Objectively measured eating behaviors and their relation to food intake in school and hospital settings

Introduction. The measurement of food intake (what and how much we eat and drink) is of great importance due to its involvement in three great challenges facing humanity: 1) obesity/overnutrition, 2) undernutrition and 3) climate change, as well as their related health consequences. However, measuring food and energy intake in humans is complicated since traditional self-reported methods have systematic bias while traditional objective laboratory methods have generalizability and upscaling issues. Therefore, novel methods to measure food and energy intake in humans have often been requested. A plethora of factors have been associated with variation in food intake in humans. For example, internal behavioral factors such as eating rate, internal disease conditions such as Parkinson’s disease (PD) as well as external environmental factors such as food proximity are notable ones. These factors have mainly been investigated by use of the traditional methods listed above.

Aims. The overarching aim with this thesis was to use novel technological tools (i.e., portable food scales and video cameras) to measure and explain variance in food intake and body mass index in school, hospital and free-living settings. Aims in school setting: To explain variance in food mass intake during school lunch with objectively measured eating behaviors (how a person eats), the proximity to food and subjective appetite measures. To assess the test-retest reliability of objectively measured food mass intake and eating rate during school lunch. To assess the concurrent validity of self- reported eating rate. Aims in hospital setting: To compare energy intake among healthy controls, early and advanced PD patients and to investigate the association between clinical features of PD as well as objective eating behaviors with energy intake during a hospital lunch. Aim in free-living setting: To distinguish differences in BMI z-scores (BMIz) among self-reported eating rate categories in populations of Swedish and Greek high school students.

Methods. School studies. Settings: The data collection was conducted in the school lunch cafeteria environment at a high school in central area of Stockholm, Sweden. Study design: A cross-sectional study design was used to explain variance in food intake and to investigate the association between objectively measured eating rate and food intake. An experimental study design was used to investigate the effects of food proximity and a repeat-measures study design was used to assess the test-retest reliability of objectively measured food mass intake and eating rate. Participants: Six high school classes including 187 students were invited to participate in monitored school lunches during 2015-2017. Out of these, 114 unique students provided complete meal data and 103 with a mean (SD) age of 16.7 (0.6) and BMIz of -0.07 (1.05) were included in the food intake variance analysis. All 114 participants (with a mean (SD) age of 16.5 (0.8) and BMIz 0.04 (1.01)) were included in the association between eating rate and BMIz. Out of the 114 unique participants, 50 students came for a repeated meal and provided complete data for test-retest analyses. Study procedures: The lunch study was conducted during normal school lunch hours (11.30-13.00). The students who participated in the snack experiment came back at 15.30 for the one-hour experimental snack session with snack foods, either a) close to the table where they were sitting (proximal condition) or further away from them (distal condition). Served food. During school lunches, usual lunch food at the included school (beef/vegetable patties, brown sauce, potatoes, fish, variety of vegetables, water/milk) was served in a buffet-like setting. For the snack experiment, chocolate lentils, crackers and grapes were served ad libitum.

Methods. Hospital study. Settings: The data collection was conducted in a dedicated room at the Department of Neurology of the Technical University Dresden (TUD), Germany. Study design: A cross-sectional study design was used. Participants: 64 participants (n = 23 healthy controls, n = 20 early and n = 21 advanced PD patients) with a mean (SD) age of 62.4 (7.8) and BMI 27.2 (4.3) were included. Study procedures: Study participants had a medical evaluation before they ate their lunch meal during normal lunch hours (11.00-15.00). Served food: A standardized meal (200g sausages, 400g potato salad, 200g apple mash and 500ml of water) was served to all participants.

Methods. Free-living study. Settings: A smartphone application was developed to gather self-reported eating rate and BMIz. Study design: A cross-sectional study design was used. Participants: Students from multiple high schools in Sweden (n = 748) and Greece (n = 1084) were recruited through school supported actions (n = 1832 in total, mean (SD) age of 15.8 (0.9), BMIz 0.47 (1.41)) that included self-reported measures of weight, height and eating rate. Study procedures: Students who chose to participate downloaded the study mobile application and self-reported their data.

Data sources and measurements: In the school and hospital setting, weight and height scales were used to measure participants body weight and height, and food mass and energy intake were measured with portable food scales. Video cameras were used to record the meals and eating behaviors were annotated onto the videos in computer software. In the free-living setting, students self-reported their age, weight, height, and their speed of eating in comparison to others at their own discretion.

Reliability and validity. In the school setting, there was no significant systematic change in mean food mass intake from lunch 1 to lunch 2 (-7.5g, 95% confidence interval: -43.1g to +28.0g). The intraclass correlation between food mass intake during lunch 1 vs. lunch 2 was 0.74 (95% confidence interval 0.58 to 0.84). There was a significant systematic change in eating rate (g/min) from lunch 1 to lunch 2 (+4.4 g/min, 95% confidence interval: +0.7 g/min to +8.1 g/min). The intraclass correlation between eating rate during lunch 1 vs. lunch 2 was 0.73 (95% confidence interval 0.59 to 0.85). When comparing the objective eating rate among the three categories of self-reported eating rate (slow, intermediate, and fast), a significant difference between the groups was obtained [F(2, 111) = 7.104, P = 0.001, partial η2 = 0.113]. Bonferroni post hoc comparisons showed that students who self-reported eating slower than others had significantly lower eating rate (- 13.7g/min, 95% confidence interval: -22.5g/min to -4.84g/min) compared to students who self-reported eating faster than others. The weighted Kappa value for self-reported eating rate categories versus objectively established eating rate categories was 0.31 (P < 0.001).

Main results. School: Eating rate, number of spoonfuls, sex, number of food additions and food taste (explanatory power in that order) were all significant explanatory variables for variance in food mass intake during school lunch, while BMI and change in fullness were not significant (effect size: adjusted R squared = 0.766 for the total model). There was a significant “large” (R = 0.667) correlation between objectively measured eating rate and food mass intake during school lunch. When dividing students into tertiles of eating rate (slow, intermediate and fast eaters), a significant difference in food mass intake between the three groups was found [F(2, 111) = 30.578, P < 0.001, partial η2 = 0.355]. Bonferroni post hoc comparisons showed that students in the “slow” objective eating rate tertile were eating 133 grams less food (95% confidence intervals = -210g to -56g) than students in the “intermediate” objective eating rate tertile, and 247 grams less (95% confidence intervals = -324g to -170g) than students in the “fast” eating rate tertile. Students who were participating in the distal snack food condition were eating significantly less energy from snacks than students in the proximal condition (mean difference = -222.7 kcal 95% confidence intervals: -428.3 kcal to -17.2 kcal). Hospital: Advanced PD patients consumed significantly less energy during lunch compared to both early PD patients (b = -202.7 kcal, 95% confidence interval: -329.2 kcal to -76.2 kcal) and healthy controls (b = -162.1 kcal, 95% confidence interval: -285.7 kcal to -38.4 kcal) when controlling for sex. Free-living: Self-reported eating rate was found to be a significant explanatory variable for variation in self-reported BMI z-scores [F(2, 1829) = 9.724, P < 0.001, partial η2 = 0.011]. Bonferroni post hoc test showed that students who self-reported eating slower than others had 0.23 units lower BMI z-scores (95% confidence intervals: -0.43 to, -0.03) than students who self-reported intermediate eating rate, and 0.37 units lower (95% confidence intervals: -0.57 to -0.17) than students who self-reported eating faster than others.

Outcome synthesis. Overall, eating behaviors were the most powerful explanatory variables, while desire to eat and food taste were the most powerful self-reported variables for food and energy intake variance when controlling for sex in the included studies. Advanced PD status (hospital study) as well as the food proximity (snack experiment) were also powerful explanatory variables, while PD-related symptomatology as well as self-reported eating rate, hunger, change in fullness and BMI had low or no explanatory power.

Conclusions. Objectively measured single-meal food mass intake and eating rate could be used to rank individuals in comparison to their peers. Subjective eating rate could be used to distinguish groups with slow and fast eating rates in large scale studies but should not be used on an individual level. The outcomes of this thesis suggest that objectively measured eating behaviors and subjective factors such as food taste and desire to eat, as well as the external condition proximity to food, are all powerful explanatory factors for variance in food mass and energy intake and might be potential targets in future interventions that aim to modify food intake. Additionally, advanced PD condition was associated with lower energy intake. Potential interventions mentioned above might be helpful in this patient group to normalize their energy intake and reduce their risk of undernutrition. Furthermore, the results suggest that novel methods that objectively measure eating behaviors could be utilized in larger-scale nutrition research. Further technological developments of these methods could also give real- time feedback on targeted eating behaviors that are related to food intake, thus ultimately reducing the risk of diseases related to over- and undernutrition.

List of scientific papers

I. Fagerberg P., Langlet B., Glossner A., Ioakimidis I. Food Intake during School Lunch Is Better Explained by Objectively Measured Eating Behaviors than by Subjectively Rated Food Taste and Fullness: A Cross-Sectional Study. Nutrients. 2019 Mar 12;11(3):597.
https://doi.org/10.3390/nu11030597

II. Fagerberg P., Charmandari E., Diou C., Heimeier R., Karavidopoulou Y., Kassari P., Koukoula E., Lekka E., Maglaveras N., Maramis C., Pagkalos I., Papapanagiotou V., Riviou K., Sarafis I., Tragomalou A., Ioakimidis I. Fast Eating Is Associated with Increased BMI among High-School Students. Nutrients. 2021 Mar 9;13(3):880.
https://doi.org/10.3390/nu13030880

III. Langlet B., Fagerberg P., Glossner A. and Ioakimidis I. Objective Quantification of the Food Proximity Effect on Grapes, Chocolate and Cracker Consumption in a Swedish High School. A Temporal Analysis. PLoS One. 2017 Aug 10;12(8):e0182172.
https://doi.org/10.1371/journal.pone.0182172

IV. Fagerberg P., Klingelhoefer L., Bottai M., Langlet B., Kyritsis K., Rotter E., Reichmann H., Falkenburger B., Delopoulos A., Ioakimidis I. Lower Energy Intake among Advanced vs. Early Parkinson's Disease Patients and Healthy Controls in a Clinical Lunch Setting: A Cross-Sectional Study. Nutrients. 2020 Jul 16;12(7):2109.
https://doi.org/10.3390/nu12072109