Accurate valuation of dietary intake is a prerequisite in a nutritional epidemiology study. Acquiring dietary reporting that could accurately represent true dietary consumption (usual intake) is undeniably difficult due to cognitive abilities and dietary habits(Reference Forrestal1). Nutritional study often relies on subjective dietary intake due to the feasibility issue of questionnaires, notably in large-scale studies(Reference Nielsen and Adair2). Inevitably, data from self-reported dietary intakes tend to be misreported, which include both under-report and over-report. Under-reporting is particularly prevalent among adolescents; it varies between ten and 50 %(Reference Jessri, Lou and L’abbé3) and appears to happen both randomly and systematically(Reference Murakami and Livingstone4).
The extent of misreporting depends on several contributing factors, including instruments for dietary assessment, the method and cut-off used to identify implausible reporters(Reference Murakami, Livingstone and Okubo5). Error while reporting also happened when a person is selective to certain kinds of foods that are likely to be socially undesirable such as high fat, added sugars and alcohol, which is also known as selective misreporting(Reference Murakami and Livingstone4–Reference Murakami, Miyake and Sasaki6) or sudden change in eating behaviour, which happens if a person eats less or more than usual (under/over-eating)(Reference Black7) or could also be linked with a person’s level of adiposity or body weight status. This occurred when respondents with certain characteristics, for example, overweight and obese persons, under-estimate (under-record) the amounts of food consumed compared with their counterparts and vice-versa, also referred to as differential misreporting(Reference Murakami and Livingstone4–Reference Murakami, Miyake and Sasaki6,Reference Rangan, Flood and Gill8) . Nevertheless, study on bias reports of energy intake (EI) served as a surrogate assessment for investigating the discrepancy between self-reported dietary intake and actual dietary consumption(Reference Forrestal1).
Energy consumption is the basis of diet, as all nutrients must be provided within the amount of food required to meet the energy requirement. Thus, if the amount of EI is underestimated, it is plausible that consumption of other nutrients is also underestimated(Reference Livingstone and Black9). Currently, one of the widely used methods to determine invalid reporting of EI is the assessment of EI against presumed energy requirements. The reported EI which was estimated from equations expressed as multiples of the mean basal metabolic rate (BMR) was then compared with the presumed mean energy expenditure of the studied population, expressed as multiples of BMR(Reference Livingstone and Black9–Reference Black, Goldberg and Jebb11). This method that was first developed by Goldberg et al.(Reference Goldberg, Black and Jebb12) and later revised by Black(Reference Black7,Reference Black10) has been employed in several studies conducted among children and adolescents(Reference Murakami and Livingstone4,Reference Murakami, Livingstone and Okubo5,Reference Rangan, Flood and Gill8,Reference Lioret, Touvier and Balin13,Reference Kobe, Kržišnik and Mis14) .
Implausible reporting affects the estimation of EI and consequently of other nutrients. Data sets that included misreporting could reduce the accuracy of reported EI by increasing the risk of type 2 error (bias towards the null). In some instances, inaccurate reporting may diminish the usefulness of nutrition data by having the possibility to attenuate the association between food/nutrient intake and health/disease outcomes. Besides, the consequent probability for differential errors in dietary data may obfuscate the interpretation of diet–disease relationships and at worst could yield ambiguous links between diet and disease(Reference Forrestal1).
Thus far, there are no clear guidelines on handling misreporting, with only a limited number of studies that examined the statistical method of handling implausible EI in children(Reference Börnhorst, Huybrechts and Hebestreit15) and adolescents(Reference Jessri, Lou and L’abbé3). Moreover, study of misreporting among children and adolescents(Reference Murakami and Livingstone4,Reference Rangan, Flood and Gill8,Reference Lioret, Touvier and Balin13,Reference Kobe, Kržišnik and Mis14,Reference Bel-Serrat, Julián-Almárcegui and González-Gross16) , particularly in Asia, is sparse(Reference Murakami, Livingstone and Okubo5,Reference Murakami, Miyake and Sasaki6) . Furthermore, differences in dietary habits due to the heterogeneity of the study population may hinder the extrapolation of findings from other countries to heterogenic countries such as Malaysia and its neighbouring countries. This warrants an exclusive investigation that can be truly representative of the study population. Thus, the current study aims (1) to identify the prevalence of implausible reporters using a method based on the ratio of EI to BMR; (2) to identify the characteristics and lifestyle determinants associated with implausible reporters and (3) to compare the viability of statistical methods to correct misreporting of EI in a varied socio-demographic population of Malaysian adolescents.
Methods
This is a cross-sectional study conducted among the public secondary school adolescents (aged 15–17 years) in the central and eastern regions of Peninsular/West Malaysia: Selangor (central part), Federal Territory of Kuala Lumpur (metropolitan area) and Pahang (eastern region) state. Data collection was conducted between September 2014 and July 2015.
Sampling
Stratified random sampling was employed to select schools and participants. In summary, public schools were randomly selected to represent different niches based on the national school system, ethnicity and location (daily school-urban, daily school-rural, daily-vernacular, boarding, sports school and Orang Asli (Indigenous) dominated school)(Reference Zulfarina, Sharif and Syarifah-Noratiqah17). All fourth-year students in the selected secondary schools were invited to participate. For the purpose of this current study, however, only students who returned the agreed written consent forms (for participants and their parents or carers) and had no concurrent medical conditions were recruited into the study. All valuations were collated at the recruitment sites by a trained enumerator. In summary, 1078 adolescents from six public secondary schools were invited. A total of 960 adolescents returned with consented form; however, only 917 participants fulfilled the eligibility criteria to be included in the final analysis; twenty-two adolescents were not in the age range, fifteen adolescents were non-resident of Peninsular Malaysia (Indigenous of Borneo Island) or of other ethnicity and six adolescents had obscured FFQ answers.
Measurements
Dietary supplement intake was extracted using a validated dietary supplement questionnaire for adolescents(Reference Zulfarina, Sharif and Sharkawi18). Demographic information (gender at birth, age, ethnicity and school types), smoking and alcohol intake status were obtained as part of the questionnaire. Eligible adolescents were grouped to Malay, Chinese, Indian or Orang Asli for ethnicity, while urban, rural, boarding or sports school for types of school(Reference Zulfarina, Sharif and Syarifah-Noratiqah17). Participants were identified as either occasional (defined as drinking at most once a month) or non-drinkers (defined as never, tried once or two)(Reference Zulfarina, Sharif and Syarifah-Noratiqah17). While smoking status was categorised as smokers (at least one puffs a day, sometimes) or non-smokers (tried once and never)(Reference Zulfarina, Sharif and Syarifah-Noratiqah17).
Anthropometric and body composition
Height was measured using a portable stadiometer to the nearest 0·1 cm (model 213 SECA). Body weight was measured in light clothing on a portable body composition analyser (BC-420 MA Tanita) that integrates a bioelectrical impedance analysis, a validated technique that was used to estimate components of body composition including lean mass and fat percentage(Reference Talma, Chinapaw and Bakker19). All measurements were performed in triplicate and followed the standard procedure(Reference Gibson20). The age- and sex-specific BMI Z-score (BMI-Z) was calculated using the least mean square method(Reference Cole and Lobstein21). Weight status was defined using the new and extended International Obesity Task Force cut-offs reference(Reference Cole and Lobstein21). The new age-sex specific BMI for children and adolescents aged two to 18 years, cut-offs must be on the BMI percentile corresponding at age 18 years, through the values less than 18·5 kg/m2, more than 25 kg/m2 and 30 kg/m2 for underweight, overweight and obese, respectively.
Physical activity and sedentary behaviour
The individual physical activity level (PAL) was determined using a validated Malay short version of the International Physical Activity Questionnaire(Reference Craig, Marshall and Sjöström22). The frequency (number of d/week) and duration (min/d) of physical activity (walking, moderate-intensity activities and vigorous-intensity activities) undertaken during the previous 7 d was assessed. Energy expenditure in terms of multiples RMR or MET-minutes per week (MET-min/week) was computed for each type of activity. The total energy expenditure was calculated as the sum of vigorous activities, moderate activities and walking MET-min/week. Adolescents who performed vigorous intensity activity on at least 3 d achieving a minimum total score of 1500 MET-min/week or combination of vigorous PA, moderate PA and/or walking for at least 7 d achieving a minimum total score of 3000 MET-min/week were classified as high PA. Adolescents were categorised as moderate PA if they performed vigorous PA for at least 20 min/d for at least 3 d, moderate PA for at least 30 min/d for at least 5 d or a combination of vigorous PA, moderate PA and/or walking for at least 5 d, achieving a total score of at least 600 MET-min/week. Conversely, adolescents who did not achieve any criteria listed above were grouped as low physical activity(Reference Craig, Marshall and Sjöström22).
Participants were also asked to estimate the total time (min/d) of sitting or lying down during the weekdays for the past 7 d. Adolescents were asked to consider the time while at school, at home, while doing coursework and leisure time, including time spent sitting at desk, visiting friends, reading, sitting or lying down in front of screen. Total daily sitting time was used as a proxy of sedentary behaviour. Adolescents were categorised as high sedentary when the total sitting time was at least 8 h/d(Reference van der Ploeg, Chey and Korda23).
Dietary intakes assessment
Dietary habit was determined from a validated FFQ, designed explicitly for the local population and has a moderate average Pearson correlation of r = 0·416(Reference Fatihah, Ng and Hazwanie24). Adolescents were required to report the frequency of consumption for each of the ninety-four selected foods and beverages commonly consumed during the preceding month and the quantities/amount consumed at each occurrence. Examples of portion size estimated using local household measures were given to facilitate answering the questions. Each FFQ was reviewed for completeness and accuracy of the information. Any omission or ambiguities were clarified with participants via direct interview. The data obtained were converted into kilocalorie and nutrient intakes based on the Nutrient Composition of Malaysian Foods(Reference Tee, Ismail and Nasir25) and Atlas of Food Exchanges & Portion Size(Reference Suzana, Nik Shanita and Zaharah26) using the Nutritionist Pro™ Diet Analysis version 4 (Axxya Systems, Stafford, Texas). This present study includes nutrients that are presented in the latest and updated Malaysia Recommended Nutrient Intakes, which is the main reference for national nutrient intake(27).
Relative energy and nutrient intake
The Goldberg cut-offs or confidence limit of agreement was applied to determine invalid reporting of EI(Reference Goldberg, Black and Jebb12). The formula compares EI with total energy expenditure when both are expressed as multiple of BMR. In stable weight conditions, the ratio of EI to BMR is equal to the ratio of total energy expenditure to BMR (EI/BMR = total energy expenditure/BMR). The total energy expenditure: BMR ratio is also implied as PAL. Hence, the cut-off values are the 95 % confidence limits of agreement between the EI to BMR ratio and PAL. The BMR in kilocalories per day (kcal/d) was estimated using population-specific BMR equations validated for Malaysian adolescents(Reference Poh, Ismail and Ong28).
BMR equation for male adolescent:
$$\rm BMR (kcal) = [55\cdot8 \times body \ weight \ (kg) + 3187]/4\cdot184 $$
BMR equation for female adolescent:
$$\rm BMR (kcal) = [53 \cdot 4 \times body \ weight \ (kg) + 2182]/4\cdot184$$
In the Goldberg equation:
$$\rm 95\% CL=PAL \times exp \ [\pm 2 \times (S/100)/\sqrt{n}] $$
the number (n) was equal to one because detection of misreporting was at the individual level. As proposed(Reference Black10), the age-specific PAL (low PA = 1·6, moderate PA = 1·8 and vigorous PA = 2·0 physical activity) adapted by European Food Safety Authority for individuals of age 15–17 years were assigned to all individuals. The application of individual PAL in the equation is highly recommended since it has been indicated to enhance sensitivity without the loss of specificity(Reference Black7). While S is the factor that accounts for variation in EI, BMR and energy requirements are given by the equation:
$$\rm S = \sqrt\ [(CV^2_{wEI})/d + CV2_{wB}+CV^2_{tP}]\ $$
The CVwEI is the within-subject CV in EI; d is the number of recording day; CVwB is the within-subject CV in repeated BMR measurements and CVtP is the total variation in PAL. To calculate S for 30 d of food consumption (d) based on the FFQ, the proposed revised CV was substituted into the equation(Reference Black10) considering 23 % of CVwEI, 8·5 % of CV2 wB and 15 % of CVtP. After substituting the estimated values into the equation, adolescents who fell below the lower limit (EI: BMR < PAL) or above the upper limit (EI: BMR > PAL) of accurate reporting were considered as under-reporters and over-reporters, respectively, whereas adolescents were considered acceptable reporters if their calculated ratio of EI: BMR fell within the two limits.
Statistics
The assumption of non-normality for each variable including EI for sample sizes greater than 300 was determined based on histograms and skewness values larger than 2(Reference Kim29). All variables were found to be within an acceptable range. The descriptive mean values between dietary reporters were tested for differences with a two-way ANCOVA followed by Bonferroni post hoc after correction for age and gender. Determinants for misreporting were analysed using logistic regression. A univariate or crude model was conducted separately for each of the potential variables. To examine the independent association, a multivariate-adjusted logistic model was created by entering all variables that were found significant in the univariate analysis simultaneously.
The utility of the statistical method when investigating the relationship between BMI and EI was tested with multivariate linear regression. The following models were examined and compared with correct implausible reporters: the first model was adjusted for all covariables (including all participants), the second model included only plausible recalls, the third model was similar to the first model but additionally adjusted for the reporting groups (under-reporter and plausible reporter). The variance inflation factor calculated for each regression model revealed no critical levels of multicollinearity that warrant corrective measures. All statistical analyses were performed using the Statistical Package for Social Science (SPSS) version 24 software (SPSS Inc.). A P value <0·05 two-tailed was considered statistically significant.
Results
Characteristics of the study sample are presented in Table 1. The prevalence of under-reporter was 17·4 %. No adolescents were identified as over-reporters. Male adolescents were twice highly (23·4 %) to under-reporting than female adolescents (11·5 %). Adolescents of Chinese descent and urban areas under-reported more than adolescents of other races and school types. Adolescents who under-report had BMI-Z scores, percentage of fat, lean mass and BMR higher than acceptable reporters.
Table 1 Participant characteristics across categories of reporting status in Malaysian adolescents (n 917)
wt, weight; kcal, kilocalories per day; BMR, basal metabolic rate; PA, physical activity; SB, sedentary behaviour; DS, dietary supplements; MET-m/w, multiples RMR-min/week; BMI-Z, BMI Z-score.
* Significant difference between under-reporters and acceptable-reporters (P < 0 05) by two-way ANCOVA adjusted for age and gender.
† Adjusted mean.
Table 2 shows the dietary characteristics for different qualities of reporters. Under-reporters had significantly low dietary intakes compared with plausible reporters for all nutrients except for vitamin C, chromium and fluoride.
Table 2 Dietary characteristics across categories of reporting status in Malaysian adolescents (n 917)
kcal, kilocalories per day.
* Significant difference between under-reporters and acceptable-reporters (P < 0 05) by two-way ANCOVA adjusted for age and gender.
† Adjusted mean.
Table 3 shows the univariate and multivariate logistics regression model for determinants associated with under-reporting of EI. After significant variables in the univariate analysis were entered and adjusted in the multivariate analysis, the adjusted model demonstrated that adolescents with overweight had 5·5 times higher odds to under-report, while adolescents with obesity had thirty-nine times higher odds to under-report than under-weight/normal adolescents (P < 0·001).
Table 3 Factors associated with under-reporting in Malaysian adolescents (n 917)
Under-Rep, under-reporter; Accept Rep, acceptable reporter; wt, weight; PA, physical activity; SB, sedentary behavior; DS, dietary supplements; Ref, reference.
Table 4 shows the methods used to correct implausible reporters by examining the relationship between BMI-Z and EI. Model 1 which includes total sample found no significant association between the BMI-Z score and the reported EI (β = -7·66, P = 0·089), after the adjustment of demographic variables. After excluding under-reporters, the adjusted multivariate model shows a significant positive association between the BMI-Z score and the reported EI (β = 36·58, P < 0·001). Model 2 contributes to 29·3 % of EI variation (P < 0·001). After the additional adjustment for reporting status, model 3 exhibits a significant positive association between the BMI-Z score and the reported EI (β = 54·50, P < 0·001) and contributed to the highest regression coefficient (R 2 = 0·404, P < 0·001) compared with the previous models.
Table 4 Impact of under-reporting on the relationship between BMI-Z and energy intake (kcal)
β, unstandardised coefficient; R2, regression coefficient; Ref, reference; Under-Rep, under-reporter; Accept Rep, acceptable reporter; BMI-Z, BMI Z-score; Report, reporting status.
Models were tested and compared to correct for misreporting: model 1 was adjusted for all variables (including all participants) (n 917).
Model 2 includes only acceptable reporters (n 757), model 3 is the same as model 1 but with adjustment of reporting status (n 917).
Discussion
This is the first study to evaluate the prevalence, characteristics and determinants associated with misreporting of EI by applying the Goldberg cut-offs among the ethnically diverse population of Malaysian adolescents. Approximately one-fifth of adolescents were classified as under-reporter, whereas no over-reporter was identified in this cohort. Compared with normal weight, adolescents with overweight and obesity were associated with greater odds of being an under-reporter of EI.
The prevalence of implausible reporters varies widely by study population. On the basis of the Goldberg principles (EI/BMR ratio), the prevalence of under- and over-reporters was, respectively, 26·0 % and 0 % among the French adolescents aged 11–17 years(Reference Lioret, Touvier and Balin13), 24·1 % and 2·7 % in a representative sample of US children and adolescents aged 2–19 years (Reference Murakami and Livingstone4) and 4·9 % and 0·4 % among Japanese adolescents aged 15–19 years(Reference Murakami, Livingstone and Okubo5), while the prevalence of under- and over-reporters among the Australian boys and girls aged 14–16 years was 10–11 % (male), 14–15 % (female) and 2·0–2·1 % (male), 14·2–15·3 % (female), respectively(Reference Rangan, Flood and Gill8). In Slovenian adolescents aged 14–16 years, the prevalence of under-reporters was 34 % (male), 27 % (female), while the prevalence of over-reporters was 10 % (male), 11 % (female)(Reference Kobe, Kržišnik and Mis14). High prevalence of under-reporters was observed among French(Reference Lioret, Touvier and Balin13), USA(Reference Murakami and Livingstone4) and Slovenian adolescents(Reference Kobe, Kržišnik and Mis14), moderate in Australian adolescents(Reference Rangan, Flood and Gill8) while low in Japanese adolescents(Reference Murakami, Livingstone and Okubo5). Except for a study conducted in Australia(Reference Rangan, Flood and Gill8), most of the studies did not measure PAL, thus indicative PAL was assigned to represent average physical activity(Reference Murakami and Livingstone4,Reference Murakami, Livingstone and Okubo5,Reference Lioret, Touvier and Balin13,Reference Kobe, Kržišnik and Mis14) . In particular, study that uses FFQ showed a high prevalence of under-reporters(Reference Kobe, Kržišnik and Mis14); however, other studies that use dietary recall also reported a diverse prevalence of under-reporters ranging from low(Reference Murakami, Livingstone and Okubo5), moderate(Reference Rangan, Flood and Gill8) to high(Reference Murakami and Livingstone4).
The variabilities observed in the occurrence of under-reporting across countries may either reflect the actual discrepancy in the accuracy of reporting or disparities of the characteristics of the sample population or the cut-off applied to identify implausible reporters or the dietary assessment instruments or the food composition database. Nevertheless, the current study adds to the accumulating evidence that shows misreporting of EI is a focal issue in dietary surveys, which occurs non-randomly in children and adolescents(Reference Murakami and Livingstone4,Reference Murakami, Livingstone and Okubo5,Reference Börnhorst, Huybrechts and Hebestreit15) .
In the current study, implausible reporters were identified using the Goldberg equation and by applying individual PAL. Misreporting in the adolescent population was predominantly under-reporting. Previously, a study among female adolescents has shown that self-reported EI differs approximately 17 % less than the actual intake as estimated using doubly labelled water(Reference Kimm, Glynn and Obarzanek30). The current study also noted a moderate percentage of under-reporters (17·4 %). This result describes the acceptable estimation of EI (and possibly food and nutrient intakes) in this cohort. Among the factors that contribute to the high plausible reporters is the application of household-based dietary assessment and the application of wide confidence limits.
Several studies had indicated that under-reporting was more prevalent in female adolescents than in male adolescents(Reference Murakami, Miyake and Sasaki6,Reference Rangan, Flood and Gill8,Reference Bel-Serrat, Julián-Almárcegui and González-Gross16) . However, equal prevalence of misreporting between boys and girls has also been observed in some studies(Reference Murakami and Livingstone4,Reference Murakami, Livingstone and Okubo5,Reference Lioret, Touvier and Balin13) . On the contrary, in the present study, under-reports were more pronounced among adolescent boys (23·4 % among boys and 11·5 % among girls). The application of age- and sex-specific cut-offs to identify implausible EIs could justify the higher percentage of under-report in boys(Reference Kobe, Kržišnik and Mis14). Boys have higher total energy expenditure compared with girls and thus require higher EI. On the contrary, by applying single EI:BMR cut-off for both sexes, more boys fall above the cut-off and consequently more girls would be identified as under-reporters(Reference Livingstone and Black9).
While misreporting of EI was common among older children and adolescents, most national studies disregard examining implausible reporters among this age range. The tendency to under-report food intakes among this age group has also been consistently observed in several previous studies(Reference Murakami and Livingstone4–Reference Murakami, Miyake and Sasaki6,Reference Rangan, Flood and Gill8,Reference Lioret, Touvier and Balin13,Reference Kobe, Kržišnik and Mis14,Reference Bel-Serrat, Julián-Almárcegui and González-Gross16) . The higher possibility to under-report implies that dietary assessment in this age group is exceptionally challenging. Some of the reasons why this particular age group was more susceptible to under-report were probably due to the extra demands exerted on reporting by increased energy requirements that lead to a more significant amount of food to recall, unstructured habits of eating, substantial numbers of dining out and concerns with self-image(Reference Livingstone, Robson and Wallace31). Aligning with previous studies, under-reporting of EI affects the estimation of other nutrient intakes in which the macro- and micro-nutrient intakes were significantly lower in under-reporters than acceptable reporters(Reference Mirmiran, Esmaillzadeh and Azizi32). Other studies have also observed selective misreporting of socially desirable food, including high fibre intake among under-reporters(Reference Jessri, Lou and L’abbé3,Reference Murakami, Miyake and Sasaki6) .
A list of potential correlates was examined to identify lifestyle determinants of under-reporting in a multi-ethnic population. The outcome was consistent with the previous studies in which overweight and obese subjects were more likely to under-estimate their food/nutrient intake(Reference Murakami and Livingstone4–Reference Murakami, Miyake and Sasaki6,Reference Lioret, Touvier and Balin13,Reference Bel-Serrat, Julián-Almárcegui and González-Gross16) . Under-reporting may indicate the low ability of self-reporting of nutrient intake and denial in this age group (under-recording), but may also reflect real under-eating in an attempt to lose weight. However, body weight must be monitored to assess the degree of under-eating, which is best assessed in a longitudinal study(Reference Jessri, Lou and L’abbé3). For other determinants such as physical activity or sedentary, research in adolescents was limited to only one study that found under-reporting was significantly associated with increased physical activity and low screen time(Reference Bel-Serrat, Julián-Almárcegui and González-Gross16). This may be attributed to increased energy requirements or inaccurate reporting of frequency of eating or inaccurate reporting of portion sizes of substantial quantities of foods(Reference Rangan, Flood and Gill8).
The prevalence of such systematic differentials in this age group accentuates the significant need to correct for reporting errors. Thus far, only a limited number of studies have investigated/explored different statistical approaches to deal with implausible recalls among adolescents. In the first model, no significant association was observed between EI and BMI-Z, this result was in line with previous studies(Reference Jessri, Lou and L’abbé3,Reference Rangan, Flood and Gill8) . Conversely, among the acceptable reporters only (as identified in the second model), assessment on the association between EI and BMI-z exhibits that the exclusion of under-reporters led to the establishment of a significant positive association between EI and BMI-Z. Even though method 2 can strengthen the diet–disease relationships, it is clear that the exclusion of inaccurate reporters is not an appropriate approach and could result in selection bias(Reference Jessri, Lou and L’abbé3,Reference Börnhorst, Huybrechts and Hebestreit15,Reference Vainik, Konstabel and Latt33) .
The regression model with the highest regression coefficient (R 2 = 0·4) was displayed in model 3, after adjustment of reporting status. The overall model explained 40 % variance (model 3) compared with only 29 % of the variability of the data (model 2) when only acceptable reporters were comprised in the model. Under-reporting in the current study was found to influence the association between reported EI and BMI-Z, which emphasises the necessity to contemplate the impact of inaccurate reporters on the data’s validity. These findings corroborated the earlier study in a mixed population of adolescents and adults(Reference Jessri, Lou and L’abbé3). Despite the limitation of each method, researchers suggested that the best method to recover an association between BMI and EI is to create a dummy variable for the reporting group. Moreover, adjusting for the reporting group would provide a maximum sample size. On the other hand, removing a substantial number of subjects from the analyses would lead to loss of statistical power, and the outcomes can no longer be generalisable to the entire population.
Among the strengths of the current study is the assessment of PAL. Adolescents were assigned according to their individual PALs to increase the sensitivity of Goldberg cut-off method. Yet, the favourable assessment should be of objective measurement such as accelerometer. Still, with the absence of objective measurement, valid estimates of physical activity via questionnaire are better than no indication of specific physical activity categories. The current findings might be specific to this dietary evaluation and should be cautiously interpreted in this notion. The use of a shorter period of dietary assessment may contribute to the low level of misreporting(Reference Burrows, Martin and Collins34). However, it is known that several days of dietary assessment is preferred in providing precise estimates of usual dietary intake and reducing the within-person variability(Reference Thompson, Subar, Coulston, Boushey, Ferruzzi and Delahanty35). It is worthwhile noting that the substitution of the number of days (d) equal to 30 in the current study was contradicted with the previous study that also used FFQ to measure habitual dietary intake, d was taken as infinite and CVwEI disappeared(Reference Kobe, Kržišnik and Mis14). Nonetheless, this discrepancy is of no real consequence since the prevalence for the two approaches demonstrated a similar value, given that the differences in cut-off between the two approaches for all levels of physical activity were minor.
Misreporting that is linked with overweight and obesity (differential misreporting) should not be entirely precluded because removing subjects with particular traits could lead to selection bias. Likewise, a data set that included misreporting was found to attenuate the association between energy and BMI by increasing the risk of type 2 error. On the other hand, correction for misreporting strengthens the relationship, increases statistical power and better represents the population. Uncovering the problems associated with dietary assessments due to errors in estimation of EI does not suggest that dietary survey should be disregarded because the study of nutrition cannot be secluded from the reality of food intake(Reference Livingstone and Black9). The Goldberg cut-offs have demonstrated good accuracy in distinguishing inaccurate reports of EI, but misclassifications of under- or over-reporters may still have occurred(Reference Tooze, Krebs-Smith and Troiano36). However, in the absence of objective measures, the Goldberg cut-off is the preferred alternative to mitigate, at least in part, for the problem of energy misreporting. Moreover, the causality of the association presented in the current study cannot be ascertained owing to the cross-sectional nature of the study. To amend this issue, future studies should consider monitoring body weight longitudinally.
Conclusions
The present study demonstrated the presence of under-reporting in a dietary survey among various demographics of adolescents in Malaysia. However, over-reporting was not evident in this survey. Adolescents with overweight and obesity were more likely to under-report their food intake and consequently affect the estimates of nutrient intakes. As inaccurate energy reporting is the inevitable consequence of differential misreporting, future analyses (based on this data) should consider correcting for implausible reporters so that the influence of misreporting of EI on study outcomes can be acknowledged and consequently improve the precision and accuracy of results from dietary surveys.
Acknowledgements
Acknowledgements: The authors wish to express gratitude to the participants and teachers from the participating schools, Tg Mohd Ikhwan Tg Abu Bakar Sidik (Faculty of Medicine, UKM) for the statistical support, Mohd Amirul Mohd Suhaimi from Nutritional Science Program and Dr. Nurul Huda Razalli from Dietetics Program, Faculty of Health Sciences, UKM. Financial support: This research was financially supported by a university grant, Dana Impak Perdana (Premier Impact Grant) grant number DIP-2013-002. Authorship: Methodology, M.S.Z., R.S. and I.N.-M.; validation, I.N.-M., R.S. and A.N.S.; formal analysis, M.S.Z. and I.N.-M.; investigation, M.S.Z., A.M.S. and R.S.; resources, S.A.M., A.N.S. and I.N.-M.; data curation, M.S.Z., A.M.S. and I.N.-M.; writing – original draft preparation, M.S.Z.; writing – review and editing R.S. and I.N.-M.; supervision, I.N.-M., R.S., S.A.M. and A.N.S.; project administration, I.N.-M.; and funding acquisition, I.N.-M., S.A.M. and A.N.S. Ethics of human subject participation: The current study was conducted according to the guidelines laid down in the Declaration of Helsinki, and all procedures involving research study participants were approved by the Medical Ethics Committee of University Kebangsaan Malaysia Medical Centre (UKMMC) (Code: UKM 1.5.3.5/244/FF-2014). Formal approval was also obtained from the Ministry of Education (MOE), the Department of State Education (JPN), the Department of Orang Asli Development (JAKOA) and school authorities. Written informed consent was obtained from all participants and their parents or carers.
Conflicts of interest:
There are no conflicts of interest.
Open access
