The pretreatment albumin to globulin ratio predicts survival in patients with natural killer/T-cell lymphoma

Patient selection and baseline evaluation

The medical records of 385 previously untreated patients with NKTCL who were treated at Sun Yat-sen University Cancer Center between 2001 and 2013 were reviewed. The inclusion criteria included: (1) diagnosis of NKTCL according to the World Health Organization (WHO) classification of lymphomas (Chan, Jaffe & Ralfkiaer, 2001; Chan et al., 2008); (2) complete laboratory data before initial treatment; and (3) complete follow-up data. The exclusion criteria were: (1) known active inflammatory disorders including autoimmune disease and infection; (2) known active liver or kidney disease; and (3) receiving supportive care alone, without chemotherapy or radiotherapy (RT). A total of 331 patients were included in this study. Written informed consent for the collection of medical information was obtained from all patients at the first visit. The ethics committee of Sun Yat-sen University Cancer Center approved this study (No. B2015-054-01).

Baseline evaluations included a medical history and physical examination; complete blood count; serum biochemistry (including lactate dehydrogenase (LDH) level); computed tomography and/or magnetic resonance imaging of the head and neck; computed tomography of the chest, abdomen, and pelvis; and a bone marrow examination. Patients were staged according to the Ann Arbor system. The International Prognostic Index (IPI, including age, performance status, stage of disease, LDH level, and number of extranodal lesions) and the natural killer/T-cell lymphoma prognostic index (NKPI, including stage of disease, the involvement of regional lymph nodes, B symptoms, and LDH level) were calculated for all patients (Shipp et al., 1993; Lee et al., 2006). The primary sites of disease were classified into the upper aerodigestive tract (UAT, including the nasal cavity, Waldeyer’s ring, hypopharynx, larynx, and oral cavity) and the extra-UAT (any site other than a UAT site) (Kim et al., 2008). Total serum protein and serum albumin were measured using an automated immunoturbidimetric analyzer (7600-020; Hitachi High-Technologies, Tokyo, Japan) within seven days before initial treatment. The AGR was calculated as albumin/(total protein−albumin).

Treatment

The treatment modality has been described previously (Bi et al., 2015a). Patients with early stage disease received induction chemotherapy followed by involved-field radiotherapy (IFRT). Patients with advanced disease received chemotherapy as the primary treatment, and IFRT could be delivered as consolidation, palliative or salvage therapy according to the physician’s clinical judgment. As previously reported (Bi et al., 2015a; Bi et al., 2015b), chemotherapy regimens varied over the study period but were categorized as an asparaginase (ASP)-containing or anthracycline-based regimen, depending on whether L-ASP or pegaspargase was incorporated. IFRT was administered in daily fractions of 2.0–2.5 Gy (5 days a week), with a median dose of 54.6 Gy (range, 18.0–74.0 Gy).

Follow-up and statistical analysis

The follow-up schedule has been reported previously (Wang et al., 2014). The overall survival (OS) was measured from the date of diagnosis to the date of death due to any cause or the date of the most recent follow-up. Progression-free survival (PFS) was measured from the date of diagnosis to the date of disease progression, relapse, death due to any cause, or the most recent follow-up. The survival data were calculated using the Kaplan–Meier method and compared using the log-rank test. Continuous variables were presented as medians (range) and were compared using the Mann–Whitney U-test. Categorical variables were reported as frequencies and percentages and were compared using the Chi-squared test. The Cox proportional hazard model was used for a univariate screen of all potential predictors of survival. Variables with statistical and clinical significance were included in the multivariate analysis using the stepwise forward Cox regression model. Results were considered statistically significant with a two-sided P value < 0.05. The statistical analysis was performed using SPSS version 17.0 software (SPSS A, Inc., Chicago, IL, USA).

Determination of the optimal cut-off value for AGR

The optimal cut-off point to define a high or low AGR was determined using the method proposed by Igarashi et al. (2001). Using the log-rank test, we examined the discriminative power of different cut-off values for the prediction of OS from 1.0 to 2.0 in steps of 0.1. The cut-off value at which the OS curves separated most significantly (i.e., 1.3) was selected to categorize the AGR as low or high (Table 1). A total of 117 (35.3%) patients had a low AGR and 214 (64.7%) had a high AGR.

Clinical features and treatment

In Table 2, the clinical characteristics and treatment modalities of the cohort are presented and compared between patients with a low or high AGR. The cohort primarily comprised males (male/female ratio, 2.2:1). The median age of the patients was 42 years, with 15.4% of patients aged older than 60 years. Most of the patients presented with good performance status (84.3%), primary disease originating from UAT (88.8%), and early-stage disease (81.3%). A total of 52.3 and 29.0% of the patients had B symptoms and elevated LDH levels, respectively. The IPI was scored as 0–1 in 78.2% of patients, and the NKPI score was 0–2 in 81.0% of patients.

The baseline characteristics differed significantly between patients with a low or high AGR. As shown in Table 2, patients in the low AGR group presented with significantly more adverse clinical features, including advanced disease (stage III-IV: 25.6 vs. 15.0%, P = 0.017), elevated LDH (40.2 vs. 22.9%, P = 0.001), B symptoms (73.5 vs. 40.7%, P < 0.001), involvement of regional lymph nodes (49.6 vs. 28.5%, P < 0.001), and a higher IPI score (P < 0.001) and NKPI score (P < 0.001). Patients with a low AGR were slightly older (>60 years: 20.5 vs. 12.6%, P = 0.057), had a significantly worse Eastern Cooperative Oncology Group (ECOG) performance score (≥2: 28.2 vs. 8.9%, P < 0.001), and had a lower median body mass index (BMI, 20.4 vs. 21.4, P = 0.010). Serum albumin (median: 38.0 vs. 43.5, P < 0.001) and absolute lymphocyte count (median: 1.4 vs. 1.6, P = 0.032) were significantly lower in the low AGR group compared with the high AGR group, while C-reactive protein (CRP; median: 12.8 vs. 3.9, P < 0.001) and the erythrocyte sedimentation rate (ESR, 41 vs. 11, P < 0.001) were markedly higher in the low AGR group. In addition, primary lesions in the upper aerodigestive tract were less frequently observed in the low AGR group (83.8 vs. 91.6%, P = 0.031). Serum creatinine was higher in the high AGR group than compared with the low AGR group (median: 69.3 vs. 64.1, P = 0.011).

As shown in Table 2, 67.7% of patients underwent RT with or without chemotherapy as the first-line treatment, while 32.3% received chemotherapy alone. Anthracycline-based and ASP-containing chemotherapeutic regimens were administered to 60.7 and 39.3% of patients, respectively. The median dose of radiation was 54.6 Gy (range, 18.0–74.0 Gy), and the median number of chemotherapy cycles was 4 (range, 0–10). The treatment modalities differed significantly between patients in the low and high AGR groups (Table 2). There were significantly more patients receiving chemotherapy alone without RT in the low AGR group than in the high AGR group (43.6 vs. 26.2%, P = 0.001). ASP-containing chemotherapeutic regimens were used more frequently in the low AGR group than in the high AGR group (49.1 vs. 34.0%, P = 0.008). However, the dose of radiation and number of chemotherapy cycles were comparable between the two groups.

Prognosis and univariate screen of prognostic factors

At a median follow-up of 37.1 months (range, 4.5–153.8 months) for surviving patients, the 5-year OS and PFS for the entire cohort were 58.1 and 49.6%, respectively (Fig. 1). A total of 127 (38.4%) patients died from lymphoma (n = 119), treatment-related complications (n = 3), comorbidities (n = 4), or unknown causes (n = 1). Patients with an AGR < 1.3 had a significantly poorer prognosis compared with patients with an AGR ≥ 1.3. The 5-year OS rates were 44.5 and 65.2% in the low and high AGR groups, respectively (P < 0.001, Fig. 2A). The 5-year PFS rates were 33.1 and 57.4% in the low and high AGR groups, respectively (P < 0.001, Fig. 2B).

Table 3 shows the results from the univariate screening of prognostic factors for OS and PFS. The hazard ratio (HR) of each unit increase in AGR was 0.32 (95% CI [0.17–0.61], P < 0.001) for OS. The HR for the OS of the patients in the low AGR group was 2.09 (95% CI [1.47–2.97], P < 0.001) compared with the high AGR group. Higher mortality risks were observed in patients aged >60 years, those with an ECOG score ≥2, a primary lesion in the EUAT, Ann Arbor stage III-IV, B symptoms, elevated LDH, involvement of regional lymph nodes, and those with two or more extranodal lesions. Each unit increase in CRP, IPI, or NKPI was associated with an increase in mortality, while each unit increase in serum total protein, albumin, white blood cell (WBC) count, or lymphocyte count was associated with a significant decrease in mortality. In terms of treatment modality, a significantly worse prognosis was observed in patients who received chemotherapy alone (compared with RT with or without chemotherapy) or anthracycline-based chemotherapy (compared with ASP-containing regimens). The BMI, neutrophil count, ESR, and serum creatinine had no significant impact on mortality risk.

Multivariate analysis of prognostic factors

The variables that had a significant impact on survival in the univariate analysis were included in the multivariate Cox regression model (Table 4). Serum total protein and albumin were excluded because they were used to calculate AGR. The IPI and NKPI were excluded because they were calculated using several individual factors (i.e., age, ECOG score, stage, B symptoms, LDH, regional lymph node involvement, and extranodal sites ≥2) that were already included in the model. CRP was excluded as it is a constituent of serum globulin and affects the calculation of AGR. We ran two separate Cox regression models using AGR as a binary variable (<1.3 vs. ≥1.3) and as a continuous variable (per unit increase). After adjusting for confounding variables, an AGR < 1.3 remained an independent adverse predictor for both OS (HR: 1.74, 95% CI [1.18–2.56], P = 0.005) and PFS (HR: 1.73, 95% CI [1.21–2.48], P = 0.003). Similarly, each unit increase in AGR was an independent predictor for improved OS (HR: 0.33, 95% CI [0.17–0.66], P = 0.002) and PFS (HR: 0.40, 95% CI [0.22–0.75], P = 0.004) in the other model.

Additional analyses

Further subgroup analyses were performed to determine if the prognostic significance of AGR was merely a function of hypoalbuminemia or early death. Among the 293 patients with normal serum albumin levels (defined as albumin ≥35 g/L), an AGR < 1.3 was associated with significantly worse 5-year OS (53.3 vs. 67.7%, P = 0.005) and PFS (44.3 vs. 59.6%, P = 0.005). Among the 281 patients who survived for more than 6 months after diagnosis, those with an AGR < 1.3 had a significantly worse 5-year OS (58.6 vs. 72.3%, P = 0.017) and PFS (44.0 vs. 63.7%, P = 0.004).

Subgroup analyses were also performed based on treatment modalities. An AGR < 1.3 significantly worsened the 5-year OS (19.9 vs. 36.6%, P = 0.013, Fig. 3A) and PFS (11.6 vs. 31.8%, P = 0.009) in the 107 patients who received chemotherapy alone. Among the 224 patients who received RT, those with an AGR < 1.3 had a slightly worse 5-year OS (63.9 vs. 74.6%, P = 0.070, Fig. 3B) and significantly inferior 5-year PFS (52.6 vs. 66.4%, P = 0.037) compared with those with a higher AGR. In the 193 patients who received anthracycline-based chemotherapy, an AGR < 1.3 was associated with significantly worse 5-year OS (33.0 vs. 58.8%, P < 0.001, Fig. 3C) and PFS (21.9 vs. 48.9%, P < 0.001). Similarly, an AGR < 1.3 was also associated with significantly inferior 5-year OS (63.5 vs. 82.0%, P = 0.013, Fig. 3D) and PFS (51.5 vs. 79.3%, P = 0.008) in patients who received ASP-containing chemotherapy.

As shown in Table 2, the IPI stratified patients into low-risk (scored 0–1, 259 cases, 78.2%), intermediate-risk (scored 2–3, 50 cases, 15.1%), and high-risk groups (scored 4–5, 22 cases, 6.6%). These three groups differed substantially in their prognoses (3-year OS: 70.5, 31.0, and 4.5% for the low-, intermediate-, and high-risk groups, respectively, P < 0.001, Fig. 4A). However, the IPI low-risk group could be further divided into two separate subgroups with significantly different survival (3-year OS: 75.3 vs. 58.4%, P = 0.016) using an AGR < 1.3 or ≥1.3. Accordingly, the entire cohort could be classified into four groups with different survivals using the IPI in combination with AGR (Fig. 4B). AGR failed to distinguish prognostic subsets among patients in the IPI intermediate- or high-risk groups (Fig. S1).

Similarly, the NKPI stratified patients into low-risk (scored 0, 101 cases, 30.5%), intermediate-risk (scored 1–2, 167 cases, 50.5%), and high-risk groups (scored 3–4, 63 cases, 19.0%) with significantly different prognoses (3-year OS: 77.9, 61.6, and 28.1% for the low-, intermediate-, and high-risk groups, respectively; P < 0.001, Fig. 4C). AGR differentiated patients with different prognoses in the NKPI high-risk group (3-year OS: 38.9 vs. 20.1%, P = 0.047) and divided the entire cohort into four risk groups in combination with the NKPI (Fig. 4D). AGR failed to distinguish prognostic subsets among patients in the NKPI low- or intermediate-risk groups (Fig. S1).