This was a retrospective, observational, two-center cohort study conducted at Bouaké University Teaching Hospital and Treichville General Hospital (Abidjan), in Côte d’Ivoire, from 1 January 2019 to 31 December 2024, in accordance with the STROBE recommendations (Figure 1). Cases were ascertained by consecutive screening of the neurosurgery admission and hospitalization records of the two centers over the study period; a traumatic spinal injury was defined as a traumatically induced injury of the spinal column—vertebral fracture, dislocation, or ligamentous injury—with or without neurological involvement, confirmed by clinical examination and spinal imaging (radiography and/or computed tomography, with magnetic resonance imaging when available). All TSI with a complete record were included, with no age limit (n = 120). Of 250 records identified, 130 were not included: incomplete record (82), loss to follow-up before one year (21), non-traumatic origin (12), and missing analysis data (15) (Figure 1). The mechanism, recruiting center, and admission ASIA grade of the non-included records—in particular the 21 patients lost to follow-up before one year—could not be systematically retrieved and are therefore not reported; this restricts the assessment of selection related to the outcome and is addressed among the study limitations. Survivors had to have a minimum follow-up of one year, or follow-up until death for patients who died before that time point, so as not to underestimate mortality. The database was frozen on 31 January 2026.

Figure 1. Study flow diagram according to the STROBE recommendations: 250 records identified, 130 not included (incomplete record 82, loss to follow-up before one year 21, non-traumatic origin 12, missing data 15), and 120 patients analyzed.

The main exposure was the mechanism: artisanal mine collapse (n = 26) versus another mechanism (n = 94), the detail of which appears in Table 1. Occupation (gold miner or not) was recorded separately. Injury location was documented by main spinal segment (cervical, thoracic, lumbar), with the thoracic sub-segment specified (T1 - T4, T5 - T8, T9 - T12) for thoracic injuries in the collapse group only. Other variables included the type of vertebral injury, associated injuries, ASIA grade, mode of evacuation, times (in days) to hospitalization and to surgery, the surgical procedure, and overall one-year outcome. For the binary analysis, evacuation without medical support grouped together transport by third parties and public rescue services lacking an onboard medical team; a single collapse-group patient, transferred secondarily by hospital ambulance after extraction from the site by third parties, was classified as “medically staffed transport”, giving a final count of 25/26 evacuations without medical support.

Table 1. Detailed traumatic mechanisms (n = 120).

The patients who fell into an artisanal shaft did so outside a collapse episode and are classified in the comparison group.

The outcome was considered favorable when functional improvement was documented in the record during follow-up—partial neurological recovery, standing or assisted walking, or partial autonomy in activities of daily living; it was classified as “stationary” in the absence of change, or as “death”. In the absence of standardized ASIA grading at follow-up, this assessment relied on the clinical data recorded by the referring neurosurgeon. The admission neurological examination was performed by the neurosurgery team on arrival, before or immediately after initial resuscitation, and the ASIA grade was assigned according to the International Standards for the Neurological Classification of Spinal Cord Injury (ISNCSCI). The ASIA grade was available only at admission. A severe neurological injury was defined as an ASIA grade A or B at admission.

Qualitative variables are expressed as counts (percentages), quantitative variables as median (IQR). Comparisons of proportions used the Fisher exact test, and the Fisher-Freeman-Halton exact test for contingency tables larger than 2 × 2 with low expected counts; continuous variables were compared with the Mann-Whitney test and the ordinal distribution of ASIA grades with a Cochran-Armitage trend test. Effect size was reported systematically (Cramér’s V for contingency tables, Cohen’s h for comparisons of proportions). Crude odds ratios (OR) and their 95% confidence intervals were estimated by univariate logistic regression, without continuity correction.

Four multivariable Firth penalized logistic regression models—suited to small samples and to separation, which it handles by providing finite estimates—tested the factors associated with: (i) a severe injury (ASIA A or B), adjusted for age, sex, and injury location; (ii) a thoracic location, adjusted for age and sex; (iii) evacuation without medical support, adjusted for age and sex; (iv) one-year death, adjusted for ASIA grade, location, and mode of evacuation. Models (i) and (iv) adjust for variables lying on the presumed causal pathway: they therefore estimate a direct effect and not the total effect of collapse, the latter being reported as the unadjusted OR. No formal mediation analysis was conducted, the sample sizes not lending themselves to it. The number of covariates was limited according to the events-per-variable principle. Sex was not retained in the mortality model, its estimate being unstable.

Two sensitivity analyses were conducted. The first introduced the recruiting center into four models: severity, on mechanism and center only, location being deliberately omitted in order to estimate a total effect adjusted for center, to be compared with the unadjusted OR and not with the direct effect; thoracic location and evacuation without medical support, on mechanism, age, sex, and center; one-year death, on mechanism, ASIA grade, location, mode of evacuation, and center. The second restricted the analysis to gold miners, comparing collapse (n = 22) and road traffic accident (n = 22), in order to distinguish the effect of the mechanism from that of the occupation. Given the exploratory nature and the multiplicity of comparisons, effect sizes were favored; the threshold was set at p < 0.05, without adjustment for multiple comparisons. Analyses: Epi Info™ 7.2 (descriptive) and R 4.4.0, logistf package (Firth models).

Gold miners were the most represented socio-occupational category (45/120, 37.5%), ahead of farmers (25.8%). Mine collapse was the second traumatic circumstance (26/120, 21.7%), behind road traffic accidents (65%), and occurred predominantly in gold miners (22/26) (Table 1). Among the 45 gold miners, injuries resulted from a collapse in 22 cases and from a road traffic accident in 22 cases, the latter frequently occurring during trips to remote sites; the 45th patient had fallen into an artisanal shaft outside any collapse episode and was therefore classified in the comparison group.

All collapse victims were men (100%), with a median age of 31 years (IQR: 26 - 34) versus 36 years (IQR: 29 - 45) in the comparison group (p = 0.058). Injuries more frequently involved the thoracic spine (50% vs 26%; OR 2.92; 95% CI: 1.19 - 7.16; p = 0.029; Cohen’s h = 0.51), whereas lumbar locations were less frequent (12% vs 32%; p = 0.048), with more frequent compression and burst-type fractures (69% vs 39%; p = 0.008) (Figure 2, Figure 3, Table 2). Among the 13 thoracic injuries in the collapse group, the lower T9 - T12 segment was the most frequent location (6 cases, 46.2%; T9 n = 2, T10 n = 1, T11 n = 1, T12 n = 2), ahead of the upper T1 - T4 segment (4 cases, 30.8%) and the middle T5 - T8 segment (3 cases, 23.1%); the sub-segmental distribution was not available for the comparison group.

Figure 2. Distribution of injury location (cervical, thoracic, lumbar) by traumatic mechanism, artisanal mine collapse versus other mechanisms.

Figure 3. Forest plot of the associations of artisanal mine collapse in univariate analysis (reference: other mechanisms). Crude odds ratios and 95% confidence intervals.

Severity was markedly greater: 85% of injuries were ASIA A or B versus 45% (p < 0.001; Cohen’s h = 0.87), with no intermediate grade (ASIA C or D) in the collapse group (Table 3; Fisher-Freeman-Halton exact test p < 0.001; Cramér’s V = 0.39; Cochran-Armitage trend test p = 0.012). Conversely, four collapse victims (15%) were free of any neurological deficit (ASIA E) but had radiological vertebral instability: three burst-type fractures with retropulsion of the posterior wall and substantial loss of vertebral height, and one compression fracture associated with injury of the posterior ligamentous complex and progressive kyphosis. All patients in the collapse group underwent surgical treatment (100% vs 67%; p < 0.001; h = 1.22).

The unadjusted total effect of collapse on neurological severity corresponded to an OR of 6.81 (95% CI: 2.18 - 21.3; p < 0.001). After adjustment for age, sex, and injury location—the latter itself a consequence of the mechanism—the direct effect remained significant (ASIA A/B: OR 4.29; 95% CI: 1.19 - 15.5; p = 0.026) (Table 4). Collapse likewise remained associated with thoracic location after adjustment for age and sex (OR 2.84; 95% CI: 1.10 - 7.29; p = 0.030) (Table 5) and with evacuation without medical support (OR 37.2; 95% CI: 6.4 - 214.8; p < 0.001) (Table 6). The total effect of the mechanism on one-year mortality was not significant (unadjusted OR 1.64; 95% CI: 0.67 - 4.01; p = 0.35). In a model adjusted for ASIA grade, location, and mode of evacuation—that is, for the whole set of presumed intermediates—no direct contribution of the mechanism was found (OR 1.18; 95% CI: 0.19 - 7.42; p = 0.86), mortality being dominated by the presence of a complete injury (Table 7). This model, limited by the number of events and by adjustment for intermediate variables, remains exploratory and does not constitute a formal mediation analysis.

Evacuation was almost always without medical support (25/26, 96% vs 36%; OR 44.1; 95% CI: 5.7 - 340.2; p < 0.001; Cohen’s h = 1.46). Overall, one-year outcome was markedly less favorable after collapse: only 12% of patients progressed favorably versus 48% (p < 0.001), for a mortality of 42% vs 31% (p = 0.35). In the absence of standardized ASIA grading at follow-up, neurological recovery could not be quantified more finely.

Cohort mortality was almost entirely concentrated on complete injuries: 39 of the 44 patients classified as ASIA A at admission died within the year (88.6%), versus 1 of the 76 patients with grade B to E (1.3%) (Table 8). Deaths occurring after collapse were, according to the records, exclusively attributed to postoperative infection (8/11) and to thromboembolic complications (3/11); the share of deaths attributed to postoperative infection was higher there than in the comparison group (8/11 vs 3/29; p < 0.001) (Table 9). This comparison, however, is not conditioned on having been operated on, whereas the entire collapse group was, against 67% of the comparison group. The documented times (in days) to hospitalization (median 1 day) and to surgery (median 3 days) did not differ significantly between groups (Table 10).

Table 8. One-year mortality by ASIA grade at admission (n = 120).

Fisher exact test p < 0.001. Firth penalized OR from the multivariable model of Table 7: 194.3 (95% CI: 43.4 - 1875.0).

Table 9. Probable causes of death by mechanism (n = 40 deaths).

Probable causes recorded in the medical files, without independent validation or systematic confirmation (microbiological, imaging, or autopsy). The share of deaths attributed to postoperative infection differed between groups (8/11 vs 3/29; Fisher exact test p < 0.001), but this comparison is not conditioned on having been operated on, whereas the entire collapse group was, against 67% of the comparison group. The registry did not document the non-fatal complications of survivors.

Table 10. Prehospital and therapeutic times (median [IQR], in days).

No significant difference between groups. Times collected in days (limited granularity); no precise hourly prehospital time was available. Admission-to-surgery time is computed among operated patients only (26/26 in the collapse group and 63/94 in the comparison group); the 31 non-operated patients of the comparison group are excluded from that row. Mann-Whitney test.

Introducing the recruiting center as a covariate altered neither the direction nor the order of magnitude of the main associations: collapse remained associated with severe injury (OR 6.94; 95% CI: 2.24 - 25.88; p < 0.001; a value to be compared with the unadjusted total effect of 6.81, this model not including injury location) and with evacuation without medical support (OR 25.57; 95% CI: 5.90 - 241.54; p < 0.001), and exerted no direct effect on mortality (OR 1.42; 95% CI: 0.11 - 15.89; p = 0.772). The association with thoracic location retained a comparable magnitude but lost precision, its confidence interval including unity (OR 2.55; 95% CI: 0.96 - 6.89; p = 0.061): this result, more fragile than the others, must be regarded as suggestive (Table 11). The analysis restricted to the 44 gold miners recovered the same associations at reduced sample size: severe injury 81.8% vs 45.5% (OR 6.00; 95% CI: 1.54 - 29.16; p = 0.009) and evacuation without medical support 100% vs 50.0% (OR 41.69; 95% CI: 4.02 - 5811.47; p < 0.001) (Table 12). At comparable occupation, the excess severity and the lack of medically staffed evacuation therefore remained associated with collapse in this restricted comparison rather than with membership of the gold-mining sector alone; residual confounding by the accident setting and by differences between collapse and road-traffic injuries cannot, however, be excluded.

Table 11. Effect of collapse after adjustment for the recruiting center.

Firth regression; each row corresponds to a distinct model. The severity model includes neither location, nor age, nor sex: it estimates a total effect adjusted for center, to be compared with the unadjusted OR of 6.81 and not with the direct effect of Table 4.

Table 12. Analysis restricted to gold miners: collapse versus road traffic accident.

Firth regression. The complete absence of medically staffed evacuation in the collapse group produces complete separation: the upper bound of the confidence interval has only indicative value. The 45th gold miner, injured by a fall into an artisanal shaft outside a collapse episode, belongs to neither group.

In this cohort, gold mining occupied a major place, being both the most represented socio-occupational category (37.5%) and the second traumatic circumstance (21.7%). This share is rarely reported in African hospital series of TSI [3]. It matches the accident profile of gold mining described in Ghana, Zimbabwe, and Ethiopia, where shaft collapse is the leading cause of serious accidents [6]-[8], as well as the TSI from traditional mine collapses reported in Mali [9]. As informal gold mining spreads across sub-Saharan Africa, the TSI related to it could constitute a growing share of the global burden of spinal cord injury, which justifies their consideration in the global neurosurgery agenda [1][3][10][11].

The main originality of this work is the identification of a distinct injury pattern, characterized by a predominance of thoracic locations, a high frequency of compression and burst-type fractures, and increased neurological severity. The association with thoracic location was maintained after adjustment for age and sex (OR 2.84; p = 0.030) but lost its precision after adjustment for center: it must be regarded as suggestive. It could be explained by the vertical compressive load of burial: unlike road traffic accidents, which often induce flexion-distraction or hyperextension injuries, the collapse of earth and rock subjects the worker to a massive axial load, resulting in burst-type fractures and severe cord compression [12]. This morphology corresponds to type A fractures of the AO Spine thoracolumbar classification [13], but the absence of systematic grading in the records did not allow this to be verified.

Three reservations, however, limit the scope of this biomechanical interpretation. The sub-segmental distribution was available only in the collapse group, so that the observed predominance of the T9 - T12 segment (6 of 13 cases) rests on no comparator; moreover, the T9 segment does not belong to the thoracolumbar junction in the strict sense (T10 - L2), and nearly a third of the thoracic injuries were located at the upper T1 - T4 segment, which the axial-load model does not explain. Above all, an axial load concentrated on the junction should be accompanied by an excess of high lumbar injuries, whereas lumbar locations were, on the contrary, less frequent after collapse (12% vs 32%). This injury signature, consistent with West and Central African series of thoracolumbar spine trauma [14][15], suggests that a potentially unstable thoracolumbar injury should be sought in any victim of a mining collapse, but it remains a hypothesis to be confirmed by studies combining standardized radiological analysis with reconstruction of the mechanisms at mining sites.

The proposed conceptual model (Figure 4) synthesizes the observed relationships and the plausible pathophysiological mechanisms linking collapse to neurological severity and then to prognosis.

Collapses were accompanied by higher neurological severity: 85% of injuries were ASIA A/B, with a large effect size (Cohen’s h = 0.87). The unadjusted total effect reached an OR of 6.81 (95% CI: 2.18 - 21.3), and the direct effect, after accounting for injury location, remained significant (OR 4.29; p = 0.026). In this model, location strongly modulated severity: lumbar injuries were markedly less severe than cervical ones (OR 0.09; p < 0.001), reflecting involvement of the cauda equina rather than the cord, while thoracic injuries were of a severity comparable to cervical ones. The absence of intermediate grades (ASIA C/D) could reflect a polarized spectrum (Figure 5), the energy of burial leading either to minor injuries or to complete injuries [16][17]; the observation covers 26 patients and calls for caution.

Figure 4. Conceptual model of the proposed pathophysiological sequence, from burial to neurological severity and then to prognosis. The odds ratios shown come from the multivariable Firth models (Tables 4-7). The dashed arrow represents the expected link between the absence of medically staffed evacuation and death, not found in the adjusted model (OR 1.58; 95% CI: 0.46 - 5.61).

Figure 5. Distribution of the admission ASIA grade (A - E) by traumatic mechanism, artisanal mine collapse (n = 26) versus other mechanisms (n = 94). No intermediate grade (ASIA C or D) was observed in the collapse group (Fisher-Freeman-Halton exact test p < 0.001; Cramér’s V = 0.39; Cochran-Armitage trend test p = 0.012).

The prognostic value of the ASIA grade [18][19] is clearly confirmed here, but the divide runs between grade A and grades B to E, and not between the A/B composite and the rest: almost all deaths occurred in complete patients, mortality of grades B to E remaining marginal. The ASIA A/B composite, relevant for describing initial severity, must therefore not be read as a homogeneous prognostic marker. Our results are compatible with the hypothesis of a prognosis mediated by the high frequency of complete injuries, but they do not establish it: the mortality model adjusts for ASIA grade, that is, for the presumed mediator, which by construction cancels out the share of the mechanism’s effect passing through it. A formal mediation analysis, which the current sample sizes do not allow, would be needed to demonstrate it.

The most concerning public-health finding is the near-total absence of medically staffed evacuation: 25 of the 26 collapse victims (96%) were transported by means without medical support, versus 36% in the comparison group (Cohen’s h = 1.46). Among the associations of the mechanism, the one linking it to the absence of medically staffed evacuation had by far the greatest magnitude (adjusted OR 37.2), which designates the isolation of clandestine gold-mining sites and the absence of structured emergency services as major determinants of initial management. It resists adjustment for center and is found among gold miners alone, where all collapse victims, but only half of road-traffic-accident victims, were evacuated without medical means: it therefore relates to the mechanism and the place of the accident more than to the occupation. This finding matches regional data, where more than 90% of the African population has no access to formal emergency services and where prehospital care of the spinal-injured patient remains rudimentary [20]. The World Health Organization has made emergency care systems a priority of universal health coverage [21]; in the most deprived settings, community first-aid programs and alert networks in mining areas constitute a realistic first step [20][22]. The documented times, collected with a granularity limited to the day, did not differ between groups: the deficiency lies more in the quality of immobilization and transport than in the delay, which prospective studies will have to clarify. The absence of medically staffed evacuation, however, exerted no independent effect on mortality once the ASIA grade was taken into account (OR 1.58; 95% CI: 0.46 - 5.61).

One-year mortality in this cohort was almost entirely borne by complete injuries: 39 of the 44 patients classified as ASIA A at admission died within the year (88.6%), versus a single one of the 76 patients with grade B to E (1.3%), that is, a penalized OR of 194.3 (95% CI: 43.4 - 1875.0). Lethality of this order exceeds what most sub-Saharan hospital series report and must first be read in the light of recruitment: the exclusion of patients lost to follow-up before one year removes from the cohort subjects whose survival is probable, which mechanically inflates the observed mortality among complete-injury patients. It nonetheless illuminates the distribution of causes of death: in the collapse group, postoperative infection accounted for 8 of the 11 deaths, versus 3 of 29 in the comparison group. Several factors specific to this mechanism may contribute—telluric contamination of wounds during burial, a median delay of three days before surgery, prolonged recumbency, limited resources for surgical-site infection prevention—but the crude comparison does not account for the fact that all collapse-group patients were operated on, versus two-thirds of the comparison group, which alone suffices to raise the expected share of postoperative infections. In the absence of microbiological validation, this hypothesis cannot be tested here; it does, however, designate an actionable target: initial decontamination, antibiotic prophylaxis, and shortening of the time to surgery in burial victims.

The concentration of these injuries among men of working age makes them a socio-economic problem of the first order: beyond the individual disability, often permanent in cases of complete injury, they represent a loss of productivity and a factor of household impoverishment in regions where gold mining is frequently the main source of income. These results argue for a dual strategy: upstream, the prevention of artisanal mining accidents (regulation of gold mining, securing of galleries, protective equipment) and its integration into national policies against serious injury; downstream, the strengthening of prehospital care, spinal immobilization, surgical-site infection prevention, and neurosurgical capacity [17][23][24].

The retrospective nature exposes the study to information and selection biases, two of which act in opposite directions: the inclusion of complete records only and the absence of any count of deaths occurring at mining sites or during transport lead to underestimation of the true lethality of collapses, whereas the exclusion of patients lost to follow-up before one year, which bears directly on the outcome, leads on the contrary, to overestimation of the mortality measured in the analyzed cohort. Patients came exclusively from two neurosurgical centers, which limits national representativeness: the most benign forms probably never reach them, hence an over-representation of severe forms linked to referral bias. Adjustment for center leaves the associations with severity and evacuation unchanged but renders imprecise the one observed with thoracic location, which correspondingly weakens the biomechanical argument developed above. A gold miner injured by a fall into an artisanal shaft, a mechanism close to burial, appears in the comparison group, which tends to attenuate rather than exaggerate the observed contrasts.

The severity and mortality models adjust for variables lying on the presumed causal pathway: they estimate a direct effect and do not allow the mediation discussed above to be established. The size of the collapse group (n = 26) limits power, particularly for mortality, whose absence of a significant difference by mechanism cannot be read as evidence of equivalence; it also explains the width of several confidence intervals, such as that of the analysis restricted to gold miners for evacuation, where separation is complete, having only indicative value. The multiplicity of comparisons requires cautious interpretation, which is why effect sizes were favored. Neurological outcome could be assessed only globally, in the absence of standardized ASIA grading at follow-up; non-fatal complications of survivors were not recorded; the thoracic sub-segments were available only for the collapse group; the absence of standardized radiological data (AO Spine classification, TLICS score [13]) limits the fine morphological characterization of injuries, and the causes of death, collected from records, received no independent validation. These results, drawn from a referral hospital cohort, are therefore exploratory and hypothesis-generating: based on observational associations, they allow neither estimation of the population incidence of gold-mining-related injuries nor inference of a causal relationship. Multicenter prospective studies incorporating artisanal mining sites and standardized neurological assessments are needed to confirm these results and guide prevention policies.