One hundred and fifty six retrospective cases diagnosed with histologically proven DLBCL at the Canberra Hospital from 1986-2005, on whom staging BM biopsies had been performed, were identified for the purpose of the study. After approval was obtained from the Australian Capital Territory (ACT) Human Research Ethics Committee, clinical information on patients was collected from the medical records department at The Canberra Hospital.

The average age of the patient cohort was 61 years (range 20-87 years), and the male to female ratio was 1.5:1. Baseline staging data using routine staging procedures [computed tomography (CT) scan, gallium/positron emission tomography (PET) scan and histological examination of BM] was available in 150 patients. Thirty nine (26%), 35 (23%), 45 (30%), and 31 (21%) were found to have stage I, stage II, stage III and stage IV disease respectively. Baseline assessment of IPI was possible in 148 patients. Thirty seven (25%) had an IPI of ≤ 1 of which 14 (9.5%) had an IPI of 0, and 23 (15.5%) an IPI of 1. IPIs of 2 and 3 were noted in 36 (24.3%) and 46 (31.1%) cases respectively. Twenty nine (19.6%) cases had an IPI of ≥ 4 of which 22 (14.9%) and 7 (4.7%) and an IPI of 4 and 5 respectively. The mean baseline IPI of the patient cohort was 2.41 with a standard deviation of 1.3.

Treatment data showed that of 152 patients on whom data was available, most were treated with anthracycline based regimens. One hundred and twenty nine patients (82.7%) were treated with Cyclophosphamide, Doxorubicin, Vincristine and Prednisolone (CHOP) 7 or variations of CHOP chemotherapy protocols 89. Two patients were treated with Ifosphamide, Carboplatin and Etoposide (ICE) 10, 5 patients with Prednisolone, Etoposide and Novantrone (PEN) 11, 1 with Etoposide, Vincristine, Doxorubicin, Cyclophosphamide and Prednisolone (EPOCH) 12, 2 with Hyper-CVAD 13 comprising of hyperfractionated Cyclophosphamide, Vincristine, Doxorubicin and Prednisolone courses alternating with courses of Methotrexate and Cytarabine, 3 with Trans-Tasman Radiation Oncology group (TROG) protocol 14 and 1 with Methotrexate, Doxorubicin, Cyclophosphamide, Vincristine, Prednisolone and Bleomycin (MACOP-B) 15. Nine patients were treated with palliative intent with steroids alone or in combination with non-anthracycline based drugs. Treatment details were not available in 2 patients and 2 were lost to follow-up. Thirty six patients (22.2%) received Rituximab. The median overall survival of the entire patient group was 6 years (95% confidence interval [CI]: 3.8, 8.4 years).

BM biopsies are performed as a routine assessment for all cases with DLBCL at first diagnosis in our institution. All trephines are fixed in buffered formalin and acetic acid for 24 hours and then decalcified using 5% nitric acid. Samples are then embedded in paraffin and sections stained with Haematoxylin and Eosin (H&E), giemsa stain and silver impregnation for reticulin. Archived H&E, giemsa and reticulin preparations on the trephine biopsy were retrieved for review. The mean trephine length for the patient cohort was similar to our previous reports, 17.6 mm with a range of 8-36 mm and the mean number of levels on H&E sections were 3.7 (range 1-8).

Two haematologists reviewed all slides blindly, and discrepant cases (n = 20) were resolved by consensus. Standardised criteria were used to classify trephine biopsy samples as positive, negative or indeterminate 16.

Raw immunophenotypic data on all bone marrow biopsies was retrieved from laboratory records and re-analysed.

Multiparametric flow cytometric analysis is performed in our laboratory, with marrow cells immunophenotypically labelled by direct four-colour immunofluorescence using a panel of antibodies (CD45, CD19, CD20, CD22, CD10, HLA-DR, Kappa, Lambda, CD2, CD3, CD5, CD7; Becton-Dickinson).

Red cells are lysed by incubation with ammonium chloride, and cells washed in phosphate buffered saline after centrifugation. A cell-suspension of 1 × 10⁶ cells per tube is incubated with the monoclonal antibody at room temperature, then washed and resuspended in a solution of phosphate buffered saline and foetal calf serum. Isotypic controls used are IgG1 and IgG2.

Data acquisition is on a Becton-Dickinson flow cytometer with a minimum of 2000 lymphocytes counted in each sample. Bright CD45 fluorescent staining and intermediate side scatter are employed as the primary gating strategies to identify the lymphocyte population, and further gating is performed as required based on cell size or using back gating on CD19 positive events.

Previously archived raw data were reanalysed, including blinded re-determination of the population of lymphocytes to be gated. Positive results on flow cytometry were defined as light chain clonal restriction with a kappa: lambda ratio of >3:1 or <0.3:117. Predominance of B-cells in the gated population alone without light chain restriction was not considered as a positive result.

Immunohistochemical analysis was performed on a Ventana Benchmark NexES machine. Sections from archived formalin-fixed decalcified paraffin-embedded (FFDPE) trephine biopsies were immunostained using the streptovidin-biotin method. The following monoclonal antibodies were used: CD3 [Dako clone CD3, 1:200 dilution], CD45RO [Novacastra clone UCLH-1, 1:1000 dilution], CD20 [Zymed clone L26, 1:50 dilution], CD79a [Dako clone JCB117, 1:500 dilution], Kappa [Novacastra clone kp-53, 1:750 dilution], and Lambda [Novacastra clone Hp-6054, 1:750 dilution]. All antibodies are validated and routinely used in our laboratory. CD20 and CD3 are reported to be sensitive at assigning lineage in diffuse aggressive NHL 18 and CD79a and CD45RO were selected over others owing to familiarity and to maintain consistency. These are the antibodies used for diagnostic tissue sections in our laboratory. Heat retrieval was used for all antibodies and tonsillar tissue was used as a positive control. A standardised system of reporting was adopted and was followed for all stains by two pathologists blinded to previous assessment on histology.

Features used to define involvement on immunohistochemistry reflected standardised histology criteria. The presence of B-cell aggregates was classified as abnormal or malignant when there were large numbers of aggregates, the aggregates were large-sized, or contained disproportionate numbers of larger cells. Controls (six morphologically normal marrows) were used to create a visual impression of normal amounts of background T and B-cells. Scattered small or large B-cells were classified positive only when the numbers were substantially increased as compared to controls. A conservative approach was adopted to avoid false positives. Discrepancies between the two pathologists were resolved consensually.

Samples for molecular studies were obtained from formalin-fixed decalcified paraffin-embedded (FFDPE) trephine sections. DNA extraction was performed manually using the Roche High Pure PCR Template Preparation Kit from two 10-micron FFDPE trephine sections according to the manufacturer's instructions. To verify the integrity of the DNA extracted from the paraffin sections, and to validate results, all samples were amplified with the control master mix provided in the Immunoglobulin heavy chain (IgH) gene clonality kit from Invivo Scribe Technologies based on the BIOMED2 protocols (IgH Gene Clonality Assay - Gel Detection; InVivo Scribe Teachnologies, USA). This is a multiplex PCR that targets multiple genes and generates a series of amplicons 100, 200, 300, 400 and 600 base pairs (bp) in length.

IgH gene rearrangement analysis was performed on all cases, targeting the conserved framework regions (FR) FR1 [IGH_A: V_HFR1-J_H] and FR3 [IGH_C: V_HFR3-J_H] using the Invivo scribe kit based on the BIOMED2 protocols 19. Only FR1 and FR3 were analysed owing to limited amounts of DNA and based on reports from other groups 20. This was combined with light chain gene rearrangement analysis and included two reactions targeting Ig Kappa (IgK) variable and joining regions [IGHK_A: V_k-J_k;] and IgK variable and intragenic regions [IGK_B: V_k-K_de]. The PCR reactions consisted of 45 μL of the FR1, FR3 or IGK master mix solution, 2.5 units of Amplitaq Gold (Applied Biosystems, USA) and 5 μl of template DNA (with an average template DNA concentration of 300-400 ng/μl). Thermo cycling was performed according to the kit protocol with no modifications on a Perkin Elmer 9600 thermocycler. Controls consisted of a positive DNA control, negative extraction control and negative PCR control. Water was used as a negative control in both cases.

Non-denaturing polyacrylamide gel electrophoresis was used to resolve the FR1 and FR3 PCR products. 25 uL of PCR product was loaded onto a 6% polyacrylamide gel and 250 V applied for 1.25 hours for FR1 and 1.5 hours for FR3 reactions. After electrophoresis, the gels were stained with ethidium bromide and visualised under UV light.

For the IGK reactions, PCR products were denatured at 94°C for five minutes and subsequently cooled at 4°C for 60 minutes to induce duplex formation. Non-denaturing polyacrylamide gel electrophoresis was used to resolve the PCR products. 25 uL of PCR product was loaded onto a 6% polyacrylamide gel and 250 V applied for 1.5 hours each for both reactions.

FR1, FR3 and IgK gene rearrangements were reported as clonal, polyclonal or not detected. The expected sizes of the PCR products were 310-360 bp for FR1 and 100-170 bp for FR3 which together are estimated to account for approximately 70% of all rearrangements 20 IGK PCR products were expected to be in the following ranges: 120-160 bp, 190-210 bp, 260-300 bp for IgKA and 210-250, 270-300, 350-390 bp for IgKB.

Survival data were recorded for each patient. Besides descriptive analysis, Kaplan Meier curves were created with cumulative survival as the outcome. Forward stepwise multivariate Cox regression analysis using the likelihood ratio method was used to establish a comparison between baseline IPI and two revised IPI models. The first model (rIPI1) was based on routine use of immunophenotyping alone (flow cytometry and immunohistochemistry) and the second (rIPI2) was based on routine use of immunophenotyping and molecular results. A probability of 0.05 was used as the entry criterion and 0.1 was considered for removal. Patients treated with palliative intent were excluded from all survival analyses. All analyses were performed using the software programme Statistical Package for Social Sciences (SPSS) version 14.0.

Of the 156 cases on which bone marrow histology slides were available, 24 were positive on routine histology. Six cases were reported as indeterminate using Cheson criteria, and agreed upon as being positive for involvement after consensual review. H&E stains showed no evidence of involvement in 126 cases.

Flow cytometry data was evaluable in 152 cases, of which 27 (17.3%) cases were noted to be positive for involvement using standardised light chain ratios. Ten of these 27 cases were also positive on routine histology.

Immunohistochemistry using T and B-cell markers showed involvement in 43 cases of 154 available cases of which involvement on routine histology was noted in 25/42 cases. One case was not comparable due to loss of H&E slides. Flow cytometry and immunohistochemistry each detected histologically inapparent involvement in 17 cases (11%).

Amplification was obtained in 133/155 cases (84.7%) with amplification at 96 base pairs (BP), 200 bp, 300 bp, 400 bp and 600 bp noted in 125 (79.6%), 74 (47.1%), 32 (20.4%), 25 (15.9%) and 18 (11.5%) cases respectively. Forty one cases of 155 evaluable ones were positive on immunoglobulin heavy and light chain gene analysis. Three showed no amplification on amplification controls. Thirty four cases were positive on light chain analysis with all showing a clonal band with kappa A; three cases also showed clonal reactions with kappa B. Overall, 19 cases were positive on heavy chain analysis (FR3: 18 cases, and FR1: 4 cases) Of these, three cases were positive on both reactions. Overlap with light chain analysis is shown in table 1. Overall, 12/41 cases were positive and 29 negative on routine histology.

To establish tumour origin, DNA was extracted from 17 available primary FFDPE tissue blocks and gene rearrangement analysis performed. Comparable clonal bands could be identified in only 10 cases. Of these, 2 were positive on routine histology.

Thus, using stringent criteria to account for false positivity, routine molecular staging on FFDPE trephine biopsy tissue yielded positive results in eight (5.1%) histologically negative cases.

Thirty cases were upstaged using immunophenotyping alone with 6 cases upstaged from stage I to IV, 12 from stage II to IV, and 12 from stage III to IV. When molecular results were added, two additional cases were upstaged from stage I, 2 from stage II and 4 from stage III.

Two new revised IPI (rIPI) models were computed for all cases. The first (rIPI1) was based on immunophenotyping results alone i.e. flow cytometry and immunohistochemistry and the second (rIPI2) on immunophenotyping and molecular results. Changes to the IPI essentially occurred when stage of disease was upgraded from I or II to stage IV diseases. Of 148 cases where IPI and rIPI were assessable, three cases were upgraded from IPI 0 to a rIPI1 of 1, 4 cases of IPI 1 changed to a rIPI of 2, 5 cases of IPI 2 were upgraded and 6 cases of IPI 3. No changes were noted in IPI 4-5 group. Overall, 18 patients had a change in their IPI. Of these, three changed their IPI from 0 to 1, which was not apparent when only four prognostic categories were considered. Incorporating molecular results, rIPI2 was found to be upgraded in 6, 7 and 6 cases of IPI 0-1, 2 and 3 respectively.

Kaplan Meier curves were created to assess the impact of baseline IPI and the two new revised IPI models rIPI1 and rIPI2 on overall survival. Figure 1 shows the cumulative survival of the four IPI categories using a baseline IPI model, a revised model using immunophenotyping (rIPI1) and a revised model incorporating molecular studies and immunophenotyping (rIPI2) respectively. All three models were statistically significant with p values of < 0.0001.

Using a multivariate forward stepwise (likelihood ratio method) Cox regression, the three IPI models were competitively considered for their contribution to predicting survival. The score tests before inclusion into the model were: Baseline IPI: 28.5 (df = 3, P < 0.001), rIPI1: 31.2 (df = 3, p < 0.001) and rIPI2: 27.9 (df = 3, p < 0.001). The revised rIPI1 model was then entered into the regression model. The hazard ratios of dying relative to the 0/1 IPI prognostic categories were: rIPI1 category 2 = 2.0 [95% CI, 0.61, 6.76, p = 0.248], rIPI1 category 3 = 7.6 [95% CI, 2.61, 22.36, p < 0.0001], rIPI1 category 4/5 = 8.9 [95% CI, 2.94, 27.17, p < 0.0001]. The baseline IPI model and rIPI2 models were excluded from the regression model because they did not further contribute to explaining survival [p = 0.5, p = 0.6 respectively].

Table 2 provides a summary of the relative performances of the three IPI models in predicting survival. It can be seen that rIPI1 provides the best differentiation between the IPI categories and the largest point estimate hazard ratios.

To study the effect of treatment with Rituximab on survival, this too was considered, but not found to contribute significantly to the Cox regression model (p:0.96).