Change in mobility independence over 5 years for persons with chronic spinal cord injury

Abstract

Objectives

To evaluate change in mobility independence (MI) in community dwelling persons with spinal cord injury (SCI).

Participants

Community Survey.

Design

Cohort study. Rasch analysis was applied to the mobility subscale of the Spinal Cord Independence Measure – Self-Report data from years 2012 to 2017, resulting in a Rasch Mobility Independence Score (RMIS). We employed multilevel modeling to examine RMIS and its change over 5 years, adjusting for demographics and SCI severity; random forest regression was applied to determine the impact of modifiable factors (e.g. environmental factors, home-support) on its change.

Results

The analysis included 728 participants. The majority (≈85%) of participants demonstrated little or no change in RMIS from 2012 to 2017; however, a smaller proportion (15%) showed considerably large change of more than 10 on the 100-point scale. A mixed-effects model with random slopes and intercepts described the dataset very well (conditional R² of 0.95) in terms of demographics and SCI severity. Age was the main predictor of change in RMIS. Considering SCI severity, change in RMIS was related to age for the subgroup with paraplegia, and to time since injury for the subgroup with tetraplegia. No impact of modifiable factors was found.

Conclusion

RMIS in persons with SCI changes over a period of 5 years, especially in elder patients with paraplegia and persons with incomplete tetraplegia with more than 15 years of time since injury. During routine follow-up change in mobility independence should be assessed in order to timely intervene and prevent mobility loss and participation limitations.

Introduction

For persons with SCI, mobility is a prerequisite for independent functioning and participation in all aspects of everyday life such as ability to work, leisure activities (sports, hobbies) and quality of life. Several studies in the UK, Australia and Sweden demonstrated an increase in functional deficits over time, especially in persons with a tetraplegia, of older age, and with secondary health conditions such as shoulder pain, weight gain and fatigue.^1–3 Hinrichs et al. showed, based on the Swiss Spinal Cord Injury Cohort Study (SwiSCI) 2012 survey, that mobility independence (MI) was negatively associated with age in wheelchair users.⁴ Since life expectancy for individuals with SCI has increased,⁵ more elderly persons with SCI and more persons with a longer time since injury (TSI) live in the community. In order to understand which persons are at risk for decline in MI, it is important to study the relationship between MI with chronological aging, but also with TSI. The influences of age and TSI on functional status have always been difficult to delineate because the two variables are often highly correlated.⁴ However, a differentiation between the effects of age and TSI is highly clinically relevant in anticipating problems in MI, and/or in the prescription of assistive technology or extra care. The objective of this study is to examine the relationship of MI with age and TSI, and to examine differences in change in MI over five years across SCI severity subgroups, i.e. lesion level and completeness.

The specific aims of this study are:

1. To describe the relationship between the following: mobility characteristics; non-modifiable person characteristics (socio-demographic traits); and injury characteristics (such as TSI, SCI severity and etiology, wheelchair use), and examine which SCI severity subgroups have a higher likelihood of change (positive, negative) in MI over time.

2. To investigate the prognostic value of modifiable factors associated with a change in MI for the subgroups in aim 1, while controlling for non-modifiable factors.

Methods

We adhere to the guidelines of International SCI Core Data Set committee of ISCoS and to the STROBE guidelines in standardizing the data analysis and reporting the results.^6,7 However, for lesion severity only basic information was addressed in the survey questionnaires, resulting in four categories of paraplegic and tetraplegic lesion levels with complete or incomplete lesions.

Sample

SwiSCI is a nationwide cohort study among the Swiss SCI population, in which participants fill out surveys (among which questions on aspects of functioning, health, well-being and life expectancy) every 5 years, beginning in 2012. SwiSCI includes persons (16 years or older) diagnosed with traumatic or non-traumatic SCI and residing in Switzerland. Details on the study design and the in-/exclusion criteria have been reported in former publications.^8,9 The current study includes participants who answered the relevant modules of both surveys (Survey 2012 and Survey 2017).¹⁰

Outcome measure

Variables

Both surveys used the mobility items from the Spinal Cord Independence Measure – Self-Report (SCIM-SR) shown in Table 1, with the coding and response options.¹¹

Table 1.

SCIM-SR Mobility Subscale Items used in the Rasch analysis with their response coding and corresponding label.

SCIM-SR number	Item Label	SwiSCI code	Response Options
10	Transfers from the bed to the wheelchair	1	I need total assistance.
		2	I need partial assistance, supervision or adaptive devices (e.g. sliding board).
		3	I do not need any assistance or adaptive devices.
		4	I do not use a wheelchair.
11	Transfers from the wheelchair to the toilet/tub	1	I need total assistance.
	Transferring also includes transfers from the	2	I need partial assistance, supervision or adaptive devices (e.g. grab-bars).
	wheelchair or bed to a toilet wheelchair.	3	I do not need any assistance or adaptive devices.
		4	I do not use a wheelchair.
12	Moving around indoors	1	I need total assistance.
	I use a wheelchair. To move around, …	2	I need an electric wheelchair or partial assistance to operate a manual wheelchair.
		3	I am independent in a manual wheelchair.
	I walk indoors and I …	4	need supervision while walking (with or without walking aids).
		5	walk with a walking frame or crutches, swinging forward with both feet at a time.
		6	walk with crutches or two canes, setting one foot before the other.
		7	walk with one cane.
		8	walk with a leg orthosis(es) only (e.g. leg splint).
		9	walk without walking aids.
13	Moving around moderate distances (10–100 meters)	1	I need total assistance.
	I use a wheelchair. To move around, …	2	I need an electric wheelchair or partial assistance to operate a manual wheelchair.
		3	I am independent in a manual wheelchair.
	I walk moderate distances and I …	4	need supervision while walking (with or without walking aids).
		5	walk with a walking frame or crutches, swinging forward with both feet at a time.
		6	walk with crutches or two canes, setting one foot before the other.
		7	walk with one cane.
		8	walk with a leg orthosis(es) only (e.g. leg splint).
		9	walk without walking aids.
14	Moving around outdoors for more than 100 meters	1	I need total assistance.
	I use a wheelchair. To move around, …	2	I need an electric wheelchair or partial assistance to operate a manual wheelchair.
		3	I am independent in a manual wheelchair.
	I walk more than 100 meters and I …	4	need supervision while walking (with or without walking aids).
		5	walk with a walking frame or crutches, swinging forward with both feet at a time.
		6	walk with crutches or two canes, setting one foot before the other.
		7	walk with one cane.
		8	walk with a leg orthosis(es) only (e.g. leg splint).
		9	walk without walking aids.
15	Going up or down stairs	1	I am unable to go up and down stairs.
	I can go up and down at least 3 steps …	2	but only with assistance or supervision
		3	but only with devices (e.g. handrail, crutch or cane).
		4	without any assistance, supervision or devices.
16	Transfers from the wheelchair into the car	1	I need total assistance.
	Transfers include also putting the wheelchair into and	2	I need partial assistance, supervision or adaptive devices.
	taking it out of the car.	3	I do not need any assistance or adaptive devices.
		4	I do not use a wheelchair.
17	Transfers from the floor to the wheelchair	1	I need assistance.
		2	I do not need any assistance.
		3	I do not use a wheelchair.

Participants are classified as “wheelchair-users” or “non-wheelchair-users”. Participants are classified as “non-wheelchair-user” if (a) they answered “I do not use a wheelchair” in all of the “transfer-questions” they responded to (SCIM 10, SCIM 11, SCIM 16, SCIM 17) and (b) stated that they walked (with or without supervision / walking aids) for all three of the “moving around” questions they responded to (SCIM 12, SCIM 13, SCIM 14). All others are classified as “wheelchair users.”

MI and its change over 5 years are the outcomes for this study. We transformed scores from the SCIM-SR mobility items into a 0–100 Rasch Mobility Independence Score (RMIS). Rasch analysis allows for testing core measurement assumptions including item independence, monotonicity and stochastic ordering, uni-dimensionality, and the absence of differential item functioning. If the 8 mobility items of the SCIM-SR fit the Rasch model, the logit-derived scaled person ability estimates provide a scaled interval metric that can then be rescaled into a RMIS ranging from 0 to 100 (the higher the score, the greater the independence). In this study, the Partial Credit Model was applied to test the metric properties of items from the SCIM-SR mobility subscale (Table 1).^12,13 Appendix 1 describes the applied Rasch analysis method and outcome in full detail.

Data analysis

761 Persons completed both Survey 2012 and Survey 2017. Of these, the following were eliminated because of non-plausible response patterns (e.g. reporting a lower age in 2017, N = 13); outlier TSI based on visual inspection (N = 9); the inability of the Rasch model to calculate a score in presence of missing SCIM-SR data (N = 3); at least one missing value for a participant's non-modifiable factor; incomplete cases (N = 8), resulting in an analysis sample of 728 persons. For aim 2, imputation of missing values conserved the number of participants available for the study. The missing value imputation in this study was performed with “MissForest,” a robust multiple imputation method for mixed-type data.¹⁴

For aim 1, to examine the relationship of RMIS with non-modifiable population characteristics, a random effects (RE) mixed-model approach with random intercept and random slope was applied to control for the repeated measures from 2012 to 2017.^15,16 The random slopes inform at the individual level on the change in RMIS from 2012 to 2017, when controlling for non-modifiable person characteristics (demographics, SCI severity).¹⁷ Marginal distribution graphs for age and TSI show the association of relevant predictors with the adjusted RMIS considering the RE of the applied modeling approach. The graphical approach depicts expected RMIS across the age and TSI continua in injury specific subgroups.

Starting from a main effect model and stepwise including interactions and higher order polynomials, the optimal RE model is selected by comparing the changes in the Akaike Information Criterion (AIC) criterion and comparison of its deviation across plural models with a likelihood ratio test.¹⁶ For the optimal model, the marginal R² and conditional R² are calculated with the R-package “Performance.”¹⁸ Marginal R2 is concerned with variance explained by fixed factors, and conditional R2 is concerned with variance explained by both fixed and random factors. We use the generalized variance inflation factor to test the absence of multicollinearity between the covariates of the regressions and Cook's distance to detect leverage outlier observations.¹⁹ The RE analysis was performed with R-package “nlme.”²⁰ For further details see Appendix 2.

For aim 2, to examine the prognostic value of modifiable factors on change in RMIS, a selection of the most relevant modifiable factors assessed in 2012 using random forest regression was followed by a weighted linear regression (with inverse probability weights for non-response correction).^8,21 The estimated slopes represent the change in RMIS, adjusted for non –modifiable factors as in aim 1. The modifiable factors included are:

• International Classification of Functioning (ICF) based functioning scores for mental functions, self-care and involvement in life situations,²²

• SCIM-SR score for self-perceived functional independence,²³

• Nottwil Environmental Factor Inventory – Short Form (NEFI-SF) for the impact of the environment on participation,²⁴

• Utrecht Scale for Evaluation of Rehabilitation – Participation (USER-P) for participation restriction and satisfaction,²⁵

• Health problems (spasticity, respiratory problems, contractures, tiredness, and pain)

• Variables considered relevant such as time in sports activities, home support, living condition, household income, having a paid employment, hours in doing household tasks, and activity limitations.

All of the above variables were selected based on the literature and expert opinion.^4,26–28 Appendix 3 describes all continuous variables with mean and standard deviation (SD) and all ordinal or categorical variables with their frequency and percentages. A random forest regression, which ranks predictors based on an estimate of variable importance, was used for variable selection.²⁹ Only variables that showed consistently high importance were retained for the linear regression.

To investigate the prognostic value of modifiable factors associated with a change in RMIS for the subgroups of aim 1, while controlling for unmodifiable factors, we used bootstraps to assure the stability of the findings, and to provide more accurate confidence intervals for the parameter estimation. The number of bootstraps was set to N = 1000. Results of a graphical analysis were considered as supportive for the robustness of the bootstrapped approach as the distribution of the standardized regression coefficients appeared bell-shaped.

Based on preliminary analyses, consistent high importance of a variable was defined as being among the top-5 ranking variables in 95% of the bootstraps. The results of the repeated regression analyses on single imputed datasets were pooled to report the mean, SD, and range of the regression estimates, as well as the mean standard error of the estimates, the mean and range of t-tests values, and the proportion of significant findings per variable.³⁰ Finally, within each bootstrap, the data was also tested for multicollinearity and outliers.

Results

Sample

The non-modifiable population characteristics in Survey 2012 as well as the RMIS in 2012 and in 2017, and the RMIS change between years 2012 and 2017, are shown in Table 2. Characteristics are provided by SCI severity and wheelchair use. Neurological status of SCI (level and severity) did not change in the majority of the participants (N = 700, 96.15%).

Table 2.

Descriptive statistics of the variables for aim 1, stratified by injury subgroups and wheelchair usage. Only 2012 data is displayed for the non-modifiable predictors.

				Paraplegia		Tetraplegia		Wheelchair user
			Total study sample	Complete	Incomplete	Complete	Incomplete	No	Yes
Non-Modifiable Predictors		N	728	253	270	70	135	174	549
Age	years	mean (SD)	50.9 (13.3)	49.4 (12.5)	52.3 (14.1)	47.8 (10.6)	52.2 (13.9)	53.6 (14.3)	49.9 (12.9)
Time Since Injury	years	mean (SD)	17.2 (12.4)	20.9 (12.8)	13.6 (10.8)	22.5 (11.7)	14.6 (11.9)	10.8 (8.6)	19.3 (12.7)
Sex	male	N (%)	527 (72.4)	193 (76.3)	182 (67.4)	60 (85.7)	92 (68.1)	121 (69.5)	401 (73.0)
SCI Cause Type	traumatic	N (%)	618 (84.9)	231 (91.3)	205 (75.9)	69 (98.6)	113 (83.7)	134 (77.0)	479 (87.2)
Outcome: RMIS
2012	0–100 Score	mean (SD)	48.5 (27.5)	37.9 (9.9)	65.1 (26.6)	18.3 (16.2)	52.4 (33.4)	89.9 (13.1)	35.7 (15.4)
2017	0–100 Score	mean (SD)	47.6 (28.0)	37.0 (11.1)	63.5 (26.9)	17.9 (15.9)	52.7 (35.3)	90.7 (13.0)	35.1 (16.3)
Change (2012–2017)	0–100 Score	mean (SD)	−1.4 (8.9)	−1.3 (7.2)	−1.3 (10.3)	−0.6 (5.4)	−1.9 (10.4)	−1.2 (8.1)	−1.4 (9.1)

Outcome measure

A preliminary analysis supported the psychometric properties of the 8 mobility items from the SCIM-SR. This allowed the derivation of a Rasch score with the Partial Credit Model, resulting in an interval scaled RMIS. The model reliability, as measured with the Person Separation Index (PSI) and Cronbach Alpha (α) was very good (PSI = 0.97, α = 0.94). All item difficulty parameters were invariant across the 2012 and 2017 data collections. See Appendix 1 for further details of the Rasch analysis.

The bubble plot in Fig. 1 shows the proportional distribution of RMIS in 2012 and 2017 overall participants. The bubbles on the diagonal indicate no change, the bubbles above the diagonal indicate an increase and all bubbles below the diagonal indicate a decrease in RMIS from 2012 to 2017. The size of the bubble indicates the proportion of participants with the same RMIS.

Figure 1.

Proportional bubble plot of RMIS. Bubble plot showing the proportional distribution of RMIS in 2012 and 2017 over all participants. The change in RMIS is the deviation from the diagonal, while the size of the bubble indicates the number of participants with the same values for the 2 time points.

A majority of participants (40%) show no change, around 28% show a decrease of less than 10, and around 17% show an increase of less than 10 points on their RMIS. 15% of the participants have a change of more than 10 points on the 100-point scale, either an increase or a decrease.

Aim 1: relationship of RMIS with non-modifiable person characteristics

All predictor variables are free of multicollinearity. A mixed-model analysis including polynomial effects for age and TSI and interactions of age with the SCI severity, and TSI with SCI severity, resulted in the most significant improvement of the model fit over simpler models (See Appendix 3, Likelihood Ratio = 83.78, df = 14, P < 0.001). Mixed-model analysis with only additive effects showed a marginal R² of 0.69 and a conditional R² of 0.94, suggesting that a high proportion of the variance in the RMIS can be explained by non-modifiable factors (demographics and SCI severity). The explained variance for the final model with interactions was even better with a marginal R² of 0.74 and a conditional R² of 0.95. Table 3 depicts the predictors of the RMIS resulting from the final mixed-effect model analysis.

Table 3.

Predictors of the RMIS resulting from the mixed effect model analysis. The RMIS is the dependent variable, random effect are the participants and survey year.

Variable	Reference	Value	SE	DF	T-value	P value
(Intercept)		80.7	1.3	727	60.7	<0.001
Sex	Male	−4.2	1.0	666	−4.1	<0.001
Injury
Paraplegia Incomplete (PI)	Paraplegia Complete (PC)	8.0	1.1	666	7.4	<0.001
Tetraplegia Complete (TC)		−14.9	2.1	666	−7.0	<0.001
Tetraplegia Incomplete (TI)		−2.0	1.3	666	−1.5	0.128
Age
Linear		−121.4	33.3	666	−3.6	<0.001
Quadratic		−79.4	28.2	666	−2.8	0.005
SCI cause type	Non-Traumatic	−0.5	1.2	666	−0.4	0.694
Wheelchair user	No	−42.4	1.1	666	−37.7	<0.001
Time since SCI
Linear		16.0	31.4	666	0.5	0.611
Quadratic		−31.1	22.9	666	−1.4	0.173
Injury×Age
PI×Age Linear	PC×Age Linear	41.0	40.0	666	1.0	0.306
TC×Age Linear		236.1	111.2	666	2.1	0.034
TI×Age Linear		234.3	48.4	666	4.8	<0.001
PI×Age Quadratic	PC×Age Quadratic	51.6	35.0	666	1.5	0.141
TC×Age Quadratic		79.7	81.0	666	1.0	0.325
TI×Age Quadratic		49.1	44.0	666	1.1	0.265
Injury×Time since SCI
PI×TSI Linear	PC×TSI Linear	−43.5	42.2	666	−1.0	0.303
TC×TSI Linear		−241.2	82.1	666	−2.9	0.003
TI×TSI Linear		−300.9	51.7	666	−5.8	<0.001
PI×TSI Quadratic	PC×TSI Quadratic	2.8	32.4	666	0.1	0.932
TC×TSI Quadratic		13.6	60.8	666	0.2	0.823
TI×TSI Quadratic		−153.1	41.7	666	−3.7	<0.001

With the exception of SCI type (traumatic versus non-traumatic), all other selected non-modifiable person characteristics contribute significantly in explaining the RMIS. Female participants show lower RMIS compared to male participants. Having incomplete paraplegia compared to complete paraplegia is associated with higher RMIS, while having complete tetraplegia is associated with lower RMIS. Being a wheelchair user is associated with lower RMIS. The effects of age and TSI are intricate. In terms of model fit, the RMIS is better explained when including age as a linear factor and TSI as a quadratic one. However, as a main effect, only age can be retained as a significant predictor of the RMIS. The interaction of incomplete tetraplegia and TSI became significant with both a linear and a quadratic effect of the TSI. While the linear effect indicates a difference in mean RMIS between the para- and tetraplegic subgroup, the quadratic effect indicates that the decrease in RMIS with increasing TSI tends to “accelerate”. There is also a significant effect for TSI and complete tetraplegia, but the difference is only found, when TSI is entered as a linear term and not in the progression across TSI.

The linear impact of the age was also highly significant in persons with tetraplegia. There was no other highly significant interaction with the other injury types. Figure 2 depicts the RMIS for age and TSI, in interaction with SCI severity. The tick-marks in the lower part of the graph illustrate the frequency of persons of each age or TSI by SCI severity. The confidence intervals of the predicted curves show flaring when going towards the maximum and minimum age or maximum TSI due to case distribution. Most remarkable is the rapid decrease in RMIS for the subgroup of persons with an incomplete tetraplegia with a TSI above 15 years.

Figure 2.

Marginal mean values of the RMIS by age and over TSI. Marginal mean values of the RMIS by age and over TSI, across SCI severity, based on the estimates of the mixed-model from aim 1. The tick-marks in the lower part of the graph illustrate the frequency of persons of each age or TSI by SCI severity. This model output is based on cross sectional data, from 2 time points and thus also considers the change in RMIS from 2012 to 2017. However, the plots should not be interpreted as a longitudinal estimation of RMIS over the whole range of Age or TSI shown.

Aim 2: prognostic value of modifiable factors

The descriptive statistics of the modifiable factors included in the analysis are shown in Appendix 3. Variable importance analysis with random forest regression allowed for reducing the number of predictors. The following 5 predictors (see Table 4) were retained for the prediction of the slopes of change of RMIS: the 0–100 mental health score (MHI-5), the net income of the household, the question about activity limitations, and 2 reported health problems (namely respiration and spasticity). The bootstrapped linear regression revealed that the additional variance in the change of RMIS explained by these 5 predictors is small (R² adj = 0.03). The most stable predictor appears to be respiratory problems, which became significant in 73% of the bootstraps. All other predictors became significant in less than 55% of bootstraps of the regression.

Table 4.

The relevant predictors that were retained in the random forest analysis and their significance in N = 1000 bootstrapped weighted linear regressions including mean and confidence interval for the regression parameters, the T-test for significance and the proportion of significant findings.

Scale	Coding	Mean of Estimate	SD of Estimate	Range Estimate [min / max]	Mean of T-test	SD of T-tests	Range T-test [min / max]	P value (% <0.05)
Mental Health – MHI-5 score	0 = no problems	−0.05	0.03	[−0.13; 0.03]	−2.1	1.12	[−5.83; 1.6]	55%
Activity Limitation question	0 = no limitations	0.05	0.15	[−0.49; 0.51]	0.34	1.07	[−3.53; 3.59]	8%
Resources	Net household income	0.09	0.08	[−0.15; 0.32]	1.36	1.15	[−2.15; 4.73]	32%
Health Problem question
Respiratory	0 = no problem	0.39	0.17	[−0.24; 0.92]	2.58	1.1	[−1.65; 5.97]	72%
Spasticity	0 = no problem	0.08	0.1	[−0.22; 0.45]	0.83	1.03	[−2.31; 4.38]	14%

Discussion

Our study shows that, although the RMIS for the total study sample did not change very much over the five years, there was a remarkable decline in RMIS in persons with incomplete tetraplegia. There are a few other findings in literature on SCIM-SR-based mobility independence, which indicate a slow decline by age,^3,4 but no studies have reported a decline in this subgroup of persons with an incomplete tetraplegia.

Using a mixed-effect approach with random intercept and slopes was essential for this study. Adding the random effects allowed to account for the repeated measures of the individuals and both survey years, when modeling the RMIS with demographics and SCI severity. The model fit to the data was very good, with a marginal R2 of 0.74 and a conditional R2 of 0.95.

Overall, age is the single most important contributor to RMIS decline. In Fig. 2 the marginal plots of the interaction of Age with SCI severity (left panel) and TSI with SCI severity (right panel) are shown. This model output is based on cross sectional data, from 2 time points and thus also considers the change in RMIS from 2012 to 2017. However, the plots should not be interpreted as a longitudinal estimation of RMIS over the whole range of Age or TSI shown. The plot in the left panel shows an unexpected higher RMIS for persons with tetraplegia and higher Age, when compared to those with lower Age. This might be due to the already lower baseline level of RMIS due to lesion level, combined with an improvement in access to aids and assistive technology for mobility. When examining the effect of the interaction of SCI severity with TSI, a subgroup of persons with a remarkably lower RMIS at higher TSI was identified. The subgroup of participants with a complete tetraplegia shows a steady decline over TSI, while the subgroup of participants with an incomplete tetraplegia and TSI more than 15 years shows an even more remarkable decline in RMIS.

Although we did not study what is causing the decline in RMIS (for example shoulder pain, weight gain, fatigue) the current study shows that for individuals with paraplegia, a decline in RMIS appears to be mostly age dependent after the age of 45, while in individuals with an incomplete tetraplegia appear to face a decline in RMIS after 15 years of injury. This finding is relevant for clinical practice, in annual follow-up controls, changes in mobility independence should be actively asked for in those subgroups to detect a possible modifiable cause in an early stage, or in order to adapt equipment for mobility if the cause is found to be non-modifiable.

Considering the modifiable factors associated with a change in RMIS, only one factor appeared consistently across bootstraps as a significant predictor: respiratory problems. However, the adjusted R² of 0.03 indicates that respiratory problems do not explain much of the change in RMIS.

When applying the mixed-effect model, controlling for the non-modifiable factors left only little variance in the change in RMIS that could be possibly attributable to the modifiable factors. However, without the modeling applied, the regression outcomes used for aim 2 would not be discriminative to the effects of non-modifiable factors like demographics and SCI severity, in comparison to modifiable factors.

Limitations and future study

The main limitation of this study is its rather short time span of 5 years between only 2 surveys. From a statistical point of view, longitudinal analysis ideally should comprise more than 3 time points, to enable a clear analysis of the path of change over time. In addition, the distribution of participants limits generalizability of study results, with most participants in age range 42–61 years (Table 2), and TSI between 6 and 26 years.

The RMIS ranges from 0 (complete dependency) to 100 (complete independency), however, a critical threshold for such a RMIS, or a value for a clinically relevant change is not available (yet).

We are aware that some persons might use different methods for mobility in different settings (indoors, outdoors, different distances). In the current study, we allocated persons as wheelchair users or non-wheelchair users. This was done to assure optimal contrast. Since we found that especially persons with incomplete tetraplegia show a remarkable decline in RMIS, a more detailed study of the changes in equipment in different settings would be a logical next step.

Another approach for future research is the analysis of the characteristics of contrasting clusters of participants, the cluster with 10% greatest decrease, and the 10% greatest increase in RMIS. Such an approach might reveal protective factors for maintaining mobility independence over time.

Conclusion

Although RMIS in persons with SCI remained rather stable over the 5 years in the total group, for which age was the single most important contributor to decline, we found two subgroups with a decline in RMIS. For the subgroup with paraplegia, this decline in RMIS appears to be related to age, while for the subgroup with tetraplegia decline in RMIS appears to be related to time since injury, with the strongest effect for the subgroup with incomplete tetraplegia and more than 15 years of time since injury. However, neither clinical thresholds nor modifiable factors for such decline could be indicated in this study.

Statement of ethics

Ethical approval for Survey 2012 was approved by the principal ethics committee on research involving humans of the Canton of Lucerne (KEK Luzern, internal application 11042, approved 28.06.2011) and subsequently endorsed by the additional involved cantonal ethics committees of Cantons Basel-Stadt (EK Basel, internal application 306/11, approved 06.09.2011) and Valais (CCVEM Sion, internal application CCVEM042/11, approved 06.12.2011). Ethical approval for Survey 2017 was granted by the leading ethical institution Ethikkommision Nordwest-und Zentralschweiz (EKNZ, Project-ID: 11042 PB_2016-02608, approved Dec 2016). We certify that all applicable institutional and governmental regulations concerning the ethical use of human volunteers were followed during the course of this research.

Data availability statement

Owing to our commitment to SwiSCI study participants and their privacy, datasets generated during the current study are not made publicly available but can be provided by the SwiSCI Study Center based on reasonable request (contact@swisci.ch).

Disclaimer statements

Contributors All authors have contributed to the conception, data analysis and/or interpretation, writing and reviewing of the manuscript.

Funding This study has been financed in the framework of SwiSCI, supported by the Swiss Paraplegic Foundation.

Conflicts of interest Authors have no conflict of interests to declare.