Features and test metrics of tested risk classifiers for previously untreated patients with symptomatic, advanced stage FL from 2 independent cohorts
| . | POD24* . | FLIPI . | m7-FLIPI . | POD24-PI . |
|---|---|---|---|---|
| Type | Posttreatment | Pretreatment | Pretreatment | Pretreatment |
| Primary endpoint | OS | OS | FFS | POD24 |
| Validated on independent cohort | Yes | Yes | Yes | Yes |
| Clinical predictors | POD24 | Age >60 y, no. of nodal sites >3, elevated serum LDH, hemoglobin <12 g/L, Ann Arbor stage III/IV | High-risk FLIPI, ECOG performance status >2 | High-risk FLIPI |
| Molecular predictors | None | None | Nonsilent mutations in 7 genes (ARID1A, CARD11, CREBBP, EP300, EZH2, FOXO1, MEF2B) at VAFs ≥10% | Nonsilent mutations in 3 genes (EP300, EZH2, FOXO1) at VAFs ≥10% |
| Calculation of risk score | Single variable | Cumulative sum of predictor values, all predictors have weight = 1 | Cumulative sum of predictor values, predictors have individual weights† | Cumulative sum of predictor values, predictors have individual weights (Figure 4A) |
| Number of risk groups | 2 (no POD24 = low-risk, POD24 = high-risk) | 3 (0-1 = low-risk, 2 = interm-risk, 3-5 = high-risk) | 2 (<0.8 = low-risk, >0.8 = high-risk) | 2 (<0.71 = low-risk, >0.71 = high-risk) |
| High-risk cases among patients with symptomatic, advanced stage FL (%) | GLSG: 23/132 (17) | GLSG: 77/151 (51) | GLSG: 43/151 (28) | GLSG: 63/151 (42) |
| BCCA: 23/102 (22) | BCCA: 53/107 (50) | BCCA: 24/107 (22) | BCCA: 39/107 (36) | |
| Predictive for POD24 | n/a | Yes | Yes | Yes |
| GLSG: evaluable patients 132/151 | OR = 4.6 (P = .0059) | OR = 5.8 (P = .00031) | OR = 7.3 (P = .00016) | |
| Sens 78%, Spec 56%, AUC 0.67 | Sens 61%, Spec 79%, AUC 0.70 | Sens 78%, Spec 67%, AUC 0.73 | ||
| Accuracy 60%, PPV 27%, NPV 92% | Accuracy 76%, PPV 38%, NPV 91% | Accuracy 71%, PPV 33%, NPV 94% | ||
| BCCA: evaluable patients 107/107 | OR = 3.2 (P = .035) | OR = 4.8 (P = .0052) | OR = 4.3 (P = .0051) | |
| Sens 70%, Spec 58%, AUC 0.64 | Sens 43%, Spec 86%, AUC 0.65 | Sens 61%, Spec 73%, AUC 0.67 | ||
| Accuracy 61%, PPV 33%, NPV 87% | Accuracy 77%, PPV 48%, NPV 84% | Accuracy 69%, PPV 40%, NPV 87% | ||
| Predictive for FFS | n/a | Yes, but only as a binary classifier (low-/interm-risk vs high-risk) | Yes | Yes |
| GLSG: evaluable patients 151/151 | 5-y FFS 57% vs 76% | 5-y FFS 38% vs 77% | 5-y FFS 50% vs 77% | |
| HR = 2.11 (P = .0034) | HR = 4.14 (P = 6.313·10−9) | HR = 3.06 (P = 5.989·10−6) | ||
| BCCA: evaluable patients 107/107 | 5-y FFS 47% vs 70% | 5-y FFS 25% vs 68% | 5-y FFS 36% vs 72% | |
| HR = 2.18 (P = .0075) | HR = 3.58 (P = 4.924·10−6) | HR = 3.01 (P = 7.178·10−5) | ||
| Predictive for OS | Yes‡ | Yes | Yes | Yes |
| GLSG: evaluable patients 151/151 | 5-y OS 40% vs 96% | 5-y OS 75% vs 91% | 5-y OS 65% vs 90% | 5-y OS 71% vs 91% |
| HR = 10.91 (P = 5.262·10−14) | HR = 2.59 (P = .0083), C-Index 0.75 | HR = 3.38 (P = .00031), C-Index 0.78 | HR = 3.55 (P = .00026), C-Index 0.79 | |
| BCCA: evaluable patients 107/107 | 5-y OS 26% vs 93% | 5-y OS 60% vs 89% | 5-y OS 42% vs 84% | 5-y OS 48% vs 89% |
| HR = 13.602 (P = 6.661·10−16) | HR = 3.90 (P = .00034), C-Index 0.81 | HR = 4.89 (P = 7.661·10−7), C-Index 0.84 | HR = 5.35 (P = 9.996·10−7), C-Index 0.86 | |
| Comments | Risk classification not possible for patients who died/were censored within 24 mo of first-line treatment. Only predictive for OS. First described in 2015. | Most widely used and best validated pretreatment classifier. Contains only clinical variables not necessarily directly reflecting disease biology. First described in 2004, not widely used to guide treatment decisions. | Highest accuracy and specificity to predict POD24. Most discriminative classifier for patients without POD24 into high- and low-risk groups. Sequencing of 7 genes required. First described in 2015, requires further validation. | Highest sensitivity to predict POD24. Sequencing of 3 genes required. Requires further validation. |
| . | POD24* . | FLIPI . | m7-FLIPI . | POD24-PI . |
|---|---|---|---|---|
| Type | Posttreatment | Pretreatment | Pretreatment | Pretreatment |
| Primary endpoint | OS | OS | FFS | POD24 |
| Validated on independent cohort | Yes | Yes | Yes | Yes |
| Clinical predictors | POD24 | Age >60 y, no. of nodal sites >3, elevated serum LDH, hemoglobin <12 g/L, Ann Arbor stage III/IV | High-risk FLIPI, ECOG performance status >2 | High-risk FLIPI |
| Molecular predictors | None | None | Nonsilent mutations in 7 genes (ARID1A, CARD11, CREBBP, EP300, EZH2, FOXO1, MEF2B) at VAFs ≥10% | Nonsilent mutations in 3 genes (EP300, EZH2, FOXO1) at VAFs ≥10% |
| Calculation of risk score | Single variable | Cumulative sum of predictor values, all predictors have weight = 1 | Cumulative sum of predictor values, predictors have individual weights† | Cumulative sum of predictor values, predictors have individual weights (Figure 4A) |
| Number of risk groups | 2 (no POD24 = low-risk, POD24 = high-risk) | 3 (0-1 = low-risk, 2 = interm-risk, 3-5 = high-risk) | 2 (<0.8 = low-risk, >0.8 = high-risk) | 2 (<0.71 = low-risk, >0.71 = high-risk) |
| High-risk cases among patients with symptomatic, advanced stage FL (%) | GLSG: 23/132 (17) | GLSG: 77/151 (51) | GLSG: 43/151 (28) | GLSG: 63/151 (42) |
| BCCA: 23/102 (22) | BCCA: 53/107 (50) | BCCA: 24/107 (22) | BCCA: 39/107 (36) | |
| Predictive for POD24 | n/a | Yes | Yes | Yes |
| GLSG: evaluable patients 132/151 | OR = 4.6 (P = .0059) | OR = 5.8 (P = .00031) | OR = 7.3 (P = .00016) | |
| Sens 78%, Spec 56%, AUC 0.67 | Sens 61%, Spec 79%, AUC 0.70 | Sens 78%, Spec 67%, AUC 0.73 | ||
| Accuracy 60%, PPV 27%, NPV 92% | Accuracy 76%, PPV 38%, NPV 91% | Accuracy 71%, PPV 33%, NPV 94% | ||
| BCCA: evaluable patients 107/107 | OR = 3.2 (P = .035) | OR = 4.8 (P = .0052) | OR = 4.3 (P = .0051) | |
| Sens 70%, Spec 58%, AUC 0.64 | Sens 43%, Spec 86%, AUC 0.65 | Sens 61%, Spec 73%, AUC 0.67 | ||
| Accuracy 61%, PPV 33%, NPV 87% | Accuracy 77%, PPV 48%, NPV 84% | Accuracy 69%, PPV 40%, NPV 87% | ||
| Predictive for FFS | n/a | Yes, but only as a binary classifier (low-/interm-risk vs high-risk) | Yes | Yes |
| GLSG: evaluable patients 151/151 | 5-y FFS 57% vs 76% | 5-y FFS 38% vs 77% | 5-y FFS 50% vs 77% | |
| HR = 2.11 (P = .0034) | HR = 4.14 (P = 6.313·10−9) | HR = 3.06 (P = 5.989·10−6) | ||
| BCCA: evaluable patients 107/107 | 5-y FFS 47% vs 70% | 5-y FFS 25% vs 68% | 5-y FFS 36% vs 72% | |
| HR = 2.18 (P = .0075) | HR = 3.58 (P = 4.924·10−6) | HR = 3.01 (P = 7.178·10−5) | ||
| Predictive for OS | Yes‡ | Yes | Yes | Yes |
| GLSG: evaluable patients 151/151 | 5-y OS 40% vs 96% | 5-y OS 75% vs 91% | 5-y OS 65% vs 90% | 5-y OS 71% vs 91% |
| HR = 10.91 (P = 5.262·10−14) | HR = 2.59 (P = .0083), C-Index 0.75 | HR = 3.38 (P = .00031), C-Index 0.78 | HR = 3.55 (P = .00026), C-Index 0.79 | |
| BCCA: evaluable patients 107/107 | 5-y OS 26% vs 93% | 5-y OS 60% vs 89% | 5-y OS 42% vs 84% | 5-y OS 48% vs 89% |
| HR = 13.602 (P = 6.661·10−16) | HR = 3.90 (P = .00034), C-Index 0.81 | HR = 4.89 (P = 7.661·10−7), C-Index 0.84 | HR = 5.35 (P = 9.996·10−7), C-Index 0.86 | |
| Comments | Risk classification not possible for patients who died/were censored within 24 mo of first-line treatment. Only predictive for OS. First described in 2015. | Most widely used and best validated pretreatment classifier. Contains only clinical variables not necessarily directly reflecting disease biology. First described in 2004, not widely used to guide treatment decisions. | Highest accuracy and specificity to predict POD24. Most discriminative classifier for patients without POD24 into high- and low-risk groups. Sequencing of 7 genes required. First described in 2015, requires further validation. | Highest sensitivity to predict POD24. Sequencing of 3 genes required. Requires further validation. |
AUC, area under the curve; CR30, complete response rate at 30 months; FLIPI, Follicular Lymphoma International Prognostic Index; HR, hazard ratio; NPV, negative predictive value; OR, odds ratio; POD, progression of disease; POD24-PI, early progression prognostic index; PPV, positive predictive value; Sens, sensitivity; Spec, specificity.
The modified definition of POD24 was used for this analysis (see text).
For this analysis only, OS for POD24 was calculated from time of treatment initiation for better comparison (nota bene, numbers differ from the text, wherein OS was calculated from time of risk-defining event).