A new prognostic model is proposed for follicular lymphoma

Early disease progression after first-line therapy is associated with poor outcomes in patients with advanced-stage follicular lymphoma (FL). The ability to identify patients who are at risk of early progression is important because these high-risk patients may benefit from alternative treatments, however, existing clinical prognostic models for FL are rarely used in routine practice. This may in part be attributed to insufficient data on the performance of these models for patients receiving standard of care treatment regimens.

In this study published in the American Journal of Hematology, Farheen Mir et al. sought to establish a new model, the FL Evaluation Index (FLEX), which would incorporate clinical variables that are easy to measure and have applicability across different treatment regimens currently being used as standard of care.¹

Methods

The FLEX model was established and evaluated using a training cohort of 1,202 patients from the randomized phase III GALLIUM trial, comparing frontline obinutuzumab and rituximab in combination with chemotherapy in previously untreated FL patients (NCT01332968, previously described on the Lymphoma Hub).

The model was validated using an independent cohort of 342 patients from the SABRINA trial, a randomized phase III study comparing subcutaneous and intravenous delivery of rituximab in patients with untreated FL (NCT01200758, also reported previously on the Lymphoma Hub)

• In all, 17 clinical variables were selected and assessed for inclusion in the model based on hazard ratios (HR) and 95% confidence intervals (CI) for progression free survival (PFS).
• Performance was evaluated by assessing the impact of categorisation on PFS and overall survival (OS). Comparisons were made to established prognostic models (FL International Prognostic Index [FLIPI], FLIPI-2, and PRIMA-Prognostic Index [PRIMA-PI]).
• The impact of different treatment regimens and accuracy of the model for predicting POD24, defined as progressive disease or death due to disease within 24 months of randomization, were also assessed.

Results

The nine clinical variables included in the final FLEX model are shown in Table 1. Novel variables included in FLEX, but not FLIPI or FLIPI-2, were the sum of the products of lesion diameters, gender, and natural killer cell count. Variables which were not retained were age, Ann Arbor stage, B symptoms, bone marrow involvement, bulky site, and geographic area.

Clinical variables in the final model were given equal weight and a score (FLEX score) was assigned by summating the number of risk factors. Patients with 0–2 risk factors were categorised as low risk, whereas patients with 3–9 factors were deemed high risk.

Of note, the cyclophosphamide, vincristine, and prednisone (CVP) chemotherapy variable was excluded to increase general applicability of the model, and positron emission tomography (PET) imaging variables were not included because PET results were only available for around half of the patients in the GALLIUM trial.

Table 1. Variables included in the FLEX score¹

CI, confidence interval; CT, computed tomography; ECOG PS, Eastern Cooperative Oncology Group performance status; FLEX, Follicular Lymphoma Evaluation Index; HR, hazard ratio; LDH, lactate dehydrogenase; NKCC, natural killer cell count; SPD, sum of the products of lesion diameters; ULN, upper limit of normal.
Variable	HR (95% CI)	p value
Sex, male	1.65 (1.3–0.29)	0.0000336
SPD in 4th quartile > 9,320 mm2 on CT scan	1.62 (1.13–2.31)	0.00851
Histology grade 3a	1.48 (1.1–1.98)	0.00849
Extranodal involvement > 2	1.46 (1.04–2.05)	0.0292
ECOG PS at baseline > 1	1.44 (0.84–2.49)	0.186
Hemoglobin < 12g/dL	1.35 (1.01–1.81)	0.0430
β2-microglobulin > ULN	1.29 (0.98–1.69)	0.0690
NKCC < 100/µL	1.26 (0.88–1.79)	0.205
LDH > ULN	1.25 (0.97–1.61)	0.0884

Model performance for the training cohort

• Overall, 36% of 1,004 patients in the GALLIUM trial were categorised as high-risk.
• Superior PFS and OS were observed for low-risk patients (2-year PFS, 91%; 3-year PFS, 86%; 2-year OS, 98%; 3-year OS, 97%) compared with high-risk patients (2-year PFS, 74%; 3-year PFS, 68%; 2-year OS, 90%; 3-year OS, 87%).
• Differences in 2-year and 3-year PFS and OS between low-risk and high-risk groups were greater for FLEX than for FLIPI, FLIPI-2, or PRIMA-PI (Figure 1). 

FLEX, Follicular lymphoma Evaluation Index; FLIPI, Follicular Lymphoma International Prognostic Index; OS, overall survival; PRIMA-PI, PRIMA-Prognostic Index; PFS, progression-free survival. 

The performance of the FLEX model for PFS was consistent for the different chemotherapy backbone subgroups (Table 2).

• The greatest difference between low and high-risk groups was observed for patients treated with bendamustine (low risk 2-year PFS 92.4% and 3-year PFS 88.8%; high risk 2-year PFS 74.5% and 3-year PFS 68.4%).
• Excluding CVP treatment had little effect on the model.

Table 2. Progression free survival according to chemotherapy backbone.¹

CHOP, cyclophosphamide, doxorubicin, vincristine, and prednisone; CVP, cyclophosphamide, vincristine, and prednisone; PFS, progression-free survival.
Treatment	Low-risk group		High-risk group
2-year PFS, %	3-year PFS, %	2-year PFS, %	3-year PFS, %
Bendamustine	92.4	88.8	74.5	68.4
CHOP	89.7	84.0	73.2	66.1
CVP	84.0	71.2	76.8	68.2

Predicting disease progression

• Sensitivity and specificity for a FLEX high-risk score to predict POD24 were 60% and 68%, respectively.
• By contrast, FLIPI and FLIPI-2 had a lower sensitivity (53%) and specificity (59%).
• The sensitivity for PRIMA-PI was greater than for FLEX (69%), however the specificity was considerably lower (47%).

Model validation

• Using 342 evaluable patients from the SABRINA trial, the FLEX model showed intergroup differences of 17.7% and 19.4% for 2-year and 3-year PFS, respectively, and better discrimination between high-risk and low-risk patients compared with FLIPI-2 (12.8% and 14.4%, respectively) and PRIMA-PI (8.4% and 9.3%, respectively). FLEX showed similar discrimination to FLIPI (16.9% and 19.2%, respectively).
• In terms of OS, FLEX performed similarly to FLIPI and FLIPI-2, but better than PRIMA-PI.
• Inconsistent performance across the training and validation cohorts may be attributed to differences in study design.

Conclusion

Using a large training dataset, the authors have developed FLEX, a new clinical prognostic model for FL that incorporates widely accessible clinical variables. For patients in the GALLIUM training cohort, the model better discriminated between a good and poor PFS and OS than existing models. FLEX was also better able to predict PFS in the SABRINA validation cohort, however ability to predict OS was similar to FLIPI and FLIPI-2. FLEX showed a greater accuracy for predicting early disease progression than existing models, however in this context the authors acknowledge that the performance of the model is still not optimal. Consistent results across treatment regimens would suggest that the model is applicable to patients receiving current standard of care therapies.