Return to work after acute spinal trauma in five hospitals in four low- and middle-income countries: a retrospective review of a prospectively collected multicentre cohort study

¹Department of Health Sciences, Queen´s University, Kingston, Canada, ²Department of Bioengineering and Computer Science, Stanford University, Stanford, United state America, ³Department of Orthopaedics, Spine Unit, Hawassa University, Hawassa, Ethiopia, ⁴Sign Fracture Care International, Richland, United state America

^&Corresponding author
Abdou Salam Gueye, Department of Health Sciences, Queen´s University, Kingston, Canada

Introduction    

Spinal trauma in low- and middle-income countries (LMICs) occurs predominantly in young men of working age, most commonly after road traffic collisions and falls [1-3]. In a systematic review of African studies, 81.1% of patients with traumatic spinal injury were male, and their mean age was 34.5 years [3], a profile also reported from Tanzania [4], whereas in high-income countries a growing proportion of spinal cord injuries follows falls in older adults [5]. Where social protection is limited, the costs of injury and lost income fall directly on patients and their households [6,7]. In a national Tanzanian household survey, households headed by a manual laborer were more likely to incur catastrophic health expenditure for illness of any cause [8]. The economic burden of not returning to work (RTW) after spinal trauma is therefore likely to be greatest among manual workers, a group that has received little study. Despite this burden, outcome reporting after spinal trauma in LMICs has concentrated on neurological recovery, complications, and mortality [3,4,9]. Return to work is reported less often and is usually drawn from high-income cohorts or from cross-national surveys in which LMICs are underrepresented. A survey of employment after spinal cord injury in 22 countries included no participants from low-income countries [10]. Findings from high-income settings, where employment protection, sick pay, and vocational rehabilitation are available, may not be generalizable to LMICs [11].

In high-income settings, RTW after spinal cord injury has been associated with younger age, less severe injury, higher education, and greater functional independence [12-14]. The physical demand of the pre-injury occupation has rarely been examined as an independent predictor [15], and one study excluded occupation type because existing classification schemes were too simplistic or too complex [16]. Where measured, RTW was 25% or less after heavy or strenuous pre-injury work, substantially lower than after light-to-moderate work [13], although this association did not persist after adjustment in a Dutch cohort [17]. Where redeployment to lighter duties or vocational retraining is rarely available, as in many LMICs, physical job demand may be a stronger determinant of RTW than in more flexible labor markets. The timing of RTW is also poorly described, because most studies measure employment once, often years after injury [12,16,18]. Repeated assessment during the first year would show when return occurs and how long follow-up and vocational support should continue. The primary objective of this study was to estimate the cumulative incidence and timing of RTW within 12 months of acute spinal trauma among adults working before injury in four LMICs. The secondary objective was to identify factors associated with RTW and, specifically, whether the physical demand of the pre-injury occupation was associated with RTW independently of admission neurological status.

Methods     

Study design and setting: this was a retrospective cohort study of data collected prospectively in a multicentre spinal trauma registry at five referral hospitals: Muhimbili Orthopaedic Institute (Tanzania), Hawassa University Comprehensive Specialized Hospital and St Paul’s Hospital Millennium Medical College (Ethiopia), Moi Teaching and Referral Hospital (Kenya), and Lilongwe Institute of Orthopaedics and Neurosurgery (Malawi). We scheduled assessments at 2 and 6 weeks and at 3, 6, and 12 months after injury. Return to work was an outcome of the registry protocol, but the present analysis was planned after data collection. It includes patients enrolled from July 2024 to August 2026, with follow-up to the data export on 18^th August 2026. Reporting follows the STROBE guideline [19], and the checklist is provided as a supplementary file.

Data collection and quality control: neurosurgical or orthopaedic trainees screened admissions and referrals daily, and research assistants trained in the protocol obtained consent. All sites entered data into a single REDCap database hosted at the University of California, San Francisco, using common case report forms in English, Swahili, Chichewa, or Amharic, with required fields, branching logic, and date validation. The treating surgeon graded neurological status with the ASIA Impairment Scale at admission, discharge, and follow-up. Follow-up was conducted in clinic or, when the patient could not attend, by telephone. Follow-up consultation and radiograph fees were waived, reminders were sent by telephone and text message, and missed visits were documented on a standard form. The coordinating centre oversaw data quality.

Participants and eligibility criteria: the registry enrolled adults (18 years or older) with acute spinal trauma. It excluded moderate or severe traumatic brain injury, major concomitant injury (long-bone, chest, abdominal, or pelvic), pathological fracture, burns over more than 10% of body surface, death in the emergency department or before arrival, and presentation more than 30 days after injury, so that outcomes could be attributed to isolated, acute spinal trauma. The analysis included patients who were working, in paid or unpaid work, before injury and had at least one post-injury assessment of work status. Work status was first asked at the 6-week visit, so patients with no follow-up or only the 2-week visit were not analysable. No formal sample size calculation was performed, and all eligible registry patients were included.

Outcome: return to work was defined by an affirmative response to the question "Since your last visit, have you been working? This includes paid and unpaid workʺ asked at the 6-week follow-up onwards. Because the question covers the interval since the previous visit, work begun before the 6-week visit was captured at that visit. Time zero was the date of injury. For patients who returned to work, the event date was the earliest recorded estimated date of return on or after injury. For one patient who reported working but had no recorded return date, we used the date of the first visit at which work was reported. Unpaid work was included because most employment in Africa is informal [20]. The question did not distinguish paid from unpaid work or record hours.

Handling of follow-up and censoring: patients who did not return to work were censored at the date of their last visit with a recorded work status. For one patient without a valid such date, the last attended follow-up visit was used. Dates preceding the injury were treated as data-entry errors and disregarded (one return date and five visit dates). Median follow-up was estimated using the reverse Kaplan-Meier method. Six patients in the analysis cohort died during follow-up. In the primary analysis, these patients were censored at their last visit with a recorded work status. In a post hoc sensitivity analysis, death was treated as a competing event and cumulative incidence re-estimated with the Aalen-Johansen estimator.

Variables: baseline characteristics were ascertained at enrolment. The exposures of primary interest were admission neurological status and occupational physical demand. Admission neurological status was classified from the American Spinal Injury Association (ASIA) Impairment Scale as no deficit, incomplete deficit (ASIA B to D), or complete deficit (ASIA A), with ASIA E counted as no deficit. One patient with a recorded deficit of missing grade was retained in descriptive summaries but excluded from regression models. Pre-injury occupation, recorded as free text, was classified by the physical demand of the job using a fixed lookup table (Annex 1) as high (regular heavy lifting or carrying, sustained manual work performed standing or in awkward postures, or operation of two- or three-wheeled vehicles), low (predominantly seated or light standing work, including light-vehicle driving, trading and sales, and clerical or professional roles), or indeterminate (free text insufficient to infer job tasks), retained as its own group and examined in sensitivity analyses. Occupations were classified independently by two investigators blinded to patient outcomes, and disagreements were resolved by consensus. Age, sex, and workplace injury were prespecified model covariates. Age was taken at injury, with two birth-year entries converted accordingly. Education, health insurance, injury region (derived from the marked vertebral levels), and operative treatment were examined in unadjusted analyses only, because the number of events limited model size.

Statistical analysis: continuous variables are reported as medians with interquartile ranges (IQRs) and categorical variables as counts and percentages. Cumulative incidence of return to work was estimated using the Kaplan-Meier method, with Greenwood variances and complementary log-log 95% confidence intervals (CIs), and compared across groups using the log-rank test. Associations with time to return to work were examined using Cox proportional hazards regression with Efron’s method for ties. A hazard ratio below 1 indicates a lower rate of return. A prespecified multivariable model included admission neurological status (ASIA B to D vs no deficit), high versus other physical-demand occupation, age (per 10 years), sex, and workplace injury. Patients with complete injuries (ASIA A) were excluded from the model because none returned to work, which made the hazard ratio inestimable, and this group is instead described using an exact binomial upper bound. The proportional-hazards assumption was assessed by adding an interaction between each exposure and an indicator for the period beyond 120 days. The proportion of patients reporting work at each visit was modelled using generalized estimating equations (GEE) with a logistic link, an exchangeable working correlation structure, and robust standard errors. In an exploratory analysis, return-to-work proportions were compared across occupation classes within each neurological stratum using the Fisher exact test. A multivariable model stratified by site and the competing-risk analysis described above were performed as sensitivity analyses. In a post hoc analysis addressing loss to follow-up, stratum-specific 12-month estimates were standardised to the admission neurological distribution of all pre-injury workers with a recorded grade (n=399), assuming similar return within strata in analysed and non-analysable patients, with a bootstrap 95% CI (2000 samples). Analyses were performed using Python version 3.11 (statsmodels 0.14). Statistical significance was defined as a two-sided p-value less than 05.

Ethical considerations: the study was conducted in accordance with the Declaration of Helsinki. All participants, or a surrogate when the patient could not consent, provided written informed consent at enrolment into the registry, including consent for export of de-identified data for analysis. The registry protocol was approved by the institutional review board of the University of California, San Francisco (#23-39621, 26 April 2024), and the consent documents used at every site were approved under this reference. The research ethics committee at each site provided local ethical oversight.

Results     

Cohort characteristics: between July 2024 and August 2026, 500 patients with spinal trauma were enrolled across five sites (Figure 1). Of these, 421 were working before injury, 70 were not, and 9 had no pre-injury work status recorded. Among the 421 pre-injury workers, 258 (61.3%) had no post-injury work assessment (195 had not yet attended a follow-up visit, and 63 had attended only the 2-week visit, at which work is not ascertained). Of these 258 patients, 238 (92.2%) had passed the 6-week assessment point at data export. This left 163 patients (38.7%), who constitute the analysis cohort. Of these, 35 (21.5%) returned to work during follow-up, and 128 were censored. Median follow-up, estimated by the reverse Kaplan-Meier method, was 172 days. Baseline characteristics are shown in Table 1. Median age was 35 years (IQR 28 to 41.5), 87.7% were male, 67.5% held high physical-demand occupations, and 57.1% were injured at the workplace. At admission, 39.3% had no neurological deficit, 38.0% had an incomplete injury (ASIA B to D), and 22.1% had a complete injury (ASIA A). Most patients (81.6%) were treated operatively. Patients who could not be analysed because of loss to follow-up differed systematically from those analysed (Annex 1). They more often had complete injuries (40.7% vs 22.1%) and less often no deficit (23.3% vs 39.3%; p<.001), more often had cervical injuries (34.9% vs 20.2%, p=.002), were less often treated operatively (39.9% vs 81.6%, p<.001), less often had health insurance (24.4% vs 42.3%, p<.001), and differed by site (p<.001; analysable proportion 26.1% to 70.5%). Age, sex, occupation class, education, and workplace injury were similar. Sixteen pre-injury workers (3.8%) were known to have died, 10 non-analysable and 6 in the analysis cohort. All six had an admission neurological deficit, and none had returned to work before death. Of 399 follow-up work assessments, 155 (38.8%) were by telephone, mostly because of transport or distance (118 of 155, 76.1%).

Cumulative incidence and timing of return to work: the 12-month Kaplan-Meier cumulative incidence of return to work was 46.0% (95% CI 33.9 to 60.1; Figure 2), compared with 11.7% (7.3 to 18.4) at 3 months and 17.9% (12.0 to 26.2) at 6 months. The 12-month estimate exceeds the crude proportion (35 of 163, 21.5%) because most patients had not completed 12 months of follow-up. The Kaplan-Meier method estimates return among those still observed, and only 22 were observed at 12 months. With death as a competing event, the 12-month cumulative incidence was 43.5% (Aalen-Johansen estimator). Standardised to the neurological distribution of all pre-injury workers, it was 33.5% (95% CI 25.3 to 42.3). The cohort median time to return could not be estimated because cumulative incidence stayed below 50%. Among the 35 who returned, the median was 116 days (IQR 66.5 to 265.5, range 25 to 323). Of these returns, 45.7% occurred within 3 months, 17.1% between 3 and 6 months, and 37.1% after 6 months. The modelled proportion reporting work at each visit rose from 4.4% (95% CI 2.0 to 9.1) at 6 weeks to 16.3% (10.0 to 25.4) at 6 months and 37.7% (25.2 to 52.1) at 1 year (Figure 3).

Factors associated with return to work: return to work was most strongly associated with admission neurological status (log-rank p<.001; Figure 4). The 12-month cumulative incidence was 69.0% (95% CI 51.1 to 85.3) with no deficit, 36.0% (20.1 to 58.7) with incomplete injury (ASIA B to D), and 0% (0 to 9.7) with complete injury (ASIA A). Within the incomplete group, 9.5% of ASIA B (2 of 21), 4.5% of ASIA C (1 of 22), and 36.8% of ASIA D patients (7 of 19) returned (Annex 1). Of returners with incomplete injury, 90.0% (9 of 10) had no deficit at their last follow-up neurological assessment, compared with 29.4% of non-returners with a follow-up neurological assessment (15 of 51). Among patients with complete injury, 16.7% (6 of 36) improved and none returned. In the multivariable model (n=126, 35 events), which excluded ASIA A injuries, two factors were independently associated with a lower rate of return (Table 2): incomplete neurological injury (HR 0.30, 95% CI 0.14 to 0.66) and injury at the workplace (HR 0.38, 95% CI: 0.17 to 0.85). Unadjusted 12-month cumulative incidence was 35.9% after workplace injury and 63.3% otherwise (Annex 1). Workplace injuries accounted for similar proportions of each neurological stratum (59.4% with no deficit, 59.7% incomplete, 50.0% complete), and within strata fewer patients injured at work returned (no deficit, 34.2% vs 46.2%; incomplete, 8.1% vs 28.0%). In the site-stratified model, both associations were in the same direction, but the confidence interval for workplace injury included 1 (HR 0.44, 95% CI: 0.18 to 1.04). High physical-demand occupation was not independently associated with return to work (HR 1.07, 95% CI 0.47 to 2.45), including when indeterminate occupations were reassigned to the high-demand group (HR 0.97, 95% CI: 0.42 to 2.24) or excluded (n=121, HR 1.00, 95% CI: 0.43 to 2.32). Among incomplete injuries, 24.0% of patients in low-demand occupations (6 of 25) and 11.8% in high-demand occupations (4 of 34) returned (Fisher exact p=.297; Annex 1). Age and sex were not independently associated with return. There was no evidence that the proportional-hazards assumption was violated, except for age (interaction p=.019; HR per 10 years 1.23, 95% CI 0.88 to 1.71, in the first 120 days and 0.65, 0.40 to 1.06, thereafter), which had a null association overall.

Education, sex, and other patient factors (unadjusted): return to work in unadjusted analysis increased with education level, from 34.9% (95% CI: 22.0 to 52.3) with primary education or less to 54.1% (29.5 to 82.3) with secondary and 85.0% (53.0 to 99.2) with college or university education (log-rank p=.005). It was also higher among women than men (69.6% vs 43.7%, log-rank p=.035), but this was based on 20 women and was not significant after adjustment (HR 1.42, 95% CI: 0.47 to 4.22, Table 2). Operative treatment was more frequent with more severe injury and was not associated with return after adjustment for neurological status (HR 0.66, 95% CI: 0.31 to 1.42). Health insurance, injury region, and age group were not significantly associated with return to work.

Discussion     

Fewer than half of previously working patients in this cohort returned to work within a year. The estimate of 46.0% lies within the 21% to 67% range reported among previously working adults in a systematic review of predominantly high-income studies [12] and near the pooled 45.8% of a meta-analysis of North American and European cohorts [15]. The true rate in this population is probably lower. The outcome included unpaid work, a lower threshold than the paid-employment endpoints of most previous studies [14], and patients lost to follow-up were more severely injured. Standardising for this difference reduced the estimate to 33.5%. This is consistent with evidence that people with spinal cord injury in lower-income groups face greater barriers to employment [11]. Return also occurred late. More than a third of returns occurred after 6 months, and cumulative incidence rose from 17.9% at 6 months to 46.0% at 12 months. Previous studies report similarly long intervals, with a median of 13 months in a Dutch cohort [17] and a mean of almost 5 years to the first post-injury job in cross-sectional surveys [16,18]. Follow-up and vocational support that end at 6 months would miss many returns. Neurological status was the strongest correlate of return. No patient with a complete injury returned, and nine of ten returners with incomplete injury had recovered to no deficit. Injury severity is a consistent predictor of RTW [12-14], and where rehabilitation, accessible transport, and workplace adaptation are limited, patients with a persisting deficit may have few routes back to work [12,21]. Contrary to our hypothesis, occupational physical demand was not independently associated with return, although it has been in some high-income cohorts [13] but not others [17]. Where workplace accommodation and accessible transport are limited, even low-demand work such as trading or driving may require walking, standing, and travel that are difficult with a persisting deficit.

Return may then depend on near-complete neurological recovery regardless of job category, as observed among returners here. In high-income settings, people often return to less physically intense work [17], an option that depends on redeployment and retraining services rarely available in our setting. Unlike neurological status, this mismatch can in principle be addressed through job modification or supported employment [22]. The estimate was also imprecise. Among incomplete injuries, fewer patients in high-demand occupations returned (11.8% vs 24.0%), but the group was too small to show whether this difference was real, so an effect of physical job demand cannot be ruled out. Injury at the workplace was associated with lower return after adjustment for neurological status. Because workplace and non-workplace injuries had similar neurological distributions and fewer patients injured at work returned within each stratum, the association does not appear to reflect injury severity, although it was less precise in the site-stratified model and needs confirmation. A workplace injury may end the employment relationship through job loss, employer reluctance to re-engage an injured worker, or closure of the worksite. Return to the pre-injury employer is the fastest route back to work after spinal cord injury [16,18], so losing that employer may remove it. These mechanisms could not be distinguished in our data but, unlike neurological injury, are open to employer and policy intervention. Return to work rose steeply with education in unadjusted analysis, as in previous studies [13,16], but 35 events were insufficient to examine this in the adjusted model without overfitting. The higher unadjusted return among women was based on only 20 women and did not persist after adjustment. Loss to follow-up was concentrated among patients with complete or cervical injuries, without surgery or health insurance, and at particular sites. Transport, financial, and family constraints were the reasons most often recorded for missed visits, and three-quarters of telephone assessments were needed because of distance or transport, even though follow-up fees were waived. A Tanzanian cohort showed a similar pattern, with insurance, lumbar injury, and neurological improvement predicting clinic follow-up and only 13.3% of patients followed at 12 months [23]. Unrecorded deaths may also contribute, and some patients who resumed work may have stopped attending. On balance, loss to follow-up probably inflated the estimate, because non-analysable patients were almost twice as likely to have a complete injury, and no patient with a complete injury returned.

Strengths: this study offers multicountry return-to-work data from LMIC settings, a population underrepresented in existing literature, using prospectively collected data with common procedures across sites. Repeated assessment of work status allowed time-to-event rather than single-timepoint estimation. Occupation was classified with a fixed lookup table by investigators blinded to outcomes, and death and loss to follow-up were addressed in sensitivity analyses.

Limitations: the main limitation is loss to follow-up. Only 163 of 421 pre-injury workers could be analysed, excluded patients were more severely injured, and the standardised estimate assumes similar return within neurological strata in analysed and non-analysable patients. Mortality was incompletely ascertained among patients lost to follow-up, so the competing-risk estimate likely understates the mortality impact. The outcome was a single self-reported question that included unpaid work and did not capture hours, type of work, or sustained return, which limits comparability with the paid-employment literature. With only 35 events, the model could include five variables, subgroup analyses were underpowered, and residual confounding by education or other factors is possible. Physical demand was assigned from job titles rather than measured, which could bias the occupation effect toward the null, and ASIA B to D spans a wide functional range. Only 22 patients remained at risk at 12 months, and enrolment at referral centres limits generalisability.

Future directions: future studies should reduce and document loss to follow-up through structured telephone follow-up, community tracing, and support for transport costs, and should record vital status for all patients. They should use standardised employment measures that distinguish paid from unpaid work and record hours, job change, and sustained return [14]. They should also extend follow-up beyond 12 months and measure job demands directly. Larger cohorts are needed to examine occupation within neurological strata and workplace injury across neurological grades, with qualitative and policy research to explain lower return after workplace injury. Interventions such as supported employment [22] have not been evaluated in these settings.

Conclusion     

Fewer than half of previously working adults returned to paid or unpaid work within a year of spinal trauma in this multicentre cohort, and no patient with a complete injury returned. Neurological status and workplace injury were associated with return, whereas occupational physical demand was not. Because more than a third of returns occurred after 6 months, these findings argue for follow-up and vocational support through at least the first year, early counselling of patients and their families, and social protection mechanisms that reach households during the long interval before return, where return occurs at all.

Competing interests     

The authors declare no competing interests.

Authors' contributions     

Abdou Salam Gueye: conceived the analysis, performed the statistical analyses, and drafted the manuscript. Hizkyas Kassaye Fisseh: critically revised the manuscript and assisted with interpretation of the data. Taryn Han Boonpongmanee: contributed to interpretation of the data and critically revised the manuscript. Sariah Khormaee: supervised the study, contributed to the study design and interpretation of the data, and critically revised the manuscript. All authors have read and agreed to the final manuscript.

Acknowledgments     

We thank the clinical and research staff at Muhimbili Orthopaedic Institute (Tanzania), Hawassa University Comprehensive Specialized Hospital and St Paul’s Hospital Millennium Medical College (Ethiopia), Moi Teaching and Referral Hospital (Kenya), and Lilongwe Institute of Orthopaedics and Neurosurgery (Malawi) for their contributions to patient enrolment and data collection. Generative AI was used to review the manuscript for compliance with the Pan African Medical Journal's author guidelines, including structure, word count, and reference formatting. All scientific content, analysis, and conclusions are the authors’ own. The authors reviewed and take full responsibility for the final manuscript.

Tables and figures     

Table 1: baseline demographic, clinical, and injury characteristics of adults working before acute spinal trauma with at least one post-injury work assessment, five referral hospitals in Tanzania, Ethiopia, Kenya, and Malawi, July 2024 to August 2026 (N = 163)

Table 2: factors associated with return to work within 12 months among adults working before acute spinal trauma without complete (ASIA A) injury, five referral hospitals in Tanzania, Ethiopia, Kenya, and Malawi, July 2024 to August 2026 (n = 126; 35 returns to work)

Figure 1: selection of the analysis cohort of adults working before acute spinal trauma, five referral hospitals in Tanzania, Ethiopia, Kenya, and Malawi, July 2024 to August 2026 (N = 500 enrolled; n = 163 analysed)

Figure 2: kaplan-Meier cumulative incidence of return to work within 12 months among adults working before acute spinal trauma, five referral hospitals in Tanzania, Ethiopia, Kenya, and Malawi, July 2024 to August 2026 (n = 163), with 95% confidence interval (shaded) and numbers at risk

Figure 3: proportion of adults working before acute spinal trauma who reported work at each follow-up visit, five referral hospitals in Tanzania, Ethiopia, Kenya, and Malawi, July 2024 to August 2026 (n = 163), estimated by generalized estimating equations with 95% confidence intervals (vertical bars). Numbers below the axis are patients reporting work over patients assessed

Figure 4: kaplan-Meier cumulative incidence of return to work within 12 months by admission neurological status among adults working before acute spinal trauma, five referral hospitals in Tanzania, Ethiopia, Kenya, and Malawi, July 2024 to August 2026 (n = 162), with numbers at risk. No patient with a complete injury (ASIA A) returned to work; one patient with an unrecorded grade was excluded. American spinal injury association Impairment scale (ASIA)

Annex     

Annex 1: supplementary materials (PDF 403KB)

References