Predictors of patient-reported fatigue symptom severity in a nationwide multiple sclerosis cohort

Background: Fatigue is a debilitating symptom of multiple sclerosis (MS), but its relation to sociodemographic and disease-related characteristics has not been investigated in larger studies. The objectives of this study were to evaluate predictors of self-reported fatigue in a Swedish nationwide register-based MS cohort. Methods: Using a repeated cross-sectional design, we included 2,165 persons with relapsing-remitting and secondary progressive MS with one or multiple Fatigue Scale for Motor and Cognitive Functions (FSMC) scores, which was modelled using multivariable linear regressions for multiple predictors. Results: Only associations to expanded disability status scale (EDSS) and Symbol Digit Modalities Test (SDMT) were considered clinically meaningful among MS-associated characteristics in our main model; compared to mild disability (EDSS 0-2.5), those with severe disability (EDSS ≥ 6) scored 17.6 (95% CI 13.1-22.2) FSMC points higher, while the difference was 10.7 (95% CI 8.0-13.4) points for the highest and lowest quartiles of SDMT. Differences between highest and lowest quartiles of health-related quality of life (HRQoL) instruments were even greater and considered clinically meaningful; EuroQoL Visual Analogue Scale (EQ-VAS) 31.9 (95% CI 29.9-33.8), Multiple Sclerosis Impact Scale (MSIS-29) psychological component 35.6 (95% CI 33.8-37.4) and MSIS-29 physical component 45.5 (95% CI 43.7-47.4). Conclusion: Higher self-reported fatigue is associated with higher disability level and worse cognitive processing speed, while associations to other MS-associated characteristics including MS type, line of disease modifying therapy (DMT), MS duration, relapse and new cerebral lesions are weak. Furthermore, we found a strong correlation between high fatigue rating and lower ratings on health-related quality of life instruments.


Introduction
Fatigue is considered a common and disabling symptom of multiple sclerosis (MS), affecting up to 80% of all patients (Brenner and Piehl, 2016) with major impact on patients' daily life (Penner and Paul, 2017).Nevertheless, it is one of the least studied symptom dimensions of MS (Penner and Paul, 2017).
Fatigue is suggested to impact negatively on patient's health-related quality of life (HRQoL) (Brenner and Piehl, 2016;Kratz et al., 2017), social activities and employment status (Krause et al., 2013;Penner and Paul, 2017), and has been associated with cognitive dysfunction and psychiatric comorbidity, such as depression, anxiety and sleep disorders, although these associations may also represent feature overlap rather than a causative association (Bakshi et al., 2000;Brenner and Piehl, 2016;Penner et al., 2007;Penner and Paul, 2017).In contrast, previous studies addressing associations between fatigue and clinical variables such as MS-type, disease duration and MS disability have produced inconsistent findings (Bakshi et al., 2000;Koch et al., 2008;Patrick et al., 2009).As fatigue is assessed with self-administered instruments, the resulting scores may also be affected by personality and sociodemographic background (Penner and Paul, 2017).
Existing evidence regarding predictors of fatigue is still limited in terms of sample sizes, range of adjustment for sociodemographic characteristics, comorbidity, and other clinical aspects of MS, which limits generalizability.The objective of this study was to evaluate the extent to which MS-associated characteristics and cognitive processing speed are independently associated with self-reported fatigue and to assess if these associations are connected to sociodemographic characteristics, social benefits, comorbidity and HRQoL.Fatigue was assessed by the Fatigue Scale for Motor and Cognitive Functions (FSMC), a standardized selfreported questionnaire developed and validated in MS patients (Penner et al., 2009).

Methods
Using a repeated cross-sectional design, we conducted a nationwide register-based study of patients with relapsing-remitting MS (RRMS) and secondary progressive MS (SPMS), linking the Swedish MS Registry (SMSreg) to national health and census registries.

Study population
SMSreg is a nationwide web-based resource covering approximately 80% of all prevalent MS patients in Sweden (Hillert and Stawiarz, 2015), with high validity for collected data (Alping et al., 2019).For this study, we included all patients with RRMS and SPMS who had ≥ 1 complete FSMC measurement(s) registered from January 1 st , 2012 to December 31 st , 2018, where a participant could contribute with more than one FSMC score.FSMC was recorded mainly in context of observational prospective follow up studies involving larger MS centers, such as the Combat-MS study (clinicaltrials.govNCT03193866).Through the unique personal identification numbers assigned to all Swedish residents (Ludvigsson et al., 2009), we added covariate data by linking the study population to national health and census registries.We excluded 399 patients with no disease modifying therapy (DMT) recorded in the SMSreg and 5,236 FSMC measurements (from 3,943 patients) with missing values in any of the covariates studied, leaving 2,165 patients and 3,213 FSMC measurements in the final data set.For sub-analyses of sick leave and disability pension we limited FSMC measurements to those aged 18-64 years old (N=3,179 FSMC measurements performed in 2,137 patients).

Outcome Definitions
We assessed fatigue through the FSMC total, and the contributing cognitive and motor sub-domain scores, which were categorized as no/ low, mild, moderate, and severe (Table 1) (Penner et al., 2009).In alignment with a previous study, we regarded mean total FSMC score differences of ≥10 as being clinically meaningful (D'Hooghe et al., 2018).

Covariate Definitions
Covariates were defined in relation to timing of each FSMC measurement.Sociodemographic characteristics included age, sex, geographical region of residence, country of birth and educational attainment from census registers (Ludvigsson et al., 2016) and Swedish Longitudinal Integrated Database for Health Insurance and Labor Market Studies (Ludvigsson et al., 2019) maintained by Statistics Sweden.We assessed the proportion of patients who received sick leave and disability pension in the year before FSMC measurement from MiDAS (maintained by the Swedish Social Insurance Agency).
We assessed medical history and comorbidity through the 10 th version of International Classification of Diseases (ICD) coded diagnoses in The National Patient Register, covering all inpatient and nonprimary outpatient care in Sweden (Ludvigsson et al., 2011) and classified prescribed drugs according to the Anatomical Therapeutics Chemical (ATC) system in The Prescribed Drug Register, with complete coverage of prescription medications dispensed at pharmacies in Sweden (but not hospital-administered drugs) (Wettermark et al., 2007).To define comorbidities, we included ICD coded diagnoses and prescribed drugs according to previous publications (Longinetti et al., 2021;Luna et al., 2019;Marrie et al., 2021) through diagnoses given in the last five years before FSMC measurement (invasive cancer as ever occurring) or medications claimed in the year before FSMC measurement (see eTable 1 in the Supplement for complete definitions including ICD-10 codes and ATC codes).Psychiatric comorbidity included use of antidepressants, anxiolytics, sleeping pill or antipsychotics, and diagnoses of depression, anxiety disorder or any other mental or behavioral disorder.Somatic comorbidity included hospital treated infections, cardiovascular diseases (i.e., arrhythmia and major adverse cardiovascular events; MACE), use of diabetes mellitus medication and prior history of invasive cancer.Symptomatic pharmacological fatigue treatment included central stimulants and amantadine.
We obtained information on MS-associated variables from the SMSreg, including MS type, DMT line, year of diagnosis, disease duration, relapses in the preceding year, new cerebral lesions in the preceding year detected by magnetic resonance imaging (MRI) and expanded disability status scale (EDSS) scores (Kurtzke, 1983), information on processing speed by Symbol Digit Modalities Test (SDMT) (Benedict et al., 2017), HRQoL variables including EuroQoL Visual Analogue Scale (EQ-VAS) (The Euroqol Group, 1990) and Multiple Sclerosis Impact Scale (Hobart et al., 2001) (MSIS-29; excluding item 23, "Feeling mentally fatigued?",as it overlaps with FSMC total score, Pearson´s correlation 0.74, p≤0.0001).We considered EDSS scores recorded within 180 days before and 15 days after collected FSMC measurements; for the MSIS-29, SDMT, and EQ-VAS, we considered measurements 30 days before and 30 days after each collected FSMC measurement (in most cases SDMT and HRQoL variables were collected on the same occasion as FSMC).

Statistics
To describe our study population, we tabulated covariates across FSMC total groups.For the main analysis, we modelled FSMC total, cognitive and motor sub-domain scores as a function of covariates using multivariable linear regression.Since participants could contribute several measurements, we calculated robust (sandwich estimator) 95% confidence intervals (CIs).Due to the risk of type I statistical errors, clinically meaningful differences were also evaluated against a Bonferroni corrected significance level of 0.001, accounting for the 52 separate tests done in the regression models.
We first performed univariate regression to assess crude associations between FSMC scores and each covariate (Model 1).In subsequent steps, we performed multivariable regression to explore associations to each covariate, while adjusting for other covariates (Models 2-6, described in eTable 2 in the Supplement).These are expressed as point estimates with confidence intervals.
We used complete case analysis to handle missing data, implying that FSMC measurements were included only if there were no missing values in any of the covariates.For respondents with missing data in the MSIS-29 psychological and physical scale, but where at least 50% of the items had been completed, we computed a respondent-specific mean score from the completed items (Hobart et al., 2001).We used SAS version 9.4 (SAS Institute) for all statistical analyses.

Standard Protocol Approvals, Registrations, and Patient Consents
This study was approved by the Regional Ethical Board of Stockholm (reference number: 2017/700-31/4).As is common for pseudonymized linkages of national registers in Sweden, the need for individual informed consent was waived.However, all participants had given consent to registration in the SMSreg, including a consent to use registered data for research.

Data Availability Statement
Data is available upon approval from the respective register holders.
Missing data were negligible (<3%) for the investigated variables except for EDSS, SDMT and EQ-VAS which were missing in the predefined time windows for 26-35% of all FSMC scores, while MSIS-29 was missing in 11% (see eTable 3 in the Supplement for detailed information).Patient characteristics were largely similar across patients with and without complete data, except for region of residence, disability pension and MS type where differences between the two groups were modest.

MS-associated characteristics, processing speed and HRQoL
Crude (Model 1) and adjusted (Models 2-6) mean differences (MDs) in FSMC total scores by MS-associated characteristics, processing speed and HRQoL are presented in Table 3.After adjusting for sociodemographic characteristics (Model 2), MS type, SDMT and EDSS were the only MS-associated characteristics presenting a clinically meaningful association (FSMC total score MD ≥ 10).Additionally, all measures of HRQoL also clearly surpassed this threshold.
When additionally adjusting for MS-associated characteristics and processing speed (Model 3), associations were further attenuated.We consider this our main model, and results are displayed in Table 3 and Fig. 2.Among MS-associated characteristics, all covariates except for EDSS fell below the 10-point threshold for being clinically meaningful, where MS type was most affected.Patients with severe MS disability (EDSS ≥ 6) reached a mean of almost 18 FSMC total scores higher than those with mild MS disability (EDSS score 0-2.5).The same pattern emerged with processing speed (SDMT), both components of MSIS-29 and EQ-VAS.Hence, differences between the highest and lowest quartiles of SDMT, EQ-VAS, MSIS-29 psychological component and MSIS-29 physical component translated into MDs of 10.7, 31.9, 35.6 and 45.5, respectively.These clinically meaningful mean differences were also statistically significant at the Bonferroni-adjusted threshold (all p<0.001).
When additionally adjusting for HRQoL (Model 4) all covariates, except for the HRQoL variables, previously considered as clinically meaningful now fell below the threshold, indicating that previous associations to fatigue in different ways could be explained by their associations to HRQoL measures.Among the HRQoL variables, the MSIS-29 physical component captured the association best.Additional adjustment for comorbidities (Model 5) and symptomatic drug treatments for fatigue (Model 6) did not change the associations.

Sociodemographics, social benefits, medical history and comorbidities
Crude (Model 1) and adjusted (Models 2-6) MDs in FSMC total scores by sociodemographic characteristics, medical history and comorbidities are presented in Table 4.When adjusting for sociodemographic characteristics, MS-associated characteristics and processing speed (model 3), associations to most sociodemographic characteristics were no longer clinically meaningful.In contrast, patients with any sick leave and disability pension ongoing in the preceding year reported a mean of 11.8 and 15.5 FSMC total scores higher respectively than those without sick leave or disability pension.Furthermore, patients with psychiatric comorbidity, except for antipsychotic use, reported a mean of FSMC total score 10.4 to 13.3 higher than those without.Similar patterns emerged for use of sleeping pills and symptomatic treatment for fatigue, but not for somatic comorbidity.Clinically meaningful mean differences from model 3 were also significant at the Bonferroni-adjusted threshold (all p<0.001), except for Amantadine (p=0.0018).After additionally adjusting for HRQoL (Model 4), all variables previously considered as clinically meaningful now fell below the threshold.Additional adjustment for comorbidities (Model 5) and symptomatic drug treatments for fatigue (Model 6) produced virtually similar results.Abbreviations: DMT = disease modulatory therapies; EDSS = expanded disability status scale; EQ-VAS = euroqol visual analogue scale; FSMC = fatigue scale for motor and cognitive function; HRQoL = health-related quality of life; MACE = Major adverse cardiovascular event; MS = multiple sclerosis; MSIS-29

Associations with FSMC cognitive and motor sub-domain
Crude and adjusted MDs for the FSMC total scores and the cognitive and motor sub-domain scores yielded similar results (data not shown), with a few exceptions.In the unadjusted model (Model 1), patients with SPMS had on average 6 FSMC cognitive scores higher than RRMS patients (MD 6.3 [95% CI, 4.2; 8.4]), while FSMC motor scores were on average 11 points higher than in RRMS patients (MD 10.8 [95% CI,9.0;12.6]).The same pattern was evident with EDSS.In Model 3, when adjusting for sociodemographic characteristics, MS-associated characteristics and processing speed, patients with severe MS disability (EDSS ≥ 6) reported a MD of 6.9 (95% CI, 4.3; 9.4) FSMC cognitive scores higher than those with mild disability (EDSS score 0-2.5), while the corresponding value for FSMC motor was 10.7 (95% CI, 8.5; 12.9).
The relation between fatigue and MS associated characteristics has been investigated only in a limited number of studies before, a few of which demonstrated a strong association with disability (Broch et al., 2021;Mills and Young, 2011), while associations were weak (Kroencke et al., 2000) or non-existent in other studies (Bakshi et al., 2000;Koch et al., 2009).This inconsistency may depend on how fatigue was rated and what covariates were included in statistical models.Thus, studies not adjusting for other MS-associated characteristics including disability and disease duration found a strong association between fatigue and disease course (Colosimo et al., 1995), while those adjusting for these variables found a weak or no association (Kroencke et al., 2000).By applying a stepwise multivariable regression analysis with successively more comprehensive corrections for differences in characteristics, we were here able to uncover novel patterns regarding FSMC scorings.Thus, the magnitude of the association between self-reported fatigue and MS type was clearly reduced after further adjusting for other MS-associated characteristics and processing speed, in addition to sociodemographic characteristics, indicating that disability level and processing speed rather than MS type drives the association.
The association between increased fatigue and higher disability level measured with EDSS may have multiple explanations.Although the exact pathophysiological mechanisms of fatigue remain incompletely understood (Amato and Portaccio, 2012;Penner and Paul, 2017), it seems clear that fatigue and disability may share underlying mechanisms in MS.Thus, evolving brain atrophy in specific brain regions plays an important role both in the onset of MS fatigue (Yaldizli et al., 2011) and in increased EDSS (Rocca et al., 2017).Moreover, as a consequence of neuroaxonal degeneration and reduced nerve conduction velocity due to demyelination, people with MS with impaired walking ability (EDSS > 4) tend to not only report motor fatigue but also cognitive fatigue in conjunction with physical effort (Amato and Portaccio, 2012;Penner and Paul, 2017), which may explain why we saw a clinically meaningful association not only between EDSS and the FSMC subdomain motor score, but also with the mental fatigue.
A striking finding is the strong association between fatigue and the HRQoL measures MSIS-29 and EQ-VAS.Thus, especially for MSIS-29 physical, a strong correlation remained throughout all steps in the multivariable regression models.These results are in line with previous cross-sectional studies of fatigue and MSIS-29 (Mills and Young, 2011), which strengthens the notion that fatigue greatly overlap with MS patients' sense of well-being.This also explains why associations to disability level, processing speed and psychiatric comorbidity were weakened upon adjustment for MSIS-29 and EQ-VAS (Model 4).Fatigue was also a major predictor for claiming of social benefits, i.e., sick leave and disability pension.These findings indicate that fatigue, apart from exerting a substantial toll on HRQoL, also contributes a relatively stronger effect on (or is a consequence of) socioeconomic outcomes than many traditional measures of MS severity.Results also corroborate a large Canadian cross-sectional study linking HRQoL to disability status and depressive symptoms (Berrigan et al., 2016).In part, the impact of these predictors, together with physical comorbidity, was indirect and mediated through the intermediary role of self-reported fatigue.However, in a recent prospective observational cohort study we observed that risk of depression was associated with choice of DMT, suggesting that therapeutic management of MS impacts also on other relevant outcomes than relapse risk and disability progression (Longinetti et al., 2021).
Our results indicate a clinically meaningful association between the lowest range of processing speed and higher fatigue, which strengthens the notion of an overlap between fatigue, cognitive impairment and depression (Brenner and Piehl, 2016).As shown by the results from Model 3, we found a clinically meaningful association between higher fatigue and worse processing speed when comparing the lower quartiles of SDMT to the highest quartile.Cut-off values for SDMT to define cognitive impairment have been proposed in previous studies and varies between 49 to 55 (López-Góngora et al., 2015;Parmenter et al., 2007).This corresponds well to the lowest quartile in our sample, i.e., 5-49 SDMT scores, which indicate that fatigue likely is associated with cognitive impairment rather than general variability in processing speed.Even if longitudinal studies will be needed to establish the 1.2 (-4.6; 7.1) 1.9 (-3.7; 7.4) -6.4 (-11.9;-0.9) -3.9 (-9.3; 1.5) -3.5 (-8.3; 1.3) -2.8 (-7.6; 1.9) Female vs. Male 6.1 (3.8; 8.5) 7.1 (4.9; 9.3) 7.5 (5.5; 9.6) 4.0 (2.7; 5.3) 3.6 (2.3; 4.9) 3.6 (2.4; 4.9) Born abroad vs. Swedish-born 3.5 (0.2; 6.9) 4.1 (0.8; 7.4) 2.7 (-0. dynamics of this association more in detail, the present results highlight a group of patients where cognitive impairment and fatigue co-exist. Though larger and more detailed than previous studies on fatigue in MS patients, this study also suffers from some important limitations.First, as we used complete case analysis to handle missing data, a number of patients were excluded due to missing information on one or more covariate potentially resulting in selection bias.However, since the mean FSMC score was approximately similar in our full cohort as in our analyzed sample, this alleviates concerns that associations between covariates and FSMC have been affected.Second, the study was limited to cross-sectional analyses due to a restricted number of repeated measures, therefore not allowing for probing the impact of interventions such as choice of DMT (Longinetti et al., 2021).Still, limited evidence suggests that initiation of certain DMTs can be associated with meaningful lowering of fatigue scores (Svenningsson et al., 2013), however this needs to be replicated in larger real-world cohorts, preferably comparing across several DMTs.Third, an important aspect is the possible impact of central stimulants on our results, since a relatively high proportion used such medications despite absence of proven efficacy in controlled studies (Nourbakhsh et al., 2021).Fourth, although we were able to determine the proportion who used sleeping pills as an indicator of sleep problems, we did not have access to data on sleep-related disorders.A fifth limitation is that we did not have access to data on level of physical activity.Meta-analyses show that physical activity is associated with reduced reported fatigue in people with MS, although there is a need for more high-quality studies, especially such that explicitly include people who experience fatigue (Moss-Morris et al., 2021).Sixth, due to the exploratory nature of the study, we conducted multiple comparisons, which increases the risk of type I statistical errors.However, the differences highlighted as clinically meaningful were all statistically significant, except for amantadine use, when we applied a Bonferroni corrected significance level of 0.001, meaning that they are unlikely to represent chance findings.Finally, as the majority of the FSMC measurements in the study was collected from RRMS patients with ongoing DMT, the generalizability of our findings to other patient populations, such as those with SPMS or with longer disease duration, should be done with caution.Importantly, our sample did not include patients with primary progressive MS.

Conclusion
Collectively, this study contributes information on the extent of, and the predictors of, self-reported fatigue, mainly in patients with RRMS.Among MS characteristics, fatigue primarily was associated to the level of disability and processing speed.The strong correlation with HRQoL measures underscores the importance of screening for fatigue and directing efforts to identifying contributing causes.Further studies involving longitudinal registrations of fatigue will be required to address the impact of interventions, such as MS DMTs, management of comorbidities and symptomatic pharmacological and non-pharmacological interventions specifically directed at reducing fatigue symptoms.

Fig. 1 .
Fig. 1.Distribution of FSMC total scores.Distribution of recorded Fatigue Scale for Motor and Cognitive Functions (FSMC) total score (A) and stratified into the categories; no/low, mild, moderate and severe fatigue (B).The distribution of measurements across the four categories in subjects with relapsing-remitting MS (RRMS) and secondary progressive MS (SPMS) is depicted in C.
Model 4: further adjusted for MSIS-29 and EQ-VAS in addition to the variables adjusted for in model 3. Model 5: further adjusted for depression, anxiety, other psychiatric comorbidities, antidepressant use, anxiolytics use, antipsychotic use, sleeping pills use, days hospitalized, history of infections, diabetes, malignant conditions, MACE, arrythmia in addition to the variables adjusted for in model 4. Model 6: further adjusted for central stimulants, amantadine in addition to the variables adjusted for in model 5. Abbreviations: DMT = disease modulatory therapies; EDSS = expanded disability status scale; EQ-VAS = euroqol visual analogue scale; FSMC = fatigue scale for motor and cognitive function; HRQoL = health-related quality of life; MACE = Major adverse cardiovascular event; MS = multiple sclerosis; MSIS-29 = MS impact scale; Q = quartile; RRMS = relapsing-remitting MS; N = number of individuals; SPMS = secondary progressive MS; SDMT = symbol digit modalities test.

Fig. 2 .
Fig. 2. Associations with FSMC total scores.Associations of MS-associated characteristics, cognitive processing speed and the health-related quality of life (HRQoL)-related instruments Multiple Sclerosis Impact Scale (MSIS-29) and EuroQol Visual Analogue Scale (EQ-VAS) with Fatigue Scale for Motor and Cognitive Functions (FSMC) total scores.Mean differences with 95% confidence intervals.DMT= disease modifying therapy; EDSS= Expanded Disability Status Scale (disability level); SDMT= Symbol Digit Modalities Test (cognitive processing speed).

Table 1
Cut-off values for the FSMC total score and sub-domain scores.

Table 2
Characteristics of patients with RRMS and SPMS across FSMC categories collected in the Swedish MS registry from January 2012 to December 2018.
b Diagnosed within five years prior to collected FSMC score.c All mental and behavioral disorders except depression and anxiety disorders.d Ever occurring before FSMC measurement.e dispensed prescription drugs within one year prior to collected FSMC score.

Table 3
Associations of MS-associated characteristics, cognitive processing speed and HRQoL with FSMC total among RRMS and SPMS patients (N=2,165).Mean differences with 95% CIs are presented.
adjusted for MS-type, line of DMT, year of diagnosis, MS-duration, relapses last year, any new cerebral lesions last year, EDSS, any new cerebral lesions last year, SDMT in addition to the variables adjusted for in model 2.

Table 4
Associations of sociodemographic characteristics, sick leave, disability pension, medical history and comorbidities with FSMC total among RRMS and SPMS patients (N=2,165).Mean differences with 95% CIs are presented.
Model 3: further adjusted for MS-type, line of DMT, year of diagnosis, MS-duration, relapses last year, any new cerebral lesions last year, EDSS, SDMT in addition to the variables adjusted for in model 2.Model 4: further adjusted for MSIS-29 and EQ-VAS in addition to the variables adjusted for in model 3. Model 5: further adjusted for depression, anxiety, other psychiatric comorbidities, antidepressant use, anxiolytics use, antipsychotic use, sleeping pills use, days hospitalized, history of infections, diabetes, malignant conditions, MACE, arrythmia in addition to the variables adjusted for in model 4.Model 6: further adjusted for central stimulants, amantadine in addition to the variables adjusted for in model 5. b Restricted to patients 18-64 years old (3179 FSMC total scores) c Diagnosed within five years prior to collected FSMC score.dAll mental and behavioral disorders except depression and anxiety disorders.Dispensed prescription drugs within one year prior to collected FSMC score.Abbreviations: DMT= disease modulatory therapies; EDSS= expanded disability status scale; EQ-VAS= euroqol visual analogue scale; FSMC= fatigue scale for motor and cognitive function; HRQoL= health-related quality of life; MACE= Major adverse cardiovascular event; MS= multiple sclerosis; MSIS-29= MS impact scale; RRMS= relapsing-remitting MS; N= number of individuals; SPMS= secondary progressive MS; SDMT= symbol digit modalities test.
a Model 1: unadjusted Model 2: adjusted for age, sex, geographical region, educational level, country of birth.e Ever occurring before FSMC measurement.f