Daily Spousal Responsiveness Predicts Longer-Term Trajectories of Patients’ Physical Function

Abstract

Everyday interpersonal experiences may underlie the well-established link between close relationships and physical health, but multiple-timescale designs necessary for strong conclusions about temporal sequence are rarely used. The current study of 145 patients with knee osteoarthritis and their spouses focused on a novel pattern in everyday interactions, daily spousal responsiveness—the degree to which spouses’ responses are calibrated to changes in patients’ everyday verbal expression of pain. Using couple-level slopes, multilevel latent-variable growth models tested associations between three types of daily spousal responsiveness (empathic, solicitous, and punishing responsiveness), as measured during a 3-week experience-sampling study, and change in patients’ physical function across 18 months. As predicted, patients whose spouses were more empathically responsive to their pain expression showed better physical function over time compared with those whose spouses were less empathically responsive. This study points to daily responsiveness, a theoretically rooted operationalization of spousal sensitivity, as important for long-term changes in patients’ objective physical function.

The quality of close relationships influences a range of meaningful physical-health outcomes, including the rate at which wounds heal (Kiecolt-Glaser et al., 2005), the likelihood of a cardiac event (Orth-Gomér et al., 2000), and mortality (Bulanda, Brown, & Yamashita, 2016). A recent meta-analysis likened a high-quality marital relationship to regular exercise and consumption of fruits and vegetables in the level of health benefit conferred over time (Robles, Slatcher, Trombello, & McGinn, 2014). Indeed, because social interactions are central to everyday life, there is growing interest in how couples’ daily experiences contribute to changes in their physical health.

Developmental frameworks and conceptual models of stress and allostatic load suggest that the effects of daily experiences—both hassles and uplifts—accumulate over time to meaningfully alter longer-term health trajectories (Almeida, 2005; Magnusson & Cairns, 1996; McEwen, 1998; Schilling & Diehl, 2014). The stress literature indicates that interpersonal stressors and uplifts are among the daily experiences with the greatest impact on health and well-being (Bolger, DeLongis, Kessler, & Schilling, 1989; Maybery & Graham, 2001) and consistently relate to concurrent changes in physiological markers (e.g., Joseph, Kamarck, Muldoon, & Manuck, 2014).

Despite the strong conceptual rationale for proposing that daily interactions and experiences contribute to long-term health outcomes, empirical evidence is sparse, likely because multiple-timescale study designs necessary to address such questions have rarely been used (Gerstorf, Hoppmann, & Ram, 2014). Ross, Martin, Chen, and Miller (2011) discovered that female adolescents who reported more negative daily interactions also experienced increased metabolic risk across 2 years. A national study of U.S. adults found that participants who were more emotionally reactive to daily stressors were also more likely to report a chronic physical-health problem 10 years later (Piazza, Charles, Sliwinski, Mogle, & Almeida, 2013). No studies have yet linked couples’ daily interactions with long-term prospective changes in physical health using a design that allows for strong conclusions regarding temporal sequence.

Among the encounters most relevant to couples’ health are their support-related interactions (Feeney & Collins, 2015), and chronic pain has provided a fruitful context for the study of such exchanges. The literature on this topic demonstrates that the effects of spouses’ responses on their partners’ physical function, the ability to safely move about and carry out daily activities unassisted, differs by response type. Empathic and solicitous responses to pain are pain-specific analogues to received emotional and instrumental support, respectively. Spouses’ empathic responses to pain—affection, active listening, and encouragement—often relate to better physical function (Romano, Jensen, Schmaling, Hops, & Buchwald, 2009). In contrast, solicitous responses are linked to worse patient functioning (Romano et al., 2009). Punishing responses—irritation, ignoring, and hostility—also relate negatively to physical function (Cano, Johansen, & Franz, 2005).

Most studies of spouses’ responses and patients’ physical function have been limited by cross-sectional designs that conflate pain sensation and physical function with the expression of pain. Prospective studies of patients’ level of disability and the emotional support they receive have varied widely in the length of follow-up and have produced mixed results (Fekete, Stephens, Druley, & Greene, 2006; Khan et al., 2009); this has prevented strong directional conclusions. Additionally, many studies have examined self-reported outcomes. Objective measures that assess basic functions, such as gait speed and balance, are critical to understanding the long-term effects of spouses’ responses on patients’ health given the strong association of these measures with institutionalization, rehospitalization, and mortality (Volpato et al., 2011).

According to Ainsworth (1969), sensitive attachment figures, sources of reliable support, are attuned to and able to discern cues and are also able to respond contingently and appropriately. For caregivers of infants, this involves noticing fussiness, reading from context what it denotes (e.g., hunger), and promptly satisfying the need (e.g., offering food). Similarly, we argue, sensitive, responsive spouses maintain awareness of patients’ pain cues and calibrate their responses accordingly. Indeed, the consequences of spouses’ responses for patients’ physical function likely depend on whether the spouses’ behavior occurs in response to or in the absence of pain expression. Building on the theoretical concept of interpersonal sensitivity, we suggest that empathic behaviors that covary with the expression of pain may be particularly beneficial for patients, whereas the same actions in the absence of pain expression may not be as salient or, worse, may be received as problematic or indicative of overinvolvement (Martire, Stephens, Druley, & Wojno, 2002). Furthermore, previous studies suggest that solicitous responses to pain threaten patients’ sense of control and competence (Martire et al., 2002), and this effect may be accentuated when attempts to help co-occur with the expression of pain day in and day out. Similar to the approaches taken in studies of daily stress and emotional reactivity (Piazza et al., 2013), our approach was to estimate slopes, for each couple, describing the strength and direction of the association between the patient’s pain expression and the spouse’s response. We expected that patterns of daily responsiveness would predict patients’ health beyond the effects of average perceptions of support and general relationship quality.

With a combined daily-diary and longitudinal approach, we examined the association between spouses’ daily responsiveness to osteoarthritis (OA) patients’ expression of pain and long-term changes in the patients’ physical function across 18 months. Specifically, all patients had knee OA, a common cause of chronic pain that can profoundly limit functional ability by reducing mobility, which exacerbates joint degradation and comorbidities (Owen, Salmon, Koohsari, Turrell, & Giles-Corti, 2014). Patients’ physical function was assessed by objective physical performance rather than subjective ratings, which are susceptible to self-report biases (Pryce et al., 2012). For daily responsiveness to be detectable, the precipitating problem and spouses’ responses to it had to vary from day to day. OA served as an ideal context because OA pain occurs regularly, fluctuates in severity, is a problem many partners communicate about, and is specific enough for relevant daily communication to be measured precisely.

We used multilevel latent-variable growth models to test whether the extent to which spouses’ daily responses tracked with patients’ everyday verbal pain expression predicted prospective changes in patients’ physical function. We hypothesized that patients whose spouses were more empathically responsive (i.e., showed stronger positive links between expression of pain and response) would function better 18 months later than patients whose spouses were less empathically responsive. Because solicitous responses may threaten patients’ sense of control and competence (Martire et al., 2002), we predicted that patients whose spouses were more solicitously responsive would experience steeper declines in physical function; we expected the same for punishing responsiveness.

Method

Study design

The data presented in this report came from a larger study of patients diagnosed with knee OA and their spouses. The parent study combined in-person interviews conducted at baseline (i.e., Time 1, or T1), 6-month follow-up (i.e., Time 2, or T2), and an 18-month follow-up (i.e., Time 3, or T3) with a 22-day assessment of daily experiences immediately after the T1 interview (Martire, Keefe, Schulz, Stephens, & Mogle, 2013). During the daily-assessment protocol, the patients and spouses used a handheld computer to answer questions regarding their health and affect at the beginning of each day, in the afternoon, and at the end of the day and questions regarding their marital and pain-related interactions at the end of the day. For the current study, we used data from the T1, T2, and T3 interviews and the end-of-day diary assessments.

Participants

We used PINT Version 2.11 (Snijders & Bosker, 1993) for the parent study’s multilevel power calculations (described in greater detail in the Supplemental Material available online). Our goal was to evaluate the design’s ability to identify effects with small to moderate effect sizes (ds < 0.30) as statistically significant, assuming power of .80 and an alpha level of .05. The typical multilevel equation in the parent study included a single Level 1 predictor, modeled as a random effect; two Level 2 predictors, modeled as random effects; and the cross-level interactions between the Level 2 and Level 1 predictors, also modeled as random effects. The two Level 2 predictors were grand-mean centered (which set their mean values to zero) and were assumed to be uncorrelated. Given the typical estimates of pain and affect found in multilevel analyses of daily-diary data from chronic-pain patients and their spouses, we set the residual variance unaccounted for by the prediction equation at .7 and the covariance of intercepts and slopes at .3. We found that for a sample size of 132 participants and 21 diary days, our design would have power to detect effects with a magnitude (d ) of 0.22 or greater for all predictors. Thus, with the proposed sampling design, random-coefficient multilevel analyses were expected to detect effects of small magnitude as statistically significant. The study plan therefore called for data collection to conclude when a sample size of 132 was attained.

To be eligible for the study, patients had to be diagnosed with knee OA by a physician and to report that their usual knee pain was of moderate or greater intensity. Patients were also required to be at least 50 years of age, to be married or in a long-term relationship (self-defined), and to share a residence with their partner (also referred to as spouse throughout this article regardless of marital status). Exclusion criteria were a comorbid diagnosis of fibromyalgia or rheumatoid arthritis, use of a wheelchair, or a plan to have hip or knee surgery within the following 6 months. Couples were excluded from the study if the spouse reported arthritis pain of moderate or greater intensity, used a wheelchair, or required assistance with personal care. Both partners were required to be cognitively functional, as indicated by the accuracy of their answers to questions regarding the current date and day of the week, their age, and their date of birth. Both partners also had to be free of any major hearing, speech, or language problems that would interfere with their comprehension of and response to interview and diary questions presented in English.

The primary sources of recruitment were research registries for rheumatology patients and older adults interested in research in the Pittsburgh area, flyers distributed to University of Pittsburgh staff and faculty, and word of mouth. A total of 606 couples were screened for eligibility. Of these, 221 couples declined to participate; the most frequent reasons were lack of interest (n = 87) and illness in the family (n = 55). In addition, 233 couples were not eligible, most often because neither partner had OA in the knee (n = 55) or the OA pain in the knee was mild (n = 47). Thus, the total enrolled sample comprised 152 couples (i.e., 304 individuals).

Of the 152 patients, 150 completed the interviewer-administered physical assessment at T1, 139 had data for the physical assessment at T2, and 120 had data for the physical assessment at T3. A total of 145 couples completed the diary component of the study; their demographics are summarized in Table 1. Altogether, 113 patients underwent all three physical assessments and completed daily diaries. Of the 32 patients who did not complete some part of the relevant study components, 7 did not report on their daily experiences, 20 were missing only their T3 assessment, 6 were missing only their T2 assessment, and 4 lacked both the T2 and the T3 assessments. Missing data were handled in Mplus Version 7.4 (Muthén & Muthén, 2015) using full information maximum likelihood, which makes use of all available data. This approach assumes that the data are missing at random. Thus, it allows systematic patterns of missingness as long as contributing factors are taken into account (Bolger & Laurenceau, 2013). The most likely predictors of missing physical-function data were relationship problems or illness-related disability, both of which were explicitly modeled in our analyses (i.e., marital satisfaction and initial physical function).

Table 1.

Demographic Characteristics of the Patients and Spouses Included in the Analyses (N = 145 Couples)

Variable	Patients	Spouses
Age (years)	M = 65.6, SD = 9.8	M = 65.3, SD = 11.5
Male	43%	58%
White	88%	86%
Education (years)	M = 16.1, SD = 2.0	M = 15.9, SD = 2.1
Employed full time	33%	33%
Duration of knee osteoarthritis (years)	M = 12.6, SD = 11.3
Marital satisfaction (0–50)	M = 39.61, SD = 6.26	M = 38.94, SD = 6.47

Note: The average household income of the couples was in the range from $40,000 to $59,000 per year, and the couples had been in their relationships for 34.3 years, on average (SD = 16.6). Full-time employment status was defined as typically working 30 hr or more per week. Marital satisfaction was assessed with the 10-item Dyadic Satisfaction subscale of the Dyadic Adjustment Scale (Spanier, 1976).

Data-collection procedure

Trained staff interviewed the patients and their spouses separately in their homes. Following the T1 interview, the couples were trained to use the handheld computer (i.e., the Palm TX; Palm, Inc., Sunnyvale, CA) and were familiarized with the format and content of the diary questions. The handheld computer and questionnaire were designed for easy use by older adults and people with minimal computer experience; features for easy accessibility included a large font size and an oversized stylus for registering responses. Each patient and spouse was provided with a handheld computer that was clearly labeled with his or her name, and the importance of completing diary assessments independently was emphasized. Surveys were intended to be completed in the morning, afternoon, and evening. More specifically, participants were instructed to answer questions (a) within 60 min of rising in the morning, (b) between 2:00 and 4:00 p.m., and (c) upon retiring at night. The current study focused on the end-of-day assessments.

Completion and compliance rates were examined for the diary data. Out of a potential 6,380 end-of-day observations (290 individuals in 145 couples × 22 days), a total of 5,863 were completed (92%). Compliance with the requested timing of the end-of-day assessment was evaluated by comparing the time of the computer entries with participants’ written logs of daily bedtimes. End-of-day assessments that were completed more than 120 min before bedtime were excluded from analysis. This left 5,327 of the 5,863 completed observations in the analysis (i.e., 91% of the completed observations, or 83% of the total possible observations). Completion and compliance rates were very similar for patients and spouses.

Measures

Both patients’ verbal expression of pain and spouses’ responses to patients’ pain were reported by the other partner for conceptual and methodological reasons. The conceptual reason was that using partner reports of behavior better captured the transactional nature of communication: Spouses’ interpretation of patients’ pain expression is most important for how spouses respond to the pain, and spouses’ responses are meaningful to the degree that they are successfully communicated to patients. This approach builds on theoretical notions of interpersonal sensitivity (Ainsworth, 1969), reflecting the degree to which a spouse detects the patient’s cues and also responds in a way that the patient can understand. The methodological reason for this approach was that using reports from each partner reduced shared reporter variance.

Verbal pain expression

Patients’ verbal expression of pain was assessed by spouses at the end of each day using an item developed for this study. On a 3-point scale (1 = not at all, 2 = somewhat, 3 = very much), spouses responded to the following question: “Today, to what extent did your spouse describe his/her pain to you?”

Spouses’ responses to patients’ pain

Spouses’ responses to patients’ pain were assessed at the end of each day by patients’ answers to nine Likert-type items with 3-point rating scales (1 = not at all, 2 = somewhat, 3 = very much). Six of the items were adapted for daily assessment from the West Haven-Yale Multidimensional Pain Inventory, a valid and reliable measure used in chronic-pain research (Kerns, Turk, & Rudy, 1985). Three of these items were intended to measure spouses’ solicitous behaviors, such as offering food or drink, taking over tasks, and encouraging rest (e.g., “Today, your spouse took over your jobs or duties to help you avoid pain”). The other three items were designed to assess punishing behaviors—ignoring the patient, acting frustrated, and seeming irritated (e.g., “Today, your spouse got frustrated with you when you seemed to be in pain”). The three remaining items, intended to measure empathic responses, were adapted from a scale assessing spouses’ emotional support in response to patients’ pain (Stephens, Martire, Cremeans-Smith, Druley, & Wojno, 2006). These items referred to showing affection, understanding the patient’s feelings about the pain, and providing attention (e.g., “Today, your spouse tried to just be there for you when you seemed to be in pain, by giving you his/her undivided attention”).

On the basis of our findings from a factor analysis (described later in this section), we summed responses to the items referring to punishing behavior to produce a punishing-response scale. The responses to the items referring to solicitous behavior were summed to create a separate scale. Finally, the responses to the items referring to empathic behavior were summed to create an empathic-response scale. These items (showing affection, understanding feelings, and providing attention) map onto the three components of perceived partner responsiveness: care, understanding, and validation, respectively (Reis & Shaver, 1988). Each scale ranged from 3 to 9, with higher scores indicating more of the response. Empathic and solicitous responses were highly correlated at the within-person level (r = .70), and both were weakly negatively correlated with punishing responses (r = −.14 and r = −.06, respectively).

We calculated generalizability coefficients to estim-ate between- and within-person reliability (Bolger & Laurenceau, 2013). The between- and within-person reliabilities were, respectively, .55 and .50 for punishing responses, .81 and .58 for solicitous responses, and .86 and .67 for empathic responses. According to Nezlek (2016), traditional Cronbach’s α benchmarks are not directly applicable to within-person reliability estimates because fewer items are used in successful ecological momentary assessment than in other study designs and because current multilevel-modeling algorithms reduce the weight of low-reliability observations in significance tests.

Because none of the nine items had been examined at the daily level in previous research, a two-level exploratory factor analysis was conducted to investigate the factor structure of spouses’ responses at both the between-couples and the within-couple levels. Using Mplus Version 7.4 (Muthén & Muthén, 2015), we specified factor parameters such that solutions ranged from one factor for each level to three factors for each level (see the code in the Supplemental Material). Accordingly, all possible combinations of factor structures were generated with and without an oblique Geomin rotation, as were their respective fit statistics. The three-factor/within-person, three-factor/between-person solution provided the best fit (root mean square error of approximation = .01; comparative-fit index = .96; Hu & Bentler, 1999), and significantly improved fit compared with the multilevel two-factor model, χ²(16) = 186.7, p < .05. Item loadings at the within- and between-person levels approximated simple structure on the three separate scales—punishing responses, empathic responses, and solicitous responses.

Patients’ physical function

The Short Physical Performance Battery (Guralnik et al., 1994) was administered by an interviewer to assess patients’ balance, gait speed, and ability to rise from a chair at T1, T2, and T3. Patients’ balance was assessed by observing them standing with their feet in a tandem position, their gait speed was assessed by timing how long it took them to walk 8 ft, and their ability to rise from a chair was assessed by timing how long it took them to alternate between standing and sitting in a chair five times. Scores on this scale range from 0 to 12, with higher scores indicating better physical performance. Performance of these tasks has strong predictive validity for the ability to independently carry out activities of daily living and the likelihood of institutionalization, rehospitalization after discharge, and mortality (Volpato et al., 2011).

Covariates

Covariates for the multilevel models were chosen on a conceptual basis. Lagged spousal responses (i.e., responses from the previous day) were included in each model to allow us to conclude whether same-day pain expressions and spouses’ responses were related above and beyond stability in responses across days. Because patients’ experience of pain serves as an interpretive backdrop that may color their perceptions of spouses’ responses, patients’ reports of pain were included as an important covariate (Stephens et al., 2006). Patients’ pain severity was assessed with one item in the daily diary at the end of the day: “How would you describe your overall arthritis pain today?” The response scale ranged from 0 (none) to 3 (severe). In order to allow conclusions regarding the effects of daily responsiveness above and beyond individuals’ overall levels of pain expression and response, we covaried the growth-curve component of the model for person-level means of daily pain expression and spousal response. Because average levels of a spouse’s empathic responses to pain represent a pain-relevant characterization of perceived partner responsiveness, it can be argued that controlling for this measure generates a particularly strict test of daily spousal responsiveness. Further, person-level means for empathic responses were derived from daily reports and, therefore, did not tap into recall biases and heuristics as strongly as general measures would.

Between-person covariates also included age and years elapsed since the diagnosis of knee OA because physical function typically declines with age and time spent living with the disease. Patients’ marital satisfaction at T1 was controlled to test the unique contributions of everyday responsiveness to physical function, above and beyond general perceptions of the relationship, unlike in previous studies of perceived partner responsiveness (Khan et al., 2009; Slatcher, Selcuk, & Ong, 2015). The 10-item Dyadic Satisfaction subscale of the Dyadic Adjustment Scale (Spanier, 1976) was used. On a scale from 0 to 50, patients reported high marital satisfaction on average (M = 39.61, SD = 6.26), and ratings showed adequate internal consistency (α = .89).

Data reduction and analysis

Prior to analysis, the daily predictors and covariates were person-mean centered to allow conclusions about deviations from a person’s usual behavior, which was of greater conceptual interest than scores on an arbitrary scale. That is, we expected that changes relative to patients’ typical levels of pain and pain expression would be more meaningful than raw scores for predicting spouses’ daily responses. Between-person predictors were grand-mean centered. Next, a series of multilevel models were fit using Mplus Version 7.4 (Muthén & Muthén, 2015) to test random-slope variance for each combination of verbal pain expression and response type (i.e., punishing, empathic, and solicitous; see Fig. 1). Because the predictors were spouse reported and the outcomes were patient reported, each multilevel model contained two levels: day and couple. For models with significant random-slope variance in daily responsiveness, as indicated by significant improvement in fit from the random-intercept-only model (determined by a likelihood-ratio test), growth models with a multilevel latent slope were estimated. Assigning a latent variable to the slope term that subsumed estimates of daily associations served as a more powerful alternative to a two-step approach of estimating multilevel models and then using the slopes as input to other models (Raudenbush & Sampson, 1999). Time scores were set to 0, 1, and 3 to index a linear trend across the three unequally spaced interview occasions, T1 (baseline), T2 (6 months after baseline), and T3 (18 months after baseline). Models were fit with and without covariates in their growth-curve portions. The effects of interest did not differ when the covariates were omitted (see Tables S1 and S2 in the Supplemental Material for these estimates; Mplus code for the multilevel latent-variable growth models is also included in the Supplemental Material).

Fig. 1.

Schematic representation of the model used to test the primary research questions. The circles and oval denote latent factors; the squares represent manifest variables. The double-headed curved arrow represents covariance, and the circular arrows signify estimated variances. The numbers next to the arrows index the effect coding used to capture the intercept and linear slope of the growth-curve model.

Our measure of daily spousal responsiveness builds on the hallmark components of interpersonal sensitivity as described by attachment theorists (e.g., Ainsworth, 1969) in a way that general responsiveness scales have not. According to Ainsworth (1969), sensitive attachment figures—sources of reliable, predictable support—are attuned to and able to discern cues, and can respond contingently and appropriately. We captured the degree to which spouses detected and interpreted patients’ cues as pain expression with spouses’ daily reports of patients’ verbal pain expression. The way in which spouses communicated various responses to patients—empathic and instrumental support, as well as hostility—was indexed with patients’ daily reports of spouses’ empathic, solicitous, and punishing reactions to pain. The predictability of spouses’ calibration of their responses to patients’ verbal pain expression day in and day out was captured with the latent couple-level slope of spouses’ daily responses as a function of patients’ pain expression. A stronger, more positive estimate suggested that spouses’ responsiveness was more consistent day to day. Failure either to detect pain cues regularly or to respond in a way interpretable for the patient resulted in a lower daily responsiveness estimate. In this way, our dyadic, dynamic measure of daily spousal responsiveness directly followed and tested the mechanics of interpersonal sensitivity theorized to confer benefit for recipients.

According to our hypothesis, slope estimates were expected to be significant—positive for empathic responsiveness and negative for solicitous and punishing responsiveness. No specific predictions were made for the association between spouses’ responsiveness and the intercept parameter, which represented concurrent physical function. A significant association between responsiveness and change in physical function (i.e., the slope) along with a null association between responsiveness and baseline physical function (i.e., the intercept) would suggest that the effects of spouses’ responsiveness patterns emerged over time. A significant link between responsiveness and the slope paired with a significant association between the intercept and responsiveness would indicate stability of spouses’ responsiveness over time, as the most recent diagnosis was 1 year prior to the study, or an immediate effect on physical function, a less likely possibility.

Results

Multilevel models

Multilevel models for empathic, solicitous, and punishing responses were examined to confirm that there was significant variability in the expression-response random slopes (i.e., couple-specific associations between pain expression and spousal response) before relating the terms to longitudinal change in patients’ physical function. Because the objective was to relate differences in daily spousal responsiveness to changes in physical function, fixed slopes, or average sample-level associations from the multilevel models, were not of interest in this study. All three models converged, and solutions were positive definite. Likelihood ratio tests revealed that the random-slope empathic-responsiveness model, χ²(2) = 70.41, p < .0001, and the random-slope solicitous-responsiveness model, χ²(2) = 31.82, p < .0001, fit the data better than the intercept-only models. According to empirical Bayes predictions (Bolger & Laurenceau, 2013), daily empathic responsiveness ranged from −0.32 to 1.48 (M = 0.46, SD = 0.32). In other words, on average, a 1-point increase in verbal pain expression (from “not at all” to “somewhat,” or from “somewhat” to “very much”) was associated with a half-point increase in the spouse’s empathic response (on a 7-point scale). The empathic response of the most empathically responsive spouse increased 1.5 points, or almost 1 SD, on days when the patient’s pain expression increased 1 point. The empathic response of the least empathically responsive spouse decreased slightly less than a half-point on days when the patient’s pain expression increased 1 point. Spouses’ solicitous responsiveness ranged slightly less than their empathic responsiveness, from −0.26 to 1.31 (empirical Bayes predictions: M = 0.33, SD = 0.28). According to a likelihood ratio test of the punishing-responsiveness models, the fit of the random-slope model was not better than that of the random-intercept-only model, χ²(2) = 0.12, p > .05, so punishing responsiveness was not analyzed further.

Multilevel latent-variable growth curves

The growth-curve model of physical function was first estimated without predictors. On average, patients’ linear trends of physical function were not significantly different from zero (95% confidence interval, CI = [−0.01, 0.04], p = .152). Nevertheless, there was significant variability in initial physical function (σ² = 3.53, SE = 0.13, 95% CI = [3.28, 3.78], p < .0001), as well as variability in change across the 18-month study period (σ² = 0.11, SE = 0.03, 95% CI = [0.05, 0.18], p = .001). This implies that, though the average trajectory of physical function was not different from zero, there were significant individual differences in trajectories, which were modeled in the next step.

Empathic-responsiveness model

Daily empathic resp-onsiveness did explain significant variance of linear change in physical function across the 18-month follow-up period, above and beyond the effects of patients’ average levels of pain expression and spouses’ average empathic responses across the diary period, as well as the effects of age, years since the diagnosis of knee OA, and initial marital satisfaction (estimate = 0.337, SE = 0.152, 95% CI = [0.039, 0.635], p = .027; see Table 2 for a summary of model results). As predicted, patients in dyads that exhibited stronger daily association between pain expression and empathic responses (i.e., higher responsivity) showed steeper linear improvements in physical function across 6 and 18 months. More specifically, patients whose spouses were 1 SD above the mean on empathic responsiveness improved 0.26 points in their physical function over 18 months, whereas patients whose spouses were 1 SD below the mean on empathic responsiveness made a gain of 0.05. Differences in daily empathic responsivity did not significantly relate to initial levels of patients’ physical function (95% CI for the effect = [−0.515, 1.532], p = .330). However, patients whose spouses were perceived to be more empathic on average had lower initial levels of physical function, a common cross-sectional finding (e.g., Selcuk & Ong, 2013). This pattern arguably reflects a greater need for help that results from poorer function.

Table 2.

Results for the Latent-Variable Growth Curve of Patients’ Physical Function as Predicted by Spouses’ Daily Empathic Responsiveness (N = 145 Couples)

Effect	Estimate	SE	95% confidence interval	p
Intercept
Empathic responsiveness	0.509	0.522	[−0.515, 1.532]	.330
Pain expression	−0.805	0.472	[−1.730, 0.120]	.088
Empathic response	−0.285	0.114	[−0.508, −0.062]	.012
Patient’s age	−0.071	0.018	[−0.106, −0.036]	< .0001
Years since diagnosis	−0.020	0.013	[−0.046, 0.005]	.118
Patient’s marital satisfaction	0.060	0.023	[0.015, 0.105]	.009
Slope
Empathic responsiveness	0.337	0.152	[0.039, 0.635]	.027
Pain expression	−0.102	0.162	[−0.420, 0.216]	.530
Empathic response	0.039	0.045	[−0.049, 0.127]	.386
Patient’s age	0.006	0.005	[−0.005, 0.016]	.309
Years since diagnosis	−0.004	0.006	[−0.016, 0.009]	.578
Patient’s marital satisfaction	0.003	0.012	[−0.020, 0.025]	.804

Note: Empathic responsiveness refers to daily empathic responsiveness as captured with the latent slope term, and empathic response denotes the person-level means of spouses’ empathic responses across the 22-day diary period.

Ancillary analyses examined the significance of daily empathic responsiveness as a predictor of the trajectory of physical function controlling for daily solicitous responsiveness and daily punishing responsiveness. The model solution with daily punishing responsiveness included was nonpositive definite. Daily empathic responsiveness remained significant with daily solicitous responsiveness in the model (b = 0.666, SE = 0.286, p = .02). For the sake of parsimony, these additional covariates were not included in the final models.

Solicitous-responsiveness model

Differences in the association between pain expression and solicitous responsivity did not relate significantly to initial levels (95% CI for the effect = [−0.695, 1.922], p = .358) or prospective changes (95% CI for the effect = [−0.524, 0.384], p = .762) in patients’ physical function (see Table 3 for a summary of model results). Patients whose spouses were perceived to be more solicitous on average had lower initial levels of physical function, as did those whose spouses were perceived to be more empathic on average. Again, this may indicate that more disabled people require more help.

Table 3.

Results for the Latent-Variable Growth Curve of Patients’ Physical Function as Predicted by Spouses’ Daily Solicitous Responsiveness (N = 145 Couples)

Effect	Estimate	SE	95% confidence interval	p
Intercept
Solicitous responsiveness	0.613	0.668	[−0.695, 1.922]	.358
Pain expression	−0.661	0.504	[−1.649, 0.327]	.190
Solicitous response	−0.464	0.151	[−0.760, −0.169]	.002
Patient’s age	−0.070	0.018	[−0.105, −0.035]	< .0001
Years since diagnosis	−0.019	0.013	[−0.044, 0.007]	.148
Patient’s marital satisfaction	0.053	0.021	[0.011, 0.095]	.013
Slope
Solicitous responsiveness	−0.070	0.231	[−0.524, 0.384]	.762
Pain expression	−0.066	0.159	[−0.378, 0.245]	.678
Solicitous response	0.021	0.051	[−0.079, 0.120]	.685
Patient’s age	0.004	0.006	[−0.007, 0.014]	.527
Years since diagnosis	−0.004	0.006	[−0.017, 0.009]	.536
Patient’s marital satisfaction	0.005	0.011	[−0.016, 0.026]	.630

Note: Solicitous responsiveness refers to daily solicitous responsiveness as captured with the latent slope term, and solicitous response denotes the person-level means of spouses’ solicitous responses across the 22-day diary period.

Discussion

The cumulative effects of everyday interpersonal experiences may explain the impact of intimate relationships on health. However, few studies have demonstrated that daily experiences influence long-term changes in health (Piazza et al., 2013; Ross et al., 2011), and none have shown such a link for couples’ everyday interactions. Associations between daily responsiveness and trajectories of physical function were examined in a sample of chronic-pain patients, whose condition represents an established support context and public-health challenge. As hypothesized, patients whose spouses were more empathically responsive to daily expression of pain demonstrated better physical function 18 months later than patients whose spouses were less empathically responsive. However, neither solicitous nor punishing responsiveness predicted change in physical function. To our knowledge, this study is the first both to link patterns of couples’ day-to-day interactions prospectively to an objective clinical end point and to capture the dyadic, dynamic nature of spouses’ responsiveness.

Several features make the finding for spouses’ daily empathic responsiveness a unique contribution to the literature. Our study combined daily-diary data with prospective longitudinal data, in line with the developmental principle that accrual of short-term effects leads to long-term changes (Magnusson & Cairns, 1996). The fact that couples’ interactions preceded the outcome measure of physical function supports the hypothesized temporal sequence. Furthermore, empathic responsiveness was not associated with baseline physical function, and therefore was not merely a reflection of long-standing individual differences. Indeed, daily responsiveness is a joint, dynamic process whose effects become manifest only over time. In our study, it related to changes in how fast participants walked, how well they balanced, and how well they were able to rise from a chair unassisted. Performance of these tasks predicts the ability to independently carry out activities of daily living, as well as the likelihood of institutionalization, rehospitalization, and mortality (Volpato et al., 2011). Couples’ daily communication was assessed over 22 days in a sample of both patients and spouses (as recommended in seminal reviews: Leonard, Cano, & Johansen, 2006; Robles et al., 2014). This approach reduced shared reporter variance and captured the transactional nature of communication (Hadjistavropoulos et al., 2011).

Daily spousal responsiveness, indexed with couple-specific slopes, emerged as a distinct construct that explained significant variance in physical-function trajectories above and beyond spouses’ support levels and patients’ pain expression, marital satisfaction, age, and time since diagnosis. In fact, spouses’ overall empathic support (i.e., their person-level average of empathic responses across days) did not explain change in physical function when daily responsiveness was excluded from the model. That is, empathic responses promoted physical function only when they tracked with daily pain expression.

Responsiveness in intimate adult relationships and parent-child dyads has been defined as the proclivity to respond to bids in an appropriate and timely, or “sensitive,” way (Ainsworth, 1969). In this study, our measure of daily responsiveness reflected the dynamic calibration of spouses’ empathic behavior to patients’ pain cues. Other studies of adult couples have indexed perceived partner responsiveness with individuals’ overall perceptions of their partners’ understanding, care, and validation (Khan et al., 2009; Reis & Shaver, 1988; Slatcher et al., 2015; Selcuk & Ong, 2013), rather than the partners’ context-dependent awareness of and titration of responses to needs. This distinction may explain why Khan et al. (2009) found a null effect for perceived partner responsiveness in their prospective study of patients’ postsurgical knee limitations. Slatcher et al. (2015) reported a link between perceived partner responsiveness and cortisol 10 years later, but could not control for baseline cortisol with the data that were available. Selcuk and Ong (2013) treated general perceived partner responsiveness as a moderator of received emotional support and mortality. Unlike these previous studies, ours operationalized responsiveness with a daily measure explicitly linked to pain expression, thus capturing the dyadic nature of interpersonal sensitivity. This measure relied on interaction dynamics across multiple days and predicted physical-function trajectories after we accounted for patients’ initial physical function, pain-related perceived partner responsiveness (i.e., average person-level empathic re-sponses to pain), and marital quality. Our analyses also separated the day-to-day coupling of behaviors from the content of spouses’ responses, an important theoretical advance called for in the literature (Selcuk & Ong, 2013). Indeed, different kinds of responses (empathic, solicitous, or punishing) should have disparate effects on patients (Romano et al., 2009).

Contrary to our hypothesis, solicitous responsiveness did not relate to prospective declines in physical function, though the association was in the expected direction. Excessive instrumental support may challenge efficacy and discourage patients from attempting tasks independently (Martire et al., 2002), but people likely differ in whether or not they experience support as disproportionate to their needs. Unstable patterns of solicitous responsiveness may account for the null effect, as spouses who are solicitously responsive at one time may not be at another time if their assistance is titrated to patients’ fluctuating needs.

Our hypothesis regarding punishing responsiveness could not be tested, as this effect lacked significant variance across couples. Punishing responses were infrequent, and their severity fluctuated little. The couples in our sample, married for 35 years on average, were highly satisfied (M = 39 on a 50-point scale) and likely avoided turbulent daily patterns.

The present study has limitations. Future studies would benefit from a wider range of negative responses from spouses and samples that include more dissatisfied couples, so that punishing responsiveness can be assessed. Also, because we examined same-day associations between pain expression and responses, we cannot confirm temporal sequence within days, though our models included spouses’ responses on the previous day to strengthen interpretation. A measurement-burst design that includes multiple intensive assessments could be used to test whether daily responsiveness exhibits time-varying properties across age, relationship length, and time since diagnosis (Sliwinski, 2008). Such designs should also be used to appropriately test proximal and distal mechanisms of daily responsiveness, which likely include a cascade from momentary mood, pain, interactions, and appraisals to thresholds for engaging and persisting in physical activity. Future studies should also test the generalizability of the effects of daily responsiveness on health for a range of stressors.

The current study demonstrates that couples’ everyday interactions affect patients’ long-term trajectories of objectively assessed physical function. Our conclusions are enhanced by the external validity of experience sampling and the temporal sequence of the longitudinal design. Daily spousal responsiveness, captured using within-couple, across-day slopes, is a unique relational construct that reflects important interpersonal processes in couples’ everyday lives. Our findings support a dynamic developmental perspective, according to which brief exchanges assemble into longer patterns that over days, months, and years accrue to wield measurable effects on physical health and function.