Usefulness of Charlson comorbidity index to predict early mortality and overall survival in older patients with acute myeloid leukemia

Abstract

Introduction:

Older adults with acute myeloid leukemia(AML) often have significant comorbidities. We hypothesized that greater comorbidity burden predicts worse one-month mortality and overall survival(OS) in patients≥60 years with AML

Materials and methods:

We included 50,668 patients≥60 years diagnosed between 2004–2014 from the National Cancer Database; patients were divided into three groups with Charlson comorbidity index(CCI) 0, 1, and≥2. Chi-square tests were used to examine the association between CCI and different variables. We utilized logistic regression and cox proportional hazard models to determine predictors of one-month mortality and OS, respectively.

Results:

Among the entire cohort, 65% had CCI 0, 24% had CCI 1, and 11% had CCI≥2. Thirty-four percent did not receive chemotherapy. Patients with CCI 0 were more likely to receive chemotherapy, especially multiagent chemotherapy and undergo upfront HCT. In multivariate analyses, one-month mortality and OS were significantly worse with CCI 1 or≥2, compared to CCI 0 in the entire cohort as the subgroup of only those patients who received chemotherapy. Younger age, male gender, higher annual income, academic facility, longer travel distance, and acute promyelocytic leukemia were associated with improved OS.

Conclusion:

In one of the largest real-world studies of older adults with AML, we demonstrated that greater comorbidity, measured by higher CCI, independently predicted worse early mortality and OS in older patients with AML. Higher CCI was more common with increasing age and correlated with lower likelihood of receiving chemotherapy and HCT. Whether optimal comorbidity management and supportive care may improve outcomes needs to be studied further.

Microabstract

We hypothesized that higher Charlson comorbidity index(CCI) predicts worse one-month mortality and overall survival(OS) in patients≥60 years with acute myeloid leukemia(AML). In our National Cancer Database study, patients with CCI 0 were more likely to receive chemotherapy and undergo upfront hematopoietic cell transplant. One-month mortality and OS were significantly worse with CCI 1 or ≥2, compared to CCI 0.

Introduction

Older adults ≥ 60 years comprise more than half of total acute myeloid leukemia (AML) cases; the median age at diagnosis of AML is 65–70 years (1, 2). Several studies report worse outcomes and higher mortality in older patients with AML compared to their younger counterparts, which may be associated with both patient- and disease-related factors (2, 3). Outcomes for older patients, in fact, has remained largely unchanged over past few decades (4, 5). Older patients generally have poor performance status, abnormal organ dysfunction, and significant comorbidities (6, 7). Older adults are also reported to have poor disease biology with higher incidence of unfavorable cytogenetics, decreased response to intensive chemotherapy, and multidrug resistance (3, 7–9). Thus, intensive therapy such as chemotherapy and hematopoietic cell transplant (HCT) may not be offered as an option with concerns for higher mortality and morbidity (8, 10, 11).

In clinical practice, different scores are used to assess the risk of treatment-related toxicities and predict outcomes in patients with multiple comorbidities. One of the scores, Charlson comorbidity index (CCI), calculated based on 19 different medical conditions, weighs the comorbidities to measure patients’ burden of disease (12). CCI is known to be of prognostic significance in various underlying diseases, including different malignancies and has been studied and modified extensively since its development. Deyo modification of CCI is commonly used, which consists of 17 variables; 3 variables- leukemia, ly(2)mphoma, and localized solid tumors in original CCI are combined into a single variable- any malignancy other than metastatic solid tumors (supplement table 1) (13). Increase in number of comorbidities increases the CCI score and predicts worse outcomes (14).

We hypothesized that comorbidity burden predicts outcomes in AML. In this context, we analyzed the effect of CCI on one-month mortality and overall survival (OS) in older patients with AML.

Materials and Methods

We utilized the National Cancer Data Base (NCDB) Participant User File (PUF) to identify patients aged ≥60 years, who were diagnosed between 2004 and 2014. NCDB, a joint program of the Commission on Cancer of the American College of Surgeons and the American Cancer Society, captures approximately 70% of new diagnoses of cancer in the United States (15). Adult patients with AML were captured using International Classification of Diseases for Oncology, Version 3, codes 9840–9861, 9865–9874, and 9891–9931. The CONSORT diagram (Figure 1) illustrates the cohort selection. We excluded patients who received part of first course treatment or decided not to treat outside of the reporting facility as well as the patients with missing data (n=10,516). For a subgroup analysis of patients who received chemotherapy, we excluded the patients who did not receive chemotherapy (n=17,077). Deyo modification of CCI was used to divide patients into 3 groups with CCI scores of 0, 1, and ≥2. In NCDB database, the calculation of the Charlson-Deyo score excludes any comorbidity code identified as malignant neoplasms from the total score, since all patients have a diagnosis of cancer. Other variables analyzed in our study included age, sex, race, education, income, insurance, facility location (urban/rural), facility type, receipt of chemotherapy, receipt of hematopoietic cell transplant (HCT), distance traveled to the treatment facility, and histology. The educational and income status are not patient-level data; they are an estimated average of the population residing in the zip code of the patient, as determined by census data of the year 2012. Further information on classification of other variables in NCDB database is detailed in our prior publication (16). Our study, with deidentified data, was not considered a human subject research by the institutional review board at the University of Nebraska Medical Center.

Figure 1.

CONSORT diagram for cohort selection. AML indicates acute myeloid leukemia

Statistical analysis

Chi-square test was used to determine the relationship between one-month mortality and each of the categorical potential predictors. Variables with p-values <0.20 from the univariate analyses were used in a multiple logistic regression model to predict one-month mortality, and backward selection was used to determine the final model. In the final model, each of the variables was presented as odds ratios (OR) and 95% confidence interval (CI). For OS, we performed univariate survival analyses and generated Kaplan-Meier curves. Within each variable, the KM curves for each stratum were compared using the log rank test. The p-values from the log rank test of the survival analysis for each variable were presented. The variables with p-values <0.20 were included in a multivariate cox-proportional model and the model was reduced using backward elimination where 0.05 was the criteria for a term to remain in the model. A subgroup analysis using the same methods was performed among patients who received chemotherapy. PC SAS version 9.4 was used for all summaries and analyses.

Results

A total of 50,668 patients were included in the analysis: 35% were 60–69 years of age, 44% were female, 81% were white, 64% received chemotherapy, and 4% received HCT. Among the entire study population, 65% had CCI 0, 24% had CCI 1, and 11% had CCI ≥2 (Table 1). Patients with CCI 0, compared to those with CCI 1 or ≥2 were slightly younger, had higher annual income and private insurance, received treatment at academic center, and travelled longer distance for treatment. Patients with CCI 0 were more likely to receive chemotherapy, especially multiagent chemotherapy (42% vs 39% vs 28%) and slightly more likely to undergo upfront HCT (4% vs 3% vs 1%).

Table 1:

Patient characteristics

Variable	CCI 0 N=32,983 (%)	CCI 1 N=12,018 (%)	CC score ≥2 N=5,667 (%)	P-value
Age				<0.001
60–69	11848 (36%)	4011 (33%)	1681 (30%)
70–79	12087 (37%)	4510 (38%)	2197 (39%)
80–89	7863 (24%)	3022 (25%)	1553 (27%)
≥90	1185 (4%)	475 (4%)	236 (4%)
Sex				0.01
Male	18185 (55%)	6762 (56%)	3220 (57%)
Female	14798 (45%)	5256 (44%)	2447 (43%)
Race				<0.001
White	29448 (89%)	10600 (88%)	4990 (88%)
African American	2111 (6%)	955 (8%)	466 (8%)
Other	1424 (4%)	463 (4%)	211 (4$)
No high school diploma *				<0.001
≥21%	4748 (14%)	1941 (16%)	974 (17%)
13%−20.9%	8230 (25%)	3097 (26%)	1563 (28%)
7%−12.9%	11380 (35%)	4174 (35%)	1959 (35%)
< 7%	8625 (26%)	2806 (23%)	1171 (21%)
Annual household income *				<0.001
<$38,000	5183 (16%)	2215 (18%)	1125 (20%)
$38000-$47999	7895 (24%)	2962 (25%)	1493 (26%)
$48000-$62999	9010 (27%)	3269 (27%)	1514 (27%)
≥$63,000	10895 (33%)	3572 (30%)	1535 (27%)
Location type				0.02
Urban	32331 (98%)	11730 (98%)	5548 (98%)
Rural	652 (2%)	288 (2%)	119 (2%)
Facility type				<0.001
Academic	15342 (47%)	5294 (44%)	2253 (40%)
Non-academic	17641 (53%)	6724 (56%)	3414 (60%)
Insurance type				<0.001
Medicare	24133 (73%)	9252 (77%)	4545 (80%)
Private	7223 (22%)	2179 (18%)	841 (15%)
None/Medicaid/Other	1627 (5%)	587 (5%)	281 (5%)
Chemotherapy				<0.001
No chemotherapy	10255 (31%)	4274 (36%)	2548 (45%)
Single agent	7993 (24%)	2889 (24%)	1414 (25%)
Multi agent	13956 (42%)	4636 (39%)	1605 (28%)
Unknown	779 (2%)	219 (2%)	100 (2%)
Received HCT				<0.001
Yes	1426 (4%)	303 (3%)	54 (1%)
No	31557 (96%)	11715 (97%)	5613 (99%)
Distance traveled (miles) †				<0.001
0–4.9	8731 (26%)	3360 (28%)	1709 (30%)
5–10.9	7555 (23%)	2773 (23%)	1323 (23%)
11–30.9	8222 (25%)	2984 (25%)	1369 (24%)
≥31	8475 (26%)	2901 (24%)	1266 (22%)
Histology				<0.001
APL	1337 (4%)	622 (5%)	358 (6%)
Core binding factor AML	699 (2%)	253 (2%)	133 (2%)
Therapy-related AML/AML	3941 (12%)	1223 (10%)	570 (10%)
MRC
All others	27006 (82%)	9920 (83%)	4606 (81%)

^*based on aggregate census data from the patient’s zip code

^†distance between the center of the patients’ zip code of residence and the treatment facility

AML- Acute myeloid leukemia; APL- Acute promyelocytic leukemia; CCI- Charlson comorbidity index; CI- Confidence interval; HCT- Hematopoietic cell transplant; MRC- Myelodysplasia related changes

One-month mortality was 24%, 34% and 45% for patients with CCI 0, 1, and ≥2 respectively (p<0.001). In a univariate analysis, lower CCI (24% vs 34% vs 44% for CCI 0, 1, and ≥2, respectively, p<0.001) was associated with lower one-month mortality (supplement table 2). In a multivariate analysis, one-month mortality was significantly worse in patients with CCI 1 (OR 1.5, 95% CI 1.4–1.6) or ≥2 (OR 2.3, 95% CI 2.2–2.5), in comparison to those with CCI 0 (Table 2). Other independent factors with improved one-month mortality included younger age, male gender, higher annual income, academic facility, availability of private insurance, longer distance traveled, and acute promyelocytic leukemia.

Table 2:

Multivariate logistic regression for one-month mortality

Variable	Odds ratio	95% CI	p-value
Charlson comorbidity index
0	1
1	1.55	1.47–1.62	<0.001
≥2	2.35	2.21–2.5	<0.001
Age
60–69	1
70–79	1.62	1.53–1.71	<0.001
80–89	2.72	2.57–2.89	<0.001
≥90	4.41	3.97–4.89	<0.001
Sex
Female	1
Male	0.9	0.91–0.98	0.01
Annual houelhold income *
≥$63,000	1
$48000–62999	1.09	1.03–1.51	0.001
$38000–47999	1.14	1.08–1.21	<0.001
<$38,000	1.21	1.14–1.29	<0.001
Location type
Urban	1
Rural	1.16	1.01–1.34	0.02
Facility type
Academic	1
Non-academic	1.52	1.45–1.58	<0.001
Insurance type
Private	1
Medicare	1.2	1.13–1.27	<0.001
None/All	1.25	1.12–1.39	<0.001
Distance traveled (miles) †
0–4.9	1
5–10.9	0.9	0.85–0.95	0.0003
11–30.9	0.86	0.82–0.91	<0.001
≥31	0.82	0.77–0.87	<0.001
Histology
APL	1
Core binding factor AML	0.56	0.47–0.67	<0.001
Therapy-related AML/AML MRC	0.47	0.42–0.53	<0.001
All other	0.84	0.77–0.93	0.0006

^*based on aggregate census data from the patient’s zip code

^†distance from patients’ residence to the treatment facility

AML- Acute myeloid leukemia; APL- Acute promyelocytic leukemia; CI- Confidence interval; MRC- Myelodysplasia related changes

The median one-year OS for the entire study population was 27% (supplement figure 1). The median OS for patients with CCI 0, 1, and ≥2 was 31%, 22%, and 15% respectively (p<0.001) (Figure 2). In a univariate analysis, treatment at academic center, receipt or chemotherapy, and receipt of HCT were associated with higher OS (supplement table 3). In a multivariate analysis, OS was worse with CCI 1 (Hazard ratio [HR] 1.27, 95% CI 1.24–1.31) and ≥2 (HR1.53, 95% CI 1.49–1.58), compared to CCI 0 (Table 3). Age 60–69 years, male gender, higher annual income, academic facility, longer travel distance, and acute promyelocytic leukemia were associated with improved OS.

Figure 2:

Overall survival stratified by Charlson comorbidity index (0 vs 1 vs ≥2)

Table 3:

Cox proportional hazard model for overall survival

Variable	Hazard ratio	95% CI	p-value
Charlson comorbidity index
0	1
1	1.27	1.24–1.3	<0.001
≥2	1.53	1.49–1.58	<0.001
Age
60–69	1
70–79	1.24	1.21–1.27	<0.001
80–89	1.46	1.41–1.5	<0.001
≥90	1.73	1.65–1.83	<0.001
Sex
Female	1
Male	1.02	1.007–1.04	0.006
Race
White	1
African American	0.99	0.95–1.02	0.5
Other	0.92	0.88–0.96	0.001
Annual household income *
≥$63,000	1
$48000–62999	1.04	1.01–1.06	0.001
$38000–47999	1.06	1.03–1.09	<0.001
<$38,000	1.11	1.07–1.14	<0.001
Facility type
Academic	1
Non-academic	1.07	1.05–1.09	<0.001
Insurance type
Private	1
Medicare	1.17	1.08–1.14	<0.001
None/All	1.05	1.01–1.1	0.03
Type of chemotherapy
No chemotherapy	1
Single agent	0.5	0.49–0.52	<0.001
Multi agent	0.38	0.37–0.39	<0.001
Receipt of HCT
Yes	1
No	2.14	2.01–2.28	<0.001
Distance traveled (miles) †
0–4.9	1
5–10.9	0.97	0.94–0.99	0.04
11–30.9	0.98	0.95–1.006	0.13
≥31	1.009	0.98=1.03	0.5
Histology
APL	1
Core binding factor AML	1.67	1.53–1.81	<0.001
Therapy-related AML/AML MRC	2.05	1.93–2.17	<0.001
All other	2.4	2.27–2.53	<0.001
Year of diagnosis
2010–2014	1
2004–2009	1.05	1.03–1.07	<0.001

^*based on aggregate census data from the patient’s zip code

^†distance from patients’ residence to the treatment facility

AML- Acute myeloid leukemia; APL- Acute promyelocytic leukemia; CCI- Charlson comorbidity index; CI- Confidence interval; HCT- Hematopoietic cell transplant; MRC-Myelodysplasia related changes

Subgroup analysis among treated patients

A total of 33,591 patients in our study population received chemotherapy- 45% were 60–69 years of age, 89% were white, 60% received multiagent chemotherapy, and 5% received HCT. Sixty-eight percent had CCI 0, 23% had CCI 1, and 9% had CCI ≥2 (supplement table 4). Patients with CCI 0 were more likely to receive multiagent chemotherapy (61% vs 60% vs 51%) and more likely to undergo upfront HCT (6% vs 4% vs 2%).

One-month mortality was 13%, 19% and 26% for patients with CCI 0, 1, and ≥2 respectively (p<0.001). In a univariate analysis, lower CCI (13% vs 19% vs 26% for CCI 0, 1, and ≥2, respectively, p<0.001) was associated with lower one-month mortality (supplement table 5). Multivariate analysis confirmed significantly worse one-month mortality in patients with CCI 1 (OR 1.5, 95% CI 1.4–1.6) or ≥2 (OR 2.2, 95% CI 2.05–2.5), in comparison to those with CCI 0 (supplement table 6). Younger age, race, higher annual income, academic facility, availability of private insurance, and acute promyelocytic leukemia were associated with improved one-month mortality. Male gender, distance traveled, and facility location (urban/rural) did not influence one-month mortality.

The median one-year OS for patients receiving chemotherapy was 37% (supplement figure 2); median OS for patients with CCI 0, 1, and ≥2 was 40%, 32%, and 24% respectively (p<0.001) (Figure 3). Factors such as treatment at academic center, receipt of multiagent chemotherapy, and receipt of HCT were associated with higher OS in univariate analysis (supplement table 7). In a multivariate analysis, OS was worse with CCI 1 (HR 1.24, 95% CI 1.22–1.28) and ≥2 (HR 1.52, 95% CI 1.46–1.58), compared to CCI 0 (supplement table 8). Other independent factors with improved OS included age 60–69 years, male gender, higher annual income, academic facility, and acute promyelocytic leukemia.

Figure 3:

Overall survival stratified by Charlson comorbidity index (0 vs 1 vs ≥2) for patients who received chemotherapy

Discussion

Outcomes of older adults with AML is generally poor; however, some older adults can achieve remission and attain long-term survival. Predicting chances of survival can help educate patients and families and inform decision-making. While characteristics of AML such as karyotype are frequently used in clinical practice for prognostication, validated measures of comorbidities are less commonly used. CCI is a widely used and validated comorbidity measure that can be easily calculated at the time of diagnosis of AML, thus, providing a tool to further prognosticate patients’ outcomes.

We report the results of one of the largest studies assessing the effect of comorbidity burden on outcomes of older adults with AML. In this real-world analysis, approximately 1/3^rd of patients were found to have CCI of 1 or ≥2; older patients, unlike their younger counterparts, were more likely to have CCI 1 or ≥2. A greater comorbidity burden was associated with lower use of chemotherapy or HCT, lower use of multiagent chemotherapy, and independently predicted worse early mortality and OS in older patients with AML, consistent with results of prior studies (11, 17–22). Different reasons for lower use of chemotherapy or HCT in older adults with significant comorbidities include concerns for chemotherapy intolerance and higher risk of early mortality, as well as paucity of data from clinical trials (3, 8, 23).

In our study, 2/3rd of all patients received chemotherapy; prior studies report utilization of chemotherapy in 30–60% of older adults with AML, although specific treatment type or the intent of treatment is not always reported (6, 17–20, 22, 24). Our results indicate that chemotherapy, one of the most important prognostic factors in AML, is less commonly used in patients with higher CCI (11). Thus, to remove the effects of chemotherapy use on outcomes, we performed a subgroup analysis on only those patients who received chemotherapy. Greater comorbidity burden, indeed, had negative effects on early mortality and OS of patients who received chemotherapy, similar to the results for the entire cohort.

Usefulness of CCI as a prognostic tool has been analyzed in a few studies in older adults with AML with variable results (19, 20, 24–26). Our results are supported by two registry-based studies in which higher CCI predicted early death and worse OS, and chemotherapy was less likely to improve OS in patient with CCI ≥2, compared to CCI 0 or 1 (19, 20). A Danish registry database study of AML patients receiving intensive therapy analyzed the association of outcomes in AML with modified CCI and the number of comorbid diseases (24). In contrast to our results, higher modified CCI score or the presence of comorbidities, after adjustment of performance status and other factors, was not predictive of OS in adult AML patients of any age, including those ≥60 years. Similarly, Tawfik et al. analyzed 144 older patients out of total 277 AML patients treated with intensive induction therapy (25). No significant association between CCI and outcomes, including OS, remission or early mortality was reported; although, presence of diabetes and renal disease independently affected 30-day mortality and remission rates, respectively. The studies that did not demonstrate the effect of CCI analyzed impact of comorbidities only among patients receiving intensive induction therapy, which excludes a sizeable population of older adults who are not fit for intensive therapy and are receiving supportive care or less intensive treatment (25–29).

Other measures of comorbidities are also available. HCT comorbidity index (HCT-CI) has gained popularity in patients undergoing HCT (supplement table 1) (30, 31). The studies that have analyzed HCT-CI as a predictor of outcome in AML patients have reported mixed results (28, 29, 32, 33). For example, a multicenter retrospective study demonstrated that HCT-CI as well as an AML composite model incorporating comorbidity and other AML measures predicted the risk of mortality; however, this study included all adults ≥18 years (34). A single center study of patients ≥ 60 years with newly diagnosed AML and planned intensive chemotherapy did not report any effect of HCT-CI in OS (29). Two studies analyzed effects of both CCI and HCT-CI in AML patients- CCI influenced OS but HCT-CI did not (26, 27). While HCT-CI is valuable to predict outcomes, HCT-CI is not easily computable at the time of diagnosis of AML.

Comorbidities, as an individual disease or as a cumulative burden, can affect outcomes in AML in several ways; the adverse impact may be related to the direct biologic interaction between the comorbidity and the cancer, or the indirect effects of the comorbidity on cancer diagnosis, treatment, and complications (30, 35, 36). Mohammadi et al. reported higher all-cause as well as cancer-specific mortality in AML patients with cerebrovascular, rheumatologic, renal, liver, and psychiatric diseases; highest mortality rate was observed with renal disease (37). Comorbidities may preclude optimal treatment of patients with AML and affect utilization of different drugs; for example, anthracyclines may be contraindicated in patients with severe cardiomyopathy, or total cycles of gemtuzumab ozogamicin may be decreased with hepatic dysfunction (38, 39). Organ dysfunction also results in lower tolerance to treatment-related toxicities and increased risk of life-threatening complications in AML, which may lead to chemotherapy dose-reductions, less intensive therapy, treatment delays, and reduced treatment adherence (40). Drug interactions between comorbidity-specific and cancer-specific treatment may affect management and subsequently, outcomes in AML. However, comorbidities do not always reflect performance status and fitness of the patients, which are important in treatment decisions and overall outcomes in AML (24, 41). Thus, treatment of AML in an older patient with comorbidities should be a collective team effort from oncologists, primary care physicians, geriatricians, pharmacists, nurses, and other ancillary members, with detailed assessment prior to treatment initiation, close supervision during and after the treatment, prompt management of any toxicities or complications, and appropriate supportive measures (29).

Comorbidities, as discussed earlier, have been a significant factor influencing use of intensive chemotherapy which may be associated with superior outcomes compared to lower intensity chemotherapy (11, 24, 25). With significant increase in therapeutic options in last few years, patients can receive effective lower intensity treatments (42). As the data with these newer agents evolves in next few years, it will be interesting to see effects of comorbidities on outcomes.

Our retrospective study has potential limitations. NCDB does not provide data regarding patients’ performance status, detailed molecular and cytogenetic features of AML, and specific chemotherapeutic agents used for treatment, which can affect outcomes. Older patients tend to have high percentage of adverse cytogenetics, adverse mutations, and secondary- or therapy-related disease, which predict lower response rate to frontline therapies (3, 43–46). NCDB does not provide data on response, an important predictor of early mortality and OS (47, 48). In addition, comorbidities alone do not take into consideration patients’ physical fitness. Measures such as geriatric assessment can provide useful information in addition to comorbidity and has been shown to predict OS in AML (29). Our study has several strengths including a large population of unselected older adults with AML treated in real world, inclusion of health system and other factors, and use of CCI, an easily computable measure at the time of diagnosis of AML.

Conclusion

Our real-world study highlights the impact of comorbidity burden in early mortality and OS of older adults with AML. Our results indicate that greater comorbidity is more common with increasing age, correlates with lower likelihood of receiving chemotherapy, and predicts worse outcomes. Trials are warranted to determine the effect of optimized comorbidity management and supportive care to reduce early mortality and improve OS.

Clinical practice points.

• Older adults often have significant comorbidities. Charlson comorbidity index (CCI), which measures comorbidity burden, is known to be of prognostic significance in various underlying diseases.

• We investigated our hypothesis that higher CCI predicts worse one-month mortality and overall survival (OS) in patients ≥60 years with acute myeloid leukemia (AML). In our analysis, patients with CCI 0 were more likely to receive chemotherapy and undergo upfront hematopoietic cell transplant. One-month mortality and OS were significantly worse with CCI 1 or ≥2, compared to CCI 0. Younger age, male gender, higher annual income, academic facility, and longer travel distance were associated with improved OS.

• Greater comorbidity burden is associated with worse early mortality and OS in older patients with AML. Whether optimal comorbidity management may improve outcomes needs to be studied further.