PDF
Citation
Print
INTRODUCTION
Although novel therapies have recently been introduced [1, 2], allogeneic hematopoietic stem cell transplantation (alloSCT) is still regarded as the only curative modality for acute leukemia [3-6]. However, a low overall survival (OS) related to relapse or treatment-related mortality is an important obstacle that compromises the efficacy of alloSCT [7]. Thus, a precise risk-adapted approach remains an unmet need in clinical practice.
In addition to emerging biological factors (such as adverse cytogenetics, failure to achieve minimal residual disease, and intolerance to chemotherapeutic toxicities), the comorbidity of individuals could have tremendous impacts on the prognosis of alloSCT. In this regard, various models, including the hematopoietic cell transplantation-specific comorbidity index (HCT-CI) [8], comorbidity-age index [9], and Charlson comorbidity index [10], have been used for pre-transplant risk stratification prior to alloSCT based on the patient’s comorbidities. These indices also provide a well-defined risk stratification for mortality. Nevertheless, these indices were not established in a nationwide cohort of patients with acute leukemia. Furthermore, the prognostic impact of commonly emphasized variables among the aforementioned comorbidity indices, such as diabetes, hypertension, cerebrovascular disease, pulmonary dysfunction, and prior non-hematologic malignancy, has not yet been validated in a Korean nationwide cohort.
In Korea, the Korean National Health Insurance Service (KNHIS) program is a mandatory public health insurance system that covers approximately 98% of the overall Korean population. The KNHIS database contains universal medical claims and mortality for the entire Korean population [11] and has been used in various epidemiological research studies on hematologic diseases, as described in detail elsewhere [12-15]. These strengths enable research on a nationwide cohort with the endpoint of identifying personal comorbidities related to patient prognosis after alloSCT.
Accordingly, this study aimed to verify whether well-known patient comorbidity-related prognostic factors, including age, sex, previous non-hematologic malignancy, hypertension, diabetes, dyslipidemia, chronic obstructive pulmonary disease (COPD), cerebrovascular or cardiovascular disease (CVA), anxiety disorder, and depression, have prognostic impacts on outcomes of alloSCT using KNHIS data. We were also interested in the development of a prognostic scoring system based on identified individual comorbidities, followed by external validation of the developed system using an independent database of the development cohort from the KNHIS data.
MATERIALS AND METHODS
Data source of the development cohort
Data for the development cohort were extracted from the KNHIS database. We first included adult patients (≥18 yr) who underwent alloSCT between January 2002 and December 2018 in the KNHIS database using the procedure codes V073, X5061, and/or X5063. These claims codes represent the performance of alloSCT, the collection of bone marrow stem cells, and the collection of mobilized peripheral blood stem cells. Thereafter, we confirmed that the final cohort comprised cases of acute leukemia classified based on the International Classification of Diseases, Tenth Revision (ICD-10) codes C92, C93, and/or C94 for acute myeloid leukemia, C91 for acute lymphoblastic leukemia, and/or C95 for unclassified acute leukemia. The requirement for individual patient consent was waived because of the anonymous nature and public availability of the data.
Data source of the validation cohort
Data from consecutive adult patients with acute leukemia who underwent alloSCT between January 2019 and April 2020 at the Catholic Hematologic Hospital were used to construct the validation cohort. Data were collected until April 2021 to have at least a 1-year follow-up period. This study was approved by the Institutional Review Board of Seoul St. Mary’s Hospital, Seoul, Korea (KC19ZNSI0396) and was conducted in accordance with the Declaration of Helsinki.
Definition of the development cohort
For individuals treated with multiple alloSCTs, the baseline date for calculating OS was defined as the date of the first alloSCT. We used the ICD-10 codes to define whether the patient presented with comorbidities, including previous non-hematologic malignancy (ICD-10 code: Cxx, except for C83, C86, C90, C91, C92, C93, C94, and/or C95 indicating hematologic malignancies), hypertension (ICD-10 code: I10, I11, I12, I13, I14, and/or I15), diabetes (ICD-10 code: E10, E11, E12, E13, and/or E14), dyslipidemia (ICD-10 code: E78), chronic obstructive pulmonary disease (COPD; ICD-10 code: J44), anxiety disorder (ICD-10: F30), and depression (ICD-10: F32 and/or F33). CVA was indicated when the patient was administered antiplatelet agents (drug codes: 117101ATB, 136901ATB, 157702ACR, 162403ATR, 165001ACH, 194930ATB, 492501ATB, 495201ATB, 498801ATB, 498900ATB, 501501ATB, 517900ACE, 517900ATE, 597301ATB, 597302ATB, 615901ATB, 615902ATB, 659501BIJ, and/or 667500ACE) and/or anticoagulants (drug codes: 511401ATB, 511403ATB, 511402ATB, 511404ATB, 249103ATB, 249105ATB, 613701ACH, 613702ACH, 617001ATB, 617002ATB, 643602ATB, 643601ATB, and/or 643603ATB).
Construction and validation of the scoring system
In the development cohort, all risk factors potentially related to OS with a P-value of <0.05 in the univariable analysis were entered into the multivariable model to confirm the factors associated with OS. Multivariable analysis was performed using the Cox proportional hazards regression model. The final parameters used in the scoring system were defined by a P-value of <0.05 in the final multivariable model. The risk score for each significant parameter was assigned by adjusting the hazard ratio (HR) values to a loge scale.
Using the scoring system created using the development cohort, we performed an analysis to validate whether the scoring system performed well as a reliable prognostic tool in the validation cohort. OS and the incidence of non-relapse mortality (NRM) and relapse were compared between the risk groups. Finally, discriminatory performance was assessed by receiver operating characteristic (ROC) curve analysis.
Statistical analysis
Numerical variables not exhibiting a normal distribution are presented as medians (range, minimum–maximum). Categorical variables are presented as numbers (%). OS was defined as the time from the date of alloSCT to death (from any cause) or the date of the last follow-up. OS rates at 1 year were calculated using the Kaplan-Meier method and compared using the log-rank test. We calculated the NRM probability and relapse rates using cumulative incidence estimation based on the competing risks of relapse and NRM. The area under the curve (AUC) of the ROC curve was calculated to predict the accuracy of the validation cohort analysis. An AUC value >0.7 was considered reliable [16]. DeLong’s test was used to compare the statistical differences between the AUCs. All statistical analyses were conducted using R statistical software (ver. 3.6.1, R Foundation for Statistical Computing, Vienna, Austria, 2019). Statistical significance was set at P<0.05.
RESULTS
Patient characteristics
Development cohort: In total, 8,230 patients who underwent alloSCT between January 2002 and December 2018 were identified. Patients aged <18 years (N=1,060) and those having other hematologic diseases (N=3,361) were excluded from the analysis. Accordingly, 3,809 patients with acute leukemia were included in the development cohort (Fig. 1A). The baseline demographics are summarized in Table 1. Overall, the median age of the patients was 47 years (range, 18–74 yr), and 54.0% of the patients were males (N=2,055). Regarding the underlying comorbidities at baseline, previous non-hematologic malignancy, hypertension, diabetes, dyslipidemia, COPD, CVA, anxiety disorder, and depression were present in 387 (10.2%), 1,224 (32.1%), 1,125 (29.5%), 2,135 (56.1%), 191 (5.0%), 166 (4.4%), 900 (23.6%), and 613 (16.1%) patients, respectively. With a median follow-up of 50.2 months [95% confidence interval (CI), 47.7–53.2], the 1-year, 3-year, and 5-year OS rates were estimated as 68.2% (95% CI, 66.7–69.7), 51.5% (95% CI, 49.8–53.2), and 46.8% (95% CI, 45.0–48.6) in the total cohort, respectively (Fig. 2).
Validation cohort: Data on 576 consecutive patients treated with alloSCT were collected. Following the exclusion of patients aged <18 years (N=128) and those having other hematologic diseases (N=135), a validation cohort of 313 patients was eligible for analysis (Fig. 1B). The baseline characteristics of the validation cohort are shown in Table 1.
Analysis of factors associated with OS in the development cohort
The univariable analysis identified the following potential factors related to poor OS: age (≥65 years or 50–64 years compared with <50 years); male sex; comorbidities such as previous non-hematologic malignancy, hypertension, diabetes, dyslipidemia, CVA, and anxiety disorder/depression. In the multivariable analysis, we confirmed that 5 variables, including advanced age (50–64 years or ≥65 years), male sex, non-hematologic malignancy, hypertension, and CVA, were significantly associated with poor OS, as shown in Table 2.
Comorbidity index score for survival outcome in the development cohort
Based on the multivariable analysis, an assigned risk score of each variable, obtained by the log-scale of the HR (Table 3) are as follows: age (50–64 years, 0.21 points; ≥65 years, 0.55 points), male sex (0.13 points), previous non-hematologic malignancy (0.17 points), hypertension (0.13 points), and CVA (0.4 points). The risk score was computed as the sum of each variable score, resulting in a median risk score of 0.2 and a range of 0–1.38. We then classified the patients into 10 subgroups, whose ranks were determined by the order of the decile risk scores. Based on the OS rates according to these 10 subgroups (Fig. 3A), the risk score was finally stratified into 4 risk groups (Fig. 3B): the 1-year OS/5-year OS rates were 73.5% [(95% CI, 71.4–75.6)/52.9% (95% CI, 50.4–55.6)], 66.2% [(95% CI, 63.6–68.9)/44.0% (95% CI, 41.0–47.2)], 61.9% [(95% CI, 57.2–67.0)/37.3% (95% CI, 31.9–43.6)], and 50.9% [(95% CI, 44.8–57.7)/29.6% (95% CI, 23.4–37.5)] in the low-risk (score ≤0.17), intermediate-risk (0.17< score ≤0.4), high-risk (0.4< score ≤0.55), and very high-risk (score >0.55) groups, respectively. The developed scoring system calculator is presented in the Supplementary calculator.
Validation cohort analysis: validation of the scoring system
With a median follow-up of 18.8 months (95% CI, 18.1–20.0) in the validation cohort, the 1-year OS rate and 1-year cumulative incidence of NRM and relapse were 74.4% (95% CI, 69.2–78.9), 11.5% (8.3–15.3), and 23.0% (18.5–27.8), respectively (Supplementary Fig. 1). When the developed scoring system was applied to the validation cohort, the 1-year OS rates were measured differently according to the risk group (P=0.085), which were 79.4% (95% CI, 72.2–84.9) in the low-risk group, 74.2% (95% CI, 64.3–81.8) in the intermediate-risk group, 60.0% (95% CI, 35.7–77.6) in the high-risk group, and 61.1% (95% CI, 35.7–77.6) in the very high-risk group. There were significant differences in both the 1-year OS rate between a combined group of the low-risk and intermediate groups and another combined group of the high-risk and very high-risk groups at 77.4% (95% CI, 71.8–82.1) and 60.7% (95% CI, 46.7–72.1), respectively (P=0.018) (Fig. 4A).
The 1-year cumulative incidence of NRM was significantly different according to the risk groups (P=0.035), which was 7.5% (95% CI, 4.1–12.3) in the low-risk group, 11.3% (95% CI, 6.0–18.6) in the intermediate-risk group, 20.0% (95% CI, 5.9–40.0) in the high-risk group, and 25.0% (95% CI, 12.2–40.0) in the very high-risk group (Fig. 4B). However, we observed no significant difference in the 1-year cumulative incidence of relapse according to the risk group (P=0.349, Fig. 4C). In the ROC curve analysis, the developed scoring system achieved an AUC of 0.715 (95% CI, 0.658–0.772), indicating reliable discrimination of NRM events at 1 year in the validation cohort (Fig. 4D).
DISCUSSION
This study developed a new scoring system to predict patient prognosis after alloSCT in 3,809 patients with acute leukemia using a development cohort derived from a nationwide database. With a 5-year OS probability of 46.8% in the development cohort, we created a comorbidity index scoring system comprising age, sex, previous non-hematologic malignancy, hypertension, and CVA. The risk score was significantly stratified into 4 risk groups: low-risk, intermediate-risk, high-risk, and very high-risk groups associated with 1-year/5-year OS rate probabilities of 73.5%/52.9%, 66.2%/44%, 61.9%/37.3%, and 50.9%/29.6%, respectively. The validation cohort analysis indicated that the developed comorbidity index scoring system was statistically feasible for predicting OS and NRM. AlloSCT is generally planned as a consolidative procedure following intensive chemotherapy in patients who achieve complete remission after intensive chemotherapy. In the validation cohort analysis, we observed interesting findings that there were significant increasing trends in the proportion of patients presenting with comorbidities at pre-alloSCT compared with the diagnosis of their disease (Supplementary Fig. 2). Therefore, our results suggest that updated comorbidity profiles at the time of alloSCT, as well as the diagnosis of acute leukemia, should be monitored rigorously because complications following intensive chemotherapy could indicate a new comorbidity, although it was absent at the time of diagnosis.
In the validation cohort of the current study, we evaluated the feasibility of the developed scoring system, which provided discriminative predictability for OS and NRM. In addition to our system, HCT-CI, which comprises 17 different categories of organ dysfunction [8], has not only been used as the most reliable tool for the comorbidity-based risk assessment of survival and NRM after alloSCT but has also been successfully validated in transplant institutions worldwide [17-19]. However, the cohort size of the original study (N=1,055) for the establishment of HCT-CI, as well as studies of validation (the largest among all studies to the best of our knowledge, N=324), was relatively small compared with our study. Moreover, multi-institutional validation of the predictive power of HCT-CI has not yet been documented. Interestingly, a multicenter prospective study by a Japanese group found that HCT-CI failed to predict NRM [20]. Therefore, there is an unmet need for better tools to optimize comorbidities based on risk stratification for alloSCT. Although our developed comorbidity index model should be further validated in a future study, the supplemental analysis showed that our new index demonstrated better 1-year NRM prediction than the HCT-CI system in the validation cohort, with an AUC of 0.688 vs. 0.509, P<0.001 (Supplementary Fig. 3). This result also illustrated an AUC of 0.509 for HCT-CI in predicting 1-year NRM, which is not discriminated, as in the abovementioned results from the multicenter Japanese prospective study.
Compared with prior studies depicting risk stratification according to comorbidities, our results had several strengths. First, to the best of our knowledge, the risk factors for survival were analyzed using a nationwide cohort comprising the largest number of patients from a Korean multicenter cohort. Second, unlike the aforementioned scoring systems that require endoscopy, echocardiography, pulmonary function tests, and various laboratory tests to generate the results, our scoring system was created using easily accessible data such as age and underlying comorbidities. Third, despite the small number of variables comprising the scoring system, we believe that the discriminatory power, as well as the validity of our system for predicting survival in the validation cohort, which was distinctively independent of the development cohort, appeared to be reliable.
There are several limitations to our study. In the development cohort analysis, we were unable to investigate all variables comprising the HCT-CI because the KNHIS database has the inherent limitation that concrete results derived from laboratory tests, echocardiography, and pulmonary function tests were absent. In this regard, the severity of the comorbidity could not be assessed. Second, due to unreliable data related to the cause of death in the KNHIS data, the cumulative incidence of relapse-related death or NRM was not explored in the development cohort analysis. Third, this study could be limited by the relatively small size, short-term follow-up, and retrospective design of the validation cohort. Fourth, we believe that the results of the current study cannot be generalized because other important factors, including characteristics of acute leukemia and treatment factors, were not entered as covariants to verify the prognostic impact of patient-related comorbidities.
Despite these limitations, our study provides a feasible scoring system based on comorbidity information that is easy to obtain for predicting survival prognosis among patients with acute leukemia who undergo alloSCT. Therefore, the current results could be attributed to better clinical decision-making based on patient-driven risk-adaptive strategies for alloSCT.
XML Download