Abstract
Background:
Accurate dietary assessment remains a challenge, particularly in free-living settings. Continuous glucose monitoring (CGM) shows promise in optimizing the assessment and monitoring of ingestive activity (IA, i.e., consumption of calorie-containing foods/beverages), and it might enable administering dietary Just-In-Time Adaptive Interventions (JITAIs).
Objective:
In a scoping review, we aimed to answer the following questions: (1) Which CGM approaches to automatically detect IA in (near-)real-time have been investigated? (2) How accurate are these approaches? (3) Can they be used in the context of JITAIs?
Methods:
We systematically searched four databases until October 2023 and included publications in English or German that used CGM-based approaches for human (all ages) IA detection. Eligible publications included a ground-truth method as a comparator. We synthesized the evidence qualitatively and critically appraised publication quality.
Results:
Of 1,561 potentially relevant publications identified, 19 publications (17 studies, total N = 311; for 2 studies, 2 publications each were relevant) were included. Most publications included individuals with diabetes, often using meal announcements and/or insulin boluses accompanying meals. Inpatient and free-living settings were used. CGM-only approaches and CGM combined with additional inputs were deployed. A broad range of algorithms was tested. Performance varied among the reviewed methods, ranging from unsatisfactory to excellent (e.g., 21% vs. 100% sensitivity). Detection times ranged from 9.0 to 45.0 min.
Conclusion:
Several CGM-based approaches are promising for automatically detecting IA. However, response times need to be faster to enable JITAIs aimed at impacting acute IA. Methodological issues and overall heterogeneity among articles prevent recommending one single approach; specific cases will dictate the most suitable approach.
1 Introduction
Nutrition has a major impact on people’s health and well-being (1–8). However, accurately assessing nutrition and dietary intake remains challenging, with the most precise tools often involving high costs, participant and staff burden, or privacy issues (9–14). Yet, valid and reliable measurement of dietary behavior is essential to accurately detect changes in research settings and guide patient counseling in clinical practice (e.g., weight loss programs). Technological advances in recent years have led to new approaches for accurately assessing dietary intake that try to overcome some of the shortcomings of traditional dietary assessment methods (9, 15–18).
An attractive technology-based option for assessing the consumption of calorie-containing foods and beverages (ingestive activity, IA) is continuous glucose monitoring (CGM). CGM involves using a sensor that measures glucose concentrations in the interstitial fluid (19, 20) as a proxy for blood glucose levels (20, 21). CGM has become an important tool in diabetes care (19, 22–25). For instance, it is an integral component of artificial pancreas (AP) systems designed to automate and improve blood glucose regulation in individuals with type 1 diabetes mellitus (T1DM) via the utilization of CGM, an insulin infusion pump, and a control algorithm (26). Beyond diabetes management, CGM is gaining popularity for use in healthy individuals and athletes (20, 27). Several CGM devices show satisfactory accuracy data (20, 28, 29).
The automatic and (near-)real-time detection of IA via CGM could offer benefits in (clinical) practice, including a reduced participant and staff burden. In addition, interventionists could monitor meal plan adherence more closely and detect deviations from intervention goals as they occur. Consequently, targeted and personalized countermeasures could be deployed proactively. One particularly useful approach would be CGM-based detection of IA in the context of Just-In-Time Adaptive Interventions (JITAIs). JITAIs aim to exploit the full potential of remote monitoring combined with delivering intervention content in the moment/context when it is most needed and the patient is likely to be (most) receptive (30). Preliminary results show promising effects of JITAIs on predicting and preventing dietary lapses (31). If detection times of the CGM-based approaches in question were extremely short (e.g., less than a few minutes), JITAIs could aim at acutely impacting IA (e.g., sending a prompt asking a person to terminate a meal). If detection times were relatively short (e.g., less than an hour), JITAIs could aim at altering subsequent IA (e.g., a dinner meal). In both cases, information on IA would be much more readily available than with traditional dietary assessment methods (e.g., 24-h recalls).
However, there may also be challenges associated with the use of CGM for the automatic monitoring of IA. On the one hand, there are system-inherent challenges. For example, postprandial rises in blood glucose vary in timing and extent depending on meal composition, meal quantity, inter-individual variability, and many other factors (32–40). Further, there is a delay between interstitial fluid and blood glucose concentrations (20, 41, 42). On the other hand, blood glucose levels are not only influenced by IA but also by other factors such as physical activity, stress, and diurnal fluctuations (20, 41, 43–55). Thus, false positive detections (e.g., erroneously flagging a meal due to glucose increases caused by stress) and false negative detections (e.g., erroneously not flagging a meal because other factors render the glucose response too flat) might occur.
In the past years, research examining the use of CGM for the automatic detection of IA has accumulated. Recent publications reviewed options for the automatic detection of IA using wearable−/sensor-based methods (15–18), but they did not specifically address CGM. The present article aims to close this research gap and answer the following guiding questions:
Which approaches using CGM for the automatic detection of IA in (near-)real-time have been investigated, and have these approaches relied solely on CGM or also used other data (e.g., sensors/wearables)?
How accurate are these approaches in detecting IA?
Can these approaches be used in the context of JITAIs?
2 Methods
The reporting of this review is based on the updated Preferred Reporting Items for Systematic Reviews and Meta-Analyses Extension for Scoping Reviews (PRISMA-ScR) guideline (56).
2.1 Search strategy
The primary systematic literature search was conducted on 09 September 2022, using the IEEE Xplore, PubMed, Scopus, and Web of Science databases. An identical supplementary search was conducted on 02 October 2023. The search term was developed and refined by two authors (JB, CH) to capture all relevant publications, and the search term contained: (intake OR uptake OR eating OR ingest* OR meal OR drink* OR beverage OR consum* OR oral) AND (monitor* OR assess* OR detect* OR estimat* OR measur* OR sens*) AND (“continuous glucose monitoring” OR “real time continuous glucose monitoring” OR “real-time continuous glucose monitoring” OR “flash glucose monitoring” OR “intermittently scanned continuous glucose monitoring” OR CGM OR rtCGM OR isCGM OR “artificial pancreas” OR “artificial beta cell*” OR “artificial beta-cell*” OR “artificial β-cell*” OR “artificial β cell*”) AND (algorithm OR “deep learning” OR “machine learning” OR “neural network*” OR AI OR “artificial intelligence“). Because a fully-closed-loop AP system must first detect meals to adequately manage the following increases in glucose by delivering insulin to the patient (57), the search also included AP systems.
JB conducted the database searches and removed duplicates for the primary and supplementary searches. Two authors (JB, CH) screened the titles and abstracts against the predefined eligibility. In discrepancies, a consensus was reached via discussions and ineligible publications were discarded. For the primary search, JB screened the full texts of the remaining publications for eligibility and consulted with CH, who then independently screened these full texts in cases of uncertainty. In addition, one other author (SHF) conducted independent cross-checks for a randomly selected 20% of the full texts. For the supplementary search, two authors (JB, CH) independently screened the full texts of the remaining publications. Again, in case of discrepancies, a consensus was reached via discussions, and subsequently, ineligible publications were discarded. Another author (CG) was consulted for her technical/mathematical expertise during the screening process. JB hand-searched the reference lists of eligible publications for any additional relevant literature. In several cases, the corresponding authors of articles were contacted, e.g., to receive full texts or raw data or to clarify results.
2.2 Eligibility criteria
We included publications if the following inclusion criteria were met: (1) publications were written in English or German and published until 2 October 2023; (2) publications are original articles published in peer-reviewed scientific journals or conference papers; (3) at least one performance measure of the automatic detection of IA is reported explicitly. For example, the accuracy was calculated by comparing the CGM-based (did not have to exclusively rely on CGM as input) approach against a ground-truth method (e.g., self-reported or observed IA); (4) a CGM-based approach was used to detect IA in vivo in free-living, semi free-living, or laboratory settings. This also included trials of AP systems if criterion 3 was met; (5) only the most recent publication on a specific approach by a particular research group was included if it supersedes preceding publications.
We excluded publications for the following exclusion criteria: (1) the approach was not tested in human participants (e.g., in silico studies); (2) no outcome results were reported (e.g., study protocol publications); (3) outcomes did not include an explicit performance measure describing the results of the automatic detection of IA (e.g., only figures showcasing the CGM trends over time); (4) the methodology was described without sufficient detail. We did not apply restrictions regarding publication date or participant age.
2.3 Data extraction
The following information was extracted: (1) first author and publication year; (2) a summary of the study; (3) sample size and, if available, sex and age of participants; (4) participants’ diabetes status [no diabetes, prediabetes, T1DM or type 2 diabetes mellitus (T2DM)]; (5) scope of the study (duration/number of IA events) and, if available, information on the IA events (e.g., meal composition); (6) ground-truth/criterion method(s); (7) performance measure(s); (8) details on the CGM device and if applicable other relevant devices used in the study.
JB extracted relevant information from the original publications, and in cases of uncertainty, the respective publications were double-checked by CH. Two other authors (CG, SHF) also double-checked the extracted information. One other author (CG) further extracted technical details of the tested approaches.
2.4 Data synthesis
We synthesized the evidence qualitatively, focusing on answering the three research questions outlined above. Although using an explicit cutoff (e.g., ≥80% F1-score or accuracy) is desirable for performance evaluation and has been used in a related review (16), this approach was not feasible, as only a few publications reported accuracy and/or F1-score values.
Furthermore, we appraised the included publications critically. We considered the following aspects of being of concern: (1) error-prone methods for identifying the ground truth of IA [e.g., self-reported IA (58) or retrospective identification from CGM data, whereas inpatient settings with observed IA were generally assumed to be less error-prone]; (2) a sample consisting exclusively of individuals with diabetes as this might limit generalizability to non-diabetic populations; (3) meal announcement/meal-accompanying insulin boluses, as there might be an interference with the (early) postprandial blood glucose levels that are relevant for the automatic detection of IA; (4) algorithm inputs other than CGM since ultimately a CGM-only approach would be desirable to minimize costs and effort.
3 Results
The literature search identified a total of 1,561 potentially relevant publications. Nineteen publications reporting data from 17 studies (for 2 studies, 2 publications each were relevant, see Table 1), including 311 participants, met the inclusion criteria (59–66, 68, 70, 71, 73–75, 82–85, 87). Figure 1 shows the process of the literature search, screening, and selection in a PRISMA-style flow diagram (88).
Table 1
| Author, year | Study summary | N (male/female), age (if available) | Diabetes status | Scope of the study/information on the assessed eating events | Ground-truth method(s) | Performance measure(s)a | Details on the CGM device and other input devices |
|---|---|---|---|---|---|---|---|
| Atlas et al. (59) | Pilot feasibility clinical trial investigating the performance of the MD-Logic Artificial Pancreas (MDLAP) System, a fully CL system utilizing a patient’s diabetes treatment management in conjunction with a fuzzy logic-based control-to-range module and a control-to-target module to control blood glucose levels. | 7 (2/5); 23.9 ± 3.4 years, range 19–30 years | T1DM | 8 h CL sessions in a resting state, with 9 fasting sessions (n = 6) and 3 meal challenge sessions (n = 2; mixed meal with 40–60 g of CHO after 8 h fast) | Inpatient study | Overall mean detection time: 23 min after consumption | CGM: Freestyle Navigator (Abbott Diabetes Care, Alameda, CA) or STS-Seven System (DexCom, San Diego, CA); sampling rate: 5 min |
| Two participants completed 1 additional 24 h CL session each, in which mixed meals (CHO content for each meal was 17.5–70 g) were consumed at 1,930 h, 0800 h, and 1,300 h, with participants entering the sessions after ≥3 h of fasting. | |||||||
| Bertrand et al. (60)b | Comparison of eating activity detection systems using (1) wearable wristbands vs. (2) wearable wristbands + CGM. Three machine learning algorithms were applied for the classification of eating and non-eating events: support vector machine (SVM), random forest (RF), and extreme gradient boosting tree (XGB). For each algorithm, one model based on the CGM data was compared to a model without CGM data. | 10 (5/5); range 19–51 years | Healthy, non-diabetic | Up to 2 weeks | An app (“aTimeLogger”) was used to log the ground truth | Mean (standard deviation)c: | CGM: FreeStyle Libre 2 |
| MCCXGB: 0.35 (0.10) − MCCSVM: 0.37 (0.06) | |||||||
| F1-scoreXGB: 0.49 (0.10) − F1-scoreSVM: 0.50 (0.08) | |||||||
| SensitivityXGB: 0.67 (0.19) − SensitivitySVM: 0.63 (0.20) | Wearable wristbands measuring steps and heart rate: Fitbit Charge 3 (dominant hand) and Mi Band 4 (non-dominant hand) | ||||||
| SpecificityXGB: 0.74 (0.07) − SpecificitySVM: 0.77 (0.14) | |||||||
| PrecisionXGB: 0.41 (0.10) − PrecisionSVM: 0.46 (0.11) | |||||||
| Bertrand et al. (61)b | This publication used the same dataset as Bertrand et al. (60). Two tree-based ensemble learning algorithms were used: random forest (RF) and extreme gradient boosting tree (XGB). Compared to Bertrand et al. (60) different resampling techniques were investigated for their performance in detecting eating activities vs. non-eating activities: no resampling (-N), random up-sampling (-U), random down-sampling (-D), and SMOTE resampling (-S).The combination of the two machine learning algorithms and the different resampling techniques resulted in eight classification models that were compared. | 10 (5/5), average age: 32 years | Healthy, non-diabetic | Up to 14 days; in total, 1,361 activity events were collected | Free-living environment ➔ participants used an app (“aTimeLogger”) to log the ground truth of their activities | Mean (standard deviation)d: | CGM: a FreeStyle Libre 2 |
| MCCXGB-N: 0.34 (0.13) − MCCXGB-U: 0.38 (0.12); | |||||||
| F1-scoreXGB-N: 0.33 (0.16) − F1-scoreRF-S: 0.51 (0.10); | Wearables: Mi Band 4 (non-dominant wrist), FitbitCharge 3 (dominant wrist) | ||||||
| SensitivityXGB-N: 0.23 (0.14) − SensitivityXGB-D: 0.67 (0.19); | |||||||
| SpecificityXGB-D: 0.74 (0.07) − SpecificityXGB-N: 0.98 (0.03); | |||||||
| PrecisionXGB-D: 0.41 (0.10) − PrecisionXGB-N: 0.80 (0.21); | |||||||
| AccuracyXGB-D: 0.73 (0.03) − AccuracyXGB-N: 0.82 (0.05). | |||||||
| RF-S was deemed to have the best overall performance of the eight models. | |||||||
| Dassau et al. (62) | Using data from a pilot study, four different detection methods were compared: backward difference (BD), Kalman filter estimation (Kalman), combination of BD and Kalman (BD + Kalman), and second derivative of glucose (G“). Central aim was to reduce FP detections to reduce the risk for erroneous insulin injections in the context of an AP. To do so, a voting algorithm was implemented, using either a two-out-of-three (BD, BD + Kalman, and G”) or three-out-of-four (BD, Kalman, B + Kalman, and G”) scheme to check for concordance in meal detections by the above-mentioned methods. Importantly, insulin meal boluses were purposefully delayed by 1 h and thus did not confound the postprandial BG changes. | 17 (40% girls); 11 ± 4 years, range 4–17 years | T1DM | 17 breakfast meals (one per participant) with an average of 56 g of CHO (range: 22–105 g of CHO), the content of which was decided upon by the participants | Inpatient study | ΔT (min) = Detection time from the onset of the meal; ΔG (mg/dL) = Difference in the glucose level when detection took place minus the preprandial value. | CGM: FreeStyle Navigator (Abbott Diabetes Care, Alameda, CA) with a sampling rate of 1 min |
| Average values for the four different detection methods: ΔTBD = 29 min, ΔGBD = 13 mg/dL; ΔTKalman = 35 min, ΔGKalman = 30 mg/dL; ΔTBD + Kalman = 31 min, ΔGBD + Kalman = 18 mg/dL; ΔTG” = 30 min, ΔGG” = 16 mg/dL. | |||||||
| ΔG using the Kalman algorithm was statistically significantly higher compared with ΔG using the other methods (p < 0.001). | |||||||
| Average values for the different voting schemes: ΔTtwo-out-of-three = 30 min, ΔGtwo-out-of-three = 15 ± 10 mg/dL; ΔTthree-out-of-four = 32 min, ΔG three-out-of-four = 21 ± 9 mg/dL. | |||||||
| Dovc et al. (63) | Double-blind, randomized, two-period crossover study on the safety and efficacy of fully CL insulin therapy/glucose control using two different insulin solutions (faster vs. standard insulin aspart). Atlas et al.’s (59) fuzzy logic-based control algorithm was used (see above). | 20 (9/11); 21.3 ± 2.3 years | T1DM | Two 27-h (1,500 h on the first day to 1,800 h the next day) CL inpatient stays with meals that were unannounced to and thus uncovered by the fully CL device. Standardized and identical meals were given on both study visits. Main meals contained ~1 g of CHO/kg of body mass and snacks about half of this amount. Macronutrient distribution was about 50% CHO, 20% proteins, and 30% fats (<10% saturated fats). Meals were scheduled at: ~1,500 h (snack); 1 h after the end of an exercise protocol between 1,900 and 2,000 h (dinner); 0800 h (breakfast); 1,200 h on the next day (lunch). | Inpatient study | Median time of delivered prandial bolus was 38.4 min (32.7, 55.8) for meals in the faster insulin aspart arm and 30.1 min (26.9, 54.6) in the standard insulin aspart arm (p = 0.388). | CGM: Enlite II sensor (Medtronic Diabetes); CL algorithm: DreaMed GlucoSitter (DreaMed Diabetes, Petah Tikva, Israel) |
| El Fathi et al. (64)e | Preliminary results from a randomized three-way experiment on the safety and efficacy of CL insulin delivery with or without a meal detection module (an adaptive model-based meal detection algorithm) versus conventional pump therapy after a missed insulin bolus. The data stem from the same study as Palisaitis et al. (65). | 4 adolescents | T1DM | Per participant, three 9-h inpatient visits with one uncovered lunch meal with 60 g of CHO per visit were conducted ➔ 4 participants x 3 visits x 9 h = 108 h of data | Inpatient study | Comparison of the incremental AUCs after the missed insulin bolus across the three conditions: conventional pump therapy = reference standard (29.6 ± 6.5 h mmol/l), CL without meal detection: −16% incremental AUC (24.8 ± 11.5 h mmol/l), CL with meal detection: −39% incremental AUC (18.0 ± 2.7 h mmol/l); The collected data were also used to run the meal detection algorithm offline over the 108 h (4 patients × 3 visits × 9 h) of clinical data: 12/12 unannounced meals detected successfully; no FPs; time until meal detection = 35 [30–40] min; glucose increase at meal detection time = 2.89 ± 1.72 mmoL/L | NIA |
| Faccioli et al. (66) [supplemented with information from Anderson et al. (67)] | Data from a multicenter clinical trial on the feasibility of a long-term automated insulin delivery system were used for a retrospective evaluation of a super-twisting-based meal detector. 14 days of SAP therapy under free-living conditions preceded the main phase of the study; these data were used for the evaluation of the meal detector. Due to the use of SAP therapy, manual meal insulin boluses were given, and the results need to be interpreted taking this into consideration. | 30 (17/13); median age 44 years, range 18–66 years | T1DM | 14 days | Patient-reported meal times. Of note, 11/30 participants had <20 registered meals for 14 days. Since some missed meal announcements occurred, the authors only selected portions of data with certain meal information. | All values refer to median (interquartile range): TP = 16 (10); FN = 6 (4); FP = 7 (3); Recall = 70% (13%); Precision = 73% (26%); F1-score = 68% (16%); FP per day = 1.4 (1.4); CHO content related to FNs = 32 g (32 g); detection time = 45 min (45 min) | CGM: DexCom G4 (DexCom, Inc., San Diego, CA, United States); sampling time of 5 min |
| Fushimi et al. (68) [supplemented with information from Sánchez-Peña et al. (69)] | CGM data obtained during a clinical trial were used to evaluate the Automatic Regulation of Glucose (ARG) algorithm with an additional automatic switching signal generator (SSG), i.e., a meal detection module. Importantly, in the clinical trial meal announcement was used, so the results need to be interpreted in light of this potential bias. | 5 (2/3), 43 ± 6 years, range 32–48 years | T1DM | Five standardized meals per participant: one breakfast, one lunch, one afternoon snack, two dinners. Breakfast and afternoon snack: a cup of tea or coffee, two slices of whole-meal bread or five crackers, diet jam, spreadable cheese (≈28 g of CHO). Dinners: whole pasta, lean meat, fresh fruit (≈55 g of CHO). Lunch: same as dinners, but mashed potatoes instead of whole pasta (≈55 g CHO). One meal was excluded due to pump occlusion ➔ 24 eligible meals | Inpatient study | 2 FPs (8.3%); 2 FNs (8.3%); efficiency = 83.3% | CGM: Dexcom G4 sensor (Dexcom, San Diego, CA), sampling rate: 5 min |
| Godoy et al. (70) | A feedback scheme-based meal detection and CHO estimation algorithm was developed and evaluated retrospectively on a clinical dataset. | 11 adults | T1DM | 5 days, whereby the first 3 days were used for identification/calibration and the following 2 days were used for the validation of the proposed model | Free-living data with CGM measurements, insulin pump recordings, participant-recorded CHO estimates, etc. | 184 TPs; 263 TNs; 9 FPs; 2 FNs; 98.92% sensitivity; 96.69% specificity; 97.60% accuracyf; Mean time gap (estimated meal onset time – actual meal onset time) = 9.0 min and 25 min delay time | Insulin pump MiniMed 640G; CGM sampling time = 5 min, but up-sampled to 1 min to increase the detection sensitivity |
| Of note, insulin boluses were used as another algorithm input. The results need to be interpreted considering this. | |||||||
| Hoyos et al. (71) [supplemented with information from Aleppo et al. (72)] | Data from a study assessing the reliability of CGM measurements were used to compare two scenarios: one scenario with the original meal events announced by the participants and one with the meal events generated automatically by the super-twisting-based meal detector introduced in Faccioli et al. (66). “An unsupervised clustering algorithm based on Fuzzy C-Means was applied to classify event-to-event segments of CGM data. Events defining data partitioning were automatically generated based on: (1) an automatic meal detection algorithm (for day periods) and (2) time of day (for night periods).” (p. 576) | 44 adults | T1DM | 26-week study; only participants with an average of 3 to 5 reported meals per day were considered | Free-living data with CGM measurements, insulin pump recordings, etc. | Results (M ± SD) for automatically detected meals: Number of clusters (c*) = 8.09 ± 1.67; Fukuyama-Sugeno index (VFS) = −16,893 ± 5,838; Compactness (Vcom) = 0.236 ± 0.063; Variance (Vvar) = 966 ± 653.4; Time in range = 45.2 ± 15% | CGM: Dexcom G4 Platinum; Sampling time = 5 min |
| There was lower variance in the clusters of the automatically detected meals as compared to the announced meals, thereby underscoring the algorithm’s value. | |||||||
| Kölle et al. (73) | Four different detection methods were retrospectively compared using a clinical dataset: two methods based on the classification of horizons (classification of estimated Rα horizons [LDA Rα] and classification of CGM horizons [LDA CGM], respectively) and two methods based on threshold violations (threshold on current Rα estimate [Threshold] and the previously published Glucose Rate Increase Detector algorithm (79), respectively). Note that often meals were accompanied by insulin boluses, so, again, results need to be interpreted in light of this. | 12 (8/4); 7.3 ± 4.7 years | T1DM | 492 of 849 identified meals were included, whereby the authors focused on meals that would necessitate automatic meal detection (e.g., larger meals); meals were divided into categories of pre-meal, at-meal, post-meal and no insulin bolus | Two experienced diabetologists independently retrospectively identified meals from free-living CGM data and logged information from insulin infusion pumps. | Averages across 10 cross-validated Monte Carlo runs: | CGM: Medtronic Enlite 2 |
| LDA Rα: sensitivity = 0.92; 1.50 FPs/day; time of detection after meal start = 18.59 min | |||||||
| LDA CGM: sensitivity = 0.90; 1.37 FPs/day; time of detection after meal start = 11.78 min | |||||||
| Threshold: sensitivity = 0.64; 1.28 FPs/day; time of detection after meal start = 32.67 min | |||||||
| GRID: sensitivity = 0.21; 2.81 FPs/day; time of detection after meal start = 42.53 min | |||||||
| Mosquera-Lopez et al. (74) | A single-center, randomized crossover trial was conducted to compare postprandial (4 h) glucose control following unannounced meals using a hybrid model predictive control algorithm and a newly developed robust artificial pancreas (RAP) system (i.e., two intervention visits per participant). The RAP system used machine learning for automated meal detection and meal size estimation. CGM and insulin data were used. | 15 (6/9) participants enrolled (age: 37.6 ± 10.4 years), 2 participants withdrew from the study ➔ 13 included in analysis | T1DM | Only the intervention visit involving the RAP system was used to determine meal detection performance; this visit involved a total of 24 participant-chosen study meals with a CHO content of 45-66 g. | Participant confirmation of alerts sent out by the meal detection system; data on mealtime records entered into a cloud-based database by a study investigator. | Sensitivity = 83.3% (95% CI 62.6–95.2%); false discovery rate = 16.6% (95% CI 4.7–37.4%); detection time (M ± SD) = 25.9 ± 0.9 min | CGM: Dexcom G6 (DexCom, Inc., San Diego, CA, United States); Insulin pump: Omnipod (Omnipod Insulet Corporation, Acton, MA, United States) |
| Ornetzeder et al., 2019 (75) [supplemented with information from Zschornack et al. (76)] | Three previously published meal detection algorithms (79–81) were compared by using data from two clinical trials, one with participants with T1DM and one with insulin-treated participants with T2DM. Importantly, in both datasets insulin meal boluses were given, so the results need to be interpreted in light of this. Furthermore, small meals (<20 g of CHO) were treated differently from larger meals in that they did not contribute to FN and FP counts. | 10 ➔ 5 (3/2) participants with T1DM (mean age: 48 years) and 5 (4/1) participants with T2DM (mean age: 65 years); both samples were random subsamples of the respective study samples | T1DM, T2DM | T1DM participants: datasets with 7 days per participant but the first 48 h after the insertion of the CGM sensor were not considered for the performance assessment; T2DM participants: datasets with 2–3 days per participant | T1DM: no further information | T1DM (fixed parameters; all means): ΔTHarvey = 19.1 min; ΔTSamadi = 12.7 min; ΔTRamkissoon = 19.6 min; FP/dayHarvey = 0.6; FP/daySamadi = 0.9; FP/dayRamkissoon = 0.4; SensitivityHarvey = 77.0%; SensitivitySamadi = 73.7%; SensitivityRamkissoon = 75.0% | T1DM: prototype CGM system using a sensor in an early development phase (Roche Diagnostics GmbH, Mannheim, Germany) |
| T2DM: outpatient study | T2DM (fixed parameters; all means): ΔTHarvey = 27.8 min; ΔTSamadi = 24.6 min; ΔTRamkissoon = 26.9 min; FP/dayHarvey = 1.3; FP/daySamadi = 1.3; FP/dayRamkissoon = 0.9; SensitivityHarvey = 70.5%; SensitivitySamadi = 67.9%; SensitivityRamkissoon = 70.7% | ||||||
| T2DM (patient-specific parameters; all means): ΔTHarvey = 30.7 min; ΔTSamadi = 30.5 min; ΔTRamkissoon = 26.1 min; FP/dayHarvey = 1.0; FP/daySamadi = 1.1; FP/dayRamkissoon = 0.4; SensitivityHarvey = 79.9%; SensitivitySamadi = 80.2%; SensitivityRamkissoon = 67.3% | T2DM: CGM sampling rates: 1 min and 5 min for the T1DM and T2DM datasets, respectively | ||||||
| For both trials CGM data and information on the ingested amount of CHO, meal timing, and insulin were recorded | ΔT was defined as the time between meal ingestion and the detection event | ||||||
| Palacios et al. (82) | Data was collected in a cross-over study on post-resistance exercise nutrient timing comparing immediate vs. 3-h delayed post-exercise nutrition. Tree-based machine learning models (random forest model and gradient boosting machines) using a cold-start and a non-cold-start approach were applied, respectively. Six primary variables were used for the machine learning model: glucose, heart rate, physical activity, core temperature, skin temperature, and respiration rate. As physical activity was found to not aid in prediction it was removed and new lag variables were included. | 9 (9/0); 24.3 ± 4.6 years | Healthy, non-diabetic | 48 h; Total daily energy intake had an approximate macronutrient distribution of 52% CHO, 32% fat, 16% protein; meals were consumed at approximately 08:00, 11:20, 16:00, and 18:00 on both days. | 48-h inpatient study in a whole-room calorimeter ➔ exact mealtimes were recorded | (1) area under the receiver operating characteristic curve (AUC-ROC); (2) area under the precision-recall curve (AUC-PR) ➔ cold-start: k = 110 min: AUC-ROC = 0.891; AUC-PR = 0.803; non-cold-start: k = 20 min, AUC-ROC = 0.996; AUC-PR = 0.964 | CGM: iPRO Professional CGM System (Medtronic MiniMed, Inc., Northridge, CA) placed on the abdomen; Equivital Sensor Electronics Module (Equivital I: Hidalgo Ltd., Cambridge, United Kingdom) combined with a heat-sensitive transmitter (pill) to assess heart rate, heat flux, core body temperature; To match the calorimeter data, the Equivital and CGM data were resampled using cubic spline interpolation |
| k = minimal window size | |||||||
| Palisaitis et al. (65)e | Randomized, three-way, crossover trial comparing an AP system equipped with a meal detection algorithm (AP + MDA) with the AP alone and conventional pump therapy (CSII) in controlling blood glucose levels after a meal without accompanying insulin bolus. The data is from the same study as El Fathi et al. (64). | 11; 14.9 ± 1.3 years | T1DM | One self-selected lunch meal which was standardized between interventions for each participant ➔ Mixed meals with 55-65 g of CHO served 4 h after the start of the intervention | Inpatient setting or at home with a member of the research staff with three 9-h interventions from 0800 to 1,700 or 0900 to 1,800 | Median meal detection time in the AP + MDA condition: 40.0 min (interquartile range 40.0–57.5 min) after consumption of the meal; Incremental glucose from the start of the meal until time of meal detection: 2.6 mmol/L [2.4–4.8], and a rate of change of 10.1 [7.3–12.5] mmol/L/h | CGM: Dexcom G5 |
| Popp et al. (83) | Self-reported (SR; using a smartphone app) timing of eating occasions (consumption of foods and beverages >0 kcal) was compared to objective assessment methods, i.e., a wrist-motion-based (WM) classifier using an ActiGraph worn on participants’ dominant wrists and a simulation-based explanation system using CGM. The data come from an ancillary study of a weight loss intervention study. | 31 completers; 62% females; age: 59 ± 11 years | Prediabetes/moderately controlled T2DM | 10 days | Free-living; Date- and time-stamped eating occasions were entered into a smartphone app; herein, we assume that these serve as the ground-truth method. | CGM method found the longest eating window and the largest number of eating occasions per day. | ActiGraph: ActiGraph GT9X-BT (Pensacola, FL, United States). |
| Pearson’s correlations: first eating occasion identified by SR and CGM: r = 0.534, p = 0.01; first eating occasion identified by CGM and WM: r = 0.325, p = 0.004; eating midpoint identified by CGM and WM: r = 0.253, p = 0.03. | |||||||
| Overlap between methods: Tolerance windows of ±0, 5, and 10 min: <40% of eating occasions identified by both WM and CGM; tolerance windows of ±30, 60, 120 min: overlap between SR and CGM: 55 to 80% of eating occasions; overlap between WM and CGM: ~23% regardless of the tolerance window used. | CGM: Abbott Freestyle Libre Pro (Abbott Park, IL, United States) providing 15-min average glucose values | ||||||
| % of meals identified by all methods: Tolerance windows of ±0, 10 and 15 min: no matching meals identified by all methods; tolerance window of 30 min: 4% of SR meals were also detected by CGM and WM; tolerance window of both 60 and 120 min: 7% overlap; overlap of the three methods over 3 days was found in only one participant | |||||||
| Samadi et al. (84) | Retrospective evaluation of meal detection and CHO estimation algorithms using clinical data collected in CL experiments using the integrated multivariable adaptive AP system (IMA-AP). Their approach relies on the qualitative analysis of the glucose trajectory and preceding insulin infusion data to detect disturbances and estimate their magnitude by estimating the amount of ingested CHO. Importantly, in these experiments, no meal announcement-based feedforward meal bolusing was used, so the data do not include manual meal-time insulin boluses. | 11; 18–35 years | T1DM | 117 meals/snacks (7–9 meals and a maximum of 6 snacks per participant) which are distinguished by a CHO threshold of 35 g | NIA | Detection rates (sensitivity): 93.5% (86/92) for meals and 68.0% (17/25) for snacks; FP rate 20.8% (27 FPs and 103 TPs); this equates to 1.05 FPs per day. | CGM: sampling time: 5 min |
| Higher probability of detection with higher CHO contents. | |||||||
| For detected meals and snacks the increase in glucose from consumption until detection is on average 8.8 ± 21.3 mg/dL (in median ± mean absolute deviation [MAD] as 10.0 ± 14.4 mg/dL) for detected meals and snacks | Real-time data on biometric variables: BodyMedia SenseWear armband and Zephyr chest-band (Bioharness-3; Zephyr Technology, Annapolis, MD) | ||||||
| Detection time (time from start of the meal to when the algorithm first reports a CHO estimate): 34.8 ± 22.8 min (in median ± MAD as 30.0 ± 16.0 min). | |||||||
| Turksoy et al. (85) [supplemented with information from Turksoy et al. (86)] | Data obtained during AP trials without meal announcement were used to test a new meal detection approach that requires only CGM data. The meal detection algorithm was meant to be integrated into the integrated multivariable adaptive AP (IMA-AP). | 9 (9/0); mean age 18.3 years | T1DM | 32-h CL sessions were conducted with each participant including breakfast, lunch, dinner and a snack as well as additional snacks if requested by participants; 63 dietary events (50 main meals and 13 snacks) ➔ Foods were selected based on subjects’ personal requirements and there was no limitation on food intake; M = 44 ± 9.38 g of CHO, whereby main meals were higher in CHO than the snacks. | Inpatient study | 61/63 (96.8%) meals/snacks detected successfully; 2 FNs; 1 FP; For the events that were detected successfully the average change in glucose from the start of the meal until the time of the meal detection is 16 ± 9.42 mg/dL. | CGM: Guardian REAL-time CGM (Medtronics, Northridge, CA, United States); sampling time of 5 min, but interpolations used to generate 1-min sampled data |
| Weimer et al. (87) | Evaluation of a physiological parameter-invariant (PAIN)-based meal detector against three established meal-detection algorithms (62, 79) (89) on a clinical dataset. Importantly, participants used insulin therapy during the monitoring period. | 61; 45.7 ± 15.3 years | T1DM | Average duration of monitoring: 17 days | Patient-reported mealtimes (time of inputting meal information into the insulin pump) | Detection rate (i.e., correctly detecting meals within 2 h of the patient-reported mealtime) of the detectors based on operating points that are closest to 2 FPs/day (i.e., relative sensitivity for a standardized specificity): PAIN = 86.9% at 2.01 FPs/day; Dassau et al. = 74.1% at 1.99 FPs/day; Lee & Bequette = 73.4% at 1.99 FPs/day; Harvey et al. = 79.4% at 1.97 FPs/day | CGM: 5-min CGM readings |
| With detection rates ≥ 55% and ≤3.7 FPs/day the PAIN-based detector performs more reliably across individuals than the other approaches (based on n = 53 since participants with <10 reported meals were excluded from this analysis). The other approaches show lower detection rates (on average and worst-case). The other approaches have lower average FP rates but display greater variance and higher FP rates in the worst-case scenarios. |
Characteristics of included publications.
Publications included in the review, ordered by the first author’s name. Some publications also include further information (e.g., in silico validation data, CHO content estimation), which are not discussed here. aOnly performance measures relevant for our research questions are reported herein. bBertrand et al. (60) and Bertrand et al. (61) are reports of the same study. cThe authors of this paper report many different performance metrics for each algorithm; for space-saving reasons we only report the lowest and highest values for each performance metric here. These values refer to the models that included CGM data. dThe authors of this paper report many different performance metrics for each algorithm; for space-saving reasons we only report the lowest and highest means (and the according standard deviations) for each performance metric here. eEl Fathi et al. (64) and Palisaitis et al. (65) are reports of the same study. fWe calculated sensitivity, specificity, and accuracy from the total number of TPs, TNs, FPs, and FNs, respectively, and thus, the values refer to the total sensitivity, specificity, and accuracy across the sample. AP, artificial pancreas; AUC, area under the curve; BG, blood glucose; CGM, continuous glucose monitoring; CHO, carbohydrates; CI, confidence interval; CL, closed-loop; CSII, continuous subcutaneous insulin infusion; FN(s), false negative(s); FP(s), false positive(s); LDA, linear discriminant analysis; M, mean; MCC, Matthew’s correlation coefficient; NIA, no information were available; SAP, sensor-augmented pump; SD, standard deviation; T1DM, type 1 diabetes mellitus; T2DM, type 2 diabetes mellitus; TN(s), true negative(s); TP(s), true positive(s).
Figure 1
PRISMA Flow diagram (88). aIn some cases, more than one reason led to the exclusion of a publication; here, only the primary reason is listed for each publication. bIn two cases, two publications of the same study were included. CGM, continuous glucose monitoring; IA, ingestive activity.
Many of the screened publications were excluded from the present review because they did not include details on the detection of IA but instead focused on measures of glycemic control (e.g., time in specific glucose ranges). Further, some publications were excluded because they included graphic CGM data with IA marked as such but did not provide quantitative data on the detection of IA. Another common reason for exclusion was investigation in silico, often using virtual patients with T1DM.
We were unable to retrieve the full text of one publication despite several efforts to contact the authors directly. This publication was excluded; however, it was considered likely ineligible based on its abstract.
3.1 Study characteristics
The included publications were published between 2008 and 2023 and reported an average sample size of 18.3 (SD = 15.1) participants. Table 1 provides an overview of the included publications and the extracted information. The publications covered a wide age range, including pediatric (62, 73), adolescent (62, 64, 65), and adult (59–61, 63, 66, 68, 70, 71, 74, 75, 82–85, 87) populations. Fourteen publications included participants with T1DM (59, 62–66, 68, 70, 71, 73, 74, 84, 85, 87), one publication included a sample of participants with T1DM or T2DM (75), one publication included participants with prediabetes or moderately controlled T2DM (83) and three publications included participants without diabetes (60, 61, 82). Publications included both controlled/inpatient (59, 62–65, 68, 74, 82, 85) and free-living settings (60, 61, 66, 70, 71, 73, 75, 83, 87).
In several aspects, there was substantial heterogeneity among the included publications. First, the number and type of performance metrics reported for the tested approaches differed substantially. Commonly reported performance metrics included the number of true and/or false positives and/or negatives (including frequencies per day and rates) (64, 66, 68, 70, 73, 75, 84, 85, 87), sensitivity (60, 61, 70, 73–75, 84, 87), specificity (60, 61, 70), accuracy (61, 70), precision (60, 61, 66), F1-score (60, 61, 66), Matthew’s correlation coefficient (60, 61), Pearson’s correlations (83), detection time or time until an insulin bolus was delivered (59, 62–66, 70, 73–75, 84), change in glucose concentrations (62, 65, 84, 85), and area under the curve (82). However, even when the same metrics were reported, their definition was sometimes inconsistent across publications. For instance, the detection window for true positive detections ranged from 60 to 180 min, depending on the publication (66, 73, 75). In addition, the general study setup varied between publications, including differences in the sample composition, the use of meal announcement/meal-accompanying insulin boluses, the ground-truth method used for identifying IA, the devices used, and the scope of the data collection (Table 1).
Table 2 shows the result of the critical appraisal of all included publications. There were some concerns regarding the applied methodology for all publications; these concerns were substantial for most publications. In 16/19 (84.2%) publications, the sample consisted exclusively of individuals with (pre)diabetes (59, 62–66, 68, 70, 71, 73–75, 83–85, 87). Further, 9/19 (47.4%) publications used error-prone methods for measuring the ground truth of IA, mostly self-reported IA (60, 61, 66, 70, 71, 73, 75, 83, 87). Moreover, 7/19 (36.8%) publications used meal announcements and/or meal-accompanying insulin boluses (66, 68, 70, 71, 73, 75, 87). Finally, 8/19 (42.1%) publications utilized other inputs besides CGM, e.g., heart rate or the insulin sensitivity factor (59–61, 63, 64, 70, 74, 82). Overall, 15/19 (78.9%) publications elicited methodological concerns in two or more appraised domains (59–61, 63, 64, 66, 68, 70, 71, 73–75, 83, 84, 87), and all publications had methodological concerns in at least one of the appraised domains (59–66, 68, 70, 71, 73–75, 82–85, 87).
Table 2
| Publication | Error-prone IA ground-truth method | Sample with (pre)diabetes | Meal announcement/insulin boluses | Other inputs in addition to CGM | Overall rating |
|---|---|---|---|---|---|
| Atlas et al. (59) | x | ✓ | x | ✓ | ★★☆☆ |
| Bertrand et al. (60) | ✓ | x | x | ✓ | ★★☆☆ |
| Bertrand et al. (61) | ✓ | x | x | ✓ | ★★☆☆ |
| Dassau et al. (62) | x | ✓ | x | x | ★★★☆ |
| Dovc et al. (63) | x | ✓ | x | ✓ | ★★☆☆ |
| El Fathi et al. (64) | x | ✓ | x | ✓ | ★★☆☆ |
| Faccioli et al. (66) | ✓ | ✓ | ✓ | x | ★☆☆☆ |
| Fushimi et al. (68) | x | ✓ | ✓ | x | ★★☆☆ |
| Godoy et al. (70) | ✓ | ✓ | ✓ | ✓ | ☆☆☆☆ |
| Hoyos et al. (71) | ✓ | ✓ | ✓ | x | ★☆☆☆ |
| Kölle et al. (73) | ✓ | ✓ | ✓ | x | ★☆☆☆ |
| Mosquera-Lopez et al. (74) | x | ✓ | x | ✓ | ★★☆☆ |
| Ornetzeder et al. (75) | ✓ | ✓ | ✓ | x | ★☆☆☆ |
| Palacios et al. (82) | x | x | x | ✓ | ★★★☆ |
| Palisaitis et al. (65) | x | ✓ | x | xa | ★★★☆ |
| Popp et al.(83) | ✓b | ✓ | x | x | ★★☆☆ |
| Samadi et al.(84) | ? | ✓ | x | x | ★★☆☆ |
| Turksoy et al. (85) | x | ✓ | x | x | ★★★☆ |
| Weimer et al. (87) | ✓ | ✓ | ✓ | x | ★☆☆☆ |
Critical appraisal of included publications.
x, no; ✓, yes;?, no information available; CGM, continuous glucose monitoring; IA, ingestive activity. “x” in all columns would signal minimal methodological concerns and thus an overall rating of four stars in the rightmost column; the more “✓,” the greater the methodological concerns, and consequently, the fewer stars are awarded in the rightmost column. Note that for the overall rating, no available information (“?”) was treated as eliciting methodological concerns (“✓”). aWhile insulin pump data were also used in this paper, the meal detection algorithm relies only on glucose data as input. bThe authors state that self-reported IA is not seen as the ground-truth method in their work; however, in the absence of direct observation of IA in this study, herein, we assume self-reported IA as the ground-truth.
3.2 Overview of detection approaches
Our review identified a wide range of methods to automatically detect IA, including fuzzy logic (59, 63), model predictive control (74), support vector machine (60), random forest (60, 61, 82), (extreme) gradient boosting trees (60, 61, 82), backward difference (62), Kalman filter estimation (62), second derivative of glucose (62), Kalman filters (64, 68), switching signal generator (68), simulation-based explanation (83), classification of horizons (73), analysis of the glucose trajectory (84), pattern recognition using linear discriminant analysis (73), and threshold violation-based approaches (73). Further, adaptive model-based (64), super-twisting-based (66, 71), feedback scheme-based (70), and physiological parameter-invariant-based (87) meal detection approaches were applied.
The reviewed approaches used different inputs to automatically detect IA. As summarized in Table 2, some methods relied solely on CGM as an input (62, 65, 66, 68, 71, 73, 75, 83–85, 87). Others also included data from insulin treatment or other sensor systems (e.g., accelerometry, photoplethysmography, temperature sensors; see Table 1) (59–61, 63, 64, 70, 74, 82).
3.3 Performance of the approaches
We identified several CGM-based approaches for the automatic detection of IA that achieved high values in the respective performance metrics (Table 1). However, the substantial heterogeneity in the applied methodology and reporting of results needs to be considered.
For instance, Godoy and colleagues achieved 98.9% sensitivity, 96.7% specificity, and 97.6% accuracy with their feedback scheme-based algorithm (70). Notably, the algorithm uses certain patient-specific parameters, such as the insulin sensitivity factor derived from participants’ usual diabetes treatment (70). Similarly, the algorithm by El Fathi et al. successfully detected 12/12 meals without any false positives and a detection time of 35 min (64). In two publications using the same dataset, Bertrand et al. investigated several IA detection approaches in individuals without diabetes (60, 61). A range of performance metrics is reported in both publications. In the first publication, the highest achieved mean sensitivity was 66.8%, and the highest achieved mean specificity was 77.3%, for example (60). In the second publication, the highest achieved mean sensitivity was 66.8%, and the highest achieved mean specificity was 97.7% (61). Importantly, in both publications, the models did not exclusively rely on CGM as input (60, 61). Similarly, Palacios et al. had a sample of individuals without diabetes (82). However, their models, too, did not exclusively rely on CGM as input, but also utilized other physiological variables such as heart rate and skin temperature (82). Palacios et al. reported the area under the receiver operating characteristic curve (AUC-ROC) and the area under the precision-recall curve (AUC-PR) (82). For cold-start cases with a window size of k = 110 min, they reported an AUC-ROC of 89.1% and an AUC-PR value of 80.3% (82). For non-cold-start cases and k = 20 min, the AUC-ROC was 99.6%, and the AUC-PR was 96.4% (82).
The performance of CGM-only approaches, which hold particularly great value for practical applications, varied substantially. Sensitivities varied between 20.8% (73) and 96.8% (85). Where reported, average false positive detections ranged from 0.4 (75) to 2.8 (73) IA events per day. Selected publications further reported a false positive rate of 20.8% (84), a false discovery rate of 16.6% (74), and a median precision value of 73.0% (66). Moreover, publications reported detection times between 11.8 min (mean) (73) and 45.0 min (median) (66). Importantly, all CGM-only approaches were tested on samples consisting exclusively of individuals with (pre)diabetes, some of which also used meal announcements/insulin boluses. A detailed description of the performance metrics for each of the included publications is provided in Table 1.
3.4 Detection times
The detection time is the relevant metric to evaluate whether the identified CGM-based IA detection approaches could be used in the context of JITAIs.
A detection time measure was reported in 11/19 (57.9%) publications (59, 62–66, 70, 73–75, 84). The detection time was commonly defined as the time between the start of the IA (i.e., typically a meal) and its (automatic) detection by the CGM-based approach. Mean (59, 62, 64, 70, 73–75, 84) and median detection times (65, 66, 84) were reported, thus impeding direct comparisons. One publication reported the median time of the delivered prandial insulin boluses (63).
Overall, the reported detection times varied between 9.0 min (mean) (70) and 45.0 min (median) (66), with most values falling into the 20-to-40-min range (Table 1).
4 Discussion
The primary objective of this review was to examine whether CGM can be used to automatically detect IA in (near-)real time. In sum, there are various promising approaches that show satisfactory to excellent performance on measures such as sensitivity and specificity. However, the performance of CGM-based methods for automatically detecting IA varies. Similarly, detection times vary, but currently, they appear too long to administer JITAIs for acutely altering IA. Methodological issues and overall heterogeneity among publications make it difficult to recommend the best-performing approach.
4.1 Which approaches using CGM for the automatic detection of IA in (near-)real-time have been investigated, and have these approaches relied solely on CGM or also used other data (e.g., sensors/wearables)?
Our results indicate that both CGM-only approaches and those supplemented with other input data (e.g., accelerometry, photoplethysmography, temperature sensors) have been tested. Moreover, various algorithms have been used to detect IA. Since approaches using different sensor modalities and/or programming methods were successful at automatically detecting IA, it is evident that various solutions can be used for automated, CGM-based IA detection.
4.2 How accurate are these approaches in detecting IA?
Our review showed that the performance evaluation of any single approach depends on the respective case and priorities. For example, if the goal is to combine a CGM-based approach with smartphone prompts to enable comprehensive diet logs, the method should have high sensitivity to avoid missing a potential IA (false negative). In this case, specificity would only play a minor role as nothing is lost by sending a prompt in response to a false positive detection – the prompt can remain unanswered by the patient/participant. In contrast, when the goal is to use the CGM-based approach as a stand-alone IA assessment tool, high specificity would be critical to avoid artificial inflation of the number of daily meals, for example. Thus, a single best approach for all scenarios could not be identified. The substantial heterogeneity of the applied methods and reporting of results, including the broad range of the number and type of reported performance metrics and their varying definitions, made it difficult to compare the performance of the different approaches.
However, collectively, our results demonstrate that there are indeed several relatively well-performing CGM-based approaches for the automatic detection of IA. One example is the feedback scheme-based algorithm by Godoy et al., which achieved near-perfect sensitivity, specificity, and accuracy (70). However, this algorithm relies on several patient-specific parameters as input that are derived from participants’ usual diabetes treatment (70). Thus, it remains to be determined whether this approach could be adapted to work equally well in individuals without diabetes, for whom these data are not routinely assessed. Similarly, methodological issues further limiting studies’ internal and/or external validity pertain to using meal announcements or insulin boluses and focusing on samples with diabetes in the reviewed studies. All included publications suffered at least one such methodological limitation (Table 2).
Several reviewed articles reported solutions that relied solely on CGM as input for their IA detection algorithms. Performance among these approaches varied, but sensitivities ≥90% were achieved by several groups (73, 84, 85), and false positive occurrences < 1 per day were reported (75). This suggests that inputs other than CGM are not necessary to achieve excellent performance in automatically detecting IA.
Of note, some algorithms that incorporated inputs other than CGM might also work with CGM as their only input for the specific goal of IA detection. For example, in Bertrand et al.’s machine-learning algorithms, data from two wrist-worn activity trackers were incorporated in addition to the CGM data (61). However, the 20 most important features were derived from the CGM data (61). Hence, it is likely that an adaptation of their algorithm that relies exclusively on the CGM data as input might also achieve good – albeit likely somewhat worse – IA detection performance. Similar cases can be made for other publications in which insulin data were used as input in addition to the CGM data (59, 63, 64). These results suggest that it is possible to automatically detect IA using CGM-based and even CGM-only algorithms.
4.3 Can these approaches be used in the context of JITAIs?
Generally, to successfully administer a dietary JITAI, IA must be detected in (near-)real-time. However, precisely how short the detection would have to be depends on the specific goal, as outlined before. Detection times as fast as 9.0 min were reported (70), but most approaches needed 20 to 40 min to detect IA (Table 1). This can generate feedback on IA much faster than traditional dietary assessment methods, such as 24-h recalls, thus creating opportunities for earlier intervention. For instance, detecting deviations from a standardized study procedure (e.g., when IA is detected in a fasting window) is likely possible. Further, when deviations from a specific meal plan (e.g., low carbohydrate) are detected, the plan could be adjusted for the subsequent meals on the same day. Automated meal detection could further trigger behavioral intervention prompts regarding portion size and eating rate (i.e., reminders to eat more slowly) for future meals. However, in most cases, detection times are too long to modify/influence IA truly in the moment it occurs (e.g., a participant on a ketogenic diet has likely already finished a carbohydrate-rich meal by the time it is detected). Regardless, it is debatable if that is really the goal and what intervening during an eating event would look like.
4.4 Implications for clinical and research practice
Our review shows that several CGM-based options for the automatic detection of IA exist. Ultimately, the specific use case will dictate the most suitable approach. Different approaches might be appropriate depending on factors such as the budget, population, targeted level of wearing comfort, and goal of the automatic IA detection.
Notably, other innovative methods for the automatic detection of IA, such as those using wearable-, sensor-, and image-based methods (9, 15–18), are also promising. These methods may even be superior to CGM-based approaches regarding detection times. Wang and colleagues identified several devices that can quickly detect IA (16), such as a headband device that can detect eating events via chewing sounds within only 3 min (16, 90). Similarly, a pilot study by Kumar and colleagues investigating the use of abdominal sounds to detect IA found an average detection time of only 4.3 min (91). It has even been demonstrated that eating events can be predicted ahead of time (16, 92). Yang et al. used a camera, a GPS device, and an accelerometer to predict eating and food-purchasing events up to 4 min in advance (92). The authors found that a trained gradient-boosting model achieved a mean accuracy of 72.9% in predicting eating events 0–4 min in advance (92). This highlights that different methodologies might have inherent strengths and limitations. The suboptimal detection times might be considered an inherent limitation of CGM-based approaches. Recent advances have tried to solve the CGM-inherent lag time issue (93), but more research is needed. It remains to be seen whether these limitations inherent to using CGM for automatically detecting IA can be overcome. On the other hand, one key benefit of using CGM might be its unobtrusiveness, which could facilitate its acceptance in practice. This unobtrusiveness contrasts many other, more obtrusive approaches such as glasses and camera-based methods (9, 16, 18).
A promising prospect might be to use a sensitive CGM-based approach that sends a prompt to the patient/participant asking them to log IA in case of a true positive detection. Thus, the CGM-based approach would serve as an automated reminder. That way, a false positive detection does not automatically lead to erroneous IA information but needs to be verified by the person. In this context, the suboptimal detection times also likely would be acceptable.
4.5 Limitations and directions for future research
4.5.1 Sample characteristics
Unsurprisingly, most publications included samples with diabetes, as the primary use case for automated IA detection is AP systems. However, to examine the potential of CGM-based approaches for detecting IA in various populations, more research in more diverse populations, including healthy individuals, should be conducted. This is particularly important as the generalizability of previous findings to non-diabetic individuals is likely limited, for instance, due to the usually far lower variations in blood glucose levels in persons without diabetes (77) as compared to persons with diabetes (78). Thus, there may likely be systematic differences in the performance of such approaches in individuals with diabetes compared to those without diabetes. Moreover, in many studies, meals were announced to the system, and/or manual insulin boluses accompanied the registered meals. For example, Ornetzeder and colleagues evaluated the detection performance of three previously published algorithms (79–81) using meals accompanied by insulin boluses (75). While the resulting performance metrics of this publication and similar others are promising, they need to be interpreted considering the applied insulin boluses. Ornetzeder et al. argue that this potential distortion was deemed acceptable due to a lack of alternative, insulin bolus-free datasets and the time it takes for the administered insulin to achieve its peak action (75). However, it is still possible that the results of CGM-based IA detection approaches might differ in scenarios without exogenous insulin infusions. Specifically, the administered insulin might flatten the blood glucose excursions from the meal’s start, making its automatic detection less likely. In line with this, Faccioli et al. state that some of their false negatives might have been related to the attenuated postprandial CGM curves following the administration of meal-accompanying insulin therapy (66). At the same time, it should be considered that the postprandial glucose excursions of individuals with insulin-dependent diabetes would be much more pronounced without insulin treatment than in non-diabetic individuals (82). As such, it could be argued that by administering meal boluses, the postprandial glucose excursions of individuals with diabetes more closely approximate those of individuals without diabetes. Direct evidence is, of course, still necessary to increase confidence in any conclusions. Thus, future studies should ultimately enroll more individuals without diabetes.
4.5.2 Research focus
Moreover, it also needs to be considered that for the initialization of closed-loop systems, background information (e.g., treatment management, physical characteristics of the patient) is typically provided to the system (59). This information may only sometimes be readily available in other contexts. In addition, the goals of algorithms geared toward use in closed-loop/AP systems might differ from approaches aimed at the use for automatic detection of IA in general. For instance, in their AP-oriented work, Kölle et al. focused on glucose excursions caused by larger meals because smaller meals or snacks, which do not cause a substantial increase in blood glucose levels, do not necessarily need to be detected and trigger an insulin bolus to ensure adequate glucose control (73). Yet, in a scenario where the automatic detection of IA via CGM is meant to provide information on any IA – irrespective of its size – this argument does not hold up. This example highlights the potential differences in the setup of algorithms depending on the goal.
Taken together, fundamentally different circumstances and goals may be pursued, and thus, algorithms may be constructed differently, depending on the research question. Consequently, it might be possible to further optimize algorithms to automatically detect IA in research or clinical settings other than closed-loop/AP systems.
4.5.3 Comparability of approaches
There was substantial heterogeneity in how the performance of the investigated approaches was evaluated across the reviewed publications. Thus, as noted by others (66), a direct comparison between the approaches is difficult due to differences in the utilized datasets, preprocessing, and evaluation methods. Differences like these ultimately hamper the search for the best-performing approaches. Performance metrics reported in publications should include at least the following measures: the number of true positives, false positives, and false negatives, which can be used to calculate important metrics such as sensitivity and precision; the detection time, defined as the time from the start of the IA to the time the algorithm detects the IA, whose reporting allows researchers and practitioners to judge whether a specific approach could be used to administer JITAIs, for example. A short detection time followed by a prompt could also allow for more immediate self-reported IA. More accurate self-reports could be the consequence due to diminished recall bias.
4.5.4 Future avenues
In general, more research should be dedicated to using CGM for the specific goal of automatically detecting IA in a broad range of populations, particularly in individuals without diabetes. Such approaches have several potential benefits, but prior research has mainly focused on using CGM for diabetes care and AP systems. However, as explained, algorithms will likely be constructed differently for the specific goal of automatically detecting IA. Moreover, previous findings will have to be replicated and extended in non-diabetic samples to overcome the currently limited generalizability.
Depending on the use case, several advancements would be necessary to rely exclusively on a CGM-based/CGM-only approach for the remote monitoring of IA. To fully automate the logging of IA times in a reliable manner, most systems would have to be even more accurate than they currently are.
If the goal is to further automate IA timing and effectively log macronutrient intake, approaches would have to incorporate specific algorithms for this task. Several publications explored whether estimating macronutrients from CGM data is possible. For instance, Samadi et al. estimated the carbohydrate content of meals (84). Results were promising, with 64.1% of the detected IA events having an absolute carbohydrate estimation error of less than 25 g (84).
Similarly, if the goal is to administer JITAIs to impact acute IA, detection times would have to decrease further. However, as mentioned above, the lag time-caused suboptimal detection times might have to be considered an inherent limitation of CGM-based approaches. Only if future studies succeed at further reducing detection times will the application of CGM-based approaches for dietary JITAIs aiming to alter IA in the moment in the truest sense of the word become possible. This is especially true for cases in which meals are followed by only small and/or delayed postprandial glucose excursions (e.g., after high-fat meals) or when meals contain only a small amount of carbohydrates, as (timely) detection appears difficult here (66, 74). It would also be necessary to explicitly test the detection performance in cases of such challenging IA (e.g., ketogenic diets). Empirical data on such cases might enable the prediction of in which settings CGM-based approaches can be used for successfully detecting IA (e.g., only in contexts where at least moderate amounts of carbohydrates are consumed).
We advise that future studies use different approaches on the same dataset, providing comprehensive CGM and objective IA data, and then compare their performance using the abovementioned metrics. A starting point could be to compare the CGM-only approaches highlighted in Table 2. Such a fair and standardized comparison could further illuminate the currently most promising approach(es).
While CGM-only approaches are highly attractive because they only necessitate one single sensor (i.e., the CGM), multi-sensor solutions also hold great potential and should thus be further investigated. Specifically, combining the strengths of different sensors (e.g., CGM and wristbands) may yield superior results as compared to relying on only one sensor, although this remains to be determined empirically.
Lastly, similar to others (10), we strongly advise that researchers use interdisciplinary collaborations to develop new CGM-based dietary monitoring tools to combine technological and biological/nutritional expertise. Interdisciplinary collaborations should ensure that the resulting tools are useful and optimized from both perspectives.
4.6 Conclusion
Based on an exhaustive and systematic literature search, this scoping review shows that it is possible to automatically detect IA using CGM-based approaches. Despite methodological issues and substantial overall heterogeneity among publications, CGM-based dietary monitoring might complement clinical and research practice.
Statements
Data availability statement
The original contributions presented in the study are included in the article/supplementary material, further inquiries can be directed to the corresponding author.
Author contributions
JB: Conceptualization, Methodology, Writing – original draft, Writing – review & editing, Visualization. CG: Methodology, Writing – review & editing. SHF: Methodology, Writing – review & editing. KK: Conceptualization, Writing – review & editing. CH: Conceptualization, Methodology, Writing – original draft, Writing – review & editing.
Funding
The author(s) declare financial support was received for the research, authorship, and/or publication of this article. This work was funded internally by the Technical University of Munich. CH was supported by a Research Fellowship from the Alexander von Humboldt Foundation. These funding sources had no role in the design, execution, analyses, interpretation of the data, or decision to submit results. Where reported, most reviewed publications received funding primarily from national agencies. Some publications have also reported funding from non-profit organizations and industry partners.
Acknowledgments
The authors want to thank the authors of Bertrand et al. (60, 61), El Fathi et al. (64), and Palisaitis et al. (65) for kindly providing additional information that helped with the interpretation of their results.
Conflict of interest
The authors declare that the research was conducted without any commercial or financial relationships that could be construed as a potential conflict of interest.
CH and KK were an editorial board member of Frontiers, at the time of submission. This had no impact on the peer review process and the final decision.
Publisher’s note
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.
References
1.
AfshinASurPJFayKACornabyLFerraraGSalamaJSet al. Health effects of dietary risks in 195 countries, 1990–2017: a systematic analysis for the global burden of disease study 2017. Lancet. (2019) 393:1958–72. doi: 10.1016/S0140-6736(19)30041-8
2.
CenaHCalderPC. Defining a healthy diet: evidence for the role of contemporary dietary patterns in health and disease. Nutrients. (2020) 12:334. doi: 10.3390/nu12020334
3.
BarbareskoJLellmannAWSchmidtALehmannAAminiAMEgertSet al. Dietary factors and neurodegenerative disorders: an umbrella review of Meta-analyses of prospective studies. Adv Nutr. (2020) 11:1161–73. doi: 10.1093/advances/nmaa053
4.
NeuenschwanderMBallonAWeberKSNoratTAuneDSchwingshacklLet al. Role of diet in type 2 diabetes incidence: umbrella review of meta-analyses of prospective observational studies. BMJ. (2019) 366:l2368. doi: 10.1136/bmj.l2368
5.
PapadimitriouNMarkozannesGKanellopoulouACritselisEAlhardanSKarafousiaVet al. An umbrella review of the evidence associating diet and cancer risk at 11 anatomical sites. Nat Commun. (2021) 12:4579. doi: 10.1038/s41467-021-24861-8
6.
WillettWCStampferMJ. Current evidence on healthy eating. Annu Rev Public Health. (2013) 34:77–95. doi: 10.1146/annurev-publhealth-031811-124646
7.
EnglishLKArdJDBaileyRLBatesMBazzanoLABousheyCJet al. Evaluation of dietary patterns and all-cause mortality: a systematic review. JAMA Netw Open. (2021) 4:e2122277. doi: 10.1001/jamanetworkopen.2021.22277
8.
SchwingshacklLSchwedhelmCHoffmannGLampousiAMKnüppelSIqbalKet al. Food groups and risk of all-cause mortality: a systematic review and meta-analysis of prospective studies. Am J Clin Nutr. (2017) 105:1462–73. doi: 10.3945/ajcn.117.153148
9.
HöchsmannCMartinCK. Review of the validity and feasibility of image-assisted methods for dietary assessment. Int J Obes. (2020) 44:2358–71. doi: 10.1038/s41366-020-00693-2
10.
KlurfeldDMHeklerEBNebekerCPatrickKKhooCSH. Technology innovations in dietary intake and physical activity assessment: challenges and recommendations for future directions. Am J Prev Med. (2018) 55:e117–22. doi: 10.1016/j.amepre.2018.06.013
11.
McClungHLPtomeyLTShookRPAggarwalAGorczycaAMSazonovESet al. Dietary intake and physical activity assessment: current tools, techniques, and Technologies for use in adult populations. Am J Prev Med. (2018) 55:e93–e104. doi: 10.1016/j.amepre.2018.06.011
12.
RavelliMNSchoellerDA. An objective measure of energy intake using the principle of energy balance. Int J Obes. (2021) 45:725–32. doi: 10.1038/s41366-021-00738-0
13.
WebbGP. Methods of nutritional assessment and surveillance In: Nutrition: Maintaining and improving health. 5th. Boca Raton, FL: CRC Press (2019). 47–87.
14.
WongWWRobertsSBRacetteSBdasSKRedmanLMRochonJet al. The doubly labeled water method produces highly reproducible longitudinal results in nutrition studies. J Nutr. (2014) 144:777–83. doi: 10.3945/jn.113.187823
15.
BellBMAlamRAlshurafaNThomazEMondolASde la HayeKet al. Automatic, wearable-based, in-field eating detection approaches for public health research: a scoping review. NPJ Digit Med. (2020) 3:38. doi: 10.1038/s41746-020-0246-2
16.
WangLAllman-FarinelliMYangJATaylorJCGemmingLHeklerEet al. Enhancing nutrition care through real-time, sensor-based capture of eating occasions: a scoping review. Front Nutr. (2022) 9:852984. doi: 10.3389/fnut.2022.852984
17.
HassannejadHMatrellaGCiampoliniPDe MunariIMordoniniMCagnoniS. Automatic diet monitoring: a review of computer vision and wearable sensor-based methods. Int J Food Sci Nutr. (2017) 68:656–70. doi: 10.1080/09637486.2017.1283683
18.
VuTLinFAlshurafaNXuW. Wearable food intake monitoring technologies: a comprehensive review. Computers. (2017) 6:4. doi: 10.3390/computers6010004
19.
BellidoVAguileraECardona-HernandezRDiaz-SotoGGonzález Pérez de VillarNPicón-CésarMJet al. Expert recommendations for using time-in-range and other continuous glucose monitoring metrics to achieve patient-centered glycemic control in people with diabetes. J Diabetes Sci Technol. (2023) 17:1326–36. doi: 10.1177/19322968221088601
20.
HolzerRBlochWBrinkmannC. Continuous glucose monitoring in healthy adults—possible applications in health care, wellness, and sports. Sensors. (2022) 22:2030. doi: 10.3390/s22052030
21.
ZhengMNiBKleinbergS. Automated meal detection from continuous glucose monitor data through simulation and explanation. J Am Med Inform Assoc. (2019) 26:1592–9. doi: 10.1093/jamia/ocz159
22.
LalRAEkhlaspourLHoodKBuckinghamB. Realizing a closed-loop (artificial pancreas) system for the treatment of type 1 diabetes. Endocr Rev. (2019) 40:1521–46. doi: 10.1210/er.2018-00174
23.
American Diabetes Association Professional Practice Committee. 6. Glycemic targets: standards of medical Care in Diabetes—2022. Diabetes Care. (2022) 45:S83–96. doi: 10.2337/dc22-S006
24.
AleppoGWebbK. Continuous glucose monitoring integration in clinical practice: a stepped guide to data review and interpretation. J Diabetes Sci Technol. (2019) 13:664–73. doi: 10.1177/1932296818813581
25.
PeaseALoCEarnestAKiriakovaVLiewDZoungasS. The efficacy of Technology in Type 1 diabetes: a systematic review, network Meta-analysis, and narrative synthesis. Diabetes Technol Ther. (2020) 22:411–21. doi: 10.1089/dia.2019.0417
26.
RenardE. Automated insulin delivery systems: from early research to routine care of type 1 diabetes. Acta Diabetol. (2022) 60:151–61. doi: 10.1007/s00592-022-01929-5
27.
IshiharaKUchiyamaNKizakiSMoriENonakaTOnedaH. Application of continuous glucose monitoring for assessment of individual carbohydrate requirement during ultramarathon race. Nutrients. (2020) 12:1121. doi: 10.3390/nu12041121
28.
AlvaSBaileyTBrazgRBudimanESCastorinoKChristiansenMPet al. Accuracy of a 14-day factory-calibrated continuous glucose monitoring system with advanced algorithm in pediatric and adult population with diabetes. J Diabetes Sci Technol. (2022) 16:70–7. doi: 10.1177/1932296820958754
29.
FreckmannGLinkMKameckeUHaugCBaumgartnerBWeitgasserR. Performance and usability of three Systems for Continuous Glucose Monitoring in direct comparison. J Diabetes Sci Technol. (2019) 13:890–8. doi: 10.1177/1932296819826965
30.
SchembreSMLiaoYRobertsonMCDuntonGFKerrJHaffeyMEet al. Just-in-time feedback in diet and physical activity interventions: systematic review and practical design framework. J Med Internet Res. (2018) 20:e106. doi: 10.2196/jmir.8701
31.
FormanEMGoldsteinSPZhangFEvansBCManasseSMButrynMLet al. OnTrack: development and feasibility of a smartphone app designed to predict and prevent dietary lapses. Transl Behav Med. (2019) 9:236–45. doi: 10.1093/tbm/iby016
32.
American Diabetes Association. Postprandial Blood Glucose. Diabetes Care. (2001) 24:775–8. doi: 10.2337/diacare.24.4.775
33.
BellKJSmartCESteilGMBrand-MillerJCKingBWolpertHA. Impact of fat, protein, and glycemic index on postprandial glucose control in type 1 diabetes: implications for intensive diabetes Management in the Continuous Glucose Monitoring era. Diabetes Care. (2015) 38:1008–15. doi: 10.2337/dc15-0100
34.
GingrasVBonatoLMessierVRoy-FlemingASmaouiMRLadouceurMet al. Impact of macronutrient content of meals on postprandial glucose control in the context of closed-loop insulin delivery: A randomized cross-over study. Diabetes Obes Metab. (2018) 20:2695–9. doi: 10.1111/dom.13445
35.
FurthnerDLukasASchneiderAMMörwaldKMaruszczakKGombosPet al. The role of protein and fat intake on insulin therapy in Glycaemic control of Paediatric type 1 diabetes: a systematic review and research gaps. Nutrients. (2021) 13:3558. doi: 10.3390/nu13103558
36.
PinskerJELeeJBDassauESeborgDEBradleyPKGondhalekarRet al. Randomized crossover comparison of personalized MPC and PID control algorithms for the artificial pancreas. Diabetes Care. (2016) 39:1135–42. doi: 10.2337/dc15-2344
37.
ShiloSGodnevaARachmielMKoremTKolobkovDKaradyTet al. Prediction of personal glycemic responses to food for individuals with type 1 diabetes through integration of clinical and microbial data. Diabetes Care. (2022) 45:502–11. doi: 10.2337/dc21-1048
38.
ReinMBen-YacovOGodnevaAShiloSZmoraNKolobkovDet al. Effects of personalized diets by prediction of glycemic responses on glycemic control and metabolic health in newly diagnosed T2DM: a randomized dietary intervention pilot trial. BMC Med. (2022) 20:56. doi: 10.1186/s12916-022-02254-y
39.
RossettiPAmpudia-BlascoFJLagunaARevertAVehìJAscasoJFet al. Evaluation of a novel continuous glucose monitoring-based method for mealtime insulin dosing--the iBolus--in subjects with type 1 diabetes using continuous subcutaneous insulin infusion therapy: a randomized controlled trial. Diabetes Technol Ther. (2012) 14:1043–52. doi: 10.1089/dia.2012.0145
40.
HallKDGuoJCourvilleABBoringJBrychtaRChenKYet al. Effect of a plant-based, low-fat diet versus an animal-based, ketogenic diet on ad libitum energy intake. Nat Med. (2021) 27:344–53. doi: 10.1038/s41591-020-01209-1
41.
ZaharievaDPTurksoyKMcGaughSMPooniRVienneauTLyTet al. Lag time remains with newer real-time continuous glucose monitoring technology during aerobic exercise in adults living with type 1 diabetes. Diabetes Technol Ther. (2019) 21:313–21. doi: 10.1089/dia.2018.0364
42.
CobelliCRenardEKovatchevB. Artificial pancreas: past, present, future. Diabetes. (2011) 60:2672–82. doi: 10.2337/db11-0654
43.
ArmarioAMartiOMolinaTde PabloJValdesM. Acute stress markers in humans: response of plasma glucose, cortisol and prolactin to two examinations differing in the anxiety they provoke. Psychoneuroendocrinology. (1996) 21:17–24. doi: 10.1016/0306-4530(95)00048-8
44.
García-GarcíaFKumareswaranKHovorkaRHernandoME. Quantifying the acute changes in glucose with exercise in type 1 diabetes: a systematic review and Meta-analysis. Sports Med. (2015) 45:587–99. doi: 10.1007/s40279-015-0302-2
45.
ParkSHYaoJChuaXHChandranSRGardnerDSLKhooCMet al. Diet and physical activity as determinants of continuously measured glucose levels in persons at high risk of type 2 diabetes. Nutrients. (2022) 14:366. doi: 10.3390/nu14020366
46.
BennetsenSLFeineisCSLegaardGELyngbækMPPKarstoftKRied-LarsenM. The impact of physical activity on glycemic variability assessed by continuous glucose monitoring in patients with type 2 diabetes mellitus: a systematic review. Front Endocrinol. (2020) 11:486. doi: 10.3389/fendo.2020.00486
47.
ColbergSRSigalRJYardleyJERiddellMCDunstanDWDempseyPCet al. Physical activity/exercise and diabetes: a position statement of the American Diabetes Association. Diabetes Care. (2016) 39:2065–79. doi: 10.2337/dc16-1728
48.
WiesliPSchmidCKerwerONigg-KochCKlaghoferRSeifertBet al. Acute psychological stress affects glucose concentrations in patients with type 1 diabetes following food intake but not in the fasting state. Diabetes Care. (2005) 28:1910–5. doi: 10.2337/diacare.28.8.1910
49.
Gonder-FrederickLAGrabmanJHKovatchevBBrownSAPatekSBasuAet al. Is psychological stress a factor for incorporation into future closed-loop systems?J Diabetes Sci Technol. (2016) 10:640–6. doi: 10.1177/1932296816635199
50.
SapolskyRMRomeroLMMunckAU. How do glucocorticoids influence stress responses? Integrating permissive, suppressive, stimulatory, and preparative actions. Endocr Rev. (2000) 21:55–89. doi: 10.1210/edrv.21.1.0389 PMID:
51.
BequetteBW. Challenges and recent Progress in the development of a closed-loop artificial pancreas. Annu Rev Control. (2012) 36:255–66. doi: 10.1016/j.arcontrol.2012.09.007
52.
ColmegnaPGarelliFDe BattistaHSánchez-PeñaR. Automatic regulatory control in type 1 diabetes without carbohydrate counting. Control Eng Pract. (2018) 74:22–32. doi: 10.1016/j.conengprac.2018.02.003
53.
ReddyRWittenbergACastleJRel YoussefJWinters-StoneKGillinghamMet al. Effect of aerobic and resistance exercise on glycemic control in adults with type 1 diabetes. Can J Diabetes. (2019) 43:406–414.e1. doi: 10.1016/j.jcjd.2018.08.193
54.
CinarATurksoyK. Multivariable control of glucose concentration In: Advances in artificial pancreas systems. Cham: Springer International Publishing (2018). 65–82.
55.
MaahsDMDeSalvoDPyleLLyTMesserLClintonPet al. Effect of acetaminophen on CGM glucose in an outpatient setting. Diabetes Care. (2015) 38:e158–9. doi: 10.2337/dc15-1096
56.
TriccoACLillieEZarinWO'BrienKKColquhounHLevacDet al. PRISMA extension for scoping reviews (PRISMA-ScR): checklist and explanation. Ann Intern Med. (2018) 169:467–73. doi: 10.7326/M18-0850
57.
DiamondTCameronFBequetteBW. A new meal absorption model for artificial pancreas systems. J Diabetes Sci Technol. (2022) 16:40–51. doi: 10.1177/1932296821990111
58.
RavelliMNSchoellerDA. Traditional self-reported dietary instruments are prone to inaccuracies and new approaches are needed. Front Nutr. (2020) 7:90. doi: 10.3389/fnut.2020.00090
59.
AtlasENimriRMillerSGrunbergEAPhillipM. MD-logic artificial pancreas system. Diabetes Care. (2010) 33:1072–6. doi: 10.2337/dc09-1830
60.
BertrandL.Cleyet-MarrelN.LiangZ. The role of continuous glucose monitoring in automatic detection of eating activities. In 2021 IEEE 3rd global conference on life sciences and technologies (LifeTech). Nara, Japan: IEEE; (2021). p. 313–314.
61.
BertrandLCleyet-MarrelNLiangZ. Recognizing eating activities in free-living environment using consumer wearable devices. In: Engineering Proceedings MDPI; (2021). p. 58.
62.
DassauEBequetteBWBuckinghamBADoyleFJ. Detection of a meal using continuous glucose monitoring. Diabetes Care. (2008) 31:295–300. doi: 10.2337/dc07-1293
63.
DovcKPionaCYeşiltepe MutluGBratinaNJenko BizjanBLepejDet al. Faster compared with standard insulin Aspart during day-and-night fully closed-loop insulin therapy in type 1 diabetes: a double-blind randomized crossover trial. Diabetes Care. (2020) 43:29–36. doi: 10.2337/dc19-0895
64.
El FathiAPalisaitisEBouletBLegaultLHaidarA. An Unannounced Meal Detection Module for artificial pancreas control systems. In: 2019 American control conference (ACC). Philadelphia, PA, USA: IEEE; (2019). p. 4130–4135.
65.
PalisaitisEEl FathiAvon OettingenJEHaidarALegaultL. A meal detection algorithm for the artificial pancreas: a randomized controlled clinical trial in adolescents with type 1 diabetes. Diabetes Care. (2021) 44:604–6. doi: 10.2337/dc20-1232
66.
FaccioliSSala-MiraIDíezJLFacchinettiASparacinoGdel FaveroSet al. Super–twisting-based meal detector for type 1 diabetes management: improvement and assessment in a real-life scenario. Comput Methods Prog Biomed. (2022) 219:106736. doi: 10.1016/j.cmpb.2022.106736
67.
AndersonSMRaghinaruDPinskerJEBoscariFRenardEBuckinghamBAet al. Multinational home use of closed-loop control is safe and effective. Diabetes Care. (2016) 39:1143–50. doi: 10.2337/dc15-2468
68.
FushimiEColmegnaPDe BattistaHGarelliFSanchez-PenaR. Unannounced meal analysis of the ARG algorithm. In: 2019 American control conference (ACC). Philadelphia, PA, USA: IEEE; (2019) p. 4740–4745.
69.
Sánchez-PeñaRColmegnaPGarelliFDe BattistaHGarcía-VioliniDMoscoso-VásquezMet al. Artificial pancreas: clinical study in Latin America without Premeal insulin boluses. J Diabetes Sci Technol. (2018) 12:914–25. doi: 10.1177/1932296818786488
70.
GodoyJLSerenoJERivadeneiraPS. Meal detection and carbohydrate estimation based on a feedback scheme with application to the artificial pancreas. Biomed Signal Process Control. (2021) 68:102715. doi: 10.1016/j.bspc.2021.102715
71.
HoyosJ. D.LagunaA. J.CarpinteroA. R.Sala-MiraI.DiezJ. L.BondiaJ. Characterization of glycemic patterns in type 1 diabetes without insulin or meal input data. In: 2022 10th international conference on systems and control (ICSC). Marseille, France: IEEE; (2022). p. 576–581.
72.
AleppoGRuedyKJRiddlesworthTDKrugerDFPetersALHirschIet al. REPLACE-BG: a randomized trial comparing continuous glucose monitoring with and without routine blood glucose monitoring in adults with well-controlled type 1 diabetes. Diabetes Care. (2017) 40:538–45. doi: 10.2337/dc16-2482
73.
KölleKBiesterTChristiansenSFougnerALStavdahlO. Pattern recognition reveals characteristic postprandial glucose changes: non-individualized meal detection in diabetes mellitus type 1. IEEE J Biomed Health Inform. (2020) 24:594–602. doi: 10.1109/JBHI.2019.2908897
74.
Mosquera-LopezCWilsonLMEl YoussefJHiltsWLeitschuhJBraniganDet al. Enabling fully automated insulin delivery through meal detection and size estimation using artificial intelligence. NPJ Digit Med. (2023) 6:39. doi: 10.1038/s41746-023-00783-1
75.
OrnetzederCReitererFChristensenMBNorgaardKFreckmannGdel ReL. Feasibility of fully closed loop insulin delivery in type 2 diabetes. In: 2019 IEEE conference on control technology and applications (CCTA). Hong Kong, China: IEEE; (2019). p. 906–913.
76.
ZschornackESchmidCPleusSLinkMKlötzerHMObermaierKet al. Evaluation of the performance of a novel system for continuous glucose monitoring. J Diabetes Sci Technol. (2013) 7:815–23. doi: 10.1177/193229681300700403
77.
ShahVNDuBoseSNLiZBeckRWPetersALWeinstockRSet al. Continuous glucose monitoring profiles in healthy nondiabetic participants: a multicenter prospective study. J Clin Endocrinol Metab. (2019) 104:4356–64. doi: 10.1210/jc.2018-02763
78.
Gubitosi-KlugRABraffettBHBebuIJohnsonMLFarrellKKennyDet al. Continuous glucose monitoring in adults with type 1 diabetes with 35 years duration from the DCCT/EDIC study. Diabetes Care. (2022) 45:659–65. doi: 10.2337/dc21-0629
79.
HarveyRADassauEZisserHSeborgDEDoyleFJ. Design of the Glucose Rate Increase Detector: a meal detection module for the health monitoring system. J Diabetes Sci Technol. (2014) 8:307–20. doi: 10.1177/1932296814523881
80.
SamadiSTurksoyKHajizadehIFengJSevilMCinarA. Meal detection and carbohydrate estimation using continuous glucose sensor data. IEEE J Biomed Health Inform. (2017) 21:619–27. doi: 10.1109/JBHI.2017.2677953
81.
RamkissoonCHerreroPBondiaJVehiJ. Unannounced meals in the artificial pancreas: detection using continuous glucose monitoring. Sensors. (2018) 18:884. doi: 10.3390/s18030884
82.
PalaciosVWoodbridgeDMKFryJL. Machine learning-based meal detection using continuous glucose monitoring on healthy participants: an objective measure of participant compliance to protocol. In: 2021 43rd annual international conference of the IEEE engineering in Medicine & Biology Society (EMBC). Mexico: IEEE; (2021). p. 7032–7035.
83.
PoppCJWangCHooverAGomezLACurranMSt-JulesDEet al. Objective determination of eating occasion timing (OREO): combining self-report, wrist motion, and continuous glucose monitoring to detect eating occasions in adults with pre-diabetes and obesity. J Diabetes Sci Technol. (2023):19322968231197205. doi: 10.1177/19322968231197205
84.
SamadiSRashidMTurksoyKFengJHajizadehIHobbsNet al. Automatic detection and estimation of unannounced meals for multivariable artificial pancreas system. Diabetes Technol Ther. (2018) 20:235–46. doi: 10.1089/dia.2017.0364
85.
TurksoyKSamadiSFengJLittlejohnEQuinnLCinarA. Meal detection in patients with type 1 diabetes: a new module for the multivariable adaptive artificial pancreas control system. IEEE J Biomed Health Inform. (2016) 20:47–54. doi: 10.1109/JBHI.2015.2446413
86.
TurksoyKQuinnLTLittlejohnECinarA. An integrated multivariable artificial pancreas control system. J Diabetes Sci Technol. (2014) 8:498–507. doi: 10.1177/1932296814524862
87.
WeimerJChenSPeleckisARickelsMRLeeI. Physiology-invariant meal detection for type 1 diabetes. Diabetes Technol Ther. (2016) 18:616–24. doi: 10.1089/dia.2015.0266
88.
PageMJMcKenzieJEBossuytPMBoutronIHoffmannTCMulrowCDet al. Statement: an updated guideline for reporting systematic reviews. BMJ. (2020) 2021:n71. doi: 10.1136/bmj.n71
89.
LeeHWayneBequette, A closed-loop artificial pancreas based on model predictive control: Human-friendly identification and automatic meal disturbance rejection. Biomed. Signal Process. Control. (2009) 4:347–54. doi: 10.1016/j.bspc.2009.03.002
90.
BiSWangTTobiasNNordrumJWangSHalvorsenGet al. Auracle: detecting eating episodes with an ear-mounted sensor. Proc ACM Interact Mob Wearable Ubiquitous Technol. (2018) 2:1–27. doi: 10.1145/3264902
91.
KumarTSSoilandEStavdahlOFougnerAL. Pilot study of early meal onset detection from abdominal sounds. In: 2019 E-health and bioengineering conference (EHB). Iasi, Romania: IEEE; (2019). p. 1–4.
92.
YangJAWangJNakandalaSKumarAJankowskaMM. Predicting eating events in free living individuals. In 2019 15th international conference on eScience (eScience). San Diego, CA, USA: IEEE; (2019). p. 627–629.
93.
HalvorsenMBenamKDKhoshamadiHFougnerAL. Blood glucose level prediction using subcutaneous sensors for in vivo study: compensation for measurement method slow dynamics using Kalman filter approach. In 2022 IEEE 61st conference on decision and control (CDC). Cancun, Mexico: IEEE; (2022). p. 6034–6039.
Summary
Keywords
meal detection, continuous glucose monitoring, dietary assessment, healthcare technology, closed loop, sensors, meal timing
Citation
Brummer J, Glasbrenner C, Hechenbichler Figueroa S, Koehler K and Höchsmann C (2024) Continuous glucose monitoring for automatic real-time assessment of eating events and nutrition: a scoping review. Front. Nutr. 10:1308348. doi: 10.3389/fnut.2023.1308348
Received
10 October 2023
Accepted
13 December 2023
Published
08 January 2024
Volume
10 - 2023
Edited by
Arpita Mukhopadhyay, St. John's Research Institute, India
Reviewed by
Andrea Tumminia, Università di Catania, Italy; Ali Cinar, Illinois Institute of Technology, United States
Updates
Copyright
© 2024 Brummer, Glasbrenner, Hechenbichler Figueroa, Koehler and Höchsmann.
This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.
*Correspondence: Christoph Höchsmann, christoph.hoechsmann@tum.de
†ORCID: Julian Brummer https://orcid.org/0000-0001-9022-5497Christina Glasbrenner https://orcid.org/0009-0008-5420-5219Sieglinde Hechenbichler Figueroa https://orcid.org/0000-0001-9147-8824Karsten Koehler https://orcid.org/0000-0002-9618-2069Christoph Höchsmann https://orcid.org/0000-0003-2007-3007
Disclaimer
All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article or claim that may be made by its manufacturer is not guaranteed or endorsed by the publisher.