2.1 Patients
The present pilot study provides two approaches: a cross-sectional approach to identify postural-control differences between CIPN patients and matched healthy control subjects and a one-armed longitudinal approach to evaluate the effects of a balance-based exercise intervention on CIPN-related postural deficits.
Therefore, we examined eight cancer patients with different cancer localizations and treatment status, all reporting severe neuropathy symptoms due to chemotherapy (CIPN). The chemotherapies applied entailed the neurotoxic agents bortezomib, carboplatin, cisplatin, paclitaxal, docetaxal and vincristine. None of the patients had any neuropathy symptom before the application of neurotoxic agents. CIPN was clinically and electrophysiologically confirmed in all patients. Moreover, we assessed patients’ subjective CIPN symptoms via the neurotoxicity subscale (NtxS) of FACT&GOG (Functional Assessment of Cancer Therapy/Gynaecology Oncology Group) scored from 0–44 (0= severe symptoms; 44= no symptoms); Table 1 summarizes our patients’ clinical information.
We excluded patients with other possible sources of neuropathy (eg hereditary, diabetes- or alcohol-induced) and patients suffering from additional deficits that might interact with their postural control such as a relevant reduction of muscular strength or certain comorbidities (eg osteolysis, severe vertebral degeneration, vestibular deficits). Specifically, all patients underwent detailed vestibular testing using a rotating chair. In addition, patients performed an incremental stress electrocardiogram on a stationary bicycle in the Institute for Exercise- and Occupational Medicine, Medical Center – University of Freiburg in order to exclude cardiovascular risks during exercise and to determine the lactate threshold for exercise control.
The control group for the postural control experiments consisted of 15 healthy subjects matched to patients’ age, weight and height. We assigned two matches to each patient (except for one patient with a relative heavy body weight) to ensure a more reliable representation of the postural behavior of healthy subjects.
Patients underwent assessments of posture control twice (before and after 12 weeks of a supervised exercise intervention) while healthy control subjects underwent the assessment only once.
Patients' recruitment and data collection took place in the Clinic of Internal Medicine I and posture analyses and clinical assessments took place in the Department of Neurology and Clinical Neurophysiology, Medical Center – University of Freiburg.
The study was approved by the Ethics Commission of University of Freiburg. All subjects provided written informed consent to the experimental procedure in accordance with the Declaration of Helsinki.
Table 1 Subjects' characteristic
APA, action potential amplitude (normal values: ≥6µV); B-NHL, B-Non Hodgkin Lymphoma; f, female; m, male; NCV, nerve conduction velocity (normal values: ≥41m/s); NtxS, neurotoxicity subscale of FACT/GOG (Functional Assessment of Cancer Therapy/Gynaecology Oncology Group) scored from 0 (severe symptoms) – 44 (no symptoms); SD, standard deviation; "-", no reflex; "(-)", reduced reflex; "+", normal reflex.
2.2 Intervention
The one-on-one training sessions took place in the division of Sports Oncology in the Clinic of Internal Medicine I, twice per week over 12 weeks. The intervention protocol included a cardiovascular warm-up of up to 20 minutes on a stationary bicycle with an intensity of 75–80% of maximum heart rate, followed by the balance-based exercises for 30 minutes and muscular endurance training for the main muscle groups. The main focus was on the balance part of the training. Balance training prescription included a progressive increase over the intervention period in the exercise amount and difficulty. Depending on the individual performance level, that could vary during the interventions period, patient performed three (beginners) to eight exercises (more advanced) with three repetitions each à 20–30s (a 20-s rest between the repetitions and a 2-min rest between the different exercises to avoid fatigue). Moreover, exercise difficulty was also adapted to patients’ performance level and successively increased by reducing the support surface (eg bipedal- to mono-pedal stance) and visual input (eyes closed), adding motor/cognitive tasks (eg moving arms or counting backwards) and inducing instability (throwing a ball or being perturbed by the sports therapist) to stimulate the sensorimotor system adequately (46,48). We documented vital parameters, training progress, and reasons for missed sessions.
2.3 Procedure and data analysis
For evaluating postural control, spontaneous sway and perturbed stance were measured with a custom-built motion platform (49,50) under two visual conditions, with eyes open and with eyes closed. Each trial lasted one minute. The participants were told to stand upright on the platform in comfortable shoes. Stance width was predetermined within a marked area. For safety reasons, participants had to hold two ropes hanging from the ceiling in a crossed-arms position so that they could not perceive a somatosensory spatial orientation signal (Figure 1A).
Data analysis was conducted off-line with custom-made software programmed in MATLAB® (The MathWorks Inc., Natick, MA, USA).
Spontaneous sway was measured on the non-moving platform. The center of pressure (COP) sway path was detected with a force transducing platform (Figure 1B-D, Kistler platform type 9286, Winterthur, Switzerland). From the COP excursions over time in anterior-posterior and medio-lateral sway directions, we calculated the root mean square (RMS) around the mean COP position. After differentiating the time series, we calculated mean velocity (MV). In addition, center frequency (CF) was extracted from the power spectrum (51,52).
Perturbed stance was measured on the moving platform to differentiate sensory contributions in reaction to external disturbances. We analyzed rotational tilts in the sagittal plane with the tilt axis passing through the participant’s ankle joints. Platform rotations were designed as pseudorandom stimuli (PRTS, pseudorandom ternary sequence, see Figure 1E) (53). This stimulus has a wide spectral bandwidth with the velocity waveform having spectral and statistical properties approximating a white noise stimulus (53). As such, this stimulus appeared to be unpredictable to the test subject. We applied two peak angular displacements (stimulus amplitude: 0.5° and 1° peak-to-peak) and analyzed at eleven stimulus frequencies (0.05, 0.15, 0.3, 0.4, 0.55, 0.7, 0.9, 1.1, 1.35, 1.75 and 2.2 Hz).
Angular excursions of the lower (hip-to-ankle: hip movement) and upper (shoulder-to-hip: shoulder movement) body segments and the platform in space were measured using an optoelectronic motion-measuring device with markers attached to shoulder and hip (Optotrak 3020, Waterloo, Canada). Each marker consisted of three light-emitting diodes (LED) fixed to a rigid triangle. The triangles were fixed to the participant’s hips and shoulders and to a rigid bar on the platform (Figure 1A). 3-D LED positions of the triangles were used to calculate marker positions (Figure 1F, G). Optotrak® and Kistler® output signals as well as the stimulus signals were sampled at 100 Hz using an analogue-digital converter. We recorded all data with software programmed in LabView® (National Instruments, Austin, Texas, USA).
To analyse postural reactions in relation to platform stimuli, transfer functions from stimulus-response data were calculated via a discrete Fourier transform. Fourier coefficients of stimulus and response time series are used to determine GAIN and PHASE with respect to stimulus frequencies. GAIN represents the size of the postural reaction as a function of stimulus size (platform angle), while PHASE is related to the relative timing between postural reaction and stimulus (54).
Furthermore, we calculated COHERENCE, a measure of reproducibility of the response. Technically, COHERENCE is calculated as the quotient between the cross power spectrum of stimulus and response, and the product of the individual spectra of stimulus and response (53). Whereas a COHERENCE value of 0 indicates that there is no linear correlation between the stimulus and response, and 1 indicating a perfect linear correlation with no noise. Values less than 1 occur in practice either because there is noise in the system or there is a nonlinear relation between stimulus and response.
[insert Figure 1.]
2.4 Parameter identification
Transfer functions served as the experimental data basis for model simulations using a specific version of an established postural control model (36,49,53,55–57) with active time-delayed proportional, derivative, and integral feedback as well as passive stiffness and damping to extract basic constituents of postural control. The physical part of the model is a single inverted pendulum model with corrective torque applied at the ankle joint. The model used here includes a negative feedback loop that relates body excursion detected by visual, vestibular, and proprioceptive sensors to a corrective torque via a neural controller. The neural controller represents the relation between sensory error, ie the difference between the current and desired position on the one hand, and the strength of the motor output, ie torque, on the other hand. With the help of an automated optimization tool (fmincon, MATLAB®, The MathWorks Inc.), which minimized the difference between experimental and simulated GAIN and PHASE curves, we estimated the neural controller’s parameters with proportional (Kp), derivative (Kd) and integral (Ki) contributions (PDI-controller). Neural controller gains are, in part, determined by mass and height of each subject’s center of mass.(53)Peterka, 2002) Because our control group presented lower masses and heights than patients, we had to correct neural controller gains for this effect. That is why we provide numbers for (Kp/mgh), (Kd/mgh), and (Ki/mgh), where (mgh) represents the gravitational pull (mass) x (gravitational constant) x (height of center of mass). Moreover, we derived time delay (Td), proprioceptive sensory weight (Wp), and biomechanical elasticity (Ppas) and damping (Dpas) of the muscles and tendons. We fitted model simulations to experimental transfer functions under different stimulus amplitudes and visual conditions.
2.5 Statistics
Statistical analyses were performed using Microsoft Excel, JMP® and Statview (SAS Institute Inc., Cary, NC, USA). We applied parametric methods after testing the normal distribution and homogeneity of variances with the Kolmogorov-Smirnov test. Due to the expected dependency between experimental conditions and outcome measures, statistical significance was tested by an analysis of variance (ANOVA) for the comparison of healthy subjects and patients. Visual condition, sway direction, and body segment (hip, shoulder) were the within-subjects’ factors for spontaneous sway. For perturbed stance, we applied visual condition, stimulus amplitude, stimulus frequency, and body segment (hip, shoulder) as within-subjects’ factors. For the analysis of the balance based exercise intervention effect on patients, we used a multivariate analysis of variance (MANOVA) with a time as the repeated measure variable, in addition. The level of statistical significance was set at p=0.05.