Prediction of severe lymphopenia during chemoradiotherapy for esophageal cancer: development and validation of a pretreatment nomogram

Abstract

Introduction

In esophageal cancer patients, occurrence of severe radiation-induced lymphopenia during chemoradiotherapy has been associated with worse progression-free and overall survival. The aim of this study was to develop and validate a pretreatment clinical nomogram for the prediction of grade 4 lymphopenia.

Methods

A development set of consecutive patients who underwent chemoradiotherapy for esophageal cancer and an independent validation set of patients from another institution were identified. Grade 4 lymphopenia was defined as an absolute lymphocyte count (ALC) nadir during chemoradiotherapy of <0.2×10³/μL. Multivariable logistic regression analysis was used to create a prediction model for grade 4 lymphopenia in the development set, which was internally validated using bootstrapping, and externally validated by applying the model to the validation set. The model was presented as nomogram yielding 4 risk groups.

Results

Among 860 included patients, 322 (37%) experienced grade 4 lymphopenia. Higher age, larger planning target volume in interaction with lower BMI, photon- rather than proton-based therapy, and lower baseline ALC were predictive in the final model (corrected c-statistic 0.76). External validation in 144 patients in whom 58 (40%) had grade 4 lymphopenia yielded a c-statistic of 0.71. Four nomogram-based risk groups yielded predicted risk rates of 10%, 24%, 43%, and 70%, respectively.

Conclusions

A pretreatment clinical nomogram was developed and validated for the prediction of grade 4 radiation-induced lymphopenia during chemoradiotherapy for esophageal cancer. The nomogram can risk stratify individual patients suitable for lymphopenia-mitigating strategies or potential future therapeutic approaches to ultimately improve survival.

INTRODUCTION

Concurrent chemoradiotherapy (CRT) with or without surgery is the mainstay of treatment with curative intent for locally advanced esophageal cancer.^{1, 2} As lymphocytes are the most radiosensitive cells of the hematopoietic system, severe lymphopenia is a frequently occurring unintended consequence of such multimodality treatment involving radiotherapy.³ Preclinical and clinical studies have demonstrated that radiotherapy directly destroys circulating lymphocytes, which exhibit significant DNA fragmentation even at low radiation doses (<1 Gy).^4–7 With growing interest in the pivotal role of the immune system in tumor surveillance and therapy, an increasing number of studies demonstrated a relationship between radiation-induced lymphopenia and detrimental survival outcomes in solid tumors.^{3, 8} Accordingly, in esophageal cancer patients, occurrence of Common Terminology Criteria for Adverse Events (CTCAE) grade 4 (G4) lymphopenia (i.e. <0.2 ×10³/μL) during CRT has been associated with worse progression-free survival (PFS) and overall survival (OS).⁹ Also, a lower absolute lymphocyte count (ALC) nadir level during neoadjuvant CRT appeared related to a lower rate of pathologic complete response.¹⁰

Although the impact of radiotherapy on reducing circulating lymphocyte counts has been known for decades¹¹, the recent recognition of a potential association with tumor control has revived the interest to understand, predict, prevent and ameliorate this phenomenon. Identification of predictors for severe radiation-induced lymphopenia could aid in further hypothesis-generating understanding, yield modifiable factors to potentially reduce the risk, and help select subgroups of patients at high risk who might benefit from modified treatment approaches or potential future therapeutic strategies. In various cancers, several factors (known before treatment commences) have been associated with the risk of severe radiation-induced lymphopenia, including age, gender, race, steroid use, smoking, tumor location, overall stage, tumor histology, baseline hematocrit, lymphocyte and platelet counts, radiation modality, planning target volume (PTV), and type of concurrent chemotherapy.^{9, 10, 12–26}

The primary aim of the present study was to investigate the relationship of pretreatment patient- and tumor-related characteristics, as well as baseline full blood counts, with the risk of G4 lymphopenia during CRT in patients with esophageal cancer. A pretreatment clinical nomogram was developed, and subsequently validated both internally and externally, to guide individualized decision-making. A secondary aim was to assess the relationship of the presented nomogram-based risk groups with PFS and OS.

MATERIALS AND METHODS

This retrospective cohort study was approved by the institutional review board of MD Anderson Cancer Center for the development cohort and by the institutional review board of Mayo Clinic for the validation cohort. The requirement to obtain informed consent was waived. The study adhered to the Health Insurance Portability and Accountability Act (HIPAA), and to the reporting guidelines of the Transparent Reporting of a multivariable prediction model for Individual Prognosis or Diagnosis (TRIPOD) statement (http://www.tripod-statement.org).²⁷

Development population

From a prospectively collected departmental registry, data from all consecutive patients who underwent concurrent chemoradiotherapy (with or without surgery) for biopsy-proven esophageal cancer at MD Anderson Cancer Center between January 2004 and November 2017 were extracted. Exclusion criteria were overall stage IV (or unknown), histology other than adenocarcinoma or squamous cell carcinoma, a history of hematologic malignancy, endomucosal resection before CRT, a planned total radiation dose other than 50.4 Gy, split course radiotherapy, radiation modality other than proton beam therapy (PBT) or intensity-modulated radiotherapy (IMRT), and simultaneous irradiation of a second primary tumor. Subsequently, cases with unrecorded baseline blood sample data, a missing baseline ALC value, or less than 3 documented weekly ALC values during CRT, were excluded. The flow of patient selection is summarized in Supplementary Figure 1.

External validation population

Medical records from Mayo Clinic were reviewed for consecutive esophageal cancer patients receiving radiotherapy and concurrent chemotherapy with curative intent between July 2015 and December 2017. Patients with histology other than adenocarcinoma or squamous cell carcinoma were excluded. All patients received concurrent CRT (41.4 to 50.4 Gy) using either PBT, IMRT or 3D-conformal radiotherapy (3D-CRT), with a planned course of 5 weekly cycles of carboplatin, paclitaxel, followed by surgical resection in operable patients.

Outcome assessment

Full blood count data, including ALC values (cells x10³/μL) were obtained prior to CRT and weekly during CRT. The minimum ALC during CRT was identified as the nadir. In accordance with the Common Terminology Criteria for Adverse Events, version 4.0, G4 lymphopenia was defined as an ALC nadir <0.2 ×10³/μL.

Predictor selection

Only potential predictors that in clinical practice are known before treatment commences were selected. Potential predictors were selected based on the published literature and clinical reasoning, and included gender, age, BMI, race, Eastern Cooperative Oncology Group (ECOG) performance status, tumor histology, differentiation grade, tumor location, tumor length, clinical T- and N-stage, overall clinical stage, induction chemotherapy (yes versus no), PTV, radiation modality, total radiation dose, type of concurrent chemotherapy, as well as baseline red blood cell count (RBC), hemoglobin, hematocrit, white blood cell count (WBC), ALC, absolute neutrophil count (ANC), neutrophil-to-lymphocyte ratio (NLR), and platelet count.

Statistical analysis

Comparison between G4 and non-G4 lymphopenia groups for continuous variables at baseline was performed using the Student’s T-test. Nominal and ordered categorical variables were compared using Pearson’s chi-square and Mann-Whitney U tests, respectively. Missing data were considered missing at random and handled with multiple imputation creating 20 imputation datasets.

Model development and validation

A detailed description on the prediction model development is provided in the Supplementary File. After univariable logistic regression analysis, a full multivariable logistic regression model including all candidate predictors and 3 interaction terms was built. Next, redundant variables were eliminated with a backward stepwise approach based on Akaike’s Information Criterion (AIC).²⁸ The discriminative ability of the final model was tested using the c-statistic. To correct for optimism in model performance, for each imputed dataset internal validation using 1,000 bootstrap resamples was performed in which all (backward stepwise elimination) modeling steps were repeated. The performance of the final models of the 20 imputed datasets was pooled and used to calculate the optimism of the original model and the shrinkage factor, after which the c-statistic and the β-coefficients were corrected. The optimism-corrected β-coefficients were used in further analysis. The final prediction model was applied to the external cohort, yielding an (externally validated) c-statistic. Calibration of the final model, reflecting the agreement between predicted versus observed (actual) outcomes, was visualized with a calibration plot for both the development cohort and validation cohort.

Nomogram, risk groups and survival

The final model was presented as a nomogram. In the development cohort, for every patient an individual nomogram sum score was calculated by summing up the nomogram points per predictor. Subsequently, this sum score was divided into risk groups in such a way that for each group the difference in predicted probability of G4 lymphopenia was at least 10%.²⁷ Progression-free survival (PFS) and overall survival (OS) were calculated from the start date of CRT to disease progression or death, respectively, and censored at the date of last follow-up. Kaplan-Meier curves were constructed to compare PFS and OS for patients with and without observed G4 lymphopenia as well as for patients in the different nomogram risk groups. Cox proportional hazard analyses were applied to calculate hazard ratios (HRs) with 95% CIs for PFS and OS. Statistical analyses were performed using SPSS (version 24.0, IBM Corp., Armonk, NY) and R open-source software (version 3.5.0, ‘mice’ and ‘rms’ packages). A p-value <0.05 was considered statistically significant.

RESULTS

The development cohort consisted of a total of 860 patients treated with CRT for esophageal cancer. The average age (±standard deviation [SD]) was 63.1 (±10.7) years. The majority of patients was male (85%), had an adenocarcinoma (84%), which was most commonly located in the distal third of the esophagus (86%), and mostly staged cT3 (86%) and cN1–3 (66%). Induction chemotherapy preceded CRT in 252 patients (29%), and these patients had a slightly lower baseline ALC compared to patients without induction chemotherapy (median [IQR], 1.43 [1.08–1.84] versus 1.55 [1.24–1.98] x10³/μL; p=0.001). The radiation modality was PBT in 35% and IMRT in 65% of patients. Radiotherapy was completed in 860 patients (100%), and all concurrent chemotherapy cycles were completed in 832 (93%). Surgery was performed in 463 patients (54%) after a median of 8.1 weeks [interquartile range 6.7–11.2] after the completion of CRT.

Among the 860 included patients, 322 (37%) experienced G4 lymphopenia during CRT. Grade 3 or 4 lymphopenia (ALC nadir <0.5 ×10³/μL) was observed in 780 (91%) patients. Detailed information on baseline patient- and treatment-related characteristics and baseline full blood count characteristics in relation to G4 lymphopenia is presented in Table 1.

Table 1.

Baseline patient- and treatment-related characteristics.

Characteristic	G4 lymphopenia (n=322)	No G4 lymphopenia (n=538)	p value	Missing
Male gender	277 (86.0)	451 (83.8)	0.387	-
Age (y)†	64.5 ± 10.3	62.3 ± 10.8	0.004*	-
BMI (kg/m2)ǂ	24.9 [21.8–28.5]	25.8 [22.2–30.2]	0.032*	1 (0.1)
Race			0.539	7 (0.8)
White	285 (89.3)	484 (90.6)
Other	34 (10.7)	50 (9.4)
ECOG performance status			0.001*	-
0	91 (28.3)	203 (37.7)
1	201 (62.4)	310 (57.6)
2	30 (9.3)	25 (4.7)
Smoking at diagnosis	38 (12.0)	94 (17.7)	0.028*	13 (1.5)
Tumor histology			0.644	-
Adenocarcinoma	272 (84.5)	448 (83.3)
Squamous cell carcinoma	50 (15.5)	90 (16.7)
Differentiation grade			0.594	2 (0.2)
Good	4 (1.2)	4 (0.8)
Moderate	142 (44.4)	253 (47.0)
Poor	174 (54.4)	281 (52.2)
Tumor location			0.029*	-
Upper third of esophagus	5 (1.6)	24 (4.5)
Middle third of esophagus	31 (9.6)	64 (11.9)
Distal third of esophagus	286 (88.8)	450 (83.6)
Tumor length (cm)ǂ	5.0 [4.0–7.0]	5.0 [3.0–6.0]	0.002*	2 (0.2)
Clinical T-stage			0.008*	2 (0.2)
cT1	6 (1.9)	13 (2.4)
cT2	22 (6.8)	62 (11.5)
cT3	282 (87.9)	453 (84.4)
cT4	11 (3.4)	9 (1.7)
Clinical N-stage			0.001*	-
cN0	90 (27.9)	199 (37.0)
cN1	128 (39.8)	215 (40.0)
cN2	93 (28.9)	109 (20.2)
cN3	11 (3.4)	15 (2.8)
Overall clinical stage			0.002*	-
I	11 (3.4)	33 (6.1)
II	89 (27.6)	189 (35.1)
III	222 (69.0)	316 (58.8)
Induction chemotherapy	104 (32.3)	148 (27.5)	0.135	-
PTV (mL)ǂ	685 [526–850]	520 [379–706]	<0.001*	1 (0.1)
Radiation modality			<0.001*	-
Proton beam therapy	65 (20.2)	232 (43.1)
IMRT	257 (79.8)	306 (56.9)
Year of CRT startǂ	2010 [2007–2013]	2011 [2009–2013]	<0.001*	-
Concurrent chemotherapy			0.251	-
Taxane/5FU	183 (56.8)	277 (51.5)
Platinum/5FU	79 (24.5)	170 (31.6)
Taxane/platinum	29 (9.0)	44 (8.2)
Taxane/5FU/platinum	16 (5.0)	28 (5.2)
Other	15 (4.7)	19 (3.5)
Surgery after CRT	146 (45.3)	317 (58.9)	<0.001*	-
Baseline RBC (M/μL)†	4.34 ± 0.53	4.46 ± 0.53	0.002*	-
Baseline hemoglobin (g/dL)†	12.9 ± 1.7	13.2 ± 1.6	0.011*	-
Baseline hematocrit (%)†	38.5 ± 4.5	39.4 ± 4.7	0.003*	1 (0.1)
Baseline WBC (K/μL)ǂ	6.30 [5.30–7.90]	6.90 [5.70–8.60]	0.004*	-
Baseline ALC (K/μL)ǂ	1.36 [1.04–1.69]	1.65 [1.30–2.11]	<0.001*	-
Baseline ANC (K/μL)ǂ	4.20 [3.28–5.42]	4.25 [3.25–5.57]	0.295	-
Baseline NLRǂ	3.15 [2.22–4.45]	2.50 [1.82–3.47]	<0.001*	-
Baseline platelet count (K/μL)ǂ	222 [181–279]	224 [183–270]	0.857	1 (0.1)

Data are presented as numbers with percentages in parentheses.

^†: Expressed as mean ± SD.

^ǂ: Expressed as median [IQR].

^*: Statistically significant (p<0.05).

ALC: Absolute lymphocyte count. ANC: Absolute neutrophil count. BMI: Body mass index. CRT: Chemoradiotherapy. G4: Grade 4. IMRT: Intensity-modulated radiation therapy. NLR: Neutrophil-to-lymphocyte ratio. PTV: Planning target volume. RBC: Red blood cell count. WBC: White blood cell count.

Fifteen variables and 3 interactions terms were entered into a full multivariable logistic regression model, and after subsequent stepwise backward elimination the final model consisted of 5 independent predictors of G4 lymphopenia (age, BMI, PTV, radiation modality, and baseline ALC) along with 1 interaction term (between BMI and PTV) (Table 2). The significant interaction term revealed that the influence of an increased PTV on increasing the G4 lymphopenia risk was greater in patients with lower BMI compared to higher BMI (Figure 1). The apparent c-statistic of the final model was 0.78 (95% CI: 0.75–0.81), and after internal validation the corrected c-statistic was 0.76 (95% CI: 0.73–0.79).

Table 2.

Univariable and multivariable logistic regression analyses for grade 4 lymphopenia.

	Univariable analysis		Multivariable analysis
Characteristic	OR (95% CI)	p value	Corrected OR (95% CI)	p value
Age (y)	1.02 (1.01–1.03)	0.004*	1.02 (1.01–1.04)	0.003*
BMI (kg/m2)	0.97 (0.95–0.996)	0.023*	See Figure 1	0.014*
ECOG performance status
0	Ref
1–2	1.54 (1.14–2.07)	0.005*
Smoking at diagnosis
No	Ref
Yes	0.64 (0.43–0.96)	0.030*
Tumor location
Upper/middle third	Ref
Distal third	1.55 (1.03–2.35)	0.038*
log_Tumor length	1.52 (1.16–2.00)	0.002*
Clinical T-stage
cT1–2	Ref
cT3–4	1.70 (1.08–2.69)	0.023*
Clinical N-stage
cN0	Ref
cN1	1.32 (0.95–1.83)	0.104
cN2–3	1.85 (1.29–2.66)	0.001*
Induction chemotherapy
No	Ref
Yes	1.26 (0.93–1.70)	0.136
log_PTV	4.36 (3.06–6.22)	<0.001*	See Figure 1	<0.001*
Radiation modality
Proton beam therapy	Ref		Ref
IMRT	3.00 (2.17–4.13)	<0.001*	2.58 (1.87–3.57)	<0.001*
Year of CRT start	0.90 (0.86–0.94)	<0.001*
Baseline RBC (M/μL)	0.66 (0.51–0.86)	0.002*
Baseline ALC (K/μL)	0.34 (0.26–0.45)	<0.001*	0.36 (0.27–0.48)	<0.001*
Baseline WBC (K/μL)	0.92 (0.87–0.98)	0.007*
Interactions
BMI*log(PTV)	See Supplementary Figure 3	0.001*	See Figure 1	0.009*
ECOG*Radiation modality	See Supplementary Figure 3	0.025*
Year of CRT start*Radiation modality	See Supplementary Figure 3	0.002*

^*: Statistically significant (p<0.05).

ALC: Absolute lymphocyte count. CI: Confidence interval. CRT: Chemoradiotherapy. OR: Odds ratio. PTV: Planning target volume. RBC: Red blood cell count. WBC: White blood cell count.

Final model logistic regression formula (with shrinkage factor applied):

Log(p/1-p) = −22.845 + 0.021*Age + 0.516*BMI + 3.579*log(PTV) − 0.086*BMI*log(PTV) + 0.949*IMRT[0=no;1=yes] − 1.019*Baseline_ALC.

Figure 1.

Visualization of interaction effect between BMI and PTV on the estimated probability of grade 4 lymphopenia in the final multivariable logistic regression model corrected for optimism using the bootstrap-derived shrinkage factor (and adjusted for ‘Age’ and ‘Baseline ALC’ which were kept constant at their median values), separately presented for A) ‘Radiation modality’ = Proton beam therapy; and B) ‘Radiation modality’ = IMRT.

For the external validation cohort 144 patients were found eligible for inclusion, in whom 58 (40%) experienced G4 lymphopenia. The average age (±SD) was 65.6 (±10.3) years. The radiation modality was protons for 45% and photons for 55% (of which 40% had IMRT and 15% had 3D-CRT). Application of the final model in the external cohort yielded a c-statistic of 0.71 (95% CI: 0.62–0.80) for the prediction of G4 lymphopenia. Visual inspection of the model calibration plots shows an excellent overall fit in the development cohort and a satisfactory fit in the validation cohort (Supplementary Figure 2).

The final model was presented as a nomogram (Figure 2). Accordingly, points can be assigned for each predictor resulting in a nomogram sum score between 0 and 20 for each individual patient, which relates to the predicted G4 lymphopenia risk. Applying the nomogram sum score, patients from the development cohort were divided into 4 risk groups with increasing sum score categories (<12, 12–13, 13–15, and ≥15, respectively). Average predicted G4 lymphopenia risk rates across the risk groups were 11%, 25%, 43%, and 71%, respectively, which were in good agreement with the observed incidence rates (Table 3).

Figure 2.

Pretreatment clinical nomogram for predicting grade 4 lymphopenia during chemoradiotherapy for esophageal cancer. The points listed at the top of the figure indicate the points to be assigned per variable. By summing all points, the predicted risk of grade 4 lymphopenia can be read out by drawing a straight vertical line from the ‘Sum score’ line to the bottom line of the figure.

Table 3.

Grade 4 lymphopenia risk and survival estimates according to nomogram risk groups.

			Progression-free survival				Overall survival
Nomogram risk group (sum score)	n (% of total)	G4 lymphopenia (%)†	Median (months)	5-year (%)	HR (95% CI)	p value	Median (months)	5-year (%)	HR (95% CI)	p value
Low risk (<12)	221 (25.7)	10.0	70.3	53.5	Ref	-	78.3	54.9	Ref	-
Low-intermediate risk (12–13)	154 (17.9)	24.0	NR	50.2	1.06 (0.77–1.46)	0.720	59.9	48.6	1.18 (0.86–1.62)	0.296
High-intermediate risk (13–15)	288 (33.5)	43.4	28.5	40.9	1.37 (1.05–1.78)	0.019*	44.8	43.7	1.39 (1.07–1.80)	0.015*
High risk (>15)	197 (22.9)	70.1	19.4	39.9	1.68 (1.28–2.20)	<0.001*	34.7	36.8	1.57 (1.19–2.08)	0.002*

^*: Statistically significant (p<0.05).

^†: Observed probability of grade 4 lymphopenia.

CI: Confidence interval. G4: Grade 4. HR: Hazard ratio. NR: Not reached.

For patients alive at last follow-up, the median follow-up duration was 49 months (range 1.2–137). Patients with versus without observed G4 lymphopenia had a significantly worse PFS (median 19.1 versus 61.7 months, respectively; HR 1.39, 95% CI 1.14–1.69; p=0.001; Figure 3A) and OS (median 34.7 versus 63.1 months, respectively; HR 1.49, 95% CI 1.23–1.80; p<0.001; Figure 3B). Based on the pretreatment clinical nomogram, patients at high-intermediate or high predicted risk of G4 lymphopenia (i.e. nomogram sum score 13–15 and >15, respectively) had a significantly worse PFS and OS compared to patients at low risk (i.e. nomogram sum score <12) (Table 3, Figure 3C and Figure 3D). Patients at low, low-intermediate, high-intermediate, and high predicted risk of G4 lymphopenia yielded five-year PFS rates of 54%, 50%, 41%, and 40%, respectively, and 5-year OS rates of 55%, 49%, 44%, and 37%, respectively.

Figure 3.

Kaplan-Meier analysis for progression-free survival (A) and overall survival (B) for patients with versus without observed grade 4 lymphopenia during chemoradiotherapy for esophageal cancer. Kaplan-Meier analysis for progression-free survival (C) and overall survival (D) according to predicted nomogram risk groups for grade 4 lymphopenia during chemoradiotherapy for esophageal cancer.

DISCUSSION

In this study a pretreatment clinical nomogram was developed and validated, both internally and externally, for the prediction of grade 4 radiation-induced lymphopenia during chemoradiotherapy for esophageal cancer, showing satisfactory model performance in terms of discrimination and calibration. Before initiation of treatment, based on patient- and (planned) treatment-related characteristics the nomogram allows for prediction of the risk of grade 4 lymphopenia for each individual patient.

Older patients, with a large PTV relative to their BMI, treated with IMRT rather than PBT, and with a lower baseline ALC were found to have the highest probability to develop grade 4 lymphopenia during CRT for esophageal cancer. The predictive ability of age is supported by two recent studies in glioma patients treated with radiotherapy and temozolomide.^{16, 25} More pronounced in previous literature, lower baseline ALC has been recognized as important predictor of severe radiation-induced lymphopenia in brain tumors^{12, 16, 26}, as well as in head and neck, pancreatic, and prostate cancers.^{15, 18–20} The cause of decreased lymphocyte counts in some cancer patients at baseline is not fully elucidated. It has been suggested that lower baseline ALC may partly just represent poorer physical condition of a patient, but it may also be a surrogate marker of tumor-induced immune suppression driving tumor progression.²⁹ This is supported by the observation that many of the pathways leading to tumor immune escape (e.g. secretion of transforming growth factor [TGF]-β by the tumor and its microenvironment, and tumor cell expression of pro-apoptotic molecules such as Fas ligand and programmed death-1 ligand-1 [PD-L1]) may also facilitate lymphopenia.^29–31 Both the tumor-induced immune suppression hypothesis, as well as a mechanistic explanation of a faster depletion of a smaller baseline reserve capacity could explain the found predictive value of baseline ALC for severe radiation-induced lymphopenia.

In line with our finding that PTV is predictive for the risk of lymphopenia risk, previous studies in breast and pancreatic cancer suggested that shrinkage of radiation fields preserve lymphocytes through reducing exposure of circulating blood.^{18, 32} Similarly, gross tumor volume (GTV) appeared significantly correlated to the ALC nadir during radiotherapy in both non-small cell lung cancer (NSCLC) and small cell lung cancer series.^{14, 33} Furthermore, several studies suggested slightly other risk factors related to high tumor burden. Examples of such factors include higher N-stage²⁴ and higher overall stage.^{10, 14} These factors appeared redundant in our model likely due to their high correlation with (the stronger predictive ability of) PTV. Our reported interaction between PTV and BMI for the risk of lymphopenia –suggesting even more lymphocyte depleting effect of increased PTV in patients with lower BMI– was not previously described nor tested. A supporting physiologic explanation of our finding could be that larger BMI -implying larger total blood volume³⁴- relates to greater lymphocyte reserve for the same baseline ALC.

The protective effect of proton-based compared to photon-based radiotherapy on the risk of lymphopenia in esophageal cancer has recently been suggested in both descriptive cohort and comparative (propensity score-matched) studies.^{9, 10, 13} The protective effect mainly appeared to arise from a significantly lower mean body dose exposure with proton-based radiotherapy.^{9, 10} This finding could have important implications as in contrast to the other found clinical predictors in the current study (i.e. age, BMI, PTV, and baseline ALC), the chosen radiation modality represents a parameter that can be modified. Also, in esophageal cancer radiotherapy, dosimetric advantages of proton-based radiotherapy have been shown to enable considerable reductions in radiation dose to specific organs such as the heart, lung, and liver, leading to reduced normal tissue complication probabilities.^35–38 It could be hypothesized that dose to specific structures or organs (e.g. the spleen^{15, 17}) may have more devastating effects on lymphocyte counts than dose to other structures, but such specific dose-volume histogram (DVH) analyses in relation to the lymphopenia risk has not yet been studied in esophageal cancer. In fact, the external c-statistic of 0.71 of the currently presented prediction model leaves room for improvement, which may be gained by addition of organ-specific and modifiable DVH parameters. Besides the spleen, other lymphopoietic and hematopoietic sites (such as the thymus and bone marrow), or large harbors for lymphocytes (such as the heart and lymph nodes), may prove to be other key organs sensitive to a direct hit of lymphocytes by radiotherapy.³

Radiotherapy rather than chemotherapy appears to be the prominent cause in the development of severe lymphopenia during CRT for thoracic malignancy. This statement is supported in our study, in which no significant association was found between the risk of G4 lymphopenia and type of concurrent chemotherapy nor with having received prior induction chemotherapy. In addition, a large study in NSCLC patients found radiotherapy-related lymphocyte count reduction to be independent of concurrent chemotherapy use, for both patients receiving and not receiving concurrent chemotherapy.¹⁴ Both patient groups exhibited similar associations between higher radiotherapy target volumes and lower ALC nadirs.¹⁴ Also, one study reported treatment-related lymphopenia in four independent cohorts with tumors located at different anatomic sites, in which treatment with radiotherapy was the only commonality.⁸

The resulting prediction model of the current study may allow for pretreatment identification of patients at high risk of severe radiation-induced lymphopenia who may benefit (most) from a modified treatment approach or potential future therapeutic strategies aimed at reduction of the impact of CRT on lymphocyte counts or at enhancement of host immunity. First, alternative radiation approaches (e.g. proton beam therapy, hypofractionated regimens, improved plan optimization, decreased target volumes) could be tested with the goal of administering “immune-sparing” radiotherapy.⁸ Second, there are ongoing studies designated to prevent or restore treatment-related lymphopenia. One approach that has shown to be feasible in glioma patients is to harvest lymphocytes prior to treatment and apply reinfusion after the completion of radiotherapy.³⁹ An alternative approach that is currently under prospective investigation (NCT01368107) is to use exogenous administration of the cytokine interleukin-7 (IL-7) in patients with treatment-related lymphopenia to stimulate lymphocyte proliferation.^{40, 41} Whether these approaches restore lymphocyte homeostasis and result in a more potent anti-tumor response remains to be elucidated. Third, an improved understanding of the immunology underlying radiation-induced lymphopenia is desired for the development of novel strategies with the potential of improving survival in selected patients.

The clinical significance of the observed decrease in tumor control and survival in cancer patients developing severe radiation-induced lymphopenia^{8, 14, 21, 22} may become particularly relevant with the recent exploration of immunotherapy in the treatment of various cancer types. Indeed, in patients with melanoma the efficacy of treatment with immune checkpoint blockade was correlated to the amount of adequate circulating lymphocytes.^42–44 Moreover, in a recent multicenter cohort study of NSCLC, metastatic melanoma, and renal cell carcinoma patients treated with immune checkpoint inhibitors, extracranial or prolonged courses of radiotherapy increased the risk of severe lymphopenia which was associated with poorer survival.⁴⁵ Some investigators have also suggested that the severe depletion of lymphocytes from current standard CRT regimens in cancer types such as high-grade glioma and pancreatic cancer might undermine tumor responsiveness to checkpoint inhibitors and other immune modulating agents.^{15, 41} With the recent explosion of interest in immunotherapy and increasing multimodality approaches, future preclinical and clinical studies are indicated to elucidate the impact of lymphopenia on immunotherapies.

The results of this study should be interpreted with an understanding of the following limitations. First, the prediction research does not answer mechanistic questions, and causality between predictors and the outcome (i.e. lymphopenia) cannot be inferred. Similarly, causality between lymphopenia and poorer survival –although likely– remains uncertain until intervention studies are performed showing improved outcomes.⁸ Second, because of the retrospective nature of the study a possibility of selection bias exists (e.g. through selection of patients with available blood samples only). However, the study was strengthened by internal and external validation, as well as by the relatively large number of patients in both G4 lymphopenia and non-G4 lymphopenia groups, providing adequate power to study multiple variables and interactions, and discerning (small) effects on the outcome.

In conclusion, age, PTV in relation to BMI, baseline ALC, and radiation modality (proton beam therapy versus IMRT) were independently associated with the risk of grade 4 lymphopenia during CRT for esophageal cancer. An externally validated clinical nomogram combining these factors yielded a satisfactory predictive performance, enabling pretreatment prediction of the risk of grade 4 lymphopenia for each individual patient, which in turn is associated with PFS and OS. As such, the nomogram can aid in the selection of patients suitable for lymphopenia-mitigating strategies or potential future therapeutic approaches, which may ultimately improve survival.