Prognostic value of immunotherapy-induced organ inflammation assessed on ¹⁸FDG PET in patients with metastatic non-small cell lung cancer

To identify and externally validate ¹⁸FDG PET-CT predictive and prognostic biomarkers in patients with NSCLC treated with ICPIs, we analyzed data of two independent cohorts prospectively started in this setting in Nice (France) and Genova (Italy). Both studies were approved by the concerned ethics committee and regulatory agencies and informed consent was obtained from all included patients (Genova cohort: NCT02475382; Nice cohort: n°ID-RCB:2018-A02116-49/ID-RCB:2018-A00915-50). The present retrospective ancillary study was approved by the French INDS (National Health Data Institute): MR 2,610,080,620.

Peak standardized uptake values, normalized by body weight (SUV_peak), were calculated in regions of interest placed on the site of the tumors or organ inflammations.

Patients were followed with regular clinical evaluations and standard of care imaging (including follow-up brain MRI, ¹⁸FDG PET, and CT exams).

Overall survival (OS) was the primary endpoint of the study. OS was recorded and defined as the time from initiation of ICPI to death from any cause.

Progression-free survival (PFS) was the second endpoint of the study and defined as the time from the initiation of ICPI to confirmed tumor progression or death. Tumor progression needed to be confirmed by a multi-disciplinary tumor board, confronting the patient’s clinical status, PET/CT results, and brain MRI results. Confirmed tumor progression necessarily implied the decision to definitively stop the treatment. For PFS, we did not take the time of first evidence of tumor progression on PET/CT due pseudo-progression or dissociated response patterns [14].

Categorical data are shown as counts and percentages, and continuous variables as minimum, median, maximum, and means with standard deviations. Available case analysis was the approach used for missing data. OS and PFS were evaluated using the Kaplan–Meier method and estimated with 95% confidence interval (95%CI). Median follow-up with a 95% confidence interval was calculated with the reverse Kaplan–Meier method. The log-rank test was used for the univariate analysis of PFS and OS. For OS, a predictive model was built, using a multivariable Cox regression method by following a step-by-step descending variable selection procedure using the AIC criterion. The proportional hazard assumption was checked using statistical tests and graphical diagnostics based on the scaled Schoenfeld residuals [20]. A classifier predicting the risk of progression or death was based on the linear predictor given by the model. The ability of the model to predict OS at t = 24 months was evaluated using the area under the ROC curve (AUC). Risk groups for predictive models were obtained using the threshold value to obtain the best compromise between sensitivity and specificity. A p value of < 0.05 was considered statistically significant and all tests were two-sided. All statistical analyses were performed with R.3.5.2 software on Windows® and the survMisc, timeROC, and Survminer packages.

Forty-nine patients were prospectively included. Four patients were excluded because of PET_Interim1 cancelation or wrong timing. Forty-five remaining patients were evaluated in this study. Patients’ characteristics are indicated in Table 1. All patients were treated with nivolumab, the two-thirds being treated in the 3rd or later line of treatment (Table 1).

Mean delay between baseline PET and initiation of immunotherapy was 17.1 ± 13.3 days. Mean time between introduction of immunotherapy and PET_Interim1 was 52.7 ± 7.9 days. Mean time between introduction of immunotherapy and PET_Interim2 was 98.3 ± 19.0 days.

Patients’ median follow-up was 56.6 months (CI 95% = 55.2–NA); Patients’ median progression-free survival was 5.9 months (CI 95% = 3.2–8.7) (Supplementary Fig. S1). Patients’ median overall survival was 9.9 months (CI 95% = 9.1–17.7) (Supplementary Fig. S2). The occurrence of at least one site of ICPI-induced organ inflammation was observed on PET_Interim1 or PET_Interim2 in 86.7% of patients (39/45).

The types of ICPI-induced organ inflammation observed on PET/CT are described in Table 2. ICPI-induced gastritis was assessed in 20.0% (9/45) of patients.

As observed in the exploratory cohort from Nice, iPERCIST tumor response on PET_Interim1 was significantly associated with better OS (Fig. 2). Patients with a tumor response/stability had a 2.9-fold improved OS compared to patients with tumor progression (2-year OS = 46% vs 16%, p = 0.002).

Patients with an immuno-induced gastritis occurrence after ICPI initiation on PET_Interim1 and/or PET_Interim2 had a 2.7-fold improved OS compared to patients without it (2-year OS = 56% vs 17%, p = 0.04) (Fig. 2).

The other sites of organ inflammatory impairment (i.e., pneumonitis, thyroiditis, midgut/hindgut, other organs’ inflammation) were not significantly associated with OS.

Finally, the prognostic model obtained in the cohort from Nice, including iPERCIST response and immune-induced gastritis criteria, was assessed in this independent validation cohort, using the optimal threshold (= − 0.19) defined in the exploratory cohort. The AUC of ROC curve to predict 2-year OS was equal to 70.7 (CI 95% = 78.1–88.3) with a sensitivity of 72.0% and a specificity of 63.6%.

Having a longer follow-up in this validation cohort, we also assessed the prognostic value of the same model to predict 3-year OS: the AUC was equal to 78.2 (CI 95% = 57.1–99.2) with a sensitivity of 69.2% and a specificity of 80.0%.

The OS curves, distinguishing poor and good prognostic groups using this model, are given in Fig. 3.

Thus far, research on predictive biomarkers of ICPI efficacy has focused on tumor or micro-environmental biological parameters. Clinical biomarkers have been less studied. Previous studies have suggested that the occurrence of symptomatic irAEs is associated with a more favorable outcome to PD-1 inhibitors, probably due to cross-reactivity between tumor and host tissues [11, 27‐31]. For example, a pooled multi-institutional cohort recently found that the incidence of grade ≥ 2 irAEs was associated with improved long-term survival across different malignancies treated with ICPI monotherapy [27]. This overall finding was confirmed in patients with lung cancer treated with an anti-PD(L)1 therapy. In another retrospective study including 270 patients with NSCLC receiving ICPIs in different lines, OS and PFS were significantly better for patients with reported irAEs than those without [28]. Outcome was not significantly different according to the different irAE grades. On a per-organ statistical analysis, thyroid dysfunction was the only specific irAE positively associated with improved OS and PFS.

The prognostic value of immune-induced inflammatory organ uptake assessed with ¹⁸FDG PET has been less investigated than clinical irAEs. As a previous study conducted on a small cohort of 25 patients with metastatic melanoma [32], we did not found a global prognostic value of immune-induced organ inflammation on ¹⁸FDG PET/CT performed in the early patients’ follow-up. But on a per-organ analysis, the occurrence of a gastritis during the first 3 months of ICPI treatment was significantly associated with improved OS (but not PFS) in the exploratory cohort. This discrepancy between OS and PFS is common when evaluating immunotherapy and may be explained by the difficulty of determining a reliable PFS in that setting. Patients with an immune-induced gastritis had a twofold improved 2-year OS compared to patients without it. This result was confirmed in the external validation cohort, in a different hospital and using a different PET/CT system. This independent validation was important to avoid false biomarker discovery and confirm that immuno-induced gastritis on PET is a reproducible predictive biomarker of improved clinical outcome. Moreover, combining tumor response (iPERCIST criteria) and gastritis occurrence can more accurately predict patient’s overall survival, as demonstrated by the predictive model. This model could identify long-term survivors with immunotherapy in advanced NSCLC (3-year OS) with a 69% sensitivity and 80% specificity in the external validation cohort. From a clinical perspective, ICPI-induced gastritis could be an additional argument for ICPI efficacy in case of doubtful or atypical response patterns on ¹⁸FDG PET/CT, such as dissociated response or pseudo-progression, often observed in the monitoring of patients treated with ICPI [14, 33].

Because gastritis symptoms can be insidious, overlapping with the common symptoms of metastatic cancer, immuno-induced gastritis may frequently be under-diagnosed without endoscopic confirmation [34]. The current literature is mostly comprised of case reports and series of patients [34‐36]. Only one large retrospective study (n = 205) has described the incidence of ICPI-gastritis at 5.4% of patients receiving immunotherapy for melanoma [37].

The occurrence of diffuse uptake of the gastric wall during ICPI treatment, related to immune-induced gastritis, has been mentioned in few previous ¹⁸FDG PET/CT case reports [21, 32, 38]. In the neoadjuvant setting of bladder cancer, Marandino et al. noted diffuse stomach uptake in 13.6% of patients (14/103) [21]. In this case, gastritis was always asymptomatic and therefore not confirmed by a gastroscopy.

There is currently no clear physio-pathological explanation of the association between immune-induced gastritis and improved outcome. The patient’s microbiota may be an interesting first track to explore this association. Indeed, the role of microbiota in regulating response and toxicities to ICPIs in cancer is increasingly recognized [39, 40], as the gut microbiome can create a pro- or an anti-inflammatory environment. Certain bacterial strains, such as Faecalibacterium, confer sensitivity to ICPI, also increasing the risk of irAEs [40], whereas dysbiosis or loss of microbial diversity is associated with poor response to ICPIs [41]. Recent studies have shown that the stomach also harbors a specific and diverse microbiota, although different from the rest of the digestive tract [42]. Interestingly, Helicobacter pylori (HP) can induce PD-L1 expression on gastric epithelial cells [43] and, in a large pooled analysis including 1512 patients with NSCLC randomly receiving atezolizumab or docetaxel, treatment with proton pump inhibitors (PPIs) was demonstrated to be specifically associated with a poorer outcome in patients treated with ICPIs [44]. Similar results were found in another retrospective study [45]. These data suggest that PPIs that induce gastric pH change may lead to subsequent alteration of gastric microbiota and influence the systemic tumor response to ICPIs. Mechanistic studies into the microbiome and gastric immunological pathways are needed to elucidate these physio-pathological mechanisms.

Despite the well-described role of the gut microbiome (assessed from stool samples) in shaping systemic immune responses, the occurrence of ICPI-induced midgut/hindgut inflammation on ¹⁸FDG PET/CT was not significantly associated with patient’s outcome. One explanation may be that we collected the occurrence of ICPI-induced organ inflammation only in a binary manner (occurrence versus nonoccurrence), without quantifying the magnitude of the intensity and extent of inflammatory uptake assessed on PET_Interim1/2. It could have been relevant to quantify the severity of ¹⁸FDG uptake to distinguish patients with minor and major inflammatory involvement of the midgut/hindgut. This quantitative work could be the subject of a future specific ancillary study.

This study has some limitations. Firstly, although most irAEs occur within the first 3 months of treatment, the 3-month follow-up may be too short to detect some types of delayed irAEs such as nephritis [24, 27]. Secondly, we did not make any correlation between ICPI-induced tissue inflammation observed on PET and the occurrence of clinically symptomatic irAEs, as this was not the main objective of this observational study. Thirdly, we focused only on patients treated with ICPI as single agent therapies. Other studies should evaluate the impact of irAEs on clinical outcomes in patients treated with a combination of PD-L1 inhibitors and chemotherapy, the new first-line gold standard for many patients. However, the relationship between irAEs and IPCI efficacy will be more challenging to assess in combination treatment strategies.

Finally, significant differences in survival were observed between the exploratory and validation cohorts. This may be explained by differences in the clinical characteristics of the two cohorts. Patients from Nice mainly received ICPI in the first and second lines, whereas two-thirds of Genova patients were treated in second or later lines. Although the early objective response rate to ICPI did not differ significantly between the 2 cohorts (29.3% vs 27.3% on PET_Interim1), tumor escape and progression occurred earlier in the more heavily treated cohort, prior to ICPI, with a median PFS of 14.3 months (Nice) vs 5.9 months (Genova).