Associations of frailty with survival, hospitalization, functional decline, and toxicity among older adults with advanced non-small cell lung cancer

Abstract

Introduction

Among older adults with cancer receiving chemotherapy, frailty indices predict OS and toxicity. Given the increased use of immunotherapy and targeted therapy for advanced non-small cell lung cancer (aNSCLC), we evaluated frailty and Karnofsky Performance Status (KPS) among older adults with aNSCLC receiving chemotherapy, immunotherapy, and/or targeted therapy.

Methods

Patients aged ≥ 65 with aNSCLC starting systemic therapy with non-curative intent underwent geriatric assessments over 6 months. We developed a deficit-accumulation frailty index to categorize patients as robust, pre-frail, or frail. To evaluate associations between frailty and KPS with OS, we used Cox proportional hazards models adjusted for race, insurance, and treatment. We used logistic regression to evaluate hospitalizations, functional decline, and severe toxicity.

Results

Among 155 patients (median age 73), 45.8% were robust, 36.1% pre-frail, and 18.2% frail; 34.8% had a KPS ≥ 90, 32.9% had a KPS of 80, and 32.3% had a KPS ≤ 70. The median OS was 17.9 months. Pre-frail/frail patients had worse OS compared to robust patients (adjusted hazard ratio [HR] 2.09, 95% CI, 1.31-3.34) and were more likely to be hospitalized (adjusted odds ratio [OR] 2.21, 95% CI, 1.09-4.48), functionally decline (adjusted OR 2.29, 95% CI, 1.09-4.78), and experience grade ≥ 3 hematologic toxicity (adjusted OR 5.18, 95% CI, 1.02-26.03). KPS was only associated with OS.

Conclusions

Our frailty index was associated with OS, hospitalization, functional decline, and hematologic AEs among older adults with aNSCLC receiving systemic therapies, while KPS was only associated with OS. Pretreatment frailty assessment may help identify older adults at risk for poor outcomes to optimize decision-making and supportive care.

Implications for practice.

Geriatric assessment, which is recommended by national guidelines, remains underutilized. One barrier is translating numerous multi-dimensional results into a clinically relevant summary score. We found that for older adults with advanced non-small cell lung cancer, a deficit-accumulation frailty index can synthesize geriatric assessment results and identify patients at risk for poor OS, functional decline, hospitalization, and hematologic toxicities who may be missed by conventional performance status assessment. Integration of pretreatment frailty assessment into clinical workflows may help oncologists better identify older adults at risk for poor outcomes to optimize decision-making and supportive care.

Introduction

Over half of all new cancer diagnoses and more than 70% of cancer-related deaths annually occur in adults aged ≥ 65 years.¹ Compared to younger patients, older adults with cancer are at higher risk for poor survival and treatment-related toxicities.² In addition, older adults are often underrepresented in the clinical trials that form the evidence base that oncologists use when caring for patients of all ages.³ Given the limited available data to help risk stratify older adults in terms of oncologic and other patient-centered outcomes, oncologists commonly rely upon clinical gestalt to assign a performance status rating (eg, Karnofsky Performance Status [KPS]), which has been shown to be associated with OS among older adults but not treatment-related toxicity.^2,4

The recently updated 2023 American Society of Clinical Oncology geriatric oncology guideline recommends that a geriatric assessment “be used to identify vulnerabilities or impairments that are not routinely captured in oncology assessment for all patients over 65 years old with cancer.”⁵ These geriatric assessment-identified impairments contribute to an individual patient’s frailty, which is defined as a “state of increased vulnerability to adverse outcomes.”⁶ Landmark clinical trials in geriatric oncology have demonstrated that geriatric assessment-guided interventions can decrease treatment-related toxicities, improve quality of life, and reduce healthcare utilization for older adults.⁵

However, geriatric assessment utilization in cancer care remains low, and oncologists report difficulty interpreting individual impairments in a clinically meaningful way.⁷ A deficit-accumulation frailty index^6,8 (DAFI) is one promising approach to help synthesize numerous individual geriatric assessment results into a more clinically digestible summary score. A DAFI assesses the proportion of pre-specified aging-related impairments that are present for each patient and can be used to categorize patients as robust, pre-frail, or frail. In geriatric oncology, frailty defined using the DAFI approach has been shown to be associated with poor OS, treatment-related toxicity, and functional decline among older adults with cancer primarily receiving chemotherapy.^8-10

While these prior studies have demonstrated the potential value of frailty indices in geriatric oncology broadly, only 4 DAFI studies have focused on older adults with non-small cell lung cancer (NSCLC),^11-14 the leading cause of cancer-related mortality in the United States and worldwide.¹⁵ Three of these NSCLC DAFI studies were retrospectively conducted and limited to existing data elements in the medical record and/or claims data.^11-13 The fourth study was prospective and included patient-reported outcomes but was cross-sectional in design such that frailty at heterogenous timepoints during treatment was examined, not pretreatment frailty.¹⁴ In addition, only 2 of these studies examined more contemporary samples of patients with advanced NSCLC (aNSCLC) treated with immunotherapy.^13,14 A small US study of 60 patients with aNSCLC found that frailty at heterogenous timepoints during immunotherapy was associated with lower patient-reported physical function and higher rates of pain.¹⁴ A retrospective Chinese study of patients with advanced lung cancer receiving immunotherapy found an association between frailty and pneumonitis specifically, but not immunotherapy toxicity in general.¹³ However, their study did not focus on older adults (median age 66) and substantially underrepresented women (82.5% men).¹³ Furthermore, no prior DAFI studies have compared a DAFI with KPS to predict outcomes.

Therefore, we used data from a prospective cohort study to evaluate the associations of both a DAFI and KPS with a primary outcome of OS, as well as secondary outcomes of unplanned hospitalization, functional decline, and severe treatment-related hematologic and non-hematologic toxicity among older patients with aNSCLC receiving chemotherapy, immunotherapy, and/or targeted therapy with non-curative intent. We hypothesized that a DAFI would be associated with OS and secondary outcomes, whereas KPS would only be associated with OS.

Methods

Study design and population

We previously published the methodologic details for our prospective cohort study “Lung cancer in older adults: Treatment experience through the patient’s lens” (Lens Study).^16,17 In brief, patients aged ≥ 65 with aNSCLC were recruited from a Comprehensive Cancer Center, Veterans Affairs clinic, safety-net oncology clinic, and 3 community oncology affiliates. Patients had stage III-IV NSCLC or recurrent disease and were starting a new systemic therapy (chemotherapy, immunotherapy, and/or targeted therapy) with non-curative intent. Patients spoke English and/or a Chinese dialect and were able to provide informed consent. Patients completed a geriatric assessment pretreatment and at 1, 2, 4, and 6 months after treatment initiation, or until treatment discontinuation, whichever came earlier.

Pretreatment assessment

Prior to treatment initiation, patients completed a demographic survey and geriatric assessment conducted by the research team, which assessed multiple domains including physical function, quality of life, nutrition status, comorbidities, cognition, mood, and social activity and support (Table S1).¹⁸ We also assessed patient-reported performance status,¹⁹ which adds independent prognostic information compared to clinician-reported KPS.²⁰ We abstracted treatment doses, creatinine clearance, and hemoglobin from the medical record. Polypharmacy (> 9 medications)²¹ was assessed by abstracting the number of daily medications from the medical record for patients enrolled from September 2017 through July 2022 or by direct patient report starting in August 2022 when this question was added to the study.

Deficit-accumulation frailty index

Using the deficit-accumulation principle and methods established by Rockwood to construct a DAFI,⁶ we developed a 46-item index (Table S1) using prospectively collected pretreatment geriatric assessment characteristics encompassing a range of patient-centered domains that substantially overlap with the Practical Geriatric Assessment recommended by the American Society of Clinical Oncology (ASCO).²² These characteristics were selected because they encompass a range of factors that are important to health status and become more common with older age.⁶ The inclusion of 46 characteristics meets the required minimum of 30-40 variables for a DAFI per the Rockwood method.⁶ For dichotomous variables, markers of frailty were coded as “1” if present and “0” if absent. For categorical variables with 3 response levels (eg, without help, need some help, unable to do), the most adverse response was coded as “1,” the intermediate condition as “0.5,” and absent as “0.”

A pretreatment DAFI score was calculated for each patient by summing scores for all non-missing items (actual deficit score) divided by the sum of maximum possible scores for all non-missing items (potential deficit score). A minimum of > 10 non-missing items is required for stable DAFI estimates.⁶ Patients with ≤ 10 non-missing DAFI items were excluded from the analysis. DAFI scores were calculated for research purposes only and were not shared with the clinical team. DAFI scores can range from 0 (no frailty deficits) to 1 (severe frailty). We categorized patients as robust (< 0.2), pre-frail (0.2-0.34), or frail (≥ 0.35) using cutoffs previously established in the geriatric oncology literature to predict clinically relevant outcomes such as OS and functional limitations.^8,9

Karnofsky performance status

Treating oncology clinicians reported a pretreatment KPS score²³ for each patient. To analyze this ordinal categorical variable,²⁴ we categorized clinician-reported KPS as ≥ 90, 80, or ≤ 70.^8,9,25

Study outcomes

We abstracted the primary outcome of OS beyond 6-month active study follow-up from the medical record and defined the index date for survival time as the date of treatment initiation. For patients without recent clinical visits in the medical record, obituaries were also searched. Patients with an unknown vital status were censored at the date of their last known vital status.

Secondary outcomes included any unplanned hospitalization (which may or may not be related to NSCLC treatment), functional decline in ability to perform instrumental activities of daily living (IADL),²⁶ and clinician-reported treatment-related grade ≥ 3 hematologic or non-hematologic adverse events (AEs)²⁷ within 6-month active study follow-up. A decline in IADL was defined as a decrease of ≥ 1 point on the Older Americans Resources and Services IADL instrument on any follow-up assessment within 6 months. We selected IADL as our functional outcome because functional decline in IADL is more common during cancer treatment among older adults than decline in activities of daily living (ADL).²⁸

Statistical analysis

All analyses were performed with Stata/SE 17 (StataCorp LLC). We used Fisher’s exact tests to compare pretreatment demographic and clinical characteristics across frailty categories (ie, robust, pre-frail, frail). For the primary outcome, we conducted a Kaplan–Meier survival analysis to estimate OS by frailty group. To evaluate the association between frailty (robust vs pre-frail/frail) and OS, we used Cox proportional hazards models. We grouped pre-frail and frail patients given the relatively small sample sizes for these subgroups. Sensitivity analyses were also conducted to evaluate OS among pre-frail and frail patients separately compared with robust patients. Based on results from our univariable analysis of characteristics associated with frailty and prior research,²⁹ we adjusted for race, insurance, and treatment type. To evaluate the independent prognostic value of frailty and KPS, we included both the DAFI and KPS in an adjusted Cox proportional hazards model for OS. Multicollinearity was assessed using variance inflation factors.³⁰

For the secondary outcomes, we used logistic regression to evaluate the associations with frailty (robust vs pre-frail/frail) adjusting for treatment type, race, and insurance. Secondary outcomes were also evaluated in sensitivity analyses evaluating pre-frail and frail patients separately compared with robust patients, though these results must be interpreted in the context of the sample size for each analysis. Analyses for primary and secondary outcomes were repeated with KPS (≥ 90 vs ≤ 80) as the independent variable instead of the DAFI frailty category. Two-sided tests with P < .05 were considered statistically significant.

Results

Pretreatment characteristics

From September 2017 to July 2022, 155 older adults with aNSCLC were included in this analysis. Four patients were excluded because they had ≤ 10 non-missing DAFI items. Median age was 73 (IQR 68-79, range 65-94) and 54.8% were female (Table 1). Most had a college-level education or higher (72.1%) and were partnered (59.5%). The cohort was diverse with 59.4% White, 27.1% Asian, 5.8% Black or African American, 5.8% Hispanic, and 0.6% American Indian/Alaskan Native patients as well as 1.3% of patients identifying as more than one race. The majority of patients had stage IV disease and adenocarcinoma histology. During the study, 30.3% received targeted therapy, 29.0% chemoimmunotherapy, 27.7% immunotherapy, and 12.9% chemotherapy; 11.0% received a primary dose reduction for cycle one. Many had received prior NSCLC treatment (64.5%). Pretreatment geriatric assessment showed that 62.0% were dependent in ≥ 1 IADL and 22.6% in ≥ 1 activity of daily living (ADL).

Table 1.

Demographic and clinical characteristics according to frailty group.

Characteristic	Overall (N = 155) n (%)	Robust (n  = 71, 45.8%) n (%)	Pre-frail (n  = 56, 36.1%) n (%)	Frail (n  = 28, 18.1%) n (%)	P
Age, year					.37
65-69	50 (32.3)	17 (23.9)	23 (41.1)	10 (35.7)
70-74	43 (27.7)	23 (32.4)	14 (25.0)	6 (21.4)
75-79	25 (16.1)	15 (21.1)	6 (10.7)	4 (14.3)
≥ 80	37 (23.9)	16 (22.5)	13 (23.2)	8 (28.6)
Female sex	85 (54.8)	38 (53.5)	32 (57.1)	15 (53.6)	.91
Race and ethnicity					<.001
American Indian/Alaskan Native, Black or African American, Hispanic, or more than 1 racea	21 (13.6)	11 (15.5)	7 (12.5)	3 (10.7)
Asian	42 (27.1)	10 (14.1)	15 (26.8)	17 (60.7)
Non-Hispanic White	92 (59.4)	50 (70.4)	34 (60.7)	8 (28.6)
Primary language					.02
English	128 (82.6)	63 (88.7)	47 (83.9)	18 (64.3)
Chinese dialect	27 (17.4)	8 (11.3)	9 (16.1)	10 (35.7)
Education					.001
High school or less	38 (27.9)	11 (16.9)	14 (28.6)	13 (59.1)
College or higher	98 (72.1)	54 (83.1)	35 (71.4)	9 (40.9)
Partnered	91 (59.5)	47 (66.2)	30 (53.6)	14 (53.9)	.29
Lives alone	33 (21.3)	12 (16.9)	15 (26.8)	6 (21.4)	.40
Currently working	18 (13.0)	10 (15.2)	7 (13.7)	1 (4.6)	.49
Insurance					.047
Medicare	110 (71.0)	55 (77.5)	40 (71.4)	15 (53.6)
Private	24 (15.5)	11 (15.5)	8 (14.3)	5 (17.9)
Medicaid +/− Medicare	11 (7.1)	1 (1.4)	4 (7.1)	6 (21.4)
Veterans affairs	10 (6.5)	4 (5.6)	4 (7.1)	2 (7.1)
Smoking status					.33
Previously smoked	98 (63.4)	50 (70.4)	32 (58.2)	16 (57.1)
Never smoked	46 (29.9)	19 (26.8)	18 (32.7)	9 (32.1)
Currently smoke	10 (6.5)	2 (2.8)	5 (9.1)	3 (10.7)
Histology					.12
Adenocarcinoma	130 (83.9)	63 (88.7)	47 (83.9)	20 (71.4)
Squamous cell	15 (9.7)	3 (4.2)	6 (10.7)	6 (21.4)
Other	10 (6.5)	5 (7.0)	3 (5.4)	2 (7.1)
Stage					.19
III	8 (5.2)	2 (2.8)	3 (5.4)	3 (10.7)
IVA	65 (41.9)	36 (50.7)	20 (35.7)	9 (32.1)
IVB	82 (52.9)	33 (46.5)	33 (58.9)	16 (57.1)
Brain metastasis	44 (28.4)	16 (22.5)	18 (32.1)	10 (35.7)	.27
Any prior lung cancer treatment	100 (64.5)	48 (67.6)	38 (67.9)	14 (50.0)	.23
Prior radiation	64 (42.1)	28 (39.4)	29 (49.2)	10 (34.5)	.36
Prior chemotherapy	51 (32.9)	26 (36.6)	20 (35.7)	5 (17.9)	.18
Prior immunotherapy	30 (19.4)	13 (18.3)	14 (25.0)	3 (10.7)	.31
Prior oral targeted therapy	36 (23.2)	16 (22.5)	12 (21.4)	8 (28.6)	.75
Prior surgery	27 (17.4)	17 (23.9)	8 (14.3)	2 (7.1)	.12
Treatment type					.75
Targeted therapy	47 (30.3)	24 (33.8)	14 (25.0)	9 (32.1)
Chemoimmunotherapy	45 (29.0)	21 (29.6)	17 (30.4)	7 (25.0)
Immunotherapy	43 (27.7)	20 (28.2)	15 (26.8)	8 (28.6)
Chemotherapy	20 (12.9)	6 (8.5)	10 (17.9)	4 (14.3)
Primary dose reduction for cycle one	17 (11.0)	6 (8.5)	8 (14.3)	3 (10.7)	.55

Missing data: education n = 19, partnered n = 2, currently working n = 16, smoking status n = 1.

^aSeveral racial and ethnic groups were combined to maintain de-identification of participants given small numbers.

Frailty and KPS

The median pretreatment DAFI score was 0.21 (IQR 0.12-0.30, range 0.03-0.72). Overall, 45.8% of patients were robust, 36.1% pre-frail, and 18.1% frail (Figure 1). Median KPS was 80 (IQR 70-90, range 40-100); 34.8% had a KPS  ≥ 90, 32.9% had a KPS of 80, and 32.3% had a KPS ≤ 70 (Figure 2A-C). Notably, 24.1% of patients with a KPS ≥ 90 were categorized as pre-frail on the DAFI (Figure 2A), and 14.0% of patients with a KPS ≤ 70 were categorized as robust on the DAFI (Figure 2C).

Figure 1.

Distribution of deficit-accumulation frailty index scores among adults aged ≥ 65 with advanced non-small cell lung cancer starting a new chemotherapy, immunotherapy, and/or targeted therapy regimen for non-curative intent.

Figure 2.

Distribution of deficit-accumulation frailty index scores among older adults with advanced non-small cell lung cancer with a Karnofsky Performance Status (KPS) (A) ≥ 90, (B) KPS 80, and (C) KPS ≤ 70.

Frailty was associated with race (P < .001) with more frailty among Asians, speaking a Chinese dialect (P = .02), lower education (P = .001), and insurance (P = .047) with more frailty among patients with Medicaid (Table 1). Frailty was not associated with age, sex, NSCLC stage, line of treatment, treatment type, or primary dose reduction (Table 1).

Primary outcome

Over a median follow-up time of 12.6 months (IQR 5.3-28.0 months) for vital status, 88 patients died (56.7%). The vast majority of deaths were abstracted from the medical record (98%) with only 2% obtained from obituaries. Of those who died, 33 (37.5%) were robust, 38 (43.2%) were pre-frail, and 17 (19.3%) were frail. Seven patients (4.5%) had an unknown vital status at the time of abstraction and were censored at the date of their last known vital status. Median OS was 17.9 months overall: 30.9 months for robust, 8.6 months for pre-frail, and 6.6 months for frail patients (Figure 3). The log-rank test for trend P-value was .003.

Figure 3.

Kaplan–Meier analysis of OS among older adults with advanced non-small cell lung cancer according to deficit-accumulation frailty index category

Pre-frail/frail patients had worse OS compared to robust patients (HR 2.06, 95% CI, 1.33-3.18; P = .001; Table 2). This difference in OS between pre-frail/frail vs robust patients remained statistically significant after adjusting for race, insurance, and treatment type (adjusted hazard ratio [HR] 2.09, 95% CI, 1.31-3.34; P = .002; Table 2). The global test of proportional hazards assumption was satisfied (P = .47). We did not adjust for primary language because it is colinear with race. We also did not adjust for education because it was not associated with any of the study outcomes (data not shown). Notably, there was no statistically significant difference in OS when frail patients were compared with pre-frail patients (HR 0.98, 95% CI, 0.55-1.75; P = .96). In unadjusted and adjusted sensitivity analyses, both pre-frail and frail patients had worse OS compared with robust patients (Table S2).

Table 2.

Associations between deficit-accumulation frailty index (DAFI) pre-frail/frail status and clinical outcomes.

Primary outcome	Characteristic	Unadjusted HR (95% CI)	P	Adjusteda HR (95% CI)	P
Overall survival	Pre-frail/frail (ref: robust)	2.06 (1.33-3.18)	.001	2.09 (1.31-3.34)	.002
Secondary outcome		Unadjusted OR (95% CI)	P	Adjusted a OR (95% CI)	P
Unplanned hospitalization	Pre-frail/frail (ref: robust)	2.27 (1.13-4.55)	.02	2.42 (1.15-5.11)	.02
IADL decline	Pre-frail/frail (ref: robust)	2.23 (1.09-4.55)	.03	2.40 (1.11-5.20)	.03
Grade ≥ 3 treatment-related hematologic adverse event	Pre-frail/frail (ref: robust)	4.14 (0.86-19.83)	.08	5.18 (1.03-26.03)	.046
Grade ≥ 3 treatment-related non-hematologic adverse event	Pre-frail/frail (ref: robust)	1.24 (0.59-2.65)	.57	1.35 (0.60-3.05)	.47

^aAdjusted for race, insurance, and treatment group.

Abbreviation: IADL, instrumental activities of daily living.

Median OS was not reached for patients with a KPS ≥ 90, 13.6 months for patients with a KPS of 80, and 7.3 months for patients with a KPS ≤ 70 (Figure 4). The log-rank test for trend P-value was .0005.

Figure 4.

Kaplan–Meier analysis of OS among older adults with advanced non-small cell lung cancer according to Karnofsky Performance Status (KPS).

Patients with a KPS ≤ 80 had worse OS compared to patients with a KPS ≥ 90 (HR 2.53, 95% CI, 1.54-4.18; P < .001; Table 3). This difference in OS between patients with a KPS ≤ 80 vs ≥ 90 remained statistically significant after adjustment (adjusted HR 2.20, 95% CI, 1.31-3.72; P = .003; Table 3). The global test of proportional hazards assumption was satisfied (P = .82). Notably, there was no statistically significant difference in OS when patients with a KPS ≤ 70 were compared with patients with a KPS of 80 (HR 1.05, 95% CI, 0.65-1.69; P = .84).

Table 3.

Associations between clinician-reported Karnofsky Performance Status (KPS) and clinical outcomes.

Primary outcome	Characteristic	Unadjusted HR (95% CI)	P	Adjusteda HR (95% CI)	P
Overall survival	KPS ≤ 80 (ref: KPS ≥ 90)	2.53 (1.54-4.18)	<.001	2.20 (1.31-3.72)	.003
Secondary outcome		Unadjusted OR (95% CI)	P	Adjusted a OR (95% CI)	P
Unplanned hospitalization	KPS ≤ 80 (ref: KPS ≥ 90)	1.98 (0.95-4.16)	.07	1.91 (0.88-4.17)	.10
IADL decline	KPS ≤ 80 (ref: KPS ≥ 90)	2.02 (0.97-4.23)	.06	1.84 (0.84-4.00)	.13
Grade ≥ 3 treatment-related hematologic adverse event	KPS ≤ 80 (ref: KPS ≥ 90)	0.93 (0.26-3.33)	.91	1.16 (0.30-4.53)	.83
Grade ≥ 3 treatment-related non-hematologic adverse event	KPS ≤ 80 (ref: KPS ≥ 90)	0.93 (0.43-2.02)	.86	0.98 (0.43-2.25)	.96

^aAdjusted for race, insurance, and treatment group.

Abbreviation: IADL, instrumental activities of daily living.

When we evaluated frailty and KPS in the same Cox proportional hazards model adjusted for race, insurance, and treatment type, frailty (adjusted HR 1.69, 95% CI, 1.02-2.79; P = .04) and KPS (adjusted HR 1.79, 95% CI, 1.02-3.14; P = .04), both remained independent predictors of OS. Variance inflation factors indicated no concerns with multicollinearity (all < 1.5, which is well under 5 or 10, commonly used thresholds to evaluate further for multicollinearity³⁰).

Secondary outcomes

During 6-month active study follow-up, 33.6% of patients had at least one unplanned hospitalization, 42.3% developed IADL decline, and 29.0% developed at least one grade ≥ 3 treatment-related AE (23.2% had at least one grade ≥ 3 treatment-related non-hematologic AE, 7.1% had at least one grade ≥ 3 treatment-related hematologic AE). Because pre-frail and frail patients had similar OS and relatively small subgroup sample sizes, they were grouped together for secondary analyses. Compared to robust patients, pre-frail/frail patients were more likely to have an unplanned hospitalization (OR 2.27, 95% CI, 1.13-4.55; P = .02; Table 2). This association remained statistically significant after adjusting for age, sex, and treatment type (adjusted OR 2.42, 95% CI, 1.15-5.11; P = .02; Table 2). In unadjusted and adjusted sensitivity analyses, pre-frail patients were more likely to have an unplanned hospitalization than robust patients (Table S2).

Pre-frail/frail patients were also more likely to develop IADL decline in both the unadjusted (OR 2.23, 95% CI, 1.09-4.55; P = .03; Table 2) and adjusted analyses (adjusted OR 2.40, 95% CI, 1.11-5.20; P = .03; Table 2). In adjusted sensitivity analyses, frail patients were more likely to develop IADL decline than robust patients (Table S2).

In unadjusted analyses, pre-frail/frail patients had similar odds of developing a grade ≥ 3 hematologic AE as robust patients (OR 4.14, 95% CI, 0.86-19.83; P = .076). However, after adjustment for potential confounding from treatment type, race, and insurance, pre-frail/frail patients were more likely to experience a grade ≥ 3 hematologic AE than robust patients (adjusted OR 5.2, 95% CI, 1.02-26.03; P = .046). There were no differences in grade ≥ 3 non-hematologic AEs between pre-frail/frail and robust patients in unadjusted or adjusted analyses (Table 2). In unadjusted and adjusted sensitivity analyses, pre-frail patients were more likely to experience a grade ≥ 3 hematologic AE than robust patients (Table S2).

Because patients with a KPS of 80 and patients with a KPS ≤ 70 had similar OS, they were grouped together for secondary analyses. When comparing patients with a KPS ≤ 80 vs ≥ 90, there were no statistically significant differences in hospitalization, IADL decline, or grade ≥ 3 hematologic or non-hematologic AEs in unadjusted or adjusted analyses (Table 3).

Discussion

Using a deficit-accumulation approach to frailty, we found that our DAFI is prognostic for OS independent of KPS and is also associated with unplanned hospitalization, IADL decline, and grade ≥ 3 treatment-related hematologic AEs among older adults with aNSCLC receiving systemic therapy. In contrast, KPS was prognostic only for OS, which highlights limitations in the clinical utility of KPS for older adults with aNSCLC. Our study adds to the growing literature demonstrating the value of a geriatric assessment-derived DAFI in identifying older patients at risk for poor outcomes during cancer treatment. In particular, our findings add to the extremely limited data on DAFI outcomes among older patients receiving contemporary systemic therapy regimens including immunotherapy and targeted therapy.

To our knowledge, this is the first prospective longitudinal study of a pretreatment DAFI among older adults with aNSCLC and the first to evaluate both a DAFI and performance status as predictors of clinical outcomes within the same cohort. In the current cancer treatment landscape where the population is aging and the use of immunotherapy and targeted therapy is rising, it is increasingly important that older adults receive geriatric assessment-informed care, which requires that oncologists have clinical tools to help facilitate the interpretation of geriatric assessment results. This is an area where a DAFI can be both practical and of clinical value. Identifying pre-frail and frail patients can facilitate targeted supportive care interventions and assessments (eg, exercise program, medication review, nutritionist evaluation) that have been previously shown to improve important patient outcomes in large randomized trials of geriatric assessment-driven interventions.^31-35

Our finding that frailty is prognostic for OS among older adults with cancer is consistent with prior studies.^8,9,11,12 Notably, we found that older adults with aNSCLC who are pre-frail have a similar OS to those who are frail. This has important clinical implications because oncologists can often identify frail-appearing patients by gestalt, but changes in pre-frail patients may be more subtle and challenging to identify. This may in turn lead to missed opportunities for tailored anticipatory guidance and increased monitoring during treatment for patients who appear more robust than they truly are. Prior studies of older adults with other cancer types (eg, gastrointestinal malignancies) have found that pre-frail patients have survival rates between those of robust and frail patients, or closer to that of robust patients,⁹ suggesting that older patients with aNSCLC may especially benefit from pretreatment identification of pre-frailty. In contrast, the commonly used KPS is poorly correlated with frailty and can underestimate impairment.³⁶ Indeed, in our study, we observed that nearly a quarter of patients categorized by clinicians as having a KPS ≥ 90 were actually pre-frail according to the DAFI.

The potential value of using a DAFI over KPS was again demonstrated with our secondary outcomes. We found an association between pre-frailty/frailty and unplanned hospitalization, but not between KPS and unplanned hospitalization. Unplanned hospitalization among older patients with cancer has been shown to be associated with comorbidity, which our DAFI heavily weighs (ie, 13 out of 46 components).³⁷ Dependence in ADLs, which our DAFI also heavily weights, has also been shown to predict hospitalization.³⁸ Unlike performance status, a DAFI systematically incorporates a wider range of health-related domains including all the above factors, which may better capture a patient’s risk of adverse outcomes such as hospitalization.

In addition, we observed an association between pre-frailty/frailty and IADL decline, which is supported by a prior study of a similar DAFI called the Cancer and Aging Resilience Evaluation-Frailty Index among older patients with gastrointestinal cancers.⁹ In contrast, we did not observe an association between KPS and IADL decline. Studies evaluating associations between individual geriatric assessment components (which comprise most of our DAFI) with IADL decline have shown mixed results; some have found no associations,³⁹ while others suggest associations of IADL decline with female sex, comorbidities, falls, fatigue, and depression risk.⁴⁰ A DAFI may capture vulnerability to outcomes such as functional decline that can be missed when relying on specific clinical characteristics or performance status alone.

Prior studies of older adults with cancer have demonstrated an association between DAFI score and grade ≥ 3 toxicity,^8,9 which we observed for hematologic but not non-hematologic AEs in our aNSCLC cohort. These prior studies evaluated patients receiving primarily chemotherapy for either gastrointestinal malignancies⁹ or multiple cancer types.⁸ Interestingly, among patients with gastrointestinal malignancies receiving primarily chemotherapy (which is usually considered more myelosuppressive than immunotherapy or targeted therapy), frail patients were more likely to experience non-hematologic but not hematologic toxicities,⁹ which is the opposite of our results. Our finding that frailty is associated with severe hematologic toxicities but not severe non-hematology toxicities among older adults with aNSCLC receiving chemotherapy, immunotherapy, and/or targeted therapy may be due to the generally more tolerable side effect profiles of immunotherapy and targeted therapy compared with chemotherapy in prior studies, even among frail older patients.

The patient factors associated with pretreatment frailty in our study have a modest overlap with those of prior studies. Frailty among older adults with cancer has been shown to be associated with lower education,⁸ which our study confirmed. However, in prior studies, associations have been variably reported between frailty and factors such as living alone, not being partnered, Black/African American race, and line of therapy.^8,9 We evaluated many of these additional factors but only found increased frailty among patients who identify as Asian, speak a Chinese dialect, have lower education, or have Medicaid insurance. Ultimately, these differences further emphasize the challenge of assessing vulnerability for adverse clinical outcomes based on a limited number of clinical or demographic factors. The strength of a DAFI stems from its ability to not overly depend on any particular characteristic and ability to function even for patients with missing data.⁴¹

Our study has several limitations. The Lens Study recruited participants from multiple sites in one urban geographic area, which may limit its generalizability to other geographic regions. Of note, our inclusion of patients who speak a Chinese dialect, the most common non-English language in our Comprehensive Cancer Center, is a strength of the study, as prior US-based DAFI studies have been conducted only among English-speaking patients. The study’s modest sample size required us to be parsimonious in selecting characteristics to adjust for in our primary and secondary analyses. Due to our sample size, we were not able to evaluate the association between frailty and the subgroup of AEs related to immunotherapy as an outcome. Additionally, our frailty index includes 46 items, many of which are part of routine geriatric assessment but may not be feasible to collect in all clinical oncology settings. However, the incorporation of frailty indices into the electronic medical record to leverage existing clinical data and patient-reported outcomes is an active area of research and implementation science.⁴² Finally, vital status was unknown for 7 patients; however, they still contributed a median of 2.2 years of survival time before they were censored at the date of their last known vital status.

In conclusion, we found that a pretreatment DAFI can identify older adults with aNSCLC receiving contemporary systemic therapy regimens who are at risk for poor outcomes including OS, unplanned hospitalization, functional decline, and grade ≥ 3 treatment-related hematologic AEs. Future work is needed to efficiently integrate DAFIs into clinical workflows to help oncologists individualize anticipatory guidance, optimize supportive care, and inform decision-making for patients with aNSCLC better than they can using performance status alone.

Supplementary material

Supplementary material is available at The Oncologist online.

Author contributions

Howard J. Lee Jr. (Conceptualization, Data curation, Formal analysis, Investigation, Writing—original draft, Writing—review & editing), John Boscardin (Conceptualization, Formal analysis, Methodology, Writing—review & editing), Louise C. Walter (Writing—review & editing), Alexander K. Smith (Conceptualization, Writing—review & editing), Harvey J. Cohen (Conceptualization, Writing—review & editing), Smith Giri (Conceptualization, Methodology, Writing—review & editing), Grant R. Williams (Conceptualization, Methodology, Writing—review & editing), Carolyn J. Presley (Writing—review & editing), Surbhi Singhal (Writing—review & editing), Li-Wen Huang (Conceptualization, Writing—review & editing), Ana I. Velazquez (Writing—review & editing), Matthew A. Gubens (Writing—review & editing), Collin M. Blakely (Writing—review & editing), Claire K. Mulvey (Writing—review & editing), Michael L. Cheng (Writing—review & editing), Lori C. Sakoda (Writing—review & editing), Lawrence H. Kushi (Writing—review & editing), Charles Quesenberry (Writing—review & editing), Raymond Liu (Writing—review & editing), Sara Fleszar-Pavlovic (Writing—review & editing), Caroline Eskandar (Data curation, Writing—review & editing), Edward Cutler (Writing—review & editing), Anne Marie Mercurio (Writing—review & editing), and Melisa L. Wong (Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Resources, Supervision, Writing—original draft, Writing—review & editing)

Funding

This work was supported by the National Institutes of Health (R03AG056439 to M.L.W., K76AG064431 to M.L.W., P30AG044281 to M.L.W. and L.C.W., KL2TR001870 to M.L.W., T32AG000212 to H.J.L., K24AG068312 to A.K.S., P30AG028716 to H.J.C., K08CA234225 to G.R.W., R03AG064374 to C.J.P., P30CA016058 to C.J.P., K76AG074923 to C.J.P., R03AG067935 to L.W.H., P30AG015272 to A.I.V., P30CA082103 to A.I.V., T32CA251064 to S.F.P.); The Permanente Medical Group and University of California, San Francisco Helen Diller Family Comprehensive Cancer Center to M.L.W. Content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.

Conflict of interest

The following authors reported conflicts of interest outside the submitted work:

S.G.—research funding and honoria from Janssen and Sanofi.

G.R.W.—consultant/advisor to Cardinal Health, Bayer, and Takeda.

C.J.P.—research funding from BMS Foundation and Rising Tide Foundation awarded directly to her institution; consultant/advisor to Jazz Pharmaceuticals and Regeneron.

S.S.—consultant/advisor to OncoHost.

A.I.V.—research funding from the National Institute on Aging, AACR, Conquer Cancer, and Lungevity; consultant/advisor to Merus, AstraZeneca, and Novocure; honoraria from BioAscend, ASCO, OptumHealth Education, MJH Life Sciences, MDOutlook, Curio Science, ObR Oncology, and CMEOutfitters; and travel, accommodations, expenses from DAVAOncology and BioAscend.

M.A.G.—research funding from Amgen, Johnson & Johnson, Merck, and Trizell awarded directly to his institution; consultant/advisor to AnHeart, AstraZaneca, Atreca, BMS, Cardinal Health, Genentech/Roche, Genzyme, Gilead, Guardant, Invitae, iTeos, Merus, Sanofi, Summit, and Surface.

C.M.B.—research funding from AstraZeneca, Novartis, Puma, and Mirati awarded directly to his institution; consultant/advisor to BMS, Gilead, and Bayer.

M.L.C.—honoraria from Lynx Group, WebMD, and Potomac Center for Medical Education; consultant/advisor to AstraZeneca, Boehringer Ingelheim, Mirati Therapeutics, Cepheid, Janssen, and Pfizer; research funding from Palleon Pharmaceuticals (Inst); and travel, accommodations, expenses from Daiichi Sankyo, AstraZeneca, and Genzyme.

L.C.S.—research funding from AstraZeneca, National Cancer Institute, and California Tobacco-Related Disease Research Program awarded directly to her institution; unpaid leadership role within the American Cancer Society (National Lung Cancer Roundtable Health Equity Task Group co-chair); stock ownership for Johnson & Johnson; and travel, accommodations, expenses from the American Cancer Society.

R.L.—research funding from AstraZeneca awarded directly to his institution.

C.E.—stock ownership for Johnson & Johnson, Pfizer, Moderna, Mural Oncology Public Limited, Walgreens Boots Alliance, Acadia Healthcare Co, and Bean Therapeutics.

M.L.W.—royalties from UpToDate; immediate family member is an employee of Genentech with stock ownership.

The remaining authors have no conflicts to report.

Data availability

Immediately following publication, de-identified individual participant data that underlie the results reported in this article and the data dictionary are available from the corresponding author to researchers who provide a methodologically sound proposal to achieve aims in the approved protocol. Investigators will need to sign a data access agreement.