Assessing Mealtime Macronutrient Content: Patient Perceptions Versus Expert Analyses via a Novel Phone App

Abstract

Background: People with type 1 diabetes estimate meal carbohydrate content to accurately dose insulin, yet, protein and fat content of meals also influences postprandial glycemia. We examined accuracy of macronutrient content estimation via a novel phone app. Participant estimates were compared with expert nutrition analyses performed via the Remote Food Photography Method© (RFPM©).

Methods: Data were collected through a novel phone app. Participants were asked to take photos of meals/snacks on the day of and day after scheduled exercise, enter carbohydrate estimates, and categorize meals as low, typical, or high protein and fat. Glycemia was measured via continuous glucose monitoring.

Results: Participants (n = 48) were 15–68 years (34 ± 14 years); 40% were female. The phone app plus RFPM© analysis captured 88% ± 29% of participants' estimated total energy expenditure. The majority (70%) of both low-protein and low-fat meals were accurately classified. Only 22% of high-protein meals and 17% of high-fat meals were accurately classified. Forty-nine percent of meals with <30 g of carbohydrates were overestimated by an average of 25.7 ± 17.2 g. The majority (64%) of large carbohydrate meals (≥60 g) were underestimated by an average of 53.6 ± 33.8 g. Glycemic response to large carbohydrate meals was similar between participants who underestimated or overestimated carbohydrate content, suggesting that factors beyond carbohydrate counting may impact postprandial glycemic response.

Conclusions: Accurate estimation of total macronutrients in meals could be leveraged to improve insulin decision support tools and closed loop insulin delivery systems; development of tools to improve macronutrient estimation skills should be considered.

Introduction

Current insulin dosing for meals by people with type 1 diabetes (T1D) is primarily based on carbohydrate counting and specific carbohydrate-to-insulin ratios used by patients to calculate the appropriate meal bolus insulin dose for each meal and snack.¹ Recent research, however, has stressed the role of protein and fat content of the meal in altering the postprandial glycemic response.^2–6 Meals with the same carbohydrate but higher fat content may have a lower early glycemic response, but the postprandial glucose peak is higher, shifted later, and remains elevated for longer in the late postprandial phase.^2,6,7 Similarly, high-protein intake at meals can increase the time to peak glucose and can increase postprandial glucose for several hours after a meal.^4,5,8,9

Indeed, it is because of these altered postprandial glycemic responses that some diabetes professionals propose adjusting the meal bolus insulin dose to account for protein and fat content by including a combined fat and protein intake of 100 g equal to 10 g of carbohydrate.^10,11 In addition, studies have used in silico models to investigate an extended wave-bolus for insulin pump users that accounts for the protein and fat content of meals in an effort to improve postprandial glycemic control.^12–14 These models have used standardized meals of known protein and fat content or computer estimates of macronutrients from food photos. But, to date, the exact approach of meal bolus dosing that improves postprandial glycemia by accounting for both fat and protein content of meals is still debated.

The carbohydrate counting method for determining mealtime insulin needs is associated with a high prevalence of postprandial hyper- and hypoglycemia, and remains inadequate for achieving postprandial glycemic control even with hybrid closed-loop insulin therapy.^15,16 Recent data from the T1D Exchange Clinic Registry suggests about 70% of people with T1D do not reach hemoglobin A1C treatment goals, and additional methods to improve postprandial glycemia are needed.¹⁷ Including estimates of fat and protein content of meals in bolus calculators or in closed-loop systems to adjust insulin delivery does improve postprandial glycemia, suggesting that this approach may be beneficial for achieving improved glycemic control.^18,19

Current systems require the user to estimate carbohydrate content of meals. Including all macronutrients in the bolus, calculations would require users to input protein and fat estimations. Estimates of carbohydrate content can be inaccurate by >20 g, especially in larger meals, and very little is known about how accurately people living with T1D can estimate the protein and fat content of meals.^20–24 To address this gap, we compared participant estimation of the macronutrient content of meals with expert nutritional analysis of those same meals using the Remote Food Photography Method© (RFPM©), among 48 adults with T1D. In addition, we assessed the relationship between the accuracy of participant estimation of carbohydrate content and the postprandial meal glucose excursion measured with continuous glucose monitoring (CGM) in a free-living environment.

Methods

Research design and methods

Adults with T1D were enrolled in a 4-week study conducted to test the methods for collection and aggregation of coordinated data around exercise events to examine the effects of various exercise modalities and nutrient intake. The protocol and informed consent forms were approved by an institutional review board; informed consent was obtained from each participant. Major inclusion criteria included age between 15 and 70 years, use of either multiple daily insulin injections or an insulin pump, and a diabetes duration of at least 2 years. Participants were trained to use a novel food and exercise tracking phone app (T1-Dexi app) to record exercise events and to collect photos of foods consumed before meals and plate waste after meals on the day of and day after scheduled, structured exercise during the 4-week protocol.

The app provided meal reminders via text on days when food photos were collected. Participants were also asked to rank meals as low, typical, or high in protein; low, typical, or high in fat; and small, medium, or large in meal size; and to record the estimated grams of carbohydrates for that meal. Meal insulin bolus and basal insulin data were collected through Tidepool (Palo Alto, CA) or Carelink by Medtronic MiniMed (Devonshire, CA) if the participant was using a Medtronic insulin pump and by a wireless Bluetooth-enabled smart insulin pen (Clipsulin) or written logs if the participant was an MDI user. CGM was used to estimate the participants' glucose. Participants used their personal CGM (50% Dexcom; 10% Medtronic, 2% Abbott) or blinded CGM (38% Dexcom G4 with 505 or G5) if they were not a current CGM user. Participant meal entries were compared to the participants' glucose levels to observe how postprandial glycemic excursions were related to the participants' carbohydrate estimation and the nutrient intake data captured via the RFPM©.

The RFPM© was used to measure energy and nutrient intake. When using this method, a smartphone or a similar device, such as an iPad, is used by the participant to capture images of their food selection before they eat, and their plate waste after they eat. Participants are asked to capture these images at an arm's distance away and at a 45° angle, and ideally a reference card or fiduciary marker is included in the image (the reference card is similar to a driver's license and is provided to participants before data collection). The food images are transmitted in real-time via the smartphone to a server for analysis. Thus, for each meal, at least two images are captured and analyzed, one “before” image of food selection and one “after” image of plate waste (additional images are used to capture second servings).

After the images arrive on the server, they are analyzed by a trained human rater using a computer-assisted approach. In brief, the rater identifies the foods in the images and links them to a nutrient reference in a nutrient database, namely, the Food and Nutrient Database for Dietary Studies.²⁵ The rater then obtains a reference image from an archive of food images. These reference images are of known quantities of foods and the rater selects a reference image of the same food or a similar food to what the participant consumed. The rater then estimates the portion size of the participants' foods by visually comparing the participants' food images to the standard food image, and this process is facilitated by the reference card that is included in the participants' images. This process relies on existing and validated visual comparison methodology^26–29 to estimate food selection and plate waste, which is used to calculate food intake by difference. These calculations are performed by the computer system, which uses the nutrient values that were matched to the participants' foods. The RFPM© has been found to accurately measure the energy^26,27 and nutrient intake of adults,²⁶ with an error of only 3.7% over 6 days in free-living conditions compared to the gold standard, doubly labeled water.²⁶ In the current project, images from 2731 meals were analyzed.

Statistical methods

All participants with at least one RFPM©-analyzed meal were included in the analysis cohort. To assess the ability of RFPM© in capturing total nutrient intake, total energy expenditure (TEE) was estimated for each participant using the Harris–Benedict basal energy expenditure equation.³⁰ This basal energy expenditure was multiplied by 1.4 under the assumption that participants were adherent to the study protocols and therefore were moderately active during the study period.³¹ Then, RFPM©-analyzed nutrient intake for days, in which participants recorded at least three meals, were compared to this estimated TEE. For the purposes of comparing the participant-reported and RFPM©-analyzed nutritional content (i.e., carbohydrate content, protein level, fat level, and meal size) for each meal, the continuous RFPM©-analyzed results were discretized into low, typical, or high meal content categories. Meal energy content was also divided into three approximate meal size subgroups with cutpoints at <200, 200 to <600, and ≥600 kCal, similar to snack, breakfast and lunch, and dinner meal sizes, respectively, in previous reports.²³

For the purposes of assessing the impact of carbohydrate estimation accuracy on postprandial metrics, meals were categorized using RFPM© into three subgroups: small (<30 g carbohydrate), medium (30 to <60 g carbohydrate), and large (≥60 g carbohydrate) and the participants' estimation of the carbohydrate content of a meal was considered relatively accurate if the participant estimated within 5 g for meals determined by RFPM© to contain <25 g of carbohydrates or within 20% for meals determined by RFPM© to contain ≥25 g of carbohydrates. Previous feeding studies have used either <20 g difference or 10% variation from the known carbohydrate content as a measure of accurate carbohydrate estimation; we have elected to use something similar, but 20% can be applied to the wide range of carbohydrate intakes observed in our free-living setting.^22,32,33 One participant was excluded from this analysis due to the participant's reported carbohydrate estimates having a mean relative absolute difference among 109 meals of 527% when compared to RFPM©-analyzed carbohydrate estimates.

For analyses, including CGM-measured postprandial glucose levels, a minimum of 2 h of postprandial CGM data were required, and researchers only included meals with no other recorded meal within 2 h before and 4 h after the recorded meal time and a nonzero insulin bolus recorded on the same day. Meal-specific insulin boluses were defined as administered boluses within 30 min before and up to 20 min after a logged meal time. Meal glucose excursion was estimated as the postprandial peak glucose minus baseline glucose (average glucose within 30 min before meal).

A linear mixed effects model with meal bolus as the response and difference in participant-reported and RFPM©-analyzed carbohydrates as the predictor adjusting for baseline age, HbA1c, pump use, baseline glucose, RFPM©-analyzed meal carbohydrate content, and participant-reported meal and fat content as fixed effects and participant as a random effect was used to determine whether or not there was an association between participant carbohydrate estimation accuracy and meal boluses. A similar mixed model was fitted to assess the impact of carbohydrate estimation accuracy on meal glucose excursion and time to postprandial peak glucose in each meal carbohydrate subgroup (e.g., small, medium, or large) adjusting for baseline age, HbA1c, pump use, baseline glucose, meal bolus, number of postprandial CGM readings, and RFPM©-analyzed meal macronutrient content (i.e., carbohydrate, protein, and fat) as fixed effects and participant as a random effect. If model residuals were skewed, a rank-based transformation was used. All statistical tests were two-sided and assessed at the α = 0.05 significance level. Multiple comparisons were adjusted using the Benjamini-Hochberg adaptive false discovery rate correction procedure. For summary tabulations, mean and SD or summary statistics appropriate to the distribution are reported.

All analyses were performed in SAS 9.4.

Results

Participant characteristics

The (n = 48) participants ranged in age from 15 to 68 years and 40% were female. They had a mean ± SD body mass index of 26.0 ± 3.1 kg/m² and T1D duration ranging from 3 to 57 years. A summary of participant characteristics is provided in Supplementary Table S1.

Food photo nutrient analysis

The average macronutrient content of the RFPM©-analyzed meals is provided in Supplementary Table S2. The participants' meals consisted of a median of 31 g (IQR: 17–54 g) carbohydrate and 2 g (IQR: 1–5 g) fiber; 12% (IQR: 5%–19%) protein, and 34% (IQR: 17%–47%) fat. As a quality control measure and to confirm that our RFPM© results were capturing the majority of food consumed by participants, the RFPM©-measured daily total energy intake was compared to the estimated TEE for each participant under the assumption that energy intake and expenditure are equal in weight-stable individuals. The participants had an average weight change of −0.2 ± 1.4 kg, and the RFPM© was able to capture 88% ± 29% of the estimated TEE.

Participant perception of high and low protein and fat meals

A total of 2203 meals were assessed for participant-estimated and RFPM©-analyzed protein level, fat level, and meal size congruence. The protein content of about half of the meals (1105/2203) contained 13% or less of the total energy in the meal (low protein). The majority (70%) of these meals were correctly identified by participants. In contrast, the protein content of one third of the meals (662/2203) was 18% or more of the total energy in the meal (high protein), but only 22% of high-protein meals were identified correctly (Fig. 1). Approximately one-third of the meals (768/2203) were low fat (<26% of the total meal energy content) and the majority (70%) of low-fat meals were correctly identified by participants. Over half of the meals (1239/2203) were high fat (≥32% of the total meal energy content), but only 17% of high fat meals were identified as high fat. About one-third (751/2203) of the meals were small meals of <200 kCal, and participants could identify a small meal 80% of the time, but only 25% of the 387 meals containing ≥600 kcal were categorized by the participant as large. In general, participants were much less likely to correctly identify large meals and meals that were high in protein and fat.

FIG. 1.

Bar plot of protein content, fat content, and meal size estimation congruency. The protein and fat level categories as a percentage of the meal energy content (kCal), meal size categories, and the number of meals (N) classified by RFPM© with that level are provided in black above the bars. The white numbers within the bars are the percent of these meals that were accurately classified by participants. RFPM©, Remote Food Photography Method©.

Participant perception of carbohydrate content of meals

One hundred fifteen meals with <30 g of carbohydrate had pre- and postprandial CGM data for analysis. Participants overestimated the carbohydrate content for 49% (56/115) of these small carbohydrate meals by an average difference of 25.7 ± 17.2 g (Table 1). One hundred thirty-nine meals with ≥60 g of carbohydrate had pre- and postprandial CGM data for analysis. Participants underestimated the carbohydrate content for 64% (89/139) of these large carbohydrate meals by an average difference of 53.6 ± 33.8 g. Accuracy of medium-sized meals with 30 to <60 g of carbohydrate was approximately evenly divided into meals where participants underestimated, overestimated, or accurately estimated the meal carbohydrate content. Figure 2 illustrates participant carbohydrate estimation accuracy over a large range of RFPM©-measured meal carbohydrate sizes. Participants' estimates of the carbohydrate content of meals appear to cluster around a range of values between 10 and 80 g and only slightly increased for large RFPM©-measured carbohydrate meals.

Table 1.

Participant Carbohydrate Estimations and Postprandial Glucose Response

Photo-analyzed meal carb content	Metric	Carbohydrate estimation accuracya			Pb
		Underestimated	Relatively accurate	Overestimated
Small <30 g	No. of participants/mealsc	11/21	25/38	21/56
	Participant-estimated – RFPM©-Analyzed Carb Content Difference (g), mean (SD)	−10.0 (3.4)	0.4 (2.5)	25.7 (17.2)
	Meal Bolus Unitsd median (quartiles)	0.0 (0.0, 3.0)	0.7 (0.0, 2.2)	2.4 (0.0, 5.2)
	Number (%) of meals with
	0 Bolus Unitsd	11 (52%)	27 (47%)	20 (36%)
	>0 to <3 Bolus Unitsd	5 (24%)	23 (40%)	12 (21%)
	3 to <6 Bolus Unitsd	4 (19%)	5 (9%)	14 (25%)
	≥6 Bolus Unitsd	1 (5%)	3 (5%)	10 (18%)
	Baseline glucosee (mg/dL), median (quartiles)	129 (86, 183)	143 (91, 176)	147 (108, 187)
	Postprandial peak glucose (mg/dL), median (quartiles)	183 (134, 233)	185 (156, 254)	217 (176, 297)
	Meal glucose excursion (peak glucose minus baseline, mg/dL), median (quartiles)	32 (0, 54)	30 (3, 79)	61 (10, 147)	0.03
	Time to peak glucose (min), median (quartiles)	78 (44, 183)	79 (44, 167)	97 (43, 173)	0.35
Medium 30 to <60 g	No. of participants/mealsc	20/42	23/49	21/46
	Participant-estimated – RFPM©-Analyzed Carb content difference (g), mean (SD)	−21.4 (8.8)	0.2 (5.6)	27.0 (23.0)
	Meal bolus unitsd median (quartiles)	0.0 (0.0, 1.7)	3.5 (1.9, 6.0)	4.6 (0.0, 8.9)
	Number (%) of meals with
	0 Bolus Unitsd	27 (64%)	11 (22%)	13 (28%)
	>0 to <3 Bolus Unitsd	9 (21%)	9 (18%)	3 (7%)
	3 to <6 Bolus Unitsd	4 (10%)	16 (33%)	12 (26%)
	≥6 Bolus Unitsd	2 (5%)	13 (27%)	18 (39%)
	Baseline glucosee (mg/dL), median (quartiles)	123 (90, 171)	132 (100, 160)	138 (89, 195)
	Postprandial peak glucose (mg/dL), median (quartiles)	209 (162, 251)	201 (166, 240)	196 (151, 280)
	Meal glucose excursion (peak glucose minus baseline, mg/dL), median (quartiles)	58 (28, 128)	66 (37, 123)	64 (26, 113)	0.65
	Time to peak glucose (min), median (quartiles)	106 (62, 203)	135 (68, 178)	113 (53, 183)	0.86
Large ≥60 g	No. of participants/mealsc	27/89	18/41	4/9
	Participant-estimated – RFPM©-Analyzed Carb Content Difference (g), mean (SD)	−53.6 (33.8)	−1.0 (9.5)	51.3 (48.1)
	Meal Bolus Unitsd median (quartiles)	3.2 (0.0, 7.7)	6.0 (3.8, 9.3)	8.0 (4.5, 13.0)
	Number (%) of meals with
	0 Bolus Unitsd	29 (33%)	9 (22%)	1 (11%)
	>0 to <3 Bolus Unitsd	13 (15%)	1 (2%)	0 (0%)
	3 to <6 Bolus Unitsd	20 (22%)	9 (22%)	2 (22%)
	≥6 Bolus Unitsd	27 (30%)	22 (54%)	6 (67%)
	Baseline glucosee (mg/dL), median (quartiles)	127 (95, 194)	110 (83, 164)	73 (70, 128)
	Postprandial peak glucose (mg/dL), median (quartiles)	239 (183, 293)	214 (170, 252)	194 (156, 268)
	Meal glucose excursion (peak glucose minus baseline, mg/dL), median (quartiles)	77 (30, 141)	78 (21, 157)	74 (38, 145)	0.57
	Time to peak glucose (min), median (quartiles)	105 (49, 176)	119 (60, 187)	124 (98, 166)	0.55

Summary statistics are on a meal-level.

^aA meal was considered to be relatively accurate if either (1) the meal was analyzed in the photo to have <25 g carbs and the participant correctly estimated within ±5 g or (2) the meal had ≥25 g carbs and the participant correctly estimated within 20% of the photo-analyzed carb content.

^bFrom a mixed-effect model adjusting for baseline glucose, meal bolus, RFPM©-analyzed carb, protein, and fat content, number of postprandial CGM readings, age, baseline HbA1c, and pump use as fixed effects and participant as a random effect. Estimation accuracy was used as a continuous covariate. Due to skewed residuals, all continuous covariates, glucose excursion, and time to peak were rank-transformed. Multiple comparisons were adjusted for using the Benjamini-Hochberg adaptive false discovery rate correction procedure.

^cOnly meals with (1) at least 2 h after the previous meal and 4 h before the next recorded meal (2) at least 2 h of postprandial CGM data, and (3) a nonzero bolus reading recorded within 16 h of the meal were included. One outlying participant was excluded due to unusual inaccuracy of the participant's reported carbohydrate data.

^dMeal boluses were defined as administered boluses in the 30 min before the meal to 20 min after a logged meal time.

^eAverage glucose level in the 30 min before meal.

CGM, continuous glucose monitoring.

FIG. 2.

Scatterplot of participant-estimated versus RFPM©-analyzed carbohydrate content of meals (n = 411, one point per meal). (B) is a subset of the plot in (A) focusing on the cluster of RFPM©-analyzed meals containing 100 g carbs or less. Diagonal dashed black line is the line of identity. Solid black lines are the upper and lower thresholds, respectively, for an accurately estimated meal: ±5 g for RFPM©-analyzed meals containing <25 g carbs; ±20% for RFPM©-analyzed meals containing ≥25 g carbs. Blue, green, and red dots represent overestimated, accurately estimated, and underestimated meals, respectively. Slightly darker dots indicate multiple meals.

Total energy content of the meals increased with increasing carbohydrate content: <30 g carbohydrate meals contained a median of 160 (IQR: 108–263) kcal; 30 to <60 g carbohydrate meals contained a median of 395 (IQR: 308–516) kcal, and ≥60 g carbohydrate meals contained a median of 757 (IQR: 567–979) kcal. Interestingly, the percent of total energy from protein and fat was similar across the small, medium, and large size meals (Supplementary Table S3). The perceived carbohydrate content of the meal appeared to alter meal bolus decisions as expected: participants who overestimated the carbohydrate content of meals, regardless of size, tended to dispense larger meal bolus insulin doses, while participants who underestimated the carbohydrate content of the meals tended to dispense smaller meal boluses or did not dose at all (P < 0.001; Table 1).

Postprandial glycemic response of under and overestimated meals

The postprandial meal glucose excursion of overestimated small carbohydrate meals was significantly higher than that of underestimated small carbohydrate meals (median 61 vs. 32 mg/dL, P = 0.03) even after controlling for any differences in meal bolus doses or protein and fat content. Overestimated small carbohydrate meals also showed a delayed time to peak in comparison to underestimated small carbohydrate meals, but this was not statistically significant (median 97 vs. 78 min, P = 0.35). These results were paradoxical despite the tendency of the participants to have higher meal bolus insulin doses as their estimate of the carbohydrate content of a meal increased (Table 1 and Fig. 3).

FIG. 3.

Tracings of postprandial glucose change from baseline for (A) meals containing <30 g carbohydrates (n = 21 underestimated and 56 overestimated meals), (B) meals containing 30 to <60 g carbohydrates (n = 42 underestimated and 46 overestimated meals), and (C) meals containing ≥60 g carbohydrates (n = 89 underestimated and 9 overestimated meals). Reds represent underestimated meals. Blues represent overestimated meals. Dots and lines represent medians. Shaded bands represent the interquartile range of glucose excursions. Dashed, horizontal line represents no change from glucose level before meal.

For larger carbohydrate meals containing ≥60 g carbs, however, there was no difference in postprandial glycemic response between underestimated and overestimated meals (Fig. 3). The times to peak glucose and meal glucose excursions were also similar between these groups despite differences in insulin doses (Table 1).

Discussion

The RFPM©-analyzed nutritional content of mixed-meals in this cohort of participants with T1D is consistent with the typical nutrient distribution of the general American population.³⁴ Methods to estimate nutrient intake of free-living participants such as 3-day diet records or 24-h recall are known to be inaccurate due, in part, to unrecorded meals/snacks and participant error in portion size estimation.³⁵ RFPM© captured an average of 88% of the estimated TEE in this relatively weight-stable cohort and previous research found the RFPM© to accurately estimate energy and nutrient intake,²⁶ suggesting that the RFPM© accurately quantified nutrient intake in our participants.

Managing postprandial glycemia with current carbohydrate counting methods remains challenging for people living with T1D. In this trial, participants tended to accurately estimate the total carbohydrate content of their meals within a relatively narrow range and generally were more accurate for smaller meals with less carbohydrate content (<30 g). Results shown in Figure 2A demonstrate that for meals with carbohydrate amounts greater than 100 g, the participants almost always underestimated the carbohydrate amount. Frequent underestimation of the carbohydrate content of large meals and potentially underdosing prandial insulin could result in high meal time glucose excursions. The information presented here demonstrating substantial underestimation of carbohydrates for larger meals is important for automated insulin delivery systems and decision support apps that include meal bolus calculators. Several previous studies have attempted to improve carbohydrate counting methods and the related bolus insulin dosing to better match true carbohydrate intake. In 20 people with T1D, a GoCARB estimation app was compared with traditional patient carbohydrate counting.³⁶ The app estimated the carbohydrate content of the meal for the participant from a meal photo and the participant used this estimate to calculate and dose meal bolus insulin. When using the GoCARB estimate, participants decreased time spent in hyperglycemia, but did not change time in hypoglycemia or significantly alter total insulin delivered per day or number of insulin bolus given in this relatively small cohort. Similarly, a new decision-making tool, SMART (aSsess-Ment of the bolus cAlculation- and caRbohydrate esTimation-skill), was evaluated in 411 people with T1D and type 2 diabetes on intensive insulin therapy.³⁷ Better estimates of total carbohydrate intake associated with improved bolus insulin estimates, lowered HbA1c, and decreased hyperglycemia, but did not alter time in hypoglycemia. In sum, these studies suggest that improved tools for enabling more accurate estimation of carbohydrates before meals can help persons with T1D improve their glycemic control.

Multiple studies in adults and adolescents with T1D have demonstrated that high-fat and high-protein meals delay time to peak glucose postprandially and prolong glucose excursion during the late postprandial phase.^2,5–8,38 Furthermore, there is evidence that these effects may be additive for fat and protein. Yet, while it is possible that including insulin adjustments for fat and protein content could improve postprandial glycemic control, it is certain that making these adjustments would add to the complexity and burden of estimating meal macronutrient content and insulin bolus doses for people with diabetes. We investigated the ability of participants with T1D to estimate the protein and fat content of their meals using a simplified categorical system of low, typical, or high. We anticipated that while people with T1D are frequently educated on how to determine the carbohydrate content of foods, they are rarely asked to identify the protein and fat content of their foods. Indeed, our results were partially confirmed in that participants were much less likely to correctly identify a meal that was high in fat (17% correctly identified) or protein (22% correctly identified) compared with a meal low in fat or protein (70% of meals correctly identified for both fat and protein). Therefore, our results suggest that asking persons with diabetes to even include broad categorical estimates of protein and fat content along with their estimates of carbohydrate in their bolus insulin calculations will require extensive education on the macronutrient content of foods and/or the addition of novel decision-making support tools for insulin dosing that can accurately incorporate all of these complex estimates into routine mealtime insulin calculations.

The tendency of participants to estimate within a narrow carbohydrate content range is illustrated by our finding that participants were more likely to overestimate the carbohydrate content of small meals, while underestimate the carbohydrate content of large meals (Fig. 2). However, we were surprised by the glycemic response to small meals. Participants who overestimated the carbohydrate content of small meals trended toward larger mealtime insulin boluses, but still recorded higher peak postprandial glucose and a delay in time to peak glucose than participants who underestimated the carbohydrate content of small meals. Our statistical analysis controlled for protein and fat content. Thus, this finding suggests that perhaps other factors, including the glycemic index of carbohydrates, fiber content, or the types of foods actually consumed—including foods that were not captured by food photos—during meals may have a larger impact on the glycemic response of smaller meals.

In a separate review of the nutrient content of these meals, we found that, in general, the overestimated meals contained refined grains more often compared to the underestimated meals, adding some support for our hypothesis that the actual foods consumed in a small meal may relate more directly to postprandial glucose response than accurate carbohydrate calculations. These and other factors such as gastrointestinal absorption rate, hepatic glycogen storage, and glucose production all impact the postprandial glycemic response.³⁹ However, it should be noted that an equally plausible explanation for our finding could be that smaller carbohydrate meals, such as those containing <30 g of carbohydrates, offer less room for underestimation, which alone could have decreased the number of underestimated small meals and related variability in postprandial glucose response.

As mentioned above, participants were more likely to underestimate the total carbohydrate content of large meals containing ≥60 g of carbohydrates. Yet, despite this tendency to underestimate these meals and thus administer lower mealtime insulin doses, the postprandial glycemic response was similar to those participants who overestimated the carbohydrate content of these meals and administered a larger insulin bolus. Other studies have demonstrated that systematic errors in carbohydrate estimation leading to a higher insulin:carbohydrate ratio do not negatively impact postprandial glycemia but random estimation errors do in fact result in inferior glycemic control.⁴⁰ We are unable to determine the random versus systematic error in this particular study, but suggest that both types of error contribute to the observed postprandial response. This finding of a similar glycemic excursion between “under guessers” and “over guessers” for carbohydrate content in large meals also adds to our previous hypothesis that other factors related to the type of food consumed at meals may play as meaningful a role as accuracy in carbohydrate estimation and that people with diabetes may benefit from education beyond simply carbohydrate counting to learn how to optimally dose for insulin at meals and achieve a more normative postprandial glycemic response.³⁹

This is the first study to use RFPM© among participants with T1D in free-living conditions. Unlike more controlled feeding studies, the current report examined nutrient intake and the postprandial glycemic response in an uncontrolled “real world” setting and provides insight into the daily challenges of T1D management for our participants. Using unique technology, we were also able to determine participants' perception of their meal macronutrient content and compare that to expert analysis. Some limitations of this study include the potential for missing nutrient data, including unrecorded meals/snacks, treatments for hypoglycemia, missed insulin boluses, and increased levels of physical activity from the exercise component of the trial. Another limitation is the smaller sample size of the meal sizes included in our analysis; only 115 small meals and 139 large meals had sufficient postprandial glycemia data, no other meals within 2 h before 4 h after, and sufficient meal bolus data. Finally, our results may be limited because this study represents a secondary analysis of a preexisting data set and we did not conduct a priori power calculation.

In conclusion, we observed that participants with T1D were rarely able to accurately identify that high-protein or high-fat meals were less likely to estimate carbohydrate content of large meals accurately and tended to estimate carbohydrate intake within a relatively narrow range regardless of the meal size. Surprisingly, however, individuals who are less able to accurately estimate their meal size and composition tend to have glycemic excursions similar to those who are better able to estimate their meal size and composition. The complexities of accurately estimating carbohydrate content, and potentially the fat and protein content of meals, are clearly a challenge for many people with diabetes. It is possible that additional decision support tools can assist people with T1D when making these estimates and may help with bolus insulin dosing decisions. But at this point, these systems are limited and more research is needed to determine how helpful they may be for improving postprandial glycemic control.

Author Disclosure Statement

The intellectual property surrounding the Remote Food Photography Method© (RFPM©) is owned by Pennington Biomedical Research Center/Louisiana State University and C.K.M. is an inventor of the technology. M.B.G. has no disclosures. Z.L. has no disclosures. R.W.B. reports receiving consulting fees, paid to his institution, from Insulet, Bigfoot Biomedical, vTv Therapeutics, and Eli Lilly, grant support and supplies, provided to his institution, from Tandem and Dexcom, and supplies from Ascenia and Roche. P.C. has no disclosures. J.C. reports receiving personal fees from Novo Nordisk, personal fees from AstraZeneca, personal fees from Zealand Pharma, personal fees from Dexcom, personal fees from Adocia, and other from Pacific Diabetes Technologies, outside the submitted work. M.C. has no disclosures. E.D. is currently an employee of Eli Lilly and Company. The work presented in this poster/article was performed as part of ED's academic appointment and is independent of his employment with Eli Lilly and Company. F.J.D. reports equity, licensed IP, and is a member of the Scientific Advisory Board of Mode AGC. R.L.G. has no disclosures. P.J. has no disclosures. S.R.P. has no disclosures. M.R.R. has received consulting honoraria from Semma Therapeutics and Sernova Corp. and research support from Xeris Pharmaceuticals. M.R. has received speaker's honoraria from Medtronic Diabetes, Insulet, Ascensia Diabetes, Program), Xeris Pharmaceuticals, Lilly Diabetes and Lilly Innovation.

Funding Information

This work was supported by a grant from the Leona M. and Harry B. Helmsley Charitable Trust and by NORC Center Grant #P30 DK072476 entitled “Nutrition and Metabolic Health Through the Lifespan” sponsored by NIDDK, and by grant #U54 GM104940 from NIGMS, which funds the Louisiana Clinical and Translational Science Center. This research was supported by a grant of no charge materials from DIABNEXT LLC.

Supplementary Material

Supplementary Table S1

Supplementary Table S2

Supplementary Table S3