Affective reserve in older adults with and without multiple sclerosis: A residual-based approach

Abstract

Background:

The concept of affective reserve has been proposed, but not operationalized, as a buffer designed to regulate mood and behavior in response to adverse events. This study derived a residual-based metric of affective reserve and evaluated whether it mediated associations between having multiple sclerosis and subjective well-being in older adults.

Methods:

Sociodemographics and brain pathology metrics were regressed against a composite mood, extraversion, and consciousness variable to derive affective reserve in a sample of older adults with multiple sclerosis (n = 107) and without (n = 104). Linear regressions were run to examine the associations between affective reserve and well-being, measured with the Short Form-12 Mental Composite Summary and Physical Composite Summary. Hayes’ PROCESS was used to evaluate affective reserve as a mediator between group status and well-being.

Results:

Affective reserve was positively associated with both Mental Composite Summary and Physical Composite Summary. After covariate adjustments, affective reserve fully mediated the relationship between having multiple sclerosis and Mental Composite Summary (b = 1.350, 95% confidence interval: 0.142–2.591) and partially mediated the association between multiple sclerosis and Physical Composite Summary (b = 0.915, 95% confidence interval: 0.073–1.906).

Conclusion:

This residual-based metric offers a novel approach for measuring affective reserve. Not only is it associated with both physical and mental well-being, but it explained the relationship between the latter and the presence of multiple sclerosis.

Introduction

Multiple sclerosis (MS) can result in a wide range of physical and psychological symptoms that can affect daily functioning, with many individuals reporting negative effects on their emotional well-being, coping abilities, future outlook, and social interactions.¹ The odds of having any mental health comorbidity, such as depression and anxiety, are 2.94 times higher in people with MS compared to those without MS after adjusting for demographics.² While several factors have been associated with these mental health conditions in MS, including disease-related physiological changes and psychosocial stressors,³ not everyone develops symptoms. One potential explanation for this discrepancy is affective reserve (AR).⁴

AR has been defined as the “capacity to regulate affective states and dispositions,” which can help explain why individuals have different responses when they experience stress due to disease activity, life events, or aging.⁵ An expansion from the previous concept of “emotional reserve,” AR includes one’s ability to regulate mood, emotions, and motivation, as well as social support and coping behaviors, all within the same framework (Figure 1).⁵ When an individual has higher AR, their brain has a stronger ability to adapt to these stressors, buffering the effects of biopsychosocial stressors (e.g., MS-related disability and life changes) and thus lowering their likelihood of developing emotional distress.⁴ AR can be considered as a potential predictor of resilience, with the latter being one’s response to stressors.⁵

Figure 1:

Framework of affective reserve (AR), which expands upon previous conceptualizations by simultaneously considering multiple processes within the same mechanism (5). This conceptualization of AR considers not only one’s capacity for regulating mood, emotions, and motivation, but also coping behaviors and social support (5). This approach is in line with conceptualizations of cognitive reserve (CR), in which multiple cognitive processes are considered within the same mechanism to capture “the adaptability of all the different cognitive processes,” such as with a composite measure that assesses several cognitive domains like the Repeatable Battery for the Assessment of Neuropsychological Status (5, 6). As the different processes in AR should not be evaluated separately, we propose using a composite proxy variable to derive AR that includes mood (reverse scored so higher scores indicate better functioning) and personality traits that are associated with coping behaviors, social support, emotional experiences, and self-motivation (25–28).

A challenge in investigating AR is quantifying it. Proxy measures, such as those assessing coping styles, have limitations due to their possible bidirectionality with depression.⁴ An alternative approach is using a residual-derived metric, in which brain pathology and sociodemographic variables are regressed on an outcome measure, with the unexplained variance used to quantify the individual’s reserve (Figure 2).⁶ This approach has previously been used in MS to assess cognitive reserve (CR) and physical reserve (PR), with the Repeatable Battery for the Assessment of Neuropsychological Status (RBANS) and Short Physical Performance Battery (SPPB) as the regression outcome measures, respectively.⁶ Both proxy measures reflected multiple processes within their respective domain, allowing these processes to be simultaneously considered within the same residual-derived metric. Notably, the residual-based approach does not directly use these composite proxy variables to measure reserve, such as when education and premorbid verbal intelligence are used to quantify CR.⁷ Instead, it is used as an outcome for their functioning in the domain (e.g., RBANS for cognition), with reserve being the remaining variance in the model that is not explained by sociodemographic or brain pathology variables.⁸ In the case of quantifying AR, measures that tap into the different “affective states and dispositions” can be used as a composite proxy variable in the derivation model, with the residual representing their capacity to regulate them.⁵ Furthermore, as changes over time can be evaluated using residual-derived scores,⁸ it is congruent with the conceptualization of AR as a modifiable factor and provides an advantage over using more static proxy measures.⁴

Figure 2:

Derivation of a residual-based reserve metric. A proxy variable is selected, which captures the different processes in the domain, either through an index measure (e.g., Repeatable Battery for the Assessment of Neuropsychological Status for deriving cognitive reserve (6)) or composite variable of several measures (see Figure 1). This variable is entered into a regression model as the dependent variable, with socio-demographics and brain pathology metrics entered as the independent variables. The remaining variance after accounting for these covariates is the final reserve metric, with higher values indicating greater reserve (e.g., higher capacity to regulate the outcome variable). In other words, when an individual has better functioning in the domain than what can be accounted for by socio-demographics and brain pathology, then that is attributed to their reserve.

Therefore, this study aimed to derive a residual-based metric of AR in older adults with MS and healthy controls (HCs) and evaluate its relationship with subjective well-being. It was hypothesized that higher AR would be associated with greater mental and physical well-being. In addition, as AR has been proposed as a factor that may explain who does and does not develop emotional distress,⁴ we tested the hypothesis that AR would mediate the relationship between the presence of MS and subjective well-being.

Patients and methods

Participants

This study was a secondary analysis of the “Brain Predictors of Mobility and Falls in Older Adults in Multiple Sclerosis” study,⁹ which enrolled both HC and people with MS, the latter who were at least 6 months stable on a disease-modifying therapy. In addition, their MS diagnoses met the 2017 revised McDonald criteria,¹⁰ which were physician-confirmed through a medical record review.⁹ All participants were (1) 60 years or older; (2) able to walk independently, had no (3) other major neurological or psychiatric conditions; (4) significant hearing or vision impairments; and (5) magnetic resonance imaging (MRI) contraindications. In addition, participants were excluded from the current analyses if they met criteria for a major neurocognitive disorder¹¹ or they did not complete the measures used to derive AR. Consistent with other work deriving residual-based reserve metrics,⁶ both people with MS and HC were included in the analyses, allowing for a wider range of brain pathology to be examined in the models. Furthermore, as AR is applicable in normal aging and other conditions besides MS,⁵ it was important that the variables used to derive the metric be replicable across populations and not limited to a specific disease (e.g., MS phenotype). A total of 211 participants (MS: n = 107; HC: n = 104) were included, with this sample size having 95% power to detect a medium effect in a linear regression with eight variables and an alpha of 0.01.¹²

Procedures

Participants signed an informed consent form approved by the Albert Einstein College of Medicine Institutional Review Board (IRB). Demographics, including age, education, sex, and race, were collected from all participants, while information on MS phenotype and disability via the self-report Patient Determined Disease Steps (PDDS)^13–16 was gathered from the MS cohort. A comorbidity index (global health status (GHS)) was calculated, which ranged from 0 to 10, based on the presence of the following 10 conditions: angina, arthritis, chronic obstructive lung disease, chronic heart failure, depression, diabetes, hypertension, myocardial infarction, Parkinson’s disease, and stroke.¹¹

Study assessments were conducted between September 2019 and February 2025 over two in-person visits. Questionnaires were completed during both visits, with the brain MRI done during the second appointment. On average, there were 8.86 ± 12.30 days between visits, excluding two participants who had longer intervals (218–219 days) due to COVID-19 restrictions. As previously noted,⁶ there were no notable deviations in their data and they were included in the analyses.

Study MRIs were done at the Gruss Magnetic Resonance Research Center of Albert Einstein College of Medicine using a 3T Philips scanner. The procedures for image acquisition and analyses have been described elsewhere.⁶ Total white matter (WM) lesion volume was selected as one of the brain pathology metric for the regression models to derive AR due its clinical relevance in MS,¹⁷ aging,¹⁸ and depression,¹⁹ while thalamus and amygdala volume were included given their associations with depression in MS.⁴ As with the derivation of CR and PR,⁶ intracranial volume was included in the models as a covariate as gray matter volumes were included in the model.

As previously described,⁶ T1 and fluid-attenuated inversion recovery (FLAIR) images were used to identify WM hypo- and hyperintensities, with total lesion volume computed using the lesion segmentation toolbox (LST).²⁰ This process does not differentiate between MS-related FLAIR hyperintensities and other age-related hyperintensities (e.g., due to small vessel disease, expected in both MS and HC), so that the WM lesion load measure included both factors.⁶ Automatic segmentation was done using the FreeSurfer software²¹ to compute the subcortical and intracranial volume metrics, which have been detailed in prior publications.^6,22

Measures

Derivation of AR.

The process for deriving AR followed prior work on CR and PR,⁶ with variable selection considering both theoretical conceptualizations and data-based findings. First, an outcome measure was selected for the regression model. Based on the AR framework,⁵ a composite variable was created that would capture mood and other “affective states and dispositions” (Figure 1). Three measures were selected: the GDS²³ and the Extraversion and Conscientiousness subscales from the 15-item version of the Big Five Inventory (BFI).²⁴ All measures were coded so that higher scores would be indicative of better affective functioning. Extraversion and Conscientiousness were selected as they have been associated with problem-centered coping and subjective well-being in MS.²⁵ In addition, Extraversion has been connected to social interactions and emotional experiences,²⁶ while Conscientiousness is positively related to aspects of self-management.^27,28 As these selected measures captured the different processes in the AR framework,⁵ Neuroticism was not included in the proxy measure. In addition, Spearman’s correlations were done to evaluate that the selected measures correlated with at least one of the MRI metrics^4,19 selected for the AR derivation model.

All three measures were given equal weighting in the composite variable, like the RBANS and SPPB total scores.⁶ The composite outcome measure was then entered into a regression model, with sociodemographics and brain pathology metrics as covariates (see Statistical Analyses; Figure 2). The standardized residual from the final model was extracted as the AR metric (e.g., capacity to regulate the affective outcome), with higher values indicative of greater AR.

Subjective well-being.

The 12-item Short Form Health Survey (SF-12) was used to measure subjective well-being, with the two summary measures calculated using the normative-based methods in the manual.²⁹ The Mental Composite Summary (MCS-12) included items measuring mental health, vitality, social functioning, and emotional roles, while the Physical Composite Summary (PCS-12) included items assessing physical functioning, bodily pain, physical roles, and general health. Both the MCS-12 and PCS-12 were transformed so that a mean of 50 (standard deviation (SD): 10) was representative of the general US population.²⁹

Statistical analyses

All analyses were conducted using SPSS version 26. Descriptive statistics were run to characterize both groups (MS and HC), with chi-squares and Mann–Whitney U tests used for bivariate analyses, given the non-normal distribution of the continuous variables. All participants were included in the regression analyses. To derive AR, a linear regression was conducted, with the composite GDS, BFI Extraversion, and BFI Conscientiousness variable as the outcome. Consistent with the process used to derive CR and PR,⁶ a stepwise backward selection approach was used. All sociodemographic variables (age, education, sex, race, and GHS score) and brain pathology metrics (WM lesion volume, thalamic atrophy, amygdala atrophy, and intracranial volume) were entered into the model, with race dichotomized. Non-significant variables (p ⩾ 0.10) were removed individually until the remaining variables all met statistical inclusion criteria (p < 0.10). Intracranial volume was retained as a covariate in the final model as gray matter volumes were included in the derivation of AR.⁶ Differences in AR between participants with and without a history of depression were evaluated using independent t-tests.

Separate linear regressions for each subjective well-being outcome (MCS-12 and PCS-12) were then run with the residual-derived AR as the predictor in both unadjusted and adjusted models. For the latter, group status (MS versus HC), age, sex, education, and GHS were included as covariates.⁶ Hayes’ PROCESS³⁰ was used to conduct simple mediation analyses, with group status as the predictor, AR as the mediator, and MCS-12 and PCS-12 as separate outcomes. Both unadjusted and adjusted models were examined, using the same covariates (age, sex, education, and GHS). Percentile bootstrapping confidence intervals (CIs) were computed for the indirect effects, using 5000 samples. Mediation was considered to have occurred if the CI did not cross zero, with it being characterized as full mediation if the direct effect was non-significant or partial mediation if the direct effect was significant. All assumptions for conducting a mediation analysis were met.

Results

Participants in the MS group had mild to moderate levels of disability on the PDDS (Mdn = 2, interquartile range (IQR) = 4), with 69.2% having relapsing–remitting MS. Compared to the HC (Table 1), the MS group was younger and had fewer years of education, but they did not significantly differ in terms of sex, race, or comorbidity. As expected, the MS group had a higher WM lesion volume than the HC group (7.851 vs 2.494, p < 0.001). The MS group also had lower mood, conscientiousness, mental and physical well-being, left and right thalamus volume, and right amygdala volume, but similar levels of extraversion, left amygdala volume, and intracranial volume (Table 1). Nearly 24% of the total sample (n = 50) endorsed a history of depression, with the MS group having a significantly greater depression history than the HC group (30.841% vs 16.346%, p = 0.013).

Table 1:

Participants’ demographics, MRI metrics, and affective measure scores.

Variables	Group				p-value
	MS (n = 107)		Control (n = 104)
	Median	IQR	Median	IQR
Age (Years)	64.000	7.000	67.000	10.750	.001
Education (Years)	16.000	5.000	16.000	2.000	.002
Sex (% Female)	69.159%	--	67.308%	--	.773
Race (% non-Hispanic White)	83.178%	--	72.115%	--	.054
MS Phenotype (% RRMS)	69.159%	--	--	--	--
MS Duration (Years)	21.000	18.000	--	--	--
Patient Determined Disease Steps	2.000	4.000	--	--	--
Global Health Score	1.000	2.000	1.000	1.000	.751
History of Depression (% Yes)	30.841%	--	16.346%	--	.013
WM Lesion Volume	7.851	13.651	2.494	4.251	<.001
Left Thalamus Volume	5,864.700	992.700	6,357.650	921.500	<.001
Right Thalamus Volume	5,626.800	1,109.900	6,126.800	1,052.100	<.001
Left Amygdala Volume	1,326.900	228.500	1,369.050	359.500	.069
Right Amygdala Volume	1,460.500	301.200	1,555.650	366.000	.005
Intracranial Volume	1,301.038	182.791	1,334.003	206.277	.317
Geriatric Depression Scale *	24.000	7.000	28.000	4.000	<.001
BFI-Conscientiousness *	12.000	4.000	14.000	4.000	.011
BFI-Extraversion *	10.000	4.000	10.500	3.750	.131
MCS-12 *	50.552	18.000	56.384	9.740	.005
PCS-12 *	40.013	17.280	53.381	8.640	<.001

BFI: Big Five Inventory; IQR: interquartile range; MCS-12: Short Form Health Survey-12 Mental Composite Summary; MS: multiple sclerosis; PCS-12: Short Form Health Survey-12 Physical Composite Summary; WM: white matter

^*Higher scores are indicative of better functioning

The final AR derivation linear regression model (Table 2), which retained age, GHS, intracranial volume, left thalamus volume, and WM lesion volume, explained 19.5% of the variance in the composite GDS, BFI Conscientiousness, and BFI Extraversion variable (F(5,205) = 11.165, p < 0.001). Correlations between the individual measures in the AR derivation composite measure and MRI metrics are provided in Supplementary Table 1. Participants who reported a history of depression had significantly lower AR scores than those who did not (−0.486 ± 1.031 vs 0.151 ± 0.927; t(209) = 4.129, p < 0.001). AR was positively associated with both MCS-12 (b = 4.390, 95% CI: 3.206–5.575, p < 0.001) and PCS-12 (b = 3.556, 95% CI: 2.134–4.978, p < 0.001). The relationships remained significant after adjusting the models for covariates (Table 3).

Table 2:

Linear regression deriving affective reserve (n = 211).

Variables	b	SE	95% CI		p-value
			Lower Limit	Upper Limit
Intercept	47.404	8.283	31.074	63.735	<.001
WM Lesion Volume	−0.183	0.057	−0.294	−0.071	.001
Left Thalamus Volume	0.002	0.001	0.001	0.004	.007
Intracranial Volume	−0.016	0.004	−0.024	−0.008	<.001
Age	0.157	0.082	−0.004	0.318	.056
Global Health Score	−1.534	0.464	−2.448	−0.620	.001

WM: white matter

Table 3:

Linear regressions examining the associations between affective reserve and subjective well-being (n = 211). Unstandardized coefficients (b) are reported.

Model	Variable	MCS-12					PCS-12
		b	SE	95% CI		p-value	b	SE	95% CI		p-value
				LL	UL				LL	UL
Unadjusted	Intercept	51.008	0.592	49.841	52.176	<.001	45.247	0.711	43.846	46.648	<.001
	AR	4.390	0.601	3.206	5.575	<.001	3.556	0.721	2.134	4.978	<.001
Adjusted	Intercept	33.682	7.187	19.511	47.853	<.001	37.002	7.589	22.039	51.965	<.001
	AR	4.179	0.598	3.000	5.359	<.001	2.834	0.632	1.588	4.080	<.001
	Group	1.799	1.258	−0.681	4.278	.154	9.437	1.328	6.819	12.056	<.001
	Age	0.175	0.101	−0.024	0.374	.085	0.014	0.107	−0.196	0.225	.894
	Sex	0.297	1.310	−2.286	2.879	.821	0.168	1.383	−2.559	2.895	.903
	Education	0.262	0.250	−0.231	0.754	.296	−0.236	0.264	−0.756	0.284	.372
	GHS	−1.033	0.548	−2.113	0.047	.061	−2.488	0.578	−3.629	−1.348	<.001

AR: affective reserve; GHS: Global Health Score; LL: lower limit; MCS-12: Short Form Health Survey-12 Mental Composite Summary; PCS-12: Short Form Health Survey-12 Physical Composite Summary; UL: upper limit

AR partially mediated the relationship between group status and MCS-12 in the unadjusted model (indirect: b = 1.342, 95% CI: 0.249–2.561; Figure 3(a)). However, after adjusting for covariates, AR fully mediated the relationship (indirect: b = 1.350, 95% CI: 0.142–2.591; Figure 3(b)). AR partially mediated the relationship between group status and PCS-12 in both the unadjusted (indirect: b = 0.892, 95% CI: 0.161–1.864; Figure 4(a)) and adjusted (indirect: b = 0.915, 95% CI: 0.073–1.906; Figure 4(b)) models.

Figure 3:

Mediation model of the effects of group status on mental well-being (MCS-12), with AR as the mediator (n = 211). Both unadjusted (A) and adjusted (B) models were conducted, with age, sex, education, and GHS included as covariates in the latter model.

AR: affective reserve; GHS: Global Health Score; HC: healthy controls; MCS-12: Short Form Health Survey-12 Mental Composite Summary; MS: multiple sclerosis

* p < .05, ** p < .001

Figure 4:

Mediation model of the effects of group status on physical well-being (PCS-12), with AR as the mediator (n = 211). Both unadjusted (A) and adjusted (B) models were conducted, with age, sex, education, and GHS included as covariates in the latter model.

AR: affective reserve; GHS: Global Health Score; HC: healthy controls; MS: multiple sclerosis; PCS-12: Short Form Health Survey-12 Physical Composite Summary;

* p < .05, ** p < .001

Discussion

This study is the first to operationalize AR, expanding upon previous work that provided a theoretical framework for the concept.^4,5 As hypothesized, the residual-derived AR metric was found to be positively associated with both mental and physical aspects of subjective well-being, with lower AR scores observed in individuals with a history of depression. In addition, AR fully mediated the relationship between having MS and mental well-being, with the presence of MS having no direct effect on mental well-being after adjusting for age, sex, education, and comorbidities. This suggests that while the MS cohort has lower MCS-12 scores than the HC, the disease itself does not independently predict lower emotional functioning. Rather, AR acts as an intermediary, with higher levels of reserve serving as a protective factor. These results are consistent with the conceptualization that higher AR increases the brain’s ability to adapt to stressful events and thus decreases the odds that symptoms will develop.⁴

AR partially mediated the relationship between physical well-being and the presence of MS. In contrast to the other mediation model, while AR partially buffers the effects of disease, MS itself has a direct impact on perceived physical functioning. This is broadly consistent with the literature, in which depression and physical disability were independently associated with physical well-being.³¹ Considering that having any mental health disorder has been associated with consistently low physical well-being in MS,³² the role of AR as a partial mediator fits conceptually. Furthermore, higher AR may contribute to certain behaviors that promote higher physical well-being. For example, higher resilience, a potential outcome of AR, is associated with greater physical activity in people with MS,³³ which in turn is related to better physical and mental functioning.³⁴

These findings provide the initial evidence for a quantifiable way to measure AR, which was operationalized in a mixed sample that included normal aging. By using a residual-derived approach, not only were the different processes involved in AR captured⁵ (Figure 1), but this type of metric is sensitive to change⁸ and thus can be evaluated as a modifiable factor.⁴ As with other residual-derived reserve metrics,⁶ this method of measuring AR considers the effects of sociodemographics and brain pathology on the individual level. Furthermore, the residual-derived AR can be computed using routine MRI scans, facilitating replication in future studies.

The next steps will be to validate this residual-derived AR metric in other samples, including younger people with MS, and examine its contribution in the broader concept of individual reserve along with CR and PR.⁶ As this paper examined AR as a mediator between group status and subjective well-being, future studies can further explore AR as a moderator between MS-specific biopsychosocial adversity factors (e.g., phenotype and level of disability) and emotional distress. In addition, considering that elevated depressive symptomatology can have small negative effects on several cognitive domains,³⁵ evaluating the effects of AR on MS-related cognitive changes should be considered.

While this study has several strengths, including the inclusion of both MS and HC, broader outcome measures that are applicable to multiple populations, and a robust sample size, there are limitations to consider. There has been limited research on AR, and as such there is no consensus on the outcome measures to use.⁴ While the GDS-30 was selected to assess depressive symptomatology due to the study’s focus on older adults, future studies should evaluate whether other depression and personality measures utilized in MS research can be considered when deriving AR, particularly in younger adults. In addition, the MS group had relatively low levels of disability. As such, these findings should be replicated not only in younger people with MS, but also in individuals with greater disability. Although the SF-12 was not used to derive AR, there is still a potential concern about overlaps with the proxy composite variable as it contains items assessing mood, social engagement, and level of care in completing activities.²⁹ Furthermore, despite compelling mediation analyses, given the cross-sectional nature of the study, causality cannot be determined. As such, longitudinal studies will be crucial for determining whether AR has protective effects.

A residual-based approach offers researchers a way to measure AR that captures within-person differences. AR fully mediated the relationship between the presence of MS and mental well-being, which is congruent with the theoretical conceptualization of AR as a protective factor that may explain discrepancies in how individuals respond to the presence of disease and adverse events.

Supplementary Material

Supplemental material

Supplemental material for this article is available online.

Data availability statement

Data will be provided to qualified investigators upon written request to the senior author (roee.holtzer@einsteinmed.edu).*