This retrospective research study was approved by the Santa Casa de Misericórdia de São Paulo Medical School (SCMSP) and by the A. C. Camargo Cancer Center (ACCC) Ethics Committees. All participants were studied in accordance with the Helsinki Declaration and the Nuremberg Code, also respecting the Brazilian National Health Council (Resolution CNS 466/2012). Informed consent was signed by all patients who were still in follow-up at both institutions (excluding those lost to follow-up and deceased whose family could not be localized).

As this study included two different institutions, the selection of patients was distinct, so as to have a comparable cohort. At the SCMSP, patients were selected using a pathology base, while at the ACCC, the information system was used to select the C85.9, which is the international disease code, and/or the “diffuse large B-cell”. Patients were deemed eligible for the analysis if they were sequentially diagnosed with primary de novo DLBCL, between January 2007 and January 2015, as well as if they presented the complete chart data and pathological review and had started treatment at the respective institution (SCMSP or ACCC).

Patients were excluded if they had primary mediastinal lymphomas, primary CNS lymphoma and secondary CNS infiltration at diagnosis. They were also excluded if they were less than 18 years old; had human deficiency virus (HIV); were treated without rituximab-based chemotherapy or in cases of early death [< 1 month of follow-up, unless among those started on therapy – mostly postmortem diagnosis (Supplement Figure 1A–C)].

Patients were submitted to physical examination, routine chemistry profiles and image procedures, such as computed tomography (CT), magnetic resonance imaging (MRI) and/or fluorodeoxy-glucose-positron emission tomography (FDG-PET). Lymphoma staging was defined according to the Ann Arbor system12 and patients were also categorized by the International Prognostic Index – IPI13 at baseline. Stage, nodal and extranodal sites had been evaluated on patients charts and/or image (when available) and/or image report from radiology. All patients were treated with rituximab-based chemotherapy, mostly R-CHOP (rituximab, cyclophosphamide, doxorubicin, vincristine and prednisone). Variations observed were R-mini-CHOP and R-CHOP (adapting for tolerance and age). However, considering the retrospective nature, they were not individualized for statistical analysis. Response to chemotherapy followed the Revised International Working Group response criteria.14,15

The CNS prophylaxis included at least one cycle of methotrexate (12–15 mg), plus dexamethasone (4 mg) intrathecal with or without intrathecal cytarabine 40 mg. Intravenous high-dose methotrexate (MTX-HD) at 3.0–3.5 g/m2 was also included.

CNS infiltrations were diagnosed by cerebrospinal fluid-CSF (positive cytology and/or immunophenotype), CT, MRI and/or FDG-PET, requested only in symptomatic patients when secondary CNS infiltration was clinically suspected. Confirmed cases were classified as parenchymal, leptomeningeal or both, following the Halderson et al. classification.16

Patient characteristics were summarized by average, medians and ranges for continuous variables and frequencies and percentages for categorical variables. Patient characteristics were compared between patient groups using the Mann-Whitney test for continuous variables and the Chi-square test/Fisher's exact test for categorical variables.

The Kaplan–Meier method was utilized for survival analysis. The OS was calculated from the date of pathological diagnosis to death, or to the last date of follow-up, while the time to CNS relapse [(TTNS) 1-survival probability] was calculated from the date of pathological diagnosis to CNS relapse, or death, or to the last date of follow-up (the last two were censored) [1-survival probability]. The impact on survival of clinical and therapeutic variables [Kaplan–Meier (1-survival probability)] was evaluated by comparing survival curves by means of the log-rank test.

The difference between groups was determined by univariable analysis using the Chi-square test or Fisher's exact test. Moreover, the differences were analyzed considering time to the event using the Kaplan–Meier method and log-rank test.

The Cox model (Supplement Table 1) was used for analyses of the potential risk factor. The univariate analysis followed what was described by the original CNS-IPI11 (IPI factors individually and kidney/adrenal). The multivariate analysis was not performed.

The CNS‐IPI was assessed by graphical comparison between our results of Kaplan–Meier curves (risk stratification groups: low-, intermediate- and high-risk) and those previously defined by the CNS‐IPI11 and by calibration.

The significance level was fixed at 5% for all tests. Statistical analyses were performed using the IBM SPSS Statistics version 23.0 (IBM Corp., Armonk, NY, USA) and the R software version 3.4 (R Foundation for Statistical Computing, Vienna, Austria).

From January 2007 to January 2015, 388 patients were diagnosed with de novo DLBCL (nodal and extra nodal) at the SCMSP. After applying the inclusion and exclusion criteria described in the methods section, 151 patients were available for analysis. In the same period, 403 patients were diagnosed with de novo DLBCL (nodal and extra nodal) at the ACCC. Using the same inclusion and exclusion criteria, 171 patients were selected. Therefore, the total number of available DLBCL patients for analysis was 322. Distributions between the two institutions were of 55% at ACCC vs. 46% at the SCMSP. The detailed consort is available in Supplement Figure 1A–C.

Patient characteristics are summarized in Table 1. Considering the entire cohort, the median follow-up was 60 months (ranging from 1 to 96 months) and the median age was 58 years. There was a similar distribution between males and females (47.8% vs. 52.2%, respectively) and also a similar distribution between localized – including nodal and extranodal – (stages I–II) and advanced disease (stages III–IV), 47.2% vs. 51.2%, respectively. More than half of the patients had a low or low-intermediate IPI (59.6%) and low and intermediate CNS-IPI 0–3 (77.6%). The minority was submitted to CNS prophylaxis (19.9%), and the majority achieved complete response at the end of therapy (73.3%). The statistics are described in Table 1.

A total of 7 patients experienced CNSi (7/322). The median time from diagnosis of DLBCL to CNSi was six months (range 2–52 months), while median time of CNSi to death was 19 days (range 0 days–9.3 months). Four patients presented with brain parenchymal involvement, two patients presented with leptomeningeal disease and one patient presented with both. The CNS-IPI was retrospectively calculated for these patients (Table 2).

As previously described, comparison between groups demonstrated at the baseline that the LDH, number of extranodal sites, IPI, kidney/adrenal were statistically different in patients without and with CNSi (Table 1). Furthermore, regarding the treatment, the absence of complete response was also statistically different. Differences between patients without and with CNSi were also analyzed considering time to the event (Kaplan–Meier curves), with similar results, except for positive bone marrow (BM) biopsy (p = 0.059) (Figure 1).

Figure 1.

Baseline characteristics in patients without and with central nervous system infiltration considering time to the event (Kaplan–Meier curves).

(0.37MB).

A univariate analysis model was also performed in our cohort. Nonetheless, a multivariable analysis was not possible, given the small number of events/relapses (7/322) and the existence of 6 predictors in the original model (age > 60; LDH > normal; ECOG > 1; Stage III–IV; extranodal involvement > 1; kidney and/or adrenal glands) – Supplement Table 1.

A demonstrative table comparing our univariate and multivariate results from the CNS-IPI model (including original model and validation cohort) can be seen in Supplement Table 2.

The CNS-IPI was retrospectively calculated in 286/322 patients. These 36 missing data were mainly due to an inadequately performed ECOG performance and staging. All CNS-IPI evaluations described above were based on 286 patients.

We observed that most of the characteristics of the cohort without CNSi were comparable or even equivalent to the entire cohort. However, there was a statistical difference between patients with and without CNSi in some specific characteristics. Patients with CNSi had a higher IPI, 4–5 (57.1% vs. 8.3%, p < 0.000), were more frequently in the high-risk CNS-IPI group (57.1% vs. 10.2%, p = 0.001) and most of them did not achieve complete response at the end of therapy (42.9% vs. 12.1%, p < 0.000). In addition, patients with CNSi had more frequently 3 or more extranodal sites (42.9% vs. 4.4%, p = 0.006) and specific sites, such as adrenal/kidney involvement (28.6% vs. 4.1%, p = 0.037).

The original CNS-IPI model stratified patients in a three-risk group model, low- [0–1 point], intermediate- [2–3 points] and high-risk [4–6 points]) and demonstrated a 2-year rate of CNS relapse of 0.6%, 3.4% and 10.2%, respectively. In our cohort, we respected the same stratification: low-, intermediate- and high-risk. Similarly, the 2-year rate of CNS relapse was of 0.0%, 0.8% and 13.8%, respectively (Figure 2).

Figure 2.

Validation of the CNS‐IPI assessed by graphical comparison.

(0.2MB).

Calibration of the CNS-IPI model was also performed. First, we calculated the CNSi probability in 2 years for the entire cohort (CNS-IPI original model). Then, we compared the estimated probability with the events observed in our cohort (real probability). Based on the results, a calibration curve was built. This analysis, nevertheless, is limited by the small number of events (7 patients with CNSi) in Figure 3.

Figure 3.

CNS-IPI calibration curve.

(0.08MB).

The majority of patients without CNSi were not submitted to any modality of CNS prophylaxis (80.5%), while 3 patients with CNSi were submitted to some modality of CNS prophylaxis (MTX-HD- 2 patients; Intrathecal + MTX-HD - 1 patient), p < 0.000; Table 1).

The CNS prophylaxis was administered according to a physician's decision, due to the inexistence of an institutional protocol at SCMSP and at ACCC. We sought to understand the rationale to perform CNS prophylaxis by calculating the CNS-IPI retrospectively (Table 3). We observed that the majority of patients considered high-risk CNS-IPI were not submitted to prophylaxis (61.1% vs. 13.9%, p = 0.015).

Comparison between the types of CNS prophylaxis applied would be inadequate in our study, considering there were only 7 events (2.17% incidence of CNSi).

The median OS for the entire cohort was not reached and was estimated at 88% in 5 years (average: 8.8 years [95% CI: 8.3–9.3]). OS in ACCC cohort was estimated at 86% in 5 years (average: 8.6 years [95% I: 8.1–9.1]) and OS in the SCMSP cohort was estimated at 74% in 5 years (average: 8.1 years [95% CI: 7.3–8.9]). There was a statistical survival difference between the cohorts of patients without or with CNSi. The median OS for patients without CNSi was not reached (average: 9.0 years [95%CI: 8.5–9.5]), while the median OS for patient with CNSi was 8 months (95% CI: 2.8–13.1), log-rank p < 0.000 (Figure 4).

Figure 4.

Overall survival. Patients without and with CNSi.

(0.1MB).