Introduction

Multiple sclerosis (MS) is a chronic disease affecting the central nervous system in which those affected typically have intermittent neurologic symptoms and signs early in the course (relapsing-remitting multiple sclerosis [RRMS]) but often, subsequently (or, less commonly, from onset), slowly accumulate disability (progressive multiple sclerosis [PMS]). Currently available tools to measure disability in MS are limited. The Expanded Disability Status Scale (EDSS) is the most widely used clinical outcome measure and is considered by the Food and Drug Administration to be the gold standard for phase 3 trials in PMS. The EDSS, however, is semiquantitative, has limited reliability, and only captures a person’s state at one short point in time during a clinical visit []. People with MS may have different disability states at different times throughout the day, as symptoms can fluctuate with temperature, fatigue, stress, and other factors [,]. The limitations of the EDSS lead to long delays in confirming that a person has transitioned from RRMS to PMS and also inflate the sample size and follow-up time required for MS trials for which disability progression is the primary end point [,].

Triaxial accelerometry is a safe and relatively inexpensive tool that may offer an objective and sensitive measure of disability in people with MS. With the use of an accelerometer worn on the wrist, real-time information about physical activity and circadian rhythmicity patterns can be collected in a person’s natural environment. Such data may allow detection of variation in activity that may be missed during clinical visits [,]. Several studies have looked at the use of accelerometry in MS to identify associations with sleep, fatigue, depression, or disability outcomes (EDSS and Patient-Determined Disease Steps); however, whether wrist-worn accelerometry can differentiate between people with RRMS and PMS remains an existing question [-]. We hypothesized that people with RRMS and PMS have different physical activity and circadian rhythmicity patterns. Due to the interplay between physical activity and circadian rhythm domains within person, we also aimed to study joint and individual variations in these 2 domains and their associations with MS subtype.

MethodsParticipant Selection and Study Design

This study includes an analysis of the baseline data from the longitudinal observational study HEAL-MS (home-based evaluation of actigraphy to predict longitudinal function in multiple sclerosis; Figure S1 in ). Eligible participants followed at the Johns Hopkins Multiple Sclerosis Precision Medicine Center of Excellence were recruited between January 2021 and March 2023. Participants had to have a diagnosis of MS, were aged ≥40 years, had no apparent comorbidities that may limit physical activity (eg, heart failure and end-stage renal disease), had no MS relapse within the last 6 months prior to enrollment, and had baseline EDSS score ≤6.5. These criteria were chosen so that the participants with RRMS were of the age at which transition to PMS might begin within the next few years, the changes in accelerometry measures were more likely to be due to MS and not comorbid conditions, and the baseline measures were not affected by recovery from a recent relapse. Medical records were reviewed by an MS-trained neurologist to ensure eligibility.

Participants were divided into 3 groups with a target of 85 participants each: those with stable RRMS who had no suspected or confirmed progression (RRMS-Stable), those with PMS who had confirmed disability worsening on EDSS (PMS), and those with RRMS who were suspected to be progressing clinically (inferred by an MS-trained neurologist from medical record notes or personal knowledge of the participant) but did not have sustained disability worsening on exam (RRMS-Suspected progression). Confirmed disability worsening on exam was defined as an EDSS change of ≥1.0 point if baseline EDSS was ≤5.5, or of ≥0.5 points if baseline EDSS was ≥6.0; this change should be sustained for ≥24 weeks, not in the context of an explanatory relapse. The 2 RRMS groups were matched based on age (target ±2 years with prespecified ability to relax the criterion if necessary), sex at birth, race or ethnicity, and efficacy class of current disease-modifying therapy (no treatment, first-line treatment [injectables and oral therapies (except for cladribine or ofatumumab)], or higher-efficacy treatment [infusion therapies and cladribine]). The PMS group was expected to be a little older and to be prescribed different (or even no) medications due to the likely lack of neuroinflammation and relapse activity in this group. However, enrollment of a comprehensive cohort of participants with respect to sex and race or ethnicity was attempted.

Clinical Measures

The EDSS exam was conducted after the fitting of the accelerometer and was performed by a masked EDSS-trained physician. A modified multiple sclerosis functional composite (MSFC), which includes the 9-hole peg test, the timed 25-foot walk test, the high- and low-contrast letter acuity (binocular, 2.5% contrast Sloan charts), and the Symbol Digit Modalities Test, was performed by a masked, trained study team member. Participants also completed patient-reported outcomes such as the International Physical Activity Questionnaire (IPAQ).

Accelerometry Measures

Accelerometry metrics were derived from accelerometry data collected with the GT9X Link ActiGraph using a built-in triaxial accelerometer []. All participants were instructed to wear the ActiGraph device on the wrist of their nondominant hand for 24 hours a day over a duration of 2 weeks. Accelerometers were set to capture 3D acceleration at 30 Hz with the acceleration range of ±8 G. The raw acceleration data (.gt3x) were downloaded from the device using ActiLife v6.134 Lite Edition. Binary raw activity data (Hz-level accelerometry data) were read by read.gt3x package into an R data frame (R Foundation for Statistical Computing) and transformed into 60-second epochs activity count data in 1440 minutes per day (12 AM to 11:59 PM) analytic format. The activity counts are vector magnitude-based activity counts.

The following criteria were applied to define valid days: (1) intervals of 90 minutes or longer with all minute-level activity counts equal to 0 were defined as nonwear intervals [], (2) valid days were defined as those with total wear-time ≥90% of the day (≥1296 minutes of wear), and (3) each participant should have ≥3 valid days of accelerometry data.

When a wear period overlapped with the transition to or from daylight saving time, the following adjustments were made. If the clock was set 1 hour forward, 1 hour of data were missing, so imputation was used by averaging the same hour from other valid days for this participant. If the clock was set 1 hour back, duplicate data were generated for that hour; in this instance, we used the average of the duplicated hour, and the loss of 1 hour on the final day was imputed by averaging the same hour from the other valid days for this participant. The wear period overlapped with the transition to or from daylight saving time for only 5 participants.

Accelerometry measures in this study are presented as averages over all valid days and included (1) measures of volume and intensity: total activity count (TAC; daily and 2-hour specific), total log-transformed activity count (TLAC), the total daily number of nonactive minutes, the number of minutes spent in moderate to vigorous physical activity (MVPA), and the number of minutes spent in light intensity physical activity (LIPA); (2) measures of composition: MVPA/LIPA (the ratio of the time spent in MVPA over the time spent in LIPA) and MVPA/nonactivity (the ratio of the time spent in MVPA over the time spent nonactive); (3) measures of fragmentation: sedentary-to-active transition probability (SATP) and active-to-sedentary transition probability (ASTP) [,]; and (4) diurnal landmarks of rest-activity rhythms: average log acceleration during the most active 10 hours of the day (M10), midpoint of M10, average log acceleration during the least active 5 hours of the day (L5), midpoint of L5, daytime activity ratio estimate (DARE), amplitude, relative amplitude (RA), midline estimating statistic of rhythm (MESOR), acrophase, intradaily variability (IV), and interdaily stability (IS) [].

The following sample-specific cut points were developed for discriminating between nonactive, LIPA, and MVPA minutes: (1) nonactive minute was defined as a minute with activity counts ≤2000, (2) LIPA minute was defined as activity counts >2000 and ≤6750, and (3) MVPA minute was defined as activity counts >6750. These cut points were specific to our cohort and defined based on an internal calibration procedure that maximized concordance between accelerometry-based estimates and the participants’ self-reported physical activity outcomes in the IPAQ (Table S1 and Figure S2 in []). Our derived cut points were similar to those defined in other population cohorts [-]. Rather than using predetermined cutoffs reported in the literature, we believe this method was more suitable for this study due to different cohort demographics and clinical conditions that may alter average physical activity intensity.

Statistical Analyses

All statistical analyses were conducted using R (version 4.3.1; R Foundation for Statistical Computing). Descriptive statistics of the demographics, clinical data, and accelerometry measures for each of the 3 groups were summarized as mean (SD) or median (IQR). Differences between the combined RRMS group (RRMS-Stable and RRMS-Suspected progression) and PMS group were compared using t tests and Pearson χ² test as appropriate. After adjusting for age, sex, and BMI, a logistic regression model was fitted to evaluate the association of each accelerometry metric with odds of PMS (vs RRMS). We also evaluated the association of accelerometry metrics in differentiating the RRMS subtypes (RRMS-Suspected progression vs RRMS-Stable). To account for differences in duration and timing of sleep, sensitivity analyses were performed by re-exploring the same accelerometry-derived measures within the most active 10 hours of each day (M10) rather than the 24-hour daily total.

Functional Principal Component Analysis

Functional principal component (fPC) analysis, applied to diurnal rest-activity rhythms (minute-level activity profiles), captures main patterns of temporal allocation of activity and provides important diurnal landmarks. The choice of the first 5 fPCs as measures of circadian rhythmicity is based on 2 factors. First, they capture most of the diurnal variability (76%) in our data. Second, the use of fPC analysis in accelerometry studies of diurnal rest-activity rhythms in large national cohorts such as the National Health and Nutrition Examination Survey (NHANES) and the UK Biobank mostly focuses on the first 4 fPCs since functional principal components beyond the 4th typically capture a relatively low proportion of variance in the data, and their shapes tend to be very difficult to interpret due to their very dynamic shape [-]. Nevertheless, in our case, the fifth fPC explained a significant proportion of the total variability (4%) and was also included. Figure S3 in shows that the additional fPCs beyond the 5th did not provide meaningful explanations of variability.

Joint and Individual Variation Explained

The Joint and Individual Variation Explained (JIVE) is an integrative dimension reduction technique that can be applied to multiple features grouped within several domains []. The physical activity (PA) domain was characterized by measures of total volume of physical activity, times spent in different intensities of activity and composition ratios of those times, measures of activity fragmentation/continuity, and temporal-local measures of total activity presented in 2-hour bins. The circadian rhythmicity (CR) domain was characterized by measures of the strength of circadian rhythms (RA and related components such as M10 and timing of M10), measures of variability including IS and IV, parametric measures extracted from the cosinor model (MESOR, amplitude, acrophase), and the first 5 fPCs. There is a significant amount of interdependence between the PA and CR domains since the measures belong to the same participants over the same time frame. Variations in PA and CR domains that are independent from each other are also expected. JIVE is a model developed to explore such situations and to decompose the joint and individual effects []. demonstrates the conceptual representation of the benefits of JIVE to better understand the association between MS subtype and the joint and individual variation of PA and CR domains. We applied JIVE to the domains of PA and CR to create three groups of latent variables that explain (1) joint variation shared across PA and CR, (2) individual variation specific to each of the domains, and (3) remaining unexplained variation.

‎Figure 1. Conceptual representation of associations between multiple sclerosis (MS) subtype and the physical activity (PA) and circadian rhythm (CR) domains performed in standard regression modeling (left) versus the associations between MS subtype and 3 sets of independent (uncorrelated) latent variables representing joint-PA-CR, individual-PA, and individual-CR information after JIVE (Joint and Individual Variation Explained) decomposition (right). Regression analysis with JIVE components as predictors can reveal and distinguish joint and individual associations between the PA and CR domains and MS subtype. Ethical Considerations

The study was approved by the institutional review board at Johns Hopkins University (IRB00243681). All participants provided written informed consent and had the ability to opt out at any point. All data are stored on REDCap (Research Electronic Data Capture), and only study members who are approved by the institutional review board have access to the data. The data are deidentified to the greatest extent possible. All participants received US $40 as compensation for the baseline visit to offset study-related travel and parking costs. All study procedure costs, including shipping of the ActiGraph, were covered as part of the study.

ResultsDemographics and Clinical Outcomes

A total of 275 people with MS completed a baseline study visit as part of the HEAL-MS study; 253 had retrievable accelerometry data and were included in this baseline analysis (16 never returned the device, while 6 had insufficient valid days). Participants were divided into three groups: 86 with RRMS-Stable, 82 with RRMS-Suspected progression, and 85 with PMS. Baseline visits occurred between January 2021 and March 2023. shows the demographic and clinical data for the 3 groups.

Table 1. Demographics and clinical data collected at the baseline visit for HEAL-MS participants.RRMSPMS (n=85)P valueRRMS-Stable (n=86)RRMS-Suspected progression (n=82)Demographics Age (years), mean (SD)53 (7)54 (8)57 (9)<.001 Sex (female), n (%)61 (71)61 (74)57 (67).36 Race, n (%).67  White64 (74)63 (77)67 (79)  Black19 (22)14 (17)13 (15)  Other3 (4)5 (6)5 (6) Hispanic or Latino, n (%)4 (5)3 (4)3 (4).79 BMI, mean (SD)29.4 (6.4)28.2 (5.6)28 (6.2).35 Employment status, n (%).18  Full time45 (58)37 (51)32 (43)  Part time4 (5)6 (8)7 (9)  Homemaker7 (9)2 (3)4 (5)  Retired13 (17)13 (18)18 (24)  Unemployed3 (4)3 (4)0 (0)  Disability6 (8)12 (16)14 (19)Clinical data EDSS score, median (IQR)2.0 (1.5-3.0)3.0 (2.0-3.5)5.0 (3.5-6.0)<.001 MSFC, median (IQR)  25-foot walk4.5 (4.0-5.1)5.3 (4.3-6.2)7.2 (5.8-9.5)<.001  D-9hpg21.3 (19.4-24.0)23.8 (20.9-27.3)26.9 (22.1-34.4)<.001  ND-9hpg22.7 (20.8-25.2)25.5 (21.9-28.9)27.4 (23.5-34.6)<.001  SDMT51 (47-59)48 (38-57)42 (34-50)<.001 Tremor, median (IQR)0 (0-0.5)0 (0-0.5)0 (0-1).20

aHEAL-MS: home-based evaluation of actigraphy to predict longitudinal function in multiple sclerosis.

bRRMS: relapsing-remitting multiple sclerosis.

cPMS: progressive multiple sclerosis.

dP value from t test or Pearson χ² test comparing RRMS (RRMS-Stable and RRMS-Suspected progression) vs PMS.

eP<.05

fEDSS: Expanded Disability Status Scale.

gMSFC: multiple sclerosis functional composite.

hD: dominant hand.

i9hpg: 9-hole peg test.

jND: nondominant hand.

kSDMT: Symbol Digit Modalities Test.

lBain Score for Tremor Severity (BSTS) was used for tremor evaluation.

Accelerometry Measures Over 24 Hours

Figure S4 in shows the raw accelerometry data for 1 participant over the course of the week. The fPCs captured important temporal or diurnal landmarks () and are comparable to those captured in large national surveys such as UK Biobank and NHANES [,]. Accelerometry metrics for the 3 groups are shown in . TAC and MVPA, the 2 main measures of daily physical activity volume, as well as M10, amplitude, and MESOR, measures of circadian rhythmicity, were significantly different between the RRMS and PMS groups, with greater intensity of activity in the RRMS groups compared to the PMS group. ASTP was also significantly different, with higher values (more frequent transitions to sedentary time) in the PMS group. When TAC was divided into 2-hour windows, it was significantly lower in the PMS group at specific times of the day (6 AM to 12 PM and 2 PM to 10 PM; ). Figure S5 in demonstrates the correlation plot for all accelerometry measures; the green box demonstrates the temporal profiles of the circadian measures associated with TAC over 24 hours. MESOR, amplitude, and fPC1 are highly correlated with TAC between 9 AM and 8 PM.

‎Figure 2. First 5 functional principal components (fPCs). fPC 2, 3, and 4 values are multiplied by −1. Table 2. Mean (SD) baseline accelerometry metrics over 24 hours for HEAL-MS participants.RRMSPMS (n=85)P valueRRMS-Stable (n=86)RRMS-Suspected progression (n=82)Metrics of physical activity volume and fragmentationTAC (x103)2113 (566)2181 (724)1909 (717).008TLAC6277 (868)6278 (933)6196 (999).51Nonactive minutes1067 (102)1052 (114)1087 (130).08MVPA59 (34)61 (52)41 (43)<.001LIPA314 (84)327 (83)312 (105).52SATP0.09 (0.03)0.09 (0.03)0.09 (0.03).21ASTP0.27 (0.06)0.26 (0.06)0.30 (0.11).003Metrics of circadian rhythmM102482 (665)2540 (844)2161 (806)<.001Midpoint of M102:20 PM (85)2:15 PM (86)2:14 PM (77).72L5111 (67)135 (109)146 (124).09Midpoint of L53:58 AM (68)3:58 AM (80)4:05 AM (76).57RA0.91 (0.05)0.89 (0.07)0.87 (0.09)<.001DARE0.70 (0.05)0.70 (0.05)0.69 (0.05).049Amplitude2290 (1081)2218 (847)1923 (839).008MESOR1313 (623)1264 (390)1120 (429).01Acrophase14.56 (1.92)14.61 (1.35)14.76 (1.31).40IV0.52 (0.11)0.49 (0.09)0.49 (0.08).02IS0.26 (0.07)0.27 (0.07)0.25 (0.06).08fPC11912 (17,574)4242 (22,747)−5206 (21,441).003fPC2933 (11,249)−1596 (11,922)507 (10,181).59fPC3235 (9505)−166 (8702)−108 (6432).89fPC4−1489 (6339)−585 (8694)451 (4989).10fPC549 (6677)−489 (5772)677 (5744).27

aHEAL-MS: home-based evaluation of actigraphy to predict longitudinal function in multiple sclerosis.

bRRMS: relapsing-remitting multiple sclerosis.

cPMS: progressive multiple sclerosis.

dP value from t test comparing RRMS (RRMS-Stable and RRMS-Suspected progression) vs PMS.

eP<.05

fTAC: total activity count.

gTLAC: total log-transformed activity count.

hMVPA: moderate to vigorous physical activity.

iLIPA: light intensity physical activity.

jSATP: sedentary-to-active transition probability.

kASTP: active-to-sedentary transition probability.

lM10: average log acceleration during the most active 10 hours of the day.

mMidpoints of M10 and L5 are represented as time of the day in hour:minutes (±SD in minutes).

nL5: average log acceleration during the least active 5 hours of the day.

oRA: relative amplitude.

pDARE: daytime activity ratio estimate.

qMESOR: midline estimating statistic of rhythm.

rIV: intradaily variability.

sIS: interdaily stability.

tfPC: functional principal component.

‎Figure 3. Total activity counts (TAC) for each 2-hour interval over the course of 24 hours for PMS versus RRMS. The 2-sample t test P values show the 2-hour intervals when TAC was significantly different between the 2 groups. As an additional reference, the logistic regression P values are shown with the TAC 2-hour intervals used as predictors, after adjusting for age, sex, and BMI. MS: multiple sclerosis; PMS: progressive multiple sclerosis; RRMS: relapsing-remitting multiple sclerosis.

In multivariable-adjusted models (), participants with PMS had lower TAC (β=−0.32, 95% CI −0.61 to −0.03; P=.03), lower MVPA (β=−.01, 95% CI −0.02 to −0.004; P=.004), lower M10 (β=−0.0006, 95% CI −0.001 to −0.0002; P=.003), and higher ASTP (β=5.68, 95% CI 1.86-9.50; P=.004) when compared to those with RRMS. PMS (vs combined RRMS) was also associated with lower TAC from 8 AM to 12 PM and from 2 PM to 8 PM (). Lower values of compositional ratios MVPA/LIPA and MVPA/nonactive were also associated with PMS (vs RRMS), indicating that individuals with PMS tend to spend relatively less time in moderate/vigorous activity compared to the time in light intensity activity or nonactivity (). Participants with PMS also had lower amplitude (β=−0.0004, 95% CI −0.0008 to −0.0001; P=.01), lower MESOR (β=−0.0009, 95% CI −0.002 to −0.0002; P=.01), higher IV (β=4.64, 95% CI 1.45-7.84; P=.004), lower IS (β=−4.43, 95% CI −8.77 to −0.10; P=.045), and lower fPC1 (β=−0.39, 95% CI −0.69 to −0.09; P=.01) (). There were no significant associations of accelerometry measures between RRMS subgroups except for a difference in RA (β=−5.26, 95% CI −10.80 to −0.20; P=.049), with lower RA in RRMS-Suspected progression (; Figure S6 in ).

Table 3. Multivariable logistic regression model to evaluate the association of each accelerometry metric with multiple sclerosis subtype. The multivariable logistic regression models were adjusted for age, sex, and BMI.PMS vs RRMS (reference group)RRMS-Suspected progression vs RRMS-Stable (reference group)EstimateP value95% CIEstimateP value95% CITAC,−0.32.03−0.61 to −0.030.14.42−0.20 to 0.50TLAC,−0.04.79−0.31 to 0.230.05.77−0.28 to 0.38Nonactive minutes0.002.16−0.0007 to 0.004−0.002.29−0.005 to 0.001MVPA−0.01.004−0.02 to −0.0040.002.70−0.006 to 0.009LIPA−0.0006.69−0.004 to 0.0020.002.25−0.002 to 0.006SATP−3.004.59−13.84 to 7.831.60.81−11.16 to 14.47ASTP5.68.0041.86 to 9.50−4.44.13−10.47 to 1.05MVPA/LIPA−4.25.002−6.99 to −1.51−0.31.81−2.91 to 2.27MVPA/nonactive−7.90.02−14.53 to −1.272.35.46−3.73 to 9.04LIPA/nonactive−0.36.74−2.50 to 1.781.74.22−0.98 to 4.58M10−0.0006.003−0.001 to −0.00020.0001.63−0.0003 to 0.0005Midpoint of M10−0.0004.82−0.004 to 0.003−0.0006.75−0.004 to 0.003L50.002.11−0.0005 to 0.0050.004.070 to 0.008Midpoint of L50.0009.71−0.004 to 0.005−0.004.16−0.01 to 0.002RA−6.00.002−9.73 to −2.26−5.26.049−10.80 to −0.20DARE−5.80.04−11.25 to −0.34−0.34.91−6.42 to 5.74Amplitude−0.0004.01−0.0008 to −0.0001−0.0001.59−0.0004 to −0.0002MESOR−0.0009.01−0.002 to −0.0002−0.0002.48−0.0009 to 0.0004Acrophase0.08.37−0.10 to 0.270.02.84−0.17 to 0.20IV4.64.0041.45 to 7.84−0.30.88−4.21 to 3.62IS−4.43.045−8.77 to −0.100.53.82−3.97 to 5.03fPC1−0.39.01−0.69 to −0.090.14.39−0.19 to 0.47fPC20.04.79−0.23 to 0.31−0.25.12−0.58 to 0.07fPC30.08.57−0.19 to 0.350.009.96−0.30 to 0.32fPC40.19.18−0.08 to 0.460.11.51−0.21 to 0.42fPC50.15.27−0.12 to 0.42−0.05.74−0.36 to 0.26

aPMS: progressive multiple sclerosis.

bRRMS: relapsing-remitting multiple sclerosis.

cP<.05.

dRegression inputs were scaled by dividing TAC and TLAC values by 1 SD.

eTAC: total activity count.

fTLAC: total log-transformed activity count.

gMVPA: moderate to vigorous physical activity.

hLIPA: light intensity physical activity.

iSATP: sedentary-to-active transition probability.

jASTP: active-to-sedentary transition probability.

kM10: average log acceleration during the most active 10 hours of the day.

lL5: average log acceleration during the least active 5 hours of the day.

mRA: relative amplitude.

nDARE: daytime activity ratio estimate.

oMESOR: midline estimating statistic of rhythm.

pIV: intradaily variability.

qIS: interdaily stability.

rfPC: functional principal component.

Accelerometry Measures Within M10

To account for differences in duration and timing of sleep, the accelerometry metrics were explored within the M10 period. The results are summarized in Table S2 in . TAC, MVPA, and ASTP remained significantly different between the RRMS and PMS groups. Additionally, the PMS group displayed more nonactive time compared to the RRMS group. Logistic regression identified significant associations between the MS subtype (PMS vs RRMS) and TAC, MVPA, nonactive minutes, ASTP, and MVPA/LIPA (Table S3 in ). No significant associations were identified when comparing the 2 RRMS subgroups.

The JIVE Components

We also applied JIVE, a novel integrative dimension reduction technique. presents the conceptualized diagram of JIVE decomposition and how JIVE can be used to better understand codependencies between PA and CR domains and their joint and individual association with MS subtype. shows the estimated JIVE components obtained from the accelerometry-derived domains of PA and CR. The directions of loadings (+ or −) and squares of loadings for the estimated JIVE joint component and individual PA and CR components are shown in and , and Table S4 in , respectively. Higher positive scores of the first joint JIVE component primarily represent lower volume of physical activity (captured by TAC) and weaker circadian rhythms (captured by M10, MESOR, amplitude, and fPC1) (). Higher positive individual PA-1 scores capture less frequent transitions to active behavior (SATP), less time spent in LIPA (LIPA, LIPA/nonactive, TLAC), and more time spent nonactive. Higher positive individual PA-2 scores capture more time spent in MVPA (MVPA, MVPA/LIPA, MVPA/nonactive) and more activity in the morning and evening hours (TAC 8 to 10 AM and TAC 6 PM to 12 AM) (). Joint JIVE component and individual PA-1 and PA-2 components were found to be significantly associated with MS subtype (). No significant associations were identified when comparing the 2 RRMS subgroups.

‎Figure 4. Joint and Individual Variation Explained (JIVE) by the 2 JIVE accelerometry-derived domains of circadian rhythm (CR) and physical activity (PA). Table 4. JIVE joint component with variables with more than 5% of proportional variation.VariableJoint component loadingRelative magnitude of joint component loadingM10−0.26653850.0710428fPC1−0.26082760.0680310MESOR−0.25195080.0634792Amplitude−0.24089920.0580324TAC−0.22565100.0509184

aJIVE: Joint and Individual Variation Explained.

bM10: average log acceleration during the most active 10 hours of the day.

cfPC: functional principal component.

dMESOR: midline estimating statistic of rhythm.

eTAC: total activity count.

Table 5. Individual JIVE components of physical activity with variables with more than 5% of proportional variation.VariablePA component loadingRelative magnitude of PA component loadingPhysical activity component 1SATP−0.44937030.2019337LIPA−0.41411490.1714912LIPA/nonactive−0.36649750.1343204TLAC−0.33659270.1132946MVPA/LIPA0.32931820.1084505Nonactive minutes0.24481850.0599361Physical activity component 2MVPA/LIPA0.47289960.2236340MVPA0.39701610.1576218MVPA/nonactive0.36734620.1349432TAC 8 PM to 10 PM0.33415790.1116615TAC 8 AM to 10 AM0.30994020.0960629TAC 10 PM to 12 AM0.24771340.0613620TAC 6 PM to 8 PM0.24302020.0590588Physical activity component 3TAC 12 AM to 2 AM0.60356900.3642955TAC 10 PM to 12 AM0.45509180.2071086TAC 4 AM to 6 AM−0.35006180.1225433TAC 2 AM to 4 AM−0.30850890.0951777Physical activity component 4ASTP−0.51348490.2636667TAC 8 PM to 10 PM0.38035490.1446698TAC 2 PM to 4 PM−0.35885930.1287800SATP−0.32833070.1078011TLAC−0.27391820.0750312TAC 10 AM to 12 PM0.25166890.0633372

aJIVE: Joint and Individual Variation Explained.

bSATP: sedentary-to-active transition probability.

cLIPA: light intensity physical activity.

dTLAC: total log-transformed activity count.

eMVPA: moderate to vigorous physical activity.

fTAC: total activity count.

gASTP: active-to-sedentary transition probability.

Table 6. Multivariable logistic regression model to evaluate the association of each JIVE component with multiple sclerosis subtype. The multivariable logistic regression models were adjusted for age, sex, and BMI.PMS vs RRMS (reference group)RRMS-Suspected progression vs RRMS-Stable (reference group)EstimateP value95% CIEstimateP value95% CIJoint component0.3665.010.0881 to 0.6558−0.1266.44−0.4525 to 0.1897PA-1−0.441.003−0.7397 to −0.159−0.0183.91−0.3404 to 0.3034PA-2−0.4147.006−0.7174 to −0.1255−0.3066.07−0.6502 to 0.0207PA-3−0.0527.71−0.3282 to 0.2204−0.2622.11−0.5887 to 0.0529PA-40.235.11−0.046 to 0.52720.1849.25−0.129 to 0.5098CR-10.2588.07−0.0162 to 0.54110.0107.95−0.3044 to 0.3275CR-2−0.2732.052−0.5536 to 0.0001−0.2991.07−0.6344 to 0.019CR-3−0.142.32−0.4266 to 0.1350.0815.61−0.2345 to 0.4031

aJIVE: Joint and Individual Variation Explained.

bPMS: progressive multiple sclerosis.