Predictive Ability of Cancer and Aging Research Group (CARG) Chemotherapy Toxicity Calculator in Hematologic Malignancy

Abstract

Introduction:

Chemotherapy toxicity tools are rarely studied in patients with hematologic malignancy (HM). The primary aim of this pilot study was to determine the predictive ability of the Cancer and Aging Research Group (CARG) chemo-toxicity calculator in estimating grade 3–5 toxicity in patients with HM.

Materials and Methods:

Patients 60 years and older with HM were prospectively evaluated using the CARG chemo-toxicity calculator. Discrimination and calibration were checked by applying the published model in our data. Additionally, a full geriatric assessment (GA), the Short Physical Performance Battery (SPPB), and health related quality of life (HRQoL) were captured longitudinally at the start of treatment and at end of study. Secondary aims explored the association of GA metrics with chemo-related toxicities and survival.

Results:

One hundred forty-five patients were approached, 118 patients consented, and 97 patients were evaluable. Most patients were newly diagnosed (n=91). The median CARG score was 9 (range 4–18). The CARG score was not validated in our cohort of older patients with HM, with area under the receiver operation characteristic curve being 0.53 (95% CI: 0.41–0.65). In multivariable analysis, after controlling for disease type, risk factors associated with grade 3–5 toxicity included living alone (hazard ratio [HR] 4.24, 95%CI: 2.07–8.68, p<0.001), increase in body mass index (HR 1.06, 95%CI: 1.01–1.12, p=0.03) and a higher social activities score (HR 1.27, 95%CI: 1.06–1.51, p=0.01). In multivariable analysis of overall survival, the only prognostic factor was an objective marker of physical function (SPPB score HR=0.85, 95%CI:0.78–0.93, p<0.001).

Discussion:

The CARG chemo-toxicity calculator was not predictive of grade 3–5 toxicity in patients with hematologic malignancy. The SPPB was associated with overall survival in multivariable analysis, suggesting future use as an objective biomarker in HM. We also report a comprehensive trajectory of function, QoL, psychosocial well-being, and cognition among older adults with HM. The predictive accuracy of the CARG chemo-toxicity calculator may be affected by the diverse range of HM treatment options that are not traditional chemotherapy.

INTRODUCTION

Aging adults with blood cancer are a growing demographic. Nearly a third of all patients diagnosed with a hematologic malignancy (HM) are older than 75 years of age.¹ Treatment approaches for older adults with hematologic malignancies are highly heterogeneous, including traditional cytotoxic chemotherapy, immunotherapy, novel targeted agents, hematopoietic stem cell transplant, and cellular therapy. The identification of factors that contribute to frailty in aging adults with blood cancer is critical in defining a treatment plan. The lack of geriatric-specific evidence perpetuates uncertainty in what to expect regarding important clinical outcomes such as toxicity, functional health, and health-related quality of life (HRQoL) in patients with blood cancer.^2–4 Without supporting data, clinicians are left to estimate the risk for drug toxicity based on clinical factors such as age, comorbidities, and performance status. Yet these metrics alone are not a reliable estimate of life expectancy, functional health, or risk of treatment complications.^5,6 Moreover, treatment intensity is highly variable for high-acuity cancers, such as hematologic malignancies. Currently, there is no consensus on a standard assessment or predictive biomarkers that can identify which older adults with hematologic malignancy are most likely to benefit from a more intensive treatment versus those patients who may be better served by a less intensive approach. Longitudinal studies embedding comprehensive metrics of physical and mental health, as well as HRQoL, are needed to track the progression of the disease and the effectiveness of the treatment over time. A comprehensive understanding of the patients’ health can improve patient outcomes and enhance the overall quality of care delivered.

Frailty is a dynamic measure that is an age-related state of increased vulnerability with limited physiologic reserve to acute stress.⁷ Frailty can be captured using an in-depth comprehensive geriatric assessment (GA) in which a patient undergoes a multi-dimensional evaluation of functional status, social support, cognitive evaluation, psychological status, nutritional deficits, co-morbidity, and polypharmacy.⁸ GAs are shown to predict mortality and toxicity^9–11, independent of performance status and age,¹² and can provide a more accurate classification of frailty than clinical judgement alone.¹³ However, patients with blood cancer are rarely the primary focus¹⁴ and GAs are often time- and resource-prohibitive in an oncology setting. For this reason, simplified oncology-specific GA tools have been created to aid clinicians’ ability to estimate chemotherapy risk. The Cancer and Aging Research Group (CARG) chemo-toxicity calculator^12,15,16 is an established tool, used widely to assess the risk of treatment toxicity in older adult oncology patients with solid tumors.¹² Importantly, the CARG chemo-toxicity calculator ¹² has not been established in patients diagnosed with HM.

The primary aim of this pilot study was to determine the predictive ability of the CARG chemo-toxicity calculator in estimating grade 3–5 toxicity in patients with HM. As secondary aims, longitudinal changes in health were measured, over time, using the full CARG GA, including physical function [Short Physical Performance Battery (SPPB)] and HRQoL (Patient-Reported Outcome Measurement Information System [PROMIS]). We then explored the relationship of these GA variables with chemotherapy-related adverse events (AEs) and survival.

METHODS

We conducted a single-institution prospective study, approved by The Ohio State University’s Institutional Review Board and written informed consent was obtained for all patients in accordance with the Declaration of Helsinki.

In this longitudinal study we evaluated the accuracy and predictive ability of the CARG chemo-toxicity calculator using the simplified online tool.^12,16,17 Patients ≥60 were recruited from the hematology clinics at The James Comprehensive Cancer Center at The Ohio State University. Eligible patients had untreated hematologic malignancy or a relapsed hematologic malignancy with intent to begin chimeric antigen receptor (CAR) T-cell therapy. Patients could go on to receive a hematopoietic stem cell transplant if indicated.

At the start of treatment, the CARG chemo-toxicity calculation was completed at baseline prior to start of treatment. Patients also completed the full CARG GA:^12,16,17 The full GA includes self-reported measures of functional status, comorbidity, medications, nutrition, mental health, and social support/function as previously published. Specific measures include the Older Americans Resources and Services (OARS) ¹⁸ instrumental activities of daily living (IADL), Medical Outcomes Study (MOS) physical function¹⁹, patient-reported Karnofsky Performance Status (KPS) ²⁰, MOS social support²¹, MOS social activity score²¹, self-report of depression²² and anxiety, and the Mental Health Index (MHI). ¹⁹ A clinical research coordinator administered the cognitive screening using the Blessed Orientation-Memory-Concentration (BOMC) ²³ (Supplemental Table 1).

Additional metrics, focused on HRQoL and objective measures of physical performance, were included to enhance the baseline performance of the GA. Evaluations included the PROMIS Global Health Scale Short Form v1.1,²⁴ and a clinical research coordinator administered the physical function assessment using the SPPB.²⁵

Clinical data and laboratory data were captured at baseline and longitudinally. The type of treatment including drugs and doses (planned and actual) were recorded along with any supportive care medications given throughout the course of chemotherapy. The National Cancer Institute (NCI) Common Technology Criteria for Adverse Events (CTCAE) (V5.0)²⁶ grade 1–5 toxicities, hospitalizations, dose delay, reductions, and discontinuation were captured with the relationship of these events to toxicity at the beginning of every treatment cycle, or monthly, whichever came first. Relative dose intensity (RDI) ²⁷was calculated as the ratio of delivered dose intensity (DDI) to the standard dose intensity (SDI) (standard dose, standard cycle length). An RDI of 100% indicates that treatment was administered at the dose planned per the treating physician, without delay or cancellation. Two physicians reviewed the patients’ chemotherapy toxicity attribution. Data were captured at baseline (within 45 days of chemotherapy receipt) and end of study (EOS). EOS visits occurred at the earliest of the following events: completion of chemotherapy, disease progression, prior to hematopoietic stem cell transplant, or one year on study (within 45 days). Baseline and EOS data are presented herein. Patients received treatment (i.e., chemotherapy, immunotherapy, targeted agents, bone marrow transplant, or other) as ordered by their treating physician.

Statistical Analysis:

Descriptive statistics were used to summarize patient characteristics with median and range provided for continuous variables, and frequency and percentage calculated for categorical variables. To estimate if GA metrics changed significantly across visits, Wilcoxon signed-rank test was used to test continuous variables and McNemar’s test was used to test categorical variables.

AE data were described and graded per the NCI CTCAE v5 guidelines. For each AE, information collected includes event description, time of onset, grade, and attribution to chemotherapy. The maximum grade for each type of toxicity was recorded for each patient and the frequency counts were tabulated accordingly. AEs were summarized both regardless of attribution and for those at least possibly related to chemotherapy.

To validate the CARG chemo-toxicity risk score, we obtained the logistic model from the original publication¹² including the intercept and the coefficients for all risk factors and calculated the linear predictor for each patient in our cohort. Then a univariable logistic regression model of AE on this linear predictor was fit. The model discrimination was evaluated through the area under the receiver operation characteristic (ROC) curve, and the model calibration was evaluated using the Wald test on the joint null hypothesis of intercept being 0 and slope being 1.

To account for EOS variations, we employed time-to-event analysis by calculating time to the occurrence of first grade 3–5 chemotherapy-related AE from the date of study enrollment to the date of such AE for each patient regardless of AE type, treating death without grade 3–5 chemo-related AEs as the competing risk. The cumulative incidence function was used to estimate the cumulative incidence rate (CIR) of chemo-related toxicities. We then fit Fine and Gray models²⁸ to estimate the correlation between the outcome and patient characteristics. Univariable models were first fit for individual risk factors, then variables reflecting a p-value of ≤ 0.2 were incorporated together in a full model, from which a backwards selection procedure was used to determine the final multivariable model where only variables with p-values of ≤0.05 were included.

Overall survival (OS) was calculated from date of study enrollment to death due to all causes, censoring patients who were still alive at date of last follow-up. A Cox proportional hazards model was used to correlate GA metrics with OS. The same model building procedure was used to obtain the final multivariable model for OS. SAS software version 9.4 was used for statistical analysis. All tests were two-sided, and statistical significance was declared at α=0.05. No adjustment was made for multiple testing.

RESULTS

Sample Cohort:

From September 2018 to March 2020, 145 patients were approached, 18 (12%) patients declined enrollment, and nine (6%) patients were ineligible. In total 118 patients consented, 10 (8%) patients died prior to GA administration, eight (8%) patients elected care at an outside hospital, and three (2%) patients withdrew (Figure 1A). The enrolled cohort includes 97 patients, with 76 patients completing an EOS visit (Figure 1B). In total, 20 patients died from the time of consent to EOS (16.9%). The median duration on study was 171.5 days (range 49–434).

Figure 1:

Study Consort

Patient Baseline Characteristics:

The study sample included 97 patients with hematologic malignancy (Supplemental Table 2) including acute leukemia (n=34, 35%), lymphoma (n=29, 30%), plasma cell disorders (PCD n=28, 29%), and chronic lymphocytic leukemia (CLL, n=6, 6%). The median age of participants was 70 (range: 60–88 years) with more males than females (58% vs 42%) and the study population was predominately White (n=94). Treatment was highly variable and included traditional chemotherapy, hypomethylating agents, immunotherapy, and targeted agents with most patients receiving multi-drug treatment (n=79) (Supplemental Table 3). Most patients were newly diagnosed (n=91) and few patients with relapsed disease planning to receive CAR-T therapy (n=6) were enrolled.

Primary Endpoint CARG Chemotherapy Toxicity Validation:

Complete toxicity data was captured in 77 patients: 334 grade 1–2 AEs occurred in 75 patients and 82 grade 3–5 AEs occurred in 42 patients (55%). Hematologic toxicities were more common with 36 (37%) patients having anemia (30 grade 1–2 and 6 grade 3–5), and 11 (11%) having febrile neutropenia (all grade 3–5). Therapeutic toxicities were documented at the start of every treatment cycle. The number of patients with first grade 3–5 chemotherapy-related adverse events is outlined in Table 1. The cumulative incidence rates of a chemotherapy-related grade 3–5 AE over time were as follows: 1-month, 10.4% (95% confidence interval [CI]: 5.3–17.5%); 3-month, 26.0% (95% CI: 17.7–35.2%); 6-month: 39.6% (95% CI: 29.8–49.2%); and 12-month: 42.7% (95% CI: 32.7–52.4%).

Table 1.

Number of patients with grade 3–5 chemotherapy-related toxicities (n, % of 97 patients)

Adverse event category	Adverse event term	Grade 3-5
	All adverse events
Hematologic		n=26
Blood or lymphatic system disorder	Febrile neutropenia*	11 (11.3)
	White blood cell decreased	7 (7.2)
	Anemia	6 (6.2)
	Platelet count decreased	1 (1)
	Blood and lymphatic system disorders - other, specify	1 (1)
Nonhematologic		n=56
Cardiac disorder	Atrial fibrillation	1 (1)
	Heart failure	1 (1)
	Myocardial infarction	1 (1)
Endocrine disorder	Endocrine disorders - other, specify	1 (1)
Gastrointestinal disorder	Nausea / diarrhea / vomiting	3 (3.1)
	Mucositis oral	3 (3.1)
	Abdominal pain	1 (1)
	Lower gastrointestinal hemorrhage	1 (1)
	Hepatic injury	1 (1)
General disorder or administration site conditions	Fatigue	2 (2.1)
Infections or infestations	Lung infection*	7 (7.2)
	Sepsis	6 (6.2)
	Urinary tract infection	3 (3.1)
	Upper respiratory infection	2 (2.1)
	Bacteremia	1 (1)
	Infections and infestations - other, specify	1 (1)
Metabolism or nutrition disorder	Anorexia	1 (1)
	Dehydration	1 (1)
Nervous system disorder	Dizzy syncope	6 (6.2)
	Peripheral sensory neuropathy	1 (1)
	Encephalopathy	1 (1)
Respiratory, thoracic or mediastinal	Dyspnea/cough*	4(4.1)
disorder	Pneumonitis*	1 (1)
	Pulmonary edema	1 (1)
Skin or subcutaneous tissue disorder	Rash	2 (2.1)
Vascular disorder	Thromboembolic event	2 (2.1)
	Hypertension	1 (1)

^*Grade 5 chemotherapy related adverse events (n=5) included febrile neutropenia (n=1), lung infection (n=2), dyspnea (n=1), and pneumonitis (n=1). Patients could have experienced more than 1 grade 3–5 event.

Abbreviations: AE, adverse event.

Validation of CARG Chemo-Toxicity Score:

Implementing the univariable logistic regression model of AE on the linear predictor based on the original CARG chemo-toxicity risk score paper, the area under the ROC curve was 0.53 (95% CI: 0.41–0.65), indicating an insufficient discrimination of the model. The Wald test of the joint null hypothesis that intercept being 0 and slope being 1 showed a p-values of <0.001, meaning we would reject the joint null hypothesis and the model calibration was inadequate. The median CARG chemo-toxicity risk score was 9 (range 4–18), with 47 patients categorized as high risk (>9), 39 patients categorized as mild risk (6–9), and 11 patients as low risk (0–5). Figure 2A presents the cumulative incidence of grade 3–5 chemotherapy-related toxicity by the CARG chemo-toxicity risk score group, there was no significant difference in CIR of toxicities between the three groups (p=0.16).

Fig. 2.

a. The CARG chemo-toxicity score was not predictive of cumulative incidence of grade 3–5 chemotherapy-related toxicity, with no significant difference in CIR of toxicities between the three groups (p = 0.16). b.In multivariable analysis, controlling for cancer type, the only factor predictive of death was the SPPB score (HR = 0.85, 95 %CI:0.78–0.93, p < 0.001).

Secondary Endpoints

Change in Geriatric Assessment Metrics over Time:

Several domains of health improved significantly with cancer treatment (Table 2). QoL, as measured by the PROMIS QoL, improved from a median of 32 (range 12–49) to 35 (range 16–47) (p=0.05). The median score in IADL improved from 13 (range 5–14) to 14 (range 7–14) (p=0.002). At diagnosis, 51% of patients (n=49) self-reported a KPS of >80, this rate increased to 80% at EOS (n=56, p<0.001). Physical function improved from a median score of 5 (range 0–12) at baseline to 9 (range 0–12) at EOS (p=0.005). Most patients had functional impairment (SPPB <9) at the start of treatment and by EOS, the proportion of patients with functional impairment decreased from 71% (n=66) to 52% (n=27). Mental health improved (MHI-17) from a median of 84.7 (range 35.3–100) to 89.4 (range 40–100) (p<0.001). Mild/moderate impairments in cognition, using the BOMC screen, was identified among 32% (n=31) of patients. Prescription and non-prescription medications were numerous, including prescribed medications (median n=5; range 0–14), over-the-counter medicine (n=1; range 0–10), and herbs or vitamins (n=1; range 0–10). Involuntary weight loss within the 6 months prior to diagnosis was common (n=36 [37%]). Body mass index (BMI) declined from treatment baseline median of 28.8 (range 18.7–53.3), to 27.8 (range 19.5–54.5) at EOS (p=0.03).

Table 2.

Change in Geriatric Metrics from Baseline to End of Study (EOS)

Domains of GA	Median (range), or n (%)		p-value
	Baseline (N=97)	EOS (N=70)
QoL- PROMIS	32 (12–49)	35 (16–47)*	0.05
Function
IADL	13 (5–14)	14 (7–14)	0.002
MOS Physical Health score	44.4 (0–100)	55.6 (5.6–100)	0.35
Self-Report KPS	80 (30–100)	80 (50–100)	<0.001
0–40	4 (4)	0 (0)
50–70	44 (45)	14 (20)
80–100	49 (51)	56 (80)
Falls within 6 months	0 (0–5)	0 (0–3)	0.14
0	78 (80)	64 (91)
>=1	19 (20)	6 (9)
SPPB score	5 (0–12)	9 (0–12)	0.005
Normal >9	27 (29)	25 (48)
Physical impairment <=9	66 (71)	27 (52)
Unknown	4	18
Comorbidities	6 (2–12)	5 (2–11)	0.03
Psychosocial Support
Mental health score	84.7 (35.3–100)	89.4 (40–100)	<0.001
Social activities score	41.7 (0–91.7)	50 (0–83.3)	0.22
Social support score	97.9 (27.1–100)	100 (33.3–100)	0.99
Cognition
BOMC score	4 (0–22)	0 (0–20) *	<0.001
Normal cognition 0–4	66 (68)	63 (91)
Questionable cognition 5–10	27 (28)	5 (7)
Cognitive impairment >=11	4 (4)	1 (1)
Pharmacy
Prescribed	5 (0–14)	7 (1–19)	0.001
Unknown	4	0
Over-the-counter	1 (0–10)	2 (0–7)	0.003
Unknown	4	0
Herbs and vitamins	1 (0–10)	2 (0–11)	0.06
Unknown	4	0
Nutrition
BMI	28.8 (18.7–53.3)	27.8 (19.5–54.5)	0.03
Unknown	1	6

^*missing (n=1) Abbreviations: BMI, body mass index; BOMC, blessed orientation memory concentration; EOS, end of study; GA, geriatric assessment; IADL, instrumental activities of daily living; KPS, Karnofsky performance status; MOS, medical outcomes study; PROMIS, patient-reported outcomes measurement information system; QoL, quality of life; SPPB, short physical performance battery

Relationship of Geriatric Metrics with Outcomes of Toxicity and Survival:

In multivariable analysis, after controlling for disease type, risk factors associated with significant grade 3–5 toxicity included living alone (hazard ratio [HR] 4.24, 95%CI: 2.07–8.68, p<0.001), a higher social activities score (HR 1.27, 95%CI: 1.06–1.51, p=0.01), and an increase in BMI (HR 1.06, 95%CI: 1.01–1.12, p=0.03) (Table 3). Risk factors associated with an increased risk of death in univariable models were advanced Eastern Cooperative Oncology Group performance status (HR 4.35, 95%CI: 2.43–7.79, p<0.001), cognitive impairment (BOMC score HR 1.08, 95%CI: 1.01–1.14, p=0.02), and an increased number of comorbidities (HR 1.26, 95%CI: 1.07–1.48, p=0.006). Factors that were protective included better self-reported physical function (MOS Physical Health: HR 0.86, 95%CI: 0.77–0.97, p=0.01), independence in IADL (HR 0.83, 95%CI: 0.73–0.95, p=0.007), better objective markers of physical function (SPPB HR 0.87, 95%CI: 0.80–0.95, p=0.001), increased social activity (HR 0.81, 95%CI: 0.68–0.95, p=0.01), better mental health (MHI HR 0.78, 95%CI: 0.62–0.97, p=0.03), and higher number of supplements (HR 0.76, 95%CI: 0.60–0.96, p=0.02). A higher CARG score was associated with an increased risk of death (HR 1.12, 95%CI: 1.02–1.23, p=0.02) (Table 4). In a post-hoc, subset analysis (n=47) evaluating patients who received traditional chemotherapy only, the CARG score was not predictive of grade 3–5 adverse events or OS (HR 1.10 95%CI 0.98–1.23, HR 0.98 95%CI 0.85–1.14, respectively). In multivariable analysis, the only factor predictive of death was related to objective markers of physical function (SPPB score HR=0.85, 95%CI:0.78–0.93, p<0.001) after controlling for cancer type (Figure 2B).

Table 3.

Treatment-Related Model for Development of Grade 3–5 Adverse Events

	Univariable Models		Multivariable Model
	Hazard Ratio (95% CI)	p-value	Hazard Ratio (95% CI)	p-value
Variables
Age, 5-year increase	0.99 (0.80–1.22)	0.94	----	----
Diagnosis, vs AML/MDS/ALL  Chronic lymphocytic leukemia  Lymphoma  Plasma cell disorder	0.37 (0.09–1.48) 0.32 (0.14–0.73) 0.37 (0.18–0.78)	0.16 0.007 0.009	0.58 (0.13–2.72) 0.31 (0.14–0.71) 0.29 (0.13–0.64)	0.49 0.006 0.002
Household composition
In marriage vs. not in marriage	0.59 (0.32–1.10)	0.10	----	----
Live alone vs. not live alone	2.29 (1.23–4.25)	0.009	4.24 (2.07–8.68)	<0.001
Function
ECOG PS, 1-unit increase	0.92 (0.55–1.54)	0.74	----	----
IADL	1.04 (0.90–1.20)	0.59	----	----
SPPB Score	0.98 (0.92–1.05)	0.53	----	----
MOS Physical Health Score, 10-point increase	0.99 (0.90–1.10)	0.90	----	----
Self-Report Health Rating score, 10-point increase	0.95 (0.79–1.13)	0.55	----	----
No. of falls in the past 6 months	1.01 (0.70–1.46)	0.96	----	----
Comorbidities score	1.07 (0.94–1.21)	0.32	----	----
Psychosocial Support
Mental health score, 10-point increase	1.04 (0.82–1.31)	0.78	----	----
Social activities score, 10-point increase	1.11 (0.96–1.28)	0.15	1.27 (1.06–1.51)	0.01
Social support score, 10-point increase	1.07 (0.88–1.29)	0.51	----	----
Religion/spirituality score	0.97 (0.92–1.02)	0.23	----	----
Cognition
Blessed score	1.01 (0.94–1.10)	0.75	----	----
Pharmacy
Prescribed	1.08 (0.99–1.19)	0.09	----	----
Over-the-counter	1.02 (0.87–1.20)	0.81	----	----
Herbs and vitamins	0.94 (0.82–1.07)	0.33	----	----
RDI, 10% increase	0.75 (0.62–0.92)	0.005	0.80 (0.65–1.00)	0.05
Nutrition
BMI, 1-unit increase	1.03 (0.99–1.08)	0.20	1.06 (1.01–1.12)	0.03
Involuntary weight loss past 6 months vs. none	1.05 (0.56–1.95)	0.89	----	----
Involuntary weight loss past 6 months, 5- pound increase	1.13 (0.99–1.29)	0.06	----	----
HRQoL
PROMIS Score, 10-point increase	1.06 (0.65–1.71)	0.83	----	----
CARG chemotherapy toxicity score	1.02 (0.95–1.11)	0.56	----	----
CARG chemotherapy toxicity score without Hemoglobin	1.00 (0.90–1.11)	0.95	----	----

42 events, death as the competing risk Abbreviations: ALL, acute lymphoblastic leukemia; AML, acute myeloid leukemia; BMI, body mass index; CARG, Cancer and Aging Research Group; CI, confidence interval; ECOG PS, Eastern Cooperative Group performance status; IADL, instrumental activities of daily living; MDS, myelodysplastic syndrome; MOS, medical outcomes study; PROMIS, patient-reported outcomes measurement information system; RDI, relative dose intensity; SPPB, short physical performance battery

Table 4.

Cox Model for Risk of Death

	Univariable Models		Multivariable Model
	Hazard Ratio (95% CI)	p-value	Hazard Ratio (95% CI)	p-value
Variables
Age, 5-year increase	1.19 (0.93–1.51)	0.17	----	----
Diagnosis, vs AML/MDS/ALL  CLL  Lymphoma  PCD	0.23 (0.03–1.72) 0.28 (0.11–0.70) 0.21 (0.08–0.56)	0.15 0.006 0.002	0.17 (0.02–1.25) 0.23 (0.09–0.58) 0.18 (0.07–0.50)	0.08 0.002 0.001
Household composition
Married/partnered	2.28 (0.93- 5.61)	0.07	----	----
Living alone	0.56 (0.19–1.60)	0.28	----	----
Function
ECOG PS, 1-unit increase*	4.35 (2.43–7.79)	<0.001	----	----
IADL	0.83 (0.73–0.95)	0.007	----	----
SPPB Score	0.87 (0.80–0.95)	0.001	0.85 (0.78–0.93)	<0.001
MOS Physical Health Score, 10-point increase	0.86 (0.77–0.97)	0.01	----	----
Self-reported KPS, 10-point increase	0.85 (0.70–1.02)	0.07	----	----
No. of falls in the past 6 months	0.87 (0.54–1.42)	0.58	----	----
Comorbidities score	1.26 (1.07–1.48)	0.006	----	----
Psychosocial Support
Mental health score, 10-point increase	0.78 (0.62–0.97)	0.03	----	----
Social activities score, 10-point increase	0.81 (0.68–0.95)	0.01	----	----
Social Support Score, 10-point increase	1.03 (0.83–1.29)	0.78	----	----
Cognition
Blessed Score	1.08 (1.01–1.14)	0.02	----	----
Pharmacy
No. of prescribed medication	1.08 (0.97–1.19)	0.17	----	----
No. of over-the-counter medication	0.98 (0.80–1.20)	0.84	----	----
No. of herbs and vitamin	0.76 (0.60–0.96)	0.02	----	----
Total number of medications (prescribed and over-the-counter)	1.05 (0.96–1.14)	0.29	----	----
RDI, 10% increase	0.82 (0.66–1.01)	0.06	----	----
Nutrition
BMI, 5-unit increase	1.22 (0.96–1.54)	0.10	----	----
Involuntary weight loss in past 6 months vs. none	1.10 (0.54–2.22)	0.80	----	----
Involuntary weight loss in past 6 months, 5-pound increase	0.98 (0.88–1.11)	0.78	----	----
HRQoL
PROMIS Score, 10-point increase	0.45 (0.27–0.77)	0.004	----	----
CARG chemotherapy toxicity score	1.12 (1.02–1.23)	0.02	----	----

33 events Abbreviations: ALL, acute lymphoblastic leukemia; AML, acute myeloid leukemia; BMI, body mass index; CARG, Cancer and Aging Research Group; CI, confidence interval; CLL, chronic lymphocytic leukemia; ECOG PS, Eastern Cooperative Group performance status; IADL, instrumental activities of daily living; KPS, Karnofsky performance status; MDS, myelodysplastic syndrome; MOS, medical outcomes study; PCD, plasma cell disorder; PROMIS, patient-reported outcomes measurement information system; RDI, relative dose intensity; SPPB, short physical performance battery.

Relationship of Chemotherapy Dose Intensity with Geriatric Metrics:

The RDI median was highest among patients being treated for CLL and lymphoma. Patients receiving treatment for acute lymphocytic leukemia (ALL) had the lowest RDI with a median of 77.7 (range 51.7–100). There was no correlation between baseline GA metrics and RDI among all patients. A third of patients were hospitalized during this study period with a median hospital admission of 1 (range 0–5) and a variable length of stay. There was no relationship between the length of hospital stay and baseline geriatric metrics.

Exploratory Analysis:

The CARG chemo-toxicity calculator includes hemoglobin laboratory data as a variable. Patients with hematologic malignancy are often anemic, inherent to the disease process. In an exploratory analysis we analyzed the relationship of different hemoglobin thresholds in patients with hematologic malignancy. The CARG chemotherapy toxicity score uses a threshold hemoglobin of 11 g/dL. Using the cut of 11 g/dL in the original CARG score, n=73 (75%) of our patients had hemoglobin of ≤11 g/dL. Within our dataset evaluating a single variable of anemia, using the target endpoint of occurrence of grade 3–5 AE, the threshold for hemoglobin that could delineate the toxicity risk group was a threshold of 8 g/dL (p=0.03). In our study cohort, n=28 (30%) of patients had a Hg <8 g/dL. Using a new threshold of 8 g/dL to construct the CARG toxicity score, the resulting composite score was not predictive of grade 3–5 AE (p=0.41).

DISCUSSION

In this prospective study in older adults with hematologic malignancy, the CARG chemo-toxicity calculator was not predictive of grade 3–5 adverse events. Chemotherapy toxicity profiles were different in patients with hematologic malignancies compared to patients with solid tumors, as previously reported,¹² with more hematologic and non-hematologic toxicities and more infectious complications. Grade 5 events were similar among patients treated with hematologic malignancies compared to prior reports among patients with solid tumors.¹⁶ Importantly, early mortality was high amongst this older adult patient population with death occurring in 16.9% of consented patients. In our exploratory study, a single variable of anemia was predictive of chemotherapy toxicity, but the composite score was not predictive in patients with hematologic malignancy.

In this study, we reported a comprehensive longitudinal analysis of age-related changes with cancer treatment focused on physical function, mental health, social activities, psychosocial well-being, nutrition, and medication changes. We identified that cancer treatment results in improvements in physical function subjectively (MOS) and objectively (SPPB), mental health (MHI), cognition, social activity, and QoL. We found that patients with CLL and lymphoma had the fewest treatment delays or dose reductions, while patients with ALL had the most treatment delays or dose reductions. In our analysis, we identified that treatment toxicities were highest among patients living alone and those that were socially active. Social support networks are an underrecognized aspect of cancer care; data have shown that unmarried patients are at higher risk of cancer death. ²⁹ In addition, most studies suggest that social relationships (and social activity) are a positive variable in cancer care and are associated with improved clinical outcomes including survival.^30–32 This stands in contrast to our dataset suggesting that patients with hematologic malignancy that are more socially active had higher risk of treatment toxicity. It is worth noting that our study completed enrollment at the start of the COVID-19 pandemic.

Several factors were associated with a higher risk of death, including known factors, such as advanced performance status and an increase in comorbidities. Cognitive impairment was also associated with an increased risk of death, yet screening of memory deficits is not routine in oncology care. This highlights an unmet need in aging adults with cancer. Several studies have recognized the impairment of functional and cognitive reserve and need for GA implementation in high-risk diseases like hematologic malignancy. ^33–35 Among patients with CLL treated homogeneously in a randomized controlled trial, the CARG chemo-toxicity tool was also not predictive of chemotherapy toxicities, albeit the geriatric and functional impairments were much lower than other reported disease cohorts. ³⁶ The assessment measure that was most predictive of death in our analysis was an objective measure of physical function using the SPPB. The SPPB can be performed readily in an outpatient oncology clinic and focuses on balance, lower body strength, and a short walk test. Using these objective markers of physical function may help delineate treatment intensity and is a better characterization of frailty relative to other subjective markers of health in our analysis.

One of the study’s strengths is the expansion of the traditional CARG GA to include objective markers of physical function using the SPPB and patient-reported PROMIS measures of QoL. We report that many of our patients at the time of diagnosis had marked functional impairment, both by subjective and objective measurements. Our team and others have identified that objective markers of function using the SPPB are associated with worse clinical outcomes and survival ³⁷ among patients with hematologic malignancy.¹³ Importantly, we also report that although patients show significant functional deficits, there is marked improvement with treatment. This is particularly relevant for clinicians when discussing the potential benefit of treatment and weighing the risks of treatments in aging adults with blood cancer. The under and overtreatment of aging adults with cancer is imprecise and the goal has primarily focused on survival.³⁸ Our data advance the science on functional trajectories and QoL to better inform patients on the expected impacts of their treatment.

The CARG chemo-toxicity calculator was created as a standardized method to estimate risk of toxicity in patients receiving chemotherapy. The model was established in 500 patients with solid tumors ages 65–91 where a GA scoring system risk stratifies patients into low, intermediate, or high risk for chemotherapy toxicities. “Other” cancers were represented at 6% and represented a limited population of patients with hematologic malignancies. Limited studies have explored other chemotherapy toxicity tools in hematologic malignancy, and the CARG tool has not been validated in this populaation.^39–42 Consequently, the chemo-toxicity calculator may have limitations when applied to patients with hematologic malignancies for several reasons. Older adults with hematologic malignancy receive a variety of treatment types and modalities including novel or targeted therapy, hypomethylating agents, monoclonal antibodies, immunotherapy, and traditional chemotherapy.⁴³ Relative to the original studies where all patients received chemotherapy, 57% of patients received traditional chemotherapy alone or in combination. The changing paradigm of treatment in cancer care may influence the future predictive modeling of the CARG chemo-toxicity tool. In addition, many patients with hematologic malignancy are treated indefinitely until disease progression or toxicity (i.e., multiple myeloma or chronic lymphocytic leukemia) and therefore the calculator may not provide an accurate reflection of chemotherapy risk over long periods of time. Longitudinal relationships of CARG chemo-toxicity scores with AE are warranted to validate the predictive ability of the chemo-toxicity calculator long-term. Other investigative teams have sought to develop and validate modifications of the CARG calculator specific for diseases at early stages, and an adaptation of the CARG chemo-toxicity tool was created for early-stage breast cancer, concluding that a breast-specific CARG score outperforms the traditional measure and was associated with treatment modifications (dose reductions, delays) and was predictive of hospitalizations.⁴⁴ As such the CARG chemo-toxicity tool is ideal for patients receiving traditional chemotherapeutic modalities.

There are several limitations to this study. The majority of participants were White, so this study lacks generalizability to different races and ethnicities. Additional institutional studies are underway to address broadening study inclusion in real time. Patients enrolled in this geriatric oncology study were 60 years and older, with the intent to capture patients prior to hematopoietic stem cell transplant, a modality more commonly used in younger patients.⁴⁵ However, this stands in contrast to the age recommendations that all adults 65 years and older undergo a geriatric assessment. ^46,47 Several studies have demonstrated the predictive ability of geriatric metrics in patients with hematologic malignancies, particularly as they apply to gait speed, IADL, and precise measures of frailty.^48–51 In contrast, we reported the relationship between repeated assessments of chemotherapy toxicity and the change over time with therapy. A second limitation is the QoL measure we selected in this study, the PROMIS Global Health Scale, which is less commonly applied in patients with hematologic malignancies. Many QoL tools are routinely used in hematologic malignancy clinical trials and numerous studies have reported results using the EORTC QLQ-C30, SF-36, or iterations of the Functional Assessment of Cancer Therapy (FACT) psychometric tools.⁵² The range of QoL instruments used may hinder comparisons across study populations. The PROMIS measures have several advantages, given their rigorous development, feasibility in administration, non-proprietary nature, accessibility, and ability to be used across different diseases. Moreover, the PROMIS data strongly corresponds to the Patient-Reported Outcomes version of the Common Terminology Criteria for Adverse Events (PRO-CTCAE).⁵³ A third limitation is that patients had grade 3–5 toxicities examined throughout the study period until progression of disease, transplant, or one-year (earliest timepoint). Importantly, attrition was 23%, primarily due to death, in our analysis a higher CARG score was associated with an increased risk of death suggesting the CARG score reflects poor physiologic reserve. Patients who remained on study may have had a longer period in which they would be at risk for toxicities; the median time on study was nearly six months. Thus, the CARG chemo-toxicity risk calculator may be better suited for a defined chemotherapy course rather than a prolonged or extended treatment duration.

In conclusion, we have demonstrated that the CARG chemo-toxicity calculator is not predictive of grade 3–5 toxicity in our data set of patients with hematologic malignancies. However, the risk score was associated with survival in univariable analysis. The wide variability of treatments options for patients with hematologic malignancy intersected with heterogeneity of aging makes treatment decisions challenging. We identified that objective measures of function using the SPPB may be a better predictor of outcomes for patients with blood cancer. We also demonstrated a comprehensive trajectory of function, QoL, psychosocial well-being, and cognition among older adults with cancer. These findings are valuable for clinicians caring for aging adults with blood cancer to better advise and inform on anticipated health trajectories with treatment.