Energy expenditure measurement in critical care: Implications for personalized nutrition support

Abstract

BACKGROUND

Accurate measurement of energy expenditure (EE) is critical for optimizing nutritional support in critically ill patients. Indirect calorimetry (IC) is the reference method used, but its availability at the bedside is limited. As a result, numerous predictive equations have been devised to estimate EE in critically ill patients, along with other more novel methods recently proposed.

AIM

To evaluate current methods of measuring EE in critical care, focusing on practical challenges, accuracy, feasibility, and limitations. We will also discuss how these methods contribute to improving nutrition support strategies for intensive care unit patients for a more personalised and effective solution.

METHODS

A comprehensive search was conducted in PubMed and EMBASE for studies published from December 2014 to December 2024. Eligible studies compared EE measurement methods in critically ill populations. Data extraction and quality assessment followed PRISMA guidelines. Adherence to reporting standards was assessed using the TRIPOD questionnaire and risk of bias was evaluated using the PROBAST tool.

RESULTS

Twenty five original studies met the inclusion criteria and were analysed.

CONCLUSION

Each method has unique strengths and limitations. We found that while IC remains the reference standard, less accurate predictive equations have greater accessibility and ease of implementation. Emerging technologies show promise for bedside applicability. Future research should address practical barriers and validate newer approaches.

Key Words: Energy expenditure; Nutrition in critical care; Indirect calorimetry; Predictive equations; Emerging technologies; Feeding in critical care

Core Tip: This systematic review evaluates current methods of measuring energy expenditure in critical care, focusing on practical challenges, accuracy, feasibility, and limitations of each one. We find that indirect calorimetry (IC) remains the gold standard with the most accurate measurements, but there remains significant difficulty in widespread use. Predictive equations are more accessible but lack accuracy. However, there are new ways of using IC and predictive equations that bring promise, and new alternative methods also show potential for application in the clinical context.

INTRODUCTION

Total energy expenditure (EE) is defined as the total amount of energy humans need to function, and consists of resting EE (REE) and activity related energy[1]. REE is often taken to be closely reflective of total EE of intensive care unit (ICU) patients due to minimal physical exertion[2]. Assessing EE has several uses in the ICU setting. Firstly, it is most often used to determine patients’ caloric requirements[3], the accurate assessment of which is vital to avoid harmful effects of under- or overfeeding[4]. Secondly, it can be used to estimate substrate-specific oxidation of proteins, carbohydrates and fats[5] to help find the optimal amount of macronutrients needed for feeding, thereby helping to guide nutritional support and improve outcomes in the ICU. Thirdly, it can help to evaluate the metabolic impact of specific treatments, such as assessing the metabolic effect of targeted temperature management in post cardiac arrest patients and how it may be associated with neurological outcome[6].

EE during critical illness can be influenced by various factors, such as the severity of the illness, the individual's pre-existing health conditions and the impact of pharmacological treatments[2,7]. It is also dependent on which metabolic phase of critical illness the patient is in at the time of measurement[1]. In reality, obtaining consistent accurate measurements of EE is not easy to achieve as each method has its potential drawbacks that limit useability and reliability in the ICU setting.

Indirect calorimetry

Calorimetry is the measurement of energy production from metabolism of macronutrients[8,9]. Indirect calorimetry (IC) is considered the reference standard for assessing EE in the clinical setting and is recommended for accurate measurement resting EE in critically ill patients[10,11]. It does this through measuring oxygen consumption (VO₂) and CO₂ production (VCO₂) and calculating the respiratory quotient (RQ) (the CO₂ production to O₂ consumption ratio)[12,13] which reflects energy metabolism[9]. The theory will be discussed in greater detail in the following sections. IC is generally preferred to direct calorimetry (DC) which measures actual heat loss as DC is expensive, requires more extensive equipment and technical skill, and is not practical for clinical implementation[9,14].

Predictive equations

Some of the most commonly used predictive equations in setting of ICU are the Harris-Benedict equation, American College of Chest Physicians weight-based method, Mifflin-St Jeor equation, Ireton-Jones equations, Faisy-Fagon equation, and the Penn State equation[15-19]. These can be found in Table 1.

Table 1 Predictive equations.

Equation	Calculation (kcal/day)
ACCP[17]	BMI (weight in kg/height in meters squared) < 25: Actual body weight in kg × 25
	BMI ≥ 25: Ideal body weight in kg × 25
Harris-Benedict[16]	Male: 66.473 + [13.7516 × weight (kg)] + [5.003 × height (cm)] – [6.755 × age (y)]
	Female: 655.0955 + [9.5634 × weight (kg)] + [1.8496 × height (cm)] – [4.6756 × age (y)]
Mifflin-St Jeor[18]	Male: [10 × weight (kg)] + [6.25 × height (cm)] − [5 × age (y)] + 5
	Female: [10 × weight (kg)] + [6.25 × height (cm)] − [5 × age (y)] − 161
Penn State[20]	2003a: 0.85 (Harris-Benedict equation) + (175 × Tmax) + (33 × Ve) – 6433
	2003b: 0.96 (Mifflin-St Jeor) + 31 (Ve) + 167 (Tmax) – 6212
Ireton-Jones[19]	Mechanically ventilated: 1784 – [11 × age (y)] + [5 × weight (kg) + (244, if male) + (239, if trauma present) + (840, if burn present)]
	Spontaneously breathing: 629 – [11 × age (y)] + [25 × weight (kg)] − 609 (when BMI > 27)
Faisy-Fagon[15]	[8 × bodyweight (kg)] + [14 × height (cm)] + [32 × minute ventilation (l/min)] + [94 × body temperature (°C)] – 4834

ACCP: American College of Chest Physicians; BMI: Body mass index; T_max: Maximum body temperature in past 24 hours (°C); Ve: Volume of exhaled gas (L/min).

Apart from IC and predictive equations, several other methods have been proposed as alternatives. One such example would be the calculation of EE based on CO₂ measurements from mechanical ventilators or the pulmonary arterial catheter. This has been proposed as a surrogate to IC, and is recommended by existing guidelines in the absence of IC as they have been shown to be superior in accuracy to predictive equations[20].

Rationale for study

Over- and underfeeding patients in the ICU have consistently been shown to be associated with poor clinical outcomes[21-24], and thus accurate measurement of EE is vital in optimising nutritional support and preventing adverse complications[25,26]. Nutritional guidelines have recommended an initial caloric dosing of 25 to 30 kcal/kg/day before advancing towards energy targets as determined by measurement of EE over next 48 to 72 hours[27].

Despite recent advances in technical developments of indirect calorimeters, bedside application remains limited by logistical challenges and availability[28], and variable performance in different patient populations. Patient factors and ongoing treatments such as agitation, fever, use of vasoactive agents, and organ support therapies such as renal replacement therapy can render REE assessment by IC inaccurate[9].

Predictive equations on the other hand tend to be unreliable and lack universal applicability, as they were formulated based on specific patient populations[3]. Significant discrepancies have also been found when predictive equations were used in isolation when compared with results taken from IC measurements[29]. Nutritional guidelines like those from ESPEN and ASPEN similarly discourage the use of predictive equations due to its inaccuracy leading to over- or underestimation of energy requirements[20]. Despite this, the INTENT trial for tailored nutrition interventions in the ICU utilizes weight-based EE calculations[30], highlighting a preference even in research trials for predictive equations. Additionally, it remains debatable whether predictive equations compromise accuracy in favour of convenience, particularly for patients at the extreme ends of body weights.

Similarly, the ventilator-derived VCO₂ based (EE: VCO₂) method of calculating energy requirements has consistently been shown to be inferior to IC[31] despite guidelines supporting its use.

Objectives

This review examines current methods of measuring EE in critical care, including IC, predictive equations and other alternatives. It also highlights the practical challenges in implementing these measurements at the bedside. By evaluating their accuracy, feasibility, and limitations, this review aims to synthesize existing evidence and provide a framework for optimizing energy delivery in critically ill patients, contributing to more personalized and effective nutrition support strategies.

MATERIALS AND METHODS

Information sources

We have applied for registration with PROSPERO-the International prospective register of systematic reviews and our registration number is CRD42025645920. We designed the screening process of our review based on the PRISMA statement[32] and extensively searched PubMed and Embase for all contemporary English language papers within the past 10 years (from December 2014 to December 2024), using the following search strings. Narrative analyses of each method were then conducted.

PubMed: 2 search strings were conducted and combined to include more relevant studies. Search string A: "Energy Metabolism"[MeSH Terms] OR "Energy Expenditure" OR calorimetry OR "predictive equations" OR "emerging technologies") AND ("Critical Care"[MeSH Terms] OR "Intensive Care Units"[MeSH Terms]) AND (accuracy OR feasibility OR limitations OR "practical challenges"). Search string B: ("energy expenditure" OR "energy metabolism" OR "indirect calorimetry" OR "predictive equations" OR "metabolic monitoring" OR "nutritional assessment") AND ("critical care"[MeSH Terms] OR "intensive care units"[MeSH Terms] OR "critically ill" OR "mechanically ventilated") AND ("nutrition support" OR "nutrition therapy" OR "energy intake" OR "personalized nutrition" OR "nutrition delivery" OR "nutritional optimization" OR "enteral nutrition" OR "parenteral nutrition").

Embase search string: ('energy metabolism' OR 'energy expenditure' OR 'energy use' OR 'calorimetry' OR 'indirect calorimetry' OR 'resting energy expenditure': OR 'REE') AND ('predictive equations' OR 'emerging technologies' AND ('critical care' OR 'intensive care unit' OR 'critical illness' OR ICU) AND (‘accuracy’ OR ‘feasibility’ OR ‘limitations’ OR 'practical challenges' OR 'bedside implementation').

Screening of records

All retrieved citations were first exported to EndNote (version X9) for reference management and compilation. Duplicate records were identified and removed using EndNote’s automated duplicate detection feature. The remaining citations were then imported into Rayyan, a web-based tool for systematic reviews for further duplicate screening and study selection, where an additional duplicate check was performed. Two independent researchers (the two authors of this paper) screened the titles, abstracts, and full texts of records within Rayyan, with conflicts resolved through discussion.

Study selection: Inclusion criteria

(1) Population: Studies involving adult patients in critical care or ICUs. Studies involving patients requiring mechanical ventilation, as they are commonly assessed for EE; (2) Interventions/methods: Studies evaluating IC, predictive equations, or emerging technologies (e.g., wearable devices, metabolic monitors) for measuring EE; (3) Outcomes: Studies reporting accuracy, feasibility, limitations, or practical challenges of these methods in clinical settings; (4) Study design: Original research articles, including observational studies, randomized controlled trials, and validation studies; (5) Language: Studies published in English; and (6) Time frame: Studies published in the last 10 years to capture current practices and technologies.

Study selection: Exclusion criteria

(1) Population: Studies focusing exclusively on paediatric or neonatal populations. Studies focusing exclusively on non-ICU populations, such as outpatient or general medical ward patients; (2) Interventions/methods: Studies not specifically assessing EE (e.g., studies focusing solely on energy intake or nutritional requirements without linking them to EE measurements); (3) Outcomes: Studies not addressing accuracy, feasibility, limitations, or practical challenges of the methods and studies focused on theoretical models or laboratory-only validations without bedside implementation; (4) Study design: Editorials, letters, commentaries, and conference abstracts without sufficient data. Systematic reviews, meta-analyses, and narrative reviews; and (5) Language: Studies published in languages other than English (unless translations are available).

Data extraction

Data on study characteristics, methods, accuracy, feasibility, limitations and applicability to nutritional optimisation of each method of energy measurement were extracted using a standardized template manually (Supplementary material). Two independent reviewers extracted data in parallel to ensure accuracy and consistency. Any inconsistencies in extracted data were resolved through cross-checking and discussion,

Quality assessment

The quality of included studies was assessed using the TRIPOD statement[33] and PROBAST for adherence to reporting standards and for risk of bias[34].

RESULTS

Included studies, adherence to reporting standards, and risk of bias

Out of 999 records screened, 25 articles fulfilled our inclusion criteria and were selected for review (Figure 1; Table 2). The database search was supplemented by reference list checks from existing systematic reviews and meta-analyses. Overall, adherence rates of publications to TRIPOD items ranged from 16% to 100%. Items concerned with model validation, performance sample size, and other miscellaneous information were reported in < 50% of the 25 publications reviewed. Items related to clinical context, study methodology, and implications were better reported, with > 90% of the publications providing adequate information regarding most of the corresponding items in the checklist. These are shown in Figure 2. Risk of bias assessment was conducted using PROBAST, and the results are shown in Figure 3.

Figure 1 PRISMA flow diagram.

Figure 2 Adherence of studies to reporting standards.

Figure 3 Risk of bias assessment.

Table 2 Studies included in review.

No.	Ref.	Method	Sample size	Study findings and outcomes	Challenges of implementation
1	Ferreruela et al[36]	IC	60	Good accuracy at FiO2 ≤ 0.6	Requires stable FiO2 settings
2	Graham et al[38]	IC	56 patients (all male)	VO2 and REE were significantly altered in sepsis, offering potential for early sepsis diagnosis	ICU operational challenges, patient-specific contraindications
3	Grguric et al[42]	IC, PE	68 patients	Significant differences between predictive equations and measured REE. Predictive equations often underestimated energy expenditure	Complexity of accurately applying predictive equations
4	Hickmann et al[35]	IC, PE	49 ICU 15 healthy	Early exercise increased REE, influenced by inflammation markers	Control of exercise conditions in ICU
5	Jonckheer et al[59]	ET	19 CVVH runs	Bioenergetic imbalances ranged from -28% to +42% of REE; citrate use added significant non-intentional calories	Complexity of CVVH settings and caloric calculations
6	Jonckheer et al[59]	ET	10 patients	CO2 removal by CVVH slightly alters REE; changes were not clinically significant	Complexity of IC during CVVH
7	Kagan et al[53]	ET	80 patients, 497 measurements	Low agreement between REE-VCO2 and indirect calorimetry. Indirect calorimetry remains the gold standard	Variability in VCO2 accuracy, reliance on predefined respiratory quotient
8	Koekkoek et al[58]	ET	31 patients	VCO2 overestimated REE; low accuracy compared to IC	Technical challenges in integrating VCO2-based REE measurements
9	Kongpolprom[55]	PE	24	No predictive equation accurately estimated REE; Penn State 2010 was the most reliable	Cost and availability of IC in routine practice
10	Liew et al[57]	PE	108	HBE overestimated REE, especially in obese patients (BMI ≥ 30)	Resource constraints for IC use
11	Lindner et al[46]	IC, PE	90 ICU patients, 58 healthy controls	Predictive equations showed low accuracy rates; IC recommended	ICU-specific variables influencing REE measurements
12	Murray et al[52]	IC, PE	326 patients	Equations underestimated REE; IC superior, particularly in obese populations	Resource and training limitations for IC use
13	Niederer et al[37]	IC	38 patients over 7 weeks	Progressive hypermetabolism peaking at 3 weeks post-intubation; prolonged stress response	Operational challenges in maintaining longitudinal IC measurements
14	Oshima et al[31]	ET	278 patients	EEVCO2 showed insufficient accuracy compared to indirect calorimetry, particularly in critical states	Reliance on stable ventilator settings, limited utility in unstable patients
15	Rehal et al[40]	IC, ET	22 patients, 48 measurements	E-sCOVX and Quark RMR overestimated VO2 and VCO2 by 10% compared to Deltatrac II. Limits of agreement within ± 20%	High variability and technical setup requirements
16	Rousseau et al[56]	IC, PE	55	None of the predictive equations were accurate; Penn State showed closest agreement	Need for accessible IC devices
17	Saseedharan et al[50]	IC, ET, PE	58 patients, 117 paired measurements	EE from IC was significantly lower than weight-based predictions, highlighting risk of overfeeding	Resource constraints during the pandemic
18	Shinozaki et al[43]	ET	10 (4 post-surgery, 6 critically ill)	Continuous and repeat measurements matched gold standard	Integration with existing ICU setups
19	Slingerland-Boot et al[41]	IC	27 patients	Beacon showed acceptable reliability but slightly underestimated REE compared to Quark	Need for device calibration and standardization
20	Sobhy et al[44]	IC, PE	50 patients	Faisy-Fagon overestimated caloric needs; PSUm showed higher accuracy. 23 kcal/kg/day offered unbiased estimates.	Complexity in applying equations accurately in varied patient conditions
21	Stapel et al[48]	IC, ET, PE	84	Comparable accuracy to IC, better than predictive equations	Requires integration with ventilator systems
22	Tah et al[51]	PE	294 (acute phase), 180 (late phase)	Single predictive equation valid for both phases, REE influenced by height, weight, age, and minute ventilation	Dynamic metabolic changes affect predictive accuracy
23	Takemae et al[60]	IC, PE	95	Novel equation (KTE) outperformed existing equations for Japanese patients	Operational complexity of IC
24	Tatucu-Babet et al[45]	ET	21 patients	Feasible for early ICU admission; EE lower than predicted by equations; increased over time	Complex setup, reliance on specific ECMO protocols
25	Vest et al[54]	IC, PE	25 patients	IC showed predictive equations underestimated energy needs; actual intake often < 70% of target	IC feasibility low; reliance on empirical equations

PE: Predictive equations; ET: Emerging technologies; ICU: Intensive care unit; BMI: Body mass index; REE: Resting energy expenditure; HBE: Harris-Benedict equation; ECMO: Extracorporeal Membrane Oxygenation; CVVH: Continuous veno-venous hemofiltration; IC: Indirect calorimetry; COVID-19: Coronavirus disease 2019.

DISCUSSION

IC

IC systems can be classified into two types: Closed circuit and open circuit. In a closed-circuit system, the patient breathes within a sealed air system of known volume[3]. However, closed-circuit systems are now rarely utilized in clinical practice. In contrast, open-circuit systems are commonly used in modern clinical settings. These systems involve the patient inhaling ambient air, with expired gases analysed through methods such as dilution, mixing chambers, or breath-by-breath techniques[4], and the Haldane transformation is used to estimate the volume of inspired air from exhaled volumes[9].

Benefits

As mentioned before, IC is the reference standard for measuring EE[10,11] and it is not surprising then that in the papers reviewed, IC measurements consistently outperformed predictive equations, with low bias and strong correlation with reference standards[31,35,36]. IC also demonstrated its advantages in hypermetabolism in specific patient groups. One study showed that IC had higher accuracy in detecting hypermetabolism among coronavirus disease 2019 (COVID-19) patients, with REE measured up to 143% of predicted values by week 3[37]. In another study, IC measurements showed that VO₂ and REE were significantly altered in patients with sepsis, offering potential for early sepsis diagnosis, demonstrating potential to improve identification and management of sepsis-related metabolic dysregulation[38].

Mixing chambers have traditionally been used for IC measurements. One of the first few IC devices available-the 30 year-old Deltatrac II^® (Datex)-employed a mixing chamber method and demonstrated high precision and accuracy[39]. It is however no longer in production, and this has spurred the development of alternative IC methods such as "breath-by-breath" instruments. Recent studies have demonstrated variable performance of these breath-by-breath systems compared to mixing chamber technology. Rehal et al[40] evaluated 2 modern devices (E-sCOVX and Quark RMR systems) against Deltatrac II and found that both overestimated VO₂ and VCO₂ compared to the reference standard Deltatrac, corresponding to a 10% higher mean REE, but remained more accurate than predictive equations.

Limitations in accuracy

There are however cases where IC has been shown to display inaccuracy. Slingerland-Boot et al[41] compared the performance of different IC devices (Beacon and Quark) in ventilated ICU patients and showed that the Beacon device underestimated REE by 96.2 kcal/day compared to Quark. The potential sources of error that contribute to these discrepancies are summarised below.

Firstly, the use of IC is limited in patients requiring high values of fraction of inspired oxygen (FiO₂) and positive end-expiratory pressure (PEEP). The Haldane transformation needed for IC measurement (which assumes stable nitrogen concentration in gas exchange) becomes unreliable at high FiO₂ values because errors in oxygen concentration measurements are amplified Therefore, while measurements at FiO₂ ≤ 0.4 were accurate, FiO₂ > 0.6 showed unacceptably high errors[36,41]. Meanwhile, high PEPP (such as a PEEP > 10 cm H₂O) can alter alveolar gas exchange dynamics, affect stability of ventilatory circuit, and make the system prone to air leaks, leading to inaccuracies gas exchange measurements[42,43]. Secondly, error can arise from instability in the patient’s physiological state, such as variations in breathing patterns, metabolic rates, or ventilator settings. Breath-by-breath systems (e.g., E-sCOVX and Quark RMR) showed desynchronization between gas flow and concentration measurements in patients with irregular respiratory patterns, such as rapid shallow breathing or asynchronous ventilation. This led to significant variability in calculated VO₂ and VCO₂, and consequently, REE[40]. Critically ill patients with rapidly changing metabolic demands, such as those on vasopressor therapy or undergoing frequent feeding adjustments, can also have erratic VO₂ and VCO₂ values, which affect accuracy[44].

Thirdly, the suitability of IC in certain patient populations may be limited. Some ICU patients, such as those requiring Extracorporeal Membrane Oxygenation (ECMO) may be unsuitable for IC[45,46]. In ECMO patients, traditional IC cannot account for extracorporeal CO2 and O2 exchange, leading to inaccuracies[45]. Continuous veno-venous hemofiltration (CVVH) introduces non-intentional caloric exchanges through citrate, glucose, and lactate in dialysis fluids, which are not captured by standard IC[47]. However, modified IC techniques have an option to adjust for these factors. These will be discussed in a later section.

Feasibility and practical challenges

There are a few factors which limit its feasibility in clinical and bedside implementation. Firstly. IC requires specialized equipment, such as Deltatrac II or COSMED Quark RMR, and trained personnel for operation[40,42,48] which limits its widespread use. Secondly, time requirements include 30-60 minutes per session for setup and data collection[47] which may not be practical in dynamic ICU settings. Additionally, there is difficulty in ensuring steady-state conditions for measurement[40,49] need for frequent recalibration and artifact exclusion increase complexity[46,50] and lack of accessibility in resource-limited settings due to cost issues[42,44,51]. While new breath-by-breath devices were easier to integrate with existing ventilators and required less space compared to traditional IC devices, their measurements were prone to transient errors, reducing real-world reliability[36]. Other logistical barriers that limit its use include the additional tools and software for data interpretation as well as staff training in device operation and interpretation of complex IC data needed to assist clinicians in tailoring nutritional support[37,38,52].

Predictive equations

Benefits: Higher feasibility, easier bedside implementation: Equations are cost-effective and require no specialized equipment, making them widely used in ICUs[48,53]. No special equipment is needed, thus making it more feasible for resource-poor settings[38]. It similarly does not require the time needed for calibration, equipment maintenance and data collection that has been consistently reported as a limitation of traditional IC methods.

Limitations-inaccuracy: Predictive equations have been shown in some studies to have constantly underestimated REE by 20%-40% compared to IC[46,52,54], while other studies showed that they have overestimated REE compared to IC, with significant bias[35,42,44,50]. One study assessing 14 different predictive equations found that none of them accurately estimated REE within ± 10%[55]. Among the traditional predictive equations, the Penn-State equation has repeatedly been shown to have the higher accuracy predictive methods, but still demonstrated variability[53,55,56].

One reason for the inaccuracy of predictive equations is their inability to capture individual variability in metabolism or stress responses[35,44,47,52]. This is especially so for specific patient groups like mechanically ventilated and obese patients[46,50], where the Harris-Benedict equation significantly overestimated REE in patients with a body mass index (BMI) ≥ 30 kg/m², leading to potential overfeeding[57]. Secondly, they rely on static factors like weight and height, which may not reflect dynamic metabolic changes[42,44]. Furthermore, many of these traditional equations were developed for the western population, and it has been shown to be unreliable in mechanically ventilated Thai patients[55]. Overreliance on these equations can lead to underfeeding or overfeeding, leading to adverse outcomes[54].

Other alternatives and emerging technologies

Ventilator-derived carbon dioxide production with nutritional RQ expenditure-EEVCO₂: An alternative way to determine EE in mechanically ventilated critically ill patients is to measure only carbon dioxide output (VCO₂) which allows use of the Weir equation to estimate VO₂[58], assuming a fixed RQ-the ratio of VCO₂ to VO₂. This approach is possible because most mechanical ventilators can continuously measure VCO₂[48].

This approach offers a few benefits. Firstly, EE: VCO₂ calculation was instantaneous once VCO₂ data were available, whereas IC required longer setup and measurement times. It also leveraged on existing ventilator equipment, reducing need for dedicated IC devise and cost. The reliance on existing ventilator-integrated sensors makes equipment maintenance easier as well[31]. Stapel et al[48] managed to achieve better accuracy using ventilator measured VCO₂ alone in ICU patients, determining RQ from the nutrition regimen to obtain more dynamic values, and obtained more accurate readings than the most frequently used predictive equations.

However, there are some limitations, the most glaring of which is its inaccuracy compared to the reference standard. Koekkoek et al[58] compared the use of EE: VCO₂ from a Hamilton-S1 ventilator with Quark RMR IC device and found that EEVCO₂ overestimated REE by 511 kcal/day. Its authors also found that including the food quotient did not improve the accuracy of EEVCO₂, and instead found that predictive equations preformed marginally better than EEVCO₂. Several reasons for its inaccuracy have been identified.

Firstly, their accuracy is affected by dependence on assumed RQ values, ventilator calibration and leaks. EE: VCO₂ has also been found to be unreliable in patients with fluctuating acid-base balance or ventilation settings, as these affect the RQ, leading to erroneous REE if these changes are not accounted for[31]. Shifts in metabolic substrate utilization have resulted in errors of ± 327 kcal/day[53], with errors margins of 5%-30% observed[31]. Meanwhile, calibration errors in VCO₂ sensors or air leaks in the ventilator circuit have been shown to cause inaccuracies in measured VCO₂, with potential biases exceeding 100 kcal/day[48,53].

Overall, despite sometimes conflicting evidence regarding its accuracy, EE: VCO₂ potentially offers a more cost-effective, less resource-intensive method to estimate REE while retaining sufficient accuracy, providing a practical alternative in settings without access to IC[31], but understanding of RQ variability and its impact on calculation of EE is needed among staff.

ECMO-adjusted IC

As mentioned before, CVVH affects the accuracy of REE measured by IC. While significant correction for CO₂ removal is not needed, variability in citrate metabolism impacts metabolic rate[59]. Adjusted IC protocols, such as for VA ECMO patients or CVVH, have been developed and these have enhanced measurement precision by incorporating real-time adjustments for metabolic contributions from ECMO and hemofiltration circuits. These protocols accounted for non-intentional caloric exchanges, such as those from citrate and glucose, allowing for more accurate REE measurements.

Feasibility of these new technologies is affected by several factors. Adjusted protocols for ECMO or CVVH are applicable bedside but require highly trained staff[45,47]. Furthermore, they added complexity and time requirements due to the need for integrated data streams and specialized analysis as it requires recalibration with changes in CVVH settings and blood gas analysis for CO₂ correction[59].

Development of new predictive equations

Takemae et al[60] implemented a novel prediction equation in septic patients in a Japanese ICU and demonstrated its superior accuracy compared to the Harris-Benedict, Ireton-Jones, and Schofield equations with smaller prediction error and bias, potentially allowing for more precise measurement in Asian populations. Tah et al[51] developed a new unified predictive equation incorporating dynamic variables that provided acceptable estimates across acute and late ICU phases. However, the main limitation to these new predictive equations lies in their limited generalizability and lack of validation in diverse ICU settings and populations.

CONCLUSION

IC remains the reference standard for accuracy in measuring EE, consistently outperforming predictive equations and emerging alternatives in critically ill patients. While it is unparalleled in accuracy, it is limited by cost, training requirements, and logistical challenges.

Predictive equations, despite their feasibility and low cost, lack precision and adaptability for dynamic ICU settings. As such, despite their practicality for bedside use, they frequently lead to under- or overfeeding.

Emerging technologies and alternatives such as EE: VCO₂ and adjusted IC for ECMO or CVVH patients hold promise but require further validation and optimization. These could bridge existing gaps but are not yet ready for routine clinical application. Lack of standardization in protocols for these emerging methods hinders widespread use, and equipment costs and training demands limit adoption in low-resource settings[45].

Interpretation

The dynamic nature of metabolic changes in critically ill patients necessitates accurate, real-time methods like IC for effective nutritional planning. Routine use of IC in ICU settings could prevent malnutrition and overfeeding, improving recovery metrics. Predictive equations should be supplemented with clinical judgment or IC measurements where available. Investment in training and infrastructure for IC and emerging technologies will be crucial. Clinicians should weigh accuracy against feasibility when choosing EE measurement methods. Training and resource allocation can enhance the implementation of advanced tools like IC.

Strengths and limitations of this paper

Strengths include adherence to PRISMA guidelines and a comprehensive search strategy with inclusion of studies representing diverse methodologies and patient conditions. This paper also included a discussion on specific patient subgroups, such as COVID-19, ECMO, and CVVH patients, for a nuanced understanding of applicability.

There are some limitations. Variability in study designs and sample sizes limited comparability and generalizability. The retrospective nature of several studies reduced control over confounding variables, and there was a lack of external validation for emerging technologies and some IC adjustments.

Future directions

Some areas of future research include the development of portable, user-friendly calorimeters, conducting large-scale validation studies for emerging technologies and newer predictive equations, and leveraging on tools such as artificial intelligence (AI). AI is not new in healthcare or in the context of ICU, and AI algorithms such as reinforcement learning (RL) have been used in multiple situations. RL aims to optimize decision making by using interaction samples of an agent with its environment and the potentially delayed feedbacks[61]. Vent AI is a RL model that was shown to be able to suggest dynamically optimized mechanical ventilation regimen for critically-ill patients[62]. RL has also been applied to optimize dosing of propofol for sedation in ICU[63] and heparin[64]. Given the potential benefits of AI, further studies could focus on integrating AI with nutritional therapy in the ICU to optimise feeding.