Prevalence, treatment and correlates of depression in multiple sclerosis

Background: The prevalence of depression in Multiple Sclerosis (MS) is often assessed by administering patient reported outcome measures (PROMs) examining depressive symptomatology to population cohorts; a recent review summarised 12 such studies, eight of which used the Hospital Anxiety and Depression Scale-Depression (HADS-D). In clinical practice, depression is diagnosed by an individual structured clinical interview; diagnosis often leads to treatment options including antidepressant medication. It follows that an MS population will include those whose current depressive symptoms meet threshold for depression diagnosis, plus those who

Background: The prevalence of depression in Multiple Sclerosis (MS) is often assessed by administering patient reported outcome measures (PROMs) examining depressive symptomatology to population cohorts; a recent review summarised 12 such studies, eight of which used the Hospital Anxiety and Depression Scale-Depression (HADS-D).In clinical practice, depression is diagnosed by an individual structured clinical interview; diagnosis often leads to treatment options including antidepressant medication.It follows that an MS population will include those whose current depressive symptoms meet threshold for depression diagnosis, plus those who previously met diagnostic criteria for depression and have been treated such that depressive symptoms have improved below that threshold.We examined a large MS population to establish a multi-attribute estimate of depression, taking into account probable depression on HADS-D, as well as anti-depressant medication use and co-morbidity data reporting current treatment for depression.We then studied associations with demographic and health status measures and the trajectories of depressive symptoms over time.Methods: Participants were recruited into the UK-wide Trajectories of Outcome in Neurological Conditions-MS (TONiC-MS) study, with demographic and disease data from clinical records, PROMs collected at intervals of at least 9 months, as well as co-morbidities and medication.Interval level conversions of PROM data followed Rasch analysis.Logistic regression examined associations of demographic characteristics and symptoms with depression.Finally, a group-based trajectory model was applied to those with depression.Results: Baseline data in 5633 participants showed the prevalence of depression to be 25.3 % (CI: 24.2-26.5).There were significant differences in prevalence by MS subtype: relapsing 23.2 % (CI: 21.8-24.5),primary progressive 25.8 % (CI: 22.5-29.3),secondary progressive 31.5 % (CI: 29.0-34.0);disability: EDSS 0-4 19.2 % (CI: 17.8-20.6),EDSS ≥4.5 31.9 % (CI: 30.2-33.6); and age: 42-57 years 27.7 % (CI: 26.0-29.3),above or below this range 23.1 % (CI: 21.6-24.7).Fatigue, disability, self-efficacy and self esteem correlated with depression with a large effect size (>0.8)whereas sleep, spasticity pain, vision and bladder had an effect size >0.5.The logistic regression model (N = 4938) correctly classified 80 % with 93 % specificity: risk of depression was increased with disability, fatigue, anxiety, more comorbidities or current smoking.Higher self-efficacy or self esteem and marriage reduced depression.Trajectory analysis of depressive symptoms over 40 months in those with depression (N = 1096) showed three groups: 19.1 % with low symptoms, 49.2 % with greater symptoms between the threshold of possible and probable depression, and 31.7 % with high depressive symptoms.29.9 %

Introduction
Depression is a common co-morbidity among people with Multiple Sclerosis (pwMS); a recent meta-analysis found the prevalence to be 27 % (Peres et al., 2022) whilst the lifetime prevalence has been reported as about 50 % (Sadovnick et al., 1996;Feinstein, 2004).It has a major negative impact on quality of life (Kołtuniuk et al., 2023), including disease management (Washington and Langdon, 2022) and employment (Srpova et al., 2022).At the more severe end of the spectrum, suicide risk is twice that in the general population (Kalb et al., 2019).
The structured clinical interview utilising the Diagnostic and Statistical Manual of Mental Disorders (DSM-IV) criteria is often viewed as the 'gold standard' for case ascertainment in depression (American Psychiatric Association 1994).However, such individualised assessment is labour-intensive.Many MS services utilise patient reported outcome measures (PROMs), such as the depression subscale of the Hospital Anxiety and Depression Scale (HADS-D) (Zigmond and Snaith, 1983) or the Beck Depression Inventory (BDI) (Beck et al., 1996).These ask about the frequency of depressive symptomatology; if a patient's depression score is above the cut-point, then the referral is classed as a clinical case (Public health profiles 2023).However, responses may be influenced by somatic symptoms; for example, the BDI has questions on inability to work, sleep disturbance, and fatigue, all of which may be related to MS rather than depression, and so incorrectly attributed by a questionnaire.This inclusion of non-mood aspects has been noted previously, with its associated risk of inflated prevalence (Nyenhuis et al., 1995).This may be one reason why case ascertainment varies by the PROM chosen, even in the same subjects at the same time (Covic et al., 2009).
An important consideration when assessing the prevalence of depression using questionnaires on depressive symptomatology is that both pharmacological and non-pharmacological interventions may have reduced symptoms below "caseness" on such case-finding questionnaires (Kurt et al., 2007).Real-world data on the use of antidepressant medication shows that the potential confounding effect of medication may apply to more than a third of patients (Meader et al., 2011).Likewise, non-pharmacological interventions may have a similar effect, given data on comparable efficacy and their use as first line strategies for treatment, consistent with the National Institute for Health and Care Excellence (NICE) guidelines on the management of depression (Gartlehner et al., 2016;NICE 2022).It follows that any attempt to derive the prevalence of depression in any given population must include a multi-attribute approach, not just relying on a single case-finding questionnaire, but also including current therapeutic interventions.
The aim of the current study was to examine the prevalence and treatment of depression in a large cohort of people with MS from across the United Kingdom.It used a case-finding questionnaire, specifically the HADS-D which was designed for use in populations where high frequency of somatic symptoms from medical conditions was anticipated, so it focuses on psychological symptoms of depression, such as anhedonia or reduced ability to experience pleasure (Zigmond and Snaith, 1983).The prevalence of current depression was estimated using data from three sources, first HADS-D score greater than published cut-points (Zigmond and Snaith, 1983;Brehaut et al., 2020), secondly current anti-depressant medication use and thirdly co-morbidity data reporting current treatment for depression.Treatment among those with depression was examined and the correlates with depression were examined for a wide range of health status measures.

Sample
Participants were recruited into the Trajectories of Outcome in Neurological Conditions-MS (TONiC-MS) study (Young et al., 2021;Young et al., 2022;Young et al., 2023) where eligibility criteria included adults with physician-verified MS (by McDonald criteria (Polman et al., 2011)) of any disease subtype and level of disability, providing they could give informed consent and complete questionnaire packs (with the help of a scribe if necessary).
Data on disease subtype at time of study entry were provided by clinicians involved in the patients' care and classified as relapsingremitting (RRMS), primary progressive (PPMS) or secondary progressive (SPMS).Duration since diagnosis and Expanded Disability Status Scale (EDSS) band were recorded from the medical records.Informed consent was obtained from all participants prior to enrolment.

Range of information collected in the questionnaire pack
In addition to the data obtained from the medical records, several Patient Reported Outcome Measures (PROMs) were included in the questionnaire pack.Following the baseline pack, further questionnaire packs were sent at approximately 9-month intervals.

A. Patient Reported Outcome Measures (PROMs)
The questionnaires relevant to the current investigation included: a.Multiple Sclerosis Impact Scale (MSIS-29) -two subscales for Physical (20 items) and Psychological (9 items) aspects of MS (Hobart et al., 2001).b.Hospital Anxiety and Depression Scale (HADS) -two subscales measuring anxiety and depression, each with 7 items, with associated cut-points delivering none-possible-probable levels of each trait (Zigmond and Snaith, 1983); the standard ordinal score of 11 (or its metric equivalent) was used for the cut-point for Probable depression (Brehaut et al., 2020).c.Neurological Fatigue Index-MS (NFI-MS) -10 item summary scale scored 0-30, where higher scores represent greater MS fatigue (Mills et al., 2010).d.Neurological Sleep Index-MS (NSI-MS) -specifically, the Non-Restorative Nocturnal Sleep Scale with 14 items, where high scores indicate worse sleep (Mills et al., 2016).e. World Health Organization Disability Assessment Schedule-2.0 (WHODAS-2.0)-the 32-item version was used, omitting the questions from the domain relating to work (D5.5-D5.8)as over half of our population were not in employment (Üstün et al., 2010).f.Multiple Sclerosis Spasticity Scale (MSSS-88) -subscale for spasticity-specific pain, totalling 9 items, each scored 0=not at all to 3=extremely (Hobart et al., 2005).
g. Qualiveen -8 items assessing urinary problems and their impact on quality of life (QoL), where high scores indicate high impact, validated in MS (Milinis et al., 2017).h.Unidimensional Self-Efficacy Scale for MS (USE-MS) -12 items scored 0-3, assessing the belief of pwMS that they can achieve actions while living with MS (Young et al., 2012).i. World Health Organization Quality of Life Scale-BREF (WHO-QOL-BREF) -24 items covering 4 domains: physical, psychological, social relationships and environment.Two stand-alone questions on QoL and satisfaction with health were not included.A total score from the 24 items, obtained from a bi-factor solution, was used in this analysis (Pomeroy et al., 2020).j.Multiple Sclerosis Vision Questionnaire (MSVQ-7) -7 items scored from 'No problem' (0) to 'Severe problem' (3), or not applicable, with a resulting score range from 0 to 21 where higher scores indicate worse vision problems (Ma et al., 2002).k.Rosenberg Self Esteem Scale -10 item scale that measures global self-worth by measuring both positive and negative feelings about the self (Rosenberg, 1965).l.Stigma Scale for Chronic Illnesses -8 item scale that assesses enacted and internalised stigma (Molina et al., 2013).m.Multidimensional Locus of Control Scale (MHLC) Form C -in the current study consisting of three domains representing Internal-, External-(Powerful Others) and Chance-Locus of Control (Wallston et al., 1978).Each domain consists of six items scored 1-6 (changed to 0-5), giving a domain score of 0-30 where a high score indicates greater emphasis on that domain.

B. Comorbidities
The ascertainment of comorbidity was derived from a list of 35 comorbidities based on existing work (Marrie et al., 2015).PwMS were asked if they had the comorbidity, when it was diagnosed, and if they were currently receiving treatment for it.

C. Health Economics and Other Indicators
a. Medication An extensive health economics set of questions included an opportunity to record all current prescribed medications, their dose and its frequency.The following drugs were considered antidepressant medications: selective serotonin reuptake inhibitors (SSRIs): citalopram, fluoxetine, paroxetine, sertraline, escitalopram, fluvoxamine, vortioxetine; serotonin-norepinephrine reuptake inhibitors (SNRI): venlafaxine, duloxetine; noradrenergic and specific serotonergic antidepressant (NaSSA): mirtazapine; serotonin antagonist and reuptake inhibitors (SARIs): trazodone.

b. Employment Status
A simple set of questions determined if the pwMS was in paid employment, and if so, whether this was full-or part-time.
c. Body map Participants ticked a stylised drawing of a person to indicate which of 17 possible body parts or functions (e.g., "thinking") were affected by MS.The results were recoded as Yes/No.

Statistical procedures
All ordinal values of PROMs have previously been transformed to interval level scaling through application of the Rasch measurement model and available nomograms (Rasch, 1960;Gray-Little et al., 1997;Hobart et al., 2006;Mills et al., 2010;Young et al., 2012;Milinis et al., 2017;Pomeroy et al., 2020;Young et al., 2023).Confidence intervals for the prevalence estimates are based upon the Wilson Score interval, suitable for small numbers, i.e. for the age-gender specific estimates (Wilson, 1927;Wallis, 2013).
A logistic regression strategy utilized a conceptual approach based upon the Wilson and Cleary model, first taking discrete clusters of clinical variables, symptoms, functioning, person and environmental factors (Wilson and Cleary, 1995).Depending on these findings, a final model is chosen to maximize McKelvey and Zavoina's pseudo R 2 , which is based on the explained variance, where the variance of the predicted response is divided by the sum of the variance of the predicted response and residual variance (McKelvey and Zavoina, 1975).It has the closest approximation to the standard interpretation of R 2 in ordinary least-squares regression (Veall and Zimmermann, 1994).The full sample was randomised into training and validation samples for cross-validation purposes.
Finally, a group-based trajectory model was applied to those with depression.This was designed to identify groups of individuals following similar developmental trajectories (Jones and Nagin, 2013;Mori et al., 2020).All analysis is undertaken in STATA17 (StataCorp 2021).

Results
This analysis included pwMS recruited from 2013 to 2019, so any effect of the COVID-19 pandemic on prevalence or treatment of depression is excluded.During this time, 5633 pwMS completed a baseline questionnaire.Differences of demographic and clinical aspects across MS subtypes were as expected; for example, 55.1 % of PPMS being female, compared to 77.3 % of those with RRMS (Table 1).Given the sample size, all group comparisons were significant apart from those who then contributed to the longitudinal study (t-test or Chi-Square < 0.05).
The derivation of the prevalence of depression from the multiattribute ascertainment is shown in Fig. 1.The cut-points for depression caseness on the metric scale are 10.3 for Possible and 12.35 for Probable, equivalent to 8 and 11 on the ordinal scale.HADS probable caseness was 12.2% (CI: 11.4-13.1);those on antidepressant medication 7.6% (CI: 7.0-8.3)and those reporting treated comorbidity 13.8% (CI: 13.0-14.8).Taken together, with some overlap, the prevalence of depression was found to be 25.3% (CI: 24.2-26.5).
Of those with depression, 29.9% (CI: 27.6-32.3)were untreated according to their medication list or patient reporting of treatment for depression (which could include non-pharmacological therapies).Conversely, of those treated, 26.1% still had a symptom level consistent with a probable case (CI: 23.5-28.9).
The age-sex specific prevalence rates are shown in Table 2.The higher rate in females was maintained in most age groups.
The discrimination of the PROMs across depression is shown in Table 3.All PROMs showed a significant difference.The strongest discrimination across depression is shown by fatigue, disability, selfefficacy and self esteem, all of which have a large effect size (>0.8),while the remainder have a medium effect size (>0.5).
After the exploration of clusters by logistic regression, a final model was chosen for further analysis.Initially it was applied to the training sample, and then the validation sample.Parameters of the cross validation between the training and validation samples are shown in Table 4. Little difference is shown between the two with all variables remaining significant, and so the samples were merged for further analysis.
The collective associations between demographic characteristics/ symptoms with depression are shown in the results of the logistic regression based on the merged sample (Table 5).The factors increasing risk of depression in descending order of influence were comorbidities, anxiety, fatigue, being a current smoker and disability.Self esteem is associated with the greatest reduction in the risk of depression, followed by self-efficacy and being married.
While the specificity of the model is high (93%), with few false positives, the sensitivity is low (44%), indicating many false negatives, i. e., those categorized by the model as not depressed although they had depression as identified by the multi-attribute analysis.To clarify why the demographic characteristics/symptoms model accurately predicted those without depression but missed some of those with depression, we conducted a further analysis of all those with depression.Table 6 shows a logistic regression of all subjects identified as having depression according to the multi-attribute analysis, with false negative as the dependent variable, contrasted against true positives.
False negatives comprise those who are depressed but misallocated by the demographic characteristics/symptoms model.Table 6 shows such pwMS were more likely to report antidepressant medication, and a higher quality of life.They also have a higher external locus of control.They were less likely to have early onset MS, and less likely to have reported problems with bladder, balance, swallowing and nonrestorative sleep.Essentially the pwMS for whom the model failed to  identify depression appeared to be those where their condition was well managed and, consequently, they had low levels of symptoms and much higher perceived quality of life.This group emerges clearly in the trajectory analysis.Fig. 2 shows the trajectory analysis of the level of depressive symptoms over the followup period for those with depression.Group 1 has low levels of depressive symptoms with a significant slight early decline, and then a significant rise after two years, nevertheless staying well below the level of caseness on the metric, even for Possible depression.Group 2 also shows a slight but significant decline over two years, thereafter stable.It remains below the threshold for Probable caseness on the metric.Group 3 represents those with a high level of symptoms, so much so that 96.5% are above the threshold for Probable depression on the HADS-D.They display an upward trend over time which just fails to reach significance.On looking at just true positives and false negatives from the previous analysis, Group 1 comprises almost entirely of false negatives (91.8%); 58.5% of Group 2 and 31.2% in Group 3 (χ 2 205.3 (df2); p ≤ 0.001).Of concern, 40.5% of Group 3 with high levels of depressive symptoms report no treatment for depression, which represents 61% of all those untreated.

Discussion
Using a multi-attribute approach, the prevalence of depression amongst a large sample of pwMS in the UK was found to be 25.3% (CI: 24.2-26.5).This is in keeping with the results of a systematic review of 118 studies across many countries, which found the prevalence of depression in MS to be 23.7% (95% CI: 17.4%− 30.0%) for population-based studies (Marrie et al., 2015).The current study found that depression prevalence differed significantly across MS subtypes, with RRMS prevalence 23.2%, PPMS 25.8% and SPMS 31.5%.A previous review found that most studies omitted subtype and quoted moderate to high heterogeneity evidence for RR and progressive MS, so the needs of   those with SPMS for depression monitoring may have not been hitherto recognized (Peres et al., 2022).Logistic regression found that comorbidities, anxiety, fatigue, disability, and being a current smoker, were all associated with increased odds of depression, while higher levels of self-efficacy and self esteem, as well as being married, were associated with reduced odds.Treating anxiety and improving self esteem and self-efficacy are targets for intervention to reduce depression in MS.
The model findings highlight that pwMS experience a range of factors which interplay to influence their risk of depression.A logistic regression found that the false negatives appeared to be treated with antidepressants, with low levels of symptomology.This recognition of a well-managed sub-group within the larger depression group also emerged in the trajectory analysis, which showed three clearly delineated groups.The first had a low level of depressive symptoms, which remained low over follow-up and 98.9% reported current depression treatment at baseline.The second group, comprising 49.2% of those with depression, followed a trajectory that remained just below the HADS-D "Probable caseness" throughout follow-up.The third group showed a high level of symptoms, with virtually all above the level for "Probable caseness".This extends our understanding of depression throughout the disease course, which has been reported to be generally stable (Freedman et al., 2023).Clinical services should recognize that depression is heterogenous in pwMS, with 20% doing well under depression management, 50% having a stable level of depressive symptoms just below the Probable cut-point and the remaining 30% of particular clinical concern, as they have high depressive symptoms, trending to worsen over time, and 40% of this group report no treatment for depression.
In the current study, we have shown how prevalence estimates can be affected by the number of attributes considered, such that the final estimate is over twice that of the HADS Probable caseness alone.The choice of cut-points for the screening questionnaire will also affect prevalence.For example, using the HADS-D, higher cut-point values will minimize false positives at the expense of false negatives and consequently estimates will vary depending on which cut-point is used (Wu et al., 2021).The current study chose the standard ordinal score of 11 for the cut-point for Probable depression (or its metric equivalent), which has been shown to be the closest to the estimates derived from the Structured Clinical Interview for DSM (SCID), albeit with substantial heterogeneity in the difference between HADS-D ≥ 11 and SCID-based estimates (Brehaut et al., 2020).
Of those with depression, 29.9% (CI: 27.6-32.3)were untreated.Of those treated, 26.1% still had a symptom level consistent with a probable case (CI: 23.5-28.9).Thus, the current study showed that almost half of those with depression had inadequate treatment, either absent or not effective.The adequacy of treatment therefore remains a subject of concern, consistent with earlier findings (Raissi et al., 2015).There is a need to develop clear and comprehensive clinical guidelines which are both specific to comorbid depression in MS and which support implementation of individualized depression management, which our trajectory findings indicate is required (McIntosh et al., 2023).A qualitative study of MS nurses and neurologists confirmed the need for evidence-based guidance and more training to improve practices including screening for depression and highlighted the lack of local referral pathways to affordable and accessible mental health services (Marck et al., 2022).The current study found many potential points of intervention which merit further investigation.These include reducing fatigue and anxiety, and increasing physical functioning, self esteem and self-efficacy; cessation of smoking may also be beneficial and amenable  to intervention (Arriaza et al., 2022;Binshalan et al., 2022;Rodgers et al., 2022).Further longitudinal work should explore causal relationships.
A limitation of the study is the use of the HADS-D for measurement of depressive symptoms rather than a Structured Clinical Interview.In a structured review examining 265 papers identified by literature search for depression in MS, focusing on those describing at least 200 adults diagnosed according to McDonald criteria and written in English, 12 papers met eligibility (Peres et al., 2022).Of these, 8 used the HADS-D and the remaining four papers used three different depression scales or no scale.This study also examined the current prevalence of depression, as only current depressive symptoms or antidepressant prescription were included, so the lifetime prevalence of depression requires further study.
Further limitations include related variables which were not addressed, such as cognition (Altieri et al., 2023) and how depression is influenced by direct neuro-biological processes in MS (Masuccio et al., 2021).White matter brain MRI lesion load and localization, particularly in the temporal and frontal lobes, correlate with the presence of depression in MS.Similarly brain atrophy (mainly of the temporal lobe) is related to depression in MS.Pathological changes in normal appearing white matter and grey matter have also been associated with depression in MS.Unsurprisingly the limbic system, which regulates memory and emotional response, is the area most strongly linked to depression, notably the hippocampus (Masuccio et al., 2021).Neurochemical changes have been linked to depression, even in the earliest stages of MS (Guenter et al., 2020).

Conclusions
In conclusion, the prevalence of depression in a large UK cohort of people with MS, determined using a multi-attribute approach, was found to be 25.3%.A wide range of symptoms, disability and personal factors were found to be associated with depression.Comorbidity was a strong factor for the presence of depression, while self-efficacy and self esteem were strongly associated with reduced prevalence.Prevalence varied by MS subtype, with SPMS having the highest rate.Over half of those with depression were found to have either an absent or ineffective therapeutic intervention.Taken together, these findings suggest that the clinical management of depression in MS remains a significant challenge.They highlight the requirement for improved screening and monitoring for depression, and the need for MS teams to support the psychological as well as physical health of pwMS.

Data access
Data supporting this study are not openly available due to reasons of sensitivity and are available from the corresponding author upon reasonable request.

Declaration of competing interest
The authors report there are no competing interests to declare.

Role of funding source
The funding sources had no role in study design; in the collection, analysis and interpretation of data; in the writing of the report; or in the decision to submit the article for publication.

Funding
This work was supported by unrestricted investigator-lead grants from Biogen, Merck, Novartis, Roche, Sanofi Genzyme, Teva; and Neurological Disability Fund 4530.Longitudinal data collection also received part-funding from the Multiple Sclerosis Society [grant number 62].The funding sources had no role in study design; in the collection, analysis and interpretation of data; in the writing of the report; or in the decision to submit the article for publication.
For the purpose of Open Access, the author has applied a Creative Commons Attribution (CC BY) licence to any Author Accepted Manuscript version arising.

Table 1
Demographic, clinical and health status aspects at baseline.
SD: standard deviation; HND: Higher national diploma; PROM: Patient Reported Outcome Measure; EDSS: Expanded Disability Status Scale; NFI-MS: Neurological Fatigue Index-MS summary scale; QoL: quality of life; WHOQOL-BREF: World Health Organization Quality of Life-BREF; HADS-D: Hospital Anxiety and Depression Scale-Depression.C.A. Young et al.

Table 3
Association with health status PROMs.Rasch-based estimates.

Table 4
Parameters of fit for the training, validation and merged samples.

Table 5
Final logistic regression model of associations with depression.

Table 6
Odds ratios of the false negatives in comparison to true positives.