Introduction

Labour market participation (LMP) of people with spinal cord injury (SCI) varies greatly between individuals: some remain employed until retirement age, while others gradually reduce their working hours or leave the labour market early before reaching statutory retirement age [1,2,3]. Employment rates of people with SCI drop drastically after SCI onset [3], and vary widely across countries ranging from 10.3% in Morocco to 61.4% in Switzerland, as evidenced by a recent study covering 22 countries worldwide [4]. Despite LMP representing a key goal of SCI rehabilitation [5], employment rates of those affected are still between 19.7% (Switzerland) to 59.3% (Brazil) lower compared to the general population [4]. The individual differences in employment trajectories of people with SCI [2, 6, 7] raise the question of why some affected individuals stay permanently employed until retirement age, while others drop out prematurely of the labour market.

Previous research showed that LMP of persons with SCI is influenced by various characteristics at the level of the person and the environment. These characteristics include sociodemographic factors such as age, sex, race, education and pre-SCI work history [8,9,10], injury-related characteristics such as age at and time since SCI, level and severity of injury [11]. Moreover, health-related aspects such as secondary health conditions (e.g. pain or depression), functional independence, psychological factors such as self-efficacy and environmental factors like workplace accessibility and insurance policies represent additional key determinants of LMP [10]. Previous research conducted in Switzerland is in line with most of these findings [7, 12,13,14]. However, due to its prevailing cross-sectional nature and because the few existing longitudinal analyses [2, 6, 7] did not focus on a comprehensive description of the determinants of within-person variation in LMP, little is known about the predictive power of the above-described factors in explaining changes in LMP over time.

The Swiss health care system performs very well regarding indicators such as life expectancy, public satisfaction and perceived quality [15]. The basic self-paid health insurance, which is mandatory, covers a wide range of goods and services for curative and rehabilitative care. The Swiss government supports the system by subsidizing the private insurance providers. Additional insurance providers from the social security scheme such as the Swiss Accident Insurance (Suva) and the Swiss Disability Insurance (IV) fund rehabilitation and vocational integration services that have the goal of returning individuals to the labour market. However, because the Swiss health care system is also highly complex, fragmented and charged with a poor case coordination, the Swiss health care costs and the share of out-of-pocket payments are exceptionally high compared to other European countries [15].

For the majority of the individuals with SCI, the situation is particularly well in the Swiss health care system. Four specialized centres for SCI provide inpatient and outpatient medical, psychological, social and vocational rehabilitation and integration services along individuals’ life course. The initial acute and post-acute rehabilitation is conducted by an interprofessional team, typically takes 3 to 9 months (depending on the SCI severity) and aims to increase the individuals’ autonomy and ability to participate in major life areas [16]. Additional inpatient and outpatient vocational integration and job coaching services support individuals in returning to and maintaining work or during vocational retraining if returning to the pre-injury job is not possible [17, 18]. These services are most often funded by the IV and partly also by the Suva. After discharge from their initial inpatient rehabilitation, individuals may seek further support from ParaHelp (i.e. a specialized home care institution for persons with SCI) or from the Swiss Paraplegic Association (SPA) that provides life and peer counselling and helps with housing, legal and financial issues [19].

However, despite of available services studies within the frame of the Swiss Spinal Cord Injury Cohort (SwiSCI) study, for instance, showed that between 20 and 30% of the affected individuals who initially return to work drop out or withdraw from the labour market before statutory retirement age [1, 12, 14]. Longitudinal evidence on the predictors of change in LMP over time may inform practitioners and policy makers on key targets of interventions that support a sustainable vocational integration of persons with SCI.

The objective of this study was thus to examine change in LMP of people with SCI living in Switzerland. Specifically, we aimed (1) to describe the change in LMP over a time of five years and (2) to investigate the predictors of increase, decrease and stability in LMP over this time.

Methods

Study design and participants

We conducted a longitudinal study using data of individuals with SCI who participated in the 2012 and the 2017 community surveys of the SwiSCI [20]. The SwiSCI community survey aims to collect longitudinal data on all Swiss residents aged over 16 years with a traumatic or non-traumatic SCI and is conducted every five years, starting for the first time in 2012. Details on the SwiSCI study design, sampling strategy and recruitment are provided elsewhere [20]. For the purpose of our study, we included individuals of working age (the statutory working age in Switzerland is 16 to 63 for females and 16 to 64 for males) who participated in the 2012 and the 2017 surveys and who were gainfully employed when they completed the 2012 survey.

Measures

The SwiSCI questionnaire modules are available online [21]. Our outcome variable was change in LMP between 2012 and 2017 of those participants who indicated to be gainfully employed in the 2012 survey. Change in LMP between 2012 and 2017 was operationalized based on information on LMP status and weekly workload that was collected in both surveys with the same multiple-choice question (“What is your current working situation?”). Participants who selected the response options “working for wages with an employer” or “self-employed” were also asked to report their weekly workload in percentages of a full-time equivalent of a standard 42-hour week in Switzerland. Based on this information, participants were assigned to one of the four following groups of LMP change between 2012 and 2017:

  1. (1)

    people who increased their weekly workload,

  2. (2)

    people whose weekly workload remained the same,

  3. (3)

    people who decreased their weekly workload,

  4. (4)

    people who changed their work status from paid work to no paid work.

Data analysis

Variable selection

First, potential predictor variables at the level of socio-demographic, health-related, functioning-related, psychological and environmental factors were selected based on the most recent international and Swiss evidence on determinants of LMP among persons with SCI [7, 12,13,14]. We then checked whether information on these variables was collected in the 2012 SwiSCI community survey and agreed upon our variable selection with vocational integration experts involved in the survey development. Table 1 presents the predictors that were finally included in the analysis, along with the collapsing strategy we applied to overcome the skewed distribution of the response options. All information on our predictor variables is from the 2012 survey.

Table 1 Predictor and outcome variables used in the study and collapsing strategy for variables with categorical response options.
Full size table

Statistical preselection of the predictors of change in LMP was carried out by implementing the least absolute shrinkage and selection operator (LASSO) in a multinomial logistic regression model. The reference category was the group with the same weekly workload in 2017 as in 2012. Similar to backward selection regression, LASSO regression is shrinking the coefficients of non-important predictors (also called discarded predictors) to 0 [22]. Unlike the backward selection regression, LASSO regression has no restriction on the numbers of considered predictors. The selection of predictors with nonzero coefficients (retained predictors) is based on a penalty to the sum of the absolute values of the regression coefficients in the minimization of the residual sum of squares [23]. The larger the value of the penalty, the more predictors are discarded. We selected the value of the penalty using the ten-cross validation procedure [24].

To enhance the stability of the estimated associations, LASSO regression was applied to 100 bootstrap samples with replacement. Each generated bootstrap sample has the same distribution in the LMP as the original data. For each predictor, the number of samples in which it was retained was calculated.

Descriptive statistics and regression analysis

Descriptive statistics of participant characteristics and predictor variables (i.e. absolute and relative frequencies for categorical variables, median and interquartile ranges for continuous variables) were calculated based on the SwiSCI 2012 survey, stratified by the four groups of LMP change.

Four multinomial regressions with predictors selected in 50%, 60%, 70%, and 80% of the 100 bootstrap samples were carried out. The predictors from the model with the smallest Akaike information criterion (AIC), that measures a model’s predictor error [25] were considered as the best predictors of our outcome.

All analyses were performed using the software R version 3.6.0 for Windows [26]. LASSO regression was performed using the R package glmnet [27] and multinomial logistic regression was conducted using the R package nnet [28]. Missing data were imputed using the R package missForest [29], which represents a distribution free missing value imputation technique based on random forests. Variables with more than 20% missing values were excluded.

Results

Sample characteristics

Figure 1 details the constitution of our study sample. A total of 311 participants fulfilled our inclusion criteria, i.e. of working age and gainfully employed in 2012. Almost half of the participants (n = 134) changed their weekly workload from 2012 to 2017: 48 increased, 49 decreased it and 37 dropped out of the labour market. The median age of the three groups was 47, 41 and 52, respectively, and 45 for participants who didn’t change their weekly workload. The median age of a whole sample was 46 and the median age at SCI event 26 years.

Fig. 1: Overview on the study selection and size of our study.
figure 1

Participants selected for the study participated in the 2012 as well as the 2017 SwiSCI community survey, were employed at the time of the 2012 survey and still of working age in 2017 (boxes with bold lines on Fig. 1). Participants were assigned to the following four groups during analysis: (1) people who increased their weekly workload between 2012 and 2017, (2) people who were employed the same weekly workload in 2017 as in 2012, (3) people who decreased their weekly workload between 2012 and 2017, (4) people who changed their work status from paid work to no paid work between 2012 and 2017 (retired prematurely or are unemployed). *These groups were combined for the analysis.

Full size image

Table 2 shows the socio-demographic, SCI-related sample characteristics and descriptive statistics of the predictor variables, stratified by the different groups of LMP change. The average weekly workload of our sample was 57.1% in 2012 and 56.6 % in 2017.

Table 2 Participant characteristics based on the SwiSCI 2012 community survey, stratified by the different groups of LMP change between 2012 and 2017.
Full size table

Predictors of change in LMP

Selection of predictor variables

In the first step of the analysis, 14 predictors where retained by the LASSO regression in at least 50% of the bootstrap samples. These were: age at the time of the 2012 survey, age at the time of SCI onset, having children, years of education, intention to change the current weekly workload, satisfaction with quality of life, satisfaction with daily routine, satisfaction with participation in sports, spasticity, sleep, SCI severity, household income, SCI-related extra-time needs for (a) managing support and (b) outdoor transportation. The mean LASSO coefficients and their confidence intervals calculated across the generated 100 bootstrap samples are provided in the Appendix 1.

Predictors of change in LMP

The multinomial logistic regression model with those predictor variables that were retained in more than 70% of the bootstrap samples showed the best fit (AIC = 674.03 compared to AIC = 690.97 for the 50%, AIC = 677.73 for the 60%, and AIC = 678.89 for the 80% model). The results of the 70% bootstrap model are presented in Table 3, while the ones of the 50%, 60% and 80% models are provided in the Appendices 2–4. The coefficients of the regression analysis describe the estimated change of the relative logit of being in a specific group compared to the reference group (i.e. no change in weekly workload between 2012 and 2017). The coefficients are to be interpreted for one unit change in a continuous predictor variable and for changing from the reference category to a specific other category in a categorical predictor variable, holding all other predictor variables constant. The main results can be summarized as follows:

  1. 1.

    The likelihood of being in the group of participants who increased their weekly workload between 2012 and 2017 as compared to being in the reference group is increased for participants who indicated an intention to work more in 2012.

  2. 2.

    The likelihood of being in the group of participants who decreased their weekly workload as compared to the reference group is lower for participants with children, a higher age at the time of the survey and higher SCI-related extra-time needs for managing support. By contrast, the likelihood is higher for persons who indicate an intention to work less in 2012 and who have more SCI-related extra-time needs for outdoor transportation.

  3. 3.

    The likelihood of being in the group of participants who dropped out of the labour market as compared to the reference group is increased by a higher age at the time of the survey and by the intention to work more in 2012. By contrast, participants with children, more years of education and higher satisfaction with their daily routine are less likely to drop out of paid work compared to those with the same weekly workload as in 2012.

Table 3 Model with those predictors that were selected in 70% of the bootstrap samples in the LASSO regression.
Full size table

Discussion

Based on longitudinal data of community-dwelling individuals with SCI living in Switzerland, we identified a number of predictors of change in LMP over a time period of 5 years. Age, education, having children, intention to change the current weekly workload, satisfaction with daily routine and SCI-related extra-time needs for transportation and managing support were most strongly associated with change in LMP. These factors should receive particular attention in the context of job retention strategies.

Our study contributes to the existing literature by identifying predictors of within-person change in LMP using longitudinal data. Our finding that more than half of the participants did not change their work status between 2012 and 2017 is in line with previous longitudinal SwiSCI research [7]. Along with our finding that 12% of the participants dropped out of the labour market between 2012 and 2017, this suggests a low likelihood of becoming unemployed once individuals have established a stable work situation and implies that dropouts tend to happen more often in the first phase after returning to the labour market or that people do not return to work at all after SCI onset. This highlights the importance of return-to-work and early job retention or coaching programs.

Beyond previous research showing that age, having children and education influences the current work status [7, 9, 12,13,14, 30], we also found these factors to be associated with change in LMP over time. For instance, education is one of the factors that was most consistently reported to positively influence LMP both in Switzerland [7, 12,13,14] and internationally [9]. Our results are also in line with qualitative research that identified having children as well as the need to support them and to act as a role model as a strong motivator for employment [31].

Chronological age was related to change in LMP in two different ways. First, participants who were older at the time of the 2012 survey were less likely to reduce their weekly workload and, second, they were more likely to drop out than to maintain their workload between 2012 and 2017. These seemingly contradictory results might be explained by the median age of the different LMP change groups (41 years for participants who decreased their weekly workload, 52 years for those who dropped out of the labour market and 45 for those who maintained their weekly workload). While middle-aged participants seem to prefer stability, early retirement becomes a more realistic and attractive option for the older ones. Our findings showed that people with SCI would stay employed with the same workload or leave the labour market prematurely than gradually reduce their workload with increasing age to the point of early retirement. It might also be that gradual reduction of weekly workload might not yet be established in the labour market, might not be possible in particular occupations or might not be feasible for individuals working with already small weekly workload.

While it is not surprising that participants who wanted to work more in 2012 were more likely to increase and those who wanted to work less in 2012 more likely to decrease their working hours, it is rather perplexing that the likelihood of labour market dropout is increased among those who wanted to increase their workload in 2012. While this result should be treated with caution due to the small number of people who wanted to increase their weekly workload in 2012 and dropped out by 2017 (n = 4), it nevertheless could be related to the fact that 3 of those 4 participants had only recently sustained their injury (1 to 4 years before the 2012 survey). Therefore, their early dropout could be an indication of an unsuccessful stabilization of their initial work situation. Reflecting on our findings, reducing one’s current workload might also be a meaningful individual strategy to stay longer in the labour market instead of dropping out prematurely because of the accumulation of work-related health issues. Yet the reduction in weekly workload or a drop out from the labour market might also have been the result of environmental factors that are beyond the control of the individual such as an organizational restructuring or a company shutdown. We tried to grasp this complexity by including a variable addressing one’s intention to change the weekly workload. However, the available data and the five-year time interval are not fine-grained enough to draw firm conclusions regarding the voluntary or involuntary nature of these observed changes in LMP.

Contrary to a previous cross-sectional study that found SCI-related extra-time needs for managing support to be a negatively associated with work status [14], we found that the devotion of more time to manage support because of SCI was associated with a lower probability of decreasing the workload. However, this comparison should be treated with caution, as the mentioned cross-sectional study treated this same variable as ordinally scaled (whereas we as continuously) and analyzed a different outcome (i.e. work status and not change in a weekly workload). One possible explanation for our result could be that people who invest more additional time in organizing support might receive more health-related services that help them to maintain their workload. This factor might be specific for Switzerland with its high availability of health care services in general and the exceptional quality of care in the specialized SCI centres in particular. In addition, the effort these people invest in organizing support might be an indication of their high motivation to stay employed. Contrary to our finding on SCI-related extra-time needs for managing support, we found that more extra-time needs for outdoor transportation increased the probability of reducing the workload. Considering the significant amount of time, the person already has to invest in selfcare and other activities resulting from living with SCI, spending extra-time for transportation and commuting might be a barrier to stable LMP.

Satisfaction with one’s daily routine turned out to be a protective factor in our analysis. Participants who were satisfied with their daily routine had a lower risk to drop out from the labour market compared to those in the reference group. Our results thus confirm findings from qualitative studies that have previously identified satisfaction with one’s daily routine and good adjustment to life after injury as important factors [32, 33]. Additionally, motivation to work more turned out to be predictive for increasing the workload.

Contrary to previous cross-sectional evidence on factors associated with work status [13, 14], secondary health conditions did not turn out to be a significant predictor of change in LMP in our study. This contradicting longitudinal finding might be explained by several factors. First, the SCI-SCS response options were collapsed differently in previous studies [13, 14] (no/mild problems vs moderate/significant problems) than in our study (no problem vs mild/moderate/ significant problem). Our decision was made because we aim to provide clinicians and vocational integration professionals with a screening tool that helps to identify people at risk of labour market dropout. Even a mild problem may develop into a moderate or severe one over time and identifying it at early stage may foster timely interventions to prevent its progression and negative effect on labour market participation [33, 34]. Second, a five-year time window between surveys might not fine-grained enough to tackle the influence of secondary health conditions on LMP change, especially because the Spinal Cord Injury Secondary Conditions Scale (SCI-SCS) [35], which was used for assessing the secondary health conditions, asks about their prevalence during the past three months before completing the survey, thus not providing a long-term perspective. Moreover, because we examined change in LMP over time of those who were employed in 2012, our study might have missed individuals with severe health issues who were not employed because of their health state already in 2012.

Strengths and limitations

A major strength of our study is the comprehensive longitudinal data set allowing us to examine determinants of within-person changes of LMP over the time period of five years. Additionally, by identifying not only predictors of reduced LMP, but also factors contributing to stable or increased LMP, this study points towards targets for preventive interventions.

However, our study has also some limitations. First, the small sample size calls for caution with regard to data interpretation, possible implications and generalizability of our findings. In particular, the relatively small number of respondents in some of the outcome groups meant that several predictor response categories were sparsely populated, resulting in wide confidence intervals for the regression coefficients. We tried, however, to mitigate this limitation by using bootstrapping modelling. Second, the lack of time up-dated information on LMP (e.g. change in LMP status and weekly workload) and the predictor variables within the longtime interval of five years between the two measurement limits the scope of our study for properly interpreting within-person changes in LMP. Finally, the specialized facilities for individuals with SCI in Switzerland provide a comprehensive spectrum of acute care, rehabilitation and vocational integration services for persons with SCI that are usually covered by the Swiss health, accident or disability insurers. These system-level factors contribute to Switzerland having the highest employment rate of people with SCI worldwide [4]. The predictors for LMP change we identified in our study may thus not be generalizable to other countries with different health and social security systems and policies.

Practical, policy and research implications

By conducting a longitudinal study on the predictors of within-person change in LMP, our analysis revealed groups at risk of not participating in the labour market as well as protective factors related to the increase of workload. The study thus provides pointers on targets of interventions to support LMP stability in the Swiss SCI population. For example, policy makers should invest in educational programs and vocational integration practitioners should address satisfaction to ensure job retention of people with SCI. Additionally, predictors of reducing the workload or dropping out of the labour market should inform the screening process in job retention programs and outpatient check-ups to identify individuals at risk. When detecting their problematic work situation sufficiently early, individuals could receive individualized support or a re-evaluation of their current job situation to prevent them from getting overburdened and dropping out of the labour market.

Longitudinal life course studies with a sufficiently granular collection of time updated data on work life transitions and trajectories would be beneficial to extend our current knowledge on LMP of individuals with SCI. Additionally, when complemented by qualitative research on the dynamics in individuals’ work life, future research could sharpen our understanding on how to support sustainable work over the life course of those affected.

Conclusion

Based on a longitudinal analysis of a community-dwelling sample of individuals with SCI living in Switzerland, we identified predictors of within-person change in LMP over a time period of five years. Age, education, having children, intention to change the current weekly workload, satisfaction with daily routine and SCI-related extra-time needs for transportation and managing support were most strongly associated with change in LMP. The identified predictors should be taken into consideration in a continuous monitoring at the workplace and during regular check-ups at medical centres to detect risk constellations timely and to subsequently provide support people who are at risk of dropping out of the labour market. In addition, key determinants of LMP stability such as education and satisfaction with daily routine should be taken up by interventions at the level of policy and practice to promote sustainable LMP of persons living with SCI.