Self-reported and measured weight , height and body mass index for the diagnosis of the nutritional status in people living with HIV / AIDS

1 Universidade Federal da Paraíba, Departamento de Nutrição, Programa de Pós-Graduação em Modelos de Decisão e Saúde. Campus Universitário I, 58051-900, João Pessoa, PB, Brasil. Correspondência para/Correspondence to: RLFC LIMA. E-mail: <rafaelanutri@gmail.com>. 2 Universidade Federal da Paraíba, Departamento de Morfologia, Programa de Pós-Graduação em Modelos de Decisão e Saúde. João Pessoa, PB, Brasil. 3 Universidade Federal da Paraíba, Centro de Ciências da Saúde, Programa de Pós-Graduação em Enfermagem. Paraíba, PB, Brasil. 4 Universidade Federal da Paraíba, Departamento de Estatística, Programa de Pós-Graduação em Modelos de Decisão e Saúde. João Pessoa, PB, Brasil. Self-reported and measured weight, height and body mass index for the diagnosis of the nutritional status in people living with HIV/AIDS

Self-reported and measured weight, height and body mass index for the diagnosis of the nutritional status in people living with HIV/AIDS Medidas de peso, altura e índice de massa corporal referidos e aferidos para diagnóstico de estado nutricional em pessoas vivendo com HIV/AIDS

I N T R O D U C T I O N
Self-reported measures of weight and height have been increasingly used owing to their simplicity, low cost and speed, but bias that may occur according to the characteristics of the individuals [1][2][3][4][5][6].Therefore, the validation of self-reported anthropometric data constitutes an important academic discussion [7].
While the self-reported weight and height method is appropriate for adults, it cannot be used for elderly individuals over 60 years old, since they demonstrably tend to overestimate their height [21].A wrong weight and height estimate leads to a variation in the resulting Body Mass Index (BMI), used as a parameter for anthropometric classification, thus leading to errors in prevalence estimates of overweight and obesity in population studies.To address this bias, it is necessary to validate the self-reported measures of weight and height [8,10,22,23].
The positive and promising point is that a high sensitivity and specificity have been observed for classifications based on BMIs calculated using self-reported measures when compared to BMIs calculated from measured values [10,24].However, these validations must take into account characteristics such as age, education level, income and the BMI of the population studied [8,10,24].
A specific vulnerable population group includes People Living With HIV/AIDS (PLWHA), and nutritional assessment constitutes a fundamental tool for decisions related to treatment and follow-up of these individuals [25], particularly because overweight has become the new nutritional problem in this group, surpassing low weight [26].
The possibility of using measures selfreported by PLWHA is an alternative that simplifies and saves time and resources and enables more frequent monitoring, but it requires validation, despite studies using these measures in their methodologies without concern for their validation [27][28][29].The validation work conducted by Duran et al. [30]. in a sample of PLWHA in the state of São Paulo constitutes an exception, showing good reliability and validity between the self-reported measures of weight and height when compared to their respective measured values.Authors recommend generating predictive equations based on the measured values of a subsample to increase the reliability and accuracy of the selfreported measures [4,30].
Therefore, the objectives of this study were to evaluate the validity of the weight and height data that are self-reported by PLWHA living in the Northeast region of Brazil, to determine the agreement and validity of the selfreported and measured values and to evaluate the sensitivity, specificity and predictive values of the individuals' nutritional status diagnoses obtained from self-reported measures, from measures corrected with their own (samplespecific) equations and from measures corrected with equations proposed by Duran et al. [30], all compared to the calculations made using the measured values.

M E T H O D S
A cross-sectional study, with prospective data colection was conducted with a sample of 481 PLWHA receiving antiretroviral therapy and followed up at the Clementino Fraga Infectious Diseases Hospital Complex in João Pessoa (PB), a reference unit in the care of these patients in the state of Paraíba.Patient selection was conducted through recruitment of all individuals who were seen at the unit from September 2, 2015 to December 23, 2015.This time period was selected independently of any department planning, representing a routine activity period at the hospital.
The inclusion criteria were being 18 years old or older, having a diagnosis of HIV/AIDS for more than six months, receiving receiving antiretroviral therapy, not presenting neurological or psychiatric disorders hindering evaluation and (for female patients) not being pregnant.
All patients were interviewed face-to-face by previously trained students and completed a structured questionnaire covering the following sociodemographic information: (a) sex: female and male; (b) age on the day of the interview in full years; (c) occupation: has an occupation/ active (employed, self-employed, student), has no occupation (unemployed, homemaker, retired or pensioner); (d) affective status: lives with partner, does not live with partner; (e) education level: incomplete elementary school, complete elementary school or higher; (f) household income, estimated as an absolute value; (g) number of people in the household in absolute value or categories: up to three residents in the household or more than three residents in the household, (h) income per capita (obtained as the household income divided by the number of individuals in the household) in absolute value and categories: up to half the minimum wage or more than half the minimum wage.The durations of disease and receiving antiretroviral therapy use were also obtained from the respective medical records, and weight and height were self-reported.Self-reported data were obtained through the questions: "What is your weight?" and "What is your height?".Thirty-three patients (6.9%) were excluded from the total sample because they did not know their weight and height.
Two of every 10 recruited individuals, starting from the first eleventh who came in the hospital, were systematically selected from the total sample included in the study.The sample size was based on the minimum sensitivity or specificity expected for the diagnostic test, as 80% and 5% precision, at 95% confidence interval.For this parameters it was necessary at least 163 measures.Forming a subgroup of 99 people who had their weight and height measured properly, we analyzed 198 measures.Each person in the sub sample did not know that he/she will be measured after self-answered his/ her height and weight.An electronic scale with a maximum capacity of 150kg and a sensitivity of 100g (Filizola ® , São Paulo, SP, Brazil) was used to measure weight.The individuals were weighed wearing light clothes and no shoes, standing and with their feet parallel, entirely supported on the scale's platform, and with their arms along their bodies [31].Height was measured using the scale-coupled stadiometer, which consists of a steel tube with an anodized aluminum ruler, measuring from 97cm to 192cm with 0.5cm increments.The individuals were standing, with their feet together and their heels touching the wall.The ear tip and the external eye corner were parallel to the floor, forming an angle of 90° with the stadiometer bar; the horizontal stadiometer bar was then lowered and supported by the head, allowing readings in centimeters [31].
The body mass index was calculated by dividing weight, in kg, by height squared, in meters.The values obtained were categorized as underweight, normal weight, overweight or obese according to the cutoff values established by the World Health Organization [32].
The study was approved by the Research Ethics Committee of the Health Sciences Center, Universidade Federal de Paraíba, under Protocol nº 0284/15.All of the ethical considerations established in Resolution nº 466/12 [33] were followed, and all of the participants consented to participate in the study by signing an Informed Consent Form.
The sample was described using central tendency and dispersion measures.The Kolmogorov-Smirnov test was used to confirm the normality of the continuous variables weight, height and BMI.
To assess the relationship between the self-reported and measured values for weight, height and BMI, the Intraclass Correlation Coefficients (ICCs), which evaluated the relationship between the groups, were calculated considering interpersonal variability, i.e., the ranked underestimation or overestimation among them.This measure is used when we wish to obtain an agreement between two measures for the same individual https://doi.org/10.1590/1678-98652017000500002[34].The Lin's Concordance Coefficients were calculated to estimate of how well a new measurement reproduces a gold standard test.It quantifies the agreement between these two measures of the same variable [35].
The agreement between the self-reported and measured values was evaluated through analysis of the mean differences between the methods with their limits of agreement, according to the method proposed by Bland & Altman [36].A plot of agreement between the mean and the difference between the measures was generated, enabling the evaluation of the magnitude of the differences and the respective 95% Confidence Intervals [37].
The predictive equations for weight and height were generated from the self-reported values using the linear regression method, using as a response variable the referred measures, and as an explanatory variable the measures measure [38].Sensitivity, specificity and positive and negative predictive values were calculated for validation by comparing the values measured against the following datasets: (a) self-reported measures; (b) measures corrected with their own predictive equation and, (c) measures corrected with the predictive equation by Duran et al. [30].The degree of agreement between measures was evaluated using the Kappa coefficient.All of the statistical analyses were performed using the statistical package Stata SE (College Station, Texas, United States), version 14 and it were significant p-value under 5% for statistic tests applied.

R E S U L T S
A total of 99 PLWHA were evaluated, with the majority being men (57.6%), with a mean age of 44 years old and a mean income per capita of R$538.50 (equivalent to US$145.50).The mean time since diagnosis for these individuals was 6.9 years (Standard Deviation [SD]=5.57),and the mean duration of receiving antiretroviral therapy use was 6 years (SD=5.33).There were no statistically significant differences (Chi-square test) between the total sample and the subsample, which had their values measured, for the variables sex, age, income per capita, affective status, time since diagnosis and duration of receiving antiretroviral therapy.The only statistically significant difference (p=0.035) between these two groups was the fact that the subsample presented a lower frequency of individuals with higher education level.
The continuous variables, weight, height and BMI, measured and self-reported, showed normal distribution according Kolmogorov-Smirnov test.
Upon calculating the linear regression, the following predictive equations for weight and height were generated for the sample according to sex (57 men and 42 women), with their respective coefficients of determination: The means of all values (measured values + self-reported values/2) and the differences (measured value -self-reported value) were determined for the subsample stratified by sex (Table 1).In general, the differences between the measures were small; the largest differences were 0.579kg for weight in women and 0.392 and 0.358kg/m 2 for BMI in men and women, respectively.The confidence intervals were generally small.According to the ICCs, these measures present good correlations between themselves (Table 2).Women tend to underestimate weight and overestimate height, which results in an underestimated BMI.Exactly the opposite occurs with men, with overestimation of weight and underestimation of height, which results in an overestimated BMI.High correlations were observed (r>0.90) between the self-reported and measured values for weight, height and BMI since the ICCs presented values very close to 1, indicating high reliability between these measures.The ICC values for the self-reported measures, the measures corrected with the predictive equations obtained with the sample itself and the measures corrected using the equations proposed by Duran et al. [30], when comparing with the measured values, are almost similar.
In the Bland & Altman [36] analyses (Figure 1), a good agreement was observed between the self-reported and measured values; practically all of the data points in the scatter plots are within the limits of agreement for the weight, height and BMI measures.People showing the highest weights have a greater tendency to underestimate weight than other individuals, which partly has effects on the BMI.
Nutritional status was diagnosed using the self-reported measures, the measures corrected based on the predictive equations generated with the sample partitioned by sex and the correction equations proposed by Duran et al. [30] (Figure 2).Based on the self-reported measures, the prevalence of low weight in the sample was small, approximately 4.0%.This value precluded the calculations of sensitivity, specificity and predictive values for those individuals.The prevalence rates of overweight and obesity were 26.2% and 31.0%for women and 49.1% and 8.8% for men, respectively.
For individuals diagnosed as normal and overweight, the diagnostic test evaluation measures were calculated (Table 3).The four underweight patients, based on the self-reported measures, were excluded from the calculation.The sensitivity, which represents the proportion of individuals effectively with normal weight or overweight who were diagnosed using the self-reported and the corrected measures as such, was high-namely, higher than 90.0% for     Mean BMI (kg/m ) 5,00 normal weight and overweight for all measures (reported, corrected based on the samplespecific equations and corrected using the equations generated by Duran et al. [30]).
Specificity represents the proportions of individuals classified as not having normal weight or being overweight using the selfreported measures and who in fact do not have the classified condition.All of the values were higher than 90.0%, and the self-reported BMI classification for overweight individuals was perfect (100.0%).The positive and negative predictive values were all higher than 85.0%.Accuracy presented equal values (92.6%) for all measures, except for overweight diagnosed from the self-reported BMI, which showed a higher value (94.6%).The degrees of agreement of the measures, evaluated using the Kappa coefficient, presented the same accuracy trend and were considered to be very good, all equal to 0.85, except overweight diagnosed by the self-reported BMI, which showed a higher value of 0.89.

D I S C U S S I O N
The results presented in this study suggest that weight and height data selfreported by PLWHA are valid for the diagnosis of nutritional status and thus corroborate the validation of BMI use in the identification and monitoring of the nutritional status of these individuals, as reported in a prior study [30].Therefore, the recognition of the use of these data as a valid tool for the nutritional assessment of these individuals may constitute an important indicator in the clinical follow-up of PLWHA, in addition to be rapid, convenience and low cost.When incorporated into the follow-up routine, this assessment can promote empowerment and self-care among these individuals.
Although there have been few studies conducted for this purpose, the results obtained in this study can fortunately be compared to those obtained in the study by Duran et al. [30] since the samples presented similar sociodemographic and clinical characteristics.Therefore, it was possible to apply and compare the equations generated in the first study together with the new equations proposed in this study.An important methodological approach adopted in this study, and which did not add much complexity from the practical point of view and regarding the use of mathematical equations, was to build two models stratified by sex.This approach improved the sensitivity of the normal weight classification and the specificity of the overweight classification.
The predictive equations showed high coefficients of determination (r 2 ) and are adequate to correct the self-reported measures.https://doi.org/10.1590/1678-98652017000500002Despite the improvement in the corrected data, the ICCs observed, which evaluate the relationship between the groups by comparing two paired measures, suffered few changes and presented good results for both the selfreported measures and the measures corrected using the equation proposed in this study or that previously proposed by Duran et al. [30].In other studies, the use of correction equations reduced the mean difference but did not resolve difference variability, classification errors and association biases [20].A positive aspect of the self-reported measures of weight and height among PLWHA is that no relevant variabilities were observed in the differences between the values for each individual, and the coefficients showed to have values close to 1, which would be the ideal situation.
The intraclass correlation coefficients observed were high and validated the use of self-reported BMI for similar population groups.Such an expressive result was not observed in prior studies with healthy adults or in studies conducted through phone surveys [13,39].The fact that the PLWHA frequently visit the hospital for monitoring their clinical status and adjusting their receiving antiretroviral therapy may contribute to a greater knowledge of their weight, considering that this measurement is important for the adjustment of drug doses.
In this study, the information selfreported by the women led to an overestimation of their nutritional status for overweight and an underestimation for obesity, whereas the normal weight estimate was very close to the actual value measured.In the case of men, the trend was slightly different: the estimates of obesity were equivalent to the true value measured, but overweight was underestimated and normal weight was overestimated.In the women's case, it is possible that the self-reported weight was lower in cases of obesity, which would explain this variation.Among men, the high value they assign to height could be responsible for the bias observed between the levels of overweight and normal weight, but this effect, if real, would also influence the classification of obesity, which did not occur.These considerations indicate that it would be necessary, in addition to using correction equations for weight and height, to consider the underestimations of weight for obese women and overweight men as potential sources of bias.These results were different from those obtained in a study conducted in women living with HIV/AIDS, which showed a tendency to overestimate their body weight [30].Additional studies should be conducted, especially because this information is highly related to subjective aspects concerning selfimage, cultural and social patterns and the exclusion stigma still associated with AIDS.
Although not representative, it is important to highlight the difference in the patterns of overweight presented between men and women.Although both present a frequency of approximately 58.0%, obesity exceeded cases of overweight among women, whereas obesity was five times less frequent than overweight among men and the overall frequency of underweight in the sample was low.These values are similar to those in other studies, for example, a study conducted among Brazilian individuals living with HIV/AIDS showed prevalence rates of underweight of 7.7% for women and 2.3% for men [26].In a study conducted in the United States, a prevalence of overweight of more than 50.0%was observed for PLWHA [40].
The Kappa coefficient, sensitivity, specificity and the positive and negative predictive values were sufficient to recommend the use of BMI estimation from self-reported values, which were further improved by using correction equations, except for the assessment of overweight, for which the self-reported measure had a specificity of 100%.However, this assessment does not take into consideration the differential behaviors of men and women when classified as overweight vs. obese.

Figure 1 .
Figure 1.Bland and Altman plots for the limits of agreement of self-reported and measured weight, height and Body Mass Index (BMI) in People Living With HIV/AIDS (PLWHA).João Pessoa (PB), Brazil (2015; N=99).

Figure 2 .
Figure 2. Diagnosis of the nutritional status in People Living With HIV/AIDS (PLWHA) according to the values self-reported, measured and corrected with their own equation or the Duran equation.João Pessoa (PB), Brazil (2015; N=99).
Overweight Obese BMI measured BMI own corrected BMI corrected by Duran BMI Self-reported

Table 1 .
Means and confidence intervals for all values (measured and self-reported) of differences between self-reported and measured values among People Living With HIV/AIDS (PLWHA), by sex.João Pessoa (PB) Brazil, 2015.

Table 2 .
Intraclass Correlation Coefficients (ICCs) and Lin's Concordance Coefficients (LC) between the self-reported values and measured values, corrected with their own equation and the equation by Duran in People Living With HIV/AIDS (PLWHA).

Table 3 .
Sensitivity, specificity, positive predictive value, negative predictive value, accuracy and Kappa coefficient of the BMIs calculated from self-reported and self-reported and corrected values of weight and height in People Living With HIV/AIDS (PLWHA), compared with the Body Mass Index (BMI) calculated from measured values.João Pessoa (PB) Brazil, 2015.