Background
With contemporary techniques, transverse myelitis following radical radiotherapy (RT) for head and neck cancer (HNC) is extremely rare. The QUANTEC paper of 2010 quotes a risk of < 1% at 54Gy with conventional fractionation [
1], and recent randomised trials have mandated maximum spinal cord (SC) doses much lower than this [
2,
3]. However, milder spinal cord toxicity in the form of L’Hermitte’s sign (LS) may be more prevalent.
LS is characterised by electric-shock sensations down the spine and into limbs on neck movement (particularly flexion). It is a well-recognised symptom of demyelinating conditions such as Multiple Sclerosis (MS) [
4], and can also occur as a side effect of radiation to the cervical or thoracic spinal cord [
5]. The mechanism is believed to be transient inhibition of oligodendrocyte proliferation leading to reversible demyelination [
6,
7]. LS usually develops in the first few months after radiotherapy, and seldom lasts more than 6 months [
7‐
9], but can be unpleasant and distressing for patients. No clear link between radiogenic LS and progressive irreversible myelitis has been established, although delayed radiation-induced myelopathy causing paralysis may be preceded by LS [
9].
Historical series, in which patients were treated with conformal, field-based techniques, report a risk of LS following RT for HNC between 3 and 13% [
9‐
11]. More recent work on LS following Intensity Modulated RT (IMRT) for thoracic and head and neck malignancy describes an incidence between 15 and 29% [
5,
12,
13]. Since IMRT permits superior sparing of critical organs at risk (OARs), and given our previous understanding of the dose-response relationship of the spinal cord [
1,
14], these results are surprising. Recent research has indicated that younger age and higher maximum dose are risk factors, and inferred that concomitant chemotherapy may also be implicated [
13,
15].
IMRT can generate steep dose gradients in order to adequately treat target volumes, whilst sparing OARs such as the SC. This often results in inhomogeneous dose distributions across OARs, and work on rat models has suggested that such inhomogeneity may be a risk factor for LS [
16]. Whilst some clinical data appears to support this notion [
12], no convincing evidence of a ‘bath and shower’ effect has yet been seen [
13].
We sought to examine the link between radiation dose, comorbidities, and concomitant systemic therapy and the subsequent development of LS using a logistic regression model, in a cohort of patients with HNC recruited to the VoxTox study.
Discussion
This is the largest prospective study of L’Hermitte’s syndrome in HNC patients in the era of IG-IMRT. LS incidence in our cohort is higher than previously reported (3.6–29%) [
5,
11‐
13,
15], although mean onset and duration of symptoms were similar. The high incidence may be due to its prospective nature, and the fact that a single positive response classified patients as having LS. As LS is transient [
9], we believe this definition is justified. Furthermore, these data come from a large, prospectively evaluated cohort of patients treated with a homogeneous protocol including daily IG and positional correction; thus the observed difference in LS incidence is credible.
Pak and colleagues suggest that concurrent neurotoxic chemotherapy may contribute to a higher incidence of LS [
13]. The odds ratio for cisplatin on univariate analysis is 1.23 (95% confidence interval is 0.5 to 2.3). Thus, we cannot conclude that there is an effect from cisplatin. It may be that our study was underpowered to detect such an effect, and there may be a confounding effect of separate factors linked with cisplatin use such as age, hypertension and diabetes.
Surprisingly, V
20% was significantly lower on univariate analysis in patients with LS, although its effect was insignificant in the multi-variate model. Conversely, V
40Gy and V
40% were insignificant on univariate analysis but significant according to the logistic regression model, consistent with previous work [
13]. Of note, TomoTherapy plans confer excellent spinal cord sparing, meaning only 27.3% of all the patients in our cohort had a partial cord volume receiving 40 Gy or more. Therefore, our V
40Gy results should be interpreted with caution. Mean D
max in our cohort was 36.4 Gy compared to 39.1 to 42.5 Gy in similar studies using VMAT and IMRT respectively [
12,
13], yet more patients in our cohort reported LS than in these studies. Interestingly, a study on 105 patients receiving thoracic IMRT for lymphoma reported a mean D
max of 33.5 Gy and had an LS incidence of 29% [
5], also suggesting factors other than dose may be important. According to our multi-variate analysis, age, diabetes, and unilateral neck radiation may be related factors, although given the sample size and degrees of freedom in the model,
p-values for all factors should be considered borderline significant, and interpreted with caution.
Patients developing LS were younger than patients without LS. The difference was insignificant on univariate analysis and binary logistic regression, but significant in the ordinal regression, suggesting younger patients have more severe symptoms if they do develop LS. These findings are not new: Mul and co-workers found a mean age of 52 in LS patients compared with 61 in non-LS patients [
15], whilst Leung et al. found a decreased risk in those over 60 [
10]. Although younger patients were more likely to receive cisplatin (
p = 0.0017) they were not more likely to receive a higher maximum or mean SC dose (
r = 0.030 and − 0.203 respectively, Pearson correlation coefficient).
Intriguingly, our data suggest that patients with diabetes are less likely to develop LS, a previously unreported finding. It should be noted that 10.2% of our cohort had diabetes compared to 3.9% and 4.1% in similar studies [
13,
15]. Nine of 13 diabetic patients in our cohort took metformin (the one diabetic patient with LS also took metformin). This drug has been suggested to have anti-inflammatory and anti-oxidant neuroprotective effects in mouse models of MS [
25,
26], whilst a significant anti-inflammatory effect of metformin and pioglitazone has also been shown in patients with MS and metabolic syndrome [
27]. However, more investigation would be needed to ascertain whether these benefits are also seen in radiation-induced demyelination.
Lastly, patients with LS were significantly more likely to have had unilateral neck radiation. A ‘bath and shower’ effect, whereby radiation tolerance is reduced if an area of high dose is surrounded by an area of low dose, was first demonstrated in rat spinal cords [
16,
28]. It is hypothesised that low dose radiation prevents oligodendrocyte migration to repair damage, and can alter gene expression [
29,
30]. This effect was sought, but not found in a previous clinical study [
13]. However, Ko and colleagues observed LS exclusively in patients that received unilateral radiotherapy (5 of 33 patients), and suggest that axial dose inhomogeneity may contribute to the development of LS [
12]. Other authors postulate that anterior-posterior dose gradients may be significant because of spinothalamic tract damage [
31].
In addition to a relationship between unilateral neck irradiation and LS, we also found that unilateral neck treatment plans had much more inhomogeneous SC dosimetry. Although this inhomogeneity was not an independent risk factor for LS in the multi-variate model, this may be due to the close association with treatment laterality, and the possibility of diluted statistical power within the model. Our interpretation of these data are as follows; firstly, to corroborate previous findings of higher LS risk in patients undergoing unilateral neck treatment, secondly to suggest that inhomogeneous SC dose distributions may be a mechanistic factor in this effect, and finally that a paradoxically rising incidence of LS may in part be due to the greater SC dose inhomogeneity that IMRT confers. It is clear however that understanding of neurological response to complex dose distributions is incomplete.