1 Introduction

Critically ill patients have a high risk of developing malnutrition [1, 2]. Their energy consumption varies greatly depending on the severity of the underlying condition and several other factors such as fever, concurrent infection, pain or surgical procedures [3]. Resting energy expenditure in patients with sepsis, burns, head trauma, mechanical ventilation or after surgery is increased of up to 50 % of normal energy requirements and omission of adjustments in substrate supply may lead to rapid malnutrition in these patients [4].

Malnutrition is associated with multiple complications as it results in impaired immune function, poor wound healing, decreased renal function, hospital-acquired infections, reduced respiratory capacity and consequently a longer ventilation time [5]. These factors contribute to increased morbidity and mortality, a prolonged hospital stay and increased costs [6–10]. Early overfeeding can also induce complications, such as hyperlipidemia, liver dysfunction, hyperglycemia and immunosuppression [5, 11–13].

Therefore both, hypocaloric and hypercaloric nutrition should be avoided [14]. Unfortunately, data regarding the adjustment of daily energy delivery to the patient’s needs are scarce [4, 15]. There is evidence that energy requirements are dynamic during ICU stay. Energy needs were first increasing and later decreasing in sepsis and trauma and both sepsis and trauma patients [16]. In order to reduce malnutrition methods are required that allow an optimal nutrient supply adjusted to different stages of disease [1, 4]. Indirect calorimetry (IC) and the double-labelled-water method are the gold standard for determining energy requirements. Technically indirect calorimetry is based on measurements of respiratory gases, i.e. consumption of oxygen (VO2) and production of carbon dioxide (VCO2). By using these parameters heat, which is generated during oxygenation of (carbon-based) substrates and an equivalent to human energy, can be assessed. Indirect calorimetry is suitable in both spontaneously breathing and mechanically ventilated patients [17]. Moreover by determination of the in- and expiratory O2 concentrations, and the CO2 concentration the respiratory quotient (RQ) can be calculated based on the Haldane transformation. However, these methods are not always available, time-consuming, expensive and often cumbersome in routine practice. VCO2-based calorimetry is an alternative non-invasive method for the calculation of energy expenditure. Normally VCO2 is measured continuously by most mechanical ventilators and is the starting point for energy measurement using a modified Weir formula [18, 19].

In recent years technological advancements contributed to the development of portable, easy to handle electronic motion detectors. One of them is the SenseWear® Pro Armband (SMT medical GmbH&Co. KG, Würzburg, Germany), a multidimensional device, which detects different physiological parameters of the human body by dedicated sensors. These include a continuous registration of skin temperature and galvanic skin resistance of heat flow. In addition a 2-axis accelerometer is localized in the armband, which records movements of the patient [20]. The SenseWear® Pro Armband is placed on the triceps of the right upper arm (Fig. 1). The upper arm allows for the best skin contact of the device and meets the requirements of the individual sensors. Once skin contact is achieved the armband automatically starts the measurement and stops with loss of skin contact [21].

Fig. 1
figure 1

Correct placement of SenseWear® Pro Armband on the triceps muscle of the right upper arm

Full size image

Three different versions of the armband are commercially available: the SenseWear Pro Armband (sold from 2001 to 2003), the SenseWear Pro2 and SenseWear Pro3 Armband, for which a detailed assessment algorithm has not been published [20]. Due to its easy manageability and its capability for continuous data acquisition it was proposed to be suitable to determine resting energy expenditure in intensive care patients as well. SWA has shown promising results for calculating energy expenditure in non-critically ill patients [22–24] and various diseases such as obesity [25], cancer [26], COPD [27], cystic fibrosis [28], rheumatoid arthritis [29] or renal insufficiency [30]. Acceptable levels of agreement were also found in older individuals [31, 32]. Data for ICU patients are sparse and only one small study investigated patients under mechanical ventilation [33]. In this study we aimed to clarify the role of SenseWear® armband in critically ill patients on intensive care units and investigated whether SWA would be a suitable alternative for REE assessment. Therefore SenseWear Pro 3 Armband was compared to indirect calorimetry, the gold standard for REE measurement. Secondly we investigated the usability of predictive equations for REE by comparing to indirect calorimetry and SenseWear Armband.

2 Methods

34 patients from two intensive care units of the University Medicine Greifswald were recruited over a period of 7 month including both surgical and medical patients. Eligible were white Caucasians fulfilling the following criteria: (1) age >18 years, (2) ICU patient (3) and mechanical ventilation of at least 24 h. Patients without mechanical ventilation, pregnant women, patients under haemodialysis, patients with MRSA infection or FiO2 > 60 % were excluded. Severity of disease was estimated using the Simplified Acute Physiology Score (SAPS II) [34]. The investigation was approved by the ethics committee of the University of Greifswald and informed consent was obtained by proxy (BB 14/10).

2.1 Anthropometry, vital signs and nutrition status

The patients were studied by anthropometry (height, weight) and BMI (kilograms per height in meters squared). Heart rate, respiratory rate and body temperature were recorded as well. All patients received enteral, parenteral or combined enteral and parenteral nutrition. The difference between estimated REE by IC and effective daily energy intake was calculated (=energy balance). Subgroups were formed according to the energy balance (>−200, −200 to +200 and >+200 kcal/day).

2.2 Indirect calorimetry (IC)

REE was measured using the Deltatrac® TM U, MBM-200 system (Datex Ohmeda, Bremen, Germany). Before each measurement CO2 and O2 analyzers were calibrated with a standard gas mixture. The Deltatrac® device was connected to the ventilator in mechanically ventilated patients to calculate the energy expenditure and the measurements were performed for a minimum of 30 min in resting conditions. The first 5–10 min were discarded to achieve steady state (VO2 and VCO2 variation of ≤10 %/min). The ratio between volume of carbon dioxide and volume of oxygen is called respiratory quotient (RQ). Normal range of RQ was between 0.7 and 1.0.

The volumes of oxygen consumption and carbon dioxide production were determined continuously (both in l/min). By using the modified Weir formula the metabolic rate was estimated as follows:

REE (kcal/day) = [3.941 × VO2 (ml/min) + 1.1xVCO2 (ml/min)] × 1.440 [35].

2.3 SenseWear Pro 3 Armband (SWA)

The SWA was applied to the upper arm 10 min before indirect calorimetry started. This was done for each measurement by the same investigator. Over a 24 h period, patients wore the SWA continuously. During this period the nursing staff recorded factors that could affect energy consumption such as physical examination, washing, diagnostic work up and vegetative parameters such as sweating, increased or decreased body temperature. After 24 h the armband was removed and the data were evaluated with the SenseWear® software (lnnerview Professional Software 6.1 BodyMedia Inc., Pittsburgh, PA, USA). Energy consumption was calculated considering “age”, “size”, “weight”, “BMI” and “gender” according to the manufacturer’s protocol. Exact time for IC and measurement by SWA was recorded to rule out a detection bias.

2.4 Predictive equations

Many equations have been developed for estimating energy consumption in critically ill patients. There are two categories of equations: first, those using static variables like height, weight, age and sex and those that use dynamic variables which varies like body temperature, respiratory rate and tidal volume [36]. The equations that have been used in the present study are shown in Supplementary Table 1.

2.5 Statistical analysis

Data were analyzed using SPSS (Inc., Chicago, II, USA). Means of two methods were compared using the paired t test. Relationship between two parameters was performed by a simple Pearson correlation. Mann–Whitney U-test was used to test significant differences between mean values for two independent samples. For comparison of three groups and more, the Kruskal–Wallis test was used. Residual plots according to Bland and Altman were performed to compare two methods. A deviation of 10 % was determined to be acceptable (accuracy in  %).

3 Results

3.1 Study population

34 patients were studied; among them 20 were medical and 14 surgical patients. Baseline characteristics of the participants are listed in Table 1. The mean age was 67 ± 12.4 years. Pulmonary disorders were the most frequent causes for ICU admission (29.4 %), followed by cardiovascular diseases and post-surgical surveillance (20.6 %). Disease severity was objectified using the SAPS II score that provides an estimation of the risk of death independently of a specific primary diagnosis [34]. Both groups showed similar mean SAPS II of around 32–34. The mean BMI was 27.6 ± 5.7 kg/m2. 44.1 % of patients had a normal weight and 55.9 % were overweight or obese. Mean daily energy intake was 1681 ± 333 kcal/day.

Table 1 Patient characteristics of the surgical and medical intensive care unit
Full size table

3.2 Correlation between estimated REE by IC and estimated REE by SWA

Energy consumption measured by IC was 1805.9 ± 413.6 and 1552.3 ± 284.9 kcal/day when measured by SWA. Results of IC and SWA showed a linear correlation (r = 0.598, p < 0.001, Fig. 2a) but also demonstrate that SWA underestimates REE compared to the gold standard. In addition root means squared error (RMSE) was 35.8 %, indicating still a poor quality of prediction. In order to evaluate measurement errors by SWA we performed a Bland–Altman analysis. Bias (ΔREE) was −253.6 ± 333.2 kcal/day, equivalent to −11.7 % (Fig. 2b). Bland–Altman plot also indicated that REE differences between indirect calorimetry and SenseWear armband are dependent on the amount of the patient’s energy intake: the higher the energy consumption, the larger differences were found between these methods.

Fig. 2
figure 2

a Correlation between estimated REE by SenseWear Armband (REESWA) and measured REE by indirect calorimetry (REElC). b Bland–Altman bias plot for REE measurement by SWA and IC (solid horizontal line mean difference between methods, dashed lines 2 standard deviations)

Full size image

3.3 Effects on measurement error

We were interested whether specific subgroups might predispose for a higher measurement error. Neither gender nor patient group (medical or surgical) showed any influence on REE measurement bias (Table 2). In females there was a trend towards higher deviations. Also, there weren’t any significant differences between individuals who underwent energy consuming interventions (physical therapy or diagnostic procedures) or those who were without interventions. The results indicate that energy balance appears to affect REE measurements. In the presence of a higher energy balances (the patient effectively received more calories than determined by indirect calorimetry) the differences between IC and SWA were greater. No significant differences were found in patients with an energy balance of less than −200 kcal (Supplementary Table 2).

Table 2 Estimated REE by IC and SWA according to sex, surgical or medical group and presence of physical examination or diagnostic procedures
Full size table

When patients were analyzed according to their vital signs significant differences of REE measurements were observed irrespective of heartbeat and respiration rate. The discrepancy was greatest in tachycardic (>100 beats/min) and tachypnoeic patients (>30 breaths/min). Even in normofrequent patients (heart rate 70–99 beats/min) who account for the majority of the study group, bias was still around 15 % (Table 3). A significant discrepancy was also seen in patients with fever (>38 °C). In non-febrile ICU patients REE measurements by SWA and IC showed comparable results, with the highest concordance in individuals with normal body temperature (<37.4 °C).

Table 3 Estimated REE by IC and SWA stratified by vital signs
Full size table

3.4 Comparison of estimated REE by indirect calorimetry with predictive equations

We compared energy consumption determined by IC or SWA with frequently used estimation formulas. Except for the formula of Ireton Jones (1992), all equations underestimated the energy consumption compared to indirect calorimetry (Supplementary Table 3). The equation of Mifflin St Jeor showed the worst correlation with an average bias of −324 kcal/24 h (−15.5 %). Significant differences were also detected when REE was assessed by Harris–Benedict, Müller, Schofield, Ireton Jones (1992) and Swinamer. There was a good correlation between indirect calorimetry and estimated REE by Penn State, Ireton Jones (1997) and Fusco formulas.

Comparable results between REE values from SWA and predictive equations were only found for Harris–Benedict (ΔREE = +27.3 kcal/day; +2.7 %) and Schofield (ΔREE = −16.8 kcal/day; −1.1 %) formula. All other equations showed significantly different results. Greatest differences were recorded when using the Ireton Jones equation (1992) with an average deviation of −412 kcal/day (−27.3 %) (Supplementary Table 4).

Results of REE measurements by indirect calorimetry, SenseWear Armband and predictive equations are summed up in Table 4. A deviation of 10 % from REE by IC was considered to be acceptable. Formulas of Penn State and Fusco showed a relative good correlation to indirect calorimetry in 41.2 and 47.1 % of patients respectively.

Table 4 Comparison of REE Measurement by IC with SWA and with predictive equations
Full size table

4 Discussion

This study was designed to compare three REE assessment tools, indirect calorimetry, SenseWear® armband and predictive equations for their usability in ventilated intensive care patients. This group of patients is critically ill, needs close monitoring of vital functions and metabolism, and both over- and underfeeding should be avoided. Up till now indirect calorimetry is the gold standard for the estimation of energy expenditure but it is time consuming and cumbersome. Only one single study exists that compared IC and SWA in intensive care patients under mechanical ventilation for at least 24 h [33]. That cohort consisted of 23 patients and there was an underestimation of 6.8 % for daily energy expenditure when using SWA with a mean bias of 135.0 ± 417.2 kcal/day. In our study we found an even greater discrepancy of 11.7 % (−253 kcal/day), but a better correlation between the two methods (r = 0.598), possibly due to the higher number of patients included.

Measurement of REE by SenseWear® armband is a reliable method in healthy individuals. Malavolti and collaborators investigated normal-weight healthy subjects and found no systematic measurement error between SWA and IC [37]. The correlation factor was high with r = 0.86 (p < 0.0001) in this group. Similar results were obtained by Fruin and Rankin who additionally found that deviation of SWA from indirect calorimetry varies depending on the intensity of physical activity. While SenseWear Armband provides valid results for energy expenditure in resting subjects it overestimated flat walking activity by 13–27 % but underestimated inclined walking energy expenditure (by 22 %) [22]. Effects of exercise intensity on energy expenditure assessment by SWA and IC were further evaluated in cystic fibrosis patients compared to healthy adults who underwent treadmill walking [28]. There was an underestimation of energy consumption by SWA individuals with high physical activity but an overestimation at very low physical activity.

Our study demonstrated that SWA measurements in ICU patients generally underestimates REE irrespective of physical or diagnostic examinations (Table 2) or physiotherapy (data not shown). However it should be noted that any comparison of energy expenditures between healthy and physically active individuals and ICU patients, who are mostly confined to bed, is rather difficult. Quantification of physical activity in intensive care patients under mechanical ventilation is even more challenging. While SWA has been established as a valid tool for healthy and normal weight persons its usability has limitations for obese individuals: while SWA underestimates REE in obese people at rest it highly overestimates energy expenditure under physical exercise [25]. In cancer patients who are often confronted with nutritional problems and metabolic derangements the practicability of SWA compared to IC was also investigated. In a small study an acceptable accuracy for SWA-based energy expenditure measurement was found in patients with acute myelogenous leukemia undergoing induction chemotherapy. Bland–Altman analysis revealed no significant differences and correlation was high (r = 0.84, p < 0.0001) [26]. Further validation studies on larger cohorts and different tumor types will be necessary for a more conclusive evaluation of the usefulness of SWA. To the best of our knowledge this study is the first one that investigates measurement errors between SWA and IC in various subgroups of ICU patients. Differences were noted irrespective of sex, medical discipline (internal medicine, surgery) and deviations tended to be higher in females and medical patients. Moreover, our study considers other potential confounders such as heart and respiratory rate and body temperature.

4.1 Predictive equations

Caloric requirements for intensive care patients are often assessed by predictive equations. We compared REE calculations by a set of predictive formulas with both indirect calorimetry and SenseWear Armband. Only a few predictive equations are useable for critically ill and ventilated ICU patients and most consistent results were obtained with equations of Ireton-Jones (1997), Fusco and Penn State. Our observations are partly in line with the results from Frankenfield and co-workers who found an accurate assessment of the metabolic rate in critically ill patients with the Penn State equation, whereas the equation by Ireton-Jones turned out to be less accurate [36]. Predictive equations by Müller and Harris–Benedict have been derived from healthy subjects and may therefore be less reliable in critically ill and ventilated patients. Notably our study showed an underestimation of energy consumption by most equations implying a risk for underfeeding and malnutrition. Prediction accuracy of formulas has been reported with an error range of 7–55 % [38]. In addition, ICU patients may fluctuate in energy needs, which restrict the usability of predictive equations. In a recently published study REE estimated by Harris–Benedict equation (HBE) was compared to SWA measurements in postoperative patients with a good agreement between HBE and SWA (mean bias: −0.569 ± 0.378 kcal/24 h, underestimation of −2.9 %) [39]. When we compared REE results by SWA with predictive equations there was a strong correlation between SWA and the formulas of Harris–Benedict and Schofield. Since SWA underestimates energy needs when compared to IC, results from predictive equations were more frequently higher than SWA results. As recently published calculation of energy expenditure on only VCO2-based methods showed accurate results and were significantly better than predictive equations [19]. Using a fixed respiratory quotient of 0.81–0.89 energy expenditure was measured by the modified Weir equation that was between 1830 and 1973 kcal/day. These results were quite close to measured REE by indirect calorimetry (1805.9 ± 413.6 kcal/day).

There are limitations of our study. Since we included patients from medical and a surgical intensive care wards, the patient cohort was more heterogeneous than in other studies and a subgroup analysis more challenging. However, our results show that the measurement bias was quite similar the surgical and medical patients. This study was designed as a feasibility and pilot study for the applicability of a new device in intensive care units since no prior data were available to base a power calculation on. We intended to evaluate a measuring tool that is practicable and inexpensive and thus could be useful for primary care hospitals that usually are not equipped with indirect calorimetry and do not have separate intensive care wards for surgical or medical patients. For a definitive assessment of the role of SWA in daily practice on ICUs a confirmatory investigation would be warranted, which can base its predictions and algorithms on our data. Application of one easy-to-use method would be more convenient for routine clinical practice.

5 Conclusions

SenseWear Pro 3 Armband is not an appropriate alternative to indirect calorimetry in intensive care patients under mechanical ventilation because it underestimates REE. Individual measurements should be interpreted with caution. There is a necessity of an improved algorithm for REE assessment based on the data obtained in this study, which will ultimately allow a more accurate measurement of energy expenditure in critically ill patients using the device.