Contributions
Abstract: EP624
Type: E-Poster Presentation
Session title: Chronic lymphocytic leukemia and related disorders - Biology & Translational Research
Background
Chronic lymphocytic leukemia (CLL) shows an extremely heterogeneous clinical course. Among cases with a favorable immunogenetic profile (mutated immunoglobulin heavy chain variable region (IGHV) genes, absence of unfavorable FISH lesions), there are those with an ultra-stable (US) clinical course who do not progress for at least 10 years from diagnosis. In a previous study (Raponi et al, Ann Oncol 2018) we described the genetic landscape of US-CLL patients and suggested a predictive model, based on the expression of six genes, capable of identifying these cases at diagnosis.
Aims
To test the previously designed classifier model in an expanded multicenter cohort of cases, including US-CLL and patients that despite favorable immunogenetic features progressed and required treatment within 5 years from diagnosis (non-US-CLL), in the context of the Campus CLL group.
Methods
Sixty new patients (31 US-CLL and 29 non-US-CLL) were included. All showed a mutated IGHV status, normal FISH or del13q only. US-CLL showed a median follow-up of 15.3 years (range: 10-23), while non-US-CLL showed a median time to progression of 2 years (range: 0-4.9). Droplet digital PCR (ddPCR) was applied to quantify the expression of the 6 previously identified genes (P2RX1, PLXND1, CPT1A, PRRC2C, GPM6A, SMCHD1) (Raponi et al, Ann Oncol 2018). These 60 cases were pooled with the 79 of our previous study (total: 89 US-CLL and 50 non-US-CLL). The absolute expression values of each gene were used to refine the original decision tree model.
Results
The decision tree model was initially fit on a training dataset obtained by the resample with bootstrap of 139 patients (n=1390). The validation of the decision tree based model and self-consistency test were performed by k-fold cross validation (CV) method. The decision-tree is derived from the best predictive model in the R output, identifying 12 subgroups (nodes) and 5 associated factors (genes).
By applying the decision tree model to the original cohort of 139 patients and according to the absolute expression values of the 5 genes included in the final model (GPM6A+P2RX1+PLXND1+PRRC2C+SMCHD1), 127/139 (91.4%) were correctly classified and 12 (8.6%) were misclassified. In particular, all the 89 US-CLL were correctly predicted (sensitivity: 100%), as well as 38 out of 50 cases who developed a progressive disease were classified by the model as non-US-CLL (specificity: 76%). Overall, the positive and negative predictive values of the model were 88.1% and 100%, respectively.
Conclusion
Although for early stage and asymptomatic CLL patients ‘watch and wait’ remains the standard of care, early treatment intervention is a tempting investigational scenario in the era of novel drugs. Besides the impact on patients’ counseling at diagnosis and subsequent monitoring, our approach could further refine the current prognostic algorithms of low-risk CLL (mutated IGHV and favorable FISH) by allowing to identify CLL patients who are very likely to remain stable for more than a decade and may never require treatment from those who may benefit from experimental early intervention.
Keyword(s): Chronic lymphocytic leukemia, Gene expression profile, Prognosis, Stable disease
Abstract: EP624
Type: E-Poster Presentation
Session title: Chronic lymphocytic leukemia and related disorders - Biology & Translational Research
Background
Chronic lymphocytic leukemia (CLL) shows an extremely heterogeneous clinical course. Among cases with a favorable immunogenetic profile (mutated immunoglobulin heavy chain variable region (IGHV) genes, absence of unfavorable FISH lesions), there are those with an ultra-stable (US) clinical course who do not progress for at least 10 years from diagnosis. In a previous study (Raponi et al, Ann Oncol 2018) we described the genetic landscape of US-CLL patients and suggested a predictive model, based on the expression of six genes, capable of identifying these cases at diagnosis.
Aims
To test the previously designed classifier model in an expanded multicenter cohort of cases, including US-CLL and patients that despite favorable immunogenetic features progressed and required treatment within 5 years from diagnosis (non-US-CLL), in the context of the Campus CLL group.
Methods
Sixty new patients (31 US-CLL and 29 non-US-CLL) were included. All showed a mutated IGHV status, normal FISH or del13q only. US-CLL showed a median follow-up of 15.3 years (range: 10-23), while non-US-CLL showed a median time to progression of 2 years (range: 0-4.9). Droplet digital PCR (ddPCR) was applied to quantify the expression of the 6 previously identified genes (P2RX1, PLXND1, CPT1A, PRRC2C, GPM6A, SMCHD1) (Raponi et al, Ann Oncol 2018). These 60 cases were pooled with the 79 of our previous study (total: 89 US-CLL and 50 non-US-CLL). The absolute expression values of each gene were used to refine the original decision tree model.
Results
The decision tree model was initially fit on a training dataset obtained by the resample with bootstrap of 139 patients (n=1390). The validation of the decision tree based model and self-consistency test were performed by k-fold cross validation (CV) method. The decision-tree is derived from the best predictive model in the R output, identifying 12 subgroups (nodes) and 5 associated factors (genes).
By applying the decision tree model to the original cohort of 139 patients and according to the absolute expression values of the 5 genes included in the final model (GPM6A+P2RX1+PLXND1+PRRC2C+SMCHD1), 127/139 (91.4%) were correctly classified and 12 (8.6%) were misclassified. In particular, all the 89 US-CLL were correctly predicted (sensitivity: 100%), as well as 38 out of 50 cases who developed a progressive disease were classified by the model as non-US-CLL (specificity: 76%). Overall, the positive and negative predictive values of the model were 88.1% and 100%, respectively.
Conclusion
Although for early stage and asymptomatic CLL patients ‘watch and wait’ remains the standard of care, early treatment intervention is a tempting investigational scenario in the era of novel drugs. Besides the impact on patients’ counseling at diagnosis and subsequent monitoring, our approach could further refine the current prognostic algorithms of low-risk CLL (mutated IGHV and favorable FISH) by allowing to identify CLL patients who are very likely to remain stable for more than a decade and may never require treatment from those who may benefit from experimental early intervention.
Keyword(s): Chronic lymphocytic leukemia, Gene expression profile, Prognosis, Stable disease