Accumulation of Chronic Disease Among Survivors of Childhood Cancer Predicts Early Mortality

PURPOSE

Cancer survivors develop cancer and treatment-related morbidities at younger than normal ages and are at risk for early mortality, suggestive of an aging phenotype. The Cumulative Illness Rating Scale for Geriatrics (CIRS-G) is specifically designed to describe the accumulation of comorbidities over time with estimates of severity such as total score (TS) which is a sum of possible conditions weighted by severity. These severity scores can then be used to predict future mortality.

METHODS

CIRS-G scores were calculated in cancer survivors and their siblings from Childhood Cancer Survivor Study cohort members from two time points 19 years apart and members of the National Health and Nutrition Examination Survey (NHANES) from 1999 to 2004. CIRS-G metrics were analyzed using Cox proportional hazards regression to determine subsequent mortality risk.

RESULTS

In total, 14,355 survivors with a median age of 24 (IQR, 18-30) years and 4,022 siblings with a median age of 26 (IQR, 19-33) years provided baseline data; 6,138 survivors and 1,801 siblings provided follow-up data. Cancer survivors had higher median baseline TS than siblings at baseline (5.75 v 3.44) and follow-up (7.76 v 4.79), all P < .01. The mean increase in TS from baseline to follow-up was significantly steeper in cancer survivors (2.89 males and 3.18 females) vs. siblings (1.79 males and 1.69 females) and NHANES population (2.0 males and 1.94 females), all P < .01. Every point increase in baseline TS increased hazard for death by 9% (95% CI, 8 to 10) among survivors.

CONCLUSION

Application of a geriatric rating scale to characterize disease supports the hypothesis that morbidity accumulation is accelerated in young adult survivors of childhood cancer when compared with siblings and the general population.

Adult pediatric cancer survivors gain chronic conditions at an accelerated pace relative to the general population.

INTRODUCTION

Five-year survival for children with cancer has improved from 62% in the 1970s to more than 87% today.¹ Nevertheless, the estimated 500,000 childhood cancer survivors in the United States² remain at risk for treatment-related chronic disease in early adulthood.^3,4 Chronic disease confers increased risk for poor health status,⁵ future hospitalization,⁶ and mortality.⁷ Cross-sectional studies indicate that chronic disease rates among young adult survivors of childhood cancer are similar to those of siblings and general population members decades older, suggesting early onset of an aging phenotype that parallels expected disease progression or an accelerated aging phenotype with exponential disease progression that deviates from that in the general population.⁸ Determining if the trajectory of chronic disease progression is accelerated in childhood cancer survivors is important as those with more rapidly progressing disease states may be at risk for early death and require earlier and more intensive clinical management. The Cumulative Illness Rating Scale for Geriatrics (CIRS-G) quantifies burden of chronic disease, characterizing the most common morbidities in geriatric populations.⁹ The scale provides organ system–specific grades, combines morbidities from multiple organ systems into one cumulative score, and provides an intensity rating of disease burden across organ systems. The ASCO guidelines for Geriatric Oncology recommend using the CIRS-G to assess comorbidities in older adults with cancer.¹⁰ We applied the CIRS-G to characterize change in disease burden in childhood cancer survivors over nearly 20 years compared with siblings and the general population. We also described treatment characteristics of survivors at greatest risk for disease burden and subsequent mortality.

CONTEXT

• Key Objective

• Does the accumulation of chronic conditions in adult survivors of pediatric cancer, as compared with their siblings or the general population over time, show evidence of accelerated aging in cancer survivors? Furthermore, is early accumulation of chronic conditions among cancer survivors associated with early mortality?

• Knowledge Generated

• This study shows by applying the Cumulative Illness Rating Scale for Geriatrics (CIRS-G) to adult survivors of childhood cancer (Childhood Cancer Survivor Study) and a general population (the National Health and Nutrition Examination Survey) that cancer survivors over a 19-year period gain additional chronic conditions at an accelerated pace relative to the general population or their siblings. Applying the CIRS-G to a population at baseline can identify a high-risk group that has an increased hazard for early mortality.

• Relevance (S. Bhatia)

• A geriatric rating scale allows identification of childhood cancer survivors at risk for accelerated accumulation of chronic health conditions and premature death.*

  *Relevance section written by JCO Associate Editor Smita Bhatia, MD, MPH, FASCO.

METHODS

Participants

Cancer survivors and siblings were participants in the Childhood Cancer Survivor Study (CCSS). Study details have been described previously.¹¹ In brief, the CCSS is a retrospective cohort of 5-year survivors of childhood cancer, diagnosed with leukemia, Hodgkin lymphoma, non-Hodgkin lymphoma, CNS tumors, neuroblastoma, Wilms tumor, soft-tissue sarcoma, and malignant bone tumors, treated at one of the 31 institutions in North America. This study was approved by institutional review boards at all sites. Participants provided written informed consent before participation. This analysis included survivors diagnosed from 1970 to 1986 and siblings who completed a baseline questionnaire from 1992 to 2004. Data from a follow-up questionnaire, completed from 2014 to 2017, were used to characterize change in chronic disease burden over time. General population estimates used data for the period 1999-2004 from the National Health and Nutrition Examination Survey (NHANES), a surveillance program in the United States of individuals age 0-85 years.¹²

CIRS-G

The CIRS-G, although initially based on in-person comprehensive medical evaluations, has since been validated,^9,13 and data were collected using self-report methodology.^14,15 Disease burden is summarized across 14 systems: heart, vascular, hematopoietic, respiratory, eyes/ears/nose/throat/larynx, upper gastrointestinal, lower gastrointestinal, liver, renal, genitourinary, musculoskeletal/integument, neurological, endocrine/metabolic/breast, and psychiatric illness. Items are scored 0-4, ranging from none to extremely severe. The detailed grading rubric using International Classification of Diseases (ICD)-9 and 10 codes to classify participant self-report of either being told by a physician that they had a specific health condition or self-report of taking medication for a particular condition is provided in the Data Supplement ([Tables S1-S3], online only). The CIRS-G severity measures include (1) total score (TS), a sum of the total number of conditions weighted by severity; (2) total organ system categories endorsed (TC); (3) severity index (SI), TS divided by TC; (4) number of organ system categories at grade 3 or 4 (G3); and (5) number of categories at grade 4 (G4). Severity grading of elements from the CCSS and NHANES questionnaires was completed by one (A.J.E.) and verified by a second investigator (K.K.N.). Discordant results were discussed to reach consensus. Coding for psychiatric illness was completed by an investigator (K.R.K.) and verified by others (A.J.E. and K.K.N.). Comparisons between survivors and NHANES estimates were limited to outcomes queried on both CCSS and NHANES questionnaires and included only persons older than 20 years. For this analysis, the CIRS-G was adapted to accommodate existing self-reported data and did not include laboratory measurements. Thus, certain disease states were not captured, and in some instances, disease severity was uncertain. When applying severity scores in uncertain cases, we conservatively assumed less severe disease.

Late Mortality

Deaths occurring 5 or more years from diagnosis among CCSS participants were ascertained through a search of the National Death Index (NDI 1979-2017) and for NHANES from the public use 2015 NDI data.

Demographic and Treatment Data

Participants reported demographic data on each questionnaire. Trained personnel abstracted diagnosis and treatment data from medical records as part of the parent study.^11,16 Exposure to and administration routes of 49 specific chemotherapy agents were recorded; cumulative dose was abstracted for 26 agents.^11,16 Radiation data included the first and last day of treatment, field size, configuration and laterality, and total treatment doses. Doses to the sites of interest were estimated by applying out-of-beam data in a water phantom to an age-specific mathematical phantom.¹⁷ Covariates evaluated for association with progression of disease burden included age at diagnosis, age at baseline assessment, sex, race, cumulative anthracycline and alkylating agent doses, exposure to platinum agents, and maximum dose of cranial, chest, abdominal, or pelvic radiation. Thoracotomy (grade 4 respiratory condition), amputation (grade 2 musculoskeletal condition), obesity (grade 1 endocrine condition), and smoking (grade 1 and 2 respiratory conditions on the basis of pack years) were not included in models as they are included as part of the CIRS-G grading rubric.

Statistics

Demographic characteristics were compared between survivors and siblings using the Chi-square or Wilcoxon rank sum test as appropriate. Generalized linear regression with a binomial distribution and a log link estimated age- and sex-adjusted relative risk of any grade 1-4 or grade 3-4 CIRS-G condition by organ system or with a normal distribution and an identity link to compare least-square means of CIRS-G composite scores between CCSS survivors and siblings.¹⁸ Models for continuous outcomes did not result in predicted estimates with negative values. Survivor-sibling comparisons included a random effect for family membership.¹⁹ Stabilized inverse probability weights (IPWs) were calculated on the basis of host demographics and, for survivors, treatment differences between participants who did and did not complete the follow-up questionnaires and applied to estimates at follow-up to assess the impact of bias because of loss to follow-up.²⁰ Prevalence of CIRS-G conditions and mean composite scores in NHANES accounted for the NHANES sampling strategy¹² and were standardized to match the age and sex distribution of the CCSS population. Cox proportional hazard models were used to estimate associations with mortality with a start time at completion of the baseline questionnaire and were censored at the time of the second survey or death whichever came first.^21,22 Uno's concordance (c-statistic) was used to evaluate model performance. Mortality analysis in cancer survivors was done both including and excluding survivors who died from cancer recurrence. To account for multiple comparisons, we used a false discovery rate within families of comparisons (survivors v siblings; survivors v siblings v NHANES, treatment-related predictors of disease severity, and mortality) ≤10% and adjusted our 95% CIs accordingly in statistical models.^23,24

RESULTS

Participants

Participants included 14,355 survivors and 4,022 siblings who completed a baseline survey (1992-2005), 6,138 survivors and 1,801 siblings who completed a follow-up survey (2014-2016; Fig 1 flow diagram), and 31,126 individuals from NHANES (1999-2004). Survivors had a median age of 6 years (IQR, 3-13) at diagnosis and 15 years from diagnosis (IQR, 12-19) at the baseline questionnaire. Survivors and siblings were similar in age at baseline and follow-up, with survivors having a median age of 24 (IQR, 18-30) and 43 (IQR, 37-48) years and siblings a median of 26 (IQR, 19-33) and 44 years (IQR, 37-51). Among survivors, leukemia was the most common diagnosis (30.2%; Table 1). Treatment exposures included surgery (68.2%), radiation (71.2%), and chemotherapy (69.4%). Survivors who were alive and eligible but did not complete the follow-up questionnaire were more likely to be male (58.4% v 47.5% among survivors completing both surveys) and less likely to be non-Hispanic White (80.7% v 86.5%; Data Supplement [Table S4]).

FIG 1.

Flow diagram. Those who were ≥20 years old were used in CCSS-NHANES comparison. CCSS, Childhood Cancer Survivor Study; NHANES, National Health and Nutrition Examination Survey.

TABLE 1.

Diagnosis and Treatment-Related Characteristics of the CCSS Cancer Survivor Population, Siblings, and NHANES Cohort

Burden of Disease Among Survivors and Siblings

At baseline, survivors had higher values than siblings on all CIRS-G severity measures: TS, 5.75 versus 3.44; total categories, 2.98 versus 2.37; SI, 1.71 versus 1.19; number of grade 3 or greater conditions (G3), 0.62 versus 0.21; and number of grade 4 conditions (G4), 0.25 versus 0.04 and at follow-up: TS, 7.76 versus 4.79; total categories, 3.84 versus 3.01; SI, 1.84 versus 1.41; G3, 0.82 versus 0.35; and G4, 0.25 versus 0.08 (all P < .01; Data Supplement [Tables S5 and S6]). At baseline, survivors were more likely than siblings to report at least one grade 1-4 condition (relative risk [RR], 1.09 [95% CI, 1.07 to 1.10]; P < .01) and at least one grade 3-4 condition (RR, 2.7 [95% CI, 2.5 to 2.9]; P < .01; Fig 2; Data Supplement [Table S7]). These estimates did not change appreciably when IPWs were applied (Data Supplement [Table S8]). Compared with siblings, survivors were at increased risk for grade 1-4 conditions in all organ systems (RR, 1.13-2.26) except for respiratory (RR, 0.99 [95% CI, 0.95 to 1.03]; P = .63) and genitourinary (RR, 1.02 [95% CI, 0.95 to 1.03]; P = .55) and for grade 3-4 conditions (RR, 1.63-11.72) in all organ systems except genitourinary (Fig 2; Data Supplement [Table S7]). Survivors of all diagnoses had higher mean scores on severity measures than siblings, with survivors of three categories of CNS tumors having the highest mean TS values at baseline and follow-up, ranging from 8.83 to 9.48 and 9.04-10.9, respectively, followed by survivors of six categories of non-CNS solid tumors (5.26-7.37, 6.3-9.37), leukemia/lymphomas (4.66-5.70, 6.7-9.99), and siblings (3.23 and 4.52), all P < .01 (Fig 3; Data Supplement [Table S9]).

FIG 2.

Distribution of CIRS-G grades and relative risks comparing survivors at baseline (n = 14,355) and follow-up (n = 6,138) with siblings at baseline (n = 4,022) and follow-up (n = 1,801). ^aRelative risk–adjusted for age and sex. CIRS-G, Cumulative Illness Rating Scale for Geriatrics; EENT, eyes, ears, nose, and throat.

FIG 3.

CIRS-G severity measures among survivors by cancer diagnosis as compared with siblings presented as means with 95% CIs. CIRS-G, Cumulative Illness Rating Scale for Geriatrics; PNET, primitive neuroectodermal tumor.

Comparison of Cancer Survivors and Siblings With NHANES

Age-standardized mean values (on the basis of the age distribution in CCSS survivor participants) for CIRS-G severity measures and percentages for specific organ system category are shown for cancer survivors, siblings, and participants in NHANES by sex. Mean values for TS, total categories, SI, number of grade 3 or 4 conditions, and number of grade 4 conditions were higher at both baseline and follow-up for survivors when compared with either NHANES data (Data Supplement [Table S10]) or siblings (Data Supplement [Table S11]; Fig 4). In addition, the estimated mean change in severity measures from baseline to follow-up was greater in cancer survivors compared with siblings and NHANES data. Among males, this was true for TS, producing a mean slope difference from baseline to follow-up in survivors (2.89), siblings (1.79), and NHANES (2.0) and extended to the other CIRS-G severity measures: total categories for survivors (1.26), siblings (0.91), and NHANES (0.91); SI for survivors (0.47), siblings (0.36), and NHANES (0.36); number of grade 3-4 categories for survivors (0.32), siblings (0.17), and NHANES (0.25); and number of grade 4 categories for survivors (0.08), siblings (0.03), and NHANES (0.04). For all measures, the slope differences were significant (P < .01) except for the SI, P = .09 (Fig 4). Trends were similar between males and females (Data Supplement [Table S12]).

FIG 4.

CIRS-G measures of severity at baseline and follow-up of cancer survivors (n = 6,138), siblings (n = 1,802) and sex-specific, age-standardized NHANES participants (n = 15,311). Slope differences from baseline to follow-up between cancer survivors and siblings/NHANES for all measures had P < .001 except the severity index (females P = .478, males P = .089). CIRS-G, Cumulative Illness Rating Scale for Geriatrics; NHANES, National Health and Nutrition Examination Survey.

Treatment-Related Predictors of Elevated CIRS-G Severity at Follow-Up

After adjusting for disease and treatment exposures previously associated with morbidity, CIRS-G baseline severity measures (TS, TC, SI, G3, and G4) were independently associated with increased severity measures at follow-up. Disease and treatment exposures associated with severity measures at follow-up included female sex, cranial radiation, chest radiation, abdominal/pelvic radiation, and cumulative platinum dose >600 mg/m² (Table 2). Other variables were only associated with certain CIRS-G measures: age at diagnosis (G4), age at baseline survey (TS, SI, G3, and G4), cumulative alkylating agent dosage (only G3 with dose over 7,000 mg/m²), and cumulative anthracycline dose (only TC with any exposure). Race/ethnicity was not associated with any severity measure. The results did not change appreciably when IPW was applied or when the analysis was limited to only those alive at follow-up (Data Supplement [Tables S13 and S14]). All severity measures at follow-up were greater in those who received chemotherapy and/or radiation vs. those who received surgery alone at follow-up except SI (Data Supplement [Table S15]).

TABLE 2.

Treatment-Related Predictors of Cumulative Disease at Follow-Up Among Survivors Who Completed Both the Baseline and Follow-Up Surveys

Mortality

In models adjusted for age at baseline survey, sex, and race/ethnicity, all CIRS-G severity measures predicted mortality in survivors, siblings, and the NHANES group (Fig 5; Data Supplement [Table S16]). Hazard ratios (HRs) estimates for each severity measure were similar across patient populations (overlapping 95% CI). The c-statistics for the models were higher in the NHANES group (0.86-0.86) than in the cancer survivors (0.67-0.70) and siblings (0.71-0.73). This hazard of mortality was increased by 9% for each unit increase in TS, by 35% for each unit increase in SI, and by 91% for each grade 4 condition. When death from cancer recurrence was excluded in the mortality analysis among survivors, the c-statistics mildly improved for all severity measures (0.68-0.70).

FIG 5.

Graphical representation of the impact of each CIRS-G severity measure on mortality (baseline-2017) using separate Cox proportional hazard models for each group: cancer survivors, siblings, and NHANES. Each model is additionally adjusted for age, sex, and race. No. at risk for each model cancer survivors (n = 13,065), siblings (n = 4,022), and NHANES (n = 15,311). C-statistics for these models ranged from 0.67 to 0.70 for cancer survivors, 0.71-0.73 for siblings, and 0.86-0.86 for NHANES population. CIRS-G, Cumulative Illness Rating Scale for Geriatrics; NHANES, National Health and Nutrition Examination Survey.

DISCUSSION

In this large and well-characterized population, we provide convincing evidence of an accelerated burden of chronic disease among childhood cancer survivors when compared either with their siblings or the general population. With approximately 20 years of follow-up, these results give the best evidence to date of accelerated aging among childhood cancer by providing direct comparison with two normative samples for longitudinal change in burden.^25,26 The results also allow comparison of disease burden with metrics available in the general population.

The CIRS-G mechanism has previously been used to characterize disease burden and severity in elderly populations including those with cancer,^27–39 hip fracture,^40,41 those requiring hospitalization,^42–45 presenting to the emergency department,⁴⁶ and delirium.⁴⁷ Although much younger in chronological age, the use of this geriatric scale provides a robust approach for comparing the trajectory of aging among childhood cancer survivors, siblings, and the general population and did a reasonable job in predicting mortality over this study period (c-statistics, 0.67-0.86). Further study will determine if this precision increases as the cohort ages.

Our data indicate that more than 90% of childhood cancer survivors had at least one grade 1-4 problem and 43% had at least one grade 3-4 problem (increasing to >50% at follow-up) despite being young adults. Survivors previously diagnosed with CNS tumors had the most morbidity. This is not surprising as patients with CNS tumor often have CNS injury from surgery, chemotherapy, and high-dose cranial radiation. Disruption of CNS top-down regulation translates into multiple organ system problems.⁴⁸ Beyond the expectation that late effects are frequent in adults previously treated for CNS disease, it is also significant that survivor participants had a higher score for every measure of disease severity relative to siblings no matter what their original cancer diagnosis, indicating that all types of cancer lead to increased morbidity.

These analyses also indicate that relative to noncancer survivor populations, the trajectory of disease burden in childhood cancer survivors is accelerated. Although differences in disease prevalence by age are reported in studies comparing cancer survivors with others,^4,7 these data provide longitudinal evidence of an accelerated accumulation of chronic disease in survivors when compared with siblings and the general population. Thus, aging in childhood cancer survivors does not appear to parallel that in those without a history of childhood cancer,⁴⁹ reinforcing that this population remains differentially vulnerable to disease throughout life. These data indicating accelerated aging is supported by mechanistic work among childhood cancer survivors in the St Jude Lifetime Cohort where telomere shortening⁵⁰ and epigenetic age acceleration⁵¹ are cross-sectionally associated with chronic disease prevalence, supporting the need for continued work designed to describe and determine longitudinal biologic changes in this population so that targets can be identified for preventive interventions.⁵²

In addition to documenting the prevalence and trajectory of disease burden, the CIRS-G reasonably predicted mortality in adult survivors of childhood cancer. Regardless of which CIRS-G metric was used, the c-statistics were robust (all approaching 0.7) and significant. Although high CIRS-G scores are associated with short-term mortality (3 months-1 year) in elderly populations with and without cancer,^{31–34,42,43,45–47} it has been infrequently used to predict long-term mortality as in this study. With just 34.2 months of follow-up, Wedding et al³⁵ reported a HR for mortality of 1.4 (95% CI, 1.01 to 2.00) among 427 patients with cancer with both hematologic and solid tumor malignancies, age 18 to ≥80 years, comparing those with and without CIRS-G grade 3-4 conditions. Using the TS rubric, Groome et al³⁶ reported a relative risk of 10-year mortality of 1.64 (95% CI, 1.52 to 1.76) among patients with prostate cancer (n = 2,132; mean age, 66.7 years), comparing TSs of 1 versus 0. These authors also reported increased hazard for mortality of 1.18 (95% CI, 1.15 to 1.21) for each one point increase in TS.³⁶ The CIRS-G is also used in elderly patients with hip fractures to predict fall risk⁴¹ and rehabilitation outcomes⁴⁰ and to predict hospital readmissions in elderly patients with and without cancer.^27,42,43 Thus, this instrument may have utility to predict health outcomes in adult survivors of childhood cancer as they move into older adulthood. As a measure of the accumulation of disease, the CIRS-G could be used by clinicians to identify individuals at the greatest risk for early mortality and target them for close monitoring of morbid conditions and potential intervention.

This study has some limitations. Because of the lack of laboratory data or reason for medication use and the intent to classify disease severity conservatively in their absence, true burden of morbidities could be even higher than reported. There were also limitations to direct comparison of CCSS with NHANES data as only comorbidities captured by both questionnaires could be used. As not all CCSS participants completed the follow-up survey, selection bias is possible. However, when IPW was applied to analysis, the results did not change. Finally, only US death statistics were available; no Canadian survivors were included in mortality analysis. Death from cancer recurrence is difficult to predict, and in fact, the discrimination of the mortality prediction improved when these deaths were excluded. For this analysis, comparison was limited to two time points where chronic disease queries were the same and to survivors treated from 1970 to 1986 because they had the longest follow-up. Future work will include additional time points and survivors treated from 1987 to 1999 as the cohort continues to collect data and survivors age.

In conclusion, childhood cancer survivors have increased burden of disease across organ systems relative to siblings and the general population. This gap increases over time. The CIRS-G provides evidence that accelerated aging is likely in cancer survivors and provides a tool for follow-up and anticipation of clinical needs over time.

DATA SHARING STATEMENT

Study data are available from the investigators upon request.

AUTHOR CONTRIBUTIONS

Conception and design: Adam J. Esbenshade, Lu Lu, Debra L. Friedman, Kevin C. Oeffinger, Gregory T. Armstrong, Wendy M. Leisenring, Amy Sketch, Leslie L. Robison, Kirsten K. Ness

Financial support: Gregory T. Armstrong

Administrative support: Gregory T. Armstrong, Kirsten K. Ness

Provision of study materials or patients: Gregory T. Armstrong, Joseph P. Neglia, Leslie L. Robison

Collection and assembly of data: Adam J. Esbenshade, Lu Lu, Gregory T. Armstrong, Joseph P. Neglia, Rebecca Howell, Robyn Partin, Leslie L. Robison, Kirsten K. Ness

Data analysis and interpretation: Adam J. Esbenshade, Lu Lu, Debra L. Friedman, Kevin C. Oeffinger, Kevin R. Krull, Wendy M. Leisenring, Kirsten K. Ness

Manuscript writing: All authors

Final approval of manuscript: All authors

Accountable for all aspects of the work: All authors

AUTHORS' DISCLOSURES OF POTENTIAL CONFLICTS OF INTEREST

Accumulation of Chronic Disease Among Survivors of Childhood Cancer Predicts Early Mortality

The following represents disclosure information provided by authors of this manuscript. All relationships are considered compensated unless otherwise noted. Relationships are self-held unless noted. I = Immediate Family Member, Inst = My Institution. Relationships may not relate to the subject matter of this manuscript. For more information about ASCO's conflict of interest policy, please refer to www.asco.org/rwc or ascopubs.org/jco/authors/author-center.

Open Payments is a public database containing information reported by companies about payments made to US-licensed physicians (Open Payments).