%)Squamous Cell Carcinoma23 (19%)Other9 (7.4%)Actionable genetic alterationEGFR mutation 14 (11.5%)ALK mutation6 (4.9%)BRAF mutation4 (3.3%)ROS1 mutation1 (0.8%)Stage at baselinemetastasized116 (95.1%)non-operable, Stage III6 (4.9%)Tumor Proportion ScoreTPS negative36 (29.5%)TPS ≥ 149 (40.2%)TPS ≥ 5037 (30.3%)median (range)5 (0-100)ECOG at treatment initiation074 (60.7%)140 (32.8%)≥ 28 (6.6%) Table 1Patients characteristics. ECOG = eastern cooperative Oncology score, TPS = tumor proportion scorePatients’ characteristicsAll NSCLCn=122 (100%)Sexmale66 (54%)female56 (46%)Age at primary diagnosismedian65range25-85Smoking historySmoker108 (88.6%)Never-Smoker12 (9.8%)Not available2 (1.6%)Packyearsmedian (range)40 (10-160)HistopathologyAdenocarcinoma90 (74%)Squamous Cell Carcinoma23 (19%)Other9 (7.4%)Actionable genetic alterationEGFR mutation 14 (11.5%)ALK mutation6 (4.9%)BRAF mutation4 (3.3%)ROS1 mutation1 (0.8%)Stage at baselinemetastasized116 (95.1%)non-operable, Stage III6 (4.9%)Tumor Proportion ScoreTPS negative36 (29.5%)TPS ≥ 149 (40.2%)TPS ≥ 5037 (30.3%)median (range)5 (0-100)ECOG at treatment initiation074 (60.7%)140 (32.8%)≥ 28 (6.6%) Patients characteristics. ECOG = eastern cooperative Oncology score, TPS = tumor proportion score Treatment response to immune checkpoint inhibitor containing therapy Within the study cohort most of the patients had stable disease (iSD, 53/122, 43.4%), 39/122 had partial remission (iPR, 31.9%), 29/122 confirmed progressive disease (iCPD, 23.8%) and one had complete remission (iCR, 0.8%) as BOR. Of all included patients 40/122 (32.8%) were classified as responder according to BOR and 82/122 (67.2%) as non-responder. DCB was achieved by 58/122 patients (47.5%). In our cohort 11/122 patients (9.0%) were identified as having experienced pseudo-progression. In total, 30/122 patients (24.6%) experienced MR. No differences in response were observed between ICI monotherapy and combination with chemotherapy (see Supplementary Table 1). Detailed information is depicted in Table 2.Table 2Treatment characteristics of included patients. Disease progression depicting reasons for progression whereas concomitant development of new target lesions (TL) and non-target lesion (NTL) possible. Progression of previous lesions indicating PD due to increase of sum of all diameters > 30% without novel lesions detected. For patients with on-going treatment, no restaging due to early deceased or discontinuation related to adverse event or patients’ choice not available depictedTreatment characteristicsAll NSCLCn=122 (100%)TreatmentICI in combination62 (50.8%)ICI monotherapy60 (49.2%)Pembrolizumab (PD-1)94 (77.0%)Atezolizumab (PD-L1)20 (16.5%)Durvalumab (PD-L1)7 (5.7%)Ipilimumab + Nivolumab (CTLA4 + PD-1)1 (0.8%)Platin + Pemetrexed51 (41.8%)Platin + Taxane8 (6.6%)Platin + Etoposide1 (0.8%)Platin + Pemetrexed + anti-VEGF1 (0.8%)anti-VEGFR1 (0.8%)Pretreated43 (35.2%)ICI in 2nd line35 (28.7%)ICI in 3rd line5 (4.1%)ICI in 4th line3 (2.5%)Treatment DiscontinuationExtracranial PD66 (54.1%)Deceased20 (16.4%)On-going treatment20 (16.4%)Observation11 (9.0%)Intracranial PD2 (1.6%)Patients’ choice2 (1.6%)Adverse event1 (0.8%)Disease progressionDevelopment new TL26 (21.3%)Development new NTL14 (11.5%)Progress previous lesion65 (53.3%)Not available21 (17.2%)Durable clinical benefit (DCB)yes58 (47.5%)no64 (52.5%)Best overall response (BOR)Responder (iCR/iPR)40 (32.8%)Non-Responder (iSD/iCPD)82 (67.2%)iCR1 (0.8%)iPR39 (31.9%)iSD53 (43.4%)iCPD29 (23.8%)Overall survival (OS)Available84 (68.8%)Median [weeks]64.4Range [weeks]4.7-192.3Progression-free survival (PFS)Available112 (91.8%)Median [weeks]31.6Range [weeks]2.9-158.4 Table 2Treatment characteristics of included patients. Disease progression depicting reasons for progression whereas concomitant development of new target lesions (TL) and non-target lesion (NTL) possible. Progression of previous lesions indicating PD due to increase of sum of all diameters > 30% without novel lesions detected. For patients with on-going treatment, no restaging due to early deceased or discontinuation related to adverse event or patients’ choice not available depictedTreatment characteristicsAll NSCLCn=122 (100%)TreatmentICI in combination62 (50.8%)ICI monotherapy60 (49.2%)Pembrolizumab (PD-1)94 (77.0%)Atezolizumab (PD-L1)20 (16.5%)Durvalumab (PD-L1)7 (5.7%)Ipilimumab + Nivolumab (CTLA4 + PD-1)1 (0.8%)Platin + Pemetrexed51 (41.8%)Platin + Taxane8 (6.6%)Platin + Etoposide1 (0.8%)Platin + Pemetrexed + anti-VEGF1 (0.8%)anti-VEGFR1 (0.8%)Pretreated43 (35.2%)ICI in 2nd line35 (28.7%)ICI in 3rd line5 (4.1%)ICI in 4th line3 (2.5%)Treatment DiscontinuationExtracranial PD66 (54.1%)Deceased20 (16.4%)On-going treatment20 (16.4%)Observation11 (9.0%)Intracranial PD2 (1.6%)Patients’ choice2 (1.6%)Adverse event1 (0.8%)Disease progressionDevelopment new TL26 (21.3%)Development new NTL14 (11.5%)Progress previous lesion65 (53.3%)Not available21 (17.2%)Durable clinical benefit (DCB)yes58 (47.5%)no64 (52.5%)Best overall response (BOR)Responder (iCR/iPR)40 (32.8%)Non-Responder (iSD/iCPD)82 (67.2%)iCR1 (0.8%)iPR39 (31.9%)iSD53 (43.4%)iCPD29 (23.8%)Overall survival (OS)Available84 (68.8%)Median [weeks]64.4Range [weeks]4.7-192.3Progression-free survival (PFS)Available112 (91.8%)Median [weeks]31.6Range [weeks]2.9-158.4 Treatment characteristics of included patients. Disease progression depicting reasons for progression whereas concomitant development of new target lesions (TL) and non-target lesion (NTL) possible. Progression of previous lesions indicating PD due to increase of sum of all diameters > 30% without novel lesions detected. For patients with on-going treatment, no restaging due to early deceased or discontinuation related to adverse event or patients’ choice not available depicted When comparing between pretreated and treatment naive patients’ early response pattern, defined as the change in SoD from baseline to first follow-up, differed significantly by treatment line. Patients receiving first-line ICI had greater early tumor shrinkage compared to those in later lines (Wilcoxon rank-sum test p < 0.001). If using a Chi square BOR differed significantly between groups, with a higher proportion of responders observed in the first-line setting (responder in first line 32 vs 8 in later line, p = 0.024). Same applies to DCB (44 vs 14, p = 0.024). Using logistic regression analysis later-line ICI treatment was associated with significantly reduced odds of achieving BOR-Response (OR 0.34, 95%CI: 0.13–0.79, p = 0.016) and DCB (OR 0.38, 95%CI: 0.17–0.82, p = 0.016). When analysed separately for ICI monotherapy and combination treatment, we observed a significant association between tumor histology and treatment response. Specifically, patients with squamous cell carcinoma were more frequently classified as responders based on BOR, whereas patients with adenocarcinoma were more often classified as non-responders (p = 0.048). In contrast, other factors, including age, sex, and histology beyond the squamous versus adenocarcinoma distinction in this subpopulation, did not significantly impact treatment response (Supplementary Table 2). Results are graphically shown in Fig. 1 A-F.Fig. 1Response, patients characteristics and organ-specific change in tumor burden. Patients with and without durable clinical benefit (DCB) regarding (A) age, (B) tumor histology and (C) sex separated by ICI monotherapy and ICI in combination. Response (BoR) separated by (D) age, (E) histology and (F) sex. Early response dynamics of for organs systems analyzed for (G) BoR response and (H) durable clinical benefit. LN = lymph node, ICI = immune checkpoint inhibitor, SSC = squamous cell carcinoma, adeno = adeno carcinoma Response, patients characteristics and organ-specific change in tumor burden. Patients with and without durable clinical benefit (DCB) regarding (A) age, (B) tumor histology and (C) sex separated by ICI monotherapy and ICI in combination. Response (BoR) separated by (D) age, (E) histology and (F) sex. Early response dynamics of for organs systems analyzed for (G) BoR response and (H) durable clinical benefit. LN = lymph node, ICI = immune checkpoint inhibitor, SSC = squamous cell carcinoma, adeno = adeno carcinoma Difference in treatment response between organ systems To determine difference of ICI treatment response between organ lesions we analysed the corresponding TLs and NTLs separately. At baseline pulmonary TLs were detectable in 113 (92.6%) patients. Furthermore, TLs were found in lymph nodes (74/122, 60.7%), adrenal glands (37/122, 30.3%), bone (22/122, 18.0%), brain (21/122, 17.2%), liver (12/122, 9.8%) and other organ systems (14/122, 11.5%). A significant decrease from baseline to 1FU of intrapulmonary tumor burden (median SoD 44 vs 40 mm, relative decrease 9.1%, p < 0.001), lymph nodes (15 vs 12 mm, relative decrease 20%, p < 0.001) and other lesions (p = 0.042) was observed. No statistically significant decrease of tumor load was seen in the other mentioned organ systems (all p > 0.05, see Fig. 1 G and H, Supplementary Table 3). Overall, we observed a significant decrease in tumor burden from baseline to first follow-up (median SoD 88 vs 73 mm, p < 0.001, see Fig. 2 A). Organ-specific change of SoD between BL and 1FU were shown graphically in Fig. 2 B.Fig. 2Tumor size change in ICI treatment. (A) Sum of all diameters (SoD, [mm]) between baseline at treatment initiation and at first restaging (FU). Analyzed by using Wilcoxon signed rank test with continuity correction. (B) Changes of SoD within different organ systems between baseline and first follow-up. Kaplan–Meier curves for (C) overall survival (OS) and (D) progression-free survival (PFS) comparing patients with early tumor shrinkage versus no shrinkage based on absolute change in the SoD from baseline to first follow-up. Early response categories based on relative SoD change are associated with differences in (E) OS and (F) PFS: iCR (−100%), iPR (≤−30%), iSD-low (−29% to 0%), iSD-high (+1% to +19%), and iUPD (≥+20%). Direct comparison of iSD-low and iSD-high subgroups revealed no significant difference in (G) OS or (H) PFS. Adrenal = adrenal gland Tumor size change in ICI treatment. (A) Sum of all diameters (SoD, [mm]) between baseline at treatment initiation and at first restaging (FU). Analyzed by using Wilcoxon signed rank test with continuity correction. (B) Changes of SoD within different organ systems between baseline and first follow-up. Kaplan–Meier curves for (C) overall survival (OS) and (D) progression-free survival (PFS) comparing patients with early tumor shrinkage versus no shrinkage based on absolute change in the SoD from baseline to first follow-up. Early response categories based on relative SoD change are associated with differences in (E) OS and (F) PFS: iCR (−100%), iPR (≤−30%), iSD-low (−29% to 0%), iSD-high (+1% to +19%), and iUPD (≥+20%). Direct comparison of iSD-low and iSD-high subgroups revealed no significant difference in (G) OS or (H) PFS. Adrenal = adrenal gland Early radiological tumor size reduction associated with durable clinical benefit To evaluate associations between early tumor size changes and later treatment response patient who died prior to first restaging were excluded (8/122, 6.6%). By using logistic regression model, we examined the response dynamics of 114 patients for each location. Overall, only the early decrease in size of lymph node (OR 0.93, 95%CI: 0.87–0.98, p = 0.016), lung (OR 0.96, 95%CI: 0.94–0.99, p = 0.006) and adrenal gland (OR 0.92, 95%CI: 0.85–0.98, p = 0.024) was significantly associated with higher chances of DCB. The other predictors (time between BL and 1FU, BL SoD) showed no significant effect in all locations (p > 0.05, see Supplementary Table 4). Next, we focused on more homogeneous groups where patients had their follow-up at 8 weeks (±2 weeks, n = 65) or 12 weeks (±2 weeks, n = 39) post treatment initiation. Early radiological size reduction showed significant correlation in predicting DCB at 8 weeks (OR 0.98, 95%CI: 0.96–1.00, p = 0.0037) and at 12 weeks (OR 0.96, 95%CI: 0.93–0.99, p = 0.035) indicating an increase in SoD predicting lower chances of DCB. Additionally, by standardization an early response pattern changes per week logistic regression model was significant (p = 0.002, see Supplementary Table 5). Change of SoD between BL and 1FU did not correlate with time between BL and 1FU (ρ=−0.02, p = 0.834, see Supplementary Fig. 1). The receiver operating characteristic (ROC) curve for the logistic regression model predicting DCB on the early response pattern is shown in Supplementary Fig. 2A (AUC 0.764). Early radiological changes associated with best overall response In the following section we focused on prediction of response according to BOR. Early response dynamics in brain (OR 0.89, 95%CI: 0.78–0.96, p = 0.020), lymph node (OR 0.89, 95%CI: 0.82–0.95, p = 0.002), lung (OR 0.92, 95%CI: 0.87–0.96, p < 0.001) and adrenal gland (OR 0.92, 95%CI: 0.85–0.98, p = 0.021) were significantly associated with occurrence of response. Other predictors showed no significant effect in all locations (p > 0.05, see Supplementary Table 6). Separated by 8 weeks (±2 weeks, n = 65) and 12 weeks (±2 weeks, n = 39) post treatment initiation early radiological size reduction showed significant correlation in predicting BOR at 8 weeks (OR 0.91, 95%CI: 0.86–0.95, p < 0.001) and at 12 weeks (OR 0.90, 95%CI: 0.82–0.95, p = 0.005). Additionally, by standardization an early response pattern changes per week logistic regression model was significant (all p < 0.001, see Supplementary Table 7). ROC curve for the logistic regression model predicting BOR responder on the early response pattern is shown in Supplementary Fig. 2B (AUC 0.701). Early SoD reduction was significantly associated with BOR response in both treatment groups (ICI monotherapy: OR 0.93, 95%CI: 0.89–0.96, p < 0.001; ICI + chemotherapy: OR 0.84, 95%CI: 0.68–0.93, p = 0.023), with comparable effect estimates. No significant interaction between early SoD change and treatment group was observed (p = 0.85). Associations of initial changes in response pattern with progression free survival and overall survival In addition to analysing response based on BOR and DCB, we evaluated OS and PFS. At data cutoff, OS events had occurred in 84 of 122 patients (68.8%), with a median OS of 64.6 weeks (range 4.7–192.3). PFS events were observed in 112 patients (91.8%) with a median PFS of 31.6 weeks (range 2.9–158.4). In univariable Cox regression, early tumor response (measured as absolute change in sum of diameters from baseline to first follow-up) was significantly associated with both OS (HR = 1.013, 95%CI: 1.004–1.023, p = 0.005) and PFS (HR = 1.010, 95%CI: 1.004–1.017, p = 0.003). These associations were further illustrated by Kaplan–Meier curves stratified by tumor shrinkage vs. no shrinkage, showing significant differences in both OS (p = 0.007) and PFS (p = 0.013, both log rank test) (see Fig. 2C and D). To further explore the prognostic value of early response patterns, patients were grouped into iCR (−100%), iPR (≤−30%), iSD-low (−29% to 0%), iSD-high (+1% to +19%), and i(U)PD (≥ +20%). Among 114 evaluable patients, most were classified as iSD-low (n = 49), followed by iPR (n = 26), iSD-high (n = 18), and i(U)PD (n = 18), while iCR was rare (n = 2). Given that most patients fall within the broad iSD category, this finer stratification may better capture prognostic heterogeneity. In grouped Cox regression, all response groups showed improved OS compared to i(U)PD: In this grouped Cox regression analysis, early response patterns were significantly associated with both OS and PFS. For OS, compared to patients with i(U)PD, those with iSD-low (HR = 0.33, 95%CI: 0.17–0.65, p = 0.001), iSD-high (HR = 0.44, 95%CI: 0.21–0.95, p = 0.036), and iPR (HR = 0.45, 95%CI: 0.20–1.01, p = 0.052) exhibited improved survival outcomes. Similarly, for PFS, all early response groups showed significant benefit compared to i(U)PD: iCR (HR = 0.10, 95%CI: 0.01–0.79, p = 0.029), iPR (HR = 0.32, 95%CI: 0.16–0.64, p = 0.001), iSD-high (HR = 0.32, 95%CI: 0.16–0.63, p = 0.001), and iSD-low (HR = 0.25, 95%CI: 0.14–0.45, p < 0.001). These findings indicate that even modest early reductions or stabilization in tumor size are associated with favourable survival outcomes in ICI-treated NSCLC (Fig. 2 E and F). However, when comparing iSD-low to iSD-high no statistically significant differences were observed in OS (HR = 0.75, 95%CI: 0.41–1.37, p = 0.35) or PFS (HR = 0.77, 95%CI: 0.46–1.30, p = 0.33, Fig. 2 G and H). Within stratified analyses performed for ICI monotherapy and ICI combined with chemotherapy, early SoD change was associated with PFS and OS in the combination group (PFS: HR 1.005, 95%CI: 1.002–1.008, p = 0.002; OS: HR 1.004, 95%CI: 1.000–1.008, p = 0.040), whereas no statistically significant associations were observed in the monotherapy subgroup (PFS: HR 1.006, 95%CI: 0.996–1.017, p = 0.214; OS: HR 1.010, 95%CI: 0.994–1.026, p = 0.209). To formally assess potential differences between treatment modalities, interaction analyses were performed. No significant interaction between early SoD change and treatment group was observed for PFS (p = 0.90) or OS (p = 0.58). In multivariable models early response pattern remained significantly associated with DCB (OR 0.98, 95%CI 0.96–0.99, p = 0.003), whereas MR was not independently associated with DCB (OR 0.87, 95%CI 0.35–2.11, p = 0.75). Comparable findings were observed for survival outcomes, where MR was not associated with PFS (HR 1.10, 95% CI 0.71–1.70, p = 0.68) or OS (HR 0.84, 95%CI 0.52–1.38, p = 0.50), while early response pattern remained significant. Inflammatory marker and associations with treatment response Lastly, we analysed previously published inflammatory parameters at BL, 1FU and the relative change to investigate any interplay between radiological response dynamics, laboratory parameter and treatment response. Here we observed a significant decrease in white blood cell count between baseline and follow-up (8.8 vs 6.9 G/L, p < 0.001), absolute neutrophiles (5.6 vs 4.3 G/L, p < 0.001), CRP (1.2 vs 1.1 mg/dL, p = 0.003), albumin levels (40.4 vs 39.1 g/L, p = 0.041), NLR (4.7 vs 2.7, p < 0.001), PLR (219.9 vs 209.3, p = 0.016), CAR (0.04 vs 0.03, p = 0.022), as well as an increase in LLR (0.16 vs 0.22, p < 0.001) (see Supplementary Table 8). Using univariable Cox regression with correction for Bonferroni and false discovery rate (FDR) to investigate associations with OS baseline values of absolute lymphocyte count (p < 0.001), PLR (p < 0.001), LDH (p = 0.003) and LLR (p = 0.002) remained statistically significant. A multivariable Cox model including baseline lymphocytes, PLR, LLR, LDH, and albumin showed significance (p < 0.001), however, none of the included variables remained individually significant (all p > 0.05). In a univariable Cox regression for PFS after multiple testing correction, baseline LDH (FDR = 0.046), PLR (FDR = 0.046), and NLR (FDR = 0.051) remained statistically significant, though not after Bonferroni correction. A multivariable Cox model including baseline lymphocytes, NLR, PLR, and LLR showed no individually significant variables (all p > 0.05). None of the relative changes in laboratory values were significantly associated with OS after correction for multiple testing. The lowest unadjusted p-values were observed for relative changes in PLR (p = 0.091), lymphocytes (p = 0.111), and LLR (p = 0.116), but none remained significant after Bonferroni or FDR correction (all adjusted p > 0.46). Similarly, no relative change in laboratory values was significantly associated with PFS. The smallest unadjusted p-value was found for relative change in LLR (p = 0.103), but this did not remain significant after Bonferroni (p = 1.0) or FDR correction (p = 0.993). None of the baseline or relative change parameters were significantly associated with early response pattern after multiple testing correction. The smallest unadjusted p-value was for relative change in LDH (p = 0.048), but this did not remain significant after Bonferroni (p = 1.0) or FDR correction (p = 0.392). Univariable logistic regression models were used to assess associations with DCB. Among baseline markers, NLR (p = 0.057), PLR (p = 0.053), LLR (p = 0.058), and lymphocyte count (p = 0.089) showed nominal significance, but none remained significant after Bonferroni or FDR adjustment (all adjusted p ≥ 0.463). No relative change variables were significantly associated with DCB (all raw p ≥ 0.183, FDR ≥ 0.743). Regarding BOR none of the baseline or relative change values were significantly associated after multiple testing correction. The most notable baseline associations were observed for PLR (p = 0.078), NLR (p = 0.102), and monocytes (p = 0.093), but these did not remain significant (all adjusted p ≥ 0.521). For relative changes, only PLR showed nominal significance (p = 0.027), but this was not significant after correction (adjusted p ≥ 0.521). All other p-values were > 0.1. Discussion By examining early radiological response patterns in NSCLC patients treated with ICIs and assessed according to iRECIST, we demonstrated that early tumor burden reduction is significantly associated with DCB and subsequent response outcomes, supporting its role as a predictive biomarker. Furthermore, organ-specific changes under ICI therapy correlated with long-term treatment response. Early changes in tumor size between baseline and first follow-up imaging were not only predictive of DCB but also significantly associated with OS and PFS. In Cox regression analyses, absolute tumor shrinkage at first follow-up was linked to improved OS and PFS. To better characterize heterogeneity within iSD, we applied a granular classification (iCR, iPR, iSD-low, iSD-high, iUPD) based on relative changes in the sum of diameters. Greater early tumor reductions were consistently associated with superior survival outcomes. All response groups demonstrated improved OS and PFS compared with iUPD, indicating that early tumor dynamics provide prognostic information even in the absence of formal iRECIST-defined response. Early iUPD may reflect primary resistance or, rarely, hyperprogressive disease under ICIs [13]. While differences between iSD-low and iSD-high were not statistically significant, a consistent trend toward improved survival with early shrinkage within the SD range was observed. In Cox models, iSD-low was associated with lower hazards for death and progression compared to iSD-high. Although not meeting significance thresholds, the direction and consistency of these estimates suggest that subtle radiographic changes within the iSD window may hold prognostic value and warrant validation in larger cohorts. The prognostic relevance of early response has been well described in metastatic melanoma, where tumor size changes at 12 weeks strongly predicted OS, with a 10% reduction proposed as an efficacy threshold [3]. Real-world data in stage IV melanoma (n = 522) showed no difference in survival between early and late responders, although early response was defined categorically (CR/PR within 3 months) [14]. In NSCLC, data on early imaging dynamics under ICIs remain scarce. Most studies were small, based on 18F-FDG-PET, and not conducted with ICIs [4, 5, 15–17]. In 16 patients receiving chemotherapy, a 0.5 SUV decrease early after treatment predicted DCB [15]. In locally advanced NSCLC (n = 28), early metabolic changes ≥ 38% were associated with improved PFS [16]. In 360 EGFR-mutated NSCLC patients treated with EGFR-TKIs, early responders (CR/PR at first follow-up) had superior PFS and OS [17]. A post-hoc analysis of five randomized trials with atezolizumab (n = 1337) demonstrated that high-risk progression patterns were associated with worse post-progression OS [18]. To our knowledge, our cohort represents the largest analysis of advanced NSCLC patients treated with ICIs evaluated for early predictive response using iRECIST. Organ-specific responses to pembrolizumab have also been described in metastatic urothelial carcinoma (n = 136), where mixed responses were associated with worse OS; lymph nodes showed prolonged benefit, while bone lesions responded poorly [19]. Consistent with our findings, bone lesions demonstrated unfavorable responses. Differences in organ-specific ORR may reflect variations in the tumor microenvironment. The presence of CD4+ and CD8+ TILs correlates with higher response to PD-1/PD-L1 inhibitors, whereas immune-desert phenotypes are associated with poor ORR [20, 21]. The liver may exert immunosuppressive effects [22, 23]. Morinaga et al. [7] demonstrated clonal and immunological heterogeneity in two melanoma lymph node metastases with divergent responses to anti-PD-1 therapy. Inflammatory biomarkers assessed at baseline and follow-up showed limited robustness after correction and multivariable adjustment, and dynamic changes were not predictive. In contrast, early radiographic tumor shrinkage demonstrated stronger and more consistent associations with clinical endpoints. Limitations include modest sample size, treatment heterogeneity, absence of pre-specified power calculation, and single-center design without external validation. Radiological images were retrospectively analyzed, although prospectively collected within the TRU biobanking program and centrally reviewed in a blinded manner. Imaging assessment was limited to baseline and first follow-up, reflecting real-world constraints. Nevertheless, consistent findings across analyses support the robustness of the associations. Furthermore, the analysis of laboratory parameters was restricted to routinely available inflammatory markers. While these markers are clinically accessible and biologically relevant, this approach does not capture the full spectrum of potential biomarkers. In particular, molecular or tumor-specific markers (e.g. circulating tumor DNA or immune-related signatures) were not assessed and may provide additional predictive value. Additionally, while SoD-based assessment according to RECIST/iRECIST is widely used and ensures standardization and reproducibility, it may not fully capture complex three-dimensional tumor changes, intralesional necrosis, or density alterations under ICI. Moreover, atypical response patterns such as pseudoprogression may not be adequately reflected by linear measurements alone. Advanced imaging approaches, including volumetric and densitometric analyses (e.g. Hounsfield unit-based), may provide a more comprehensive characterization of early treatment response. However, such data were not consistently available in this retrospective cohort. Future studies integrating these approaches may further refine imaging-based biomarkers in this setting. In conclusion, early tumor reduction following ICI initiation is associated with improved long-term outcomes in advanced NSCLC. Lung and lymph node lesions responded more favorably than other sites, supporting differential organ-level sensitivity to immunotherapy. While laboratory biomarkers showed limited predictive consistency, early imaging-based response assessment demonstrated superior prognostic utility. These findings support its integration into clinical decision-making, particularly in patients with heterogeneous response patterns within iSD. Further validation and mechanistic studies across tumor types are warranted. Electronic supplementary material Below is the link to the electronic supplementary material. Supplementary Material 1 Supplementary Material 1