Prior to the initiation of this study we received approval by the institutional ethics review board. Data collection was performed on the first six consecutive patients undertaking a KD during RT/RCT in our clinic. In most cases patients already had initiated or wished to initiate the diet at the initial consultation with the treating radiation oncologist. In one case (patient 2) we recommended the diet during RCT in an attempt to stop continuing weight loss. All patients were told, however, of the limited clinical experience with this diet in the context of cancer treatment and informed about potential side effects such as headaches, light-headedness, constipation/diarrhea and weight loss. Exclusion criteria were defined a priori as Karnofsky index less than 80, type I diabetes, pregnancy, pace maker, cognitive impairments or rare metabolic diseases that would contradict a KD such as defects in key enzymes of ketogenesis, ketolysis, gluconeogenesis or fatty acid oxidation. At study entry, all patients gave written informed consent to collect and publish their data. In particular, all patients agreed that their data could be used for scientific purposes such as publication in a scientific journal. Besides the KD and QoL assessment none of the interventions were unusual for a patient undergoing RCT in our clinic. All these procedures were performed according to the ethical standards of the declaration of Helsinki and approved by an institutional review board.
Body composition
Body weight and composition were assessed at study entry and in regular intervals thereafter (the aim was once per week) through bioimpedance analysis (BIA) on a seca 515/514 medical body composition analyzer (mBCA; seca Deutschland, Hamburg, Germany). The mBCA device has a standing platform with an integrated scale and a handrail system, with one pair of electrodes for each hand and foot (eight-electrode BIA). A current of 100 μA and a total of 19 frequencies between 1 and 1000 kHz are used to obtain impedance measurements. Instrumental precision is specified as 100 g for BW between 50 and 200 kg and 5 Ω for impedance of the left and right half of the body. Fat free mass (FFM), extracellular water (ECW) and total body water (TBW) are derived from predictive equations that have been validated against dual-energy X-ray absorptiometry, air-displacement plethysmography and D2O and NaBr dilution techniques, yielding root mean square errors of 1.9 kg for FFM, 0.8 l for ECW and 1.3 kg for TBW [12].
In order to minimize systematic errors, patients were always measured fasted and without having drunk for at least 10 h. Patients were also asked to void their bladder prior to each measurement. Following the manufacturer’s instructions, body height was measured to within the closest 5 mm with a seca 231 stadiometer before the patient stepped onto the mBCA platform and placed feet and hands on the corresponding electrodes. Besides body weight (BW), the mean electrical resistance (R) and reactance (XC) between left and right half of the body, and the quantities FFM, ECW and TBW derived therewith, the following parameters were recorded for each patient: fat mass (FM = BW − FFM), intracellular water (ICW = TBW − ECW), hydration (= ECW/ICW) and phase angle at 50 kHz (PA = arctan (XC/R)). The PA is of special interest for oncological and other critically ill patients, as small PA values have been shown to predict a poor prognosis [13, 14].
Blood parameters
In all patients except patient 5, blood draws were performed at study entry, once during RT and in the week of treatment termination. Parameters of interest included a complete blood count as well as HDL and LDL cholesterol, BHB, insulin, insulin-like growth factor 1 (IGF1) and thyroid stimulating hormone (TSH). In patient 5 blood test results were requested from the hospital where the patient underwent chemotherapy. Starting with the fourth patient, we also used occasional finger-pricktests with a FreeStyle Precision device (Abbott Diabetes Care Ltd., Range Road, Witney, UK) in order to determine BHB and blood glucose concentrations.
Ketogenic diets
Due to the limitations imposed by the ambulatory setting in our clinic, all KDs were self-administered with counselling at least once per week. Patients received short handouts providing basic information on KDs such as main foods to eat and avoid as well as sample menus. For more detailed information we also lent and recommended to buy a popular book on KDs for cancer patients [15]. In general we recommended to restrict CHO intake to <50 g/day and emphasized the intake of olive oil, coconut oil, butter, ghee, and fatty fish, cheese, meat and non-starchy vegetables. To monitor ketosis, patients were provided with urine strips (Ketostix®, Bayer Consumer Care AG, Basel, Switzerland) and a diary for protocolling their measurements, preferably done in the afternoon. Patients were also asked to record the quantity of all foods eaten for at least 2 days in order to determine their actual energy and macronutrient intake.
Questionnaires
QoL was assessed using the EORTC QLC-C30 questionnaire version 3.0 together with its disease-specific modules [16]. At termination of RT, each patient received a short non-validated questionnaire addressing several aspects of their subjective feeling with regard to the KD.
Statistical analysis
For sake of simplicity we assume a linear relationship between body composition changes and time that we model using Bayesian linear regression [17]. If we have N measurements \(y_{i} \;\left( {i = 1, \ldots ,N} \right)\) made at time points t
i
then we assume
$$y_{i} \sim N\left( {\mu_{i} ,\tau^{ - 1} } \right), \quad \mu_{i} = \alpha + \beta t_{i} ,$$
where α and β are the intercept and slope of the linear relation, modelled with a flat prior distribution, and τ is the inverse of the residual variance σ2. Since we do not know σ exactly, we model it as a random variable drawn from a uniform prior distribution in the range [0.01, 10] times the unit of the parameter under consideration. Changes in the parameters of interest were considered significant if the 95 % highest posterior density interval for the slope β excluded 0. In the figures two-sided p values are also reported in order to put the results into a frequentist framework. Statistical analysis was performed using R version 3.0.2 and OpenBUGS version 3.2.2.