Longitudinal Changes in White Matter Hypointensities in Recurrent Late-Life Depression

Abstract

Background:

White matter hypointensities (WMhs) on T1 and white matter hyperintensities (WMH) on T2-FLAIR are highly correlated and indicate elevated vascular risk in late-life depression (LLD). While larger baseline WMH are associated with depression recurrence, the literature is sparse regarding the longitudinal accumulation of WMh and its association with relapse in older adults.

Methods:

We investigated the relationship between WMh and LLD recurrence over a 2-year multi-site study. T1 images were collected every 8 months from N = 223 participants, 157 with remitted LLD and 66 comparison participants (mean age 67.5 ± 4.92 years; 67.3% female). A total of 52% of participants in remission experienced relapse. WMh accumulation rates were analyzed across groups defined on relapse status using mixed effects modeling. We extracted individual measures of WMh and divided participants into four groups: baseline WMh load (low, high) x WMh accumulation (slow, rapid) to estimate differences in relapse risk using a Cox model.

Results:

The relapse groups were associated with greater baseline WMh than the non-depressed comparison group. WMh accumulation did not differ between non-depressed and remitted LLD participants (β = 0.009, 95% CI: −0.04 to 0.06). Compared to individuals with low baseline WMh and slow accumulation, individuals with high baseline WMh and rapid accumulation had a higher risk of relapse (HR = 2.92, 95% CI: 1.31–6.47).

Conclusion:

Greater baseline WMh was associated with relapse after remission in LLD, but the rate of WMh progression was not. Patients with high baseline and rapidly accumulating WMh may warrant closer clinical monitoring due to increased risk of depression relapse.

INTRODUCTION

Late-life depression (LLD) is associated with significant disability, comorbid medical conditions, elevated risk of suicide, and cognitive impairment and mortality.^1–7 These poor outcomes are commonly associated with the recurrent nature of depression, characterized by repeated episodes throughout a patient’s lifetime.^8,9 Growing evidence suggests that structural brain changes, particularly in white matter, play a critical role in the pathophysiology of LLD.^10–13

Several cross-sectional studies found that older adults with a history of depression have higher white matter hyperintensity (WMH) volumes compared to never-depressed older adults.^11–14 Individuals with higher WMH burden tend to have worse LLD disease courses, including more brittle response to treatment and higher relapse rates.^1,15,16 Others reported that severity of WMH has been associated with poor antidepressant treatment response.¹⁷

Fewer longitudinal studies have examined how WMH accumulation may influence the disease course in LLD. Some studies found that faster WMH progression is associated with worse LLD outcomes, that is, relapse or non-remission,^18,19 and worse depression severity over time.^20–22 Chen et al. reported that while depressed participants had larger cross-sectional white matter lesion volumes than nondepressed participants, there were no significant differences in the rates of change of the longitudinal trajectories across groups.

WMH have been extensively studied as markers of cerebral small vessel disease (cSVD) and their relationship to LLD.^23,24 These hyperintensities on T2-weighted MRI can also similarly appear on T1-weighted MRI as white matter hypointensities (WMh) and have been found to be strongly correlated,²⁵ allowing for the tracking of these lesions in larger samples, as only the far more ubiquitous T1-weighted sequences are needed for their assessment.

We obtained WMh estimates every 8 months over the course of a 2-year study in a sample that included both non-depressed older adults (healthy comparison participants [HC]) and individuals recently remitted from a major depressive episode in late life (remLLD). Participants were followed for 2 years to prospectively identify relapse of a new depressive episode. We hypothesized that remLLD would have greater WMh than HC and a faster rate of WMh accumulation. We hypothesized that those who relapsed (REL) would have greater rates of WMh accumulation compared to both HC and those in stable remission (REM).

METHODS

Participants and Study Design

REMBRANDT (Recurrence Markers, Cognitive Burden and Neurobiological Homeostasis in Late-Life Depression)^9,26,27 is an ongoing 2-year observational longitudinal study of remitted LLD conducted at three sites: the University of Illinois Chicago (UIC; Chicago, IL), the University of Pittsburgh (Pitt; Pittsburgh, PA), and Vanderbilt University Medical Center (VUMC; Nashville, TN). The study design has been described in Taylor et al.

In brief, 2 sets of participants with LLD were recruited. (1) Participants with remitted LLD who were ≥60 years and met criteria for a DSM5 diagnosis of major depressive disorder, recurrent, in partial or full remission, with the last documented depressive episode occurring no more than 4 months prior to study recruitment, and a Montgomery-Asberg Depression Rating Scale (MADRS)²⁸ ≤10 at initial evaluation. (2) Currently depressed older (≥60 years) adults who met criteria for a DSM5 diagnosis of major depressive disorder, recurrent, with a MADRS score ≥15. These participants were treated to remission through an initial treatment program (ITP) for up to 20 weeks. Broadly, the ITP involved antidepressant medication treatment that is closely monitored for symptom improvement and adjusted using an algorithm informed by the STAR*D²⁹ and Duke Neurocognitive Outcomes of Depression in the Elderly studies.³⁰ Depressed participants who did not remit during the ITP were not enrolled in the study. Healthy comparison participants were recruited with a MADRS score of ≤8 and without any lifetime history of psychiatric disorders or psychotropic medication use.

All participants underwent a screening process before study initiation and were included based on the following criteria: age ≥60 years; English fluency, and a Montreal Cognitive Assessment (MoCA) score ≥24³¹ or a MoCA-BLIND score ≥18.³² Exclusion criteria included acute suicidal ideation within 3 months of beginning the study; acute grief (<1 month); a history of developmental disorders or IQ below 70; current or past psychosis; and primary neurological disorders including major neurocognitive disorders related to dementia, Parkinson’s Disease, stroke, and epilepsy. Mild cognitive impairment was not exclusionary.

The REMBRANDT study was approved by the Vanderbilt single Institutional Review Board, and we obtained written informed consent for participants prior to any study procedures. Participants were recruited through various means, including clinical referrals, community outreach, and advertisements. The REMBRANDT study was powered to detect group differences for primary measures specified in the grant, not including longitudinal WMh volumes. Power calculations are included verbatim in the supplement.

Longitudinal Phase Design

Participants were followed for up to 2 years and assessed at least every 2-months for depression severity using MADRS. RemLLD participants continued the treatment plan that led to remission. Changes to these plans were allowed as clinically indicated (e.g., adjustments in medications and/or doses) by the psychiatry providers. Participants were followed after relapse until the completion of the longitudinal phase. Relapse was assessed using the MADRS, defined using a score of 15 or greater over a duration of at least 2 weeks, and required both a research specialist and an MD. Note that we do not distinguish between the terms relapse and recurrence. Both describe any new depressive episode during the study period for remLLD participants. Further, none of the HC participants developed LLD during the study course.

Group definitions

All participants were defined as either HC or remLLD. Within remLLD, we considered those with sustained remission who did not relapse (REM) separately from those who relapsed in the study (REL). We parsed REL on the time to first relapse, defining early relapsers (EarlyREL) as those who first relapsed within 250 days of baseline and late relapsers (LateREL) as those who first relapsed after 250 days. The distinction between early and late relapsers was made based on previous results, finding that early relapsers have a larger burden of white matter lesions at baseline compared to those who relapsed later or did not relapse.¹⁴ Additionally, the 250-day mark roughly corresponded to the first follow-up appointment after 8 months.

Clinical Assessments

We used several clinical assessments as part of this study, including the Mini-International Neuropsychiatric Interview (MINI) for diagnostic assessments³³ at baseline, MoCA and MoCA-BLIND for cognitive screening, and MADRS to assess depression severity. For a complete description of the assessments used, see Taylor et al.

We used the sum of the endocrine/breast, vascular, and heart components of the Cumulative Illness Rating Scale-Geriatric (CIRS-G)³⁴ for medical burden assessment. CIRS-G provides a simple measure to quantify overall disease burden in late-life^34,35 and these components can indicate known risk factors for LLD including cardiovascular disorders, diabetes, obesity, and thyroid illness.

MRI Acquisition

Participants underwent MR scanning on a 3T MRI with a 32-channel head coil on Siemens Prisma (PITT), Philips Elition (VUMC), and GE Discovery MR750 (UIC), following the Adolescent Brain Cognitive Development (ABCD) Study MRI Protocol.³⁶ MRIs were acquired at baseline and at approximately months 8, 16, and 24. See Taylor et al. for full details on the one-hour protocol. A sagittal, whole-brain T1-weighted magnetization prepared rapid gradient echo (MPRAGE) was collected with repetition time (TR) = 2400 ms (2500 ms PITT), echo time (TE) = 2.9 ms, flip angle (FA) = 8 deg, field of view (FOV)=256 × 256 (256 × 240 VUMC) with 176 slices (225 slices VUMC), 1 mm³ isotropic resolution, and GeneRalized Autocalibrating Partial Parallel Acquisition (GRAPPA) with acceleration factor of 2 (total time 5.63–6.65 min). An axial, whole-brain fluid attenuated inversion recovery (FLAIR) sequence was collected with repetition time (TR) = 4800 ms, echo time (TE) = 445 ms (113 ms UIC), inversion time (TI) = 1650 ms (1464 ms UIC), field of view (FOV) = 256 × 256 with 176 slices, 1 mm³ isotropic resolution, and GRAPPA with acceleration factor of 3 (total time 5.96–6.5 min). Supports were given to participants for their head, back, or legs as needed. We administered monthly scanning of ADNI (Alzheimer’s Disease Neuroimaging Initiative) structural phantom that included both MPRAGE and FLAIR. The signal-to-noise ratio (SNR) was computed across sites and was within an expected tolerance (not reported).

Structural Processing

Baseline WMH volumes were assessed to determine reliability with WMh but were not a focus of this analysis. WMH volumes were segmented from T2-FLAIR sequences taken at baseline and the Lesion Segmentation Toolbox.³⁷ Analyses were conducted through SPM12 using a 0.3 threshold set by visual inspection across several participants. This toolbox co-registers the T1 image after bias correction, segmenting white matter, gray matter, and cerebrospinal fluid. After an initial identification of WMH through outlier detection, a lesion-growing algorithm is used to enlarge this lesion using a Markov random field and estimate WMH volume. We also created intracranial volume (ICV) masks by thresholding gray/white matter and cerebrospinal fluid probability maps by 0.1 and smoothing to obtain an estimate for baseline ICV.

To estimate WMh volumes, we used T1-weighted images that were available from all visits (baseline, 8, 16, and 24 months). We compared baseline WMH to baseline WMh. We used Freesurfer v6.0 to segment each T1-weighted image^38–40 to extract estimates of total WMh volume. This was done with the ‘recon-all’ stream and visually checked for accuracy.⁴¹ Briefly, this pipeline conducts automated transformation to a standardized space, removes non-brain tissue,⁴² and the gray/white matter boundary was separated with tessellation.⁴³ Hypointensities are identified using spatial intensity gradients across tissue classes.^44,45 This approach results in an automated segmentation of hypointense areas on the T1-image.

Statistical Analysis

To assess whether WMh volumes (from T1-MPRAGE) could serve as a proxy for WMH burden (from T2-FLAIR) in a longitudinal analysis, we investigated the relationship between baseline WMH and WMh volumes using a Pearson correlation, and performed two-tailed t-tests to examine group differences in baseline WMh and vascular disease burden. Baseline WMh comparisons were made between HC and each of the three LLD groups (REM, EarlyREL and LateREL). A Holm-Bonferroni correction was applied to the resulting p-values to correct for multiple comparisons. Similar t-tests were conducted comparing the vascular disease burden using the composite CIRS-G variable across groups. All t-tests were performed using a significance threshold of α = 0.05.

Next, we used linear mixed effects models with fixed slope and random intercept to compare the average WMh volume over time across groups. Models with fixed slope and random intercept were chosen because of the low within-subject variability in the change in WMh volume over time. In total, three mixed effects models were examined in our analyses: (1) HC vs. remLLD; (2) HC vs. REM vs. REL; (3) HC vs. REM vs. EarlyREL vs LateREL. Each model adjusted for baseline age and ICV, sex, race, education, study site, and disease burden using the composite CIRS-G variable. Mixed effects models were created using the “lme4” package in R (version 4.4.1). Fixed effects were evaluated using 2-tailed tests based on t statistics with Satterthwaite degrees of freedom, implemented using the ‘lmerTest’ package.

To further investigate the change in WMh volumes, linear regression was used in an exploratory slope analysis to determine if WMh progression was affected by group when a shared slope effect was not assumed. Regression models adjusted for baseline age, baseline ICV, baseline WMh volume, sex, race, education, and study site, and were implemented using the ‘lm’ function from the base ‘stats’ package in R (4.4.1). All hypothesis tests were two-tailed. To find progression rates, we fit linear models for WMh as functions of time for each participant and extracted model coefficients; the slope was used as the dependent variable in this exploratory slope analysis.

Lastly, a survival analysis was conducted using a Cox proportional hazards model to determine the effect of both baseline and longitudinal change in WMh volume on relapse risk. The change in WMh was measured using the slope of the linear regression models, described previously. A total of 4 groups were used in the analysis, defined on the median baseline WMh and median rate of change in WMh volume. Groups included those with high baseline WMh volume and high slope, high baseline WMh volume and low slope, low baseline WMh volume and high slope, and low baseline WMh volume and low slope. Because baseline WMh and WMh progression are related measures, we grouped participants in this way to test clinically meaningful patient profiles rather than the independent linear contribution of each variable to relapse risk. The resulting hazard ratios are thus easily interpretable as relative relapse risks and allow for clear comparisons between WMh trajectory patterns to be made. The analysis was implemented using the “coxph” function in the “survival” package in R (4.4.1), with statistical significance evaluated using 2-tailed Wald tests.

Any observations that contained missing data were excluded. The final mixed effects models included N=202 participants with 637 total observations (HC: 58, REM: 68, EarlyREL: 40, LateREL: 36). The exploratory slope analysis included data from N = 192 participants (HC: 62, REM: 57, EarlyREL: 36, LateREL: 37). The survival analysis was performed on N=118 remitted participants (REM: 51, EarlyREL: 32, LateREL: 35), of which 29 had low baseline WMh and low slopes, 33 had low baseline WMh and high slopes, 29 had high baseline WMh and low slopes, and 27 had both high baseline WMh and high slopes.

Partial eta squared $\left({\eta }_p^2\right)$ values were reported to quantify the effect size of the time × group interactions. Commonly, ${\eta }_p^2=0.01$ is considered a small effect, ${\eta }_p^2=0.06$ is a moderate effect, and ${\eta }_p^2=0.14$ is a large effect.⁴⁶

RESULTS

Demographic Characteristics

In total, 223 participants were included. The mean age of participants at baseline was 67.5 ± 4.92 years, and 67.3% were women. There were no differences between diagnostic groups in age at baseline or sex. MoCA scores were similar across groups. Race composition significantly differed between diagnostic groups. We report sample characteristics in Table 1. Among remLLD, 81 participants (51.6%) relapsed, while 76 participants (48.4%) maintained remission during observation. 43 participants (27.4%) experienced an early relapse (within 250 days of baseline), whereas 38 participants (24.2%) experienced a late relapse (occurring after 250 days).

TABLE 1.

Demographic characteristics of participants.

	HC (N = 66)	remLLD (N = 157)	Overall (N = 223)	Test Statistic P-value
Baseline Age, Years	67.8 (5.25)	67.4 (4.77)	67.5 (4.92)
Mean (SD)				t(112) =0.48
Median [Min, Max]	67 [60, 83]	67 [60, 80]	67 [60, 83]	P = 0.63
Age of Onset, Years
Mean (SD)	-	29.2(15.5)	29.2(15.5)	-
Median [Min, Max]	-	25.0 [4, 76]	25.0 [4, 76]
Missing	66 (100%)	28 (17.8%)	94 (42.2%)
Sex	25 (37.9%)	48 (30.6%)	73 (32.7%)
Male (0)				χ2(1) =0.82
Female (1)	41 (62.1%)	109 (69.4%)	150(67.3%)	P = 0.37
Race	16 (24.2%)	19(12.1%)	35(15.7%)
Non-White (0)	50 (75.8%)	138 (87.9%)	188 (84.3%)	χ2(1) = 4.30
White (1)				P = 0.04
Education		12.9 (0.532)	12.9 (0.457)
Mean (SD)	13.0 (0.129)	13.0 [8, 14]	13.0 [8, 14]	t(190)= 1.34
Median [Min, Max]	13.0 [12, 13]	4 (2.5%)	10 (4.5%)	P = 0.18
Missing	6 (9.1%)
Study Site	26 (39.4%)	76 (48.4%)	102 (45.7%)
VUMC				χ2(2) =2.05
Pitt	19 (28.8%)	44 (28.0%)	63 (28.3%)	P = 0.36
UIC	21 (31.8%)	37 (23.6%)	58 (26.0%)
Relapse Status	66(100%)	76 (48.4%)	142 (63.7%)
Did not relapse
Early Relapse	0 (0%)	43 (27.4%)	43 (19.3%)	-
Late Relapse	0 (0%)	38 (24.2%)	38 (17.0%)
Number of Relapses
Mean (SD)	-	2.04(1.45)	2.04(1.45)	-
Median [Min, Max]	-	2 [1, 8]	2 [1,8]
Missing	66 (100%)	76 (48.4%)	142 (63.7%)
WMh, log10(cm3)	0.279 (0.260)	0.390 (0.286)	0.357 (0.283)
Mean (SD)				t(132) =−2.79
Median [Min, Max]	0.243 [−0.133, 1.02]	0.351 [−0.236, 1.20]	0.313 [−0.236, 1.20]	P = 0.006
Missing	1 (1.5%)	4(2.5%)	5 (2.2%)
ICV, cm3	1.47e6 (1.69e5)	1.47e6 (1.60e5)	1.47e6 (1.62e5)
Mean (SD)				t(116) = −0.18
Median [Min, Max]	1.46e6 [1.18e6, 1.9e6]	1.45e6 [1.13e6, 1.95e6]	1.46e6 [1.13e6, 1.95e6]	P = 0.85
Missing	1 (1.5%)	6 (3.8%)	7(3.1%)
CIRS-G – H+E+V	1.51 (1.36)	2.64 (1.71)	2.32 (1.70)
Mean (SD)	1 [0, 6]	2 [0, 7]	2 [0, 7]	t(128) = −4.92
Median [Min, Max]	9(13.6%)	12(7.6%)	21 (9.4%)	P = 2.58e-6
Missing
MoCA	26.9(2.26)	26.8(2.15)	26.9 (2.18)
Mean (SD)	27 [22, 30]	27 [22, 30]	27 [22, 30]	t(117) = 0.10
Median [Min, Max]	0 (0%)	1 (0.6%)	1 (0.4%)	P = 0.92
Missing

Abbreviations: HC, non-depressed comparison participants, remLLD, participants with remitted late-life depression; SD, standard deviation; Age of onset, age in years of first depressive episode; VUMC, Vanderbilt University Medical Center; Pitt, University of Pittsburgh; UIC; University of Illinois Chicago; WMh, baseline white matter hypointensity volume; ICV, baseline intracranial volume; CIRS-G — H + E + V, the sum of the Cumulative Illness Rating Scale-Geriatric (CIRS-G) heart, endocrine/metabolic and breast, and vascular scores; Pearson’s Chi-squared test and Welch’s t-test were used to compute test statistics and P-values.

3.2. White matter hypointensies at baseline and follow-up visits

Baseline WMh volume (from T1-weighted image) and WMH volume (from FLAIR) are highly correlated (r = 0.88, see Figure S1 in Supplement), which indicates a strong positive linear relationship between these measurements. When the two outliers with high WMH volumes (>25) were removed, the strong correlation between measurements was preserved (r = 0.90).

Baseline WMh differences were found between groups: EarlyREL had larger initial WMh volume than any other group examined (Figure 1, Supplementary figure S2). Both EarlyREL and LateREL had statistically greater baseline WMh values than HC (EarlyREL: t(92.98) = 3.42, p = 0.0009, p_adj = 0.0028; LateREL: t(70.44) = 2.29, p = 0.0249, p_adj = 0.0498). The REM group did not differ from HC in baseline WMh (t(136.75) = 1.43, p = 0.1544, p_adj = 0.1544).

FIGURE. 1.

Mixed effects model predictions. (A), (B): Results of the model comparing HC vs. remLLD. (C), (D): Results comparing HC vs. REM vs. EarlyREL vs. LateREL. EarlyREL and LateREL represent individuals that relapse within 250 days of baseline and after 250 days of baseline, respectively. Both models adjusted for time (days since baseline), vascular disease burden using CIRS-G, age at baseline, ICV at baseline, sex, education, race, study site, group, and time * group effects.

We found no evidence of differences in WMh progression across groups based on depression status (Tables S1-S4 in Supplemental Material), where time × group effects were not significant. These results are consistent across all examined groups [HC vs. remLLD: ${\eta }_p^2<0.01$; HC vs. REM vs. REL: ${\eta }_p^2<0.01$; HC vs. REM vs. EarlyREL vs. LateREL: ${\eta }_p^2<0.01$]. Group effects were significant in all mixed effects models. WMh volumes were larger for remLLD participants compared to HC regardless of relapse status (Table S1), and EarlyREL had larger WMh volumes than HC (Table S3) and both REM and LateREL groups (Table S4). WMh volumes increased with time (Figure 1) for all examined groups.

When comparing differences in the summed CIRS-G variable across groups (Supplemental Figure S3), we found that comparison participants had lower cumulative CIRS-G heart, endocrine, and vascular scores than remLLD participants (t(127.94) = −4.92, p < 0.001). Exploratory pairwise comparisons were made between REM, EarlyREL, and LateREL groups, which resulted in no evidence that CIRS-G differed within remLLD participants based on relapse status.

3.3. Exploratory Slope Analysis

Participants with higher baseline WMh measurements had statistically slower WMh progression during the observation period (Supplementary Table S5). Group effects were not significant. Baseline age and ICV were positively associated with WMh progression. The longitudinal data for each participant is displayed with average group trends in the supplemental figure S4. The slope distributions are compared in the supplemental figure S5, illustrating that there is little variance in the progression of WMh across groups.

3.4. Survival Analysis

We fit individual slopes for each participant and divided participants into four equal groups: baseline WMh load (low vs. high WMh) × rate of WMh accumulation or slope (slow vs. fast WMh accumulation). The Cox model showed evidence that remLLD who have high baseline WMh volumes and high rates of WMh progression are at greatest risk of relapse (Figure 2). Compared to the reference group (low WMh, low slope), the hazard of relapse was significantly higher for participants with high baseline WMh and high slope. Additionally, higher levels of education were associated with a lower risk of relapse, independent of all other covariates. See Table 2.

FIGURE. 2.

Differences in time to relapse based on median split in baseline WMh volume and rate of change in WMh volume over time. We illustrate the probability of staying in remission with LLD over time (N=130, 73 relapse events) used in the survival analysis.

TABLE 2.

Results of the Cox model comparing time to relapse across four groups based on dichotomized baseline WMh volumes and the slopes or rates of change in WMh volume over time (above or below their respective medians). The reference group consisted of remitted LLD participants with low baseline WMh volumes and low slope. P-values were derived from Wald tests (df = 1). Bolded groups represent those that significantly affect time to relapse (P < 0.05). Model fit: Likelihood ratio test, χ²(11) = 23.18, P = 0.02.

Variable	HR	95% CI	P-value
Age at baseline	0.97	0.92–1.03	0.37
Sex - Female	1.58	0.74–3.36	0.24
Race - White	1.12	0.43–2.89	0.82
Site (VUMC ref.)	1	-	-
PITT	0.69	0.35–1.37	0.29
UIC	0.62	0.26–1.49	0.29
Baseline ICV	1.16	0.80–1.69	0.43
Education	0.53	0.35–0.80	0.003
CIRS-G - Heart + Endo + Vasc	0.95	0.81–1.11	0.52
Group (Low WMh/Low slope Ref.)	1.00	-	-
Low WMh/High slope	0.81	0.37–1.78	0.60
High WMh/Low slope	1.14	0.50–2.59	0.75
High WMh/High slope	2.92	1.31–6.47	0.009

Abbreviations: HR, hazard ratio; CI, confidence interval; ICV, intracranial volume; CIRS-G - Heart + Endo + Vasc, the sum of the Cumulative Illness Rating Scale-Geriatric (CIRS-G) heart, endocrine/ metabolic and breast, and vascular scores; WMh, white matter hypo- intensity volume at baseline.

DISCUSSION

In this study, we examined the association between the progression of WMh and depression relapse in older adults. We found that WMh volume was significantly greater in remLLD than HC, and those who relapsed early showed the highest WMh baseline volume compared to all other groups, consistent with our prior results.¹⁴ Notably, we found no evidence that remLLD nor REL had more rapid WMh accumulation than HC. In our survival analysis, participants that had high baseline WMh and high rates of WMh progression had greater risk of depression recurrence.

We report that WMh and WMH are highly correlated in our cohort of older adults, which replicates past studies,²⁵ indicating that WMh can be used as a reliable measure of white matter burden without the need for T2-FLAIR. Understanding the relationship between WMH and WMh volumes in older adults could make such work more accessible while allowing for the retrospective analyses of WMh as a marker of depression in T1 imaging studies. Our results support the vascular depression hypothesis²³ in some individuals with LLD showing heightened WMh as well as higher CIRS-G heart, vascular, and endocrine/breast scores in remLLD compared to HC.⁴⁷ Participants with remitted LLD have larger WMh volumes compared with HC consistently across time. Of the groups examined, early relapsers showed the largest baseline WMh volumes compared to HC. A similar result was found by Chen et al., showing that there were significant baseline differences in white matter lesion volume between depressed and comparison groups for those with late-onset depression.

We did not find evidence that remLLD or relapse was associated with greater rates of WMh progression over time. This is contrary to previous results^18,19,22 that showed an association between greater WMH accumulation and worse LLD disease courses (i.e., non-remittance or depression relapse). Note that the populations examined in these previous longitudinal studies differed from our current work. The samples used in Taylor et al. and Khalaf et al. consisted of 133 and 47 actively depressed older adults, respectively. Neither included participants with no history of depression for comparison, and instead compared non-remitted or relapsed groups (i.e., “poor outcome” groups) to participants that remitted during the studies. They each concluded that WMH accumulate at a faster rate among those who did not remit or relapsed after remitting. Nebes et al., with a total of 13 never-depressed participants, of which 1 developed late-onset LLD during the 3-year follow-up period, reported that the depressed participant’s WMH volume almost doubled during the study course and exhibited faster WMH accumulation than the 12 comparison participants.

Additionally, current clinical guidelines for aggressive and timely treatment of vascular risk factors (e.g. hypertension, cardiac disease, diabetes mellitus) may have positively influenced the overall cerebrovascular burden in the current cohort, although we do not have any direct comparison studies at this time to support this statement. Several prior studies suggest that antihypertensive drugs can moderately reduce WMH progression in older adults.^48–51 Recent meta-analyses using randomized trials further support the idea that intensive BP lowering effectively reduces or prevents WMH progression.^52,53 Another suggests that insulin may slow WMH growth in older adults with cognitive impairment or Alzheimer’s disease,⁵⁴ though results are mixed when considering diabetic patients with normal cognitive function.⁵⁵ Taken together, our results suggest that any progression of WMh that led to significant baseline differences between comparison and remitted LLD groups occurred prior to inclusion in our study. Thus, it is increasingly important to identify cerebrovascular and inflammatory risk before late life.

One of the most prominent models for understanding depression in older adults, the vascular depression hypothesis, suggests that cerebrovascular ischemic damage contributes to the development of depressive symptoms in a subset of individuals.^23,56 This hypothesis is further supported by studies demonstrating that WMHs are associated with treatment outcomes.^6,18 Importantly, our results remained significant after adjusting for CIRS-G, and thus it is likely that the observed group differences in WMh over time are only partially explained by differences in vascular disease burden.

In older adults, the etiopathogeny of white matter lesions is multifactorial, including small vessel ischemia and microvascular age-related damage, as well as gliosis, demyelination and inflammation.^57,58 It is well know that neuroinflammation has an impact on white matter microstructure in older adults^59,60 even though the underlying mechanisms are not entirely understood. Neuroinflammation has been shown to play a critical role in the development and progression of cSVD, which is intricately connected to LLD and cognitive dysfunction, a depression-executive dysfunction syndrome.⁶¹ A compelling model suggests that inflammatory pathways associated with depression contribute to long-term structural changes in the brain, particularly white matter lesions.^56,62–66

Our results demonstrated that participants with higher baseline WMh volumes exhibited slower WMh progression over time, a pattern that may reflect underlying mechanisms associated with more advanced stages of white matter injury. WMh on T1-weighted MRI are considered indicators of severe or long-standing injury, correlating with myelin and axonal loss in several neurological diseases.^67–69 These findings support the hypothesis that structural brain abnormalities associated with LLD may reflect accumulated vascular injury across the lifespan.^23,56 Such early cerebrovascular injuries may initiate a cascade of neurobiological changes, including nonspecific demyelination, axonal loss, higher levels of microglial activation, and chronic neuroinflammation - that ultimately alter fronto-limbic circuitry critical for mood regulation, executive function, and stress regulation and autonomic control.^23,56,70,71 These findings suggest that the pathophysiological processes contributing to depression recurrence in older adults may stem from longstanding, possibly subclinical, vascular and inflammatory damage, rather than ongoing or rapidly progressing disease within the observed study window. Further studies examining these patterns at earlier life stages may be important to identify this ‘inflection’ point where individuals experience a more rapid increase in WMh.

However, when we divided groups by baseline WMh load (low vs. high WMh) and rate of WMh accumulation (slow vs. rapid progression), we found that individuals who had high WMh at baseline with rapid WMh accumulation had nearly 3 times greater likelihood of relapse compared to those with low baseline WMh and slow WMh accumulation. This suggests that it may be critical to monitor patients with actively progressing cSVD as they could be at higher risk of depression relapse, and potential interventions may help reduce that risk.

There are some limitations to this study that should be considered. First, while we compared WMh and WMH volumes, the observed difference may, in part, be driven by variations in the underlying algorithms used to detect lesions. Second, although we adjusted for race in the statistical models, residual confounding may remain due to the differences in racial compositions between diagnostic groups. Third, DSM-5 criteria were not administered longitudinally to assess depression severity which may limit the precision in relapse classifications, and our study does not address the effect of antidepressants or other psychotropic medications on relapse risk. Fourth, this analysis focused on global WMh and WMH volumes, without accounting for their spatial distribution. While total WMH volume is often used as a marker of white matter integrity, several imaging studies consider periventricular WMHs separately from deep WMHs,^72,73 finding LLD is commonly associated with more severe deep WMHs.^74–77 Others argue periventricular, deep brain, and overall WMH burden are highly correlated measures and such distinctions are generally not identifiable.⁷⁸ Future studies should examine the spatial patterns of WMh and WMH to determine whether overlapping or distinct lesion distributions provide additional insight into associations with clinical outcomes. Fifth, although we included key vascular risk factors such as heart, endocrine, and vascular scores from the CIRS-G, other relevant contributors to vascular burden - like poorly controlled BP, high cholesterol, smoking, physical inactivity and objective measures of carotid atherosclerotic plaques - were not included due to limited/no availability of these metrics in our dataset. Finally, the REMBRANDT study was not powered on longitudinal WMh volumes, potentially limiting our ability to identify subtle associations.

This study has several notable strengths, including the prospective analysis of longitudinal WMh in a well-defined cohort of older adults with remitted LLD. Given the critical role of white matter architecture associated with mood regulation, the findings from this study could hold significant clinical relevance for older adults, especially those with remitted LLD, and emphasize the need for lifespan studies, including both midlife and late-life cohorts, to identify when in the disease course that WMh accumulation rates differ between those with and without depression. Further, targeted interventions to manage vascular burden and inflammatory profile over time may contribute to enhanced brain health and improved overall functioning in late life.

Supplementary Material

Supplementary material associated with this article can be found in the online version at https://doi.org/10.1016/j.jagp.2026.02.004.

Highlights.

What is the primary question addressed by this study?

We investigated the longitudinal relationship between white matter hypointensity (WMh) volume and depression relapse for older adults in a 2-year study on remitted late-life depression (LLD).

What is the main finding of this study?

We found that remitted depressed participants do not have faster WMh accumulation over time than controls, despite having larger WMh volumes. However, our results suggest those with both high baseline WMh and rapid rates of WMh accumulation were at greatest risk of depression relapse.

What is the meaning of the finding?

Monitoring vascular disease burden over time using WMh volume could inform targeted clinical intervention to mitigate depression relapse and reduce the risk of recurrence in late life.

DATA SHARING STATEMENT

Data is available upon request from study site principal investigators.