Predictive factors of prognosis after radiation and steroid pulse therapy in thyroid eye disease

To identify predictive factors of prognosis after radiotherapy with concurrent steroid pulse therapy for thyroid eye disease, retrospective analyses were performed among 77 patients. Clinical activity score and magnetic resonance imaging were used to evaluate degrees of orbital inflammation. As a pre-treatment work-up, the thyroid-stimulating antibody (TSAb) level was measured. During a median follow-up of 25.0 months, the 2-year cumulative relapse-free rate (CRFR) was 80.9%. In the univariate analysis, a worse 2-year CRFR was significantly associated with the presence of optic neuropathy (P = 0.001), a higher TSAb rate (P = 0.001), and lower standard deviation (SD) of signal intensity at the extraocular muscle in T2-weighted images (P = 0.006). In the multivariate analysis, TSAb rate and SD affected the CRFR independently. When TSAb activity of 2800% was set as a cut-off at 2 years after treatment, the predictive sensitivity and specificity of relapse were 81.2% and 90.6%, respectively. With regard to SD, the respective sensitivity and specificity values were 81.2% and 82.7% when 100 was set as a cut-off. In conclusion, high TSAb and low SD were significant risk factors for cumulative relapse in orbital radiotherapy. Cut-off values of 2800% for TSAb and 100 for SD may be suitable.

The results of Fisher's exact test performed on initial response data and univariate analysis performed on CRFR data are shown in Table 3. Only patients with higher thyroid-stimulating antibody (TSAb) rates had significantly worse initial responses (P < 0.05). In univariate analysis, a worse 2-year CRFR was significantly associated with the presence of optic neuropathy (47.7% vs. 86.3%, P = 0.001), higher TSAb rates (66.5% vs. 93.1%, P = 0.001), and lower region of interest (ROI) standard deviations (SDs) (67.7% vs. 94.1%, P = 0.006). In multivariate analysis, the TSAb rate (hazard ratio 1.010, 95% confidence interval [CI] 1.004-1.014, P < 0.001) and SD (hazard ratio 0.974, 95% CI 0.957-0.980, P < 0.001) affected the CRFR independently. Based on the results Twenty-two adverse events requiring medical intervention were observed in a total of 19 patients. These adverse events consisted of deterioration of dry eye in 13 cases, hepatitis in 3 cases, diabetes in 2 cases, cataract in 3 cases, and gastric ulcer in 1 case. All of these events were mild or moderate, and none were associated with serious sequelae. There was no significant correlation between the total dose of corticosteroid and the incidence of adverse events. The 3 patients who underwent a cataract operation were aged 58, 61, and 71 years at the time of the operation, and the relevance of irradiation and steroid administration was unclear. Although only 5 cases had been followed up for ≥7 years after treatment, secondary cancer did not develop during the follow-up period in any of these cases.

Discussion
The present study is the first report of an association between a higher TSAb rate and worse initial response or worse 2-year CRFR after radiotherapy with steroid pulse therapy. The results of the study suggest that a TSAb activity cut-off of 2800% is useful for determining the predictive sensitivity and specificity of relapse. TSAb is an autoantibody that induces orbital inflammation via receptors expressed in orbital tissues, and correlates with the degree of orbital inflammation in TED 8,9 . Accordingly, we suggest that it is reasonable that higher TSAb activity is a factor relevant to a higher relapse rate after radiotherapy with steroid pulse therapy.
We used SDs to objectively and quantitatively evaluate the homogeneity of extraocular muscles via MRI, and found that lower SDs were significantly associated with worse 2-year CRFRs. Lower SDs denote a homogeneous  extraocular muscle signal and may indicate more severe fatty infiltration and intramuscular oedema, which may have resulted in the higher relapse rate observed in the current study. Notably, SDs are not suitable for making predictions in cases of early relapse (Fig. 1). Using the marker combined with TSAb is preferable, especially in the early period.
Although TED in the active phase tends to be uniformly treated with an irradiation dose of 20 Gy, we propose a tailor-made treatment strategy whereby the intensity of treatment is based on predictive factors. Some studies have investigated reducing the total doses of irradiation or steroid administration 10,11 . Although it should not be applied to all cases uniformly, reducing the treatment intensity in patients without adverse prognostic factors is worthy of consideration. Conversely, in patients with adverse prognostic factors increases in total steroid or irradiation doses may be appropriate. Notably however, a total steroid dose of >8 g can cause fatal hepatopathy, and in a previous study there was reportedly no benefit associated with dose-escalation in routine orbital radiotherapy 12 . A combination of rituximab, cyclosporine, and somatostatin analogues may reduce the risk of relapse 13 . Further studies are needed to establish better treatment strategies for cases with and without adverse prognostic factors.
In a previous pilot study including 25 patients with moderately severe TED, pre-treatment CAS was significantly higher in patients who responded to radiotherapy (3.7 vs. 2.1; P = 0.008) 14 . Orbital inflammation was not evaluated via MRI in that study, however, mean pre-treatment CAS in the non-responders was 2.1, raising doubt as to whether all participants actually had orbital inflammation. In the current study, orbital inflammation activity was evaluated via CAS and MRI, and patients were deemed to be in active phase prior to radiotherapy with concurrent pulse therapy.
A previous study measured the SIR of enlarged extraocular muscles and cerebral substantia alba at the coronal section in STIR images from patients with TED who were treated with steroid pulse therapy 15,16 . The results of that study were inconsistent with the observation in the current study that SIR was significantly greater in patients in which treatment was effective than it was in those in which treatment was ineffective. A possible reason for this discrepancy is that different measurement parameters were used in the two studies, such as the position of the ROI, the size of the ROI, and the setting of pulse sequences.
In a previous MRI study, more favourable outcomes (reduction in muscle size or recovery from diplopia) were reported after combined treatment in patients with uniform T2-weighted image (T2WI) intermuscular intensity than in those with a non-uniform pattern 16 . There was no significant relationship between the SD and the effectiveness of combined therapy in the current study, but this difference may be due to the difference in the methods used to determine therapeutic effect in the two studies.
The current study had several limitations. Because the study design was retrospective, 16 patients were administered higher or lower methylprednisolone for pulse therapy. Furthermore, 22 patients were prescribed subsequent oral prednisolone, 34 patients underwent previous treatment for ophthalmopathy, and duration of ophthalmopathy had a large range (1.1-43.2 months). These limitations were mainly due to the selection criteria; patients who had a treatment history (in another hospital) were included in our study group. This introduced the possibility of confounding variables in the patient characteristics. Although the effectiveness of the treatment was evaluated 3 months post-treatment in most patients, some patients were evaluated at earlier or later follow-up time-points (median 3.3 months, range, 1.9-5.8 months). In conclusion, high TSAb values and low SD values were significant risk factors for cumulative relapse after orbital radiotherapy with concurrent steroid pulse therapy for TED. Furthermore, a novel parameter derived by combining TSAb and SD was a more sensitive prognostic predictor than either parameter alone. We propose the utilisation of a tailor-made treatment strategy in patients with and without prognostic predictive indicators.

Ethics approval and study design. The institutional review board of Aichi Medical University Hospital,
Japan, approved this retrospective observational study (application number 2017-H211). The study was conducted in accordance with the tenets of the Declaration of Helsinki and its subsequent amendments. Written informed consent was obtained from all patients.

Patients.
Patients with TED who were diagnosed in the active phase using CAS and/or MRI underwent radiotherapy with steroid pulse therapy at the Aichi Medical University Hospital. Data pertaining to 115 consecutive patients who underwent radiotherapy for active TED between August 2005 and March 2017 were initially reviewed. A total of 38 were subsequently excluded: 19 due to low CAS (0-1), 8 because they did not undergo steroid pulse therapy, 6 due to a short follow-up period (less than 6 months), and 5 due to insufficient ophthalmic examination data availability. The remaining 77 patients were included in the study.

Laboratory tests and ophthalmic evaluations.
Blood tests for pre-treatment work-up included a complete blood count; assessment of thyroid function, TSAb activity, liver and kidney functions; and hepatitis B and C serologic analyses. The TSAb measurement method at our institution was changed from the radio-immuno-assay (RIA) to the enzyme-immuno-assay (EIA) on 1 July 2014. Therefore, for a meaningful comparison, RIA-TSAb values were converted to the corresponding EIA-TSAb values via the following formula 17 : = .
Steroid therapy. Most patients (n = 61) underwent two or three cycles of steroid pulse therapy under a regimen of body weight ×10 mg/kg/day of methylprednisolone for 3 days per cycle. Patients treated early in the study period (n = 13) were administered 1000 mg of methylprednisolone per day, and 3 patients were administered reduced doses. Patients did not receive subsequent oral prednisolone in principle. However, in order to avoid sudden discontinuation, oral prednisolone (initial dose 0.2-0.4 mg/kg per day, tapering and discontinued 1-3 months after initial administration) was administered to 22 patients who already had been prescribed in a previous hospital.
Radiotherapy. Both the target area and orbital structures for which irradiation was avoided were contoured using treatment planning systems (either XiO, Electa, CMS, St Louis, MO, USA or Eclipse, Varian Medical System, Palo Alto, CA, USA). The clinical target volume was defined utilising the extraocular muscles and retrobulbar soft tissue, and the planning target volume was defined as the clinical target volume with an additional 0.5-cm margin. Irradiation was delivered to the planning target volume using 6-MV photons, with lateral opposing fields angled posteriorly at 2-4° to align the anterior field edges. Although a 0.5-cm multi-leaf collimator margin was utilised, the anterior-posterior margin was reduced to spare the lens and pituitary gland. A total dose of 20 Gy in 10 fractions was delivered to all but three patients during hospitalisation for steroid pulse therapy. Due to their young age, two patients received 12 Gy in six fractions, and one patient received 10 Gy in five fractions.

MRI examination.
Axial and coronal orbital MR images were obtained via a 1.5-Tesla scanner (MAGNETOM Avanto, Siemens Health-care, Erlangen, Germany) with a head surface coil. Typical pulse sequence parameters were spin-echo 4000/100 (repetition time ms/echo time ms) for T2WIs and inversion recovery 4000/180/80 (repetition time ms/inversion time ms/echo time ms) for short-tau inversion recovery (STIR) images. Section thickness, field of view, and matrix were 3-5 mm, 140 × 140 mm, and 256 × 220, respectively. Extraocular muscle thickness, proptosis, and SIR were measured using the digital calliper tool of a viewer (ShadeQuest/ViewR, Yokogawa Medical Solutions Corporation, Tokyo, Japan).
Referring to the previous study, the following parameters were measured by expert radiologists 15,16 . Extraocular muscle thickness was defined as the area of the region of interest (ROI), which was situated along the perimeter of the largest cross-section of the extraocular muscles in the coronal T2WI (Fig. 2). In conjunction with placing the ROI, the SD of the internal signal intensity was measured. Representative cases with high and low SDs are shown in Fig. 2a,b, respectively. Axial globe position was defined as the distance between the corneal eminence and the interzygomatic line on an axial T2WI through the optic nerve (Fig. 3a). The SIR was defined as the ratio of signal intensity of the muscles to that of the cerebral white matter (Fig. 3b).
Data collection and statistical analyses. Initial treatment response was evaluated in all cases based on the 10-point CAS at pre-and post-treatment. 'Responders' were defined as patients for who the CAS decreased by 2 or more points. A change in CAS of 0 or 1 was classified as 'no change' , and 'progressive disease' was assigned to patients for who the CAS increased by 2 or more points. During the follow-up period, cases requiring additional salvage anti-inflammatory therapy (including cases in which the patient refused treatment) were defined as cases of relapse. CRFR was calculated from the initial date of treatment until the last follow-up or the date of relapse. The association of characteristics with initial response was analysed using Fisher's exact test for categorical variables and for continuous variables. Student's unpaired t-test was used to determine the significance of differences between two sample means. Because CAS was an ordinal variable, the Wilcoxon signed-rank test was used to compare it before and after treatment. CRFR was estimated using the Kaplan-Meier method. Log-rank tests were used to compare the estimated CRFR among the subgroups in univariate analysis. The Cox proportional hazards model was used in multivariate analysis. P < 0.05 was considered statistically significant. Factors yielding a significance of P < 0.1 in univariate analysis were included in the multivariate analysis. The receiver operating characteristic (ROC) curve was used to evaluate relationships between relapse and selected variables. Based on the Youden index, the cut-off point was determined and well-balanced sensitivity and specificity values were obtained 18 . The predictive capacities of each of the selected variables were compared via area under the curve (AUC) analysis. All statistical analyses were performed with EZR version 1.33 (Saitama Medical Center, Jichi Medical University, Saitama, Japan), which is based on the R and R commander 19 .

Data Availability
The datasets generated during and/or analysed during the current study are available from the corresponding author on reasonable request.