The efficacy of a home-use metabolic device (Lumen) in response to a short-term low and high carbohydrate diet in healthy volunteers

ABSTRACT

Background

Based on stoichiometric assumptions, and real-time assessment of expired carbon dioxide (%CO₂) and flow rate, the Lumen device provides potential for consumers/athletes to monitor metabolic responses to dietary programs outside of laboratory conditions. However, there is a paucity of research exploring device efficacy. This study aimed to evaluate Lumen device response to: i) a high-carbohydrate meal under laboratory conditions, and ii) a short-term low- or high-carbohydrate diet in healthy volunteers.

Methods

Following institutional ethical approval, 12 healthy volunteers (age: 36 ± 4 yrs; body mass: 72.1 ± 3.6 kg; height: 1.71 ± 0.02 m) performed Lumen breath and Douglas bag expired air measures under fasted laboratory conditions and at 30 and 60 min after a high-carbohydrate (2 g·kg⁻¹) meal, along with capilliarized blood glucose assessment. Data were analyzed using a one-way ANOVA, with ordinary least squares regression used to assess the model between Lumen expired carbon dioxide percentage (L%CO₂) and respiratory exchange ratio (RER). In a separate phase, 27 recreationally active adults (age: 42 ± 2 yrs; body mass: 71.9 ± 1.9 kg; height: 1.72 ± 0.02 m) completed a 7-day low- (~20% of energy intake [EI]; LOW) or high-carbohydrate diet (~60% of EI; HIGH) in a randomized, cross-over design under free-living conditions. L%CO₂ and derived Lumen Index (L_I) were recorded daily across morning (fasted and post-breakfast) and evening (pre/post meal, pre-bed) periods. Repeated measures ANOVA were employed for main analyses, with Bonferroni post-hoc assessment applied (P ≤ 0.05).

Results

Following the carbohydrate test-meal, L%CO₂ increased from 4.49 ± 0.05% to 4.80 ± 0.06% by 30 min, remaining elevated at 4.76 ± 0.06% by 60 min post-feeding (P < 0.001, η_p² = 0.74). Similarly, RER increased by 18.1% from 0.77 ± 0.03 to 0.91 ± 0.02 by 30 min post-meal (P = 0.002). When considering peak data, regression analysis demonstrated a significant model effect between RER and L%CO₂ (F = 5.62, P = 0.03, R² = 0.20). Following main dietary interventions, no significant interactions (diet × day) were found. However, main diet effects were evident across all time-points assessed, highlighting significant differences for both L%CO₂ and L_I between LOW and HIGH conditions (P < 0.003). For L%CO₂, this was particularly noted under fasted (4.35 ± 0.07 vs. 4.46 ± 0.06%, P = 0.001), pre-evening meal (4.35 ± 0.07 vs. 4.50 ± 0.06%, P < 0.001), and pre-bed time-points (4.51 ± 0.08 vs. 4.61 ± 0.06%, P = 0.005).

Conclusion

Our findings demonstrated that a portable, home-use metabolic device (Lumen) detected significantly increased expired %CO₂ in response to a high-carbohydrate meal, and may be useful in tracking mean weekly changes to acute dietary carbohydrate modifications. Additional research is warranted to further determine the practical and clinical efficacy of the Lumen device in applied compared to laboratory settings.

KEYWORDS:

• Lumen
• respiratory
• dietary
• nutrition
• metabolism
• carbohydrate

Introduction

Nutritional intake is an essential component influencing exercise adherence and aerobic/anaerobic sports performance [1–7]. As part of this, metabolic responses to dietary intake can provide strategic advantages when planning and undertaking exercise programs [8]. The importance of individualized dietary advice has been increasingly recognized, including day-to-day tailored recommendations before, during, and after training and/or competition [3]. Estimates of metabolic responses under resting (fed/fasted) and exercise conditions are typically measured using indirect calorimetry, notably the use of Douglas bag expired air collection [9,10], through to ventilated hood or breath-to-breath analyzers [10–15]. Through assessment of expired fractional gas measures under controlled laboratory conditions [15], metabolic rate and relative contribution of energy substrates (notably fat and carbohydrate oxidation rates) can be estimated from stoichiometric equations [15–20] or extrapolated from the respiratory quotient (RQ)/respiratory exchange ratio (RER) [21]. Estimation of macronutrient contribution to energy demands is important for evaluating metabolic change and/or metabolic flexibility in relation to dietary or exercise-based interventions [22–25]. However, accurate determination of metabolic patterns typically involves controlled laboratory conditions, as well as expensive analytical equipment [13,26–28].

Moreover, while laboratory-based studies tend to report pre-to-post dietary intervention effects, it is challenging to accurately quantify non-laboratory-based metabolic measures to quantify the metabolic effects of dietary programs. Home-use devices such as glucose monitors and cardiovascular trackers are becoming more available and, with technological advancements, are becoming more cost-effective [29]. However, there are currently limited market resources to quantify expired air analysis outside of laboratory conditions (i.e. at home or other field settings). Therefore, the potential to accurately assess or track substrate utilization away from the laboratory may be beneficial in supporting nutrition or exercise-based research (e.g. participant adherence to dietary interventions) or from a consumer self-monitoring perspective (e.g. application of a personalized nutrition approach, or periodized dietary preparations for an event). One such device (Lumen, MetaFlow Ltd.) currently available to consumers is marketed as the ‘first hand-held, portable device to accurately measure metabolism’ (www.lumen.me). Based on real-time assessment of expired percentage of carbon dioxide (%CO₂) and flow rate after a breath-hold procedure, the Lumen device is proposed to support monitoring for weight loss programs and longer-term metabolic flexibility, as well as support individual responses to tailored dietary strategies. Based on stoichiometric assumptions, a lower RER (i.e. volume of expired CO₂ [$\dot{\mathrm{V}}$CO₂] per minute divided by the volume of inspired oxygen [$\dot{\mathrm{V}}$O₂] per minute) of ~0.7 indicates the prevalence of fat oxidation to total energy contribution. In contrast, a raised RER value approaching ~1.0 indicates greater dominance from total carbohydrate oxidation [30,31].

Whilst the Lumen device only measures %CO₂, previous validation work has been undertaken in comparison to laboratory-based assessment of RER [32]. Following a 150 g total glucose load, both RER and Lumen %CO₂ (L%CO₂) significantly increased, with regression analysis supporting the agreement between an acute 0.09 unit increase in RER corresponding with a 0.28% increase in L%CO₂ [32]. Therefore, the Lumen device could provide indirect estimates of substrate use [33], and potentially be used to track changes in dietary (i.e. carbohydrate) intake [34]. However, assessment of the Lumen device under free-living conditions is warranted to understand device efficacy, particularly in response to acute dietary changes (e.g. a reduction in total carbohydrate intake or carbohydrate loading). This could facilitate its use in wider research and clinical settings, benefit sport and exercise applications, and support consumer guidance. This project therefore involved two aims: i) to undertake an independent assessment of the Lumen device under controlled laboratory conditions in relation to a standardized test meal; and ii) to investigate the efficacy of the Lumen device in tracking %CO₂ responses to an acute low- and high-carbohydrate diet in healthy volunteers.

Materials and methods

Ethical approval

This project employed a two-phase approach encompassing a laboratory-based device assessment and a separate dietary intervention study. For both phases, institutional ethical approval was obtained from the School of Psychology and Sports Science Research Ethics Panel, Anglia Ruskin University (Ethical approval number: SREP/SES_Staff_19-21) and was conducted in accordance with the Declaration of Helsinki (2013).

Phase 1: laboratory-based device assessment

Study participants and eligibility criteria

Following a priori power calculation (G*power3, Dusseldorf, Germany [35]; using α = 0.05; 1 − β = 0.80, based on observed L%CO₂ responses to a test meal [32]), a minimum sample size of 10 was determined. Participants were required to meet specific eligibility criteria for this phase, including: no known history of metabolic disorders, cardiovascular disease, or hypertension; no recent illnesses or viral infections (including Covid-19); no use of medication or participating in any current diet programs. All participants were considered generally healthy, with a body mass index (BMI) of <29.9 kg·m², and were actively engaged in recreational exercise weekly (~3–5 sessions per week). Importantly, all participants were required to have access to a smartphone (iOS or Android) for the purposes of Lumen mobile application accessibility. Before study inclusion, participants attended a full study briefing and provided written informed consent. From initial recruitment, 12 participants (5 male, 7 female) satisfactorily completed this phase (mean ± standard error [SE]: age: 36 ± 4 yrs; body mass: 72.1 ± 3.6 kg; height: 1.71 ± 0.02 m; BMI: 24.6 ± 1.2 kg·m²).

Experimental protocol

All participants were initially provided with an individual Lumen device with study-specific login details and guided instructions for use. An initial 7-day period was allocated to allow participant familiarization with the device and the specific breath maneuver, and synchronization of the Lumen software to individual daily breath responses. The device utilizes a single breath measure (after a standardized seated rest period) involving a deep inhalation with a 10 s breath hold followed by a flow-rate controlled exhalation procedure. Based on the assumption that under resting state conditions individual oxygen consumption (O₂) should remain relatively constant, metabolic fluctuations should therefore reflect CO₂ changes in expired air. Device reliability was derived from duplicate %CO₂ measures within <0.2%. If the error margin exceeded 0.2%CO₂, users were required to rest a further 5 min before repeating the procedure to ensure stable reliability. A Lumen Index (L_I) from 1 to 5 is derived from the application software based on individual updated L%CO₂ range (lowest to highest), in alignment with prior validation research [32]. An L_I of 1 indicates the lowest assumed RER based on L%CO₂ regression agreement with RER (i.e. greater relative fat oxidation), and 5 indicates a higher assumed RER (hence greater relative carbohydrate utilization), with remaining values based on quintile distribution of the individual L%CO₂ range.

Following device habituation, participants attended the Human Physiology Laboratory in a rested state (avoiding exercise in the pre-24 h period) and having completed a standardized overnight fasting period (~10–12 h). Upon arrival, participants were assessed for body mass (Tanita SC-330ST, Amsterdam, The Netherlands) and height (Seca CE123, Hamburg, Germany), before resting in a comfortable supine position for 10 min in a thermoneutral environment (20.6 ± 0.9°C, 61 ± 5% humidity, 758.3 ± 2.8 mmHg). Participants then performed duplicate seated Lumen measures as guided by the mobile application platform. On completion of the Lumen measures, participants remained seated whilst duplicate expired air samples were collected using the Douglas bag method, and analyzed for percentage O₂ and CO₂ using a Servomex MINIMP 5200 gas analyzer (Servomex Group Ltd, Crowborough, UK; gas calibration accepted with a <0.2% error margin, and flow rate/volume calibration accepted with a <1.0% error margin). Total Douglas bag volume was measured using a Harvard dry gas meter (Harvard Apparatus, Holliston, USA), with sample temperature recorded during volume measurement. Estimates of $\dot{\mathrm{V}}$O₂ and $\dot{\mathrm{V}}$CO₂ were corrected against ambient environmental conditions. A 20 μL capillarized blood sample was then collected for analysis of glucose using a Biosen C_Line automated analyzer (EKF Diagnostics, Cardiff, UK).

Following resting measures, participants then consumed a test breakfast meal (sports drink and porridge) comprising a total carbohydrate load of 2 g·kg⁻¹ body mass. This load was based on similar research [32] instigating a fixed 150 g carbohydrate dose, but tailored relative to participant body mass and based on a high-carbohydrate daily plan (~6 g·kg⁻¹). The meal consisted of a sports drink (Maurten Hydrogel Sports Fuel Drink Mix 320, Maurten UK Ltd., London, UK; comprising 41 g total carbohydrate per sachet), which was standardized to 75 g of total carbohydrate mixed with 400 mL of water. The remaining carbohydrate load was calculated against individual body mass delivered via a basic porridge meal (Quaker Oats So Simple, Reading, Berkshire, UK; comprising 26 g total carbohydrate per sachet). To standardize fluid intake, the test meal was made with a total of 9 mL·kg⁻¹ water (with sports drink water volume deducted). As such, the mean carbohydrate intake for the test meal was 69.1 ± 25.1 g, and mean fluid intake (excluding sports drink) was 248.5 ± 113.0 mL. Mean total carbohydrate load was 144.1 ± 25.1 g. Participants were required to consume the test meal over a 10 min timed period. Further duplicate Lumen and Douglas bag measures and capillarized blood samples were collected post-meal at 30 min (P30) and 60 min (P60) (with the timer started at the beginning of consumption of the test meal).

Phase 2: main dietary intervention assessment

Study participants and eligibility criteria

For the main intervention, sample size power calculation (G*power3, Dusseldorf, Germany [35]) was undertaken using L%CO₂ data from our laboratory-based study and determined a sample size of 15, using an α = 0.05 and 1 − β = 0.80. Participants were recruited based on the same inclusion criteria as the laboratory-based assessment study. Prior to study inclusion, participants attended a separate study briefing and provided written informed consent. An initial pool of 31 participants took part in the study, from which 4 were eliminated from the final analysis based on non-compliance with the main study protocol (i.e. were either on a hypocaloric diet or did not conform to the study parameters). A cohort of 27 participants (9 males, 18 females) were therefore included in final analyses (Mean ± SE: age: 42 ± 2 yrs; body mass: 71.9 ± 1.9 kg; height: 1.72 ± 0.02 m; BMI: 24.5 ± 0.7 kg·m²).

Experimental design and intervention

For the main study, a randomized, cross-over, dietary controlled design was implemented. Participants initially attended the laboratory to collect study equipment, comprising a Lumen device (with individual study login/code allocation), a Xiaomi Mi Smart Band 4 heart rate monitor, and a Xiaomi Mi Body Composition Scale 2 (Xiaomi Technology UK Ltd., Reading, UK). During this initial visit, participants were measured for both height (Seca CE123, Hamburg, Germany) and body mass (Seca 780, Hamburg, Germany); with laboratory assessed body mass cross checked against the Xiaomi Mi Smart Scale (mean difference 0.2 ± 0.1 kg, P > 0.05). Participants were also provided with guided instructions into the set up and use of the Lumen device, along with an activity diary and individual MyFitnessPal application accounts (MyFitnessPal, Inc., San Francisco, CA, USA).

Participants then had a 10-day habituation period to activate their Lumen application and become familiar with the specific breath maneuver, and gather typical ranges based on minimum/maximum L%CO₂ measured values. During the final 7-days, participants recorded their normal exercise patterns using the daily activity log provided to account for session type, mean session heart rate, exercise duration, and session perceived exertion. From this mean weekly training load, training monotony and strain were estimated based on previous research [36]. In addition, participants were provided with individual guidance on daily completion of dietary intake using the smart phone application (MyFitnessPal) with due attention to meal content and brands, portion size/weight, and fluid intake, as previously reported by our group [37]. Individual records were assessed daily by the same researcher to ensure satisfactory compliance and completion detail. Food diaries were then assessed by the same researcher using Nutritics Professional Dietary Analysis software (Nutritics Ltd., Co. Dublin, Ireland).

From this initial habituation period (NORM), individual maintenance caloric intake was determined based on predicted basal metabolic rate (using the Harris-Benedict formula), adjusted against habitual training demands and estimated non-exercise activity thermogenesis [38]. Participants were randomly assigned a 7-day dietary intervention comprising either isocaloric low-carbohydrate (LOW) intake (target ratio of ~20–25% EI from carbohydrate, 15–20% protein, 55–60% fat) or isocaloric high-carbohydrate (HIGH) intake (target ratio ~55–60% EI from carbohydrate, 15–20% protein, 20–25% fat). Individual guidance on meal-planning was provided for each dietary intervention. The dietary interventions were interspersed with a ‘wash-out’ period of at least 7 days returning to habitual intake.

Throughout all dietary periods, participants were requested to perform duplicate Lumen device measures at the following daily stipulated periods: i) overnight fasted (~10 h), morning assessment (30 min post-waking) and morning body mass; ii) 45 min post-breakfast; iii) immediately pre-evening meal; iv) 45 min post-evening meal; and v) pre-bed. Instructions were provided for participants to rest in a comfortable seated position for ~5 min before each measured time-point to standardize the process.

Quality control

Every Lumen breath session undertaken by the participants was stored through the Lumen application in an online platform, and reviewed by one experienced researcher on the acceptability and repeatability criteria for each breath. Acceptability was based on ambient air %CO₂ and Lumen breath maneuver requirements for inhaled volume, breath hold time, and exhaled volume. The session %CO₂ was determined through replicate measures with a difference <0.2%. Breath sessions that did not meet acceptability criteria were removed from the final analysis.

Statistical analyses

Statistical analyses were performed using SPSS (v26, IBM, NY, USA). Dependent variable distributions were assessed for normality using a Shapiro-Wilk test, with outlier evaluation using 1.5 x interquartile range. Laboratory-based data were assessed using a one-way ANOVA, with ordinary least squares regression used to assess the model between L%CO₂ and RER. For the main intervention, a mixed design repeated measures ANOVA (diet, day) was performed for L%CO₂ and L_I score, with Bonferroni post-hoc assessment where applicable. Where sphericity was violated, a Greenhouse–Geisser correction was applied. An α level of ≤0.05 was employed for statistical significance, with effect size (partial eta squared; η_p²) also reported (small = 0.02, medium = 0.13, large = 0.26). Data are reported as the mean ± SE.

Results

Phase 1: laboratory-based device assessment

Laboratory-based assessment of the Lumen device is shown in Figure 1 for mean blood glucose, RER, and L%CO₂. Blood glucose significantly increased (F = 21.82, P < 0.001, η_p² = 0.67) from 4.57 ± 0.22 to a peak of 7.44 ± 0.49 mmol·L⁻¹ by 30 min post-meal (P = 0.001), and remained elevated at 60 min (6.89 ± 0.36 mmol·L⁻¹; P < 0.001 compared with rest). Likewise, RER peaked at 30 min post-meal (F = 16.37, P < 0.001, η_p² = 0.62), increasing by 18.1% to 0.91 ± 0.02 (P = 0.002) and remained elevated compared with resting RER at 60 min (0.86 ± 0.02; P = 0.003). The impact of the test meal also resulted in a significant increase in L%CO₂ (F = 30.49, P < 0.001, η_p² = 0.74) from 4.49 ± 0.05% to 4.80 ± 0.06% by 30 min, remaining elevated at 4.76 ± 0.06% at 60 min (P < 0.001). For all measures, no differences were observed between P30 and P60 (P > 0.05).

Figure 1. Comparison between fasted and post-meal (30 and 60 min) laboratory and Lumen device measures: (a) blood glucose (mmol·L⁻¹); (b) expired air respiratory exchange ratio (RER); and (c) Lumen%CO₂ (L%CO₂). *** Denotes significant difference between paired timepoints (P ≤ 0.003). **** Denotes significant difference between paired timepoints (P < 0.001).

Mean Douglas bag data are shown in Table 1. No significant differences were reported over time for fractional expired oxygen percentage (F_EO₂%) or $\dot{\mathrm{V}}$O₂ (L·min⁻¹; P > 0.05). Fractional expired %CO₂ (F_ECO₂%) increased significantly over time (F = 4.48, P = 0.03, η_p² = 0.31), with post-hoc analysis demonstrating a significant increase by P30 (P = 0.047) but not P60 (P > 0.05). However, $\dot{\mathrm{V}}$CO₂ (L·min⁻¹) significantly increased over time (F = 15.90, P < 0.001, η_p² = 0.61) both at P30 (P = 0.005) and P60 (P = 0.001) compared with fasted. Minute ventilation ($\dot{\mathrm{V}}$_E) also increased over time (F = 8.42, P = 0.002, η_p² = 0.46), with paired significance only demonstrated at P60 compared with fasted (P = 0.014).

Table 1. Mean Douglas bag data for fasted and post-meal respiratory measures.

Variable	Fasted	P30	P60
FEO2 (%)	18.27 ± 0.20	18.44 ± 0.19	18.39 ± 0.18
FECO2 (%)	2.13 ± 0.12	2.30 ± 0.14*	2.26 ± 0.12
$\dot{\mathrm{V}}$E (L·min^−1)	13.62 ± 1.36	15.48 ± 1.31	17.19 ± 1.35*
$\dot{\mathrm{V}}$O2 (L·min^−1)	0.36 ± 0.03	0.38 ± 0.02	0.42 ± 0.02
$\dot{\mathrm{V}}$CO2 (L·min^−1)	0.28 ± 0.02	0.34 ± 0.02*	0.37 ± 0.02*
RER	0.77 ± 0.03	0.91 ± 0.02*	0.86 ± 0.02*

Data presented as mean ± SE. F_EO₂ (%) = fraction of expired oxygen (%); F_ECO₂ (%) = fraction of expired carbon dioxide (%); $\dot{\mathrm{V}}$_E = absolute minute ventilation (STPD); $\dot{\mathrm{V}}$O₂ = absolute oxygen consumption standardized; $\dot{\mathrm{V}}$CO₂ (STPD) = absolute carbon dioxide standardized (STPD); RER = respiratory exchange ratio. * = significantly different to fasted (P ≤ 0.047).

Ordinary least squares regression was employed to assess the model fit between RER and L%CO₂ using fasted and post-meal measures. When considering peak (P30) values, a significant regression model was found (F = 5.62, P = 0.03, R² = 0.20; see Figure 2 tile A) with L%CO₂ found to significantly predict RER (β = 1.042, P = 0.03). When mean post-meal data were considered, a significant model effect was also similarly found (F = 5.35, P = 0.03, R² = 0.20; see Figure 2 tile B). Based on these models, it was estimated that each 0.1 unit increase in RER corresponded with an 0.11% increase in L%CO₂.

Figure 2. Ordinary least squares regression between expired RER and Lumen %CO₂ (L%CO₂) for: (a) fasting and 30 min post-meal (P30); (b) fasting and mean post-meal data. Open circles indicate fasting measures; closed circles represent respective fed measures.

Phase 2: main dietary intervention assessment

Dietary monitoring and training load

Dietary intake adherence to the intervention is shown in Table 2. No significant differences were observed for absolute or relative caloric intake between LOW or HIGH, or compared with prior habitual intake (P > 0.05). Furthermore, no differences were observed between estimated maintenance kilocalories (2277.1 ± 60.3 kcal·d⁻¹) compared with caloric intake during both LOW and HIGH (P > 0.05). Protein intake as a percentage of caloric intake was equally maintained across both LOW (19.9 ± 0.3%) and HIGH (19.3 ± 0.4%) interventions (P > 0.05), in accordance with the targeted maintenance kilocalories.

Table 2. Mean dietary intake at baseline and across intervention periods.

		NORM	LOW	HIGH
EI	(kcal·d^−1)	2196.3 ± 77.2	2270.9 ± 65.8	2236.3 ± 72.4
	(kcal·kg^−1·d^−1)	31.0 ± 1.3	32.2 ± 1.1	31.3 ± 1.3
CHO	(%EI)	41.7 ± 1.9 *	19.9 ± 0.6	56.3 ± 0.8 ^#
	(g·d^−1)	231.3 ± 14.4 *	113.2 ± 4.9	315.4 ± 12.3 ^#
	(g·kg^−1·d^−1)	3.3 ± 0.2 *	1.6 ± 0.1	4.4 ± 0.2 ^#
PRO	(%EI)	18.0 ± 0.6 ^a	19.9 ± 0.3	19.3 ± 0.4
	(g·d^−1)	98.1 ± 4.0 ^b	113.2 ± 3.6	107.9 ± 3.9
	(g·kg^−1·d^−1)	1.4 ± 0.1 ^b	1.6 ± 0.1	1.5 ± 0.1
FAT	(%EI)	36.9 ± 1.6 *	57.7 ± 0.8	22.8 ± 0.6 ^#
	(g·d^−1)	89.4 ± 4.5 *	145.4 ± 4.3	56.2 ± 1.9 ^#
	(g·kg^−1·d^−1)	1.3 ± 0.1 *	2.1 ± 0.1	0.8 ± 0.0 ^#

Data presented as mean ± SE. NORM = normal dietary intake pre-intervention; LOW = low carbohydrate diet week; HIGH = high carbohydrate diet week. EI = caloric energy intake; CHO = carbohydrate; PRO = protein. * = significantly different to both LOW and HIGH (P < 0.001); ^# = significantly different to LOW (P < 0.001); ^a = significantly different to LOW only (P = 0.037); ^b = significantly different to both LOW and HIGH (P ≤ 0.03).

As expected, a significant main effect was found for total carbohydrate intake (F = 149.23, P < 0.001, η_p² = 0.85). For LOW, mean carbohydrate intake was at the lower range of the targeted 20–25% EI (19.9 ± 0.6%) and was significantly lower than both NORM and HIGH (P < 0.001). For HIGH, mean carbohydrate intake satisfactorily met the targeted 55–60% EI range (56.3 ± 0.8%) and was significantly greater than both NORM and LOW demonstrating adherence to the dietary interventions overall (P < 0.001). Likewise, there was a significant main effect for total fat intake (F = 258.54, P < 0.001, η_p² = 0.91), with mean relative contribution satisfactorily meeting the targeted 55–60% EI for LOW (57.7 ± 0.8%) and 20–25% EI for HIGH (22.8 ± 0.6%), in both cases being significantly different to NORM and the opposite intervention (P < 0.001). No significant differences were reported for body mass at the start of each dietary period (NORM: 71.9 ± 1.9 kg; LOW: 71.6 ± 2.0 kg; HIGH: 72.7 ± 2.3 kg, P > 0.05). Training load comparisons across all dietary periods are shown in Table 3. No significant differences were reported between dietary periods for any of the training load variables (P > 0.05) indicating relative consistency across the study duration.

Table 3. Training load comparisons across dietary periods.

Variable	NORM	LOW	HIGH
sTIME (mins)	54.0 ± 2.1	58.9 ± 2.3	58.7 ± 3.3
sHR (b·min^−1)	133 ± 3	133 ± 4	132 ± 4
sRPE (0-10)	5.3 ± 0.2	5.4 ± 0.2	5.5 ± 0.2
TL (AU)	1581.4 ± 125.1	1666.0 ± 146.5	1676.9 ± 129.6
TM (AU)	1.3 ± 0.1	1.2 ± 0.1	1.2 ± 0.1
TS (AU)	1863.5 ± 227.3	1784.8 ± 211.3	1881.1 ± 218.4

Data presented as mean ± SE. sTIME = mean exercise session duration; sHR = mean session heart rate; sRPE = mean session perceived exertion rating using a 0–10 visual analogue scale; TL = mean weekly intervention training load (AU; arbitrary units); TM = mean weekly training monotony based on TL; TS = mean weekly estimated training strain.

Lumen response to dietary interventions

Mean L%CO₂ and L_I values across each dietary period are shown in Figures 3 and 4, respectively. Under fasted conditions, there was a significant main diet effect for L%CO₂ (F = 8.05, P < 0.001, η_p² = 0.05) and L_I (F = 10.84, P < 0.001, η_p² = 0.07). For HIGH, mean fasted L%CO₂ (4.46 ± 0.06%) and L_I (2.9 ± 0.2 arbitrary units [AU]) were significantly greater than both NORM (P < 0.005) and LOW (P < 0.001). However, no post-hoc differences were observed between NORM and LOW (P > 0.05).

Figure 3. Mean Lumen %CO₂ responses for NORM, LOW, and HIGH according to: (a) ~30 min post-waking fasted assessment; (b) ~45 min post-breakfast; (c) immediately before evening meal; (d) ~45 min post evening meal; and (e) immediately before bed. * Denotes significant difference between paired timepoints (P ≤ 0.043). ** Denotes significant difference between paired timepoints (P ≤ 0.009).

Figure 4. Mean Lumen Index responses for NORM, LOW, and HIGH according to: (a) ~30 min post-waking fasted assessment; (b) ~45 min post-breakfast; (c) immediately before evening meal; (d) ~45 min post evening meal; and (e) immediately before bed. AU = arbitrary units. * Denotes significant difference between paired timepoints (P ≤ 0.028). ** Denotes significant difference between paired timepoints (P ≤ 0.005).

There was also a main diet effect found for L%CO₂ (F = 7.30, P < 0.001, η_p² = 0.04) and L_I (F = 7.49, P < 0.001, η_p² = 0.04) when considering mean values over each dietary period in relation to the first meal of the day. In contrast to morning fasted values, post-hoc analysis revealed significantly lower values following LOW (L%CO₂ = 4.48 ± 0.09%, L_I = 2.9 ± 0.2 AU) compared with both NORM (L%CO₂ = 4.57 ± 0.07%, L_I = 3.1 ± 0.1 AU: P < 0.023) and HIGH (L%CO₂ = 4.54 ± 0.08%, L_I = 3.2 ± 0.2 AU; P < 0.004).

Lumen values typically decreased by the evening meal. For the pre-evening meal measures, a significant main effect was found for both L%CO₂ (F = 19.59, P < 0.001, η_p² = 0.10) and L_I (F = 19.12, P < 0.001, η_p² = 0.10), demonstrating a similar pattern to post-breakfast measures. Post-hoc analyses demonstrated a significantly lower L%CO₂ (4.35 ± 0.07%) and L_I (2.5 ± 0.1 AU) for LOW compared with both NORM and HIGH (P < 0.001), although no differences were reported between NORM and HIGH (P > 0.05). The consensus between L%CO₂ and L_I was consistent until post-dinner measures. At this point, there was a main effect for L%CO₂ (F = 5.87, P = 0.005, η_p² = 0.03), whereby mean L%CO₂ was significantly different between intervention diets only (LOW = 4.54 ± 0.08%; HIGH = 4.63 ± 0.07%; P = 0.01). Whilst this same comparison was also detected for L_I (F = 12.13, P < 0.001, η_p² = 0.06), values were considerably greater for HIGH (3.5 ± 0.2 AU) compared to both LOW (3.0 ± 0.2 AU; P < 0.001) and NORM (3.2 ± 0.1 AU; P = 0.005). Post-dinner measures were not different between NORM and LOW (P > 0.05).

For the final measure for each day (pre-bed), a significant main effect was found for both L%CO₂ (F = 6.18, P = 0.003, η_p² = 0.03) and L_I (F = 14.97, P < 0.001, η_p² = 0.08). For L%CO₂, values were notably lower following LOW (4.51 ± 0.08%) compared with both NORM (4.60 ± 0.06%; P = 0.04) and HIGH (4.61 ± 0.06%; P = 0.005), respectively. Similarly, following LOW, L_I values were significantly lower (3.0 ± 0.2 AU) than both NORM (3.2 ± 0.1 AU; P = 0.02) and HIGH (3.4 ± 0.2 AU; P < 0.001). However, it was also noted that a significant difference existed between NORM and HIGH (P = 0.03) in contrast to the pattern for L%CO₂.

Discussion

The aim of the current study was to undertake an independent assessment of the efficacy of the Lumen breath device to track %CO₂ in response to acute dietary changes in healthy volunteers under real-world conditions. As part of this, we also undertook an independent assessment of the Lumen device under controlled conditions. Findings from our laboratory assessment concurred with those previously reported [32], demonstrating that glycemic response to a carbohydrate-rich test-meal corresponds with distinctive increases in both RER and L%CO₂. In the current study, however, we employed a fixed meal plan in contrast to previous research [32] which provided three 50 g glucose solutions separated by 20 min each. This was to instigate a more realistic feeding strategy in line with a high-carbohydrate meal approach. In the current study using the Douglas bag method, an 18.1% (or 0.14 unit) increase in RER corresponded with a 6.9% increase in L%CO₂ (or 0.31 absolute %CO₂ increase). Lorenz et al. [32] similarly demonstrated an 11.0% increase in RER (or 0.09 unit), using a metabolic cart, corresponded with a 6.6% increase in L%CO₂ (or 0.28%CO₂ increase). We also demonstrated a significant regression model effect between RER and L%CO₂ using both peak (P30) values (P = 0.03, R² = 0.20) and mean post-fed values (P = 0.03, R² = 0.20), again comparable to previous research [32]. Expired air analysis demonstrated that changes in RER following the carbohydrate meal likely reflect an increased $\dot{\mathrm{V}}$CO₂ (~940 mL), with a smaller (~673 mL) non-significant change in $\dot{\mathrm{V}}$O₂.

Our findings are consistent with previous research [39–46], whereby post-prandial $\dot{\mathrm{V}}$CO₂ production is associated with increased $\dot{\mathrm{V}}$_E to facilitate constant arterial CO₂ tension (PaCO₂). This also highlights the critical importance of standardizing resting procedures as part of Lumen measurements to ensure stable O₂ use. Our results therefore infer that relative changes in L%CO₂ in response to a test meal may be useful for tracking corresponding metabolic changes based on acute shifts in CO₂ production, on the assumption that O₂ consumption remains relatively constant. As the Lumen device does not measure %O₂, more accurate quantification of metabolic changes cannot be ascertained. Instead, the device may be pertinent for indirect tracking of metabolic alterations to carbohydrate-based meal patterns, and further validation work is warranted.

For the main study, dietary consistency was generally met across the intervention stages, with mean macronutrient distribution ranges achieved for both LOW and HIGH carbohydrate weeks. An important finding from our study was that when macronutrient ratios (particularly for carbohydrate intake) were significantly polarized (i.e. ~20% vs. 60% of caloric load), consistent and significant differences were found for both mean L%CO₂ and L_I for all timepoints assessed. However, no interactions between diet and time were found for any measures, indicating that day-to-day changes were not significant. These findings therefore highlight that mean weekly changes in L%CO₂ and L_I measures may provide better indicators of general metabolic adaptations in response to acute dietary changes, particularly when the macronutrient ‘shifts’ are significant. This may have important implications for end-user interpretations when using the device.

One contention with this suggestion, however, is that end-users do not view real-time L%CO₂ data, and instead only view a derived L_I to provide an indicator of metabolic fuel contribution (based on regression agreement between L%CO₂ and RER). However, a critical insight is that this largely depends on the spread of variance in L%CO₂ values and, indeed, assumed RER range for a particular individual. From a metabolic perspective, whilst a low L_I score (i.e. 1) may be associated with a lower relative RER value, if the distribution of L%CO₂ is narrow, then presumed metabolic changes may be minimal. Additionally, RER values typically range from 0.7 to 1.0 under resting conditions, with extremes of this range associated with fat and carbohydrate utilization respectively (and mixed fuel utilization in between). This has important implications for end-user interpretation of results, particularly if assessing acute changes or if resting procedures are not appropriately standardized.

Furthermore, in the current study, mean L_I data were compiled to one decimal place to reflect the subtle changes observed. Indeed, across both dietary interventions, mean L_I ranged from 2.5 (morning fasted on LOW) to 3.5 (post evening meal on HIGH), supporting the small effect sizes observed. Whilst it was expected that the transition to a low-carbohydrate diet might result in a consecutive reduction in L%CO₂ and L_I under fasting conditions as example, this was only partly observed over days 2–4, after which measures fluctuated in line with potential metabolic adaptations (i.e. ketosis [47,48]) resulting in subtle increases in L%CO₂. As such, end-users should be mindful of tracking mean weekly L_I to have a more reliable understanding of individual responses to acute dietary patterns at a particular point in the day.

When comparing differences between NORM and LOW, our findings indicated that statistically significant responses occurred for both mean L%CO₂ and L_I post-breakfast, pre-evening meal, and pre-bed. However, differences were not found under fasting conditions. This may indicate that metabolic adaptations to an acute low-carbohydrate diet are better observed at these times, with lack of significance under fasting conditions potentially reflecting acute ketogenic influence on CO₂ fluctuations in the latter half of the week. Longer term dietary monitoring is therefore warranted to clarify these findings, especially in light of ketogenic adaptations to prolonged low-carbohydrate or intermittent fasting protocols [49,50].

In contrast, the main time point where significant differences were observed between NORM and HIGH for both mean L%CO₂ and L_I was under fasting conditions. Mean carbohydrate intake under NORM represented 41.7% of EI compared with 56.3% for HIGH. Increased consumption of carbohydrate over the intervention period may have influenced carbohydrate oxidation rates [51,52] leading to elevated expired CO₂ under fasted conditions. Potentially, this infers that when dietary carbohydrate is significantly increased, Lumen measures tracked under fasting conditions may offer a more reliable or standardized time-point. However, further research is required to corroborate these observations.

It is important to recognize several limitations of the current study. Whilst sample size was deemed sufficient, a larger sample pool representative of wider consumer, athletic, or clinical populations is warranted to determine the effectiveness of the Lumen device in different settings. Indeed, the current study involved recreationally active volunteers, and expired percentage CO₂ may be elevated in trained athletes [53]. Furthermore, circadian hormonal patterns (i.e. across the menstrual cycle) were not determined which may influence metabolic responses to dietary interventions. Further research on the efficacy of the Lumen device, particularly with female users, is therefore warranted to determine endocrine influence on L%CO₂ variance. As this study aimed to assess device efficacy under real-world settings, specific meal volume/quantity and set timings were not controlled for. In the current study, overall effect sizes were relatively small; therefore, when strict meal type/amount and timings are followed, metabolic responses based on expired CO₂ may be greater. Furthermore, whilst overnight fasting periods were appropriately standardized (~10 h), longer term fasting periods may yield different results.

Whilst the current study demonstrates that the Lumen device may provide useful mean weekly differences between polarized dietary conditions, day-to-day variance did not yield significant findings. Therefore, future research should aim to provide further confirmation on the efficacy and accuracy of the device under controlled laboratory conditions (i.e. with standardized test meals), or with progressive increments of carbohydrate intake to determine whether a ‘threshold’ level for meaningful detection exists. As the current study investigated acute dietary interventions, it would also be meaningful to determine longer term patterns with device use when undertaking dietary programs (e.g. ketogenic diets) alongside laboratory (e.g. RER, blood ketone assessment) or other proxy measures (e.g. glucose monitoring, breath acetone).

Translational aspects

Our findings support previous research [32] that the device detects acute changes in L%CO₂ following a significant increase in carbohydrate intake, and collectively these results demonstrate that the Lumen device could have pertinent applications in non-invasively monitoring response to acute meal settings (e.g. pre-exercise carbohydrate strategies) considering previous research highlighting the association between carbohydrate oxidation and exercise performance [18,38,54–56].

For the main aspect of this study, our findings also highlight that the device could be useful for monitoring mean weekly changes in L%CO₂ and L_I in response to acute dietary approaches (i.e. carbohydrate reduction for weight loss, or carbohydrate-loading for endurance events). However, our results did not demonstrate day-to-day changes at a particular time-point, which may have pertinent applications for end-users when interpreting acute differences between days. The Lumen device may therefore be pertinent for monitoring relative consistency to a targeted nutrition program, with applications to research and individual-user contexts (e.g. at-home monitoring of adherence to a low- or high-carbohydrate diet).

It is important to note that results may be different for chronic interventions (particularly in line with assessment of metabolic flexibility) and further studies should assess the efficacy and practical benefits of using the device longitudinally. End-users should be mindful that metabolic responses to dietary interventions are both complex and highly individualized. This is further apparent when considering differences between trained/lean and untrained/obese populations. Whilst dietary interventions (e.g. low-carbohydrate ketogenic, fat restricted or targeted energy-reduced diets) may modify substrate utilization (i.e. an increase in fat oxidation) leading to improved body composition [57,58], particularly in overweight individuals [47,59], this may not necessarily confer performance benefits (e.g. increased peak power or time to exhaustion [57,60], prolonged endurance [58], or efficient glycogen use [61]). Use of additional measures (e.g. body composition, performance data) may therefore be relevant as an adjunct strategy to support application of the Lumen device. Finally, further research should be undertaken to assess device efficacy in relation to specific end-user goals pertinent to metabolic outcomes [62].

Conclusion

A portable, home-use metabolic device (Lumen) detected significantly increased expired %CO₂ in response to a high-carbohydrate test meal, and may be useful in tracking mean weekly changes to acute dietary carbohydrate modifications, especially when protein intake is standardized. Lumen measures following a high-carbohydrate diet may be better assessed under fasting conditions, whereas measures post-breakfast may be better suited for monitoring low-carbohydrate diets. Further research comparing applied versus laboratory settings is warranted to establish clinical and practical efficacy of the Lumen device, particularly in relation to end-user goals.

Abbreviations

BMI: body mass index; EI: energy intake; F_ECO₂: fraction of expired carbon dioxide; F_EO₂: fraction of expired oxygen; HIGH: high-carbohydrate diet intervention; L%CO₂: Lumen breath percent carbon dioxide; L_I: Lumen Index; LOW: low-carbohydrate diet intervention; NORM: habitual diet intake; RER: respiratory exchange ratio; RQ: respiratory quotient; $\dot{\mathrm{V}}$CO₂: volume of expired carbon dioxide; $\dot{\mathrm{V}}$O₂: volume of oxygen consumption; $\dot{\mathrm{V}}$_E: minute ventilation.

Authors’ contributions

JR, DD, TS conceived the idea and designed the study in conjunction with SY and MM. JR, DD, and TS conducted data collection. JR, DD, TS, JP, JL, and AW analyzed and interpreted the data. JR, DD, and TS prepared the manuscript with assistance from JL, JP, and AW. All authors read and approved the final manuscript.

Data sharing statement

The data collected and analyzed in this study are available from the corresponding author upon reasonable request (https://aru.ac.uk/people/justin-roberts).

Ethics statement

All protocols and procedures were approved by the School of Psychology and Sports Science Research Ethics Panel, Anglia Ruskin University. All participants provided written informed consent.

Acknowledgments

The authors would like to acknowledge Andrew Nwajei for his support in undertaking the laboratory-based device assessment phase. The authors would also like to thank MetaFlow Ltd. for their input in providing study equipment and technical support.

Disclosure statement

Whilst this study was undertaken independently, both SY and MM are employed by MetaFlow Ltd. As such, they were only involved with conceptual design and equipment support, and were fully independent of the main research team conducting the data collection for this study. TS has acted in a scientific advisory capacity for MetaFlow Ltd. This has therefore been noted for transparency.

Additional information

Funding

Equipment support was provided by MetaFlow Ltd. Open Access funding was provided by Anglia Ruskin University.