An Ecological Momentary Assessment Study Investigating Self-efficacy and Outcome Expectancy as Mediators of Affective and Physiological Responses and Exercise Among Endometrial Cancer Survivors

Abstract

Background

Previous studies have shown affective and physiological states in response to exercise as predictors of daily exercise, yet little is known about the mechanism underlying such effects.

Purpose

To examine the mediating effects of self-efficacy and outcome expectancy on the relationships between affective and physiological responses to exercise and subsequent exercise levels in endometrial cancer survivors.

Methods

Ecological momentary assessment (EMA) surveys were delivered up to eight 5- to 7-day periods over 6 months. Participants (n = 100) rated their affective and physiological states before and after each exercise session (predictors) and recorded their self-efficacy and outcome expectancy each morning (mediators). Exercise (outcome) was based on self-reported EMA surveys and accelerometer measures. A 1-1-1 multilevel mediation model was used to disaggregate the within-subject (WS) and between-subject (BS) effects.

Results

At the WS level, a more positive affective state after exercise was associated with higher self-efficacy and positive outcome expectation the next day, which in turn was associated with higher subsequent exercise levels (ps < .05). At the BS level, participants who typically had more positive affective and experienced less intense physiological sensation after exercise had higher average self-efficacy, which was associated with higher average exercise levels (ps < .05).

Conclusions

In endometrial cancer survivors, affective experience after exercise, daily self-efficacy and positive outcome expectation help explain the day-to-day differences in exercise levels within-person. Findings from this study highlight potentials for behavioral interventions that target affective experience after exercise and daily behavioral cognitions to promote physical activity in cancer survivors’ everyday lives.

In endometrial cancer survivors, positive feelings after exercise were associated with higher confidence in and more positive expectations from exercise the next day, which led to engaging in more exercise.

Introduction

Owing to improved early detection and advancement in treatment, the 5-year relative survival rate for all cancers combined has increased 20% over the past three decades [1, 2]. There are more than 15.5 million cancer survivors in the USA, and this number is expected to grow to more than 20 million by 2026 [3]. Therefore, optimizing patients’ long-term health and well-being after cancer treatment is being increasingly emphasized as part of their standard cancer care [4].

There is strong evidence that physical activity has beneficial effects on cancer survivors’ physical outcomes (e.g., increased fitness, decreased fatigue) and psychosocial outcomes (e.g., increased self-esteem, decreased depression and anxiety; refs. 5–7). The evidence for the link between physical activity and cancer is strongest for colorectal, endometrial, and breast cancer [8]. In addition, physical activity plays an important role in energy balance and obesity, which is an independent risk factor for cancer recurrence and mortality [4]. For example, dose–response associations were found between obesity and risk of endometrial cancer [9, 10]. Physical activity is also associated with a lower risk of other common comorbidities, such as cardiovascular disease and diabetes [11]. Because of these health benefits, the American Cancer Society recommends that cancer survivors engage in at least 150 min of physical activity per week [12, 13]. However, only 18% of adult male and 15% of adult female cancer survivors meet the recommended physical activity guideline [14]. Thus, there is a pressing need to develop strategies to help cancer survivors achieve and maintain an active lifestyle to improve their physical and mental health.

A comprehensive understanding of the correlates and determinants of physical activity is crucial to developing effective physical activity interventions for this population. In recent years, there has been a growing interest in understanding the roles that affective and physiological responses to exercise have in influencing physical activity in everyday life. Studies have found that positive affective states are associated with increased subsequent physical activity, ranging from within the next few hours to the next day, in free-living settings [15]. Feeling more energetic and less tired at the moment were also found to be predictors of subsequent physical activity engagement [16]. From a behavioral theory perspective, positive affective and physiological states may influence future behaviors by increasing appetitive motivation to participate in those behaviors [17], whereas negative affective and physiological states may trigger a motivational state of behavioral avoidance [18]. Nevertheless, few empirical studies have investigated how, or through what mechanisms, affective and physiological response to exercise might predict subsequent physical activity in daily life. Understanding the mechanisms related to cancer survivors’ processes of adopting and maintaining physical activity is particularly important, given the added complexity of living with a chronic disease. For example, for cancer survivors and others with chronic illnesses, physiological states or somatic sensations during and following physical activity may be perceived as aversive or threatening given possible past experiences with negative physical symptoms [19, 20]. This, in turn, may lead cancer survivors to be less willing to engage in activities that may produce somatic sensations (e.g., racing heart, muscle pain; ref. 21).

Plausible links between affective/physiological response to exercise and subsequent physical activity could be self-efficacy and outcome expectancy. Self-efficacy is defined as one’s perception about his/her ability to perform a particular behavior; individuals may be more likely to engage in physical activity if they believe that they can exercise despite potential constraints and impediments [22]. Previous studies have shown that exercise self-efficacy is an influential predictor of actual physical activity in a broad range of populations [23–26], including cancer survivors [27, 28]. Interventions that aimed to increase exercise self-efficacy in cancer survivors have demonstrated an increase in exercise intention [29] and subsequent physical activity [20]. However, most of the previous research on self-efficacy and physical activity has focused on interindividual (i.e., between-person) effects, and treated such behavioral cognitions as relatively stable constructs. In recent years, more research has investigated and found that exercise self-efficacy fluctuates within-person from day to day and that momentary self-efficacy predicts subsequent physical activity levels [30, 31]. Since exercise self-efficacy could decline over the course of an intervention [32], it is important to consider factors that might influence daily exercise self-efficacy. According to Self-Efficacy Theory, the affective and physiological experience associated with physical activity is one of the essential sources of exercise self-efficacy [22]. For example, people might interpret their breathlessness and aching muscles from exercise as indicators of limitations in their exercise ability, thus lowering their efficacy. Previous studies have shown that a more positive affective experience with exercise can result in higher exercise self-efficacy over time in older adults [32]. To date, few studies have evaluated the affective and physiological experience as a potential predictor of daily exercise self-efficacy and their influence on subsequent physical activity.

Although being relatively less studied compared to self-efficacy, outcome expectancy has received increased interests as a direct predictor of physical activity in recent years. Outcome expectancy is the expectation of positive and negative outcomes that result from performing a behavior, with positive expectation serving as motivations and negative expectation serving as disincentives for the behavior [22]. A study in older adults found that both self-efficacy and outcome expectancy were directly associated with exercise behaviors [33]. Another study in cancer patients found that outcome expectancy predicted subsequent physical activity [34]. Further, this study also found that while cancer patients, in general, had positive expectations towards physical activity, their actual experience with physical activity was slightly more negative. The Integrative Model of Behavioral Prediction describes self-efficacy and outcome expectancy as independent predictors of health-related behaviors [35], and past behavior and affect are part of the “background” variables that may influence these constructs in deciding to perform those specific behaviors in the future. Other behavioral theories, such as behavioral conditioning and expectancy-value theories, also posit that individuals would expect certain outcomes to occur as a result of a particular behavior based on experience [36, 37]. In particular, individuals’ affective and physiological states are important sources of information that have a bearing on their perceptions of the relevant costs and benefits associated with behavioral options [38]. Further, the decision to engage in a behavior also reflects individuals’ on-going, dynamic relationship between past behavior experience and outcome expectations [39]. A study by Loehr and colleagues found that negative experience (i.e., negative affect and body soreness) in a given week was associated with negative outcome expectation for the upcoming week for physical activity, while positive experience (e.g., enjoyment, positive affect, tranquility) did not impact positive outcome expectation [40]. Nevertheless, this study did not examine whether the outcome expectations influenced physical activity levels.

To address the above-mentioned research gaps, the current study used ecological momentary assessment (EMA) data to determine the extent to which daily self-efficacy and outcome expectancy mediate the relationships between affective and physiological responses to physical activity and subsequent exercise levels in a sample of endometrial cancer survivors. EMA, a real-time data capture technique that allows individuals to self-report experiences in the context of their daily lives [41], is a useful method for studying the effects of affective and cognitive factors on subsequent physical activity. In addition to providing more ecologically valid data, EMA can capture patterns of fluctuation and changes better than cross-sectional or longitudinal studies with infrequent measures. EMA data can also be used to disentangle the within-person effects from the between-person effects of time-varying correlates on physical activity to clarify whether any observed associations are driven by enduring individual differences, intraindividual variations, or both [42]. We hypothesized that at the within-person level, more positive affect and less intense physiological sensations following exercise, compared with one’s average level, would be associated with higher subsequent exercise levels through an increase in self-efficacy and outcome expectancy. At the between-person level, we hypothesized that cancer survivors who had more positive affective and less intense physiological experiences following exercise would have higher average self-efficacy, outcome expectancy, and engage in more exercise than survivors with less positive affective and more intense physiological experiences.

Methods

Procedure

The current study used EMA data from the Steps to Health (STH) study, a 6-month longitudinal study assessing predictors of exercise behavior in endometrial cancer survivors receiving an exercise intervention [43]. The STH study focused on endometrial cancer survivors because this is one of the populations that may benefit from exercise the most, given their high prevalence of overweight/obese weight status and physical inactivity. The reporting of this study followed guidelines from the Adapted STROBE Checklist for Reporting EMA Studies [44].

The STH study included 100 women who had been diagnosed with stage I, II, or IIIa endometrial cancer. Participants had to have completed treatment at least 6 months previously and have no evidence of disease. Women were excluded from the study if they met and had maintained physical activity recommendations [45] for 6 months or longer.

Intervention and assessment time points are outlined in Fig. 1. The first EMA assessment consisted of 7 days of data collection; the rest of the EMA assessments consisted of 5 days of data collection. Participants also wore a GT1M accelerometer (ActiGraph LLC, Pensacola, FL) at the waist during waking hours throughout these monitoring periods.

Fig. 1.

Assessment and intervention schedule for the Steps to Health study. Home-based assessment included the ecological momentary assessment and the accelerometers monitoring.

EMA was delivered and completed via a handheld computer (iPAQ RX1950, Hewlett-Packard, Palo Alto, CA). The handheld computer prompted one morning assessment and one evening assessment each day during the monitoring periods. Participants were also instructed to initiate and complete a brief survey before and after each exercise session during the day. The recommended exercise and primary focus of the STH intervention was walking. Each participant received an exercise/walking prescription tailored to her fitness level assessed at the baseline (i.e., individualized exercise duration, intensity, and heart rate target). All participants were encouraged to exercise one to three times a day for three to five times a week. The ultimate goal for all participants was to achieve moderate-intensity exercise for at least 30 min a day on 5 or more days per week.

Measures

Self-efficacy

Self-efficacy was assessed in the morning using a one-item measure that asked participants to rate their confidence in meeting the recommended physical activity levels for that day. Participants rated their confidence on a five-point Likert scale, whose responses ranged from “Not at all confident” to “Extremely confident.” The intraclass correlation coefficient (ICC) for the self-efficacy item was 0.27.

Positive and negative outcome expectations

In the morning, participants were additionally asked a series of questions about their expected outcomes of exercising for that day. There were seven positive items (e.g., “I will sleep more soundly tonight if I exercise today”) and three negative items (e.g., “Exercising today will be painful”). These items were developed from the decisional balance measure for exercise [46] and revised to address the positive and negative outcomes of exercise that were identified in interviews with breast cancer survivors from previous pilot studies. Participants rated their expectations on a five-point Likert scale, whose responses ranged from “Not at all likely” to “Extremely likely.” The internal consistency reliabilities of the positive and negative scales at the between-person level were 0.93 and 0.62, respectively. The ICC for positive outcome expectation was 0.71, and for negative outcome expectation was 0.57.

Affective states

The 12-item Exercise-induced Feeling Inventory (EFI; ref. 47) was used to assess participants’ affective states. The EFI captures four distinct feeling states: revitalization (i.e., energetic, refreshed, revived), tranquility (i.e., calm, relaxed, peaceful), positive engagement (i.e., enthusiastic, up-beat, happy), and physical exhaustion (i.e., fatigued, tired, worn-out). Participants rated the affective state items before and after each exercise session on a five-point Likert scale ranging from “Do not feel” to “Feel very strongly.” The internal consistency reliability at the between-person level was 0.947. The interitem correlation at the between-person level between physical exhaustion (reversed coded) and positive engagement, revitalization, tranquility was 0.743, 0.826, and 0.642, respectively. The ICC for the EFI scale was 0.67.

Physiological states

The 10-item Pennebaker Inventory of Limbic Languidness (PILL; ref. 48) was used to assess participants’ physiological states. Participants were asked to rate their experience of several physiological sensations, including “racing heart,” “tightness in chest,” and “stiff or sore muscles” before and after each exercise session. The questions were rated on a five-point Likert scale, with responses ranging from “Not at all” to “Very much.” The internal consistency reliability at the between-person level was 0.855. The ICC for the PILL scale was 0.70.

Daily exercise minutes

The focus of the STH intervention was to promote planned/intentional exercise carried out by cancer survivors. Therefore, EMA surveys were designed to capture exercise sessions [1] directly after each exercise session (i.e., real-time minutes), and [2] each evening (i.e., nighttime diary). The use of EMA to assess exercise in real-time, or near real-time (i.e., the nighttime diary), reduces the risk of recall bias that is associated with self-reported measures and is comparable to device-measured activity levels [49]. To minimize missing data, if the EMA reports were missing, then the accelerometer data were used. Accelerometer data were processed using the ActiLife software to obtain moderate-to-vigorous activity performed in bouts of at least 10 min, following the cut-points for uni-axial accelerometers by Freedson [50]. Additional details about computing the daily exercise minutes for the STH study have been published elsewhere [43]. Overall, the intercorrelations among the three methods (real-time EMA, nighttime diary, accelerometry) were similar to those of other studies comparing different types of physical activity measures [51].

Baseline physical activity

For the assessment of baseline physical activity, participants filled out the Community Health Activities Model Program for Seniors (CHAMPS) questionnaire [52] during their first laboratory visit. The CHAMPS measures the weekly frequency and duration of physical activities commonly performed by older adults.

Weight category

Participants’ weight categories were classified based on body mass index (BMI) as normal weight (<25 kg/m²), overweight (25–30 kg/m²), or obese (≥30 kg/m²).

Statistical Analysis

Multilevel structural equation modeling (MSEM; ref. 53) was used for all analyses. Between-subject (BS) and within-subject (WS) effects were separated (i.e., partitioning the variance). The BS effect represents the individual mean deviation from the grand mean (i.e., grand mean centering), and the WS effect represents the deviation from one’s own mean at any given day (i.e., within-person mean centering; ref. 54). Since no participants reported more than one exercise session for any given measurement day, all variables in the models were at the day level. The outcome variable, daily exercise minutes, was not normally distributed and many observations had zero minutes. To address this issue, a two-piece modeling approach was used as in previous studies [16]. The Piece 1 model was a multilevel logistic regression model predicting the probability of engaging in some exercise (i.e., non-zero exercise minutes) versus no exercise (i.e., zero exercise minutes). The Piece 2 model was a multilevel linear regression model predicting the log-transformed non-zero exercise minutes. All models were fitted using Mplus (version 7.11).

Building on the two-piece model, multilevel mediation modelling was used to test whether affective and physiological states after exercise (i.e., the predictor at day t) were associated with self-efficacy and outcome expectancy the next morning (i.e., the mediator at day t+1) and the overall exercise level of that day (i.e., the outcome at day t+1). Since the antecedent, mediator, and outcome were all measured at level 1 (i.e., day-level), and days were nested in level 2 (i.e., each participant), the 1-1-1 mediation model was used. The positive and negative outcome expectations were evaluated in the same model. Since all participants were in the same intervention study and the primary interest of the current study was to explore the daily relationship among affective/physiological response to exercise, self-efficacy/outcome expectancy, and exercise in cancer survivors, the fixed-slopes modeling was used.

Fig. 2 illustrates an example of the multilevel mediation model used in the current study. The BS and WS effects for the 1-1-1 mediation models were separately tested. This way, we were able to examine (i) whether participants’ average self-efficacy and outcome expectancy (e.g., SE/OEi, which represents SE/OEBS, the person-level mean at level 2) mediated the effect of their average affective and physiological responses to exercise (i.e., AFFECTi, which represents AFFECTBS, the person-level mean at level 2) on daily exercise levels at the between-person level (i.e., the BS mediation effect); and (ii) whether participants’ relative self-efficacy and outcome expectancy each morning (i.e., SE/OE(t+1)i- SE/OEi, denoted as SE/OEWS) mediated the relationship between their relative affective and physiological responses to exercise (i.e., AFFECTti-AFFECTi, denoted as AFFECTWS) on subsequent exercise levels (i.e., the WS mediation effect). Mean-centering the antecedent and mediator within each participant (i.e., AFFECTWS and SE/OEWS) and using the person-level means on the antecedent (i.e., AFFECTBS) and mediator (i.e., SE/OEBS) in the level-2 model of outcome can “deconflate” the level-1 effects with their level-2 component [55].

Fig. 2.

Diagram for a sample multilevel mediation model that disaggregates the between-person and within-person effects.

The following equations demonstrate the three steps for a generic “deconflated” 1-1-1 mediation model:

Step 1

$${\displaystyle \begin{array}{l}{\displaystyle \mathrm{L}\mathrm{e}\mathrm{v}\mathrm{e}\mathrm{l}1P{A}_{(t+1)i}={\beta }_{0i}^{(1)}+{\beta }_{1i}^{(1)}AFFEC{T}_{WS}+r_{ti}^{(1)}}\end{array}}$$

$${\displaystyle \begin{array}{l}{\displaystyle \mathrm{L}\mathrm{e}\mathrm{v}\mathrm{e}\mathrm{l}2{\beta }_{0i}^{(1)}={\gamma }_{00}^{(1)}+{\gamma }_{01}^{(1)}AFFEC{T}_{BS}+u_{0i}^{(1)}}\end{array}}$$

$${\displaystyle \begin{array}{l}{\displaystyle {\beta }_{1i}^{(1)}={\gamma }_{10}^{(1)}}\end{array}}$$

Step 2

$${\displaystyle \begin{array}{l}{\displaystyle \mathrm{L}\mathrm{e}\mathrm{v}\mathrm{e}\mathrm{l}1SE\varnothing E_{(t+1)i}={\beta }_{0i}^{(2)}+{\beta }_{1i}^{(2)}AFFEC{T}_{WS}+r_{ti}^{(2)}}\end{array}}$$

$${\displaystyle \begin{array}{l}{\displaystyle \mathrm{L}\mathrm{e}\mathrm{v}\mathrm{e}\mathrm{l}2{\beta }_{0i}^{(2)}={\gamma }_{00}^{(2)}+{\gamma }_{01}^{(2)}AFFEC{T}_{BS}+u_{0i}^{(2)}}\end{array}}$$

$${\displaystyle \begin{array}{l}{\displaystyle {\beta }_{1i}^{(2)}={\gamma }_{10}^{(2)}}\end{array}}$$

Step 3

$${\displaystyle \begin{array}{l}{\displaystyle \mathrm{L}\mathrm{e}\mathrm{v}\mathrm{e}\mathrm{l}1P{A}_{(t+1)i}={\beta }_{0i}^{(3)}+{\beta }_{1i}^{(3)}AFFEC{T}_{WS}+{\beta }_{2i}^{(3)}SE\varnothing E_{WS}+r_{ti}^{(3)}}\end{array}}$$

$${\displaystyle \begin{array}{l}{\displaystyle \mathrm{L}\mathrm{e}\mathrm{v}\mathrm{e}\mathrm{l}2{\beta }_{0i}^{(3)}={\gamma }_{00}^{(3)}+{\gamma }_{01}^{(3)}AFFEC{T}_{BS}+{\mathrm{\gamma }}_{02}^{(3)}SE\varnothing E_{BS}+u_{0i}^{(3)}}\end{array}}$$

$${\displaystyle \begin{array}{l}{\displaystyle {\beta }_{1i}^{(3)}={\gamma }_{10}^{(3)}}\end{array}}$$

$${\displaystyle \begin{array}{l}{\displaystyle {\beta }_{2i}^{(3)}={\gamma }_{20}^{(3)}}\end{array}}$$

Therefore, the level 1 mediation effect can be quantified as:

$${\displaystyle \begin{array}{l}{\displaystyle {\gamma }_{10}^{(2)}\times {\gamma }_{20}^{(3)}}\end{array}}$$

and the level 2 mediation effect can be quantified as:

$${\displaystyle \begin{array}{l}{\displaystyle {\gamma }_{01}^{(2)}\times {\gamma }_{02}^{(3)}}\end{array}}$$

All models controlled for the respective affective/physiological state prior to exercise to examine the effect of affective/physiological response to exercise. In addition, all models controlled for participants’ BMI category since the exercise pattern over time was found to be different between BMI categories in the STH study [56]; for wave of assessment since observations across waves were included in the same model; and for other exercise-related potential cofounders (i.e., baseline physical activity levels and exercise duration).

Results

Descriptive Statistics

Of the 100 endometrial cancer survivors who enrolled in the STH study, one did not complete any EMA survey, and two did not report any exercise sessions during the monitoring periods. Therefore, there was a total of 97 participants in the analytical sample. Table 1 shows the characteristics of these participants. The average EMA compliance (calculated for pre- and postlaboratory assessment at baseline, 2 months, 4 months, and 6 months) for the morning assessment was 67.3%, 73.2%, 68.0%, 68.6%, 71.6%, 74.2%, 78.4%, and 75.8% at each assessment point. The average EMA compliance for the evening assessment was 59.7%, 61.0%, 67.8%, 65.2%, 65.0%, 60.0%, 77.2%, and 72.4% at each assessment point. Further, 81.27% of the times when the accelerometer detected some moderate-to-vigorous physical activity during the day, participants had completed the pre-/postexercise EMA for that day. On average, participants reported exercising 16 days (standard deviation [SD] = 9.1) across the 42 monitoring days. Each participant, on average, had 11 exercise days (SD = 6.8) that were matched with the next-day morning self-efficacy and outcome expectancy assessment. There were no missing data for exercise minutes. The person-level average daily exercise minutes was 18.0 (SD = 9.4). This yielded a total of 1,068 matched days across all participants.

Table 1.

Participant Characteristics (n = 97)

Characteristic	No. of participants (%)a
Mean age (SD), years	57.2 (11.2)
Weight category
Normal weight	15 (15.5)
Overweight	21 (21.6)
Obese	61 (62.9)
Education level
High school or less	15 (15.5)
Some college/technical training	41 (42.3)
Bachelor’s degree	24 (24.7)
Master’s degree and above	17 (17.5)
Marital status
Single	12 (12.4)
Married/living with significant other	67 (69.1)
Divorced/separated/widowed	18 (18.6)
Employment status
Employed (full-time)	43 (44.3)
Employed (part-time)	7 (7.2)
Not employed	47 (48.5)
Disease stage
I	78 (80.4)
II	15 (15.5)
IIIa	4 (4.1)
Treatment
Surgery only	56 (57.7)
Surgery + radiotherapy	41 (42.3)
Mean time since diagnosis (SD), years	2.2 (1.3)

SD standard deviation.

^aAll data are no. of patients (%) unless otherwise indicated.

The person-level average affect (i.e., EFI score) reported after exercise was 29.5 (SD = 8.2), and the person-level average physiological sensations (i.e., PILL score) was 6.0 (SD = 4.3). The person-level average self-efficacy, positive outcome expectation, and negative outcome expectation were 3.3 (SD = 0.7), 3.7 (SD = 0.7), and 1.8 (SD = 0.5), respectively. The ICC ranged from 0.27 (self-efficacy) to 0.71 (positive outcome expectation) for these variables, indicating a reasonable amount of within- and between-person variability in the data to be considered in multilevel models [57]. The within- and between-person correlations of all study variables are presented in the Supplementary Table 1.

Affective/Physiological Response to Exercise and Exercise Levels

The results of the two-piece multilevel model controlling for the pre-exercise affective state showed that at the WS level, affective state after exercise was not significantly associated with the probability of engaging in some exercise versus no exercise the following day. It was also not significantly associated with exercise duration for participants who did engage in some exercise the following day. At the BS level, average affective state was not significantly associated with the overall frequency of engaging in some exercise. Nevertheless, having a more positive affective state after exercise compared to others in the study was associated with more average exercise minutes (piece 2 model, BS β = .006, standard error [SE] = .002, p = .014). There was no significant association between physiological response to exercise and exercise levels at the WS and BS levels.

Affective Response to Exercise, Self-efficacy, and Exercise Levels

Table 2 shows the results of the multilevel mediation model using the affective state after exercise as antecedent, the next morning self-efficacy as the mediator, and the next-day exercise levels as the outcome. At the WS level, having a more positive affective state following exercise than one’s average was associated with higher self-efficacy the next morning (WS β = 0.025, SE = 0.006, p < .001). Higher self-efficacy in the morning was associated with higher probability of engaging in some versus no exercise that day (piece 1 model; WS β = 0.966, SE = 0.154, p < .001), as well as longer exercise duration for participants who engaged in some exercise that day (piece 2 model; WS β = 0.065, SE = 0.014, p < .001). For both piece 1 and piece 2 models, the direct effect was not statistically significant, but the indirect effect was (piece 1 model: WS β = 0.024, SE = 0.007, p < .001; piece 2 model: WS β = 0.002, SE = 0.001, p < .001), suggesting that although affective response to exercise did not predict exercise levels the following day, it might indirectly impact exercise levels through its positive influence on subsequent self-efficacy.

Table 2.

Associations Between Affective Response to Exercise and Next-day Exercise Levels, Mediated by Self-efficacy

		a Path (affect on self-efficacy)		b Path (self-efficacy on exercise)		Direct effect (c’ path)		Indirect effect
		Beta estimate (SE)	p	Beta estimate (SE)	p	Beta estimate (SE)	p	Beta estimate (SE)	p
WS Effect	Piece 1 Model (Some vs. zero exercise minutes)	0.025 (0.006)	<.001	0.966 (0.154)	<.001	−0.022 (0.020)	.257	0.024 (0.007)	.001
	Piece 2 Model (exercise minutes)a			0.065 (0.014)	<.001	0.003 (0.002)	.295	0.002 (0.001)	.001
BS Effect	Piece 1 Model (Some vs. zero exercise minutes)	0.037 (0.008)	<.001	0.474 (0.191)	.013	-0.014 (0.017)	.416	0.017 (0.008)	.027
	Piece 2 Model (exercise minutes)a			0.064 (0.022)	.004	0.006 (0.003)	.035	0.002 (0.001)	.015

BS between-subject; SE standard error; WS within-subject.

All models controlled for the pre-exercise affective state, exercise duration, participant’s weight category, baseline activity, and wave of the assessment. Bold values indicate a p value < .05.

^aExercise minutes were log-transformed.

Similar effects were found at the BS level. Participants who on average experienced a more positive affective state after exercise compared to others in the study tended to have higher self-efficacy (BS β = .037, SE = .008, p < .001). Higher average self-efficacy was associated with more frequent exercise (piece 1 model; BS β = 0.474, SE = 0.191, p = .013), as well as longer exercise duration (piece 2 model; β = 0.064, SE = 0.022, p = .004). For the piece 1 model, the direct effect was not statistically significant but the indirect effect was (BS β = 0.017, SE = 0.008, p = .027). This suggests that the average affective response to exercise did not directly influence the frequency of exercise engagement, but it may indirectly impact the frequency of exercise engagement through its positive influence on average self-efficacy. For piece 2 model, both direct and indirect effects were significant (BS β = 0.006, SE = 0.003, p = .035; BS β = 0.002, SE = 0.001, p = .015; respectively), suggesting that, compared with those who in general had less positive affective experience after exercise, endometrial cancer survivors who typically experienced more positive affect after exercise also exercised more, partly attributable to their higher self-efficacy.

Physiological Response to Exercise, Self-efficacy, and Exercise Levels

Table 3 shows the results of the multilevel mediation model using the affective state after exercise as antecedent, the next morning self-efficacy as the mediator, and the next-day exercise levels as the outcome. At the WS level, the physiological state after exercise was not associated with self-efficacy the next morning and did not have a direct or indirect effect on exercise the next day, suggesting that day-to-day variations in physiological response to exercise had no impact on subsequent exercise levels either directly or indirectly through self-efficacy in this sample of endometrial cancer survivors. At the BS level, cancer survivors who on average experienced a more intense physiological state after exercise tended to have lower self-efficacy (β = −0.048, SE = .016, p = .003), and lower average self-efficacy was associated with lower frequency of exercise overall (piece 1 model; BS β = 0.104, SE = 0.035, p = .003), as well as fewer exercise minutes on average (piece 2 model; β = 0.054, SE = 0.028, p = .050). The BS direct effects for both models were not significant, and the indirect effect was only significant for piece 1 model (BS β = −0.005, SE = 0.002, p = .035), suggesting that the typical physiological sensations that cancer survivors experienced might indirectly impact the frequency of exercise engagement through its influence on average self-efficacy.

Table 3.

Associations Between Physiological Response to Exercise and Next-day Exercise Levels, Mediated by Self-efficacy

		a Path (physiological sensation on self-efficacy)		b Path (self-efficacy on exercise)		Direct effect (c’ path)		Indirect effect
		Beta estimate (SE)	p	Beta estimate (SE)	p	Beta estimate (SE)	p	Beta estimate (SE)	p
WS Effect	Piece 1 Model (Some vs. zero exercise minutes)	−0.003 (0.013)	.845	0.168 (0.022)	<.001	0.007 (0.005)	.158	0.000 (0.002)	.845
	Piece 2 Model (exercise minutes)a			0.060 (0.014)	<.001	−0.004 (0.003)	.148	0.000 (0.001)	.843
BS Effect	Piece 1 Model (Some vs. zero exercise minutes)	−0.048 (0.016)	.003	0.104 (0.035)	.003	0.006 (0.005)	.232	−0.005 (0.002)	.035
	Piece 2 Model (exercise minutes)a			0.054 (0.028)	.050	−0.001 (0.004)	.887	−0.003 (0.002)	.110

All models controlled for the pre-exercise physiological state, exercise duration, participant’s weight category, baseline activity, and wave of the assessment. Bold values indicate a p value < .05.

BS between-subject; SE standard error; WS within-subject.

^aExercise minutes were log-transformed.

Affective Response to Exercise, Outcome Expectations, and Exercise Levels

Table 4 shows the results of the multilevel mediation model using affective state after exercise as the antecedent, next morning positive and negative outcome expectations as the mediators, and next-day exercise levels as the outcome. At the WS level, having a more positive affective state following exercise than one’s average level was associated with a higher positive and lower negative outcome expectations the next morning (WS β = 0.026, SE = 0.004, p < .001; WS β = −0.017, SE = 0.004, p < .001, respectively). No other paths were significant at the WS level for the piece 1 model, suggesting that affective response to exercise did not predict the likelihood of exercise engagement the following day; and outcome expectancy in the morning was not associated with the likelihood of exercise engagement of that day. For the piece 2 model, only positive outcome expectation was positively associated with exercise minutes of that day (WS β = 0.051, SE = 0.025, p = .040), and the indirect effect was significant (WS β = 0.001, SE = 0.001, p = .047). Taken together, these results suggest that, controlling for the effects of negative outcome expectation, affective response to exercise did not directly influence exercise minutes the following day, but it might indirectly impact exercise minutes through its positive influence on subsequent positive outcome expectation.

Table 4.

Associations Between Affective Response to Exercise and Next-day Exercise Levels, Mediated by Outcome Expectancy

		a Path (affect on outcome expectations)		b Path (outcome expectations on exercise)		Direct effect (c’ path)		Indirect effect
		Beta estimate (SE)	p	Beta estimate (SE)	p	Beta estimate (SE)	p	Beta estimate (SE)	p
WS Effect (Positive outcome expectation)	Piece 1 Model (Some vs. zero exercise minutes)	0.026 (0.004)	<.001	0.068 (0.049)	.167	−0.002 (0.003)	.527	0.002 (0.001)	.178
	Piece 2 Model (exercise minutes)a			0.051 (0.025)	.040	0.002 (0.002)	.498	0.001 (0.001)	.047
WS Effect (Negative outcome expectation)	Piece 1 Model (Some vs. zero exercise minutes)	−0.017 (0.004)	<.001	−0.003 (0.043)	.942	−0.002 (0.003)	.527	0.000 (0.001)	.942
	Piece 2 Model (exercise minutes)a			0.019 (0.027)	.485	0.002 (0.002)	.498	0.000 (0.000)	.485
BS Effect (Positive outcome expectation)	Piece 1 Model (Some vs. zero exercise minutes)	0.052 (0.007)	<.001	−0.008 (0.045)	.859	0.002 (0.004)	.592	0.000 (0.002)	.859
	Piece 2 Model (exercise minutes)a			0.045 (0.033)	.181	0.001 (0.003)	.740	0.002 (0.002)	.194
BS Effect (Negative outcome expectation)	Piece 1 Model (Some vs. zero exercise minutes)	−0.043 (0.006)	<.001	0.021 (0.045)	.633	0.002 (0.004)	.592	-0.001 (0.002)	.637
	Piece 2 Model (exercise minutes)a			−0.058 (0.037)	.116	0.001 (0.003)	.740	0.002 (0.002)	.113

All models controlled for the pre-exercise affective state, exercise duration, participant’s weight category, baseline activity, and wave of the assessment. Bold values indicate a p value < .05.

BS between-subject; SE standard error; WS within-subject.

^aExercise minutes were log-transformed.

At the BS level, cancer survivors who on average experienced a more positive affective state after exercise tended to have higher than average positive and lower than average negative outcome expectations (BS β = 0.052, SE = 0.007, p < .001; BS β = −0.043, SE = 0.006, p < .001, respectively). No other paths were significant at the BS level for both piece 1 and piece 2 models, suggesting that although average affective response to exercise was associated with average outcome expectancy among cancer survivors, it was not associated with overall exercise levels; and average outcome expectancy were also not associated with overall exercise levels.

Physiological Response to Exercise, Outcome Expectations, and Exercise Levels

Table 5 shows the results of the multilevel mediation model using physiological state after exercise as the antecedent, next morning positive and negative outcome expectations as the mediators, and next-day exercise levels as the outcome. At the WS level, having a more intense physiological state following exercise than one’s average level was associated with a lower positive and higher negative outcome expectations the next morning (WS β = −0.015, SE = 0.007, p = .019; WS β = 0.036, SE = 0.008, p < .001, respectively). No other paths were significant at the WS level for the piece 1 model, suggesting that physiological response to exercise did not predict the likelihood of exercise engagement the following day; and outcome expectancy in the morning was not associated with the likelihood of exercise engagement of that day. For the piece 2 model, only positive outcome expectation was positively associated with exercise minutes of that day (WS β = 0.049, SE = 0.025, p = .049). These results suggest that, although physiological response to exercise predicted subsequent positive and negative outcome expectations, it did not predict subsequent exercise levels either directly or indirectly through outcome expectancy.

Table 5.

Associations Between Physiological Response to Exercise and Next-day Exercise Levels, Mediated by Outcome Expectancy

		a Path (physiological sensation on outcome expectations)		b Path (outcome expectations on exercise)		Direct effect (c’ path)		Indirect effect
		Beta estimate (SE)	p	Beta estimate (SE)	p	Beta estimate (SE)	p	Beta estimate (SE)	p
WS Effect (Positive outcome expectation)	Piece 1 Model (Some vs. zero exercise minutes)	−0.015 (0.007)	.019	0.083 (0.047)	.075	0.008 (0.005)	.110	−0.001 (0.001)	.162
	Piece 2 Model (exercise minutes)a			0.049 (0.025)	.049	-0.005 (0.003)	.124	−0.001 (0.001)	.134
WS Effect (Negative outcome expectation)	Piece 1 Model (Some vs. zero exercise minutes)	0.036 (0.008)	<.001	−0.014 (0.044)	.746	0.008 (0.005)	.110	−0.001 (0.002)	.741
	Piece 2 Model (exercise minutes)a			0.026 (0.025)	.307	-0.005 (0.003)	.124	0.001 (0.001)	.339
BS Effect (Positive outcome expectation)	Piece 1 Model (Some vs. zero exercise minutes)	−0.065 (0.016)	<.001	0.015 (0.040)	.704	0.006 (0.006)	.331	−0.001 (0.003)	.707
	Piece 2 Model (exercise minutes)a			0.043 (0.028)	.123	0.003 (0.004)	.464	−0.003 (0.002)	.162
BS Effect (Negative outcome expectation)	Piece 1 Model (Some vs. zero exercise minutes)	0.066 (0.012)	<.001	−0.027 (0.046)	.558	0.006 (0.006)	.331	−0.002 (0.003)	.558
	Piece 2 Model (exercise minutes)a			−0.064 (0.034)	.061	0.003 (0.004)	.464	−0.004 (0.002)	.069

All models controlled for the pre-exercise physiological state, exercise duration, participant’s weight category, baseline activity, and wave of the assessment. Bold values indicate a p value < .05.

BS between-subject; SE standard error; WS within-subject.

^aExercise minutes were log-transformed.

At the BS level, cancer survivors who on average experienced a more intense physiological state after exercise tended to have lower than average positive and higher than average negative outcome expectations (BS β = −0.065, SE = 0.016, p < .001; BS β = .066, SE = 0.012, p < .001; respectively). No other paths were significant at the BS level for the piece 1 model, suggesting that although average physiological response to exercise was associated with average outcome expectancy among cancer survivors, it was not associated with the overall frequency of exercise engagement. For the piece 2 model, there was a trend for a negative association between negative outcome expectation and average exercise minutes (BS β = −0.064, SE = 0.034, p = .061), as well as for the indirect effect (BS β = −0.004, SE = 0.002, p = .069). These results suggest that cancer survivors’ average physiological response to exercise might have the potential to indirectly impact their overall exercise minutes through the association between physiological response to exercise and negative outcome expectation.

Discussion

To our knowledge, the current study was the first to examine the effects of self-efficacy and outcome expectancy as mediators of the relationship between affective and physiological responses to exercise and subsequent exercise levels in a sample of endometrial cancer survivors. Few studies have used EMA to assess affective and physiological responses to exercise (i.e., having both pre- and postexercise EMA entries) in free-living settings. The current study did not find evidence for direct associations between affective and physiological responses to exercise and subsequent exercise levels at the WS level. However, at the BS level, higher average affective state was associated with more average exercise minutes; this is consistent with previous studies in healthy populations [58, 59].

The results of the present study’s mediation analyses provide some insight into the way affective and physiological responses to exercise might affect other psychosocial factors that are closely related to cancer survivors’ daily exercise levels. SCT, self-efficacy theory, and behavioral conditioning and expectancy-value theories highlight the importance of affective and physiological experience in predicting self-efficacy and outcome expectancy [35–38, 60]. As hypothesized, affective response to exercise was positively associated with self-efficacy and positive outcome expectation, and negatively associated with negative outcome expectation at both WS and BS levels. For physiological response to exercise, as expected, significant negative associations were found for positive outcome expectation and significant positive associations were found for negative outcome expectation at both WS and BS levels. Physiological response to exercise was found to be negatively associated with self-efficacy at the BS level, but not at the WS level. These results imply that, when self-evaluating their ability to perform exercise for a given day, cancer survivors might be influenced by their previous affective experience from performing exercise more so than their physiological experience. On the other hand, compared to self-efficacy, the physiological experience might be more closely related to cancer survivor’s daily outcome expectancy. It is possible that cancer survivors perceive the physiological sensations from exercise (e.g., racing heart, sweating) as part of the “default” consequences from exercise that do not necessarily impact their ability to perform future exercise.

Consistent with previous studies, results from this study found that higher self-efficacy predicted daily exercise levels at both WS and BS levels. Further, through examining the mediation models, results from the current study provide some potential strategies for interventions that target to increase self-efficacy as a way to promote daily exercise levels among cancer survivors. For example, this study found that at the WS level, prior affective but not physiological experience predicted subsequent self-efficacy, which, in turn, may have influenced daily exercise levels. Interventions could incorporate messaging to normalize the aversive affective feelings that arise after exercise so that individuals do not interpret them as unpleasant. Further, this messaging would be particularly useful in a just-in-time intervention [61]. If real-time assessment detects that a cancer survivor just had an unpleasant experience after an exercise session, messaging could be used to prevent a decrease in self-efficacy, which might negatively impact physical activity levels in subsequent days. At the BS level, we found that the association between affective response and exercise minutes was partly explained by cancer survivor’s average level of self-efficacy; and that cancer survivors’ typical physiological response to exercise might indirectly impact daily exercise levels through average self-efficacy. This result helps to explain the interindividual variability in daily exercise levels among cancer survivors. It suggests that exercise interventions could identify cancer survivors who tend to negatively interpret unpleasant feelings (e.g., tired, fatigued) and intense physiological sensations after exercise (e.g., sweating, sore muscles, dizziness) as inability to perform exercise, and help them better manage or cope with these negative internal experiences. For example, in addition to personalizing the exercise intensity to the appropriate level for each individual, interventions could also aim to promote value-based behavior (e.g., exercise) even in the face of aversive internal events (e.g., negative sensations) through mindfulness and acceptance-based contextual cognitive-behavioral therapies [62].

Contrary to our hypotheses, for outcome expectancy, only positive outcome expectation was positively associated with exercise minutes for cancer survivors who engaged in some exercise that day. The null findings for negative outcome expectation suggest that, for cancer survivors, the negative aspects associated with exercise might not influence their daily decision in being active or not active. The positive aspects associated with exercise might have more weight in terms of exercise engagement for cancer survivors. It is possible that, for endometrial cancer survivors, the benefits of exercise (e.g., better sleep, less stress) matter more than the potential negative consequences of exercise (e.g., pain, fatigue). It is also possible that there are other aspects of negative consequences or barriers to daily exercise that were not assessed in this study (e.g., time commitment). Although this study identified positive outcome expectation as a potential predictor for daily exercise levels, we did not observe evidence to suggest that affective or physiological response to exercise impact daily exercise level through their influence in positive outcome expectation. Future studies could explore other factors that might affect positive outcome expectation based on behavioral theories among cancer survivors (e.g., perceived risk, attitudes towards targets; ref. 35).

It is also worth noting that in this study, the ICC for self-efficacy (0.27) was lower than the other measures (outcome expectancy and affective and physiological response; ICCs ≥ 0.50), suggesting that self-efficacy fluctuated from day to day within individuals more so than outcome expectancy and affective and physiological response to exercise. This difference in intra-individual variability might partly explain the significant indirect effect we found for self-efficacy but not for outcome expectancy. Previous studies in the general population have found that exercise-related self-efficacy, but not outcome expectancy, varied substantially within-individuals across physical contexts (e.g., indoors vs. outdoors; at home vs. at work; ref. 31) and time of the week (i.e., weekday vs. weekend days; ref. 63). In the current study, we only measured self-efficacy and outcome expectancy in the morning. Future studies with cancer survivors could consider assessing self-efficacy and outcome expectancy at different times throughout the day to better understand the nature of variations in those behavior cognitions.

Despite the novelty of the current study, it has several limitations. First, the EMA protocol did not enforce data entry right after an exercise session. Therefore, the affective and physiological states captured after exercise may not have always reflected the immediate responses following exercise. Nevertheless, based on participants’ pre-exercise entry and their reported exercise duration, it is estimated that 61.7% of the postexercise EMA entries were completed within one hour after exercise, and 80.3% were completed within two hours after exercise. Second, because walking was the primary form of physical activity targeted in the STH study, the current study may not have captured potentially different affective and physiological states in responses to various types of exercise. This limit in the type of physical activity could be one of the reasons for low within-person variability in affective and physiological response to exercise observed in this study (ICCs = 0.7). Third, because this study included only endometrial cancer survivors, its findings might not be generalizable to other populations (e.g., male cancer survivors). For example, previous research showed gender differences in exercise patterns (e.g., exercise duration, exercise location; refs. 64, 65) and reasons for exercise (e.g., women for weight loss and men for enjoyment; ref. 66). Future studies could investigate whether the associations between affective/physiological response to exercise, self-efficacy and outcome expectancy, and daily exercise levels might be different in male cancer survivors. Lastly, the STH study used a composite measure to represent daily exercise minutes based on EMA surveys and accelerometer. Although EMA reduces the risk of recall bias, it might still have other self-report-related biases (e.g., accuracy, social desirability bias). We repeated all analyses using the accelerometer data only (see Supplementary Tables 2–5). Contrary to the composite exercise measure, there was a direct association between self-reported physiological response to exercise and total moderate-to-vigorous physical activity (MVPA) minutes the next day (piece 2 model, WS β = −0.011, SE = 0.004, p = .009). Further, in the mediation model, self-efficacy was associated with the likelihood of engaging in some vs. no MVPA minutes (piece 1 model, WS ps ≤ .01), but there was no association between outcome expectancy and MVPA minutes. The differences in findings could be due to the potentially different activities that EMA versus accelerometer captured. It might be that consciously planned, intentional exercise is perceived differently by cancer survivors, and has distinct psychosocial implications, than episodes of physical exertion that meet the objective thresholds defining MVPA. Furthermore, the standard accelerometer cut-points for MVPA might underestimate the amount of exercise in this clinical population, as they have been shown to in older adults [67]. Future physical activity studies in cancer survivors should take into considerations of the different assessment methods of daily physical activity and what types of activities are being measured [68]. More research is needed to further investigate how planned, intentional exercise compares to device-measured physical activity regarding its associations with behavior cognitions and affective, physiological states.

Overall, the current study showed that, among endometrial cancer survivors, affective response to exercise predicted their subsequent self-efficacy and positive outcome expectation, which predicted their daily exercise levels. The between-person differences in typical affective and physiological responses to exercise might help to explain the differences in levels of self-efficacy, which were associated with average daily exercise levels in this sample of cancer survivors. This study also demonstrated that EMA could be a useful tool to delineate the temporal sequence of events and therefore allow the testing of causal models in real-life settings at both the WS and BS levels. At the WS level, when cancer survivors’ self-efficacy and positive outcome expectation are low, interventions may explore encouraging more enjoyable exercise sessions (e.g., a just-in-time intervention message saying “Why not turn on the music and dance to your favorite song for 10 min?”). At the BS level, interventions could consider screening for individuals who tend to have negative physiological responses to exercise at the beginning of the intervention and help them overcome this barrier with individualized exercise plans and extra support (e.g., acceptance and value-based behavioral therapies). Future studies could also consider using EMA to further explore other time-varying determinants (e.g., social and physical contexts) of cancer survivors’ daily health-related behaviors (e.g., eating behaviors). A better understanding of the challenges and opportunities to intervene in cancer survivors’ everyday lives may inform public health initiative to design more effective behavior change interventions that aim to promote healthy lifestyles in cancer survivors.

Compliance with Ethical Standards

Conflict of Interests The authors declare no conflict of interest.

Authors’ Contributions Y.L. and K.B.E. conceived the research questions. J.S. and M.C.R. cleaned the dataset. Y.L. and J.S. analyzed the data. E.C.M. contributed to the discussion of theoretical background and implications for clinical practice. Y.L. drafted the manuscript. All authors reviewed and edited the manuscript and approved the final version of the manuscript.

Authors’ Statement of Conflict of Interest and Adherence to Ethical Standards The authors have no conflict of interest to disclose.

Ethical Approval All procedures performed in studies involving human participants were in accordance with the ethical standards of the Institutional Review Board of the University of Texas MD Anderson Cancer Center.

Informed Consent All participants in this study provided written informed consent.