Real‐world outcomes of 18,186 metastatic solid tumor outpatients: Baseline blood cell counts correlate with survival after immune checkpoint inhibitor therapy

Abstract Background Patient survival in advanced/metastatic melanoma, non‐small cell lung cancer (NSCLC), and renal cell carcinoma (RCC) has improved with immune checkpoint inhibitors (ICI). Biomarkers' role in prognosis and treatment has been limited by conflicting trial results. Methods This retrospective, observational study analyzed baseline demographic, clinical, laboratory, and treatment data versus outcomes of The US Oncology Network adult outpatients. Patients with advanced/metastatic melanoma, NSCLC, or RCC treated between January 1, 2015 and November 30, 2020 were given ICI monotherapy or combination therapy with ipilimumab, pembrolizumab, nivolumab, or atezolizumab. Treatment outcomes (overall survival [OS], time to treatment discontinuation, time to next treatment) were followed longitudinally until May 31, 2021, last patient record, or date of death. Baseline blood cell counts, including absolute monocyte count (AMC), absolute lymphocyte count (ALC), monocyte‐to‐lymphocyte ratio (MLR), absolute neutrophil count (ANC), and eosinophil count, were subdivided into quintiles for univariate and multivariable Cox regression analyses. Results Data from 18,186 patients with advanced/metastatic melanoma (n = 3314), NSCLC (n = 12,416), and RCC (n = 2456) were analyzed. Better OS correlated with increased baseline serum albumin concentration, increased eosinophil and lymphocyte counts, and Western United States physician practice location. Decreased OS correlated with increased AMC, MLR, ANC, age, and worse Eastern Cooperative Oncology Group performance status. Conclusions To our knowledge, this study is the largest to date to associate baseline survival indicators and outcomes in outpatients with advanced/metastatic melanoma, NSCLC, or RCC and receiving ICIs. Results may inform disease‐specific prognostic models and help providers identify patients most likely to benefit from ICI therapy.


| INTRODUCTION
Of the estimated 1.9 million new cancer diagnoses anticipated in the United States (US) for 2022, a large proportion is expected to be melanoma (99,780), lung cancer (236,740), and renal cancer (79,000). 1For patients with metastatic disease, only 31.9% (melanoma), 7.0% (lung), and 15.3% (renal) are likely to survive for 5 years. 1 Better indicators are needed to help providers understand the relationship between patient and disease in order to select optimal therapeutic regimens, ideally those with broad relevance to multiple tumor types.
Melanoma, renal cell carcinoma (RCC), and possibly non-small cell lung cancer (NSCLC) are highly immunogenic tumors. 2,35][6] Despite improved overall survival results, many patients receiving ICIs do not respond, relapse, suffer from immune-related adverse events, and/ or incur high costs of therapy, or a combination of these factors. 7,8Established survival indicators may help providers mitigate these treatment risks. 9,10Serum albumin concentration, age, and Eastern Cooperative Oncology Group performance status (ECOG PS) are known to correlate with patient survival and are determined in routine care.Often a panel of predictors, such as the International Metastatic Renal Cell Carcinoma Database Consortium (IMDC) criteria for RCC, is used in determining prognosis. 113][14] Finding reliable survival indicators is a key component of precision medicine which is needed to optimize patient outcomes.
Biomarker research applicable to immunogenic tumor ICI therapy is ongoing. 2 Beyond PD-L1 status in NSCLC, no clinically validated prognostic biomarkers inform ICI therapeutic decision making. 13,14Recent data suggest weak utility of PD-L1 tumor status as malignancies with low PD-L1 expression may still respond to ICI therapy, albeit at a lower rate, perhaps due to heterogeneity of PD-L1 expression throughout tumor tissue or differing PD-L1 expression detection methods and cut-off criteria. 26][17][18][19][20][21][22][23][24][25][26][27][28][29][30][31] For example, increased blood absolute monocyte count (AMC) correlated with decreased overall survival for patients with melanoma receiving ICI therapy, 15,20,26 but not for those diagnosed with RCC. 17 Conversely, AMC correlated with variable survival in NSCLC across different studies. 18,25,30Despite numerous studies it remains uncertain whether individual blood cell counts 20,23,26,28,32 or their ratios 29,31,33,34 reliably correlate with patient survival.In this study we retrospectively examined baseline patient demographic, clinical, laboratory, and treatment history characteristics across a large US-based population of outpatients who had advanced/metastatic melanoma, NSCLC, or RCC that received ICI therapy.We sought to determine how routinely collected baseline blood disease-specific prognostic models and help providers identify patients most likely to benefit from ICI therapy.

K E Y W O R D S
biomarkers, check point control, medical oncology, melanoma, non small cell lung cancer, renal cancer

Lay Summary
This study analyzed data collected from an electronic health record database of 18,186 patients.Routine pre-treatment blood test data was assessed for correlation with survival for patients with advanced skin, lung, or kidney tumors receiving treatment that stimulates the person's immune system to fight cancer.Patient information came from The US Oncology Network community clinics between January 1, 2015 and May 31, 2021.Results showed blood cell count tests corresponded with patient survival.Improved or worsened survival depended on white blood cell type/ count.More research is needed to learn how this testing can help health care providers improve patient outcomes.
cell counts can inform patient survival and treatment response.

| Patients
The study population was determined from abstracted data from medical records of US adult community oncology outpatients.Electronic health record (EHR) data were drawn from practices within The US Oncology Network or from associated clinics that use the iKnowMed™ (iKM) EHR.The US Oncology Network includes a large community of 1400 affiliated, qualified physicians operating in over 500 sites across the US.Patient demographics, clinical data, laboratory data, and treatment history were collected.All data were handled in compliance with the Health Insurance Portability and Accountability Act (HIPAA) and the Health Information Technology for Economic and Clinical Health Act (HITECH).The study protocol was granted an exemption and waiver of informed consent under 45 CFR 46.104 by The US Oncology Institutional Review Board (IRB), IRB# 20-015E.

| Inclusion criteria
Patients included were at least 18 years of age, had advanced or metastatic melanoma, NSCLC, or RCC between January 1, 2015, to November 30, 2020 ("the study identification period"), and received a qualifying ICI monotherapy or a combination ICI regimen of ipilimumab, pembrolizumab, nivolumab, or atezolizumab, with or without concomitant targeted therapy or chemotherapy, within those dates.Advanced or metastatic disease was defined during EHR abstraction by the following criteria: documentation of stage III or IV disease, TNM stage of M = 1, recorded metastases location, reference to metastatic disease status, or a numbered line of therapy (LOT) and/or a LOT with references to "metastatic."Patient data must have included at least two clinic visits and included baseline blood cell counts obtained within 30 days prior to starting ICI therapy.

| Exclusion criteria
Patients were excluded if enrolled in clinical trials of investigative medication therapy or received treatment for another documented primary cancer during the observation period.Further exclusions were considered upon medical review for patients with other concurrent cancers, or if combination treatment contained <5 patients.

| Data collection
Data were obtained via programmed database abstraction queries of structured fields within the EHR.Patient treatment outcomes were followed longitudinally until May 31, 2021, last patient record, or date of death, whichever occurred first.Mortality data were obtained from the US Department of Commerce National Technical Information Service / Social Security Association: Limited Access Death Master File (https:// dmf.ntis.gov/ ), if not found in the EHR.Missing (random) data were assumed to be unavailable in our fixed dataset which reflects individual/variable EHR documentation practices.

| Outcomes
Outcomes included overall survival (OS), time to treatment discontinuation (TTD), and time to next treatment (TTNT).OS was calculated as the interval from first date of initiating a qualifying ICI regimen until date of death from any cause.Patients that did not die within the study observation period were censored on the study end date or last visit date available in the dataset, whichever occurred first.TTD was defined as the interval between the index ICI treatment initiation date and treatment discontinuation for any cause, including any treatment interruptions <90 days.Patients that did not discontinue ICI treatment during the study observation period were censored on the study end date or last visit date available in the dataset, whichever occurred first.Patients who did not receive a next treatment and were alive at end of study were censored on the study end date, or last available visit date, whichever occurred first.TTNT was defined as the interval between the index ICI treatment initiation and initiation of the next ICI treatment or death.Patients that did not receive a next treatment and were alive on the study end date were censored on the study end date, or last available visit date, whichever occurred first.

| Statistical analysis
Statistical analyses were conducted using SAS® v9.4; p-values <0.05 were considered statistically significant.AMC, absolute lymphocyte count (ALC), monocyte-to-lymphocyte ratio (MLR), absolute neutrophil count (ANC), and eosinophil count were subdivided into quintiles (Q) with Q1 including the lowest values and Q5 including the highest, based on the distribution for the sample within each tumor type (normal values in Figure 1). 35Quintiles were chosen to increase discriminative power among groups.All blood counts were derived by multiplication of white blood cell count and percentages of cells.Univariate and multivariable Cox regression methods were used to describe impacts of baseline biomarkers, including blood count quintiles, on OS, TTD, and TTNT by tumor type.Multivariable Cox regression models were adjusted for confounders and important survival indicators at baseline based on current practice per tumor type.Variables that had more than 50.0% of their data not documented were excluded from the multivariable Cox regression models.Multivariable Cox regression models employed a stepwise selection approach to identify covariates for multivariable models with p ≤ 0.25 for entry and ≤0.15 for retention.Survival indicators with significant outcome associations in univariate models and a priori variables of interest were used in multivariable modeling.Results of log-rank tests from survival curves showed whether, for each cancer type, time-to-event variables (TTD, TTNT, and OS) differed by quintiles.Covariate assessment and additional statistical analysis methods are in Data S1.

| Patient characteristics
We describe a large study population dataset in a US community-based setting that illustrates relationships among baseline blood cell counts, ICI use, and outpatient outcomes in advanced or metastatic melanoma, NSCLC, and RCC.In total, 18,186 patients at over 500 different clinical sites were treated with ipilimumab, pembrolizumab, nivolumab, or atezolizumab either as monotherapy or combination therapy (Figure 2).Patients were grouped by tumor type, including melanoma (n = 3314), NSCLC (n = 12,416), and RCC (n = 2456).Baseline demographic, clinical, biomarker, laboratory, and drug treatment data were stratified by tumor type (Table 1; Data S1).All patients had advanced or metastatic disease at time of ICI therapy.A majority of these patients had Stage IV disease at initial diagnosis (Table 1).Median (95% CI) OS per cohort was 67.80 months (59.50-not reached [NR], melanoma), 13.60 months (13.00-14.20,NSCLC), and 28.30 months (25.60-32.00,RCC). Figure 1 shows baseline blood cell quintiles subdivisions per tumor type relative to normal, low, and high values. 35

| In univariate analysis, longest median OS was associated with lower baseline AMC, ANC, and MLR, and higher baseline eosinophil cell count
Baseline blood cell counts were divided into quintiles (Q1-Q5), and survival curves were generated by quintile for cell counts by tumor type (Figures 1 and 3; Data S1).All quintiles (Q2-5) were compared to Q1 to evaluate inter-quintile relationships.For all cohorts, longest median OS was seen for lower AMC, ANC, and MLR quintiles.Conversely, median OS was longest across all cohorts for patients with higher eosinophil counts (Q5).Detailed results for AMC, ANC, MLR, and eosinophil count are in Data S1.

| In multivariable analysis, known and investigational baseline characteristics are associated with OS in all three tumor types
In multivariable analysis, baseline characteristics associated with OS across melanoma, NSCLC, and RCC included serum albumin concentration, age, region, eosinophil count, MLR, and ALC (Table 2).Forest plots (Figure 4) show cohort-specific, covariate associations with increased OS, as well as risk factors for decreased OS.Known covariates, such as high serum albumin concentration as well as investigational covariates, such as high eosinophil levels were associated with longer OS in all tumor types.Increased eosinophil count (Q5) appeared to have a greater influence on OS duration for the melanoma cohort (p < 0.0001; HR 0.45, 95% CI 0.36-0.57)than for the NSCLC (p < 0.0001; HR 0.73, 95% CI 0.66-0.81)or RCC (p = 0.002; HR 0.68, 95% CI 0.54-0.87).
Increased age, non-Western physician practice location, and increased MLR were risk factors for decreased OS for all cohorts (Figure 4).Per cohort, specific physician practice location regions were more at risk as compared to the Western region of the US.Higher MLR quintiles were associated with shorter OS duration for all cohorts.Increased MLR levels (Q5) appeared to have a greater influence on OS duration for the RCC cohort (p = 0.0003; HR

TTD and TTNT
TTD and TTNT were evaluated as proxies for progression free survival despite limitations in their ability to attribute treatment starts and stops to toxicity versus disease progression. 36Further discussion of TTD, TTNT data per cell type are available in Data S1 (Figures S2-S4).
This dataset is the largest known to date to examine associations among potential baseline survival indicators in patients with advanced/metastatic melanoma, NSCLC, or RCC who were given ICIs in community oncology outpatient settings.Consistent with expectations, serum albumin concentration, ECOG PS, and age each associated with OS outcomes in univariate analyses, as reported in other studies. 9,10These known prognostic indicators serve as positive controls to validate the relevance of this large dataset for investigation of additional survival indicators for which prior results have been less clear.][17][18][19][20][21][22][23][24][25][26][27][28][29][30][31] In our large patient series, increased baseline blood monocyte and neutrophil counts were associated with less favorable OS, whereas increased eosinophils and lymphocytes were associated with more favorable OS.]23,[26][27][28][29][30] Elevated blood monocyte and neutrophil counts are consistent with a persistent inflammatory state, whereas elevated baseline lymphocytes presumably facilitate the effectiveness of ICI therapy.Measuring these blood cell counts at baseline may help providers determine the clinical immune competence of their patients, i.e., baseline laboratory results may indicate the capacity of a patient to mount an ICI therapy-induced immune response.In addition, our study and earlier studies demonstrate that elevated neutrophil-to-lymphocyte ratio (NLR) correlated with worse outcomes for patients on ICIs (Data S1). 29,31ncreased baseline monocytes were associated with decreased OS in all cohorts.8][39][40] HLA-DR is a major histocompatibility (MHC) class II protein found on antigen presenting cells linking innate and adaptive immune response by presenting foreign peptide antigens to antigen-specific T lymphocytes. 37educed HLA-DR expression on MDSCs is associated with dysregulated myelopoiesis and impairs antigen presenting F I G U R E 3 Blood cell counts correlate with overall survival.See Figure 1 for quintile values and normal lab correlates.MLR, monocyteto-lymphocyte ratio; NSCLC, non-small cell lung cancer; RCC, renal cell carcinoma.cells processing of tumor-derived peptides, yielding an impaired immune response to cancer. 38Elevated numbers of MDSCs are also associated with diminished ICI responses in melanoma and RCC. 37,402][43][44][45][46][47] We believe this strategy is a very important initiative for future investigation.
In multivariable analysis, six variables (serum albumin concentration, age, region, eosinophil quintile, MLR quintile, and ALC quintile) were common survival indicators across all tumor types (Table 2).Other notable variables include calcium (melanoma); calcium, gender, race, body mass index, and smoking (NSCLC); and IMDC performance and IMDC calcium (RCC).1]48 Anti-CTLA-4 monotherapy was associated with longer OS for patients with melanoma, whereas anti-PD-L1 + anti-CLTA-4 combination therapy were associated with longer OS for NSCLC and RCC.These findings are surprising for the melanoma cohort since anti-PD-1 monotherapy has been shown in previously untreated patients to improve outcomes over anti-CLTA4 monotherapy 49 and was more often prescribed in this dataset than anti-CLTA-4 monotherapy.Reasons for this disparity between recommended therapy and these real-world data may include inherent patient selection bias (i.e., comorbidities impacting treatment selection), patient or provider preference (e.g., combination modality therapies).A non-standard immunotherapy regimen resulted in shortest OS across cohorts and might be attributable to prior therapy failure, immune-related adverse events, or patient preferences.
Interestingly, physician practice location (geographic region) consistently was associated with OS in this study.Patients in the Western US fared better than patients elsewhere.One explanation might be variable US regional microbiomes that influence ICI response in solid tumors based on dietary effects on microbial richness and diversity. 50,51Alternatively, provider practice patterns, access to care, insurance coverage, health disparities, timing of diagnosis, patient lifestyle, or other environmental factors might influence regional effects on patient survival.
Strengths of this study include its large representative sample size of patients that reflects US community oncology practice trends, as opposed to interventional clinical trials with restrictive enrollment.Real-world data studies are now being used with increased frequency as these approaches and reports appear to complement prospective trials. 52,53Also, this study compared survival indicators across three tumor types, allowing assessment of relationships within and among tumor types.23,[26][27][28][29][30] Limitations follow from this study's retrospective, observational, EHR-derived real-world-data design.Conclusions may be hampered by missing or incomplete staging information and other data given that EHR data are usually recorded for clinical care, not necessarily research (e.g., PD-1 status or tumor mutational burden).We recognize therapeutic practices changed over the course of the data collection period.Concomitant medications impacting blood cell counts (i.e., corticosteroids) were not queried for analysis.Generalizability may also be limited by location distribution of The US Oncology Network practices or use of the iKM EHR, not well represented in the Northeast.Some variables of interest, such as PD-L1 status, EGFR, ALK, ROS 1 , MET, and BRAF, were not consistently available.Structured EHR data limit progression-free survival (PFS) calculation due to absent disease progression information.PFS measurement requires regular and reliable assessments of disease progression, infeasible in this design.Instead, proxies TTD and TTNT were used despite limitations in their ability to attribute treatment starts and stops to toxicity versus disease progression. 36TTD and TTNT as endpoints are further confounded by differences in the LOT at study entry, e.g., nearly all (90.1%) of the melanoma cohort was receiving ICI as first line therapy (LOT1), whereas only about half of the NSCLC cohort (57.4%) and RCC cohort (44.0%) were starting ICI therapy as LOT1 (Table 1).This study examined baseline but not subsequent laboratory values, which may change over the course of ICI therapy.Furthermore, this study focused on initial and not subsequent ICI use, as was recently studied. 29achine learning longitudinal models based on data from patients with NSCLC receiving ICI-containing regimens showed that blood cell dynamics reflect pathologically-assessed ICI therapy response.Interestingly, similar to our study, elevated blood eosinophils during ICI therapy correlated with therapeutic response in both metastatic and early-stage NSCLC cohorts.

| CONCLUSION
These study results may inform practitioners by building on predictive ICI response models and disease-specific prognostic models to identify patients most likely to benefit from ICI therapy. 11,28,54Baseline blood cell counts may indicate pre-ICI clinical immune competence and present a practical way to inform likelihood of ICI response.This research focuses on aspects of precision medicine and can provide direction for future investigations of the impact of cell count relationships, characteristics, and functional qualities on patient outcomes which might yield further opportunities to improve therapy.

F I G U R E 2
Flowchart of study design (STROBE diagram).anCTLA-4: cytotoxic T lymphocyte-associated antigen 4 inhibitor; anti-PD-1: programmed cell death protein 1 inhibitor; anti-PD-L1: programmed cell death ligand 1 inhibitor; ICI: immune checkpoint inhibitor; NSCLC: non-small cell lung cancer; RCC: renal cell carcinoma.*Patients were excluded if after additional medical review by the principal investigator for patients with other concurrent cancers.Additional patients were excluded if treatment patterns were inconsistent with approved medical treatment standards or if a drug treatment class combination contained <5 patients.T A B L E 1 Patient characteristics.

F I G U R E 4
Forest plots of adjusted hazard ratios on overall survival by tumor type.Hazard ratios in blue; 95% confidence intervals in green; p-values shown per covariate.Additional covariate of interest data correlated with longer or decreased OS can be found in Data S1.BMI, body mass index; CTLA-4, cytotoxic T-lymphocyte-associated antigen 4; ECOG, Eastern Cooperative Oncology Group; IMDC, International Metastatic RC Database Consortium; LDH, lactate dehydrogenase; LOT, line of therapy; MLR, monocyte-to-lymphocyte ratio; PD-1, programmed cell death protein-1; PD-L1, programmed cell death ligand-1.
Significant survival indicators in multivariable regression by tumor type (reflection of data from Figure 4. Forest plots of adjusted hazard ratios on overall survival by tumor type).
T A B L E 2