=Paper=
{{Paper
|id=Vol-1683/hda16_guan
|storemode=property
|title=The feasibility of analysing food consumption combinations from overweight and obese participants of weight loss clinical trials
|pdfUrl=https://ceur-ws.org/Vol-1683/hda16_guan.pdf
|volume=Vol-1683
|authors=Vivienne Guan,Yasmine Probst,Elizabeth Neale,Allison Martin,Linda Tapsell
}}
==The feasibility of analysing food consumption combinations from overweight and obese participants of weight loss clinical trials==
https://ceur-ws.org/Vol-1683/hda16_guan.pdf
The feasibility of analysing food
consumption combinations from
overweight and obese participants of
weight loss clinical trials
Vivienne GUANa, Yasmine PROBSTa,b, Elizabeth NEALEa,b, Allison MARTINa,b and
Linda TAPSELLa,b
a
School of Medicine, Faculty of Science, Medicine and Health, University of
Wollongong, Australia
b
Illawarra Health and Medical Research Institute, University of Wollongong, Australia
Abstract. Overweight and obesity is a global epidemic. Investigating food
consumption combinations (FCCs) may offer useful insights into addressing eating
behaviours to manage overweight and obesity. Using food intake data generated
from a detailed dietary assessment method allows advanced analytical methods to
be employed. Food intake data collected by a diet history interview appears to be
more precise in terms of capturing the usual food intakes of individuals.
Exploration of FCCs can be conducted using the Apriori algorithm, but this
method is dependent on correct data preparation. Given the uncertainties related to
collecting food intake data via diet history interviews, the aim of this study was to
explore the feasibility of using food intake data derived from diet history
interviews from three weight-loss clinical trials to investigate FCCs. A 10%
random sample (n=62) of baseline paper-based diet history records, reflecting
usual food intake by meal, from three registered clinical trials (n=617) were
extracted. FCCs were assessed by considering the sum of single food items
consumed at the same time or in the same occasion using the United States
Department of Agriculture Food Combination Codes and the nested hierarchical
food groups of the 2011–13 Australian Health Survey food classification system.
FCCs were identified in all diet history data records at the major food group level.
A proportion of FCCs for the dinner meal (n=13) were unable to be assessed at the
specific food level due to limited detail for meat-containing FCCs. FCCs for the
dinner meal created more challenges for accurately distinguishing and naming
FCCs. Given the complexity of beverage reporting, combinations of foods and
beverages were not revealed in the selected data set. In conclusion, despite a lack
of meat-containing FCCs at dinner and food-beverage combinations, the food
intake data collected using the diet history interview method can feasibly be used
to investigate FCCs.
Keywords. Food intake, food consumption combinations, diet history, association
rule
Introduction
Overweight and obesity form a global epidemic with the prevalence of these conditions
increasing since 1980 [1]. By 2010, overweight and obesity contributed to 23% of
disability-adjusted life-years for ischaemic heart disease, but also to 3.4 million deaths
and 4% of years of life lost globally [2]. Although great effort has gone into managing
overweight and obesity including dietary guidelines that use dietary models to target
certain nutrients and foods [3], there is no reported successful case of turning the
problem around [4]. Research on nutrients, single foods and food groups has been used
as a basis for exploring diet-disease relationships that underpin obesity management
regimens. However, most foods are consumed in combinations whether as meals or
snacks [5]. For example, in Western diets, a savoury biscuit may be eaten with cheese;
steamed vegetables might be consumed with roasted meat for dinner. Thus,
investigations of food consumption combinations (FCCs) may offer an alternate
strategy for examining eating behaviours to manage overweight and obesity.
A number of methods for the collection of dietary data exist, including the diet
history interview, food record, and 24 hour recall. The diet history interview method
employs an open-ended interviewer-administrated approach to collecting data about an
individual’s usual food intake over a defined time period [6]. During the interview, a
trained interviewer asks the interviewee to describe food consumption generally from
the start of the day, such as the first food item consumed after waking, through to the
end of the day, before sleep. Based on the reported information, the interviewer applies
probing questions to assist the interviewee to recall and report what had been consumed
with the reported foods and in meals. Thus, data generated from a diet history interview
is more precise in terms of capturing the usual food combination intakes of individuals
than weighed food records and 24-hour recalls. However, the weakness of the diet
history interview is that the interviewer asks the interviewee to make judgements about
the food items and combinations through the types and timing of probing questions [7].
The effort and expertise of the interviewer, as well as the interaction between
interviewer and interviewee can play significant roles in the information captured. For
example, an experienced interviewer is able to ask further probing questions based on
the interviewee’s cues and responses to capture the ‘actual’ food consumption. This
might imply that information captured through a diet history interview may not be
compared in the same manner as other forms of dietary assessment.
With respect to the research trials themselves, different types of biases and errors
have been identified in obesity research [8]. The study design and analysis methods
used in obesity research have been criticised for their limited ability to translate to
health outcomes [9]. For example, dietary research that focuses only on single nutrients
may overlook the effects of combinations of foods or dietary patterns [10]. Conversely,
analytical methods required to examine food combinations are complicated and less
well explored than statistical methods [11]. This may imply that advanced analytical
methods need to be employed to contribute robust evidence for the associations
between body weight and food intakes, to provide meaningful insights towards more
effective obesity management strategies.
Examining FCCs is one way of looking at food patterns in a trial. FCCs can be
explored using association rules which are data mining tools used for identifying
certain relationships or combinations in a large data set [12]. To date, there are only
two published studies using a modified Apriori algorithm, which is one of algorithms
of association rules that successfully identified FCCs within a meal [13, 14]. In the
Apriori algorithm, two steps are conducted. The first step identifies frequent food item
sets in a meal. The frequent item sets are supported by a pre-defined support level,
which is the proportion of cases in the database containing the identified food item sets.
The second step is to generate rules by using identified frequent food items sets.
Apriori algorithms have previously been used in the literature to explore FCCs.
Woolhead et al only performed step 1 of the algorithm [14], while Burden et al
conducted both steps to generate the FCCs [13]. Apart from the Woolhead study using
pre-defined food groups, and Burden using specific food items, the discrepancies
between methods might be due to differences in the study aims. Using identified FCCs
to develop a generic meal code system to cover inter- and intra-variabilities of food
consumptions, required the result to provide a wide coverage of the possible FCCs. On
other hand, Burden et al aimed to use the outcome to develop software to assist dietary
data collection, where users could select a food item from a drop-down menu. This
required results to provide more accurate combination descriptions to improve the
efficacy of the questioning scheme for the software. Therefore, it appears that the two
steps of the Apriori algorithm may be used to investigate FCC behaviours to provide
robust evidence on the association between body weight and food intake.
Data preparation is a critical step of data analyses using data mining tools, to
accurately perform subsequent analyses [15]. Although FCCs were successfully
investigated using food record [14] and 24 hour recall [13] dietary data, there are a
number of uncertainties related to using food intake data collected by diet history
interviews. This paper aimed to better understand the feasibility of using food intake
data collected using a diet history interview method to examine the FCCs in an
overweight and obese population. The paper will address two specific objectives: (1) to
determine whether FCCs can be successfully identified using diet history interview
records from pooled data pertaining to three clinical trials; (2) to examine challenges
related to determining FCCs.
1. Methods
The basis of this work was diet history data from clinical trial participants. Hertzog has
suggested that a 10-15% sample for a testing group is sufficient to test the feasibility of
a study [16]. Thus, a 10% random sample (n=62) of baseline paper-based diet history
records of participants from pooled analyses of three registered weight-loss clinical
trials (n=617) were extracted as a pilot. Details of the trials have been described
elsewhere [17-19]. In an open-ended interviewer-administrated interview, self-reported
food intake data reflecting usual weekly (7 days) consumption was collected by
Accredited Practising Dietitians (APDs), followed by a food list to systematically
check for omitted food items. Meals, food items and their quantities were recorded on a
paper-based diet history proforma.
FCCs were defined as the sum of single food items consumed together at the same
time or in the same occasion, for example, toast, jam, and peanut butter reported as
eaten together at the breakfast meal. The combination was noted as one event of a FCC.
Firstly, FCC events of the extracted diet history records were identified and grouped by
meal (breakfast, lunch, dinner, mid-meals and beverages). Secondly, the nested
hierarchical food groups of the 2011–13 Australian Health Survey food classification
system, including the major, sub-major and minor groups, were used to assist in
assessing the food items of identified FCCs [20]. At the major food group level, there
are 24 groups, in which foods are grouped on the basis of the main nutrient or
ingredient (such as fruit products and dishes) [20, 21]. Foods are categorised at the sub-
major food group level based on species, family, and cooking and/or preparation
methods (for example citrus fruit) [20, 21]. Detailed or specific information related to a
food is included at the minor food group level, in order to identify and distinguish
foods from each other (for example an orange versus a lemon) [20, 21]. Lastly, the
United States Department of Agriculture (USDA) Food Combination Codes Scheme,
hereafter referred to the USDA codes was used to guide the categorisation of the
identified FCCs [22] (Table 1).
Table 1. The United States Department of Agriculture Food Combination Codes Scheme [22] and examples
Code Description Example
00 Non-combination Chocolate consumed alone
01 Beverage with additions Tea with milk and sugar
02 Cereal with additions Ready-to-eat cereal (Weet-bix) with milk and banana
03 Bread/baked products with Bread with margarine and jam
additions
04 Salad Lettuce, tomato, cucumber and avocado with dressing
05 Sandwiches Bread, butter, ham, cheese, tomato, lettuce and
mayonnaise
06 Soup Pumpkin soup or ready-to-eat soup made by powder
(liquid food)
07 Frozen meals Lean Cuisine
08 Ice cream/frozen yoghurt with Ice cream with chocolate sauce
additions
09 Dried beans and vegetable with Lentil curry (dried beans as the main ingredient for
additions the combination)
10 Fruit with additions Strawberry with yoghurt
11 Tortilla products Taco
12 Meat, poultry, fish Chicken and vegetable casserole
13 Lunchables Vita-weat biscuits with canned tuna
90 Other mixtures Omelette (eggs, cheese, ham and tomato)
2. Results
At the major food group level, FCCs were successfully identified in all extracted pilot
diet history records, such as meat with vegetables and starchy foods (for example rice,
pasta and potato products). Although FCCs of breakfast, lunch, mid-meals and
beverages were successfully identified at the sub-major and the minor food group
levels, FCCs were unable to be identified at the sub-major and minor level at the dinner
meal in 13 (21%) diet history records. This occurred when variations in meat (beef,
lamb, pork, and chicken) were recorded together. Thus, the specific meat item was
unable to be matched with subsequent vegetable and starchy foods to articulate FCCs
that were consumed together with the specific meat type.
Applying the USDA codes identified that 84% (n=52) of cases reported cereal with
additions (such as milk, sugar and/or fruit) and 55% (n=34) reported bread/baked
products with additions (such as spreads and eggs) at breakfast. A total of 92% (n=57)
of cases reported sandwiches at lunch. The number of variations in FFCs for dinner
was high (ranging from 1 to 9 combinations). However, the available USDA codes
were unable to cover all FCCs from the extracted dataset, particularly for dinner. For
example mixed dishes such as pasta dishes and shepherd’s pie were often reported for
dinner, but no USDA codes could be used to accurately reflect these FCCs.
The challenge identified for assessment of FCCs was in assessing the combination
of foods and beverages. Beverages were found to be reported with food (n=30), alone
(n=53), both with food and alone (n=13), and in the food frequency checklist at the end
of the diet history interview proforma (n=30). Additionally, beverages of eight data
records were reported with food, alone and in the food checklist. There was no
reporting trend for the characteristics of reporting non-alcoholic and alcoholic
beverages. Therefore, the reported combination of food and beverages may be unable
to be assessed using the available diet history data.
3. Discussion
The aim of this study was to determine the feasibility of using available food intake
data collected by the diet history interview method to assess FCCs of overweight and
obese participants in weight-loss clinical trials. Methodological issues associated with
identifying FCCs and challenges relating to data preparation used to undertake the
analyses were investigated. The results demonstrate that using diet history data
provides sufficiently detailed information on FCCs of breakfast, lunch, mid-meals and
beverages. However, the analysis process of this study has shown the complexity of
preparing dietary intake data to investigate FCCs, due to the variation in food
consumption between and within individuals. Specific challenges encountered related
to determining FCCs, specifically meat-containing FCCs, at dinner and the
consumption of beverages and food in combination.
The findings indicate that meat-containing FCCs at dinner could only be identified
at the major food group level (meat food group), rather than the minor food group level
(such as beef or chicken). This is due to red meat such as beef, lamb, pork, and veal and
white meat such as chicken being reported as alternate FCC options, but being
separated in the food classification system [20]. This may indicate that some eating
behaviours can still be identified, such as meat with vegetables and starchy foods;
however, the specific meat item consumption may not. Thus, given the high proportion
of unsuccessful FCCs for the dinner meal at the minor group level, eating occasions
such as main meals (breakfast, lunch and dinner) may need to be analysed separately,
particularly for the meat-containing FCCs at dinner.
In addition, this pilot demonstrated that the USDA codes might be too simple to
distinguish specific FCCs in different countries, suggesting that different eating habits
of different countries may need to be taken into consideration. The USDA codes could
only be used as a guide to categorise FCCs of overweight and obese participants of
weight loss clinical trials. Specific FCCs aligned with particular cultural food codes
may need to be created to accurately reflect true FCCs. Therefore, in order to
accurately perform subsequent analyses using the Apriori algorithm, additional
categories are required to accurately reflect true FCCs for an Australian overweight and
obese population eating context, such as pasta dishes and pie dishes.
Furthermore, food and beverage consumption combinations could not be identified
in the available data set. Although beverages may be consumed alone, such as coffee
with milk at morning tea, the results indicated that the beverage reporting practice in
the data set was not consistent. Due to the complexity of beverage reporting found
here, further investigations may need to focus on foods only.
In conclusion, food intake data collected by diet history interview can be used
successfully for investigating FCCs for breakfast, lunch, mid-meals and single
beverages. Meat-containing FCCs at dinner and the combined consumption of
beverages and food present challenges for identifying FCCs. To apply the present
methods to investigate FCCs in future studies, such as the Apriori algorithm, the tool
used to assess FCCs need to be modified or developed reflecting eating behaviours of
targeted population.
References
[1] Finucane, M.M., et al., National, regional, and global trends in body-mass index since 1980: systematic
analysis of health examination surveys and epidemiological studies with 960 country-years and 9·1
million participants, The Lancet 377(9765) (2011), 557-567.
[2] Lim, S.S., et al., A comparative risk assessment of burden of disease and injury attributable to 67 risk
factors and risk factor clusters in 21 regions, 1990-2013: a systematic analysis for the Global Burden of
Disease Study 2010, The Lancet 380(9859) (2010), 2224-2260.
[3] Hession, M., et al., Systematic review of randomized controlled trials of low-carbohydrate vs. low-
fat/low-calorie diets in the management of obesity and its comorbidities, Obesity Reviews 10(1) (2009),
36-50.
[4] Ng, M., et al., Global, regional, and national prevalence of overweight and obesity in children and adults
during 1980-2013: a systematic analysis for the Global Burden of Disease Study 2013, The Lancet
384(9945) (2013), 766-781.
[5] Scholderer, J., et al., Meal mapping, Food Quality and Preference 30(1) (2013), 47-55.
[6] Martin, G.S., et al., Relative validity of a diet history interview in an intervention trial manipulating
dietary fat in the management of Type II diabetes mellitus, Preventive Medicine 36(4) (2003), 420-428.
[7] Thompson, F.E. and A. Subar, Dietary assessment methodology, in Nutrition in the Prevention and
Treatment of Disease, A.M. Coulston, C.J. Boushey, and M.G. Feruzzi, Editors. Elsevier: Oxford,
England, 2013, 5-46.
[8] Ioannidis, J.P.A., Biases in obesity research: Identify, correct, endorse, or abandon effort? Obesity 24(4)
(2016), 767-768.
[9] George, B.J., et al., Common scientific and statistical errors in obesity research, Obesity 24(4) (2016)
781-790.
[10] Tapsell, L.C., et al., Foods, nutrients, and dietary patterns: Interconnections and implications for dietary
guidelines, Advances in Nutrition: An International Review Journal 7(3) (2016), 445-454.
[11] Leech, R.M., et al., Understanding meal patterns: definitions, methodology and impact on nutrient
intake and diet quality, Nutrition Research Reviews, 28(01) (2015), 1-21.
[12] Agrawal, R. and R. Srikant. Fast algorithms for mining association rules, in Proc. 20th int. conf. very
large data bases, VLDB. 1994.
[13] Burden, S., et al., Identification of food groups for use in a self-administered, computer-assisted diet
history interview for use in Australia, Journal of Food Composition and Analysis 22(2) (2009), 130-
136.
[14] Woolhead, C., et al., A generic coding approach for the examination of meal patterns, The American
Journal of Clinical Nutrition 102(2) (2015), 316-323.
[15] Zhang, S., C. Zhang, and Q. Yang, Data preparation for data mining, Applied Artificial Intelligence
17(5-6) (2003), 375-381.
[16] Hertzog, M.A., Considerations in determining sample size for pilot studies, Research in Nursing &
Health 31(2) (2008), 180-191.
[17] Tapsell, L.C., et al., Foods, nutrients or whole diets: effects of targeting fish and LCn3PUFA
consumption in a 12mo weight loss trial, BMC Public Health 13(1) (2013), 1-11.
[18] Tapsell, L.C., et al., Weight loss effects from vegetable intake: a 12-month randomised controlled trial,
European Journal of Clinical Nutrition 68(7) (2014), 778-785.
[19] Tapsell, L.C., et al., Interdisciplinary lifestyle intervention for weight management in a community
population (HealthTrack study): Study design and baseline sample characteristics, Contemporary
Clinical Trials 45, Part B (2015), 394-403.
[20] Food Standard Australia New Zealand. AUSNUT2011–13 food measures database file. Available
from:http://www.foodstandards.gov.au/science/monitoringnutrients/ausnut/classificationofsupps/Pages/
default.aspx. Accessed 29/02/16.
[21] Food Standard Australia New Zealand. AUSNUT2011–13 classification of foods and dietary
supplements. Available from: http://www.foodstandards.gov.au/science/monitoringnutrients/ausnut
/classificationofsupps/Pages /default.aspx. Accessed 08/07/16
[22] Centers for Disease Control and Prevention, Combination Code in Task 2: Key Concepts about the
United States Department of Agriculture Food Coding Scheme. Available from:
http://www.cdc.gov/nchs/tutorials/Dietary/SurveyOrientation/ResourceDietaryAnalysis/Info2.htm.
Accessed 30 May 2016.