Portion size and consistency as indicators of complementary food energy intake

Abstract We evaluated whether novel portion size and consistency indicators can identify children with low complementary food energy intake in southern Ethiopia. We conducted 24‐h dietary recalls with caregivers of 548 children aged 6–13 months; additionally, caregivers estimated their child's usual portion size using uncooked rice and selected which of five photographs of porridges of varying consistencies most closely matched the food their child usually ate. Complementary food energy and density from the 24‐h recall were used as reference values. We computed correlation coefficients and areas under receiver operating characteristic curves (AUC) and conducted sensitivity and specificity analyses to classify children with low complementary food energy intake. The median complementary food energy intakes for children 6–8, 9–11 and 12–13 months were 312, 322 and 375 kcal; median estimated portion sizes were 50, 58 and 64 ml, respectively. Estimated portion size correlated with total complementary food energy intake and with average energy and quantity consumed per feeding (r = 0.42, 0.46 and 0.45, respectively, all p < 0.001). Reported food consistency was weakly correlated with total complementary food energy intake (r = 0.18) and density (r = 0.10), and energy density of porridge only (r = 0.24, all p < 0.05). Predicted energy intake combining feeding frequency and portion size predicted inadequate energy intake better than did feeding frequency alone in infants 6–8 months [∆AUC = 0.16, 95% confidence interval (CI) 0.04, 0.28] and 9–11 months (∆AUC = 0.09, 95% CI 0.04, 0.14). Caregiver estimates of portion size can improve identification of infants with low complementary food energy intake when more robust dietary assessment is not feasible.

low-and middle-income countries is especially challenging (Foster & Adamson, 2014;Paintal & Aguayo, 2016;Quandt, 1987;Shankar et al., 2001). The World Health Organization (WHO) developed indicators of infant and young child feeding (IYCF) to enable populationlevel IYCF assessment for monitoring, evaluation and research (Ruel, Brown, & Caulfield, 2003) where more robust methods of assessment are resource-prohibitive. The WHO IYCF indicator definitions were published in 2008, with an ensuing implementation guide in 2010 (WHO, 2008(WHO, , 2010. The indicators have since been widely adopted but are limited in their scope (Ruel, 2017).
Feeding frequency, the number of meals and snacks consumed in a 24-h period, was previously evaluated as an indicator of energy intake using 10 data sets from nine countries in Africa, Asia and Latin America (Working Group on Infant andYoung Child Feeding Indicators, 2006, 2007) and is currently the only IYCF indicator of complementary food energy intake. Portion size and energy density are difficult to estimate using questionnaire-based approaches.
The objective of the present research was to estimate the added value of low-burden field methods for indicators of portion size and complementary food consistency when screening children at risk of low complementary food energy intake in a sample of children 6-13 months old in southern Ethiopia.

| Study setting and population
Data for this study were collected as part of a cluster-randomized nutrition-sensitive agriculture project evaluation in the Sidama and Gedeo zones in the Southern Nations, Nationalities, and Peoples' Region (SNNPR), Ethiopia. The trial consisted of a control group (n = 235), a partial intervention group (n = 143) and a full intervention group (n = 170). Both intervention groups received agricultural inputs and support for orange-fleshed sweet potato agriculture as well as community-based nutrition education that includes messaging on age-appropriate portion size and complementary food consistency. The full intervention group also received a feeding bowl and spoon designed to promote age-appropriate portion sizes and complementary food consistency (Collison et al., 2015;Kram et al., 2016).

| Data collection
Caregivers with infants less than 6 months completed a baseline survey in January 2018, immediately prior to the initiation of community-based nutrition education in intervention communities.
Households were considered eligible if they had an infant under 6 months without a serious health condition and (for intervention groups) if they were enrolled in implementation activities. A sample size of 600 households was sought at the baseline survey for adequate power to detect increases in energy intake of 100-150 kcal.
Households completed a follow-up survey in August 2018; at this time, households in the intervention group were nearing their 'graduation' from intervention activities, though activities were being scaled to other households within those communities. Because infants were all under 6 months at baseline, only data from follow-up are

Key messages
• Energy intake from complementary foods is a function of feeding frequency, energy density of complementary foods and amount consumed per feeding.
• Caregivers' estimate of usual portion size using uncooked rice is correlated with complementary food intake and energy intake.
• A five-photograph indicator of complementary food consistency is weakly correlated with complementary food energy intake and energy density.
• Portion size and feeding frequency combined better predict young children with inadequate energy intake from complementary foods, compared with feeding frequency alone.
used in these analyses. The exception is caregiver education, which was only assessed at baseline.
At follow-up, all infants were at least 6 months of age, but not more than 13 months. Therefore, we group the sample by the following age groups: 6-8, 9-11 and 12-13 months. The follow-up survey included WHO IYCF indicators (WHO, 2008(WHO, , 2010, novel indicators of usual portion size and complementary food consistency, and 24-h dietary recall of infant diets using the methodology described by Gibson and Ferguson (2008). Food composition, recipe and conversion factor databases were developed using primary data collection and/or pre-existing data from Uganda (Hotz, Abdelrahman, Sison, Moursi, & Loechl, 2012), Ethiopia (Ethiopian Health andNutrition Research Institute, 1968-1997)  Written informed consent was obtained from caregivers prior to each survey, and enumerators received training in research ethics.

| Reference intakes
Multiple-pass 24-h dietary recall was used to estimate total energy intake from complementary foods, average energy intake per feeding, average quantity (in grams) of complementary food consumed per feeding, average energy density of complementary foods and average energy density of porridges/gruels. If a child consumed no complementary food or liquid (other than water), then, his or her energy intake was set to 0 kcal, but energy density was considered missing.
Thresholds for classifying children as having low complementary food energy intake were as follows: <202 kcal for 6 to 8 months, <307 kcal for 9 to 11 months, and <548 kcal for 12 months and older (Dewey & Brown, 2003).

| Test indicators
To estimate portion size, caregivers were first asked whether their child normally eats from a shared dish or receives his or her own dish.
For children who receive their own dish, caregivers used uncooked rice to estimate a typical portion served to the child; this volume of uncooked rice was then transferred to a graduated cylinder and the volume recorded to the nearest millilitre (ml). The caregiver was asked if the child typically leaves any food remaining, and if so, was asked to use uncooked rice to estimate the amount of food uneaten, such that the amount consumed could be calculated as the volume remaining subtracted from the volume served. If the caregiver answered that the child typically eats from a shared dish, then the caregiver used the uncooked rice to estimate the amount consumed. To estimate complementary food consistency, caregivers were shown five numbered photographs of porridges with varying consistencies (Figure 1) and were asked to select which photograph most closely resembled the consistency of complementary foods eaten by their child. The five photographs depicted porridges with energy densities of 0.1, 0.4, 0.7, 0.9 and 1.3 kcal/g, respectively. Each photograph depicts a porridge prepared with maize flour and vegetable oil, which are locally available and were identified through formative research as common ingredients for porridge. Each ingredient was weighed, mixed with boiling water and stirred, then allowed to cool for 5 min before the final porridge was weighed (so that the energy density could be calculated) and photographed. These methods of assessing portion size and complementary food consistency are considered test indicators.
We computed predicted energy intake (PEI) as the product of portion size (millilitre per feeding) and feeding frequency (feedings per day), while assuming an energy density of 1.0 kcal/g and a F I G U R E 1 Photographs of porridge with energy densities (1) 0.1, (2) 0.4, (3) 0.7, (4) 0.9 and (5) 1.3 kcal/g, used to assess usual complementary food consistency in a household survey in Southern Nations, Nationalities, and Peoples' Region (SNNPR), Ethiopia complementary food density of 1.05 g/ml, which is the density of porridge prepared with maize flour according to the INFOODS Density Database Version 2.0 (FAO/INFOODS, 2012). We also computed a modified predicted energy intake (PEI-M) as the product of portion size (millilitre per feeding), feeding frequency (feedings per day) and the energy density of the photograph selected by the caregiver.

| Analytical approach
We assessed the distributions of continuous variables for normality based on skewness, kurtosis and visual inspection of histograms.
We computed correlation coefficients between references and test indicators. For portion size, references were total energy intake from complementary food, average energy intake per feeding and average amount of complementary food, in grams, consumed per feeding. For consistency, the references were total energy intake from complementary foods, average energy density of complementary foods and average energy density of porridges/gruels. We also assessed the correlations between feeding frequency and total energy intake from complementary foods and number of feeding episodes reported during the 24-h recall.
We used Pearson's correlation coefficients for normally distributed continuous variables and Spearman's rank correlation coefficients for non-normally distributed variables. Feeding frequency was treated as a continuous variable. Consistency was treated as an ordinal variable, for which we used polyserial correlation coefficients.
We computed correlations for the full sample and for subgroups of children whose caregiver reported that the child had consumed solid, semi-solid or soft foods in the previous day; by age category (6 to 8, 9 to 11 and 12-13 months); whether the child eats from a shared dish versus his/her own dish; by intervention group; by Sidama versus Gedeo zone; whether or not the child had been sick in the previous day; whether or not anyone in the household had fasted in the previous day; and whether or not the caregiver had completed primary school. We used Fisher's r-to-z-transformation to test whether correlation coefficients differed between these strata.
Using the conversion of 1.05 g/ml (FAO/INFOODS, 2012), we converted caregivers' estimates of usual portion size to grams to assess agreement between estimated portion size and average amount of food consumed per episode, both in grams, using a Bland-Altman plot.
We computed sensitivity, specificity and area under the receiver operating characteristic (ROC) curve (AUC) to assess the ability of the indicators to identify children whose total complementary food energy intake, as measured by 24-h recall, was low. For the sensitivity and specificity of portion size, we used 10-ml increments. We used the ROCCONTRAST statement in SAS 9.4 (SAS Institute Inc., Cary, NC, USA) to test whether any of the indicators' AUC was significantly different from feeding frequency. We then classified children as either at risk or not at risk of low complementary food energy intake based on whether the child met thresholds for portion size and consistency that were identified through sensitivity and specificity analyses. Using this dichotomous classification, we computed sensitivity and specificity for identifying children with low complementary food energy intake. Sensitivity, specificity and ROC curves were estimated separately for each age group (6-8, 9-11 and 12-13 months).
A p value less than 0.05 was considered statistically significant.
All analyses were conducted in SAS 9.4 (SAS Institute Inc., Cary, NC, USA).

| Ethical considerations
Ethical approval for this work was obtained from Emory University's Institutional Review Board, and from the Southern Nations, Nationalities and Peoples' Regional Bureau of Health Ethical Review Committee. The trial for which the data were collected is registered with ClinicalTrials.gov, ID NCT03423472.

| RESULTS
In total, 548 caregiver-child dyads completed the follow-up survey and were eligible for the analyses described here. General characteristics of the sample are shown in Table 1. Children ranged in age from 6 to 13 months (mean 10.0 ± 1.7 months), and 18.3% were reportedly sick in the previous day. A minority (6.4%) of households had a member who had fasted in the previous day, and 18.1% of caregivers had completed primary school. Only one child was not breastfed, and 96.2% of caregivers said their child received 'solid, semi-solid or soft foods' in the previous day. The mean number of feeding episodes was 3.6 ± 1.7. A majority of caregivers (87.3%) reported that the child typically receives his or her own feeding dish. Most children (65.1%) reportedly consume complementary food of medium consistency, matching the third photograph (0.7 kcal/g); almost 20% of children The complementary food consistency photograph number was weakly correlated with total complementary food energy intake (r = 0.18, p < 0.001), average energy density of complementary foods T A B L E 1 Characteristics and feeding practices of young children 6-13 months at follow-up, residing in Southern Nations, Nationalities, and Peoples' Region (SNNPR), Ethiopia, overall and by age category c Energy intakes greater than or equal to 202, 307 and 548 kcal for children ages 6-8, 9-11 and 12-13 months, respectively, are considered adequate. d An average energy density greater than or equal to 0.8 kcal/g is considered adequate.
(r = 0.10, p < 0.05) and average energy density of only porridges/ gruels (r = 0.24, p < 0.001). Correlation of the consistency photograph number with total complementary food energy was significantly greater in the partial (n = 164) compared with the full intervention group (n = 133), and among households in which no member fasted in the previous day (n = 470) compared with those who did (n = 32).
Reported feeding frequency was significantly correlated with both total complementary food energy intake (r = 0.41, p < 0.001) and with actual number of feeding episodes as reported in the 24-dietary recall (r = 0.66, p < 0.001). The strength of the correlations differed in some subgroups, but with few exceptions where strata sample sizes were small, remained significant in every group.
Each of the individual indicators, as well as PEI and PEI-M, significantly predicted low complementary food energy intake (Table 3).
When compared with the AUC of feeding frequency, the AUC of portion size was similar in each age group, whereas the consistency indicator performed significantly worse among children 9-11 months Portion size cut-offs of <50, <60 and <70 ml among children 6-8, 9-11 and 12-13 months, respectively, yielded sensitivities and specificities that were both at least 0.60 for classifying children with low complementary food energy intake (Table S1). No clear cut-off in consistency was identified (Table S2). In each age group, the second photograph was highly specific (0.77-1.0) but not sensitive (0.16-0.43).

| DISCUSSION
We found that caregiver estimates of portion size were correlated with their recall of complementary food intake and energy intake and that when combined with minimum meal frequency, these estimates  Table S5.
There is limited research with which to compare these findings.
Previous research on portion size estimates has been limited to recall of specific meals and has been conducted primarily in adolescents or adults and in high-income countries. Nevertheless, the observation that portion size estimates become less precise with increasing intake is consistent with previous research on portion size estimation using photographs (Ovaskainen et al., 2008;Turconi et al., 2005;Vereecken, Dohogne, Covents, & Maes, 2010). The mean difference between average portions consumed and caregivers' estimate showed a small overestimation of portion size and wide limits of agreement. Wide limits of agreement have been noted in other attempts to validate portion size estimation of specific meals (Flax et al., 2019;Turconi et al., 2005).
It is also important to note that there were some differences in indicators' performance by intervention group. Caregivers in the full intervention group were eligible to receive a feeding bowl and spoon designed to promote age-specific portion sizes and thick complementary foods, which could influence caregiver awareness and/or report.
Indeed, portion size estimates were more strongly correlated with complementary food energy and quantity per episode in the full intervention group. On the contrary, the consistency indicator was not correlated with total complementary food energy intake in the full intervention group.
Complementary food consistency has previously been assessed in surveys, though methods are not well described. To our knowledge, the validity of survey-based consistency indicators has not been assessed, and little is known about the relationship between consistency and energy intake outside of controlled feeding trials. It has previously been established that the consistency of grain-based porridges, as measured via a viscometer, is correlated with energy density (Treche & Mbome, 1999). However, viscosity is also impacted by temperature and other ingredients (Black, Pahulu, & Dunn, 2009;Mouquet & Treche, 2001

| Strengths and limitations
There are important limitations of this research. First, the sample is limited in its generalizability to older ages and contexts outside of southern Ethiopia. When feeding frequency was originally evaluated as a potential indicator of energy intake from complementary foods, it was done so in 10 data sets from nine countries, and some heteroge-

ACKNOWLEDGMENTS
Financial support for this work was provided by Emory University and the International Potato Center via a grant from the European Union.
The authors wish to thank the International Potato Center and People in Need SNNPR offices for their support with field work. We acknowledge Abdelrahman Lubowa for his support with dietary recall data collection and management.

CONFLICTS OF INTEREST
The authors declare that they have no conflicts of interest.

CONTRIBUTIONS
ADS, AWG and ECF formulated the research question; ECF and AWG led study design. ECF led the data collection and analysis; ADS and AWG provided input on the analysis and interpretation. ECF wrote the first draft of this manuscript and led manuscript revisions. All authors approved the final version.