Quality of Life in Adults With Metastatic Breast Cancer: The Impact of Between‐ and Within‐Person Changes in Symptoms and Positive Affect

ABSTRACT

Introduction

Quality of life (QoL) is important for patients with metastatic breast cancer (MBC). We examined QoL in daily life among patients with MBC using ecological momentary assessment (EMA).

Methods

From 2023 to 2024, patients with MBC were recruited from 5 Mid‐Atlantic sites. Eligible patients were English‐ or Spanish‐speaking, with a life expectancy > 6 months and access to a smartphone. Participants were invited to complete 12 EMAs over 4 weeks. EMAs assessed global QoL, symptom severity (depression, anxiety, pain, fatigue, slowed cognition, appetite loss, nausea, gastrointestinal distress, decreased libido), and positive affect (social connection, peace, joy). Multilevel mixed‐effects models analyzed the relationships between QoL and between‐person and within‐person variations in symptom severity and positive affect. We also explored associations between QoL and years since MBC diagnosis at the time of study entry.

Results

Participants (N = 108; Mean age 57.8 years) were a mean of 4.4 years post‐diagnosis (range = 0.06–20.65 years). Between‐person effects indicated QoL was worse among individuals reporting more nausea (β = −0.109, p = 0.035) and less joy (β = 0.593, p < 0.001). Within‐person effects indicated individuals' QoL was lower at times when they had more depression (β = −0.177, p < 0.001), anxiety (β = −0.049, p = 0.023), fatigue (β = −0.052, p = 0.009), appetite loss (β = −0.053, p = 0.028), and decreased libido (β = −0.058, p = 0.036) and less social connectedness (β = 0.091, p = 0.001), joy (β = 0.055, p = 0.047), and peace (β = 0.107, p < 0.001) than their personal average. QoL was not associated with years since MBC diagnosis (β = 0.106, p = 0.956). Age (β = 0.482, p = 0.013) and being partnered (β = 13.356, p = 0.005) at baseline were positively associated with estimated QoL at MBC diagnosis. Changes in QoL per year since MBC diagnosis were positively associated with being employed at baseline (β = 2.453, p < 0.001) and negatively associated with HR positive disease (β = −1.334, p = 0.001).

Conclusions

Both within‐person and between‐person variations in symptom severity and positive affect are associated with QoL, highlighting the need for adaptive, targeted intervention strategies to optimize QoL for all patients.

Individuals living with metastatic breast cancer experienced substantial within‐person and between‐person variations in symptom severity and positive affect, which affected their quality of life. These data highlight the need for adaptive, targeted intervention strategies to optimize quality of life for all patients.

1. Introduction

The number of individuals in the United States living with metastatic breast cancer (MBC) is growing due to significant increases in both MBC incidence and survival [1]. Nearly a quarter of a million people in the United States will be living with MBC by 2030, a 55% increase from the 2015 estimate of approximately 159,000 [2]. The incurable nature of MBC makes understanding and supporting quality of life (QoL) a priority [3]. Compared to survivors with earlier stage BC, more widespread disease and prolonged treatment exposure in MBC can be associated with significant physical and psychological symptoms that limit daily functioning and activities [4, 5, 6]. The MBC experience is also different from other types of metastatic cancer; individuals with MBC report more emotional well‐being concerns, more symptom burden and impact, and more concern regarding relationships and intimacy than individuals living with other types of metastatic cancer (e.g., colorectal, prostate, lung) [7]. Thus, understanding and addressing QoL in MBC could help to mitigate the burden of disease in this growing population.

As individuals live longer with MBC, there is a pressing need to understand if and how their QoL changes over time [8, 9]. Prior research examining QoL longitudinally among those with MBC is primarily restricted to pharmaceutical clinical trials [10, 11, 12, 13, 14]. While valuable, such studies may not be reflective of the broader population, as treatment trials among adult patients with cancer may enroll approximately 7.1% of affected patients [15], and pharmaceutical trials' inclusion criteria often require higher functional statuses and exclude those with significant comorbidities [16, 17]. These trials' assessment timepoints also typically align with the beginning and end of treatment cycles, overlooking more granular changes that happen within treatment and reducing generalizability to QoL between treatment cycles. Evidence outside of pharmaceutical trials, while limited, offers conflicting results [18, 19]. Whereas data from pharmaceutical trials have primarily found significant decrements to QoL over time [12, 13, 14], recent observational studies have reported no or minimal changes [18, 19, 20], perhaps due to differences in study samples and eligibility criteria, emphasizing a need for further research. Additionally, existing longitudinal assessments of QoL typically employ traditional retrospective patient‐reported outcomes and relatively long periods between study assessments [11, 21, 22, 23], which are subject to recall bias and may obscure important fluctuations in patients' everyday symptom experiences that inform QoL [24]. These types of assessments also preclude examination of between‐person (i.e., differences between individuals) and within‐person (i.e., differences within the same individual over time) variability in QoL. This is particularly relevant in MBC, as patients with advanced cancer demonstrate significant day‐to‐day, within‐person changes in well‐being [25]. Thus, existing research may not adequately reflect the real‐world daily experiences of those living with MBC, and patterns of QoL over time in this population require further clarification. Such longitudinal information on QoL may help inform treatment decision‐making, symptom management, and supportive care planning [26, 27]. Specifically, longitudinal QoL data can capture the cumulative effect of MBC treatment over time, allowing for shared decision‐making that balances therapeutic benefits against patient tolerance. Tracking longitudinal trends in QoL could also enable early, proactive symptom intervention and the timely adjustment of supportive care as patient needs evolve.

In addition to temporal patterns, further research is needed to identify potential drivers of QoL for patients with MBC. While the significant physical and psychological symptom burden of MBC has been consistently demonstrated, it is less clear which symptoms have the greatest effects on QoL. Furthermore, symptom burden has primarily been explored at the between‐person level, focusing on which patients tend to have worse symptoms. Within‐person variation in symptoms has received limited attention but could help us to identify times when patients are most vulnerable to poor QoL. In other words, it answers the question: is a patient at risk for poor QoL when symptom burden is worse than usual? Finally, the potential buffering role of positive experiences has received little attention in the context of MBC. Meta‐analytic data demonstrate the reliable effects of positive affect on QoL in a variety of clinical and non‐clinical populations, including early‐stage BC, but it is unclear if these benefits extend to the metastatic setting [28, 29, 30]. Understanding which experiences—positive or negative—are most related to QoL for patients with MBC could help drive development of targeted interventions.

To address these gaps, the present study employed ecological momentary assessment (EMA), a methodological approach that can improve ecological validity, address recall bias associated with retrospective reporting, and capture dynamic patterns of change in participants' real‐world, daily lives [24, 31, 32]. EMA asks participants to repeatedly report their experiences in the present moment and therefore provides rich, temporally‐sensitive data that offers insight into participants' well‐being between typical monitoring appointments [24, 31, 32]. Using this EMA data, the study aimed to examine the relationships between QoL and differences in symptom burden and positive affect across different individuals (between‐person differences) and among the same individual over days of assessment (within‐person differences). We also explored associations between QoL and time since MBC diagnosis.

2. Materials and Methods

This longitudinal, observational study among people living with MBC utilized EMA to examine QoL. This research was designed and conducted with the active input of patient advocate collaborators to ensure the methodology was informed by the perspectives of individuals living with MBC.

2.1. Participants and Procedures

Participants were recruited at five sites: (1) MedStar Georgetown University Hospital (Washington, DC), (2) MedStar Washington Hospital Center (Washington, DC), (3) MedStar Good Samaritan Hospital (Baltimore, MD), (4) MedStar Franklin Square Medical Center (Baltimore, MD), and (5) Hackensack Meridian Health John Theurer Cancer Center (Hackensack, NJ). All study procedures were standardized across all sites. Participants met the following inclusion criteria: receiving care at one of the five sites for a diagnosis of metastatic breast cancer, assigned female at birth, current age ≥ 18, English‐ or Spanish‐speaking, and possession of a working cell phone with SMS messaging and internet access. We restricted the sample to participants assigned female at birth due to the rare and distinct nature of breast cancer in individuals assigned male at birth [33]. We excluded individuals who were unable to complete study assessments on their own and those who had a life expectancy of less than 6 months as determined by their treating breast oncologist [34]. While study inclusion and exclusion criteria were necessary to ensure participant ability to complete the study, these criteria may have selected for those with higher functional status and health literacy [35]. Participants were recruited between January 2023 and April 2024.

After providing informed consent, participants completed a baseline survey pertaining to sociodemographic and health characteristics. Participants also underwent a standardized EMA “training” session, in which a member of the research team reviewed the process of receiving the prompt and completing the assessments on their devices to ensure functionality and understanding of the process. The training session took place via telephone or video call, as preferred by the participant. It was standardized across sites. Following this training, participants began completing EMA prompts, which consisted of timed text reminders to complete assessments of global QoL, symptom severity, and positive affect (see Table S1). EMA prompts occurred as bursts: three times a day (morning, afternoon, evening), 1 day per week (EMA assessment day was randomly assigned each week), for four consecutive weeks. This “measurement burst” design incorporates short periods of intensive, recurrent assessment that are repeated longitudinally and spaced out over longer intervals. This design reduces participant burden while maintaining density of assessments [36, 37]. On average, participants completed the assessments in 2 min and 47 s. Participants were enrolled in the study from January 2023 to April 2024, with follow‐up through May 2024. Participants received up to $35 for completing all surveys.

2.2. Measures

2.2.1. Sociodemographic and Health Characteristics

Participants self‐reported their age, gender identity, race, ethnicity, relationship status, sexual orientation, education, employment, income, insurance status, and current caregiving responsibilities (yes/no). Health characteristics were abstracted from electronic medical records, including breast cancer subtype (hormone receptor [HR] status, HER2 status) and treatment type(s).

2.2.2. Ecological Momentary Assessments: QoL, Symptom Severity, Positive Affect

Participants reported their global QoL, symptom severity, and positive affect using a single visual analog scale item [38, 39]. Response options ranged from 0 to 100, with higher scores on each item indicating better QoL, greater severity of symptoms, and greater positive affect (see Table S1 for anchors). The global QoL item asked participants, “how would you rate your quality of life right now?,” describing QoL as overall enjoyment of life, including physical, emotional, and social well‐being (see Table S1). Symptom severity items included current severity of depression, anxiety, pain, fatigue, slowed cognitive functioning, appetite loss, nausea, gastrointestinal (GI) distress, and decreased libido (higher scores reflect greater symptom severity). Positive affect items included current experiences of social connectedness, peace, and joy (higher scores reflect greater positive affect), which were selected based on the positive psychology literature [40, 41] and in collaboration with patient advocates based on their relevance to the experiences of patients with MBC. EMA questionnaire items were reviewed by patient advocates prior to use to ensure readability and comprehensibility but did not undergo formal cognitive testing. However, single visual analog scale items have been used extensively to assess QoL among individuals with cancer and demonstrate good reliability and validity [38, 42].

2.3. Analyses

All analyses were conducted using IBM SPSS Statistics for Windows, version 31 (IBM Corp., Armonk, NY, USA). Descriptive statistics were used to characterize the sample. Bivariate Pearson correlations or Analysis of Variance (ANOVA), as appropriate, were then used to examine the associations between baseline sociodemographic and health variables and the person‐mean QoL (i.e., the individual's average QoL score across all timepoints). The person‐mean was used for these analyses to account for the clustered nature of the EMA data (i.e., multiple observations within persons).

We employed mixed models to examine predictors of QoL. First, we ran unconditional no growth models for QoL, symptoms, and positive affect to determine the extent to which QoL, symptoms, and positive affect varied within‐ and between‐ persons. Characterizing the relative variance in this way indicates whether outcomes are amenable to time‐varying characteristics. Second, we ran a conditional no growth model to assess whether symptom and positive affect experiences predicted QoL using both within‐ and between‐person estimates of each predictor. That is, each symptom and positive affect construct had a corresponding within‐person variable (i.e., deviation from an individual's average; person‐mean centered), as well as a between‐person variable (i.e., deviation from the group's average; grand mean centered). This allowed the model to assess, for example, the impact of times when an individual feels more or less depressed than their personal average, as well as the impact of an individual being, on average, more or less depressed than the sample's mean level (see Figure 1 for an illustration of these within‐ vs. between‐person effects). This model adjusted for age, race and ethnicity, partner status, employment status, HR status, and HER2 status, as prior literature has demonstrated their association with QoL in breast cancer [43, 44, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55]. It also adjusted for study site. Only random intercepts were specified for this model. All predictors were entered simultaneously; symptoms and positive affect constructs where neither within‐person nor between‐person variation were significantly related to QoL at p < 0.1 were then removed from the model.

FIGURE 1.

Examples of within‐ versus between‐person variation: Depression ratings for two participants across study assessments. Figure depicts patterns of depression among two participants. Participant A has a higher overall average rating of depression, but less within‐person variability in ratings. Participant B has a lower overall average rating, but more within‐person variation.

Third, we ran an unconditional growth model to explore cross‐sectional associations between QoL and years since diagnosis. We did not follow participants from the time of diagnosis, as participants entered the study at variable points after diagnosis. Rather, we utilized participants' time‐varying time since diagnosis as a meaningful time variable within the context of this study. Finally, we utilized conditional growth models, adjusting for study site and the abovementioned sociodemographic and health variables selected based on the extant literature, to assess whether QoL and its association with time since diagnosis were related to sociodemographic and health characteristics. This model specified both random intercepts and random slopes. All predictors were entered simultaneously.

The sample size was determined a priori. We powered the study to detect a small‐to‐medium fixed between‐person effect (d = 0.25) of depression on QoL at α = 0.05 [56]. Power analyses specified α = 0.05 and a person‐level intraclass correlation coefficient (ICC) of 0.50. The target N was 120 participants, or 964 observations (assuming 10% attrition at each measurement burst and an EMA response rate of 78% of EMA prompts, based on the prior literature [57, 58, 59]). Although the final sample size (N = 108) was lower than our original target, the study's final observation count (i.e., 1101) exceeded the target number of observations. Therefore, this study had adequate power to detect the specified effect.

3. Results

A total of 295 patients were contacted, and 225 responded (Figure 2). Of those who responded, 125 consented (56%). The remainder either actively declined (31%) or were ineligible (14%). Reasons for ineligibility included not having the required technology, not speaking English or Spanish, dying before consent, moving to an outside care facility or hospice care and therefore having < 6 months to live, or being unable to provide informed consent (see Figure 2). After providing informed consent, 118 participants (94% of those who consented) completed the baseline survey, and a total of 108 participants (86% of those who consented) completed at least one EMA survey. Thus, the final analytic sample was N = 108. Loss to follow up and dropout remained below 10% throughout EMA bursts (total drop out of 8% from time of consent). To characterize potential sample bias, we examined sociodemographic and health characteristics of participants who did and did not complete EMAs (see Table S2). Compared to participants who completed at least one EMA survey (N = 108), participants who consented but did not complete any EMA surveys (N = 17) were more likely to be Spanish‐preferring (2.8% vs. 17.6%, p = 0.033), have HER2‐positive disease (25.0% vs. 47.1%, p = 0.038), and be receiving chemotherapy (26.9% vs. 52.9%, p = 0.017), and less likely to have private health insurance (68.5% vs. 35.3%, p = 0.008). The final amount of observations totaled 1101, with an average cluster size of 10.2 observations per participant (SD = 2.108, median = 11, range: 1–12).

FIGURE 2.

Study participant flow. Lost to follow up refers to individuals who became unresponsive to study team contact. Drop out refers to individuals who confirmed they no longer wished to participate in the study.

3.1. Sample Characteristics

Table 1 presents the sample's sociodemographic and health characteristics. Participants were, on average, 57.8 years of age and diagnosed 4.4 years ago (range: 0.06–20.65 years; median = 2.5 years). Participants most frequently identified as Non‐Hispanic White (44.4%), Non‐Hispanic Black (35.2%), or Hispanic or Latino/a (9.3%). Most (53.7%) were partnered; patients who were partnered (vs. single) reported higher average QoL across EMA assessments (r = 0.247, p = 0.011).

TABLE 1.

Participant sociodemographic and health characteristics and association with person‐mean quality of life scores (N = 108).

Characteristic	N (%) or mean (SD)	Association with person‐mean quality of life a
Age (years)	57.8 (12.1)	0.103
Race and ethnicity		1.871
Non‐Hispanic White	48 (44.4%)
Non‐Hispanic Black	38 (35.2%)
Hispanic or Latino/a	10 (9.3%)
Other/multiple	12 (11.1%)
Relationship status		0.247*
Partnered	58 (53.7%)
Single	48 (44.4%)
Missing	2 (1.9%)
Sexual or gender minority (SGM) status		−0.053
Yes	4 (3.7%)
No	101 (93.5%)
Missing	3 (2.8%)
Education level		−0.056
Less than bachelor's degree	46 (42.6%)
Bachelor's degree or higher	62 (57.4%)
Employment status		0.036
Employed	54 (50.0%)
Unemployed	46 (42.6%)
Missing	8 (7.4%)
Annual household income (USD)		−0.173
< $15,000	9 (8.3%)
$15,000–24,999	7 (6.5%)
$25,000–49,999	18 (16.7%)
$50,000–75,999	9 (8.3%)
$75,000–99,999	5 (4.6%)
≥$100,000	43 (39.8%)
Missing	17 (15.7%)
Health insurance status b
Private	74 (68.5%)	0.140
Public	51 (47.2%)	−0.071
Current caregiving responsibilities		−0.043
Yes	34 (31.5%)
No	73 (67.6%)
Missing	1 (0.9%)
Time since diagnosis (years)	4.4 (4.5)	−0.066
Metastatic breast cancer subtype b
HER2 status (positive)	27 (25.0%)	−0.066
HR status (positive)	80 (74.1%)	0.108
Current treatment(s) b
Chemotherapy	29 (26.9%)	0.009
Radiation	7 (6.5%)	−0.112
Targeted therapy	64 (59.3%)	−0.016
Immunotherapy	14 (13.0%)	−0.094
Hormone therapy	61 (56.5%)	0.092
Treatment changed during study	19 (17.6%)	−0.069

Abbreviations: HR, hormone receptor; SD, standard deviation.

^a Represents Pearson or point‐biserial correlation coefficient (r) for all variables except race and ethnicity. The association between race and ethnicity and quality of life was examined via ANOVA and the F‐statistic is presented.

^b Responses not mutually exclusive; counts reflect all response options that applied. Totals may therefore not equal 100%.

^* p < 0.05.

3.2. QoL, Symptom Burden, and Positive Affect

Across all EMA timepoints, participants had a mean overall QoL rating of 71.1 (see Table 2). On average, participants' most severe symptoms were decreased libido (59.5) and fatigue (35.8), while appetite loss (21.2) and nausea (10.7) were least severe.

TABLE 2.

Quality of life descriptive characteristics.

Characteristic	Range	Mean (SE)	ICC	Correlation with quality of life*
Global quality of life	15–100	71.1 (1.6)	0.722	—
Symptom severity
Depression	0–96	25.7 (2.0)	0.694	−0.750
Anxiety	0–100	30.9 (2.1)	0.663	−0.636
Pain	0–100	23.3 (2.1)	0.702	−0.342
Fatigue	0–100	35.8 (2.2)	0.646	−0.542
Slowed cognition	0–98	24.5 (1.9)	0.760	−0.573
Appetite loss	0–95	21.2 (2.0)	0.678	−0.377
Nausea	0–99	10.7 (1.4)	0.539	−0.367
Gastrointestinal distress	0–99	24.1 (1.9)	0.542	−0.493
Decreased libido	0–100	59.5 (3.2)	0.899	−0.340
Positive affect
Social connection	0–100	75.3 (1.9)	0.748	0.709
Peace	0–100	73.2 (1.9)	0.703	0.766
Joy	0–100	70.1 (1.8)	0.660	0.776

Note: Higher scores indicate better quality of life and more symptom severity and positive affect as assessed by visual analog scale items (possible score range 0–100). Mean and standard error of the coefficient obtained via unconditional no growth models.

Abbreviations: ICC, intraclass correlation coefficient; SE, standard error of the coefficient.

^*All correlations significant at p < 0.001.

Unconditional no growth models evaluating the extent to which QoL, symptoms, and positive affect varied within‐ and between‐persons indicated that 72.2% of the total variance in QoL was attributable to differences between individuals (see Table S3). Among symptoms and positive affect, the amount of between‐person variation was highest among libido (89.9%), cognition (76.0%), and social connection (74.8%). All symptom severity and positive affect scores were significantly correlated with QoL ratings at p < 0.001 (r values ranging from ±0.340–0.776). See Table 2 for full descriptive characteristics, ICCs, and correlation coefficients.

3.3. Between‐ Versus Within‐Person Predictors of QoL

Mixed models evaluating whether between‐person or within‐person variation in symptom and positive affect experiences predicted QoL (Table 3) indicated that, among between‐person predictors, QoL was worse among individuals who reported more nausea (β = −0.109, p = 0.035), and better among those who reported more joy (β = 0.593, p < 0.001). Additionally, within‐person predictors indicated that QoL was lower at times when individuals had more depression (β = −0.177, p < 0.001), anxiety (β = −0.049, p = 0.023), fatigue (β = −0.052, p = 0.009), appetite loss (β = −0.053, p = 0.028), and decreased libido (β = −0.058, p = 0.036) compared to their personal average. QoL was higher when individuals had more social connectedness (β = 0.091, p = 0.001), joy (β = 0.055, p = 0.047), and peace (β = 0.107, p < 0.001) compared to their personal average.

TABLE 3.

Multilevel model results of between‐ and within‐person predictors of global quality of life.

Predictor	β	SE	p
Between‐person
Depression	−0.130	0.070	0.069
Anxiety	0.023	0.046	0.614
Fatigue	−0.081	0.042	0.056
Appetite loss	0.028	0.040	0.487
Nausea	−0.109	0.051	0.035
Decreased libido	−0.024	0.018	0.197
Social connection	0.062	0.056	0.266
Joy	0.593	0.091	< 0.001
Peace	−0.044	0.111	0.695
Within‐person
Depression	−0.177	0.027	< 0.001
Anxiety	−0.049	0.022	0.023
Fatigue	−0.052	0.020	0.009
Appetite loss	−0.053	0.024	0.028
Nausea	−0.028	0.024	0.234
Decreased libido	−0.058	0.027	0.036
Social connection	0.091	0.028	0.001
Joy	0.055	0.027	0.047
Peace	0.107	0.030	< 0.001
Covariates
Age (years)	−0.006	0.058	0.921
Race and ethnicity (ref: NHW)	−0.129	0.594	0.828
Partner status (ref: single)	1.118	1.300	0.392
Employment status (ref: not working)	0.167	1.208	0.890
HR status (ref: negative)	1.345	1.438	0.352
HER2 status (ref: negative)	0.945	1.336	0.481
Study Site (ref: Site 1)	−0.469	0.360	0.196
Intercept	54.264	5.942	< 0.001
Variance components
Intercept	17.703	3.724	< 0.001
Residual	73.645	3.604	< 0.001

Note: Quality of life, symptom severity, and positive affect were assessed via visual analog scale (possible score range 0–100; higher scores indicate better quality of life, more symptom severity, and greater positive affect). Conditional pseudo R ² = 0.793.

Abbreviation: SE, standard error.

3.4. Association Between QoL and Time Since Diagnosis

Unconditional growth models exploring the relationship between QoL and years since diagnosis (see Table S4) did not find a significant linear (p = 0.743) association between QoL and time since diagnosis (−2LL = 8417.94; AIC = 8429.94; BIC = 8459.85). We then used conditional growth models to assess whether QoL and its association with time since diagnosis were related to sociodemographic and health characteristics (Table 4). Compared to the unconditional growth model, the conditional growth model presented in Table 4 demonstrated better model fit (−2LL = 7594.01; AIC = 7634.01; BIC = 7731.70). Participants who were older (β = 0.482, p = 0.013) and partnered (vs. single, β = 13.356, p = 0.005) at the time of the baseline survey were projected to have significantly higher QoL at MBC diagnosis. Although time since MBC diagnosis was not significantly related to QoL (p = 0.956), we did observe a significant interaction of time since MBC diagnosis and employment status and HR status. Compared to participants who were not working at the time of the baseline survey, those who were working had a projected 2.5‐point increase in QoL per year since MBC diagnosis (β = 2.453, p < 0.001). Compared to participants with HR negative breast cancer, those who had HR positive breast cancer had a projected 1.3‐point decrease in QoL per year since MBC diagnosis (β = −1.334, p = 0.001).

TABLE 4.

Multilevel model examining association between time since diagnosis and QoL, adjusting for sociodemographic and health factors.

Predictor	β	SE	p
Fixed effects—Intercept
Time since diagnosis (years)	0.106	1.948	0.956
Age (years)	0.482	0.194	0.013
Race and ethnicity (ref: NHW)	−0.871	2.421	0.719
Partner status (ref: single)	13.356	4.694	0.005
Employment status (ref: not working)	−8.679	4.667	0.063
HR status (ref: negative)	7.435	4.720	0.116
HER2 status (ref: negative)	4.609	5.637	0.414
Study Site (ref: Site 1)	−2.975	1.409	0.035
Fixed effects—Slope
Time × Age	−0.033	0.025	0.230
Time × Race and ethnicity	−0.118	0.270	0.663
Time × Partner status	−0.385	0.650	0.554
Time × Employment status	2.453	0.586	< 0.001
Time × HR status	−1.334	0.405	0.001
Time × HER2 status	−0.337	0.545	0.536
Time × Study site	0.615	0.183	< 0.001
Intercept	46.202	14.822	0.002
Variance components
Intercept	278.044	51.624	< 0.001
Residual	106.397	5.068	< 0.001

Note: Quality of life was assessed via visual analog scale (possible score range 0–100; higher scores indicate better quality of life). Conditional pseudo R ² = 0.727.

Abbreviations: HR, hormone receptor; NHW, non‐Hispanic White; SE, standard error.

4. Discussion

This prospective, longitudinal study of 108 women with MBC examined QoL by within‐ and between‐person variations in symptom severity and positive affect, as well as its relationship with time since diagnosis. The results indicated that both within‐ and between‐person variation in symptoms and positive affect were important determinants of individuals' QoL, which was not significantly associated with time since MBC diagnosis. These results have important implications for delivering timely interventions and clinical care that are tailored to individuals living with MBC.

To our knowledge, this study is the first to compare the unique influences of between‐ versus within‐person symptom variation on overall QoL among individuals with MBC. While both within‐ and between‐person differences predicted QoL, the pattern of results emphasizes the importance of within‐person predictors. Instances when individuals are feeling worse than usual may be particularly influential in determining their QoL, and these instances may therefore be ideal times to provide intervention [60]. Future research should further explore the use of adaptive or “just‐in‐time” interventions, in which dynamic symptom monitoring can effectively recognize when patients need more support and deliver the type and amount of support needed accordingly. This would allow for intervention over time when indicated by symptom reporting. Existing interventions to improve QoL in those living with MBC are sparse and demonstrate varying efficacy, leaving patients with unmet supportive care needs [26, 27, 61, 62, 63, 64]. It is possible that an adaptive intervention approach could improve reach to vulnerable patients and provide tailored, timely support that addresses their unmet needs.

Moreover, the pattern of significant symptoms and positive affect experiences present important considerations for research and practice. Overall, the physical symptoms significantly predicting patients' QoL (particularly within‐person fatigue, appetite loss, and decreased libido) may be related to the significant psychosocial experiences (within‐person depression, anxiety, and lower positive affect). Prior research in breast cancer has suggested the existence of symptom clusters that include combinations of these psychosocial and physical symptoms [65]. It may be useful to explore how these experiences may cluster specifically within MBC to inform clinical practice. For example, patients with MBC may benefit from expanded supportive care options to address these symptoms more comprehensively, such as nutrition consults, sexual health‐focused interventions, and mental health and social support resources.

Similarly, the pattern of findings highlights the importance of positive affective experiences in patients' QoL. Individuals' levels of social connection, peace, and joy all significantly predicted overall QoL, yet existing interventions to improve psychosocial well‐being in this population primarily focus on reducing negative affect rather than boosting positive affect [27, 64]. Specifically, only one positive psychological intervention has been conducted among patients with metastatic breast cancer, and it was completed nearly 10 years ago [66]. Considering advances in treatment and individuals' changing experiences living with MBC, additional research is required to further explore the impact of positive affect on QoL and the utilization of positive affect as a target of intervention.

While we did not follow patients longitudinally over time since diagnosis, we were able to explore the relationships between QoL, time since MBC diagnosis, and sociodemographic and health characteristics at study entry. Results demonstrated that the relationship between QoL and time since diagnosis varied with baseline sociodemographic and health characteristics. Specifically, participants who were older and partnered at the time of the baseline survey were projected to have significantly higher QoL at MBC diagnosis. These findings build upon previous research suggesting a significant need for support among younger, single women after cancer diagnosis: unmarried and younger patients may present with more aggressive MBC subtypes, be diagnosed at later stages, and experience high levels of psychosocial distress during and after treatment [46, 47, 48, 49, 50, 51]. Furthermore, our statistical models also projected increases in QoL over time since MBC diagnosis for participants who were working at the time of the baseline survey and decreases in QoL over time since MBC for participants with HR‐positive disease. While these effects may reflect the previously observed benefit of employment on QoL for breast cancer survivors [52] or the progression to later lines of therapy involving more toxic treatments for patients with HR‐positive disease [67, 68], results must be interpreted with caution given that change in QoL over time was not directly observed but was instead projected from the statistical model. As such, reverse causation and survival bias are also plausible explanations for these findings. Longitudinal cohort studies of patients living with MBC are needed to confirm these findings.

Finally, we observed no relationship between time since diagnosis and QoL, which was not consistent with the lived experiences of our patient advocate collaborators. While our results align with recent findings from other observational studies [18, 19, 20], they conflict with findings from pharmaceutical trials demonstrating worsening QoL over time [12, 13, 14]. Further research is necessary to explore changes in QoL over time since diagnosis, and these findings must be interpreted with caution for several reasons. First, we did not follow a cohort of patients with MBC from the time of diagnosis. Rather, we recruited patients who had been living with MBC for any period of time. As a result, our sample had a very wide range of years since MBC diagnosis (range: 0.06–20.65 years). Our statistical analyses modeled associations between QoL and time since diagnosis because it was more meaningful and interpretable than time in the study. However, the trends observed here are statistical projections based on cross‐sectional reports of time since diagnosis at study entry and may differ from QoL data from long‐term longitudinal observational studies. Thus, these null findings should be examined further in longitudinal cohort studies of patients living with MBC.

4.1. Strengths and Limitations

The present study contains a number of methodological strengths. The sample had a diverse representation of race and ethnicity, MBC subtypes, treatments, and time since diagnosis. The use of EMA allowed for frequent collection of data to provide a representative depiction of daily life with MBC, and the measurement burst design was feasible and acceptable for participants, with high completion rates across bursts (≥ 96%). This methodology yielded unique data that few other studies have collected in women with MBC. Furthermore, the methodology was designed in collaboration with patient advocates, ensuring the assessments' relevance to patients' lived experiences.

The study also has several potential limitations. Of the 225 individuals we established contact with, 56% consented to participate; we do not have data on non‐responders or those who declined, so we are not able to compare, but it is possible that the results may be subject to selection bias. The most common reason for participant ineligibility was lack of access to the required technology (i.e., cell phone capable of receiving text messages and accessing the internet). Patients with limited technology literacy may systematically differ in their QoL and symptom experiences [69]. Future studies should consider integrating lower‐burden options for EMA, such as paper‐and‐pencil diaries, which could increase accessibility for patients with limited technology literacy and thus improve generalizability of results. In this study, participants who did not complete the EMAs were more likely to be Spanish‐preferring, have HER2 positive disease, and be receiving chemotherapy, and less likely to have private health insurance. Disease, treatment, and health insurance factors may represent competing demands that could make participation in EMA research challenging. Although this study was supported by bilingual research staff, Spanish‐preferring participants may have experienced more barriers to participation than English‐preferring participants. Additional research may be needed to understand best ways to engage participants with high disease burden and/or limited English proficiency in studies involving EMA. Additionally, although we were able to compare the characteristics of those who did versus did not complete the EMAs, there was a high rate of missing sociodemographic data among participants who did not complete the EMAs, limiting the interpretability of this comparison.

Participants also self‐reported their experiences using single visual analog scale items, which may be susceptible to social desirability and demand characteristics (i.e., reports that are influenced by a desire to align with expectations or social norms). Additionally, while the use of single items reduced participant burden, they may not perform as well as multi‐item measures of QoL or symptom burden, which may better reflect the nuance of multidimensional constructs such as depression, anxiety, or fatigue. However, prior research has established single items' relative equivalence with regard to psychometric performance [38, 42]. Furthermore, participants were recruited from five sites across three metropolitan areas in the Mid‐Atlantic region of the United States. Although these sites represent both academic and safety‐net hospitals with varying patient populations, findings may differ across institutions and other regions within and outside of the United States. Finally, there is still unexplained variance within the models (i.e., as assessed via pseudo R² measures), suggesting there are important but unmeasured variables that may further predict QoL. We chose to focus on this study's symptoms and positive affect due to their relevance to our patient collaborators' lived experiences, but other constructs, such as spiritual (e.g., meaning and purpose, faith) or environmental factors (e.g., physical safety and security, home environment, etc.), may contribute to QoL, as well [70].

5. Conclusion

QoL is an important outcome for patients with MBC. The present study found substantial within‐person and between‐person variations in symptom severity and positive affect that were associated with QoL. Results suggest that within‐person variation in symptoms and positive affect is important but understudied determinants of QoL. QoL also varied by sociodemographic factors, the influence of which varied by time since MBC diagnosis. These data highlight the need for adaptive, targeted intervention strategies to optimize QoL for all patients.

Author Contributions

Brenna Mossman: conceptualization, formal analysis, visualization, writing – original draft, writing – review and editing. Brent J. Small: conceptualization, funding acquisition, methodology, formal analysis, writing – review and editing. Elizabeth L. Addington: conceptualization, writing – review and editing. Judith Tedlie Moskowitz: conceptualization, writing – review and editing. Mikaela A. Velazquez‐Sosa: investigation, data curation, writing – review and editing. Jennifer D. Rodriguez: investigation, data curation, writing – review and editing. Aigner Bobbitt: investigation, data curation, writing – review and editing. Alene Mathurin: investigation, data curation, writing – review and editing. Lesley Glenn: conceptualization, writing – review and editing. Shontè Drakeford: conceptualization, writing – review and editing. Roxana Guerra: conceptualization, writing – review and editing. Claudine Isaacs: conceptualization, funding acquisition, resources, writing – review and editing. Ami Chitalia: conceptualization, resources, writing – review and editing. Christopher Gallagher: conceptualization, resources, writing – review and editing. Deena Graham: conceptualization, resources, writing – review and editing. Nina Kadan‐Lottick: conceptualization, writing – review and editing. Suzanne C. O'Neill: conceptualization, funding acquisition, methodology, writing – review and editing. Claire C. Conley: conceptualization, funding acquisition, investigation, data curation, project administration, methodology, supervision, writing – review and editing.

Ethics Statement

This study has been reviewed and approved by the Georgetown‐MedStar Joint Oncology Institutional Review Board (Study #00005081). All participants provided written informed consent prior to participating in any study activities and for publication and sharing of de‐identified data. The study was conducted in accordance with relevant ethical standards and guidelines, including the Declaration of Helsinki and its subsequent amendments.

Conflicts of Interest

C.I. has a consulting role with AstraZeneca, Genentech, Gilead, ION, Merck, Medscape, MJH Holdings, Novartis, Pfizer, PUMA, Seagen; receives royalties from Wolters Kluwer (UptoDate) and McGraw Hill (Goodman and Gillman); and has received institutional research support from Tesaro/GSK, Seattle Genetics, Pfizer, AZ, BMS, Genentech, Novartis, Regeneron. C.G. has a consulting/advisory role with AstraZeneca, Daiichi‐Sankyo, Pfizer, Lilly, and BioTheranostics. N.K.‐L. has common stock in Astrazeneca, Biomarin, Boston Scientific, Intuitive Surgical, Eli Lilly, Eco Lab, Edwards Lifescience, Proctor & Gamble, Resmed, Thermo Fisher. S.C.O. has received grant funding from Pfizer and Gilead. C.C.C. has received grant funding from Pfizer. No other authors have conflicts to declare.

Supporting information

Table S1: Ecological momentary assessment burst items.

Table S2: Results of independent‐samples T‐tests, chi‐square tests, and Fisher's exact tests comparing sociodemographic and health characteristics between consented participants who did (N = 108) and did not (N = 17) complete ecological momentary assessments (EMAs).

Table S3: Results of unconditional no‐growth models examining within‐ and between‐person variation in quality of life, symptoms, and positive affect.

Table S4: Results of unconditional growth model examining the relationship between time since metastatic breast cancer diagnosis and quality of life.

Data Availability Statement

The data that support the findings of this study are openly available on the Open Science Framework at https://doi.org/10.17605/OSF.IO/RYZKD.