Pretreatment Eosinophil Counts in Patients with Advanced or Metastatic Urothelial Carcinoma Treated with Anti–PD-1/PD-L1 Checkpoint Inhibitors

Abstract

Eosinophils influence antitumor immunity and may predict response to treatment with immune checkpoint inhibitors (ICIs). To examine the association between blood eosinophil counts and outcomes in patients with advanced or metastatic urothelial carcinoma (mUC) treated with ICIs, we identified two ICI-treated cohorts: discovery (n=60) and validation (n=111). Chemotherapy cohorts were used as comparators (first-line platinum-based chemotherapy, n=75; second-line or more pemetrexed, n=77). The primary endpoint was overall survival (OS). Secondary endpoints were time on treatment (ToT) and progression-free survival. Univariate and multivariate analyses were performed using Cox proportional hazard models. Associations between changes in eosinophil count at weeks 2/3 and 6 after the start of ICI treatment were analyzed using landmark analyses. Baseline characteristics of the ICI cohorts were similar. In the discovery cohort, an optimal cutoff for pretreatment eosinophil count was determined (Eos-Lo: <100 cells/μL; n=9 [15%]; Eos-Hi: ≥100 cells/μL; n=51 [85%]). Eos-Lo was associated with inferior outcomes (OS: HR, 3.98 [95% CI, 1.85–8.56]; P<0.013; ToT: HR, 2.45 [95% CI, 1.17–5.10]; P=0.017). This was confirmed in the validation cohort (Eos-Lo: n=17 [15%]; Eos-Hi: n=94 [85%]) (OS: HR, 2.51 [95% CI, 1.31–4.80]; P=0.006; ToT: HR, 2.22 [95% CI, 1.29–3.80]; P=0.004), and remained significant after adjustment for other prognostic factors. Changes in eosinophil counts at weeks 2/3 and 6 were not clearly associated with outcomes. In chemotherapy cohorts, eosinophil counts were not associated with outcomes. In conclusion, low pretreatment eosinophil count was associated with poorer outcomes in patients with mUC treated with ICIs, and may represent a new predictive biomarker.

INTRODUCTION

Anti–PD-1/PD-L1 immune checkpoint inhibitors (ICIs) are now established treatment options for metastatic urothelial carcinoma (mUC).^1–5 Responses are potentially durable, although only 15% to 25% of patients with mUC exhibit response to ICIs.^1–5 Various predictive and prognostic biomarkers are at different stages of evaluation, including disease-related factors (genomic alterations in DNA damage-repair [DDR] genes,⁶ tumor mutation burden, molecular subtype,⁷ and PD-L1 expression) and host-related factors (T-cell receptor clonality).⁸. Emerging data from other disease types indicate that non-T-cell–related immune cells, including eosinophils, might also be associated with ICI response.^9–11

Eosinophils are granulocytes that are commonly elevated with parasitic infection, allergies, and autoimmune conditions.^12–13. Eosinophils were recently shown to be associated with the efficacy of ICIs in melanoma,¹⁰ lung cancer,⁹ and renal cell carcinoma.¹¹ Peripheral blood eosinophil counts are part of the standard complete blood count panel and are therefore readily measurable and inexpensive to ascertain. We hypothesized that peripheral blood eosinophil count is associated with outcomes in patients with mUC treated with anti–PD-1/PD-L1 ICIs.

METHODS

Study Design and Patients

Our primary objective was to evaluate the association between peripheral blood eosinophil count and OS in patients with mUC treated with ICIs. Secondary objectives were to evaluate the association between peripheral blood eosinophil count and time on treatment (ToT) and progression-free survival (PFS), when data were available, and to explore the association between on-treatment changes in eosinophil counts and OS and ToT. The institutional review board at Memorial Sloan Kettering Cancer Center (MSK) approved this study before any procedures were conducted.

Four different cohorts of patients with mUC were included in this study. Two ICI-treated cohorts were included: a discovery cohort, consisting of patients enrolled in completed ICI clinical trials and reported previously in a study on DDR gene alterations,⁶ and a validation cohort, consisting of consecutive patients with mUC treated with ICIs as standard of care or in a clinical trial at MSK. In addition, two cohorts of patients treated with chemotherapy, which have previously been reported upon, were included: chemo cohort 1, consisting of patients with mUC treated with first-line platinum-based chemotherapy,¹⁴ and chemo cohort 2, consisting of patients with mUC treated with second-line or more pemetrexed.¹⁵ By design, there was no overlap between the discovery and validation cohorts.

Data Collection

Clinical data, including baseline and on-treatment complete blood count (CBC) results, were extracted from individual chart review. Pretreatment or baseline eosinophil counts were obtained from CBCs drawn within 7 days of the start of treatment, inclusive of the date of treatment. For on-treatment changes, we extracted eosinophil counts at the start of cycle 2 (i.e., at week 2 or 3 [±3 days], hereafter “week 2/3,” depending on the ICI agent used) and at week 6 (±3 days). Other clinical variables with known or proposed prognostic value were extracted, including variables comprising the Bellmunt prognostic classification, tumor mutational burden, and presence of mutations in DDR genes, as previously defined.⁶ All laboratory tests were performed in an MSK clinical laboratory.

Endpoints

The primary endpoint of this study was OS, defined as the time from the start of treatment to death or last follow-up. Secondary endpoints included ToT, which was evaluated for all cohorts, and PFS, which was evaluated for the discovery cohort only. ToT was defined as the interval between the first and last dose of ICI or chemotherapy. Patients still on treatment at the time of data cutoff were censored at the date of the last follow-up. PFS was defined as the time from the start of treatment to radiological progression (RECIST 1.1), death, or last follow-up. Data cutoff was January 15^th, 2019.

Statistical Methods

Descriptive statistics were used to summarize patient characteristics in each cohort. To evaluate associations between eosinophil counts and outcomes of interest, eosinophil counts were analyzed as a continuous variable; to account for potential nonlinear effects, eosinophil counts were grouped and analyzed as a categorical variable. For categorization of eosinophil counts into high and low groups, we determined an optimal cutoff point using the discovery cohort by identifying the cut-off using the maxstat Lau94 package for R which provided the largest difference in OS. Time-to-event variables were estimated and plotted using the Kaplan-Meier method. Univariate Cox proportional hazard models were fit to evaluate associations between pretreatment eosinophil counts and OS, ToT, and PFS, and multivariate Cox models were used to adjust for other known prognostic factors. Rank-based correlation using an inverse probability censoring weighted estimator was performed to analyze Kendall’s tau between OS and PFS or ToT.

To assess associations between changes in eosinophil counts at week 2/3 and 6 and OS and ToT, landmark analyses at week 2/3 and 6 were conducted for the ICI cohorts. For these analyses, only subjects who had not experienced the event of interest before the week-2/3 or week-6 time point were considered, and that time point was treated as the new baseline time for survival analyses. We categorized the change in eosinophil count in two ways: (1) using an optimal cutoff point determined by maximizing the log-rank statistic, using OS as the endpoint in the initial cohort; and (2) by categorizing the change in eosinophil counts into three groups: decrease (change <0), no change (change=0), and increase (change >0). All analyses were conducted using R (version 3.6.1).

RESULTS

ICI Cohorts

The discovery cohort included 60 patients who started treatment between April 2014 and December 2016, and the validation cohort included 111 patients who started treatment between November 2014 and July 2018. Overall, the median age was 68 years (interquartile range, 60–76 years), 78% were male, and 71% had bladder as the primary tumor location. In total, 70% of patients received prior chemotherapy for metastases. The discovery cohort had a lower proportion of women (12% vs 27%; P=0.033) and higher baseline neutrophil counts (median, 5.6 vs 4.7 cells/μL; P=0.009). In the validation cohort, 19% of patients were treated with combined anti–PD-1 and anti-CTLA4 ICIs. Other baseline parameters, such as age, prior use of platinum chemotherapy, DDR status, tumor mutational burden, and Bellmunt prognostic group were comparable between the two cohorts (Table 1). OS was similar between the two cohorts (log-rank P=0.58), but ToT was significantly shorter in the validation cohort than in the discovery cohort (log-rank P<0.001) (Supplementary Figure 1).

Table 1.

Baseline characteristics of the immunotherapy cohorts

Characteristic	Discovery Cohort (n=60)	Validation Cohort (n=111)	P
Age at treatment start, years	67 (61–72)	68 (60–76)	0.440
Sex			0.033
Male	53 (88)	81 (73)
Female	7 (12)	30 (27)
Primary site at diagnosis			0.800
Bladder	45 (75)	77 (69)
Renal pelvis/ureter	13 (22)	29 (26)
Urethra	2 (3)	5 (5)
Metastatic sites
Bone	19 (32)	28 (25)	0.500
Liver	16 (27)	23 (21)	0.400
Lung	20 (33)	47 (42)	0.300
Nodes	41 (68)	79 (71)	0.700
Previous chemotherapy treatment for advanced disease			>0.900
Yes	42 (70)	78 (70)
No	18 (30)	33 (30)
Hematologic parameters, cells/mL
Eosinophil	0.10 (0.10–0.20)	0.10 (0.10–0.20)	0.800
Lymphocyte	1.30 (0.88–1.70)	1.20 (0.85–1.60)	0.400
Neutrophil	5.60 (4.40–7.45)	4.70 (2.60–6.20)	0.009
Neutrophil:lymphocyte ratio	4.3 (3.30–6.30)	3.8 (2.60–6.20)	0.200
DDR Status
Wild type	32 (53)	57 (51)	0.800
Deleterious	15 (25)	25 (23)
VUS	13 (22)	29 (26)
Tumor mutation burden, Mutations/Mb	7.3 (0.0 – 200.9)	7.8 (0.8 – 56.3)	0.233
Current treatment			<0.001
Anti–PD-1	43 (72)	55 (50)
Anti–PD-L1	17 (28)	35 (32)
Anti-CTLA4 + anti–PD-1	0 (0)	21 (19)
Bellmunt prognostic factors a			0.386
0	10 (17)	28 (25)
1	31 (52)	55 (50)
2 or 3	19 (32)	28 (25)

^aPresence of liver metastasis, hemoglobin level of less than 10 g/dL, and ECOG performance status of more than zero.

Abbreviations: Data are median (interquartile range) or no. (%). DDR: DNA Damage response and repair genes; VUS: variant of unknown significance

Pretreatment Eosinophil Counts and Outcomes

In the discovery cohort, no significant associations between pretreatment eosinophil counts and other known variables were observed, including age, sex, Eastern Cooperative Oncology Group performance status, presence of hepatic or other visceral metastasis, Bellmunt prognostic group, prior treatment with platinum-based chemotherapy, type of immunotherapy received, DDR status, or tumor mutational burden.

Pretreatment eosinophil count as a continuous variable was positively associated with OS (HR, 0.01 [95% CI, 0.00–0.29]; P=0.009). The optimal cutoff for eosinophil count at baseline was determined to be 100 cells/μL, and patients were divided into two groups: Eos-Lo (<100 cells/μL; n=9 [15%]) and Eos-Hi (≥100 cells/μL; n=51 [85%]). Eos-Lo patients had significantly shorter OS than Eos-Hi patients (median OS, 3.5 vs. 22.4 months; log-rank P adjusted for multiple comparisons in cutoff-point selection using maxstat Lau94 approximation = 0.013) (Figure 1). On univariate analysis, DDR status, tumor mutational burden, and Bellmunt score (of 2 – 3) were also significantly associated with OS (Table 2). On multivariate analysis, eosinophil count remained significantly associated with OS after adjustment for DDR status and Bellmunt score (Table 3).

Figure 1. Outcomes in the discovery cohort.

In the discovery cohort, patients with low baseline blood eosinophil counts (Eos-Lo; Eos <100 cells/μL) had shorter (A) time on ICI treatment and (B) overall survival than patients with high baseline blood eosinophil counts (Eos-Hi; ≥100 cells/μL).

Table 2.

Univariate analysis of overall survival in the immunotherapy cohorts

Characteristic	Discovery Cohort			Validation Cohort
	HR	95% CI	P	HR	95% CI	P
Age at treatment start, continuous variable	1.00	0.97–1.03	>0.900	0.99	0.97–1.02	0.500
Sex, male vs female (reference)	0.89	0.32–2.53	0.800	0.90	0.50–1.65	0.700
Prior chemotherapy treatment
Prior platinum	1.05	0.53–2.10	0.900	1.81	0.97–3.38	0.062
Any chemotherapy	1.12	0.55–2.26	0.700	1.94	1.00–3.77	0.049
Type of immunotherapy
Anti–PD-1/anti–PD-L1 (reference) vs anti-CTLA4 ± anti–PD-1	NA	NA	NA	0.87	0.44–1.72	0.700
Anti–PD-1 vs anti–PD-L1 (reference)	1.23	0.61–2.50	0.600	0.81	0.42–1.55	0.500
Pretreatment eosinophil counts
Continuous variable	0.01	0.00–0.29	0.009	0.28	0.04–2.18	0.200
Eos-Lo vs Eos-Hi (reference)	3.98	1.85–8.56	0.013 b	2.51	1.31–4.80	0.006
DDR status, DDRmut (deleterious + VUS) vs DDRwt	0.30	0.12–0.74	0.009	0.93	0.49–1.75	0.800
Tumor mutational burden	0.95	0.91–1.00	0.035	0.98	0.94–1.01	0.200
Bellmunt prognostic factors a
1 vs 0	1.49	0.56–4.02	0.400	1.28	0.61–2.65	0.500
2 or 3 vs 0	2.56	0.91–7.21	0.076	2.56	1.18–5.57	0.018

^aPresence of liver metastasis, hemoglobin level of less than 10 g/dL, and ECOG performance status of more than zero.

^bLog-rank P adjusted for multiple comparisons owing to cutoff point selection.

Abbreviations: DDR, DNA damage repair; DDRmut, any mutation in DDR; DDRwt, wild-type DDR; NA, not available; NLR, neutrophil-to-lymphocyte ratio; VUS, variants of unknown significance

Table 3.

Multivariate analysis of overall survival in immunotherapy cohorts

Characteristic	Discovery Cohort			Validation Cohort
	HR	95% CI	P	HR	95% CI	P
Pretreatment eosinophil count, Eos-Lo vs Eos-Hi (reference)	3.23	1.42–7.38	0.005	2.32	1.20–4.48	0.012
DDR status, DDRmut (deleterious + VUS) vs DDRwt	0.37	0.18–0.75	0.006	0.82	0.46–1.46	0.510
Bellmunt prognostic factors a
1 vs 0	1.10	0.40–3.02	0.860	1.25	0.59–2.64	0.567
2 or 3 vs 0	1.43	0.48–4.29	0.524	2.53	1.11–5.80	0.028

^aPresence of liver metastasis, hemoglobin level of less than 10 g/dL, and ECOG performance status of more than zero.

Abbreviations: DDR, DNA damage repair; DDRmut, any mutation in DDR; DDRwt, wild-type DDR

ToT was also shorter in the Eos-Lo group than in the Eos-Hi group (median, 2.3 vs 8.9 months; HR, 2.45 [95% CI, 1.17–5.10]; P=0.017) (Figure 1 and Supplementary Table 1). The association between baseline eosinophil count and ToT however lost its statistical significance after adjustment for DDR and Bellmunt risk status (Supplementary Table 2). Additionally, PFS was shorter in the Eos-Lo group than in the Eos-Hi group (HR, 3.11 [95% CI, 1.44–6.73]; P=0.004). OS was strongly correlated with both PFS and ToT (Kendall’s tau 0.78 and 0.76, respectively).

Analysis of the validation cohort revealed similar findings. Eos-Lo patients (n=17 [15%]) had shorter OS than Eos-Hi patients (n=94 [85%]) (median OS, 8.7 vs 21.6 months; HR, 2.51 [95% CI, 1.31–4.80]; P=0.006) (Figure 2). ToT was also shorter in Eos-Lo patients than in Eos-Hi patients (median, 2.4 vs 4.2 months; HR, 2.22 [95% CI, 1.29–3.80]; P=0.004) (Figure 2, Table 2, Supplementary Table 1, Supplementary Table 2). Association between eosinophil levels and ToT remained significant in this cohort after controlling for DDR and Bellmunt score.

Figure 2. Outcomes in the validation cohort.

In the validation cohort, patients with low baseline blood eosinophil counts (Eos-Lo; Eos < 100 cells/μL ) had shorter (A) time on ICI treatment and (B) overall survival than patients with high baseline blood eosinophil counts (Eos-Hi; ≥100 cells/μL).

Early Changes in Eosinophil Counts

Further analyses were performed to investigate whether changes in peripheral blood eosinophil counts at week 2/3 and 6 after the start of ICI treatment were associated with differential OS and ToT. This analysis was performed using two approaches: (1) a landmark analysis at week 2/3 and 6, with the optimally selected cutoff point for each time point determined using the discovery cohort; and (2) a comparison of levels at week 2/3 and 6 with baseline levels, which was classified into three possible categories: decreased, unchanged, and increased. In the discovery cohort, at week 2/3, 3 patients were excluded (missing lab values, n=2; follow-up time <3 weeks, n=1). At 6 weeks, 11 patients were excluded because of missing values (evaluable, n=49). In the validation cohort, 10 and 24 patients were excluded from analyses at week 2/3 and 6, respectively, because of missing values.

There was a correlation between baseline eosinophil count and change at week 2/3 (Pearson correlation = −0.64; P<0.001) (Supplementary Table 3), indicating that any effect of change could be related to residual baseline effects. The optimally selected cutoff for change in eosinophil count was <100 versus ≥100 cells/μL (determined on the basis of OS in the discovery cohort). In the discovery cohort, at week 2/3, an increase in the eosinophil counts above the established cutoff was associated with shorter ToT (HR, 4.05; P=0.020), and this finding was confirmed in the validation cohort (HR, 1.78; P=0.040). Increased counts at week 2/3 were associated with shorter ToT than decreased counts (HR, 2.52 [95% CI, 1.01–6.30]; P=0.047), and unchanged counts were associated with shorter ToT than decreased counts, although this association was not statistically significant (HR, 2.03 [95% CI, 0.87–4.76]; P=0.100). A similar signal was observed in the validation cohort. Unchanged counts at week 2/3 were associated with shorter ToT than decreased counts (HR, 1.80 [95% CI, 1.05–3.08]; P=0.032), and increased counts were associated with shorter ToT than decreased counts, although this association was not statistically significant (HR, 1.50 [95% CI, 0.89–2.54]; P=0.13). Changes at week 2/3 were not associated with OS. No associations between changes at week 6 and ToT or OS were observed (Supplementary Table 3).

Chemotherapy Cohorts

To test whether the observations to date were mUC-related or therapy-specific features, we further evaluated the association between baseline eosinophil counts and outcomes in two separate chemotherapy-treated cohorts. Chemo cohort 1 included 75 patients with mUC treated with platinum-based chemotherapy as first-line treatment between October 2012 and September 2015.¹⁴ Chemo cohort 2 consisted of 77 patients treated with pemetrexed in the second-line (47%), third-line (48%), or fourth-line (16%) setting between November 2008 and November 2013.¹⁵ Baseline characteristics are presented in Supplementary Table 4. Of note, in chemo cohort 1, 7 of 71 evaluable patients (10%) were classified as Eos-Lo; in chemo cohort 2, 33 of 77 evaluable patients (43%) were classified as Eos-Lo.

For chemo cohort 1, no associations between pretreatment eosinophil counts and OS were observed, both when eosinophil count was considered as a continuous variable (HR, 0.53 [95% CI, 0.08–3.76]; P=0.500) and when separated as categorical values using previously determined cutoffs (HR, 0.82 [95% CI, 0.29–2.31]; P=0.700). Similarly, no associations between pretreatment eosinophil counts and ToT (categorical: HR, 0.55 [95% CI, 0.25–1.21]; P=0.140) or PFS (categorical: HR, 1.06 [95% CI, 0.45–2.49]; P=0.900) were observed (Supplementary Table 5, Supplementary Figure 2). Similar outcomes were observed for chemo cohort 2, in which pretreatment eosinophil counts were not associated with either OS (categorical: HR, 0.93 [95% CI, 0.57–1.50]; P=0.800) or PFS (categorical: HR, 0.74 [95% CI, 0.47–1.17]; P=0.200) (Supplementary Figure 3).

DISCUSSION

ICIs are an established treatment option for advanced urothelial carcinoma, and their efficacy may be influenced by a series of disease- and host-related factors.⁸ To our knowledge, this is the first work to show that a low pretreatment peripheral eosinophil count (observed in ~15% of patients with mUC) is independently associated with significantly shorter OS among patients with mUC treated with ICI. The correlation was demonstrated in two independent ICI-treated cohorts. To further demonstrate that this was an ICI-specific phenomenon, we further evaluated two separate chemotherapy cohorts—no association was observed between eosinophil count and clinical outcomes in these cohorts.

Eosinophils are specialized granulocytes that are part of the innate immune system and are involved in many biological processes, including allergic response, autoimmune conditions, and response to parasitic infections.^12,16 Preclinical models have associated eosinophils with tumor immune surveillance, antitumor immunity, suppression of metastasis, and antitumor cytotoxicity,^13,17–21 but correlations with clinical outcomes have remained unclear.^22–24 Immunomodulatory therapies—including high-dose IL-2 ²⁵ and, more recently, anti-CTLA4 ²⁶ and anti–PD-1 blockade ¹⁰—have long been shown to induce eosinophilia. Various reports across different disease types, including melanoma,¹⁰, non-small cell lung cancer,⁹ and renal cell carcinoma,¹¹ have found an association between higher peripheral eosinophil count and a lower risk of progression or longer OS. Our observations of patients with mUC were not dissimilar from prior observations of patients with different tumor types. In two ICI cohorts, low pretreatment eosinophil levels were consistently associated with shorter OS, independent of DDR status and Bellmunt score. Additionally important is that no association was observed in two separate cohorts of patients treated with cytotoxic chemotherapy, confirming that eosinophil count is in fact a predictive factor, instead of a general prognostic indicator.

On-treatment peripheral eosinophil kinetics did not provide additional predictive or prognostic value in our data set. This is in contrast to the protective benefit seen with early increases in eosinophil counts with ipilimumab in patients with advanced melanoma ²⁶ and other treatment-related immune cell alterations.^27,28 We did, however, observe an early increase in peripheral eosinophil counts at week 2/3, which might be associated with shorter ToT, although the observation was not consistent across cohorts and methods of analyses, nor was it correlated with OS. It is unclear whether this lack of association is attributable to the limited sample size or whether this is a biological factor.

In our analysis, we observed that while pretreatment eosinophil levels and DDR alterations were independent predictors of OS in the discovery cohort, DDR alterations were not significantly associated with outcomes in the validation cohort. This could be secondary to inherent biological and clinical differences between clinical trial and standard of care populations, as seen in different ToT between both cohorts (Supplementary Figure 1).

Although our analysis is limited by its retrospective nature, the inclusion of an ICI validation cohort and two chemotherapy cohorts added robustness to this analysis. Furthermore, eosinophil count is part of the standard complete blood count panel, and therefore levels were prospectively obtained as patients received their treatment.

Our observations suggest that baseline eosinophil count lower than 100 cells/μL is inversely associated with OS in patients receiving ICIs. We did not observe a correlation between elevated eosinophil counts and clinical outcomes. However, eosinophilia has been associated with high-grade immune-related cutaneous adverse events,²⁹ and immune-related toxicity has, in turn, been associated with treatment efficacy,³⁰ suggesting that eosinophils play a complex and poorly understood role in the efficacy and toxicity of ICIs. As a biomarker with potential clinical utility, baseline blood eosinophil count can be assessed easily and at low cost with an established clinical-grade assay. If our findings are prospectively validated, pretreatment blood eosinophil counts should be integrated into larger-scale nomograms or biomarker efforts to improve our ability to identify patients who are most likely to benefit, or not, from ICIs.