An Immune-Clinical Prognostic Index (ICPI) for Patients With De Novo Follicular Lymphoma Treated With R-CHOP/CHOP Chemotherapy

Purpose Although the role of tumor-infiltrating T cells in follicular lymphoma (FL) has been reported previously, the prognostic value of peripheral blood T lymphocyte subsets has not been systematically assessed. Thus, we aim to incorporate T-cell subsets with clinical features to develop a predictive model of clinical outcome. Methods We retrospectively screened a total of 1,008 patients, including 252 newly diagnosed de novo FL patients with available peripheral blood T lymphocyte subsets who were randomized to different sets (177 in the training set and 75 in the internal validation set). A nomogram and a novel immune-clinical prognostic index (ICPI) were established according to multivariate Cox regression analysis for progression-free survival (PFS). The concordance index (C-index), Akaike’s information criterion (AIC), and likelihood ratio chi-square were employed to compare the ICPI’s discriminatory capability and homogeneity to that of FLIPI, FLIPI2, and PRIMA-PI. Additional external validation was performed using a dataset (n = 157) from other four centers. Results In the training set, multivariate analysis identified five independent prognostic factors (Stage III/IV disease, elevated lactate dehydrogenase (LDH), Hb <120g/L, CD4+ <30.7% and CD8+ >36.6%) for PFS. A novel ICPI was established according to the number of risk factors and stratify patients into 3 risk groups: high, intermediate, and low-risk with 4-5, 2-3, 0-1 risk factors respectively. The hazard ratios for patients in the high and intermediate-risk groups than those in the low-risk were 27.640 and 2.758. The ICPI could stratify patients into different risk groups both in the training set (P < 0.0001), internal validation set (P = 0.0039) and external validation set (P = 0.04). Moreover, in patients treated with RCHOP-like therapy, the ICPI was also predictive (P < 0.0001). In comparison to FLIPI, FLIPI2, and PRIMA-PI (C-index, 0.613-0.647), the ICPI offered adequate discrimination capability with C-index values of 0.679. Additionally, it exhibits good performance based on the lowest AIC and highest likelihood ratio chi-square score. Conclusions The ICPI is a novel predictive model with improved prognostic performance for patients with de novo FL treated with R-CHOP/CHOP chemotherapy. It is capable to be used in routine practice and guides individualized precision therapy.


INTRODUCTION
Follicular lymphoma (FL) is the second most common form of non-Hodgkin's lymphoma and accounts for 20-30% of all adult lymphomas diagnosed worldwide (1,2). It is an indolent lymphoma with heterogeneity both in treatment strategies and clinical outcomes (3,4). The treatment approaches cover a range from a watch-and-wait strategy to CD20-directed immunotherapy alone or in combination with chemotherapy (5)(6)(7). Due to the importance of the immunological microenvironment in the oncogenesis and progression of FL, immunomodulatory drugs such as lenalidomide in combination with rituximab produced significant response rates in first-line therapy (8,9). To stratify risk categories within FL patients, a variety of clinical prognostic models have been identified such as the Follicular Lymphoma International Prognostic Index (FLIPI) and FLIPI-2, and also proved as useful predictors of outcome (10,11). Despite their frequent use, there are some distinct outcomes within risk groups, suggesting the need for additional prognostic parameters (7,12). However, these models are not enough to reliably guide patient treatment options, while the GELF (Groupe d'Etude des Lymphomes Folliculaires) criteria (13) which includes parameters of tumor burden serves as an indication for initial treatment. Recently, several molecular biomarkers and gene signatures with prognostic significance have been discovered (14,15). Several clinicogenomic models have been established to clarify the patients that are expected to exhibit poor outcomes after providing standard immunochemotherapy. For instance, the m7-FLIPI risk score, integrating the FLIPI with the Eastern Cooperative Oncology Group performance status (ECOG PS) and the mutation status of seven genes (EZH2, FOXO1, EP300, CREBBP, CARD11, MEF2B, ARID1A) appears to be better for the identification of high-risk patients compared with the existing clinical models (16,17). Additionally, a developed 23-gene signature model could capture multiple aspects of the tumor biology and identify patients with markedly distinct outcomes (18). Molecular markers and gene signatures, on the other hand, are costly, technically difficult, and not routinely accessible in most hospitals. Recently, it was demonstrated that lack of expression of the intrafollicular CD4+ T-cell predicted risk of early failure, and by integrating this microenvironment biomarker with the FLIPI, termed BioFLIPI, which could further enhance the identification of FL patients at risk of early failure (19). Considering its complexity and poor reproducibility, it remains of great interest for us to investigate the impact of peripheral blood T cell subsets on patients' clinical outcomes, which provides a simple indicator of the host's immune status.
Tumor microenvironments have been recognized to be an imperative factor in the pathogenesis of follicular lymphoma (20,21). Recently, several studies found that the composition of nonmalignant cells, rather than tumor cells, influenced FL prognosis, emphasizing the modulation of the clinical course of follicular lymphoma by immune responses (22)(23)(24)(25). Subsequently, various researches also tried to identify phenotypic markers to stratify patients, but this has proven to be more difficult than anticipated and also influenced by treatments (26)(27)(28). To the best of our knowledge, T lymphocytes can directly reflect the status of the body's immune function and changes in the number of subsets of T lymphocytes had predictive value in disease progression. Further, its distribution can be different in peripheral blood and tumor microenvironment (29,30). The T-cell subset can be used as a surrogate biomarker for the tumor microenvironment. Thus, this motivated us to identify the potential role of the peripheral blood T-cell subsets in FL, since its prognostic role was previously reported in diffuse large B-cell lymphoma (DLBCL) (31,32).
Here, we have established an easily applicable prognostic nomogram and proposed a new prognostic model, termed the Immune Clinical Prognostic Index (ICPI)based on peripheral blood T-lymphocytes (allowing ease of application in routine practice). This new model possesses the capability to risk-stratify patients well and predicts survival of the currently recommended regimen based on first-line immunochemotherapy. In comparison to traditional clinical scoring systems like FLIPI, FLIPI2 and PRIMA-PI, the performance of this suggested ICPI showed enhanced predictive abilities and it was convenient to be monitored dynamically in clinical application.

METHOD Study Population
A total of 1,008 patients were retrospectively screened and reviewed by two experienced haematopathologists for diagnostic confirmation at Tianjin Medical University Cancer Institute and Hospital (TMUCIH) between January 2011 and July 2020 ( Figure S1). We excluded the 737 patients (73%) for the following reasons: unavailability of peripheral blood T lymphocytes subsets (n = 630); histologically confirmed grade 3b FL (n=28); DLBCL with an additional FL 3b component (n=39); unavailability of other NHL or pathological data (n = 15); previously accepted chemotherapy (n = 25). The residual 271 histologically confirmed grade 1-3a de novo FL patients were further screened. Patients who were assigned to a wait-and-see (WW) policy (n = 7), refused to consider any therapy (n=4), received lenalidomide plus rituximab (n = 3), received singleagent rituximab (n = 2) or received fewer than two cycles of therapy (n=3) were not included. Finally, 252 patients with indications for treatment were found to be eligible for model construction and were randomly assigned to a training set (n = 177) and an internal validation set (n = 75) by 7:3. Overall, 141 (56%) patients met the GELF criteria. External validation was performed in an independent set (n = 157) from other four centers, including Shanxi Provincial Cancer Hospital, Second Hospital of Dalian Medical University, The First Hospital of Jilin University and Hunan Cancer Hospital, who were treated with RCHOP (rituximab,cyclophosphamide, doxorubicin, vincristine, and prednisone). The study was reviewed and approved by the Research Ethics Committee of the TMUCIH and other centers.

Flow Cytometry Analysis of Peripheral Blood T-Cell Subsets
The cells were stained with the anti-CD3-fluorescein isothiocyante (FITC), anti-CD4-FITC, and anti-CD8phycoerythrin (PE) monoclonal antibodies. After staining, the results were examined through Cell Quest software and FACSCalibur flow cytometer (Becton-Dickinson, USA). The percentage of lymphocytes positive for each antibody was identified as a result.

Construction and Validation of the Nomogram
The least absolute shrinkage and selection operator (LASSO) is a widely used system for regression (33), which was used for the screening of independent variables that affect outcomes and to prevent overfitting. The "glmnet" package was used to generate this regression. To find independent prognostic factors for progression-free survival (PFS), researchers used multivariate Cox proportional hazards regression. The coefficient of regression of the selected independent variables was used to construct the corresponding nomogram prediction model. Internal validation was performed, and a C-index was calculated by investigating the area under the receiver operating characteristic curve. Calibration plots were used to evaluate whether the observed and predicted survival probabilities were in concordance with the bootstrap resampling method (B=1000). The "rms" package was used to draw the calibration curve and internal verification of the nomogram.

Comparison Between Different Models
The ICPI score was evaluated by determining the risk categorization effect on PFS and which was then compared with the results obtained through developed prognostic indices (FLIPI, FLIPI-2 and PRIMA-PI). We compared the ICPI model with FLIPI, FLIPI2, and PRIMA-PI based on C-index and Akaike's information criterion (AIC) (34)(35)(36). The AIC seems to have a relative model quality measure; a better model would be based on smaller values. As a general guide, AIC differences < 2 between models show no improved model fitting, differences > 2 but < 10 reveal an improved fit, and <10 indicates a significant improvement in model fitting (37). The C-index measured the model's predictive capability, demonstrated as the possibility of concordance between observed and predicted survival. The Cindex corresponds to the area under the receiver operating characteristics curve. The values of the C-index equivalent to 0.5, 0.7, and 1.0 represent that the random, acceptable, or perfect discrimination model, respectively, between long and short survival times. To measure homogeneity, the likelihood ratio chi-square was calculated using Cox regression; a higher likelihood ratio chi-square score indicated better homogeneity.

Statistical Analysis
PFS was calculated from the initial treatment date to the date of death occur via any cause, disease progression or relapse, or last contact. The Kaplan-Meier method was used to estimate PFS, which was then compared among the groups with the help of a 2sided log-rank test. The counts of T lymphocyte subsets were intended by using the percentages obtained by peripheral blood flow cytometry (PBFCM). Categorical variables were analyzed using the c 2 test or Fisher's exact test. Using standard clinical thresholds, continuous biological variables were dichotomized. Especially, the determination of the optimal cut-off values for peripheral blood T lymphocyte subsets was accomplished by Xtile 3.6.1 software (Yale University, New Haven, CT, USA) (38) which was based on the highest value of c 2 defined by the logrank test and Kaplan-Meier survival analysis.
Each statistical test was two-sided. Statistical significance was set at 0.05 and the analyses were accomplished with SPSS 25.0 (SPSS Inc. Chicago, IL) and R software (version 4.0.3; http:// www.r-project.org/).

Clinicopathological Characteristics of Patients
Patients' characteristics were not significantly different between the training set and internal validation set ( Table 1). A total of 252 newly diagnosed FL patients were examined, the median age was 52 years (range, 24-81), and most of them suffering from stage III-IV disease (83%). There were 72% of patients who had more than five involved lymph nodes, 17% showed the value of more than 6 cm for the longest diameter of the largest involved node (LoDLIN). Involvement of bone marrow was found in 20% of patients, and 31% showed spleen involvement. Hemoglobin (Hb) level below 120 g/L occurred in 19% and elevated lactate dehydrogenase (LDH) and b2-microglobulin (b2-MG) were found to be 15% and 28% of patients, respectively. Overall, 218 patients received CHOP plus rituximab (RCHOP)-like therapy and 34 patients received CHOP-like therapy.
Based on all three clinical risk rating systems, all patients were divided into categories. Especially, 8 patients' LoDLIN could not be obtained so that FLIPI2 was evaluated among 244 patients.
According to FLIPI, 23% patients were low risk, 41% patients were intermediate risk and 36% patients were high risk. In the proposed FLIPI2 score, specifically for the rituximab-treated patients, 65% patients were low risk, 20% patients were intermediate risk and 12% patients were high risk. In the

Cutoff Values of Peripheral Blood T Lymphocyte Subsets
In our study, the cut-off values of CD4+ T cells, CD8+ T cells levels and CD4+/CD8+ ratio were determined by X-tile program, which were found to be 30.7%, 36.6% and 0.8, respectively (Figure 1).

Clinical Value of the Nomogram
In the training set, we used the Lasso regression approach to distinguish adverse prognostic factors due to the limited sample size and a large number of variables. The cross-validation and filtering practices of the independent variables are given in Figures 2A, B,   Arbor stage, Hb, LDH, CD4+, and CD8+ were independent predictors for PFS (Table S2).
Nomogram was developed to predict 1, 3, 5-year PFS upon the multivariate analysis results ( Figure 2C). Variables included stage III/IV disease, elevated LDH, Hb <120g/L, CD4+ <30.7%, CD8+ >36.6% entered the nomogram. The predictive accuracy for 1, 3, 5-year C-index measured PFS was 0.72, 0.67 and 0.70 respectively (Figures 3A-C). The calibration curve for the likelihood of 1, 3, and 5-year PFS revealed a strong connection between the observed result and the nomogram's prediction. (Figures 3D-F). Figure S2 (Table S3). Patients repartition in each ICPI risk group in the training set were detailed in Table S4. When comparing ICPI with FLIPI, 41% of patients were in the same risk group, whereas 54% were in adjacent risk categories. Risk classifications were largely different in 8 patients (5%; 8 patients were classified as low risk by ICPI but high risk by FLIPI). As for ICPI and FLIPI2, 57% of patients were in the same risk category using ICPI and FLIPI2, whereas 41% were in adjacent risk categories. Risk classifications were largely different in 3 patients (2%; 2 patients were classified as low risk by ICPI but high risk by FLIPI2, and 1 patient was classified as high risk by ICPI but low risk by FLIPI2).
To compare the ICPI with PRIMA-PI, 47% of patients were in the same risk category using ICPI and PRIMA-PI, whereas 42% were in adjacent risk categories. Risk classifications were largely different in 18 patients (10%; 14 patients were classified as low risk by ICPI but high risk by PRIMA-PI and 4 patient was classified as high risk by ICPI but low risk by PRIMA-PI). We then conducted Kaplan-Meier curves to estimate the effect of different risk groups on patients' PFS time according to the ICPI, FLIPI, FLIPI2 and PRIMAPI (Figures 4A-D). All of the scores could differentiate subgroups of patients with substantially different prognoses except for the PRIMA-PI (P = 0.34). The ICPI (P<0.0001) is comparable to FLIPI (P=0.013) in stratifying patients into different risk groups. However, the FLIPI2 and PRIMA-PI were unsatisfactory for stratifying patients between intermediate and highrisk groups. According to ICPI, 84 (47.5%) patients were classified at low risk, 84 (47.5%) were intermediate risk, and 9 (5%) were high risk. Two-year PFS was 94% (95% CI, 88%-100%), 95% (95% CI, 90%-99%), and 33% (95% CI, 12%-96%) for the ICPI in the low, intermediate and high risk groups, respectively. In both the internal validation (P = 0.0039) and external validation sets (P = 0.04), the ICPI could also classify patient into different risk groups (Figures 4E,  F). In the external validation sets according to ICPI, 34 patients (22%) were classified at low risk, 105 (67%) at intermediate risk, and 18 (11%) at high risk. Additionally, in our training set of patients treated with RCHOP-like therapy (49%, 47%, and 4% for patients at low risk, intermediate risk, and high risk, respectively; P < 0.0001), the ICPI was also predictive ( Figure S3). Therefore, the ICPI proved to be a robust tool for predicting survival.
As for model performance, when compared with FLIPI, FLIPI2, and PRIMA-PI, the ICPI provided the best data fitting designated by the lowest value of AIC (training set: ICPI vs.  Table 2). The ICPI also distinguished between patients with favorable and poor PFS (specified by the highest C-index: training set: 0.679, 0.647, 0.645, and 0.613, respectively, for models with ICPI, FLIPI, FLIPI2, and PRIMA-PI risk categories; internal validation set: 0.636, 0.619, 0.562, 0.507) ( Table 2)

DISCUSSION
To date, several prognostic indexes such as FLIPI and FLIPI2 are available to predict the prognosis in previously untreated FL patients; however, immunity of the host and tumor microenvironment is not considered by either FLIPI or FLIPI-2. The FLIPI was well-established to predict the OS via a retrospective study, and the FLPI2 was well-established to predict the PFS as the primary efficacy endpoint for model building (10,11). However, they are not capable to accurately predict the early progress of patients (eg, POD24) after receiving first-line immunochemotherapy (39) and unable to provide evidence of risk-adapted treatment strategy. Furthermore, clinicogenomic models integrating genomic data into a clinical model, such as m7-FLIPI and 23-gene signature, may be difficult to implement because of cost and complexity. Therefore, the ability to recognize those high-risk patients who might develop progression or recurrence before treatment is still important because they can be candidates for alternative intensified therapies. Here, with the help of the obtained clinical features and laboratory indexes that are part of standard procedures used in the diagnosis, we have developed a predictor model of PFS based on peripheral blood T-lymphocytes and clinical features that stratify patients into different risk groups. The model we developed is based on immune-relevant characteristics of human bodies rather than recognized clinical prognostic factors. It is well known that the status of the host immune is closely related to the development and occurrence of follicular lymphoma. T lymphocyte subsets including CD3+, CD4+, and CD8+ are the major components of the cellular immune system playing a leading role in antitumor immunity. Identifying the number of CD4+ T cells, CD8+ T cells, and the CD4+/CD8+ ratio in peripheral blood can thus represent the immunological state of patients with malignant tumors, and it may also be useful in predicting the prognosis of FL patients. The prognostic value of peripheral blood T-lymphocytes has also been reported in patients suffering from various subtypes of lymphoma (32,(40)(41)(42). However, no study has yet precisely reported the prognostic value of the T-cell subset in FL and its added value to clinical prognostic indices that identify patients especially at high risk of early death and progression even after the application of modern immunochemotherapy.
FL is a heterogeneous disorder, and over the last decade, a number of clinical parameters have emerged as significant prognostic factors (43,44). Aiming to establish a prognostic system, we utilized variables of clinical significance that had been previously related to outcomes for FL, as well as peripheral blood T-lymphocytes which were immunologically relevant effective and simple biomarkers. Using the Cox Regression Multivariate Analysis, five variables (Ann Arbor Stage, Hb, LDH, CD4+, CD8+) have been identified as the most significant parameters affecting the survival of FL patients, and all these variables are easily obtainable at the time of diagnosis. Additionally, we established a nomogram to assess the probability of 1, 3, 5-year PFS, which could provide the risk assessment for each patient. The results showed that the nomogram possesses a good accuracy level for the prediction of PFS.
To our knowledge, this is the first research to combine clinical features with peripheral blood T-lymphocytes to predict outcomes in FL. Based on the risk factors offered by patients, we developed a new immune clinical prognostic index (ICPI). This risk score is feasible since it is based on features that are readily available at the time of diagnosis. Unlike existing prognostic scoring systems, ICPI could better stratify patients into different groups and revealed an upgraded ability for predicting PFS in comparison with PRIMA-PI or FLIPI-2, and a comparable ability to FLIPI. Most importantly, peripheral blood T-lymphocytes are simple to attain in clinical routine, and their complex changes over time can have significant clinical implications. In comparison to traditional prognostic models, the ICPI's C-index for PFS prediction was superior to the predictive power of the FLIPI, FLIPI2, and PRIMA-PI. The key benefit of ICPI is that it combined peripheral blood T-lymphocyte subsets with traditional prognostic factors, which was found to stratify patients more efficiently than models that only used traditional prognostic factors, according to PFS. Hence, the model *The AIC provided a relative measure of the quality of the model; lower values indicate a better model fitting. AIC differences of < 2 designate no progress in fit, differences of > 2 but < 10 exhibited increasing progress in fit, and differences of more than 10 indicate significant improvement in model fit. † The C-index provided a measure of the model's predictive ability, defined as the likelihood of concordance between observed and predicted survival. The C-index corresponds to the area under the receiver operating characteristics curve. C-index values of 0.5, 0.7, and 1.0 suggest that the model discriminates between short and long survival periods in a random, permissible, or ideal manner, respectively. ‡ A higher likelihood ratio chi-square score indicates better homogeneity.
may be useful to identify the candidates for further immune-related research and clinical trial of the novel anti-immune drug. Additionally, although the ICPI helps to well stratify patients after the first-line treatment with rituximab plus chemotherapy, its prognosis value needs to be further explored among patients who received other frontline treatment like lenalidomide plus rituximab (R2), bendamustine and rituximab (BR) and single-agent rituximab in a large cohort study.
In conclusion, when compared to other models, the ICPI proposed in this study integrates peripheral blood Tlymphocytes, which represent the immune status of the host, and clinical risk factors and performs well in stratifying patients into distinct outcomes. This predictor is suitable to be used in routine practice and able to better adjust the existing options of therapy according to the patient risk category.

DATA AVAILABILITY STATEMENT
The original contributions presented in the study are included in the article/Supplementary Material. Further inquiries can be directed to the corresponding authors.

ETHICS STATEMENT
The study was reviewed and approved by the Research Ethics Committee of the Tianjin Medical University Cancer Institute and Hospital and other four centers. The patients/participants provided their written informed consent to participate in this study.

AUTHOR CONTRIBUTIONS
HZha, XW, and SZ contributed to conception and design of the study. LS, XS, OB and HZha provided external validation cohort data. XG, TZ and BM performed the research and data analysis. YL wrote the first draft of the manuscript. JY, WG wrote sections of the manuscript. LL, LQ, and ZQ were responsible for collecting clinical information. XR supported the essential reagents. All authors listed have made a substantial, direct, and intellectual contribution to the work and approved it for publication.