Setting
In our department radiation therapy treatments can be performed using the TomoTherapy® Hi-Art® System 3.0 (Accuray Incorporated, Madison, WI), which involves a helical fan beam scan using 3.5MV photons and an arc-shaped xenon CT detector array mounted on the opposite side of the ring gantry Mega Volt Fan Beam Computed Tomography (MVFBCT).
All patients with head-neck tumour were immobilised using a 5-points mask: four fixing points for the neck and shoulders plus an additional one on the head (Acquaplast RT® Fibreplast ™ Thermoplastics, Avondale, PA, USA). The patient’s position was maintained by means of a standard rigid base made of carbon fibre, created specially to stabilise the patient’s head and neck during IMRT treatments. This supporting base of the patient is always fixed to the Computer Tomography and the Tomotherapy beds.
To maintain the patient’s position standard neck supports were used in different sizes (Large, Small size), connected to the base. In some cases, to improve patient comfort, an additional customised neck support was used (MOLDCARE® CUSHION For Head, Neck and Small Regions Rt -4492S, ALCARE Co., Ltd. Tokyo, Japan). All the patients enrolled were subjected to Virtual Simulation Computed Tomography (GE Medical System HiSPEED NX/I) using the same technical protocol (120 kV, 250 mA, Thickness of acquisition 3 mm, Pitch 3, Thickness of reconstruction 3 mm, Reconstruction interval 3 mm).
Before each treatment, patient positioning was obtained by means of alignment of the lasers of the room with 4 reference points positioned on the “mask” at the time of CT simulation.
In IGRT-IMRT treatments dispensed in Tomotherapy, the MVFBCT images were acquired daily before each treatment and matched with those obtained during the CT simulation. Alignment of the images was obtained by means of automatic software tools and, if necessary, also with manual adjustments [
10‐
12].
We considered the translational mediolateral (ML), craniocaudal (CC), and anteroposterior (AP) deviations; antero-posterior correction of 3.8 mm is applied for the systematic AP setup error caused by couch sag [
13‐
16] that occurs when the patient is moved to the treatment isocenter (inside the bore) from virtual isocenter located 70 cm outside. Rotational corrections like pitch, roll and yaw were not recorded.
Patients
We prospectively collected data for a sample of consecutive patients with head and neck cancer, treated with the IMRT technique by our department from September 2011 to August 2013.
The eligibility criteria were: female and male patients with head and neck cancer who underwent intensity modulated radiation therapy, ≥14 years old and ≤85 years old, without cognitive disfunctions, performance status score of 0–1 [
17] and with written informed consent.
Data collection
For all the patients, the thermoplastic mask and other custom immobilization devices were prepared before CT simulation, on the basis of the same internal operating protocol.
We created an ad-hoc report to record all the environmental and technical variables that might have a correlation with the setup error. On the baseline (BL) day of CT simulation we recorded: anxiety [
18] and pain [
19‐
24], before starting the simulation procedure (more on the measurements scales below); patient’s weight and height [
25,
26]; intent of radiation therapy; variable influencing patient positioning such as shoulders forced down, scoliosis or kyphosis; some patient and environment variables such as presence of mobile dental prosthesis, use of bite for locking of the jaw, tracheostomy, voluminous beard, voluminous hair, room temperature of the CT room and temperature of the water used for shaping the thermoplastic mask; type of standard neck supports used and the eventual addition of customized neck support; number of radiation therapy team members who worked together to shape the thermoplastic mask and temporary removal of the mask before the acquisition of CT simulation. On the first day of every week of treatment we measured the patient’s weight. Finally on each treatment day we recorded: Mediolateral (ML), craniocaudal (CC), and anteroposterior (AP) deviations; radiation therapists executing the procedure; anxiety and pain, before starting the treatment session; changes related to the presence of mobile dental prosthesis, use of dental bite, tracheostomy, beard or voluminous hair, temperature of the treatment room.
The anxiety was measured using the STAI Y form, a test validated worldwide [
27] consisting of two scales: the TRATTO-A scale, which measures the predisposition for anxiety, and the STATE-A scale which evaluates the anxiety state, i.e. the emotional state at a given moment. Only STATE-A was used in the study, which consists of 20 items; the compilation, done directly by the patient, requires an average of 5–7 min. The test was administered to the patient in the waiting room.
Pain was measured using the Numeric Pain Rating Scale, [
19‐
24] a numeric scale that describes the intensity of pain, ranging from 0 (no pain) to 10 (max value of pain); the measurement was performed directly by the patients themselves in the waiting rooms.
Study size
In absence of a-priori hypothesis given the exploratory nature of the study, no formal sample sizing was performed. Nonetheless, a limited measurement period (recruiting: September 2011 to August 2013, data collection: first 6 weeks of treatment for each patient) was defined, in order to ensure general feasibility, homogeneity of the procedures and of the data collected.
Statistical methods
Clinical and demographic data were expressed in terms of frequency and percentage for categorical variables, mean ± standard deviation for symmetric quantitative variables, median + IQR for skewed ones.
Overall mean error (M), systematic SD (∑) and random SD (σ) were calculated, both for full sample and by groups (determined according to site) of patients.
In order to explore the correlation between relevant variable and treatment setup errors, we adopted a single Euclidean distance measure. Setup error (in any of the three directions) can assume positive or negative values; regardless of the direction, the more the error deviates from value 0 the more unfavourable the it is. Therefore, to collapse all the three errors (cranial-caudal, medial-lateral and anterior-posterior) in one single measurement readily available for multivariate analysis, we considered each setup error as a vector and calculated 3-dimension modulus as an overall measure of error severity; accordingly, total error
e
tot
was simply calculated as Euclidean distance, as follows:
$$ {e}_{tot}=\sqrt{e_{ml}^2+{e}_{cc}^2+{e}_{ap}^2} $$
where
e
ml
is medial-lateral (ML),
e
cc
is cranial-caudal (CC) and
e
ap
is anterior-posterior (AP) error (to simplify notation we avoided patient and time subscripting; however
e
tot
has to be considered as overall setup error for patient
i at time
t). The total error, expressed in mm, was log-transformed (in order to mitigate the positive-skewness effects on the residuals normality) and then analysed using a linear mixed effects model (complete-case analysis). This method was choosed to accommodate non-independency of the observations sampled (eg. due to clustering within individuals): the method assumes that setup errors can be explained in terms of both fixed and random effects. Fixed effects represent the effects of factors of intrinsic interest (in our case those that can be associated to setup error) because of being repeatable in other populations (eg number of therapists, mask removal etc.); random effects represent random deviation (from the relation depicted by the fixed effects) associated to factors of no intrinsic interest (not strictly repeatable, eg the single patient) which are nonetheless considered in the estimation process [
28].
Covariates used in the model were suggested by clinical/technical interest and feasibility (some patient characteristic, eg voluminous beard etc., were highly imbalanced and therefore were ignored for the correlation analysis): fixed effect variables were shoulders down, scoliosis/kyphosis and mask removal (No/Yes variables), type of neck support device (categorical variable), number of radiation therapists involved at baseline, pain, anxiety, number of radiation therapists encountered during the treatment weeks and body mass index (continuous variables treated as such also in the statistical model); a random patient intercept complete the model presented. We compared this model with one including some additional variables (fixed effects: pain and anxiety before each treatment session, both hair and temperature at baseline and before each treatment session; random intercept: treatment technician), concluding that these latter were not needed (P = 0.98).
The base group for the model reported is represented by patients without ‘Shoulders down’, ‘Scoliosis/kyphosis’ and ‘Mask removal’ at baseline setup; with ‘Large size standard plus customized neck support device’ (the most common); followed by 2 radiation therapists at baseline for mask and setup (the minimum/standard encountered); with no pain (pain = 0); with mid-point anxiety (anxiety = 50); patient who during the entire treatment cycle, was followed by 12 radiation therapists (minimum) during treatment weeks; with a mean BMI.
In order to ease interpretation of results, coefficients (and their confidence intervals) were transformed using (((exp(coefficient) – 1)*100)-100) and presented using a forestplot-like graph: this leads to an estimate of each factor contribution as percentage variation (compared to an otherwise identical patients without it).
Confidence intervals (considering a 0.95 confidence level) and models comparison were performed with bootstrap methods [
29]; a confidence interval not including the threshold of no differences (0% for transformed coefficients) denoted the contribution of the factor analysed as statistically significant. Statistical analysis was carried out using R 3.3.2.