Predictors of Paid Employment Status Among Individuals With Spinal Cord Injury in South Korea: An Exploratory Study Using Random Forest Analysis

Abstract

Objective

To explore factors affecting the paid employment of individuals with spinal cord injury (SCI) in South Korea.

Methods

Data collected from previously and currently community-dwelling adults with SCI (n=538) who responded to the 125-item self-reported International Spinal Cord Injury Survey questionnaire were analyzed using random forest and decision tree modeling.

Results

In total, 75.1% responders were male, 53.9% were unmarried, 90.1% lived with family, and 84.8% received daily living assistance; the most frequent education level was high school or below (56.1%), and the percentage of participants with paraplegia and complete injuries was 57.2% and 58.7%, respectively. Average age and mean SCI duration were 46.31 and 15.25 years, respectively. Factors with a positive effect on engaging in paid employment included greater total monthly household income, younger age, higher self-efficacy, better quality of life and satisfaction, access to vocational rehabilitation services, absence of disability benefits, no need for daily living assistance, greater independence, no difficulties engaging in activities, absence of sleep disturbances, higher education level, and being unmarried.

Conclusion

Employing the random forest method and a large dataset, we identified key predictors of paid employment, corroborating previous findings and establishing a foundation for policy development for employment opportunities.

GRAPHICAL ABSTRACT

INTRODUCTION

Spinal cord injury (SCI) often results in lifelong disabilities, functional impairments, and complications [1], significantly restricting social participation and employment prospects [2]. Employment is crucial for livelihood and remaining independent, providing a sense of satisfaction and accomplishment as well as social inclusion and connections through goal-oriented social engagement [3].

Although the prevalence of SCI in South Korea is estimated to be around 85,000, the exact number remains uncertain because individuals with SCI are not separately classified within the broader population of people with physical disabilities [4]. However, approximately 1,100 new cases are reported annually in South Korea, and this number has been increasing over time [5].

Employment among individuals with SCI has been widely studied across countries [6,7], which has been shown to improve financial circumstances and foster social integration and security, boost self-esteem as active members of society, and help overcome the prevailing stereotypes associated with disability [7]. Furthermore, studies have indicated that employment can alleviate pain and depression and reduce negative thoughts [8]. A Korean qualitative study on job experiences found that employment leads to positive psychological changes, increased life satisfaction, and a shift from negative to positive societal perceptions [9].

However, the employment rate for people with disabilities aged ≥15 years in Korea in 2022 was 36.4%, similar to the 35%–40% employment rate for people with SCI in high-income countries [10,11]. Furthermore, the employment rate for individuals with severe disabilities was only 22.9%, and those with physical disabilities, including SCI, was 36.8% [12], compared with the overall population employment rate of 63% [13]. Employment challenges for individuals with SCI are often attributed to health-related complications such as urinary infections, pressure injuries, autonomic dysreflexia, and depression [6]. However, the relevant research findings are somewhat inconsistent—a US cohort study comparing pre-injury and post-injury employment rates found no significant differences [14], and a Korean study found no significant impact of physical functions (sensory functions and pain, and cardiovascular, respiratory, genitourinary, and digestive functions) and structures on employment in its analysis of factors impacting work and employment based on the International Classification of Functioning, Disability and Health (ICF) [15]. Whereas studies in other countries often report a strong positive correlation between pre-injury education level and employment rate [16], a Korean study found no significant difference in employment rates based on education level [17].

Thus, although several studies have identified various demographic, personal, physical, and environmental factors as key determinants of employment among individuals with SCI in other countries [14,16,18], there is a need for further research on these topics in the Korean context. Consequently, this study aimed to comprehensively investigate factors affecting paid employment among individuals with SCI in South Korea by considering diverse categories, including demographic, personal, and environmental characteristics; SCI lesion severity; mental and physical health; social activity and participation; independence; and quality of life. This study seeks to establish a foundational evidence base to support employment initiatives for this population.

METHODS

Data collection and participants

The ICF-based International Spinal Cord Injury Survey (InSCI Survey) was conducted in 22 countries in 2017–2019. In South Korea, a nationwide survey targeting previously and currently community-dwelling individuals with SCI was conducted from March to October 2017. Participants were identified through the National Rehabilitation Center and the Korean Spinal Cord Injury Association databases. Of 6,355 individuals contacted via text messages, emails, phone calls, face-to-face interactions, and websites of hosting institutions, 892 completed the survey. The survey was conducted in accordance with the InSCI Study Center’s protocol, requiring a minimum sample size of 200 per country, based on person ability scores in five relevant ICF domains using Rasch analysis of data from the Swiss Spinal Cord Injury (SwiSCI) community survey [19].

The 125-item self-reported survey questionnaire was administered electronically, in-person, or via email and was tailored to enhance participant convenience. When self-reporting was not feasible, an interview was conducted by a researcher.

Inclusion criteria were as follows: a history of community dwelling, current hospitalization not for the initial acute or rehabilitation care related to the disability, age ≥19 years, medical history of traumatic or non-traumatic SCI (American Spinal Injury Association Impairment Scale category A–D), and consent to participate in the survey. Exclusion criteria were congenital cause of SCI, no community-living experience, or any neurodegenerative disease or peripheral nerve damage.

This study was approved by the Institutional Review Board of the National Rehabilitation Center (approval number: NRC-2024-02-010), and approved a waiver of informed consent, as the study involved analysis of de-identified previously collected data.

Data collection tool

The InSCI Survey incorporated 47 ICF domains, including the Brief ICF Core Set for SCI and ICF Rehabilitation Set, which were specifically designed to reflect the living conditions of individuals with SCI [19]. The resulting 125-item questionnaire was translated into Korean and refined through rigorous consultations and discussions among experts by utilizing various data collection tools known for their reliability and validity. Key domains assessed included energy and mood, secondary health issues, social activity and participation, independence in performing activities of daily living (ADLs), employment, environmental factors, personal factors, quality of life, and health status.

Secondary health issues were assessed using the 36-Item Short Form Survey (SF-36; vitality and mental health domains), Brief Pain Inventory, and Spinal Cord Injury Secondary Conditions Scale. Activity and participation were assessed using the World Health Organization (WHO) Model Disability Survey (MDS), Spinal Cord Independence Measure for Self-Report, and Spinal Cord Injury Functional Index.

Employment status was assessed using modified MDS and SwiSCI version, and International Labour Market Integration Assessment in SCI surveys. Environmental factors were assessed using the Nottwil Environmental Factors Inventory Short Form and MDS. Personal factors were assessed using the General Self-Efficacy Scale, Moorong Self-Efficacy Scale, General Belongingness Scale, and MDS items. Health utility and quality of life were assessed using the SF-36 General Health Scale and 5-item WHO Quality of Life Brief Version instruments.

Analysis

The dependent variable was employment status, defined as whether the respondent was engaged in paid work at the time of the survey, including all forms of paid employment such as self-employment and part-time work. To address missing data, a listwise deletion approach was applied to both dependent and independent variables. A total of 107 predictive variables were included in the analysis (Table 1).

Table 1.

Details of the items extracted as predictors for the random forest analysis

Domain (no. of items)	Item (sequential code)
Demographics (8)	Sex, age, marital status, household membership, need for assistance in ADLs, education level, net monthly household income, and subjective social status
SCI lesion (4)	Degree of paralysis, complete injury, cause of SCI, duration of injury
Energy and feelings (9)	Vigorous, full of life (Q15), nervous (Q16), down in the dumps (Q17), calm and peaceful (Q18), depressed (Q20), worn out (Q21), happy (Q22), and tired (Q23)
Health problems (15)	Sleep problems (Q24); bowel dysfunction (Q25); urinary tract infections (Q26); bladder dysfunction (Q27); sexual dysfunction (Q28); contractures (Q29); muscle spasms, spasticity (Q30); pressure sores, decubitus (Q31); respiratory problems (Q32); injury caused by loss of sensation (Q33); circulatory problems (Q34); autonomic dysreflexia (Q35); postural hypotension (Q36); pain (Q37); and current smoking status (Q40)
Activity and participation (18)	Difficulties performing ADLs (Q41), stress management issues (Q42), problems using hands and fingers (Q43), problems finding a destination (Q44), difficulties using public transportation (Q45), difficulties using private transportation (Q46), health management issues (Q47), difficulties performing housework chores (Q48), providing care or support (Q49), difficulties in interactions (Q50), difficulties in intimate relationships (Q51), activities related to relaxation and enjoyment (Q52), breathlessness during physical exertion (Q53), ability to sit unsupported (Q54), ability to stand unsupported (Q55), ability to get up from the floor from a lying position (Q56), ability to push open a heavy door (Q57), and ability to move from sitting at the side of the bed to lying down on the back (Q58)
Independence in performing ADLs (16)	Eating and drinking (Q59), washing the upper body and head (Q60), washing the lower body (Q61), dressing the upper body (Q62), dressing the lower body (Q63), grooming (Q64), using an indwelling catheter (bladder management) (Q65), need for assistance with bowel management (Q66), using the toilet (Q67), ability to perform activities without assistance or devices, each with 5 items (Q68), transfers from the bed to the wheelchair (Q69), and moving around moderate distances (Q70)
Occupation (4)	Job before spinal cord injury (Q71), vocational rehabilitation service (Q72), disability benefits (Q74), paid employment (Q76)
Environmental factors (14)	Accessibility of public places (Q87); accessibility to homes of friends and relatives (Q88); unfavorable climatic conditions (Q89); negative societal attitudes (Q90); negative attitudes of family and relatives (Q91), negative attitudes of friends (Q92); negative attitudes of neighbors, acquaintances, and work colleagues (Q93); lack of or inadequate adapted assistive technology for short-distance mobility (Q94); lack of or inadequate adapted means of transportation for long distances (Q95); lack of or insufficient nursing and support services (Q96); lack of or insufficient medication and medical aids and supplies (Q97); problematic financial situation (Q98); lack of or insufficient communication devices (Q99); and lack of or insufficient state services (Q100)
Personal factors (10)	Getting what one wants (Q108), confidence in efficient problem-solving (Q109), maintaining contacts (Q110), maintaining good health (Q111), resilience gained through spinal cord injury (Q112), concern about the future (Q113), achieving desires (Q114), decision-making on important matters (Q115), feeling of belonging (Q116), and experience of negative events within the year (Q117)
Quality of life and health status (6)	Overall quality of life (Q118), subjective health satisfaction (Q119), satisfaction with performing ADLs (Q120), satisfaction with oneself (Q121), satisfaction with personal relationships (Q122), and satisfaction with current living conditions (Q123)

ADLs, activities of daily living; SCI, spinal cord injury.

Given the large number of variables available for analysis, we employed the random forest (RF) methodology instead of traditional logistic regression. As a non-parametric approach, RF provides greater flexibility when statistical assumptions are not met and can inherently account for nonlinear relationships and interactions among variables without requiring explicit model specification [20,21]. Therefore, RF was considered appropriate for analyzing high-dimensional data in this study. Specifically, we used the RF algorithm and constructed multiple decision trees to enhance predictive accuracy and variable importance assessment [22].

The RF technique was implemented using the randomForest package in R [23], and the detailed analysis methodology involved was as follows. The total dataset was split into training and validation sets at a 7:3 ratio. The RF model was generated using the training set (hyperparameters: 500 trees, 10 variables per split). The optimal mtry 10 was selected to minimize the out-of-bag classification error rate using the tuneRF() function in the randomForest package. Then, an importance analysis was performed to identify the top 20 variables contributing to the prediction of paid employment status. We used the impurity-based variable importance measure, also referred to as the mean decrease in impurity. In classification tasks, impurity is typically measured using the Gini index [22,24].

RESULTS

Performance of the prediction model

The performance of the RF prediction model developed in this study was evaluated using metrics including accuracy, sensitivity, specificity, precision, and F1 scores for training and validation sets. The validation set provided good classification performance without overfitting (accuracy, 78.26%; sensitivity, 45.10%; specificity, 93.64%; precision, 76.67%; F1-score, 56.79%).

Classification prediction results

Of the 892 survey respondents, 538 were included in the analysis after excluding 354 who did not report their employment status. As Korea has one of the fastest aging populations and highest employment rate for older adults (age ≥65 years; 34.9%) [24], among the Organisation for Economic Co-operation and Development member countries, older respondents were intentionally included in the analysis. The demographic characteristics of both the overall and analyzed samples are presented in Table 2, showing no significant differences across all items. The analyzed sample had higher proportions of males (75.1%), unmarried individuals (53.9%), and individuals living with family (90.1%) than their respective counterparts. Additionally, 84.8% were receiving daily living assistance, 56.1% were educated till high school level or below, 57.2% had paraplegia, 58.7% had complete injuries, and 93.1% had injuries due to traumatic causes. The mean age was 46.31 years, and the average duration of SCI was 15.25 years.

Table 2.

Demographic characteristics

Characteristic	Collected sample (n=892)	Analyzed sample (n=538)
Sex
Male	677 (75.9)	404 (75.1)
Female	214 (24.0)	134 (24.9)
No response	1 (0.1)	0 (0.0)
Marital status
Married	426 (47.8)	248 (46.1)
Unmarried	460 (51.6)	290 (53.9)
No response	6 (0.7)	0 (0.0)
Household type
Living alone	108 (12.1)	48 (8.9)
Living with the family	770 (86.3)	485 (90.1)
Institutionalized	9 (1.0)	5 (0.9)
No response	5 (0.6)	0 (0.0)
Daily living assistance
Yes	756 (84.8)	456 (84.8)
No	133 (14.9)	82 (15.2)
No response	3 (0.3)	0 (0.0)
Education level
College or higher	381 (42.7)	236 (43.9)
High school or lower	507 (56.8)	302 (56.1)
No response	4 (0.4)	0 (0.0)
Degree of paralysis
Tetra	353 (39.6)	230 (42.8)
Para	529 (59.3)	308 (57.2)
No response	10 (1.1)	0 (0.0)
Completeness
Complete	509 (57.1)	316 (58.7)
Incomplete	377 (42.3)	222 (41.3)
No response	6 (0.7)	0 (0.0)
Etiology of injury
Traumatic	829 (92.9)	501 (93.1)
Non-traumatic	62 (7.0)	37 (6.9)
No response	1 (0.1)	0 (0.0)
Age	47.91±11.76	46.31±11.86
Net monthly income	3.08±2.30	3.21±2.37
Duration of SCI (yr)	15.55±10.09	15.25±9.29

Values are presented as number (%) or mean±standard deviation.

SCI, spinal cord injury.

Importance analysis of predictors of paid employment

An importance analysis was conducted to identify the most influential variables among those included in the RF prediction model. Table 3 presents the top 20 variables contributing to predicting paid employment status. The highest importance was assigned to “net monthly household income” (Q9), followed by “age” (Q2), both of which fall in the demographics domain. The third highest importance was assigned to “resilience gained through spinal cord injury” (Q112), a personal factor indicating self-perception of strength gained from SCI. Other factors with high importance were “achieving desires” (Q114) and “decision-making on important matters” (Q115).

Table 3.

Top 20 predictors of employment status

Rank	Domain	Predictive variable	Importance
1	Demographics	Net monthly household income (Q9)	8.2580
2	Demographics	Age (Q2)	7.1972
3	Personal factors	Resilience gained through spinal cord injury (Q112)	5.6106
4	Quality of life and health status	Satisfaction with oneself (Q121)	5.5720
5	Occupation	Vocational rehabilitation service availability (Q72)	5.0950
6	Personal factors	Achieving desires (Q114)	4.7338
7	Quality of life and health status	Satisfaction with performing ADLs (Q120)	4.5615
8	Independence in performing ADLs	Using the toilet (Q67)	4.1100
9	Occupation	Disability benefits (Q74)	4.0579
10	Environmental factors	Lack of or inadequate adapted assistive technology for short-distance mobility (Q94)	3.8518
11	Activity and participation	Activities related to relaxation and enjoyment (Q52)	3.7711
12	Demographics	Need for assistance in ADLs (Q6)	3.5069
13	Quality of life and health status	Satisfaction with interpersonal relationships (Q122)	3.4307
14	Activity and participation	Ability to push open a heavy door (Q57)	3.0875
15	Secondary health issues	Sleep problems (Q24)	2.9715
16	Independence in performing ADLs	Dressing and undressing the lower body (Q63)	2.9510
17	Personal factors	Decision-making on important matters (Q115)	2.9431
18	Quality of life and health status	Quality of life (Q118)	2.9124
19	Demographics	Education level (Q7)	2.8171
20	Demographics	Marital status (Q4)	2.7536

ADLs, activities of daily living.

Four “quality of life and general health” domain variables ranked among the top 20 predictors, with “satisfaction with oneself” (Q121) being the highest (fourth overall), followed by “satisfaction with the ability to perform daily living activities,” “satisfaction with interpersonal relationships,” and “quality of life.” Receiving “vocational rehabilitation services” and “disability benefits” were also significant predictors of paid employment. Independence in performing ADLs, such as using the toilet and dressing/undressing the lower body, was also crucial in predicting paid employment status. The only environmental domain variable that ranked among the top 20 predictors was “lack of or inadequate adapted assistive technology for short-distance mobility.”

Additionally, activities for relaxation and enjoyment as well as the ability to independently open heavy doors were significant predictors of paid employment status within the “activity and participation” domain. Demographic characteristics, such as the availability of assistance for daily living, also significantly influenced paid employment status. Education level and marital status were further identified as important predictors. Among the various secondary health issues considered, that of sleep problems was the most impactful.

Partial dependence plot analysis results

Following the importance analysis, which identified the top 20 predictors, a partial dependence plot (PDP) analysis was performed to visualize relationships between individual variables and the likelihood of paid employment. Accordingly (Figs. 1, 2), the likelihood of paid employment dramatically increased for households with a monthly income of 0.96–1.9 million won (approximately USD 713–1,412) and above, with net monthly household income being assigned the highest importance. Notably, the average monthly income of households with disabled members in Korea in 2017 was 2.42 million won (USD 1,799) [25]. Age was identified as the next most significant variable, and the likelihood of paid employment significantly decreased from 40s and markedly decreased from 50s.

Fig. 1.

Partial dependence plots (PDPs, rankings 1–10). The PDP shows the relationships between individual variables and the likelihood of paid employment rankings, ranging from 1 to 10. ADLs, activities of daily living.

Fig. 2.

Partial dependence plots (PDPs, rankings 11–20). The PDP shows the relationships between individual variables and the likelihood of paid employment rankings, ranging from 11 to 20. ADLs, activities of daily living.

Furthermore, individuals with stronger self-perceived resilience, goal orientation, and decision-making confidence—particularly those scoring ≥3 on a 5-point Likert scale (1=not at all, 5=fully)—were more likely to be employed. Individuals who rated their self-satisfaction, with performing ADLs, and with interpersonal relationships as “high” had a markedly higher employment rate than those who rated them as “neutral” or “low.” Interestingly, these variables were stronger predictors than quality of life. Moreover, although quality of life generally correlated with employment, the highest likelihood of paid employment was observed among those rating their quality of life as “high,” followed by those rating it as “very high.”

Vocational rehabilitation services positively impacted employment opportunities, while the availability of disability benefits was inversely correlated with employment rates. Independence in ADLs, particularly the ability to use the toilet without assistance or assistive devices, was associated with higher employment rates, with the chance rapidly increasing when no assistance was needed, even if assistive devices such as handrails and specially designed toilets were used for people with disabilities. A similar pattern was observed for independently dressing and undressing the lower body.

Among environmental factors, lack of or inadequate adapted assistive technology for short-distance mobility, such as stair lifts and walking aids, negatively impacted employment prospects. Those reporting related inconveniences in daily living had lower employment rates. Enhanced ease in engaging in activities related to relaxation and enjoyment, along with greater ability to open heavy doors, was associated with higher employment rates. Among demographic factors, lower employment rates were associated with receiving assistance in ADLs and being married. Individuals with a college degree or higher education level had increased chances of securing paid employment. Among secondary health issues, individuals with severe sleep problems (scores of ≤3) faced significantly lower employment chances.

DISCUSSION

This cross-sectional study investigated predictors of paid employment among community-dwelling individuals with SCI in South Korea and identified significant factors associated with employment status. A total of 107 items were extracted from the InSCI Survey. Few studies have examined the complex interplay between employment and multiple factors among individuals with SCI in South Korea. To our knowledge, this study is among the first to investigate a wide range of variables potentially associated with paid employment in this population. The analysis results corroborated the complex interplay between employment and multiple factors including personal characteristics, independence in performing ADLs, activity and participation, quality of life, secondary health issues, and environmental factors. Significant predictors of paid employment among individuals with SCI included socioeconomic factors (higher net monthly household income, higher educational level, younger age, and being unmarried), personal factors (higher self-efficacy), independence and participation-related factors (greater independence, no need for assistance in ADLs, and fewer participation difficulties), health-related factors (absence of sleep problems), and contextual factors (sufficient vocational rehabilitation services and absence of disability benefits), as well as improved quality of life and satisfaction.

Average monthly household income, ranking first among the top 20 predictors in this study, was a strong predictor in a recent South African study [18]. However, the relationship between income and employment should be interpreted with caution due to the possibility of reverse causality. In particular, while employment may lead to higher income, higher income may also facilitate access to resources that support employment. Household income sharply declines, along with employment rates, following SCI [26], presumably due to increased medical costs, as well as additional expenditures on assistive devices and home modifications, while opportunities for employment decrease due to the injury [27]. At the same time, higher household income may be associated with better access to rehabilitation services, assistive technologies, and environmental adaptations, which are known to influence employment outcomes after SCI (prevalence and predictors of return to work following a SCI using a work disability prevention approach, Employment After SCI). Consistent with this, a longitudinal study found that individuals with lower household income experienced greater life difficulties, including health problems and environmental barriers, compared with those with higher income [28].

Age is associated with employment status [29-31]; in our study, age was a strong predictor of paid employment, second only to monthly household income, particularly among those in their 20s to early 40s. However, given that employment rates for individuals without disabilities in South Korea remain high in individuals in their 50s (75.3%) and those aged 60−64 years (60.6%), similar to rates in those aged 20s−40s [32], the decline in employment prospects for individuals with SCI after the early 40s warrants further consideration. This decline may be related to the cumulative effects of functional limitations and secondary health conditions over time, potential mismatches between job demands and physical capacity, and challenges in job retention and opportunities for retraining or career transition [10,33].

Higher self-efficacy, reflecting the anticipation and belief in one’s ability to perform appropriate actions in any situation, and higher perceived control, the belief in one’s ability to make decisions independently, were found to increase the likelihood of paid employment. This aligns with Umucu et al.’s [34] findings, which examined the relationship between self-efficacy and employment. Interestingly, a systematic review [35] exploring the relationship between psychological factors and employment reported a weak impact of self-efficacy on employment. These findings underscore the potential benefits of targeted psychological support to enhance self-efficacy among individuals with SCI.

Quality of life and satisfaction, confirmed as significant predictors of employment in previous studies [36,37], also emerged as key factors in this research. The InSCI Survey, conducted across 22 countries, consistently associated higher quality of life with higher employment rates, irrespective of the country’s gross domestic product [38].

Furthermore, the availability of vocational rehabilitation services was identified as one of the most important predictors of paid employment prospects. These findings are consistent with the findings of an Australian study that recognized vocational rehabilitation services as a significant factor positively influencing employment intentions following SCI [16]. However, a Korean survey evaluating vocational rehabilitation services for individuals with SCI found that the satisfaction score for met needs was only 3.7 out of 10 [39]. This finding highlights the necessity for further research to thoroughly analyze unmet needs in vocational rehabilitation services from multiple perspectives, providing foundational data for developing various expansion strategies for efficient service delivery and utilization.

Receiving disability pension or financial benefits due to disability was found to reduce the likelihood of paid employment, supporting previous findings [40,41]. In South Korea, various policy measures support income security for people with disabilities depending on the degree of disability and income level, including disability pensions, industrial accident disability benefits, and disability allowances. To prevent these supports from reducing employment prospects for individuals with SCI, it is essential to evaluate work capability during disability assessment and ensure it is effectively linked with vocational rehabilitation services, aimed at facilitating employment [42].

As in recent studies [16,18], the need for assistance in ADLs, both inside and outside the home, was identified as a significant predictor for paid employment in this study. The finding that the greater the independence in performing ADLs, such as using the toilet and dressing, the less the assistance required from caregivers [43], suggests that lower degrees of independence necessitate more intensive assistance. This finding is supported by the high positive correlation between functional independence—which combines mobility and self-care abilities—and employment [44], clearly indicating that higher independence in performing ADLs is a good predictor of paid employment.

As an environmental factor, lack of or inadequate adapted assistive technology for short-distance mobility, such as stair lifts and walking aids (including wheelchairs), have been reported to lower the likelihood of paid employment. This supports previous findings [45] that identified manual mobility and independent living devices such as wheelchairs as the most crucial tools for employment. Mobility aids are the most frequently used tools by individuals with SCI and are essential for enhancing independence at work. However, the cost of assistive devices necessary for work is 3.5 times higher than those not essential for employment [45], and the average cost of assistive devices for individuals with SCI in South Korea is 3.54 million won [39], which surpasses the average household income for people with disabilities. Under these circumstances, it is imperative to establish policy measures such as securing sufficient public funding for assistive devices and building an integrated public service delivery system for these devices [46].

Surprisingly, sleep problems have emerged as the most critical secondary health issue-related variable predicting paid employment for individuals with SCI. Despite existing research indicating that individuals with SCI commonly experience sleep disturbances [47], the relationship between sleep and employment remains underexplored. Interestingly, a recent study in Australia [16] observed that people with lower sleep quality were more likely to be employed, although causality could not be confirmed. Another study [48] showed that poor sleep quality was strongly associated with unemployment among individuals with SCI. This discrepancy underscores the need for further research into how sleep problems affect the employment status of individuals with SCI.

While numerous studies [10,49] have established strong evidence that higher pre-SCI education is associated with higher post-SCI employment rates, a Korean study [17] found no significant association between higher education and employment. However, our study demonstrated a strong correlation between education and paid employment among individuals with SCI. Regarding marital status, our results contradict those of other studies [16,18,50], highlighting the need for further research. Nonetheless, marital status and other demographic characteristics were found to be strong predictors of paid employment in this study, with five demographic variables ranked in the top 20. This finding aligns with previous studies [29,30,51] that observed strong correlations between sociodemographic characteristics and employment.

Limitations

Some limitations of this study should be noted. First, the cross-sectional design precluded the establishment of causality between variables. Second, while we used a nationwide sample, relying on databases from two institutions rather than using random sampling could have been a source of selection bias, which may have influenced the results. Finally, unlike traditional statistical methods, the RF technique used in this study does not yield statistical significance. To address these limitations, we used PDPs to visualize and analyze the relationships between variables. However, there are limitations in interpreting the relationships between the identified variables and employment status. The deep learning model is designed for prediction rather than causal inference, and some variables may reflect consequences of employment. Also, the model showed relatively low sensitivity (45%) in identifying employed individuals. This may be partly explained by class imbalance, as machine learning models tend to favor the majority class [52]. The lower sensitivity compared to specificity suggests difficulty in detecting the minority class. Future studies should consider newly proposed RF–based methods designed to address imbalanced classification problems [53].

Conclusion

This study employed the RF method to analyze a complex array of factors that could potentially influence employment among previously and currently community-dwelling individuals with SCI in South Korea. The prediction model developed in this study highlighted monthly income, age, self-efficacy, self-satisfaction, vocational rehabilitation, and toilet use as key predictors of paid employment status in the study cohort, corroborating previous research findings and providing a solid foundation for policy development aimed at enhancing employment opportunities for individuals with SCI. To deepen our understanding, future research should delve into the specific occupational environments conducive to both securing and maintaining employment for individuals with SCI.