Early depressive symptoms and disability accrual in Multiple Sclerosis: a UK MS Register study

Understanding the associations and potential drivers of long-term disability in Multiple Sclerosis (MS) is of clinical and prognostic value. Previous data have suggested a link between depression and disability accrual in MS. We aimed to determine whether depression in early MS predicts subsequent accrual of disability. Using data from the UK MS Register, we identified individuals with and without symptoms of depression and anxiety close to disease onset. We used Cox proportional hazards regression to evaluate whether early depressive or anxiety symptoms predict subsequent physical disability worsening, measured using the Expanded Disability Status Scale (EDSS). We analysed data from 862 people with MS of whom 134 (15.5%) reached an EDSS of ≥ 6.0. Early depressive symptoms were associated with an increased risk of reaching an EDSS of 6.0 (HR 2.42, 95% CI 1.49–3.95, p < 0.001), however this effect dissipated when adjusting for baseline EDSS (HR 1.40, 95% CI 0.84–2.32, p = 0.2). These data suggest that early depressive symptoms in MS are associated with subsequent disability accrual, but are likely the result of disability rather than its cause.

www.nature.com/scientificreports/ Variable definitions. Expanded Disability Status Scale (EDSS 13 ) values were extracted from two sources: a validated online self-administered EDSS scale (the webEDSS 14 ), and clinician-determined EDSS available for a subset of the population. The primary outcome was defined as the probability of reaching EDSS ≥ 6.0. EDSS 6.0 reflects use of a unilateral walking aid, and is frequently used as an endpoint in observational studies. All other data used in this study were solely derived from participant-reported measures. Anxiety and depression were defined using the Hospital Anxiety and Depression Scale (HADS) 15 , which UKMSR participants are encouraged to complete online via the web portal at 6 monthly intervals from the time of registration. HADS is one of a battery of PRO measures regularly ascertained via the UKMSR. Anxiety and depression were dichotomised using the established cutoff of 7 15 . We classified individuals as 'anxious' or 'depressed' if one of these respective scores was > 7, i.e. 8 or higher. Only HADS scores obtained before the baseline EDSS recording or in the 6 months following the baseline EDSS recording were considered, and the HADS score closest to their baseline EDSS recording was used.
Inclusion and exclusion criteria. We defined the study population using the following inclusion criteria, all of which had to be satisfied for inclusion in the analysis: • Complete demographic data (year of birth, year of diagnosis, gender, MS type at diagnosis); • At least one 'baseline' EDSS value within 5 years of self-reported MS diagnosis date; • At least one subsequent EDSS recording more than 6 months after the baseline reading; • Baseline EDSS score < 6.0; • At least one HADS score before or within 6 months of the baseline EDSS reading.
From the entire UKMSR population (n = 23,268), we defined the analysis population for the study as those individuals with complete baseline demographic data (n = 18,401), at least one 'baseline' EDSS reading (i.e. an EDSS reading within 5 years of their stated MS diagnosis date), one or more EDSS readings at least 6 months after their baseline reading, a valid HADS score before or in the 6 months after their baseline EDSS reading, and a baseline EDSS reading of < 6.0 ( Fig. 1). Figure 1 depicts the number of individuals excluded at each stage of filtering (Fig. 1). Baseline EDSS were ascertained at a median of 2.4 years after MS diagnosis and 5.3 years after symptom onset.
Statistical analysis. We used Spearman rank tests and chi-squared tests for univariable comparisons of variables between individuals who reached EDSS 6.0 and those who did not. Unless specified, categorical variables are reported as n (%) and continuous variables as median (interquartile range). To determine the association between anxiety and depression at baseline and time to EDSS 6.0, we used Cox proportional hazards models. In the primary analysis, we adjusted for age at baseline EDSS, gender, and Primary Progressive MS (PPMS) vs other MS types. We conducted a range of secondary sensitivity analyses using the following model specifications: • Primary Analysis: age at baseline EDSS + Gender + PPMS-onset • Secondary analyses: • As for primary analysis + disease duration at time of EDSS www.nature.com/scientificreports/ • As for primary analysis + baseline EDSS • Just Relapsing-Remitting MS (RRMS), adjusted for age at baseline EDSS + Gender • Just males, adjusted for age at baseline EDSS + PPMS-onset • Just females, adjusted for age at baseline EDSS + PPMS-onset • Just participants with baseline EDSS < 4.0, covariates as per primary analysis • Age at baseline EDSS + Gender + exposure to high-efficacy disease modifying therapy (DMT) (alemtuzumab, cladribine, ocrelizumab, fingolimod, and natalizumab) prior to baseline EDSS recording. Individuals with no DMT data were considered to be unexposed to high-efficacy DMT.
We used Schoenfeld residuals to check the validity of the proportional hazards assumption, checked for unduly influential observations, and checked the linearity of the relationship between continuous covariates and disability progression.
All analyses were conducted in R version 4.1.3 within the UKMSR secure research environment (UKSERP 16 ).
Ethical approval. The  We examined the differences in demographic characteristics between our study population and the wider UKMSR population (n = 17,539) to assess the risk of collider bias. The study population was diagnosed at an older age (median 43.0 vs 38.0, p < 0.001), with a roughly similar gender split (76.5% female vs 74.5%, p = 0.2), a higher proportion of people with RRMS compared to the non-study cohort (83.1% vs 72.5%, p < 0.001), and were diagnosed more recently (median 2016 vs 2007, p < 0.001).
Anxiety and depression were common in both groups. The proportion of people with anxiety at baseline did not differ significantly between those who reached EDSS 6.0 and those who did not (46.3% vs 40.9%, p = 0.29), whereas depression at baseline was more common among those who reached EDSS 6.0 (43.3% vs 22.4%, p < 0.001).
Depression at baseline was associated with an increased hazard of reaching EDSS 6.0 (Multivariable Cox regression models adjusted for age at EDSS measurement, gender, and MS subtype; HR 2.42, 95% CI 1.49-3.95, p < 0.001; Fig. 2A). We observed similar effects in sensitivity analyses, including adjustment for disease duration, sex-stratified analyses, analyses including only participants with relapse-onset MS, and adjustment for exposure to high-efficacy disease-modifying therapy. However, adjustment for baseline EDSS diminished the effect (HR 1.40, 95% CI 0.84-2.32, p = 0.20; Fig. 2B). We did not observe a statistically significant association between anxiety at baseline and subsequent EDSS 6.0 (HR 1.49, 95% CI 0.92-2.41, p = 0.10). Table 1. Demographic characteristics and key variables of the study population are shown. Demographics are shown for individuals who reached EDSS 6.0 during the study follow-up period and those who did not. P values reflect univariable statistical comparisons between the two groups.  www.nature.com/scientificreports/

Discussion
We provide evidence from the UK MS Register for a relationship between depression and subsequent disability accrual in MS, corroborating previous findings from Swedish 11 and Canadian cohorts 17 . Importantly, this relationship dissipates almost entirely on adjustment for baseline disability. There are several plausible explanations for these data: psychiatric illness could be triggered by physical disability, could stem from a common causal origin (i.e. a higher brain lesion load may predispose to depression/ anxiety and physical disability), or could arise independently but contribute to overall disability via a separate process. Adjusting for baseline EDSS substantially diminished the signal, suggesting that depression may simply reflect more advanced disease rather than being an independent predictor of progression. An important limitation of this work is the relatively small proportion of the cohort with serial EDSS measures and HADS scores, which raises the possibility of collider bias as both increasing physical disability and depression are likely to influence an individual's probability of participating in these assessments. Furthermore, the use of self-reported EDSS measures, although broadly accurate, may be less accurate than clinician-determined EDSS. Both HADS and EDSS scores fluctuate over time, for instance during a relapse (often termed relapse-associated worsening); while the small proportion of patients with a recorded relapse near the time of HADS recording is reassuring, relapse data were only available for a minority of the cohort, and therefore this is a weakness of the present study.
Given the stringent filtering we applied from the broader cohort, it is plausible that collider bias could distort the results, for instance potentially disguising a genuine association and leading to a false negative result. Our study has focussed on a more recently-diagnosed, younger, and more heavily relapse-onset cohort than the whole UKMSR population. By definition, individuals who are able to complete the webEDSS are likely to be skewed towards individuals with less aggressive disease, more preserved upper limb and cognitive function. Furthermore, and 95% confidence intervals for the hazard of progression to EDSS 6.0 given the presence or absence of depression at baseline. The dashed line at 1 indicates the null. The x axis is on a log-10 scale. These HRs were derived from a variety of multivariable cox proportional hazards models. The y axis describes the model used. We performed a variety of sensitivity analyses including using different confounding covariates, outcome definitions, and cohort definitions. 'Primary analysis' refers to the primary analysis model, in which we adjusted for age at baseline EDSS recording, MS subtype (PPMS vs other types), and gender. www.nature.com/scientificreports/ depression itself is likely to influence both the completion of the HADS score and the webEDSS score, and so it is likely that collider biases have an impact on these findings. As participants are enrolled after being diagnosed with MS, we did not have data on premorbid psychiatric symptoms. This precludes any comment on the temporal relationship between depressive symptoms and MSrelated disability based on these data alone. For parsimony we considered the presence or absence of psychiatric comorbidity at baseline as a static predictor of future disability, however this approach does not account for fluctuations in psychiatric symptoms over time. We were unable to fully address the influence of disease-modifying therapy (DMT) or of relapses on disability accrual due to data availability. Although we do adjust for DMT exposure, these data are only available for a small subset of participants and so this analysis is unlikely to fully capture the effect of treatment.
In this study, we replicate the previously-observed association between depressive symptoms and subsequent disability progression. Our results raise the suspicion that depression is not a causal risk factor for future disease progression, but a reflection of prior disease activity and present severity. Although these data do not support a causal role for depression in influencing subsequent disability outcomes, the high prevalence of depressive illness in this cohort underscores the importance of recognising and treating depression in people with MS 18 .

Code availability
All code used in these analyses is available at https:// github. com/ benja cobs1 23456/ UKMSR_ depre ssion. Access to UKMSR data is open to all researchers on application and subject to suitable governance review. Details of how to apply for the data can be found here: https:// ukmsr egist er. org/ Resea rch/ Worki ngWit hUs.