Introduction

Diffuse large B cell lymphoma (DLBCL) is the most common lymphoid malignancy in adults, accounting for approximately 32.5% of non-Hodgkin lymphomas diagnosed annually [1]. In spite of the inspiring achievements in the treatment of DLBCL since adding rituximab (R) to CHOP (cyclophosphamide, doxorubicin, vincristine, and prednisone) [2], about 30% of patients experience disease relapse or refractory to induction therapy. Thus, it is crucial to seek an efficient method that could early identify high-risk DLBCL patients who may benefit from a prompt therapy change.

Interim response assessment is performed to identify patients whose disease has not responded to or has progressed on induction therapy. The Lugano Response Criteria has been widely used for interim response assessment for DLBCL [3]. Based on it, patients with interim reduction percentage of the sum of the products of the greatest perpendicular diameters for multiple lesions (iΔSPD) between baseline and interim contrast-enhanced (CE) computed tomographic (CT) scan less than 50% are regarded as non-sensitive to induction therapy and may benefit from a prompt therapy change. However, the prognosis of patients with iΔSPD ≥ 50% was not always promising either. Actually, there is no consensus on the criteria for early therapy change during interim response assessment yet, which led us to investigate whether there was an iΔSPD cutoff value that could discriminate DLBCL patients with poor prognosis. To the best of our knowledge, no data has been published until now.

Materials and methods

Ethics

The study was approved by the Ethics Committee of the First Affiliated Hospital of Nanjing Medical University. Informed consent and permissions were obtained from all the subjects involved in this study at the time of first DLBCL admission.

Patients

Two hundred and sixty-five subjects with newly diagnosed DLBCL between August 2009 and November 2017 from the First Affiliated Hospital of Nanjing Medical University were enrolled in this single-center retrospective study if both baseline and interim CECT scans were performed and no malignancy other than DLBCL was present at the time of examination. Patients diagnosed with high-grade B cell lymphoma, primary mediastinal large B cell lymphoma, and primary central nervous system lymphoma were excluded. All patients were treated with R-CHOP.

Data collection

Baseline clinical characteristics and laboratorial data including age, gender, serum lactate dehydrogenase (LDH) level, Ann Arbor stage, extra-nodal site involvement, Eastern Cooperative Oncology Group (ECOG) score, B symptoms, the Hans classification of germinal center B cell–like (GCB) or non-GCB, and Ki-67 of the immunohistochemistry were collected from medical records and hospital-based laboratory service.

CECT imaging technique

Whole body CECT inclusive of the neck, thorax, and abdomen/pelvis was performed using a 64-row multidetector CT scanner (Sensation 64; Siemens). Intravenous non-ionic iodinated contrast (ioversol, 320 mgI/mL, HengRui Medicine Ltd.) was administered via the antecubital vein by a double-cylinder high-pressure syringe at a flow rate of 2.5–3.0 mL/s. A total of 80–120 mL of contrast agent was injected, followed by 40 mL of normal saline at a flow rate of 3.0 mL/s. Each patient was scanned using the same parameters: 120 kV voltage, 180 mAs current, pitch of 0.8, 256 × 256 matrix, and slice thickness of 5 mm. Arterial-phase thoracic CT was done in deep inhalation with a delay of 35 s, followed by portal-phase abdominal CT with a delay of 65–70 s scanning from the top of the liver to the thighs. Ultimately, scanning from the bottom of the skull to the supraclavicular region with a delay of 90 s was done. A baseline whole body CECT was performed within 28 days before the start of treatment. An interim whole body CECT was performed after 3 courses of chemotherapy.

CECT imaging analysis

Two radiologists with more than 10 years of experience who were unaware of clinical findings and outcomes evaluated all CT images independently on a picture archiving and communication system (PACS; Medvis), and recorded the number, size, and location of nodal and extra-nodal lesions. Then the two radiologists selected target lesions strictly according to the Lugano Response Criteria [3] together. Any discrepancies in interpretation between two radiologists were resolved by consensus. Lymph nodes greater than 15 mm in longest diameter (LDi) and extra-nodal lesions with LDi greater than 10 mm could be considered measurable lesions and used to calculate SPD. At most six of the largest target nodes, nodal masses or other lymphomatous lesions that are measurable in LDi and shortest diameter were selected from disparate body regions representative of the patient’s overall disease burden and mediastinal and retroperitoneal disease sites were included, if involved. Measurable extra-nodal diseases such as hepatic nodules were included in the six target lesions. All the remaining lesions were considered non-measured lesions.

Statistical analyses

Cutoff value was determined by use of X-tile [4], a software tool that can divide tumors into subgroups based on the expression level of clinical parameters. By inputting the survival time, survival status, and expression level of the clinical parameters studied, the software can automatically give the optimal cutoff value. Receiver operating characteristic (ROC) curve analysis and Cox proportional hazards regression analysis were further used to validate the optimal cutoff value. Statistical analyses were performed using the SPSS software for Windows (version, 24.0) and MedCalc for Windows, version 15.2.2.0 (MedCalc Software). Progression-free survival (PFS) was defined as interval from diagnosis to first progression or relapse, and overall survival (OS) was defined as interval from diagnosis to death or last follow-up. Survival curves were constructed by the Kaplan–Meier method, and log-rank test was used to test for significant differences. To address the multiple comparisons, a Bonferroni correction was applied. ROC curve analysis and the corresponding areas under the curve (AUCs) were used to assess the predictive accuracy of the cutoff values. Prognostic significances of three iΔSPD cutoff values (50%, 80%, and 100%) and clinical variables (age, gender, International Prognostic Index (IPI) score, LDH level, Ann Arbor stage, extra-nodal sites, ECOG score, B symptoms, the Hans classification, and Ki-67) were assessed by univariate and multivariate Cox proportional hazards regression analysis. The results were evaluated in 95% confidence interval and at a significance level of two-sided p < 0.05.

Results

Patients’ characteristics

Clinical characteristics of the 265 patients were outlined in Table 1. The median age at diagnosis was 58 years (range, 18–88 years), and 111 (41.9%) of them were more than 60 years old. One hundred and thirty-nine patients (52.5%) were male; 100 (37.7%) had an IPI score ≥ 3; 112 (42.3%) had elevated LDH; 179 (67.5%) were at stage III or IV according to the Ann Arbor classification; 78 (29.4%) had more than one extra-nodal sites of involvement; 50 (18.9%) had an ECOG score ≥ 2; 100 (37.7%) had B symptoms. One hundred and seventy-two patients (64.9%) were non-GCB type according to the Hans classification and 10 (3.8%) were unavailable. Ki-67 ≥ 80% were observed in 115 (43.4%) patients and 24 (9.1%) were unavailable.

Table 1 Baseline characteristics of all patients (N = 265)
Full size table

Selection of the optimal iΔSPD cutoff value

The follow-up results of all 265 patients enrolled in the study were at the end of Nov. 2018 with a median follow-up of 47 months (range, 18 − 111 months). Ninety-eight (37.0%) patients progressed or relapsed and 71 (26.8%) died. In the whole cohort, the estimated 3- and 5-year PFS were 65.2% ± 2.9% and 59.9% ± 3.4%, respectively. The estimated 3- and 5-year OS were 75.2% ± 2.7% and 71.8% ± 3.0%, respectively.

According to the analysis by X-tile, the cutoff value of iΔSPD that yielded the highest difference in OS was 80%. The prognostic values of three cutoff values (100%, 80%, and 50%) were compared using the ROC curves. AUCs of 100%, 80%, and 50% cutoff values were 0.667 [0.606, 0.723] (p < 0.001), 0.737 [0.680, 0.789] (p < 0.001), and 0.641 [0.580, 0.699] (p < 0.001), respectively, when OS was considered the entire follow-up period (Fig. 1).

Fig. 1
figure 1

ROC curve analysis for three iΔSPD cutoff values

Full size image

Sensitivity, specificity, Youden’s index, positive predictive value (PPV), and negative predictive value (NPV) were also used to compare the prognostic values of three cutoffs (Table S1). PPV and NPV of 80% cutoff value were 78.69% and 75.49% for PFS, but 67.21% and 85.29% for OS, respectively. For PFS, the 100% cutoff value had the maximal sensitivity (82.65%), but the minimal specificity (50.90%) and the intermediate Youden index (0.3355). Fifty percent cutoff value had the maximal specificity (95.21%), but the minimal sensitivity and Youden index (23.50% and 0.1871, respectively). Although the sensitivity and specificity of the 80% cutoff value were intermediate (48.98% and 92.22, respectively), the Youden index (0.4120) was maximal. For OS, which was in line with the result for PFS, the sensitivity and specificity of the 80% cutoff were intermediate (57.75% and 86.69%, respectively) while the Youden index (0.4744) was maximal.

Univariate Cox regression analysis showed that all baseline clinical characteristics except gender were significantly associated with PFS, and all baseline clinical characteristics except gender and number of extra-nodal sites were significantly associated with OS. However, in multivariate Cox regression analysis, only advanced stage and iΔSPD < 80% remained significantly associated with PFS (HR 3.215, 95% CI 1.515–6.825, p = 0.002; HR 4.137, 95% CI 2.321–7.374, p < 0.001; respectively) and OS (HR 4.993, 95% CI 1.759–14.172, p = 0.003; HR 4.988, 95% CI 2.486–10.006, p < 0.001; respectively). Non-GCB type remained significantly associated with OS (HR 2.316, 95% CI 1.136–4.724, p = 0.021) but not with PFS. Notably, iΔSPD < 50% and iΔSPD = 100% were not significant for both PFS and OS anymore (Table 2).

Table 2 Cox regression analysis for PFS and OS of all patients (N = 265)
Full size table

Combining the results of ROC curve analysis and Cox regression analysis, we selected 80% as the optimal cutoff value to compute for further analysis.

Survival analysis of the whole cohort

The whole cohort was divided into 4 groups according to three iΔSPD cutoff values: iΔSPD = 100% group (102 (38.5%)), 80% ≤ iΔSPD < 100% group (102 (38.5%)), 50% ≤ iΔSPD < 80% group (30 (11.3%)), and iΔSPD < 50% group (31 (11.7%)). The iΔSPD = 100% group had significantly superior PFS than the 80% ≤ iΔSPD < 100% group (Bonferroni-corrected p = 0.027), while no statistical significance (Bonferroni-corrected p = 0.21) for OS was observed between them. There were no statistical differences for PFS and OS between the 50% ≤ iΔSPD < 80% group and iΔSPD < 50% groups Bonferroni-corrected (p = 0.741, Bonferroni-corrected p = 0.267, respectively). The 50% ≤ iΔSPD < 80% group had significantly inferior PFS and OS than the 80% ≤ iΔSPD < 100% group (both Bonferroni-corrected p < 0.001) (Fig. 2a, b).

Fig. 2
figure 2

a Kaplan–Meier analyses of progression-free survival according to three iΔSPD cutoff values: (a) p value between the iΔSPD = 100% group and the 80% ≤ iΔSPD < 100% group; (b) p value between the 80% ≤ iΔSPD < 100% group and the 50% ≤ iΔSPD < 80% group; (c) p value between the 50% ≤ iΔSPD < 80% group and the iΔSPD < 50% group. b Kaplan–Meier analyses of overall survival according to three iΔSPD cutoff values: (a) p value between the iΔSPD = 100% group and the 80% ≤ iΔSPD < 100% group; (b) p value between the 80% ≤ iΔSPD < 100% group and the 50% ≤ iΔSPD < 80% group; (c) p value between the 50% ≤ iΔSPD < 80% group and the iΔSPD < 50% group. Log-rank p values are indicated. p values have been adjusted after Bonferroni’s correction

Full size image

According to the optimal iΔSPD cutoff value 80%, 204 (77.0%) patients were iΔSPD ≥ 80% and 61 (23.0%) were iΔSPD < 80%. Not only did PFS estimation differ between patients with iΔSPD ≥ 80% and iΔSPD < 80% (hazard ratio (HR) 4.137, 95% confidence interval (CI) 2.321–7.374, p < 0.001), but also worse OS was observed in patients with iΔSPD < 80% than in those with iΔSPD ≥ 80% (HR 4.988, 95% CI 2.486–10.006, p < 0.001); the 3- and 5-year PFS were 78.4% ± 2.9% and 72.4% ± 3.6% in patients with iΔSPD ≥ 80% vs. 19.7% ± 5.3% and 15.7% ± 5.5% in patients with iΔSPD < 80%, respectively. The 3- and 5-year OS were also higher for patients with iΔSPD ≥ 80% than for those with iΔSPD < 80%, with 87.4% ± 2.4% and 83.8% ± 2.9% vs. 34.4% ± 6.3% and 31.8% ± 6.4% (Fig. 3a, b).

Fig. 3
figure 3

Kaplan–Meier analyses of progression-free survival (a) and overall survival (b) according to the optimal iΔSPD cutoff value 80%. Log-rank p values are indicated

Full size image

Discussion

Interim response assessment is performed to identify patients who can benefit from a prompt therapy change. However, there is no consensus on the criteria for early therapy change during interim response assessment yet. Although end-of-treatment assessment is more accurate with PET-CT than CECT for patients with DLBCL [5, 6], different from the Hodgkin lymphoma, the predictive value of interim positron emission tomography/CT (iPET/CT) in DLBCL remains a matter of significant debate [7,8,9,10,11,12,13,14,15,16]. Some studies have revealed that interim positron emission tomography/computed tomography (iPET/CT) has encouraging prognostic value [7,8,9,10], while some studies hold the opposite opinion [11,12,13,14,15]. And the optimal timing for iPET/CT scanning still needs to be further studied [17,18,19,20]. Besides, iPET/CT scans are associated with an increased false positive rate and some patients turn out to achieve a favorable long-term outcome despite a positive iPET/CT. In a prospective study that investigated the role of iPET/CT in DLBCL patients (after 4 cycles of R-CHOP), only 5 of 37 (13.5%) patients with a positive iPET/CT had a positive biopsy result; PFS outcome in patients who were iPET/CT-positive and biopsy-negative was similar to that in patients with a negative iPET/CT [21]. Therefore, NCCN guidelines still have reservations for patients with a positive iPET/CT, and iPET/CT imaging is not yet recommended to be used to guide changes in therapy at present. If treatment modifications are considered based on iPET/CT results, a repeat biopsy of residual masses should be strongly considered to confirm PET positivity prior to additional therapy. In addition, PET/CT is expensive and technically challenging and is not standardized for clinical use in developing countries. Thus, we tried to investigate whether there was an iΔSPD cutoff value that could discriminate patients with DLBCL who may benefit from an early therapy change just using interim CECT. Actually, this is the first time to date to find out an optimal iΔSPD cutoff value for patients with DLBCL in interim response assessment.

In our study, we found that according to the analysis by X-tile, the cutoff value of iΔSPD that yielded the highest difference in OS was 80%. Then, we compared the prognostic value of it with 50% and 100% through ROC curve analysis and Cox regression analysis. Eighty percent cutoff value showed the maximal AUC and Youden index. For PFS, the sensitivity, specificity, PPV, and NPV of 80% cutoff value were 48.98%, 92.22%, 78.69%, and 75.49%, respectively; for OS, these were 57.75%, 86.69%, 67.21%, and 85.29%, respectively. Burggraaff et al [22] performed a systematic review and meta-analysis from 18 studies comprising 2255 DLBCL patients treated with first-line immunochemotherapy regimens and showed NPV, PPV, sensitivity, and specificity of iPET/CT for progression of 64–95%, 20–74%, 33–87%, and 49–94%, respectively, which was in line with our results. This means our approach using interim CECT on the basis of 80% cutoff value is comparable, at least not inferior to iPET/CT. Cox regression analysis also revealed that the iΔSPD < 80% cutoff value could independently predict an inferior PFS and OS while neither iΔSPD < 50% nor iΔSPD = 100% could. Given its superiority, we selected 80% as the optimal cutoff value to identify patients with DLBCL who need a prompt therapy change.

In conclusion, iΔSPD with the cutoff value 80% is an independent predictor of PFS and OS for patients with DLBCL. Treatment modification is recommended for patients with iΔSPD < 80% at interim response assessment. However, whether such modification will improve survival should be proven in further prospective, multicenter studies.