Prognostic Score and Cytogenetic Risk Classification for Chronic Lymphocytic Leukemia Patients: Center for International Blood and Marrow Transplant Research Report

In recent years, we have seen a rapid development of targeted therapies against pathways that are constitutively activated in chronic lymphocytic leukemia (CLL). However, despite the clinical effectiveness of these therapies, durability of response to these agents is limited and discontinuation is high due to toxicity or progressive disease in high-risk patients (1). Allogeneic hematopoietic cell transplantation (HCT) thus remains the only potentially curative therapeutic modality for patients who fail targeted therapies.

The Dana-Farber Cancer Institute (DFCI) transplant group previously reported an excellent 5-year progression-free survival (PFS) and overall survival (OS) after reduced intensity conditioning (RIC) HCT for CLL (2). In that study, a prognostic model for PFS was proposed using remission status, HCT comorbidity index (HCT-CI), lymphocyte count, and lactate dehydrogenase (LDH) at HCT. Although this model was very predictive of PFS as well as OS, the model needed to be validated in a much larger and multicenter cohort.

Cytogenetic risk is critical to stratifying patients with CLL and has been extensively studied in the nontransplant setting. However, controversy remains about the importance of high-risk cytogenetics in predicting outcome after HCT. In the nontransplant setting, patients with TP53 or ATM gene mutation, or 17p or 11q deletion by cytogenetics, have markedly poorer survivals (3–5). However, a number of transplant studies have suggested that patients with CLL with del(17p) can still achieve long-term remission after transplant, but the sample size of these studies was limited and thus the long-term remission rate for these patients would need to be confirmed in a larger study (2, 6–8). Furthermore, a recent study has suggested that transplant outcome is significantly worse for those patients with CLL with complex karyotype, defined as five or more abnormalities rather than the traditional three or more abnormalities (9). However, the sample size of this study was also small. The Center for International Blood and Marrow Transplant Research (CIBMTR) patient population provides the best opportunity to determine the optimal definition of complex karyotype for patients with CLL who undergo HCT and to truly assess whether HCT overcomes the adverse prognosis of high-risk cytogenetics, including del(17p), del(11q) and complex karyotype, which often includes the respective cytogenetic aberrations.

The goals of this study are therefore (i) to validate and/or modify the prognostic score developed by the DFCI transplant group (2) using a large independent cohort, (ii) to assess whether HCT overcomes the adverse prognosis of high-risk cytogenetics, and (iii) to develop a cytogenetic risk classification for patients with previously treated CLL who underwent RIC HCT. Steps in data analysis are shown in Fig. 1A.

CIBMTR is a combined research program of the Medical College of Wisconsin (Milwaukee, WI) and the National Marrow Donor Program. CIBMTR comprises a voluntary network of more than 450 transplantation centers worldwide that contribute detailed data on consecutive allogeneic and autologous HCT to a centralized Statistical Center. All CIBMTR research studies are conducted in compliance with U.S. Office of Human Research Protection (OHRP) common rule regulations (45 CFR Part 46) and the Federal Drug Administration (FDA) regulations [21 Code of Federal Regulations (CFR) Part 50 and 21 CFR Part 56)]. Studies utilizing the CIBMTR database are conducted under its Research Database Protocol. The National Marrow Donor Program/Be the Match central Institutional Review Board (IRB) is fully accredited by the Association for Accreditation of Human Research Protection Programs and has oversight responsibility for all research conducted under the CIBMTR Research Database Protocol. In addition, CIBMTR requires all reporting centers to maintain local IRB approval for the Research Database Protocol and seek consent from patients for use of their data in CIBMTR research. Only patients' data that is consented for use in research are included in CIBMTR studies. Additional details regarding the data source are described elsewhere (10).

Between 2008 and 2014, 1,505 patients with a diagnosis of CLL underwent allogeneic HCT with 7/8 or 8/8 matched related or unrelated donors, peripheral blood, or bone marrow transplants reported to CIBMTR. Of these, 758 patients had disease specific data available from the CIBMTR comprehensive report form or provided by centers. To ensure this cohort of 758 patients was a representative subset of the larger cohort of 1,505 patients, OS and PFS were examined. Both OS and PFS were superimposable between patients with and without available information (Supplementary Fig. S1) and baseline characteristics were similar (data not shown), suggesting the study cohort is representative of the entire cohort. Of 758 patients, 606 (80%) patients underwent RIC and 152 (20%) underwent myeloablative conditioning (MAC) HCT according to the CIBMTR definition of conditioning intensity (11). In this report, we focus on the 606 patients who underwent RIC during the study period and data on MAC HCT recipients will be reported in a subsequent paper.

We performed expert clinician review of all available primary cytogenetics reports submitted to CIBMTR (J.R. Brown, M.S. Davids, and V.O. Volpe). Patients with a standard metaphase karyotype analysis with at least five cells, although the majority had the usual 20, were considered evaluable and categorized as normal karyotype or by the total number of abnormalities, with complex karyotype defined as either three or more, or five or more, for purposes of further evaluation. Patients with a FISH analysis were separately categorized as to presence or absence of del13q, del11q, del17p, and trisomy 12.

The primary endpoint for development of the prognostic score and cytogenetic risk classification was progression-free survival (PFS). Other endpoints of interest included overall survival (OS), nonrelapse mortality (NRM), and relapse. PFS was defined as the time from stem cell infusion to disease relapse, progression or death from any cause, whichever occurred first. Patients who were alive without disease relapse or progression were censored at the time last seen alive and relapse or progression-free. OS was defined as the time from stem cell infusion to death from any cause. Patients who were alive or lost to follow-up were censored at the time last seen alive. The 1- and 4-year follow-up completeness indices were 99% and 96%, respectively. The Kaplan–Meier method was used to estimate PFS and OS, whereas cumulative incidence of NRM and relapse was estimated in the context of a competing risks framework. NRM and relapse were treated as competing events. The log-rank and Gray tests (12) were used for comparing estimates of PFS and OS and estimates of cumulative incidence of NRM and relapse, respectively. In addition, univariable Cox regression analysis for PFS was performed (Supplementary Table S1A); multivariable Cox regression analysis for PFS and OS and multivariable competing risks regression analysis (13) for NRM and relapse were performed. Center effect was tested using a frailty model and the effect was not significant (P = 0.52). Because adding Center to the model did not affect the final models, the center effect was not further adjusted.

To modify the prognostic model, we performed multivariable Cox regression analysis using a bootstrap validation method with 10,000 resamples of size 606 with replacement (14). The bootstrap method was used to adjust the model to decrease the impact of overfitting to the original dataset and reducing the influence of unusual or outlying values. Although the split-sample method is commonly used for validation, it greatly reduces the sample size for both the training and validation sets. Furthermore, if the process is repeated with a different split, different regression coefficients may be obtained from the validation set. This is a concern particularly when the sample size is not large (15). The bootstrap validation method overcomes these drawbacks and obtains nearly unbiased estimates without sacrificing sample (15). Using 10,000 resamples, the bootstrap sampling distribution of each estimator was established and HRs and respective 95% confidence intervals were taken from the 50th, 2.5th, and 97.5th percentiles of the distribution, respectively.

It is known that many factors in CLL are correlated and collinear (2). Collinearity can cause predictors to compete with each other and SEs of the regression coefficient estimates can be inflated. As a result, some predictors arbitrarily become nonsignificant (15). However, collinearity does not affect the joint influence of correlated variables, that is, some variables are correlated and partially redundant but not identical [e.g., lymphocyte count and white blood cell count (WBC)]. We therefore performed unsupervised hierarchical clustering analysis to assess potential collinearity among variables and found that no complete remission (CR)/partial remission (PR), high WBC, and high lymphocyte count were clustered (expected predictors of relapse) and HCT-CI and low WBC were clustered (expected predictors of NRM; Supplementary Fig. S2A). Because collinearity makes it difficult to estimate regression coefficients while holding (highly correlated) variables constant, we performed multivariable analysis including each of these prognostic variables separately but adjusting for all other variables (Supplementary Table S1B). Prior to performing Cox regression analysis, the proportional hazards assumption was examined and two-way interaction terms were assessed. The linearity assumption for all continuous variables was examined using the methods of restricted cubic spline function on relative hazard (15) and classification and regression tree for survival data (16, 17). From this analysis, WBC was found to be nonlinear, indicating both low and high WBC were associated with shorter PFS (Supplementary Fig. S2B). On the basis of this analysis, WBC was categorized as low (<2), normal (2–10), or high (>10 × 10⁹/L), lymphocyte count was dichotomized as low (<2) or high (≥2 × 10⁹/L) and HCT-comorbidity index (HCT-CI) was categorized as 0–1 versus ≥2. The cut-off values for WBC are consistent with our previous study of WBC after HCT (18). We also utilized the Akaike information index (AIC) for the assessment of model fit and C-index (15) for predictive ability of models. All P values were two-sided at a significance level of 0.05. All calculations were performed using SAS 9.3 (SAS Institute, Inc.) and R version 3.3.2.

The median age was 58 (range, 26–73) and 72% were male. At the time of HCT, the median number of prior treatments was three (range, 1–10), with 13% in CR and 51% in PR. Thirty-seven percent received an HLA-identical sibling donor HCT and 53% received an 8/8 matched unrelated donor HCT. Thirty-two percent of patients had del(17p) and 27% had complex karyotype defined as ≥3 abnormalities (Table 1). For the entire cohort, the median survival time among survivors was 49 months (range, 4–99); the 4-year PFS was 41% [95% confidence interval (CI), 37%–45%] and the 4-year OS was 56% (95% CI, 52%–60%).

We first examined whether the previously reported prognostic scoring system (2) applied in this cohort. Using this system, PFS for patients with a high (score = 2) and very high risk (score ≥ 3) are very similar; OS for patients with low (score = 0) and intermediate risk (score = 1) are similar and OS for patients with high (score = 2) and very high risk (score ≥ 3) are also similar. As a result, patients were stratified into three groups in PFS and two groups in OS instead of the initially intended four groups (Supplementary Fig. S3). Because this stratification is not optimal, we examined each of these four factors separately and found that elevated LDH was not associated with PFS (Supplementary Fig. S4A–S4D).

It is widely believed that WBC and lymphocyte count are closely related. To investigate whether one metric is a surrogate of the other, we plotted lymphocyte count against WBC and found that for WBC ≤10 (×10⁹/L), there was no correlation between these two metrics (Fig. 1B). For WBC >10 (×10⁹/L), WBC is nearly linearly associated with lymphocyte count (Fig. 1C). To assess the prognostic implication of these metrics, we performed multivariable analysis. Because there is a correlation in high WBC and high lymphocyte count, two multivariable analyses were performed for each metric adjusting for other factors. For high lymphocyte count, HR for PFS was 1.76 (P < 0.0001); for low and high WBC, the HR was 1.87 (P < 0.0001) and 1.62 (P = 0.01), respectively. We then created a joint variable from these two metrics as: (i) low lymphocyte count with WBC ≥ 2 × 10⁹/L (low risk), (ii) high lymphocyte count with normal WBC (intermediate risk), (iii) low WBC or high lymphocyte count with high WBC (high risk; Fig. 1D). The HR for PFS was 1.73 (P = 0.0012) for intermediate and 2.05 (P < 0.0001) for high-risk group compared with low-risk group (Supplementary Table S1B).

We then sought to rebuild an improved prognostic scoring system using the CIBMTR cohort. On the basis of HRs from multivariable models with 10,000 times resampling, one point (HR < 1.5) was assigned to no CR/PR and HCT-CI ≥ 2. For WBC and lymphocyte count, 1.5 points were assigned to the intermediate risk group (1.5 ≤ HR ≤ 2) and 2 points (HR > 2) were assigned to the high-risk group. The summary of score assignment is presented in Table 2A. The sum score, ranging from 0 to 4, was then grouped into four categories: 0 (low), 1 (intermediate), 1.5–2.5 (high), ≥3 (very high). Relative to the low-risk group, the HR was 1.66 (P = 0.0024), 2.36 (P < 0.0001), and 2.7 (P < 0.0001) for the intermediate-, high-, and very-high-risk group, respectively (Table 2B). This prognostic scoring system also stratified patients for OS (HR 1.68, 2.2, 2.76 for intermediate-, high-, and very-high-risk group, respectively), NRM, and relapse (Table 2B).

Using this scoring system, the 4-year PFS was 58%, 42%, 33%, and 25%, P < 0.0001 and the 4-year OS was 70%, 57%, 54%, and 38%, P < 0.0001, for low-, intermediate-, high-, and very-high-risk group, respectively (Fig. 2A and B; Supplementary Table S2A). The 4-year cumulative incidence of NRM was 19%, 31%, 25%, 44%, P < 0.0001, respectively. The 4-year cumulative incidence of relapse was 23%, 27%, 41%, 31% (P = 0.007), respectively (Supplementary Fig. S5A and S5B; Supplementary Table S2A).

Of 606 patients with RIC, 469 patients had cytogenetics information, which led us to analyze this information separately and then combine with the prognostic score later. To ensure that patients with cytogenetics information are representative of the cohort of 606, we examined PFS and OS as before and the curves are superimposable between patients with and without this information (Supplementary Fig. S6A and S6B). We then assessed each individual cytogenetic abnormality and found that patients with del(17p) or complex karyotype with ≥5 abnormalities (complex 5) had significantly worse PFS (Supplementary Fig. S7). We also found that having complex karyotype with three or four abnormalities (complex 3–4) was not associated with poor PFS. Using both classification and regression tree (16, 17) for survival data and a multivariable Cox model with 10,000 times resampling, we built a cytogenetics risk classification tree (Fig. 3A) and stratified patients into three groups: (i) no del(17p) and no complex 5 (low risk), (ii) del(17p) without complex 3–4, or 5; or complex 5 without del(17p) (intermediate risk), and (iii) del(17p) with complex 3–4, or 5 (high risk). Relative to the low-risk group, HR for PFS was 1.56 (P = 0.012) and 2.0 (P < 0.0001) for the intermediate- and high-risk group, respectively. For OS, HR was 1.56 (P = 0.048) and 1.89 (P = 0.0016) for the intermediate- and high-risk group, respectively. For relapse, the subdistribution HR (ref. 13; sHR) was 1.52 (P = 0.12) and 2.11 (P = 0.006) for the intermediate- and high-risk group, respectively (Table 2C). Using this classification, the 4-year PFS was 48%, 37%, 23% (P < 0.0001), the 4-year OS was 62%, 53%, 42% (P = 0.0008), and the 4-year cumulative incidence of relapse was 26%, 31%, 46% for the low-, intermediate-, and high-cytogenetic-risk groups, respectively (P = 0.0004; Fig. 3B–E; Supplementary Table S2B).

We then examined cytogenetic risk within each prognostic score and found that a small proportion of patients with poor cytogenetic risk within each prognostic score group did poorly, particularly within the low- and intermediate-risk group (Supplementary Fig. S8). We therefore attempted to combine the prognostic score and cytogenetic risk classification by adding one and two points for patients with intermediate and high cytogenetic risk, respectively, to the prognostic score (Table 2A). The sum score was again grouped into four categories based on HRs: low, intermediate, high, and very high. The low-risk group now includes patients with low prognostic score and low cytogenetic risk. Relative to the low-risk group, the HR for PFS was 1.92 (P = 0.0006), 2.62 (P < 0.0001), and 3.71 (P < 0.0001) for the intermediate-, high-, and very-high-risk group, respectively (Table 2D). The result for OS is similar (Table 2D). Using this stratification, the 4-year PFS was 61%, 44%, 34%, and 18% (P < 0.0001) and 4-year OS was 75%, 59%, 51%, and 34% (P < 0.0001) for the low-, intermediate-, high-, and very high-risk group, respectively (Fig. 2C and D; Supplementary Table S2C). Because patients with high cytogenetic risk had poor outcome irrespective of the prognostic risk, if the cytogenetics information is also available, both sets of information should be utilized to risk stratify patients.

To further identify which factors contribute to NRM and relapse, we performed multivariable competing risks regression analysis using the bootstrap method. Less than PR (sHR 1.89, 95% CI, 1.13–3.52, P = 0.013), low WBC (sHR 2.05, P = 0.002), and high HCT-CI (sHR 1.59, P = 0.005) were significant factors for NRM, whereas less than CR (sHR 2.02, P = 0.0046), high lymphocyte count (sHR 2.0, P = 0.0002), and high WBC (sHR 1.71, P = 0.034) were significant factors for relapse (Supplementary Table S3). The cumulative incidences of NRM and relapse of these factors are presented in Fig. 4. PFS and OS according to remission status and WBC group are presented in Supplementary Fig. S9.

We examined the impact of the conditioning intensity (nonmyeloablative vs. reduced intensity conditioning) in a multivariable model and found it not significant. In fact, the PFS and OS curves are superimposable (Supplementary Fig. S6C and S6D). We therefore did not consider this factor further.

We also explored other potential prognostic factors that had incomplete data. The sample size for these factors ranged from 180 to 328 out of 606. These included bulky adenopathy, prior spleen involvement, prior chemotherapy, refractoriness to fludarabine, refractoriness to chemotherapy, prior rituximab, and number of lines of prior treatment. Of these factors, refractoriness to fludarabine and refractoriness to chemotherapy were significantly associated with PFS (HR 1.54, P = 0.023; HR 1.53, P = 0.017, respectively). In addition, these variables were highly correlated with CR/PR (r = 0.79 P < 0.0001 with refractoriness to chemotherapy, r = 0.53, P < 0.0001, with refractoriness to fludarabine) and were therefore well represented by disease status in the model. For example, 89% of patients who were not chemo-refractory were responders, whereas chemo-refractory patients were all nonresponders. In addition, the number of prior treatments was significantly associated with PFS (HR, 2.48 for ≥ 2 vs. <2, P = 0.0007). The number of prior treatments was also strongly associated with disease status as 74% of patients without CR/PR had 3 or more prior treatments (P = 0.005).

In this large cohort of previously treated patients with high-risk CLL who underwent RIC HCT in a multicenter setting, we first assessed the previously developed DFCI model. We then further modified the model and developed a more robust prognostic scoring system that effectively risk-stratifies patients with CLL at the time of HCT. We also developed a novel cytogenetic-based risk stratification system that has been lacking in the HCT setting. When the prognostic score was combined with the cytogenetic risk, patients in the low-risk group had a prolonged survival (4-year OS 75%) compared with patients in the very-high-risk group (4-year OS 34%). This result shows that despite the fact that these patients had already failed multiple other therapies and thus transplant was offered as the last resort, transplant outcome is promising, particularly for patients with low risk. By analyzing for the first time the impact of complex karyotype with or without del(17p) in the context of HCT, we were able to establish the prognostic impact of both complex karyotype, and del(17p), particularly in the context of complex karyotype.

In this study, we refined cut-off values of lymphocyte count and HCT-CI. In the previous DFCI study (2), the cut-off value for lymphocyte count was 1 × 10⁹/L. However, this cut-off value was based on a small number of patients (N = 76). With a much larger sample size, we were able to determine the prognostic cut-off value for lymphocyte count to be 2 × 10⁹/L and HCT-CI ≥2. In many clinical studies, WBC is typically assumed to be linear in relation with outcome. That is, the higher the value, the worse the outcome. Because 14% of our patients had low WBC at baseline and these patients had poor outcomes, WBC loses its prognostic significance if WBC is simply dichotomized into >10 vs. ≤10. More importantly, because low WBC may reflect bone marrow dysfunction or a defective marrow microenvironment, whereas high WBC reflects disease burden, both low and high WBC are prognostic factors and must be analyzed separately. Indeed, consistent with this result, in our previous study of WBC, low WBC was associated with an increased risk of NRM, whereas high WBC was associated with an increased risk of relapse (18), as also demonstrated in the current data. Furthermore, we found that high lymphocyte count is correlated with high WBC. However, for WBC < 10, there is no clear correlation between these two metrics and neither predicts the other. We also found that low WBC is a significant prognostic factor in patients with CLL.

Patients with CR at the time of HCT had lower relapse and thus better PFS compared with those with PR (Fig. 4B; Supplementary Fig. S9A). However, the OS is similar (Supplementary Fig. S9B), indicating that repeat HCT, DLI, or other salvage therapies, likely provided benefit to patients with PR who experienced relapse. When those patients with CR in the low-risk prognostic score group (N = 38) were compared with those with PR (N = 146), the HR for PFS was not significantly improved (HR, 0.84; P = 0.57), that is, if patients were in remission without any other risk factors, the outcome was similar between CR and PR. We therefore collapsed CR and PR in the prognostic model, but this warrants further investigation in future studies.

Controversy remains about the importance of high-risk cytogenetics in predicting outcome after HCT. In a nontransplant setting, due to the difficulty of obtaining metaphase karyotypes from nondividing CLL cells, the Dohner hierarchical model (3), which utilizes FISH and not standard karyotype analysis, came into widespread use. As a result, complex karyotype is not well defined in CLL and has typically borrowed the definition from acute leukemias (i.e., ≥3 or ≥4 abnormalities; refs. 19, 20). Recent studies have evaluated the impact of complex karyotype in the context of standard therapy and found it to be adverse (21–23). In the transplant setting, a small study (N = 51) performed by Jaglowski and colleagues (9) reported that outcome is significantly worse for those patients with CLL with ≥5 abnormalities (HR, 4.75, 95% CI, 2.12–10.6, P = 0.0001). Consistent with this finding, our study found that HCT outcome is significantly worse for patients with ≥5 abnormalities. Our study also defined a novel interaction between complex karyotype and del(17p); HCT overcomes any adverse effect of three or four abnormalities unless these include del(17p). When we assessed the interaction between high-risk FISH and complex karyotype, patients without del(17p) and with fewer than five abnormalities had low cytogenetic risk, whereas patients with del(17p) with ≥3 abnormalities had high cytogenetic risk. Furthermore, patients with del(11q) alone or del(11q) with ≥3 abnormalities did not have a poor outcome after HCT.

We acknowledge limitations of this study including its retrospective design. In this particular setting, however, this approach can also be a strength that improves upon our prior single-center study (2) because the study cohort represents all patients from a comprehensive dataset derived from a diverse spectrum of transplant programs worldwide and thus avoids potential selection bias, inflated efficacy, and underestimated real world toxicity. Although data on CLL karyotype is a great strength of our study, a potential limitation is that other data were not available. For example, CIBMTR did not collect information on prior targeted therapy during the study period, although the majority of this study period pre-dates the era of approved targeted therapy in CLL. Also, CIBMTR does not capture certain disease-specific prognostic information, such as IGHV mutation status, TP53 mutation status, and beta2-microglobulin (B2m). Again, neither IGHV mutation status nor TP53 mutation status were commonly performed in clinical practice at the time that most of these patients were diagnosed or undergoing transplant. Furthermore, at time of HCT, the prognostic value of these markers is not well established, nor do most patients have sufficient circulating disease to measure these even if desired. In our previous study of del(17p) in the nontransplant setting (24), 88% of patients with TP53 mutation had del(17p) in parallel. Because of this overlap, we can assume that TP53 mutation information has likely been well represented through del(17p). In our previous study (2), B2m was not included in the prognostic model due to missing data. However, we noticed that B2m was highly correlated with disease status and may therefore be well represented, as many prognostic factors in CLL are highly correlated. A final limitation is that we excluded patients receiving less than 7/8 matched donor transplants. Therefore, caution is required before extrapolating these results to heavily mismatched transplants such as umbilical cord blood or haploidentical transplant recipients.

Some known prognostic factors in HCT are not included in the prognostic model. These include age, HLA matching and patient and donor sex mismatch. Our HR for patients age 70 or higher was 1.8 (P = 0.07). Although this is not significant at the 0.05 level because of the small number of patients in this age group (N = 18), the importance of advanced age should not be discounted. For patients age 60–69, the HR was 1.17 and 1.1 for age 50–59 compared with patients age <50. The HR for the 7/8 matched unrelated cohort here was 1.04 (P = 0.82) compared with HLA-identical sibling donors. In a recent retrospective study of patients with CLL by the European Society for Blood and Marrow Transplantation (EBMT) (25), risk factors for PFS included age and patient donor sex mismatch [HR = 1.4 for male patients with female donors (F→M) compared with male patients with male donors (M→M), 95% CI, 1.1–1.8, P = 0.01]. We examined sex mismatch in a multivariable model and the HR for PFS was 1.09 (95% CI, 0.82–1.43) for F→M compared with M→M. This is consistent with our previous report for patients with all hematologic malignancies using the CIBMTR data (26). In that study, we reported that patient sex and not sex mismatch was detrimental. An important difference between the EBMT and our current study is that the EBMT study included patients with myeloablative conditioning and largely relied on HCT risks rather than on CLL risks. Furthermore, the EBMT study did not assess the impact of complex karyotype as the cytogenetic abnormalities were classified according to Dohner and colleagues (3) and thus cytogenetic study results between these two studies are not directly comparable. Nonetheless, patients with del(17p) in the EBMT study had a poor prognosis compared with patients with other abnormalities (25). We did not investigate NRM-specific or relapse-specific prognostic factors. Because PFS is a composite endpoint of NRM and relapse, factors associated with increased risk of NRM but decreased risk of relapse (or vice versa) were therefore not selected because the net effect was negated due to the opposing effects.

In summary, using readily available data, we have developed a prognostic scoring system and cytogenetic risk classification that risk stratifies patients with CLL who undergo RIC HCT. We demonstrate in a large well-characterized cohort that patient factors as well as del(17p) and complex karyotype with more than 5 abnormalities are both associated with poorer outcomes, consistent with recent observations also with novel agents. These can be used for counseling patients, comparing data across studies, and providing a benchmark for the evaluation of future interventions. For future study, we plan to validate all three systems using an independent dataset for patients receiving targeted therapies prior to HCT, although we suspect that the genetic and/or risk factors used for our risk scoring systems will still be significant as disease burden (represented by remission status, lymphocyte count, and WBC), patient fitness (HCT-CI) and cytogenetic risk are the main drivers of clinical outcomes in CLL. Because the durability of response to targeted therapy in a high-risk patient population is relatively low, ultimately an integrative approach of transplant and targeted therapy, that is, inducing remission in high-risk patients with targeted therapy, offering HCT during remission, and potentially reinstating targeted therapy for consolidation post-HCT, might enhance the clinical outcome of these patients. Alternative approaches, such as CAR T therapy, are also under investigation.

M.S. Davids is a consultant/advisory board member for AbbVie, Genentech, Gilead, Janssen, Pharmacyclics, TG Therapeutics, Celgene, InCyte, Merck, AstraZeneca, Verastem, MEI Pharma, Syros, Acerta and Adaptive Biosciences, and reports receiving commercial research grants from Genentech, Pharmacyclics, TG Therapeutics, Verastem, Bristol-Myers Squibb, MEI Pharma, Surface Oncology, and Acerta. M. Shadman is a consultant/advisory board member for AbbVie, Genentech, Sounds Biologics, Verastem, ADC Therapeutics, Atara Biotherapeutics, and AstraZeneca. J. Mehta has ownership interests (including patents) at Celgene, Bristol-Myers Squibb, and Bluebird, and reports receiving speakers bureau honoraria from Celgene and Millennium. J.P. McGuirk reports receiving speakers bureau honoraria from Kite and Juno, and reports receiving commercial research grants from Novartis, Fresenius Biotech, Astellas, Bellicum Pharmaceuticals, Kite, Gamida Cell, and Pluristem Ltd. S. Giralt is a consultant/advisory board member for Celgene, Sanofi, Amgen, Jazz, Kite, Novartis, Johnson & Johnson and Pfizer, and reports receiving commercial research grants from Johnson & Johnson, Sanofi, Amgen, Miltenyi, and Celgene. M.A. Perales is a consultant/advisory board member for Incyte, Merck, AbbVie, Bristol-Myers Squibb, Novartis, Takeda, and Nektar Therapeutics. G. Hale is an employee of Medpace. R.P. Gale is an employee of Celgene. M. Hamadani reports receiving speakers bureau honoraria from Sanofia, is a consultant/advisory board member for ADC Therapeutics, Janssen, Incyte, and reports receiving commercial research support from Astellas, Takeda, and Spectrum. S. Ganguly reports receiving speakers bureau honoraria from Seattle Genetics, is a consultant/advisory board member for Amgen, Janssen, and Kite, and reports receiving commercial research support from Daiichi Sankyo. H. Lazarus reports receiving speakers bureau honoraria from Pharmacyclics. R. Reshef is a consultant/advisory board member for Kite/Gilead, Bristol-Myers Squibb, Incyte, Pfizer, Pharmacyclics, Magenta, and Jazz Pharmaceuticals. G.C. Hildebrandt has ownership interests (including patents) at AbbVie, Aetna, Axim, Bluebird, Cardinal Health, Celgene, Cellectis, Clovis Oncology, Crisper, CVS Health, Endocyte, GW Pharmaceuticals, Idexx Labs, Immunomedics, Insys Therapeutics, Johnson & Johnson, Juno Therapeutics, Kite Pharma, Novartis, Pfizer, Procter & Gamble, Sangamo Bioscience, and Vertex Pharmaceuticals, is a consultant/advisory board member for Onyx Pharmaceuticals, Pfizer, Kite, Incyte, Jazz Pharmaceuticals, and reports receiving commercial research grants from Takeda, Jazz Pharmaceuticals, and Pharmacyclics. M. Solh has ownership interests (including patents) at Gelgene, and reports receiving speakers bureau honoraria from Amgen. M.A. Kharfan-Dabaja is a consultant/advisory board member for Pharmacyclics. N. Ghosh reports receiving speakers bureau honoraria from Pharmacyclics, Janssen, Abbvie, and Gildead, and is a consultant/advisory board member for Pharmacyclics and Janssen. A. Saad is a consultant/advisory board member for Actinium Pharma Inc. and reports receiving royalty fees from Incysus Therapeutics. B.T. Hill is a consultant/advisory board member for Pharmacyclics, Abbvie, Genentech, and AstraZeneca. N.S. Saba reports receiving speakers bureau honoraria from Pharmacyclics and Janssen, is a consultant/advisory board member for Pharmacyclics, I-Mab Biopharma, and Kyowa Kirin, and reports receiving commercial research grants from Celgene. B. Savani reports receiving speakers bureau honoraria from Sanofi Genzyme and Jazz Pharmaceuticals. J. Cerny is a consultant/advisory board member for Pfizer Inc., Incyte Inc. and Daiichi-Sankyo. M.R. Grunwald has ownership interests (including patents) at Medtronic, is a consultant/advisory board member for Agios, Abbvie, Amgen, Cardinal Health, Celgene, Incyte, Pfizer, Merck, Daiichi-Sankyo, ARIAD, and reports receiving commercial research support from Amgen, Forma Therapeutics, Incyte, Janssen, Novartis, and Genentech/Roche. J.R. Brown reports receiving speakers bureau honoraria from Janssen, Teva, and Abbvie, is a consultant/advisory board member for Abbvie, Verastem, Gilead, Acerta, Beigene, Genentech/Roche, Juno/Celgene, Kite, Loxo, Novartis, Pfizer, Pharmacyclics, Sunesis, TG Therapeutics, and reports receiving commercial research grants from Gilead, Loxo, Sun, and Verastem, and served on data safety monitoring committees for Invectys and Morphosys. No potential conflicts of interest were disclosed by the other authors.

The views expressed in this article do not reflect the official policy or position of the NIH, the Department of the Navy, the Department of Defense, Health Resources and Services Administration (HRSA) or any other agency of the U.S. Government.

Conception and design: H.T. Kim, Z.-H. Hu, M. Sorror, O.W. Press, J. Pidala, R. Negrin, A. Langston, M. Battiwalla, G. Hale, R.P. Gale, D.I. Marks, M. Hamadani, S. Ganguly, R. Reshef, G.C. Hildebrandt, J.-Y. Cahn, M. Solh, M. Aljurf, H.C. Schouten, A.M. Beitinjaneh, B. Savani, J. Cerny, B. Wirk, R. Olsson, A. Bashey, U. Popat, E.P. Alyea, W. Saber, J.R. Brown

Development of methodology: H.T. Kim, Z.-H. Hu, M. Sorror, O.W. Press, J. Pidala, A. Langston, S. Giralt, M. Battiwalla, D.I. Marks, M. Hamadani, S. Ganguly, M. Solh, M. Aljurf, B. Savani, J. Cerny, J. Yared, B. Wirk, E.P. Alyea, W. Saber

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): Z.-H. Hu, M.S. Davids, V. Volpe, J.H. Antin, M. Sorror, M. Shadman, O.W. Press, S. Devine, J. Uberti, E. Agura, R. Nash, J. Mehta, S. Giralt, M.A. Perales, M. Battiwalla, S. Ganguly, R. Reshef, H.C. Schouten, A. Pawarode, E. Copelan, J. Cerny, J. Yared, R. Olsson, W. Saber, J.R. Brown

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): H.T. Kim, K.W. Ahn, Z.-H. Hu, V. Volpe, M. Sorror, M. Shadman, J. Pidala, J.P. McGuirk, M.A. Perales, M. Battiwalla, G. Hale, M. Hamadani, S. Ganguly, V. Bacher, H. Lazarus, G.C. Hildebrandt, M.A. Kharfan-Dabaja, N. Ghosh, A. Saad, M. Aljurf, H.C. Schouten, B.T. Hill, N.S. Saba, S. Nathan, A.M. Beitinjaneh, B. Savani, J. Cerny, M.R. Grunwald, J. Yared, B. Wirk, S. Chhabra, R. Olsson, A. Bashey, U. Popat, R.M. Sobecks, E.P. Alyea, W. Saber, J.R. Brown

Writing, review, and/or revision of the manuscript: H.T. Kim, Z.-H. Hu, M.S. Davids, J.H. Antin, M. Sorror, M. Shadman, J. Pidala, W. Hogan, R. Negrin, S. Devine, E. Agura, R. Nash, J. Mehta, J.P. McGuirk, S.J. Forman, A. Langston, S. Giralt, M.A. Perales, M. Battiwalla, G. Hale, R.P. Gale, D.I. Marks, M. Hamadani, S. Ganguly, V. Bacher, H. Lazarus, R. Reshef, G.C. Hildebrandt, Y. Inamoto, J.-Y. Cahn, M. Solh, M.A. Kharfan-Dabaja, N. Ghosh, A. Saad, M. Aljurf, H.C. Schouten, A. Pawarode, T. Kindwall-Keller, N.S. Saba, E. Copelan, S. Nathan, A.M. Beitinjaneh, B. Savani, J. Cerny, M.R. Grunwald, J. Yared, B. Wirk, T. Nishihori, S. Chhabra, R. Olsson, A. Bashey, U. Gergis, U. Popat, R.M. Sobecks, E.P. Alyea, W. Saber, J.R. Brown

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): Z.-H. Hu, V. Volpe, W. Saber, J.R. Brown

Study supervision: H.T. Kim, M. Hamadani, J.-Y. Cahn, R.M. Sobecks, W. Saber

Other (initial conception; see note on page X): H.T. Kim, J.R. Brown

This work was supported by NIH grants R01CA183559-02 (to H.T. Kim). J.R. Brown acknowledges support from National Comprehensive Cancer Network, from NCI (R01CA213442 and P01CA206978), and from the Susan and Gary Rosenbach Fund for Lymphoma Research and the Melton Family Fund for CLL Research. The CIBMTR is supported primarily by Public Health Service Grant/Cooperative Agreement 5U24CA076518 from the NCI, the National Heart, Lung and Blood Institute (NHLBI), and the National Institute of Allergy and Infectious Diseases (NIAID); a Grant/Cooperative Agreement (4U10HL069294) from NHLBI and NCI; a contract (HHSH250201200016C) with Health Resources and Services Administration (HRSA/DHHS); two grants (N00014-17-1-2388 and N0014-17-1-2850 from the Office of Naval Research; and grants from Actinium Pharmaceuticals, Inc., Amgen, Inc., Amneal Biosciences, Angiocrine Bioscience, Inc., anonymous donation to the Medical College of Wisconsin, Astellas Pharma US, Atara Biotherapeutics, Inc., Be the Match Foundation, Bluebird bio, Inc., Bristol–Myers Squibb Oncology, Celgene Corporation, Cerus Corporation, Chimerix, Inc., Fred Hutchinson Cancer Research Center, Gamida Cell Ltd., Gilead Sciences, Inc., HistoGenetics, Inc., Immucor, Incyte Corporation, Janssen Scientific Affairs, LLC, Jazz Pharmaceuticals, Inc., Juno Therapeutics, Karyopharm Therapeutics, Inc, Kite Pharma, Inc., Medac, GmbH, MedImmune, The Medical College of Wisconsin, Mediware, Merck & Co, Inc., Mesoblast, MesoScale Diagnostics, Inc., Millennium, the Takeda Oncology Co., Miltenyi Biotec, Inc., National Marrow Donor Program, Neovii Biotech NA, Inc., Novartis Pharmaceuticals Corporation, Otsuka Pharmaceutical Co, Ltd., PCORI, Pfizer, Inc, Pharmacyclics, LLC; PIRCHE AG, Sanofi Genzyme, Seattle Genetics, Shire, Spectrum Pharmaceuticals, Inc., St. Baldrick's Foundation, Sunesis Pharmaceuticals, Inc., Swedish Orphan Biovitrum, Inc., Takeda Oncology, Telomere Diagnostics, Inc., and University of Minnesota.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.