T-cell Lymphoma Forum
Photo by Larry Young
SAN FRANCISCO—Researchers have used data from the T-Cell Project (TCP) to create a prognostic model for peripheral T-cell lymphoma not otherwise specified (PTCL-NOS).
Analyses have suggested the TCP model is more accurate for PTCL-NOS than 4 other prognostic models—the International Prognostic Index (IPI), the Prognostic Index for T-cell Lymphoma (PIT), the International Peripheral T-cell Lymphoma Project score (IPTCLP), and the modified PIT (mPIT).
Massimo Federico, MD, of the University of Modena and Reggio Emilia in Italy, described the TCP model at the 9th Annual T-cell Lymphoma Forum.
Creating the model
TCP is a prospective registry that includes data from T-cell lymphoma patients in 15 countries located in 5 different regions of the world. As of December 31, 2016, 1523 cases of T-cell lymphoma have been registered with TCP.
Dr Federico and his colleagues used these data to create their prognostic model. There were 311 patients with PTCL-NOS who had adequate data for analysis. The 5-year overall survival (OS) for these patients was 36%.
The researchers chose 13 variables from the literature that have been reported to have a prognostic impact on survival in PTCL-NOS:
- Age > 60
- Lactate dehydrogenase > upper limit of normal
- Albumin < 3.5 g/dL
- Hemoglobin < 12 g/dL
- Platelets < 150/mm3
- Lymphocyte to monocyte ratio ≤ 2.1
- Neutrophil to lymphocyte ratio > 6.5
- Absolute neutrophil count (ANC) > 6.5/mm3
- ECOG performance status > 1
- Stage III-IV disease
- B symptoms
- Extra nodal sites > 1
- Male gender.
In univariate analysis, nearly all of these factors were significantly associated with OS in the cohort of TCP patients. (The 2 exceptions were age older than 60 and having more than 1 extranodal site.)
However, Dr Federico and his colleagues said the factors with the greatest prognostic impact were:
- ECOG performance status > 1, with a hazard ratio (HR) of 2.12 (P<0.001)
- Albumin < 3.5 g/dL, with an HR of 2.03 (P<0.001)
- ANC > 6.5/mm3, with an HR of 1.85 (P<0.001)
- Stage III-IV disease, with an HR of 1.74 (P=0.010).
So the researchers used these factors in their model, which has 3 risk categories.
Risk categories
Patients were considered low-risk if they had 0 of the 4 risk factors. These patients had a 3-year OS of 76% and a 5-year OS of 69%.
Patients were considered intermediate-risk if they had 1 to 2 risk factors. These patients had a 3-year OS of 43% and a 5-year OS of 31%. Compared to low-risk patients, the HR was 3.08 (P<0.001).
Patients were considered high-risk if they had 3 to 4 risk factors. The 3-year OS was 11% for these patients, and the 5-year OS was 8%.
The HR was 8.88 (P<0.001) for high-risk compared to low-risk patients and 2.88 (P<0.001) for high-risk compared to intermediate-risk patients.
Validation
The researchers tested the TCP model in a validation cohort of 98 patients from the COMPLETE registry. As with the training cohort of TCP patients, the model revealed 3 different risk groups (in terms of OS) in the validation cohort.
Dr Federico noted that there were no significant differences between the training and validation cohorts, except when it came to follow-up. The median follow-up was 46 months in the TCP group and 18 months in the COMPLETE group.
The researchers also found the TCP could classify patients into 3 different risk groups according to progression-free survival.
Comparison
Finally, Dr Federico and his colleagues compared the TCP model to the IPI, PIT, IPTCLP, and mPIT models using 208 patients.
“The discriminant power of the proposed model is superior to the others in terms of all of the statistical tests we adopted,” Dr Federico said.
| Model | c-Harrell*
(95% CI) |
D-Royston
(SE) |
R2 | AIC (95% CI) | AUC,
3-year OS |
| TCP | 0.666 (0.618-0.713) | 1.152 (0.191) | 0.31 (0.14-0.46) | 983 | 0.714 |
| PIT | 0.614 (0.563-0.664) | 0.750 (0.195) | 0.15 (0.06-0.31) | 1004 | 0.696 |
| IPI | 0.645 (0.594-0.696) | 0.883 (0.191) | 0.22 (0.08-0.38) | 987 | 0.704 |
| IPITCLP | 0.606 (0.549-0.663) | 0.631 (0.188) | 0.12 (0.03-0.28) | 1006 | 0.704 |
| mPIT | 0.640 (0.586-0.694) | 0.762 (0.170) | 0.16 (0.05-0.33) | 999 | 0.681 |
In closing, Dr Federico said the TCP model clearly defines risk groups in PTCL-NOS and identifies patients with relatively good prognosis.
However, there is a need for emerging biologic variables to be tested for prognostic value and included in prognostic tools to allow for better risk stratification.
*c-Harrel: Harrell’s concordance index, 95% CI: confidence interval, D-Royston: Royston/Sauerbrei’s D statistic (Stat Med 2004 Mar 15, 23[5]:723-48), SE: standard error, R2: explained randomness, AIC: Akaike information criterion, AUC: area under the curve (according to Heagerty et al, Biometrics, 2000 Jun, 56[2]:337-44).