Psychometric Properties of the Functional Assessment of Cancer Therapy-General for Evaluating Quality of Life in Patients With Life-Limiting Illness in the Emergency Department

Abstract

Background:

The Functional Assessment of Cancer Therapy-General (FACT-G) is a widely used quality-of-life measure. However, no studies have examined the FACT-G among patients with life-limiting illnesses who present to emergency departments (EDs).

Objective:

The goal of this study was to examine the psychometric properties of the FACT-G among patients with life-limiting illnesses who present to EDs in the United States.

Methods:

This cross-sectional study pooled data from 12 EDs between April 2018 and January 2020 (n = 453). Patients enrolled in the study were adults with one or more of the four life-limiting illnesses: advanced cancer, Congestive Heart Failure, Chronic Obstructive Pulmonary Disease, or End-Stage Renal Disease. We conducted item, exploratory, and confirmatory analyses (exploratory factor analysis [EFA] and confirmatory factor analysis [CFA]) to determine the psychometric properties of the FACT-G.

Results:

The FACT-G had good internal consistency (Cronbach's alpha α = 0.88). The simplest EFA model was a six-factor structure. The CFA supported the six-factor structure, evidenced by the adequate fit indices (comparative fit index = 0.93, Tucker–Lewis index = 0.92, root-mean-square error of approximation = 0.05; 90% confidence interval: 0.04 – 0.06). The six-factor structure comprised the physical, emotional, work and daily activities-related functional well-being, and the family and friends-related social well-being domains.

Conclusions:

The FACT-G is a reliable measure of health-related quality of life among patients with life-limiting illnesses who present to the ED.

Clinical Trial Registration:

NCT03325985.

Introduction

During the last six months of life, 75% of adults with life-limiting illnesses visit the emergency department (ED).¹ Adults with life-limiting illnesses are a heterogeneous group of critically and terminally ill patients, and ED visits are inflection points in these patients' illness trajectories and may signal a more rapid rate of decline.^2–4 These patients usually present to the ED with advanced stages of terminal illnesses such as Congestive Heart Failure (CHF), Chronic Obstructive Pulmonary Disease (COPD), End-Stage Renal Disease (ESRD), and cancer. More than 70% of these patients express priorities focused on comfort and quality of life rather than life extension.⁵

Quality of life is an important aspect of palliative care.⁶ Assessing the quality of life of patients with life-limiting illnesses in the ED may present an opportunity to identify and initiate discussion on hospice and palliative care.⁷ In addition, palliative care interventions in the ED can capture high-mortality risk patients at a time of crisis.^8,9 Palliative care has been shown to decrease depression,¹⁰ pain, and anxiety¹¹ and increase patient and caregiver satisfaction.¹² Palliative care is also associated with decreased hospital admissions,¹¹ intensive care utilization,¹³ and overall hospital costs.¹⁴ Early initiation of palliative care, therefore, may improve self-reported health-related quality of life (HRQOL) among patients with life-limiting illnesses.

Several measures of HRQOL exist.^15–19 However, few HRQOL measures focus on patients with life-limiting illnesses. The Functional Assessment of Cancer Therapy-General (FACT-G), a 27-item survey, is one of the most common measures of HRQOL (see Table 1). Although primarily designed for cancer patients, the FACT-G has been used to assess changes in HRQOL among individuals with chronic illnesses such as nursing home patients,^20,21 older adults with Parkinson's disease,²² and women with stress urinary incontinence.²³ To our knowledge, no research has validated the FACT-G among adults with life-limiting illnesses who present to the ED. We, therefore, aim to validate the FACT-G and examine its psychometric properties (i.e., item analysis, reliability, exploratory and confirmatory factor analyses) among patients with life-limiting illnesses presenting to the ED.

Table 1.

Psychometric Properties in Prior Studies Using the Functional Assessment of Cancer Therapy-General

Study author and year	Country	Language	Study sample	Full scale Cronbach's alpha (α)	Number of factors	Subscale Cronbach's alpha (α)
Winstead-Fry and Schultz (1997)62	United States	English	344 rural cancer patients	0.92	5	Range: 0.68–0.90a
Song et al. (2020)63	United States	English	Caregivers of 747 cancer patients	0.91 (prostate cancer) 0.91 (advanced cancer)	4	PSEF: 0.81, 0.82, 0.83, 0.86 PSEF: 0.83, 0.84, 0.81, 0.84
Serrano et al. (2016)64	Brazil	Portuguese	975 cancer patients	Not reportedb	4	Range: 0.71–0.82a
Yu et al. (2000)65	China	Chinese	1262 cancer patients	0.85	5	PSDEF: 0.75, 0.53, 0.37, 0.65, 0.75
Lee et al. (2004)66	Korea	Korean	193 breast cancer patients	0.89	5	PSSEF: 0.82, 0.86, 0.70, 0.78, 0.90
Dapueto et al. (2001)67	Uruguay	Spanish	140 cancer patients	0.83	5	PSDEF: 0.79, 0.65, 0.66, 0.60, 0.72
Thomas et al. (2004)68	India	Malayalam, English	214 cancer patients	0.80 (Malayalam patients) 0.89 (English patients)	4	PSEF: 0.75, 0.83, 0.77, 0.64 PSEF: 0.89, 0.63, 0.78, 0.84
Mullin et al. (2000)69	South Africa	3 South African languages	167 cancer patients	0.92 (Pedi patients) 0.89 (Tswana patients) 0.82 (Zulu patients)	5	PSDEF: 0.78, 0.60, 0.80, 0.87, 0.92 PSDEF: 0.79, 0.61, 0.92, 0.76, 0.83 PSDEF: 0.64, 0.36, 0.86, 0.77, 0.80
Conroy et al. (2004)70	France	French	310 cancer patients	0.89	5	PSDEF: 0.82, 0.69, 0.65, 0.74, 0.80
Fumimoto et al. (2001)71	Japan	Japanese	180 lung cancer patients	0.85	5	PSDEF: 0.81, 0.63, 0.63, 0.81, 0.69
Costet et al. (2005)72	France	French	493 cancer patients	0.90	4	PSEF: 0.86, 0.77, 0.83, 0.85
Smith et al. (2007)73	United Kingdom	English	465 cancer patients	Not reportedb	4	Range: 0.72–0.85a
Overcash et al. (2001)74	United States	English	112 cancer patients aged 65+	0.86	5	PSDEF: 0.78, 0.62, 0.65, 0.60, 0.85
Sanchez et al. (2011)75	Colombia	Spanish	473 cancer patients	0.89	4	PSEF: 0.85, 0.79, 0.85, 0.73

^a Range of subscale Cronbach's alpha reported otherwise subscales reported with respect to the order of listed subdomains.

^b Total scale internal consistency reliability not reported; range of internal consistency reliabilities for FACT-G subscales.

NR, not reported; PSDEF, physical well-being, social/family well-being, patient relationship with doctor, emotional well-being, and functional well-being; PSEF, physical well-being, social well-being, emotional well-being, and functional well-being; PSSEF, physical well-being, social well-being-family, social well-being-friends, emotional well-being, and functional well-being.

Methods

Study design and sample

This cross-sectional study pooled data of patients enrolled across 12 EDs in the Emergency Medicine Palliative Care Access (EMPallA) trial between April 2018 and January 2020. EMPallA is a multicenter, Patient-Centered Outcomes Research Institute (PCORI)-funded, two-arm randomized control trial (RCT). The RCT was registered on Clinical Trials.gov,²⁴ and approval was obtained from all appropriate institutional review boards. Further details of the EMPallA protocol have been published.^25,26 Since the index study used a cross-sectional design, nonrandomized data for patients that were enrolled in the aforementioned period were used for this psychometric analysis.

Patients enrolled in the study were 50 years or older, scheduled for ED discharge or observation status, spoke either English or Spanish, and had one or more of the four qualifying life-limiting conditions: advanced-stage metastatic solid tumor (hereafter referred to as “cancer”), class III or IV CHF, ESRD, or stage III or IV COPD. In addition, enrolled patients had clinic-accepted health insurance, resided in the geographic area of the recruiting participating EDs, and possessed a working telephone. Study respondents were patients with life-limiting illnesses, and the self-administered survey was administered and completed between day zero and five of the ED presentation.

Functional Assessment of Cancer Therapy-General

The 27-item FACT-G has four subscales: physical well-being (PWB) (seven items), social/family well-being (seven items), emotional well-being (EWB) (six items), and functional well-being (seven items), and these subscales assess the quality of life of patients. The FACT-G has a strong internal consistency, with an aggregate internal consistency of 0.88, averaged across 78 published studies.²⁷ Responses across all four subscales are scored on a 5-point Likert scale (0 = not at all; 1 = a little bit; 2 = somewhat; 3 = quite a bit; and 4 = very much). The FACT-G instrument provides an option for respondents to skip the seventh item in the social/family well-being domain (item GS7: “I am satisfied with my sex life”). About a third (n = 148) of the study participants skipped this item and we removed the item from the analysis. Study participants whose response to item GF1 (“I am able to work (include work at home)”) was “not at all” were asked to skip item GF2 (“My work (including work at home) is fulfilling”). A total of 191 study participants skipped item GF2, and we assigned the value of “not at all” to these GF2 responses, consistent with the principles of scoring valid skip responses.^28–30

Handling of missing data

We performed multiple imputations for missing values, using the Multivariate Imputation via Chained Equation (MICE) model,³¹ after establishing that the missing values were not missing completely at random (Little's test not significant).³² Across the 26-item FACT-G, missingness per item ranged from 1 (0.2%) to 29 (6.4%). For the imputations, we ran 100 iterations, generated 100 predicted values for each missing value, and the final value that replaced the missing value was obtained after generating the mean of the predicted values.^33,34

Scoring of the FACT-G

Based on the scoring guidelines,^35,36 all items in the PWB domain and five of the six items in the EWB domain were reverse scored. We obtained the subscale scores by summing the individual items in each subscale and we obtained the total FACT-G score by adding the scores across each of the four subscales. Higher scores reflect better quality of life.

Sample characteristics, item analysis, and reliability

We computed the frequency distribution of the sociodemographic and health characteristics of the study sample. Our item analysis consisted of four measures: mean, standard deviation (SD), item discrimination, and consistency. Mid-range values of the items' mean and SD are considered ideal.³⁷ Item discrimination assesses the correlation between the item score and the total score of the full scale.³⁸ The range of values is from −1 to +1, and values >0.20 are considered adequate.^38,39 The internal consistency of the FACT-G, a measure of reliability, was examined using Cronbach's alpha (α). We calculated Cronbach's α for the full scale and the four subscales. In addition, we calculated Cronbach's α for the subscales of the final factor structure after the confirmatory factor analysis (CFA). Following recommendations,^40–42 we considered Cronbach's α of 0.70–0.79 as “acceptable,” 0.80–0.89 as “good,” and equal to or >0.90 as “excellent.” We computed the recalculated α with each item deleted for the four-factor model. The recalculated α was used to justify that excluding an item will not improve the scale's reliability.

Split-half validation

To address the practical limitations of model building, we conducted split-half (i.e., cross) validation analyses. Specifically, because we could not collect new data to evaluate the fit of the model we tested, we used the same data to assess model fit. Cross-validation techniques focused on using half the data set for modeling the data (i.e., “training set” or “derivation sample”) and the other half for testing the performance of the model that was built (i.e., “testing set” or “replication sample”). The sample, therefore, was randomly divided into a derivation sample (n = 227) for the exploratory factor analysis (EFA) and a replication sample (n = 226) for CFA.

Exploratory factor analysis

The factor structure was examined by repeatedly exploring the latent structure using maximum likelihood methods. To identify the simplest factor model (i.e., the model with minimal or no cross-loadings of items on more than one factor) that best fits the observed data, we chose an iterative analytical approach with increasing complexity—from an unrotated model to orthogonal and oblique rotations. The number of possible factors was determined by examining the number of factors that correspond to Eigenvalues >1, the major bend in the scree plot, and more than 50% of the cumulative proportion of the variance of the FACT-G scores.⁴³ The results of unrotated, orthogonal rotated, and oblique rotated solutions were compared. To facilitate model interpretation, we suppressed factor loading coefficients <0.25 from the results.

Confirmatory factor analysis

To determine the reliability of the factor structure that emerged from the EFA, we performed a CFA. Model fit and factor loadings were evaluated. Specifically, we reported comparative fit index (CFI), Tucker–Lewis index (TLI), and root-mean-square error of approximation (RMSEA). CFI and TLI >0.90 are deemed adequate fit, while values >0.95 are deemed a good fit.^44,45 RMSEA values <0.05 indicate good fit, values 0.05–0.08 suggest adequate fit, values 0.08–0.10 suggest marginal fit, and values 0.10 and higher indicate poor fit.^44–46 Furthermore, we compared the factor structure that emerged from this study with the original four-domain factor structure (hereafter referred to as the “original FACT-G”), and we compared the models using Akaike information criteria (AIC) and Bayesian information criteria (BIC). The lower the AIC and BIC, the better the model. Post hoc model modifications were limited to placing covariance on the item-specific residual variance (represented by item error terms). These modifications were guided by the standardized expected parameter change (SEPC), modification index (MI), and conceptual relationship of the items identified by the SEPC and MI. Item residual was allowed to covary within their latent factor if there was a valid conceptual explanation, the SEPC was a value >0.2, and the MI was greater than the critical value of 3.84.⁴⁷ To minimize modification, the critical MI value was set to 10.0.

Data analysis

Frequency distribution and missing data analysis were conducted in R version 4.1.0.^48,49 Item analysis and EFA were conducted using the IBM Statistical Package for Social Sciences (SPSS) version 27.⁵⁰ CFA was conducted using IBM SPSS Amos, version 27.⁵¹

Results

Sample characteristics

Our sample consisted of 453 adults with life-limiting illnesses. The sample was predominantly aged 50–64 years (42%), female (54%), non-Hispanic White (57%), married (40%), with high school education or less (41%), and a yearly income <$25,000 (42%) (Table 2). With regards to functional status, ∼27% of the sample required occasional assistance with activities of daily living, 27% required considerable assistance, and 6% had a disability. The mean (SD) FACT-G score was 60.8 (18.1). There were no differences in the sociodemographic and health characteristics of the sample pooled into the derivation and replication samples.

Table 2.

Demographic and Health Characteristics Sample of Seriously Ill Adults Enrolled in the Emergency Medicine Palliative Care Access Study (n = 453)

Variables	Total population, n (%)	Derivation sample, n = 227 (%)	Replication sample, n = 226 (%)	p
Age categories
50–64 years	192 (42.4)	96 (42.3)	96 (42.5)	0.277
65–74 years	153 (33.8)	69 (30.4)	84 (37.2)
75–84 years	79 (17.4)	45 (19.8)	34 (15.0)
85 years and older	29 (6.4)	17 (7.5)	12 (5.3)
Sex
Male	209 (46.1)	110 (48.5)	99 (43.8)	0.321
Female	244 (53.9)	117 (51.5)	127 (56.2)
Race/ethnicity
Non-Hispanic White	257 (56.7)	123 (54.2)	134 (59.3)	0.500
Non-Hispanic Black	131 (28.9)	70 (30.8)	61 (27.0)
Hispanic	17 (3.8)	7 (3.1)	10 (4.4)
Other races	48 (10.6)	27 (11.9)	21 (9.3)
Marital status
Married	179 (39.5)	93 (41.0)	86 (38.1)	0.468
Never married	104 (23.0)	49 (21.6)	55 (24.3)
Separated/divorced	93 (20.5)	42 (18.5)	51 (22.6)
Widow/widower	77 (17.0)	43 (18.9)	34 (15.0)
Educational attainment
High school or less	186 (41.1)	93 (41.0)	93 (41.2)	0.114
Some college	142 (31.3)	69 (30.4)	73 (32.3)
Bachelor's degree	76 (16.8)	46 (20.3)	30 (13.3)
Graduate degree or higher	49 (10.8)	19 (8.4)	30 (13.3)
Yearly income categories
<$25,000	191 (42.2)	97 (42.7)	94 (41.6)	0.533
$25,000–$49,999	129 (28.5)	60 (26.4)	69 (30.5)
$50,000–$99,999	78 (17.2)	38 (16.7)	40 (17.7)
$100,000 and higher	55 (12.1)	32 (14.1)	23 (10.2)
Comorbid illness
Cancer only	173 (38.2)	93 (41.0)	80 (35.4)	0.128
CHF only	71 (15.7)	32 (14.1)	39 (17.3)
COPD only	85 (18.8)	36 (15.9)	49 (21.7)
ESRD only	65 (14.3)	32 (14.1)	33 (14.6)
2 or more comorbidity excluding cancer	45 (9.9)	29 (12.8)	16 (7.1)
2 or more comorbidity; cancer inclusive	14 (3.1)	5 (2.2)	9 (4.0)
Population type
Cancer population	187 (41.3)	98 (43.2)	89 (39.4)	0.413
Noncancer population	266 (58.7)	129 (56.8)	137 (60.6)
Functional status
Normal activity	123 (27.2)	70 (30.8)	53 (23.5)
Cares for self	102 (22.5)	47 (20.7)	55 (24.3)
Requires occasional assistance	123 (27.1)	59 (26.00)	64 (28.3)
Requires considerable assistance	76 (16.8)	35 (15.4)	41 (18.1)
Disabled	29 (6.4)	16 (7.0)	13 (5.8)
FACT-G score, mean (SD)
Physical Well-Being	15.7 (5.9)	16.1 (5.7)	15.3 (6.2)	0.156
Social Well-Being	17.2 (5.9)	17.0 (6.1)	17.5 (5.7)	0.330
Emotional Well-Being	15.4 (5.8)	15.8 (5.9)	15.0 (5.7)	0.131
Functional Well-Being	12.5 (6.7)	12.7 (6.7)	12.2 (6.6)	0.395
Total Fact-G Score	60.7 (18.1)	61.6 (18.0)	60.0 (18.2)	0.348

CHF, Congestive Heart Failure; COPD, Chronic Obstructive Pulmonary Disease; ESRD, End-Stage Renal Disease; FACT-G, Functional Assessment of Cancer Therapy-General; SD, standard deviation.

Item analysis and reliability of the original FACT-G factor model

Using the total sample, the mean (SD) of the 26 items in the FACT-G ranged from 1.23 to 3.14 (1.19–1.50) (Table 3). The mean (95% confidence interval [CI]) score for the FACT-G scale was 2.34 (2.16–2.52) with the social (2.87 [2.69–3.05]) and functional well-being subscales (1.78 [1.60–1.96]) having the highest and lowest mean scores, respectively. Item-total correlations ranged from 0.244 to 0.699. The internal consistency (Cronbach's α) for the total FACT-G scale was 0.884 and deleting any of the 26 items would not substantially improve the internal consistency of the scale. Since the item means and SDs were midrange values, the item-total correlations were higher than 0.20, and deleting any item would not improve the scale, all items were retained. The internal consistency for the physical, emotional, functional, and social well-being subscales was 0.72, 0.83, 0.81, and 0.80, respectively.

Table 3.

Functional Assessment of Cancer Therapy-General Item Analysis Using the Full Sample (n = 453) of Seriously Ill Adults Enrolled in the Emergency Medicine Palliative Care Access Study

ID	FACT-G question	Item code	Mean	SD	Item-total correlation	Subscale metrics (n = 453)
GP1	I have a lack of energy	Lack of energya	1.66	1.27	0.376	Physical Well-Being Average Score: 2.24 95% CI: 2.06–2.42 Cronbach's α: 0.715
GP2	I have nausea	Nauseaa	3.07	1.19	0.274
GP3	Because of my physical condition, I have trouble meeting the needs of my family	Meeting family needsa	2.46	1.49	0.396
GP4	I have pain	Paina	1.75	1.44	0.400
GP5	I am bothered by side effects of treatment	Side effectsa	2.59	1.46	0.244
GP6	I feel ill	Feel illa	1.99	1.41	0.545
GP7	I am forced to spend time in bed	Time in beda	2.18	1.48	0.388
GS1	I feel close to my friends	Close friends	2.57	1.45	0.404	Social Well-Being Average Score: 2.87 95% CI: 2.69–3.05 Cronbach's α: 0.828
GS2	I get emotional support from my family	Family support	3.00	1.33	0.440
GS3	I get support from my friends	Friend support	2.57	1.42	0.381
GS4	My family has accepted my illness	Family accepting illness	3.06	1.24	0.334
GS5	I am satisfied with family communication about my illness	Family communication	2.90	1.36	0.408
GS6	I feel close to my partner (or the person who is my main support)	Close to partner	3.14	1.30	0.355
GE1	I feel sad	Feel sad	2.39	1.31	0.536	Emotional Well-Being Average Score: 2.56 95% CI: 2.39–2.73 Cronbach's α: 0.814
GE2	I am satisfied with how I am coping with my illness	Coping with illness	2.30	1.32	0.559
GE3	I am losing hope in the fight against my illness	Fighting illnessa	3.08	1.25	0.500
GE4	I feel nervous	Feel nervousa	2.49	1.39	0.511
GE5	I worry about dying	Worry about dyinga	2.95	1.35	0.491
GE6	I worry that my condition will get worse	Condition worsea	2.15	1.40	0.485
GF1	I am able to work (include work at home)	Able to work	1.23	1.31	0.435	Functional Well-Being Average Score: 1.78 95% CI: 1.60–1.96 Cronbach's α: 0.799
GF2	My work (include work at home) is fulfilling	Work fulfilling	1.36	1.50	0.391
GF3	I am able to enjoy life	Enjoy life	2.02	1.35	0.699
GF4	I have accepted my illness	Accept illness	2.87	1.28	0.360
GF5	I am sleeping well	Sleep well	1.73	1.47	0.487
GF6	I am enjoying the things I usually do for fun	Fun things	1.58	1.46	0.603
GF7	I am content with the quality of my life right now	Quality of life	1.67	1.49	0.627
FACT-G Total Scale Metrics
Average Score: 2.34 95% CI: 2.16–2.52 Cronbach's α: 0.884

^a Item was reverse scored; all questions use a 5-point rating scale: 0 = not at all; 1 = a little bit; 2 = somewhat; 3 = quite a bit; and 4 = very much. Cronbach's alpha would range from 0.875 to 0.885 if any item is deleted, suggesting that reliability would not be substantially changed by deleting any one item. Item GS7 (“I am satisfied with my sex life”) was optional, and the missingness in GS7 was 33%. Item GS7 was excluded. However, when it was added to the analyses (after multiple imputations), Cronbach's alpha on the 27 items was 0.883; adding or removing the item GS7 did not improve or weaken the scale.

Exploratory factor analysis

Across the 26 items in the derivation sample, the cumulative proportion of explained variance >50% corresponded to a four-factor model, the major bend of the scree plot corresponded to a five-factor model, while the number of Eigenvalues >1 corresponded to a seven-factor model. After several iterations of EFA, the simplest model with the least cross-loading was a six-factor model (cumulative proportion = 58.7% of total variance), created through an orthogonal rotation.

Consistent with the original FACT-G, all seven items in the PWB category were loaded on one domain with estimates ranging from 0.30 to 0.66 (Table 4). Two separate domains, social well-being: family (SFM) and social well-being: friends (SFR), emerged from items that measure social well-being. Four items were loaded on SFM with estimates ranging from 0.35 to 0.85. Two items were loaded on SFR, with estimates being 0.79 and 0.87. All items but one (GE2) in the EWB in the original FACT-G were loaded on one domain, with estimates ranging from 0.58 to 0.76. Two separate domains, functional well-being: work (FWW) and functional well-being: daily activities (FWD), emerged from items that measure functional well-being. Two items were loaded on FWW, with estimates ranging from 0.75 to 0.92. All remaining four items in the original FACT-G's functional well-being scale along with an item from the EWB scale (GE2) were loaded on FWD, with estimates ranging from 0.30 to 0.71.

Table 4.

Exploratory Factor Analysis of the 26-Item Functional Assessment of Cancer Therapy-General Scale Showing the Identified Domains Among Seriously Ill Adults Enrolled in the Emergency Medicine Palliative Care Access Study for the Derivation Sample (n = 227)

Items ID	PWB	SFM	SFR	EWB	FWW	FWD
GP1	0.358				.
GP2	0.503
GP3	0.299
GP4	0.419
GP5	0.405
GP6	0.659
GP7	0.547
GS2		0.682
GS4		0.802
GS5		0.852
GS6		0.347
GS1			0.789
GS3			0.870
GE1				0.575
GE3				0.566
GE4				0.717
GE5				0.759
GE6				0.687
GF1					0.922
GF2					0.753
GF3						0.705
GF4						0.296
GF5						0.368
GF6						0.614
GF7						0.701
GE2						0.361

Rotation converged in six iterations. Item GS7 was excluded from analyses. Values represent the factor loading coefficients. Factor loading coefficients <0.25 were suppressed for ease of identifying the primary domain of each item. Extraction method was maximum likelihood. Rotation method was varimax with kaiser normalization.

EWB, emotional well-being; FWD, functional well-being: daily activities; FWW, functional well-being: work; PWB, physical well-being; SFM, social well-being: family; SFR, social well-being: friends.

Confirmatory factor analysis

A comparison of the four-factor structure of the original FACT-G domains and the EFA-derived six-factor model (without post hoc modifications) showed that in the replication sample, CFI and TLI values were higher in the six-factor model (Table 5). In addition, the RMSEA, AIC, and BIC values were lower in the six-factor model compared to the four-factor model. After assessing MI and SEPC, the model was modified post hoc by allowing covariations of item residual of two sets of items. The two sets of items are (1) GF4 and GE2 (“I have accepted my illness” vs. “I am satisfied with how I am coping with my illness”) and (2) GS4 and GS5 (“My family has accepted my illness” vs. “I am satisfied with family communication about my illness”). GF4 and GE2 were related conceptually, had high MI (12.4) and SEPC values (0.32), and the correlation coefficient of their item residuals was 0.25. Similarly, GS4 and GS5 were related conceptually, had high MI (15.0) and SEPC values (0.21), and the correlation coefficient of their item residuals was 0.48 (Fig. 1). The modified six-factor structure exhibited good factorial validity with a CFI of 0.93, TLI of 0.92, and RMSEA of 0.049 (90% CI: 0.040–0.058). Across the six domains in the replication sample (n = 226), all the path coefficients were >0.4 (Fig. 1).

Table 5.

Fit Indices for the Functional Assessment of Cancer Therapy-General Confirmatory Models Using the Replication Dataset (n = 226)

Fit index	Original FACT-G domains	EFA-based model	Final modified model (with interpretation)
	Four-factor structure	Six-factor structure	Six-factor structure
Comparative fit index	0.832	0.905	0.927 (adequate)
Tucker–Lewis index	0.814	0.891	0.916 (adequate)
RMSEA (90% CI)	0.070 (0.062–0.078)	0.056 (0.047–0.064)	0.049 (0.040–0.058) adequate
AIC	782.44	669.89	625.29 (lower is better)
BIC	805.35	695.26	651.20 (lower is better)
χ2 (df)	614.44 (293)	483.89 (284)	435.29 (282)
PCMIN/DF	2.097	1.704	1.544 (adequate)

Results are based on the replication sample.

χ² (df), Chi-square (degree of freedom); AIC, Akaike information criteria; BIC, Bayesian information criteria; CI, confidence interval; df, degrees of freedom; EFA, exploratory factor analysis; PCMIN/DF, ratio of Chi-square and the degree of freedom; RMSEA, root-mean-square error of approximation.

FIG. 1.

CFA using the replication dataset (n = 226) showing the six-factor correlated model structure for the FACT-G. Rectangles represent FACT-G items, and ovals represent latent variables. Numbers on straight arrows represent standardized path coefficients. Numbers on curved arrows represent the correlation coefficient. Item GS7 was excluded from analyses. CFA, confirmatory factor analysis; e, error; EWB, emotional well-being; FACT-G, Functional Assessment of Cancer Therapy-General; FWD, functional well-being: daily activities; FWW, functional well-being: work; PWB, physical well-being; SFM, social well-being: family; SFR, social well-being: friends.

Supplementary analyses

After confirming the six-factor structure of the FACT-G, we re-estimated Cronbach's α value across the six subscales. The Cronbach's α values for the PWB, SFM, SFR, EWB, FWW, and FWD subscales were 0.715, 0.816, 0.846, 0.816, 0.860, and 0.801, respectively (Supplementary File). Since our sample population represents patients with advanced cancer, CHF, ESRD, and COPD, we reported the results of the item reliability, convergent validity, CFA, and measurement invariance among patients with cancer and noncancer diagnoses (Supplementary File).

Discussion

We assessed the psychometric properties of the FACT-G in a sample of patients with life-limiting illnesses presenting to the ED. The 26-item FACT-G scale had good reliability, and the six-factor structure can be used to assess functional assessment among patients with life-limiting illnesses in the ED. In this study, the segregation of family and friends domains, and the distinction between work and other daily activities, in the social and functional well-being scales, respectively, emerged in the sample of patients with life-limiting illnesses presenting to the ED.

To our knowledge, this study is the only psychometric investigation of the FACT-G in a sample of adults with life-limiting illnesses who present to the ED. Indeed, a number of equally reliable and validated HRQOL questionnaires exist, some of which include Flanagan's Quality of Life Scale,⁵² McGill Quality of Life Questionnaire,⁵³ and the European Organization for the Research and Treatment of Cancer Quality of Life Questionnaire.⁵⁴ However, the FACT-G scale uniquely provides a measure of assessing the quality of the patient's relationships and support system.⁵⁵ Using the sufficiently high internal consistency (Cronbach's α) score of the FACT-G scale in our study population, the FACT-G can be considered as a reliable quality of life measure for adults with life-limiting illnesses who present in the ED.

We reported the reliabilities of the subscales in the original four-factor and our reported six-factor models. While the acceptable levels of Cronbach's α of the scales suggest that the instrument can reliably measure the quality of life of patients with life-limiting illnesses, the subscales portray the dimensionality that can be assessed.^56,57 The acceptable levels of Cronbach's α across all the other subscales further suggest good construct validity of the scale, good reliability of the subscales, and may provide additional information on which quality of life domains clinicians should address for each patient with life-limiting illnesses.

In contrast to the original four-factor structure, we report a six-factor structure, which may reflect the social and functional characteristics of patients with life-limiting illnesses at the time of presentation to the ED. Specifically, there were two social well-being (family and friends) domains with significantly different mean scores. A possible explanation may be that patients with life-limiting illnesses undergoing an acute event may place greater value on family in end-of-life care, compared to friends. In addition, there were two functional well-being (work and daily activities) domains with significantly different mean scores. This response pattern may suggest that the ability to work will likely be of less importance to adults with life-limiting illnesses experiencing acute events and this may explain why the mean scores of FWW were lower.

Given the widespread use of the FACT-G in both clinical and research settings, the present study offers numerous clinical implications. First, the FACT-G can be used to assess the quality of life among patients with life-limiting illnesses who presented to the ED. It is therefore a useful tool to identify the needs and resources of patients, initiate serious illness conversations, and monitor treatment outcomes and disease progression. Second, patients with life-limiting illnesses may view their social interactions with their family differently compared to social interactions with friends and may perceive their ability to work as a low priority compared to sleep or leisure activities. Indeed, patients' priorities vary widely, and clinicians should seek to understand and address each patient's priorities. Third, we advocate that among patients with life-limiting illnesses who present to the ED, item GS7 (“I am satisfied with my sex life”) can be skipped. Some studies have reported performing mean or multiple imputations for item GS7 that we excluded.^58,59 When we imputed the missing values in GS7, the resulting CFA model performed worse compared to the six-factor model, although substantially better than the four-factor 27-item FACT-G. Removing this item only improved our model without producing a change in the two social well-being domains. Furthermore, we advocate introducing a skip statement to item GF2 “My work (include work at home) is fulfilling” if the response to item GF1“I am able to work (include work at home)” is “not at all.” It is unlikely that patients with life-limiting illnesses will interpret “work” differently in items GF1 and GF2. We assessed the sensitivity of our introduction of a skip sequence in GF2 compared to imputing values for missing GF2. The six-factor structure was preserved in both cases, and the fit indices were not meaningfully different. Finally, we advocate that the reverse scoring of the six-factor FACT-G should be consistent with the original four-factor FACT-G, and the subscale scores should be multiplied by the number of items in the subscale divided by the number of items answered.³⁶ The range of scores of the 26-item six-factor FACT-G should, therefore, be 0–104.

This study has its limitations. The FACT-G is a self-reported measure, which cannot exclude self-report bias and social desirability responses.^60,61 While the EFA is subject to analytical bias, we mitigated this by establishing a rigorous analytical protocol and using a replication sample to confirm the constructs in the EFA. In addition, our findings may be generalized only to the patients with life-limiting illnesses in the 12 EDs from where the study sample was drawn. Our approach to handling the skipped responses in item GF2 might have added artificial homogeneity, increasing the reliability and factorial validity. Our sample population of patients with life-limiting illnesses consists of patients with cancer and noncancer diagnoses. The possibility exists that self-reported responses may differ across cancer and noncancer populations. We have addressed this limitation by comparing the results of item analysis, factorial validity, and assessing measurement invariance in cancer and noncancer populations (Supplementary File). Our result showed that the psychometric properties in these two groups were similar with subtle differences in the PWB subscale. Despite these limitations, this study provides a reliable tool that may be used to assess the functional status among patients with life-limiting illnesses who presented to the ED. While we provide some modifications to the construct and scoring of the six-factor FACT-G model, future studies may compare the predictive accuracy of the FACIT-Pal, the palliative care-modified FACT-G instrument, and our six-factor FACT-G model.

Conclusion

The FACT-G is a reliable tool for assessing functional status among patients with life-limiting illnesses admitted to the ED. Our validation highlights the different roles of family and friends in the social well-being of patients with life-limiting illnesses. In addition, patients with life-limiting illnesses may perceive their ability to work as a low priority. Given the importance of measuring HRQOL in palliative care, this study presents the FACT-G as a reliable tool that can be administered to patients with life-limiting illnesses who present to the ED.

Authors' Contributions

M.Y.: conceptualization, formal analysis, writing. O.A.: formal analysis, writing. M.F.: data curation. J.-B.B.: review and editing. K.V.A.: review and editing, project administration. A.M.C.: review and editing, project administration. K.S.G.: data curation, formal analysis. K.O.: supervision, writing. C.R.G.: supervision, investigation.

Institutional Review Board Statement

Approved by the New York University Grossman School of Medicine Institutional Review Board.

Informed Consent Statement

Written informed consent was obtained from study participants.

Disclaimer

All statements in this report, including its findings and conclusions, are solely those of the authors and do not necessarily represent the views of the PCORI, its Board of Governors, or the Methodology Committee.

Funding Information

This work was (partially) supported through a PCORI Award (PLC-1609-36306). Research reported in this manuscript is also supported by the National Cancer Institute of the National Institutes of Health under Award Number K08CA259538 awarded to Miryam Yusufov. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.

Author Disclosure Statement

No competing financial interests exist.

Supplementary Material

Supplementary File