Introduction

Non-traumatic spinal cord injury (NTSCI) is damage to the spinal cord due to causes other than trauma, most commonly resulting from degeneration of the vertebral column, metastatic disease, inflammation, infection or vascular aetiologies [1, 2]. NTSCI incidence has been reported to vary globally between 6 and 76 per million of population. While this is a relatively low incidence, it is believed to be greater than traumatic spinal cord injury (TSCI) in many developed countries [3]. The functional, social and financial impact from NTSCI is comparable to that of TSCI and places a significant burden on the healthcare system [4]. Symptoms of NTSCI can be slow to develop and diagnosis can be significantly delayed, which may impact access to specialist spinal rehabilitation and classification of injury level and severity. NTSCI generally affects older people, and as a result its incidence is only expected to increase with an ageing population [5].

People who have sustained a spinal cord injury from any cause prioritise return to walking as a key rehabilitation goal [6]. This, along with healthcare system pressures mentioned above makes predicting the ability to walk of utmost importance in the clinical setting. It enables clinicians to design individualised rehabilitation programs, guide decision-making around carer needs, undertake equipment prescription and home modifications, provide appropriate and tailored counselling regarding likely functional outcomes, and support the prioritisation of limited and expensive public healthcare resources [5,6,7,8,9].

In 2011 van Middendorp et al. developed a clinical prediction rule (CPR) for walking after TSCI. The rule utilises five simple variables: age (less than and greater than 65 years), motor scores of quadriceps and gastrocsoleus and light touch sensory scores of dermatomes L3 and S1 to predict walking ability 1 year post injury [10]. The rule was reported to be correct in predicting walking or not walking in more than 95% of cases at 1 year, area under the curve (AUC) 0.96 (95% confidence interval (CI), 0.94–0.98). This CPR has been externally validated for three other groups of people with TSCI. Two in accordance with the original paper (van Sifhoult [11] and Malla [12]) used the Spinal Cord Independence Measure (SCIM) item 12 to determine independent walking [13], while Hicks et al. determined independent walking based on the FIM score [14]. All three studies used retrospective data from registries or medical records.

While the CPR was developed for people with all categories of TSCI, two recent projects investigated the prognostication according to the International Standards for Neurological Classification of Spinal Cord Injury (ISNCSCI). The ISNCSCI classifies people as complete (American Spinal Injury Association Impairment Scale (AIS) A) or incomplete (AIS B to D) injury [15]. Phan et al. conducted a review of the van Middendorp and Hicks versions of the rule comparing the ability to predict walking in two groups: AIS B and C vs. AIS A and D [16]. They found the prognostication of walking for people with an AIS B or C classification was less accurate than for people with an AIS A or D classification. The AUC for AIS A and D (0.95, 95% CI, 0.93–0.98) was significantly higher (p = 0.00034) than for AIS B and C (0.83, 95% CI, 0.77–0.90) [16]. Malla et al. [12] found similar results with AIS A having a predicted probability of walking 5% (95% CI, 0.01–0.13) and observed probability of 9% (95% CI, 0.04–0.13) and AIS D a predicted probability of walking 95% (95% CI, 0.88–0.98) and an observed probability of 97% (95% CI, 0.88–0.98). The greatest discrepancy in prognostication was for AIS B with a predicted probability of walking of 26% (95% CI, 0.13–0.43) and an observed probability of 58% (95% CI, 0.41–0.74) [16]. These results highlight the importance of injury severity and the influence it may have on predicting who will walk. Studies of mobility outcomes in people with spinal cord damage have shown differences between the AIS grades in TSCI and NTSCI groups, with a higher prevalence of incomplete injuries (AIS B, C and D) following NTSCI [4]. Despite this difference in AIS scores the functional outcome, independent ambulation of NTSCI and TSCI appear to be similar [4, 17].

In order for the CPR to be used in the NTSCI population it must be validated to determine if it can be applied to this group of individuals that have a very different clinical and demographic profile to the TSCI population [18]. The aim of this study was to investigate whether the van Middendorp clinical ambulation rule can accurately predict walking in people with NTSCI.

Method

Data collection and setting

People with NTSCI who were consecutively admitted to the spinal rehabilitation service at Caulfield Hospital, Alfred Health, Victoria, Australia, between August 2013 and July 2017 were invited to take part in the study. People were included if they were over 18 years of age, had a confirmed diagnosis of NTSCI, and had adequate English to provide informed consent. People were excluded if they had a congenital SCI, a lower limb peripheral nerve injury, lower limb fractures that limited ambulation and cognitive, psychological or psychiatric impairments that limited participation in this study.

Study outcomes and variables

Demographic characteristics (age at onset, sex) and clinical outcomes were recorded. The clinical outcomes included level (paraplegia vs. tetraplegia), AIS grade, as determined using the ISNCSCI [15], and aetiology of NTSCI, as classified using the International NTSCI dataset [19].

As outlined earlier, the diagnosis of NTSCI can be delayed, therefore it was decided that we would not exclude people if completion of their ISNCSCI occurred beyond 15 days following onset of TSCI, as was the approach followed in the van Middendorp study [10]. Onset of NTSCI was determined as the date the participant was admitted to an acute hospital. Aligned to the original study, the primary outcome was the ability to walk at 6 months and 1 year following onset, using the SCIM version III, item 12 (ability to walk 10 m independently with or without an assistive device) [20]. A cut off SCIM indoor mobility score was used i.e. an item 12 SCIM score of 0–3 was defined as unable to walk independently, a score of 4–8 as able to walk independently. The SCIM assessment was conducted by interview in accordance with previous studies validating this approach [21].

The results were calculated in a two-stage process. Firstly, the predicted score was calculated based on the corrected equation from Malla \(\frac{{e^{ - 3.273 \,+\, 0.267 \,\times\, score}}}{{1 \,+\, e^{ - 3.273 \,+\, 0.267 \,\times\, score}}}\) [12, p. 8]. The score entered into this equation ranged between −10 and 40, in accordance with the van Middendorp study [10] and was based on the best four neurological variable scores i.e. the motor scores of quadriceps and gastrocsoleus and light touch sensory scores of dermatomes L3 and S1, regardless of side were extracted from each participant’s medical record and entered into the equation together with their age. This provided a predicted to ‘walk’ or ‘not walk’ result based on the van Middendorp original paper. Secondly, the predicted value was compared with the participant’s actual walking ability derived from their SCIM item 12 as a ‘walker’ or ‘non-walker’. Participants were then classified as true positive (predicted to walk and actually walked), false negative (predicted not to walk but did walk) and so on.

Statistical analysis

Descriptive statistics with absolute and relative frequencies were calculated for categorical variables, with mean and standard deviation for continuous variables. Data were analysed using SPSS 25. A receiver-operating curve (ROC) was plotted to assess the (AUC). Sample size analysis showed that a sample size of 50 would be sufficient to detect a minimum AUC of 0.75 with an estimated standard error of 0.07 [22].

We certify that all applicable institutional and governmental regulations concerning the ethical use of human volunteers were followed during the course of this research.

Results

A total of 142 adults were admitted to Caulfield Hospital with a confirmed diagnosis of NTSCI during the recruitment study period. Eighty-six met the inclusion criteria and were approached to take part in this project. Sixty-two people consented; one subsequently changed diagnosis so was withdrawn and was not included in any analysis. Three were lost to follow-up and six died during the project, five of whom completed solely the 6-month follow-up and are included in the 6-month analyses. Fifty-two participants completed all data collection and were classified to five aetiology groups (Table 1). The characteristics of all participants including those lost to follow-up are outlined in Table 1. Time to completion of AIS was a median of 44 days (interquartile range 33–62 days) from acute admission.

Table 1 Participant characteristics
Full size table

Analyses of participants lost to follow-up at 12 months showed no significant difference in hospital length of stay from acute admission to discharge from rehabilitation (p = 0.08), age (p = 0.57), time to completion of ASIA (p = 0.14). The AUC for the whole cohort were 0.93 (95% CI, 0.86–1.00) at 6 months and at 0.94 (95% CI, 0.86–1.00) 12 months (Fig. 1). A comparison between our results and previous studies is shown in Supplementary Appendix 1.

Fig. 1
figure 1

Area under the curve for whole cohort at 6 months (a) and 12 months (b)

Full size image

Overall, the CPR correctly predicted those who would and would not walk in 75% of participants at 6 months and 82% at 12 months. The CPR correctly predicted 100% for those with an AIS A and B classification at 12 months. At 12 months the CPR correctly predicted 22 of 23 participants who did not walk (Table 2) and 21 of 29 who walked (95% sensitivity, 7% specificity). However, differences were noted by diagnosis, age and injury classification, with poor specificity evident in those with AIS C and D (Tables 3, 4). Given there were only three participants classified with an inflammatory diagnosis, sensitivity and specificity were not calculated for this diagnostic group.

Table 2 Concordance matrix full cohort
Full size table
Table 3 Predictive accuracy for AIS classification, diagnosis and age
Full size table
Table 4 Concordance matrix by AIS classification, diagnosis and age
Full size table

Discussion

This study is the first to investigate the predictive accuracy of the van Middendorp CPR for a NTSCI population. Our results were similar to those previously recorded for people with TSCI and it appears equally good at predicting the ability to walk in an NTSCI population, as supported by the AUC analysis (present study 0.935 (CI 95% 0.855–1.00) van Middendorp study 0.956 (CI 95% 0.936–0.976) Supplementary Appendix 1). However, the CPR was better at identifying those who would not walk independently than those who would. In predicting those who cannot walk at 12 months post injury, the CPR was incorrect for one participant with an extremely rare diagnosis of infective meningoencephalitis, who had a medical research council grade of 0/5 for both quadriceps and gastocsoleus and a score of 0 for L3 and S1 sensation on initial assessment. This unexpected outcome, given the initial presentation, highlights the importance of ongoing clinical review for at least 12 months post NTSCI. This is further supported by the 14% of participants who, while predicted to walk, were not walking at 6 months, requiring the full 12 month period for their walking independence to manifest (Table 2).

This is the first time the accuracy of this CPR has been reported based on those predicted to walk, and not walk. While it is possible that AIS A and B groups could be predicted not to walk without a prediction tool, the CPR provides additional quantitative data (% likelihood to walk/not walk) for individuals with NTSCI, and their treating team. This additional data, could have significant implications in the clinical setting by supporting individuals with AIS A and B NTSCI in their understanding of, and adjustment to the likely outcome of being unable to walk through tailored education and counselling. Clinicians will also be able to prioritise intervention and facilitate early discharge planning to potentially decrease length of stay and cost on the healthcare system. Age, diagnosis and AIS classification did not impact on the CPR’s accuracy in predicting those who would not walk following NTSCI.

In this study the CPR was less accurate in predicting those who will walk, with a false positive percentage for the overall cohort of 27% at 12 months post injury (Table 2). Given the heterogeneity of the NTSCI population, the data were analysed further in line with the existing literature to determine which, if any, variables may impact the result. Phan et al. on closer examination of the original van Middendorp data, reported people with an AIS A classification had a sensitivity of 37% and specificity of 98%, while those with an AIS D classification had 100% sensitivity, results that are very different to those reported here (Table 3) [16]. Malla on the other hand found the rule was least accurate for people with AIS B, while the present study found the best prognostication was for those classified as AIS A (100%) and least for AIS C (Table 3) [12]. Six of the 11 participants with an AIS C classification predicted to walk did not, indicating that the CPR was no better than chance (50% specificity). Those with AIS D also had low specificity (66%). It has been hypothesised that the key muscle groups for functional walking are hip extensors, hip flexors and hip abductors [23], none of which are assessed in the CPR. People with AIS C classification have less than grade 3 strength in greater than 50% of muscles below the level of injury, i.e. they may not have anti-gravity strength in more than 50% of the muscles required for walking [15]. This may provide an explanation for the wide variability in predicting the ability to walk of people with an AIS C classification in this study. Phan et al. hypothesised that the original rule achieved such a high predictive accuracy by combining all classifications and including a disproportionally large number of people with an AIS A and AIS D classification [16]. The results of this study lend further credence to this with a similar AUC but very clear difference in sensitivity and specificity by AIS classification.

One of the unique features of NTSCI is the range of aetiologies that can result in damage to the cord including degenerative spine conditions, infection or tumours. The varied aetiologies may influence the ability to walk, and therefore the ability for the rule to predict who will walk. The results of this study demonstrated that the rule holds well for those with degenerative (100% sensitivity and 86% specificity), vascular causes (100% sensitivity and 82% specificity), and slightly less well for infective pathologies (88% sensitivity and 71% specificity). For the tumour diagnosis group the sensitivity and specificity was 50% indicating that the accuracy of prediction was no better than chance. The number of participants with an inflammatory pathology was too small to complete sensitivity and specificity of this CPR for this subgroup and was therefore not included in our results. The heterogeneity of NTSCI and the inherent within-group variables are difficult to control and may confound conclusions that might be drawn about NTSCI subgroups. Further research in this area is required to confirm these preliminary results.

The previous investigations of this CPR used a cut off of 15 days for initial INCSCI classification. The present study had a median of 44 days (IQR 33–62), almost treble the period of time reported previously. It was interesting to note that despite the increase in time to completion this did not appear to affect the overall accuracy of the AUC or the predictive accuracy of the rule. Whilst this requires further detailed examination, it remains an important finding.

Limitations

There are several limitations of this study. The sample size, while sufficient for the ROC analysis, was relatively small. The heterogeneous nature of NTSCI makes research in this population difficult as there is significant variability in the cohort. The current study did not take into account variability in medical management, including surgical and non-surgical interventions. While trends have been identified in relation to injury classification, age and aetiology, these were post hoc analyses and they are unable to be generalised to the entire NTSCI population. Further studies of a larger sample size may address the issue of heterogeneity in this population. While all eligible participants were approached to participate in this study, not all consented. The outcome for these individuals is unknown and may have altered the overall outcome. While a number of participants were lost to follow-up, including six who died during the project, three diagnosed with tumours and three with vascular conditions, participants are representative of people with NTSCI [2]. This study was undertaken at a state-wide NTSCI referral centre as part of routine clinical care by clinicians skilled in SCI assessment and management. The mean time to completion of AIS assessment is acknowledged as a further limitation however as discussed above does not appear to reduce the accuracy of the CPR in comparison to previous studies. Further van Middendorp identified in post hoc analysis that even if the examination was performed up to 15 days post injury there was no pronounced effect on the accuracy of the prediction [10]. This context increases the generalisability of this CPR across other healthcare settings that provide acute and rehabilitation services to people with NTSCI.

Conclusion

The van Middendorp CPR appears to work well in the NTSCI population, with high sensitivity and a similar AUC to the original study. Its ability to predict those who cannot walk is particularly useful in clinical practice. Further studies are recommended to determine how to strengthen this CPR in the NTSCI population to increase its accuracy in predicting who will walk and to determine the overall impact of AIS classification, diagnosis and age on the accuracy of this rule in clinical practice. It should be used judiciously when applied to individuals with an AIS C classification.