Can calculation of energy expenditure based on CO₂ measurements replace indirect calorimetry?

This retrospective observational study includes the measurements of indirect calorimetry performed in the mixed medical-surgical adult ICU at the Geneva University Hospital, Switzerland.

We included 278 mechanically ventilated ICU patients from the institutional database of indirect calorimetry, conducted by the Clinical Nutrition Department. We collected data regarding physical characteristics (age, gender, height, anamnestic weight, body mass index), diagnosis and severity at ICU admission (primary diagnosis, Acute Physiology and Chronic Health Evaluation (APACHE) II score [13], Simplified Acute Physiology Score (SAPS) II [14]), and patient observations at the time of the measurement (body temperature, Glasgow coma scale, Sedation-Agitation Scale [15]). We also reported the characteristics of energy provision at the time of the indirect calorimetry, consisting of the route of administration (enteral nutrition (EN), parenteral nutrition (PN), combined (EN + PN), or non-nutritional energy sources), the total administered energy (kcal/d) along with the composition of the energy substrates (carbohydrate, protein, lipids). Energy from glucose-containing solutions and propofol was included in the calculation of the total administered energy.

All measurements were conducted using the Deltatrac Metabolic Monitor® (Datex, Helsinki, Finland) which was calibrated before each measurement according to the manufacturer’s procedure. Calorimetry was not conducted in patients with ventilator settings exceeding the general limits (fraction of inhaled O₂ (FiO₂) < 60%, positive end expiratory pressure (PEEP) < 9cmH₂O), or with contraindicated treatments (e.g. chest drains for pneumothorax, inhaled nitric oxide). Patients were in resting state for at least 1 hour prior to the measurement, i.e. free of physical activity, mobilization, and transport. Measurements were conducted to obtain valid recordings for at least 20 minutes in a single measurement session for each patient after reaching steady state; recordings from the first 5 minutes of measurements and from non-stable states were excluded. As the Deltatrac® was a stand-alone device, minute-by-minute readings of VO₂ (ml/min STPD) and VCO₂ (ml/min STPD) were obtained as printouts or handwritten recordings. The results were reported in the clinical database as the means of valid recordings for the duration of the measurement. Mean VO₂ and mean VCO₂ of each patient was used for the individual calculation the RQ (=VCO₂/VO₂) and the determination of the measured EE by the modified Weir’s equation:

Fraction of inspired O₂ (FiO₂, %), and minute volume of ventilation (L/min BTPS) at the time of the indirect calorimetry were also collected. The day of the measurement was recorded as the number of days after ICU admission.

The EEVCO₂ was calculated using VCO₂ values measured by indirect calorimetry, and two different RQ values: fixed value of 0.85 (EEVCO₂_0.85) and FQ (EEVCO₂_FQ).

Patient characteristics are presented as mean ± standard deviation (SD), median (interquartile range), or number (percentage) as appropriate. Normality of distribution was confirmed based on the skewness and kurtosis analyses prior to parametric analyses. We used paired t tests to compare calculated EEVCO₂ and EE measured by indirect calorimetry, and chi-squared tests for comparisons of proportions. The agreement between the measured EE and EEVCO₂ were analyzed using Bland-Altman plots. The individual accuracy of EEVCO₂ in comparison to measured EE was expressed as 5% accuracy rates defined as the proportion of calculated EEVCO₂ values within the clinically relevant limits, i.e. 5% of the measured EE. The results were also evaluated by 10% accuracy rates, a more general standard of used in previous literature [11].

EEVCO₂ accuracy rates were further tested according to the timing of measurement, i.e. within or after the 7th ICU day, and the rate of energy provision, i.e. <90 or ≥90% of measured EE. These factors were selected to test the hypothesis that patients treated for longer duration and fed closer to the energy targets are metabolically more stable and make better candidates for EEVCO₂ [11].

Patients’ characteristics are shown in (Table 1). Patients with a wide variety of primary diagnoses were analyzed, reflecting patient recruitment in a polyvalent ICU. Ninety-nine percent of the patients were fed with nutrition formulas at the time of the measurement, mainly through the enteral route (90%, including EN + PN). The remaining patients received energy only from non-nutrition energy source, i.e. drug-related glucose or lipid administration.

Mean biases (lower, upper 95% confidence intervals) for EEVCO₂_0.85 and EEVCO₂_FQ were -21 kcal/d (-41, 1) and -48 kcal/d (-67, -28), respectively (Table 2). The limits of agreement in Bland-Altman plots were (+314, -356) and (+272, -367), respectively (Fig. 1). The 5% accuracy rates compared to measured EE were 46.0 and 46.4%, while 10% accuracy rates were 77.7 and 77.3%, respectively (Table 2).

The overall accuracy of the EEVCO₂ calculation method is dependent on the accuracy of the assumed RQ and the VCO₂ measurements. In the present analysis, VCO₂ was measured by the indirect calorimeter, thus leaving RQ as the only determinant of the difference between calculated EEVCO₂ and measured EE. The RQ can be affected by metabolic consequences related to feeding and stress response secondary to critical illness, as well as non-nutrition factors such as acid-base status. In mechanically ventilated subjects, suboptimal ventilation could lead to unstable CO₂ elimination, resulting in erroneous representation of the metabolic status by VCO₂. In such cases, RQ may be more correctly described as respiratory exchange ratio, as RQ refers to the gas exchange ratio of the true metabolic consequence of energy oxidation. Our study demonstrates the practical implication of the RQ variability by investigating its effect on the accuracy of EEVCO₂ calculation.

The measured RQ was indeed variable among patients as observed in the standard deviation (±0.09) of the measured RQ (Table 1). While optimizing the conditions for VCO₂ measurements may decrease the variability of the RQ [11], this level of variability leads to a difference of > 8% of the calculated EE. By fixing the RQ at a certain value, this variability directly results in the level of inaccuracy of the calculated EE when compared with the measured EE. The use of FQ as suggested by Stapel et al. [11] did not help to improve the EEVCO₂ in our analysis. The method is based on the hypothesis that FQ will more accurately predict RQ, which is not often the case in critically ill patients with variable metabolic and feeding status. The standard deviation of FQ was ±0.01, demonstrating the limited possibility of agreement with the measured RQ, which presented a much larger variation. Detailed investigation by McClave et al. concluded that RQ is neither a reliable indicator of the feeding status nor strongly associated with non-nutritional factors such as condition of ventilation and acid-base disturbance [5], suggesting the difficulty of predicting the RQ. Thus, measuring VO₂ by indirect calorimetry remains as the only valid solution to determine the RQ accurately.

Another important factor is that VCO₂ has less impact on the EE compared to VO₂. This phenomenon can be observed in the multiples for VO₂ and VCO₂ in the Weir’s equation (3.941 and 1.11, respectively), giving 3.6 times more weight to the VO₂ value. As a result, VO₂ has a much larger influence on the EE. However, the complexity of O₂ measurements in variable high O₂ concentration ranges precludes the VCO₂ analyses to be accurately conducted with ventilator-derived measurements. Indirect calorimeters are specially designed to solve this issue, by installing precision O₂ analyzers and implementing appropriate calibration procedures.

Mean EEVCO₂ were in better limits of agreement to measured EE than previously reported comparisons to EE calculated by predictive equations [16]. However, individualized analysis revealed the inaccuracy hidden behind the comparison of the means. Previous studies have evaluated the accuracy of the EEVCO₂ method according to 10% accuracy rates to measured EE [11, 12]. In fact, the 10% accuracy rates of EEVCO₂ in our patients (77–78%) were better than in the previous study by Stapel et al. (61%) [11, 12], perhaps due to the use of VCO₂ values measured by indirect calorimetry. In the present study, we implemented 5% accuracy rates according to the clinical relevancy of the results. For example, 5% accuracy for the mean measured EE of the present study (1956 kcal/d) means allowing ±98 kcal/d difference in the calculated EE; while it will be ±196 kcal/d for 10% accuracy. The results can vary within the ranges of 1858–2054 kcal/d for 5% accuracy, and 1760–2152 kcal/d for 10% accuracy. It is irrelevant to consider a method that allows almost 400 kcal/d difference in the calculated results as an alternative to measuring EE by indirect calorimetry. In addition, EEVCO₂ was calculated based on the VCO₂ measured by the indirect calorimeter, meaning that the difference in the results could only arise from the calculation based on assumed RQ. For these reasons, we decided to evaluate the validity of calculated EEVCO₂ as an alternative to measured EE according to 5% accuracy rates.

We anticipated that the accuracy rates for EEVCO₂ would be different when measured before or after the 7th day of ICU admission. This was not the case, and suggests that the metabolic state of critically ill patients is difficult to predict, even after the 7th day of ICU admission when clinicians expect that the initial stress of critical illness starts to resolve [17]. The shift between the acute and subacute (or later) phase of critical illness is generally characterized by the shift from the catabolic to anabolic condition, and complicates the metabolic pathways [18]. Prolonged immobilization and organ support therapies can also alter the metabolism, not to mention the effect of repeated stress due to secondary infections and organs failure [17, 19, 20]. Our data thus suggest that a similar variability and complicated metabolic pathways exists also during the acute phase of critical illness [21].

The accuracy rates of the calculated EEVCO₂ also remained unchanged when the patients were fed less or more than 90% of their measured EE. This observation can partly be explained by the relationship between feeding and RQ, as the accuracy of EEVCO₂ relies on the accuracy of the assumed RQ to the measured RQ. Higher rate of feeding correlates with higher RQ [5], suggesting that RQ is likely to deviate higher from the generally accepted value (i.e. 0.85) with higher rate of energy provision. At the same time, RQ is highly variable and unpredictable in critically ill patients, limiting its validity as an indicator of energy substrate oxidation [5]. Thus, variable energy provision rates can lead to variable differences between the assumed and measured RQ, leading to the inaccuracy of EEVCO₂ calculations. The inaccuracy of EEVCO₂ can result in misleading energy provision targets, enhancing the risk of underfeeding and overfeeding, which are both associated with a worsening of the outcome [22‐24].

A major strength of the study is the large number of patients studied. Our analysis is based on 278 mechanically ventilated patients with various diagnoses, enhancing the generalizability of the results to most ICU patients. However, it should be noted that their length of stay was longer than the mean (<4 days) at our ICU. The selection of mechanically ventilated patients also means that patients had at least one organ failure [17], while the assumption of RQ in the calculation of EEVCO₂ may be better applicable once the patients are stabilized [11].

As this was a retrospective study, the timing of the indirect calorimetry measurements was not controlled. Stratifying patients before and after the 7th ICU day may not have reflected the characteristics of acute and late-phase metabolism of ICU patients. However, we believe that this limitation can also be seen as an advantage as indirect calorimetry is usually recommended when considerable changes are observed or suspected in the patients’ conditions. In this regards, the non-significance of the stratification according to the timing of the measurement (i.e. before or after 7th ICU day) strengthened the clinical relevance of our analysis.

Another limitation arising from the retrospective nature of the study is that the analyzed measurements were conducted for clinical purpose, and not originally intended for research. Clinical conditions during the measurements such as sedation levels, ventilator types and settings, or duration of the measurements were not predefined. These factors could have affected the variability represented by the SD of the RQ, which was higher than in a previous report [11]. The VO₂ and VCO₂ values obtained from the clinical database were already the averaged results of the minute-by-minute readings by the Deltatrac® during the measurements, thus precluding the assessment of the stability of each measurement. We also had no comparison of 24-hour measurements, recently proposed as one of the benefits of the EEVCO₂ method. However, it should be noted that our team members are trained to routinely conduct indirect calorimetry strictly according to our protocol, to ensure the quality of the clinical measurements.

The measurements of VCO₂ in the present study were from indirect calorimeter, and not the measurements by mechanical ventilators as proposed in recent literature [11, 12]. We also did not conduct simultaneous measurements with EEVCO₂ devices and indirect calorimeters. Our Deltatrac® has been used over the years, but has been regularly maintained by the Biomedical Department and has been calibrated before each measurement to assure optimal performance. The accuracy of VCO₂ measurements of the Deltatrac® has been shown to be within 2–4% of expected values in in vitro validations [25], while the level of accuracy for VCO₂ measurements in ventilators can vary as much as ±9%, according to the instructions manuals. Therefore, it is unlikely that the use of VCO₂ measured by mechanical ventilators will significantly improve the accuracy of EEVCO₂ and thus influence the conclusions of our study.