点击显示 收起
1 From the Equipe de Nutrition, Laboratoire de Physiologie, Département de Biologie Animale, Faculté des Sciences et Techniques, Université Cheikh Anta Diop de Dakar, Sénégal, West Africa (AD, ASC, and SW); the Nutrition Unit, WHO Collaborating Centre for Nutrition, Institut de Recherche pour le Développement, Montpellier, France (AG, FD, and BM); and the Institute of Physiology, Faculty of Medicine, University of Lausanne, Switzerland (YS).
2 Supported by the Institute of Research for Development, Montpellier, France. AD was supported by a doctoral research fellowship from the Institute of Research for Development. 3 Address reprint requests to A Gartner, Nutrition Unit, UR 106, Institut de Recherche pour le Développement, BP 64501, 911 Avenue Agropolis, 34394 Montpellier Cedex 5, France. E-mail: gartner{at}mpl.ird.fr.
ABSTRACT
Background: Little information is available on the validity of simple and indirect body-composition methods in non-Western populations. Equations for predicting body composition are population-specific, and body composition differs between blacks and whites.
Objective: We tested the hypothesis that the validity of equations for predicting total body water (TBW) from bioelectrical impedance analysis measurements is likely to depend on the racial background of the group from which the equations were derived.
Design: The hypothesis was tested by comparing, in 36 African women, TBW values measured by deuterium dilution with those predicted by 23 equations developed in white, African American, or African subjects. These cross-validations in our African sample were also compared, whenever possible, with results from other studies in black subjects.
Results: Errors in predicting TBW showed acceptable values (1.3–1.9 kg) in all cases, whereas a large range of bias (0.2–6.1 kg) was observed independently of the ethnic origin of the sample from which the equations were derived. Three equations (2 from whites and 1 from blacks) showed nonsignificant bias and could be used in Africans. In all other cases, we observed either an overestimation or underestimation of TBW with variable bias values, regardless of racial background, yielding no clear trend for validity as a function of ethnic origin.
Conclusions: The findings of this cross-validation study emphasize the need for further fundamental research to explore the causes of the poor validity of TBW prediction equations across populations rather than the need to develop new prediction equations for use in Africa.
Key Words: Bioelectrical impedance analysis • BIA equation • total body water prediction • validity • deuterium dilution • African women
INTRODUCTION
Water is the major chemical component of the body and is an essential medium of the body's internal environment. Total body water (TBW) is constantly maintained in healthy subjects, and it is frequently measured to evaluate body composition—a sensitive indicator of health and nutritional status. Simple methods are needed for epidemiologic field studies, notably in low-income countries; the most widely used method is bioelectrical impedance analysis (BIA) (1, 2). BIA is often used to estimate body fat and muscle, but it is important to recall that the conductor of the body is its water content, and a BIA analyzer actually measures the impedance of this fluid conductor, thus enabling assessment of TBW. However, BIA measurement is an indirect method from which body composition is predicted with the use of statistical relations derived in other groups against a reference method.
Prediction formulas for body composition tend to be population-specific (3-5). Previous results, mainly obtained in white subjects, might be inappropriate in black subjects. Because blacks differ from whites in several aspects of body composition (6, 7), the validity of the white equations in Africans needs to be tested (8), as previously shown in Asians (4). We recently showed that white equations are invalid in Africans when segmental impedance measurements are used (9, 10).
Initially, differences in body-composition estimations by BIA between whites and blacks were studied in populations from the United States. BIA may be effective in predicting TBW when race is specified (11). Prediction equations derived from white or predominantly white subjects were more valid for whites than for blacks (12, 13). Despite numerous investigations of BIA, minimal data are available on Africans. Prediction formulas developed in whites with reference to dilution techniques overestimated body water in Ethiopians (14, 15) but showed no significant differences between whites and Nigerians (16). Some equations for the prediction of TBW were developed in African Americans (11, 13, 16) and one was developed in Africans (17). The usefulness of these published equations in other samples remains unknown, especially in Africans.
The central hypothesis of this study was that prediction equations developed specifically in black subjects are likely to lead to better validity when cross-validated in Africans than are prediction equations developed in predominantly white populations. A second hypothesis to be tested was that single-predictor equations based on impedance index alone are likely to yield less accurate predictions than are equations that also use anthropometric measures as additional predictor variables.
The aim of this study was to apply the BIA method to a group of African women and to compare TBW values predicted by BIA with reference values measured by the deuterium oxide dilution technique (18-20). Predicted values were obtained by applying published single or multiple predictor variable equations developed in white, African American, or African subjects.
SUBJECTS AND METHODS
Subjects
The sample included 36 women volunteers who were recruited from the community in a periurban neighborhood of Dakar, the capital city of Senegal (West Africa). The physical characteristics of the subjects are shown in Table 1. A local community center was used for recruitment. Women were included provided they were between 18 and 55 y of age, they were not pregnant, and their youngest child, if any, was 9 mo old. The ethics committee of the University of Dakar approved this study. All subjects gave their written consent to participate in the study after being thoroughly informed of its purpose, requirements, and procedures. Measurements were performed at the Institute of Research for Development Centre in Dakar. All measurements were performed in the morning between 0830 and 1300 with a room temperature of 22.5–25.5 °C. The subjects fasted overnight from food and drink, did not perform strenuous exercise, and emptied their bladders preceding the measurements. Immediately after the BIA and anthropometric measurements were performed, deuterium water was administered orally to each subject, thus allowing the BIA and reference tests to be done at the same time and under similar conditions. Consequently, the hydration status was likely to have remained throughout the measurement period.
View this table:
TABLE 1. Characteristics of the African women1
Anthropometric measures
Measurements were made by trained personnel using standard procedures (21). Wearing minimal clothing, the subjects were weighed to the nearest 0.01 kg with an electronic scale (Tanita Corp, Tokyo, Japan). Height was measured to the nearest millimeter with a portable gauge (Seca, Hamburg, Germany). Sitting height was measured to the nearest millimeter while the subjects were sitting on a stool. The height of the stool was subtracted from the measurement. Leg length was calculated as height minus sitting height. Relative leg length was calculated as leg length divided by height (cm/cm). Anthropometric measurements were the mean of duplicates. All measurements were made by the same observer (AG).
BIA
BIA was performed on the left side of the body with a body-composition analyzer (Xitron 4000B; Xitron Technologies, San Diego, CA) with a 4-electrode arrangement. The adhesive electrodes were paired, one pair acting as current electrodes and the other pair acting as detector electrodes. The electrodes were placed on the hand, wrist, foot, and ankle of each subject according to the standard placement for adults stated in the manufacturer's guidelines. At each site, the skin was cleaned with alcohol before the electrodes were placed. All metal objects were removed from the subjects before the measurements were made. The subjects were supine on a nonconductive surface with their arms and their thighs apart; the measurements were performed after the subjects had remained in the supine position for 15 min. The calibration of the instrument was checked daily with the use of standard resistors purchased with the analyzer. All BIA measurements were performed by the same person. The values used in the calculations were the mean of duplicate measurements. Only the resistance (R) or impedance (Z) data at 5, 50, and 100 kHz were used in the calculations of the present study. The impedance index was calculated as height2/R (cm2/) or height2/Z, depending on the explanatory variable used in the equation tested. The impedance index can be assumed to reflect conductor volume and has been found to correlate highly with laboratory estimates of TBW volume and fat-free mass in humans (2).
The current is able to pass through cell membranes at high frequencies but not at low frequencies, thus enabling the prediction of extracellular water (ECW) and TBW independently at 5 and 100 kHz, respectively (22). Differences in TBW distribution over the intra- and extracellular spaces are reflected by impedance ratios of 5-100 kHz frequencies (1). The length2/R5 is assumed to reflect the ECW volume, and the length2/R100 is assumed to reflect the TBW volume. Their ratio was used as a simple index of the ECW/TBW ratio, as we had done in previous studies in African subjects (23, 24).
Deuterium oxide dilution technique
Saliva from the women (5 mL) was collected directly in small sterile vials. The subjects first provided a predose saliva sample to determine the natural deuterium content. Then, an accurately weighed dose of 30 g deuterium oxide (D2O; 99.8% purity, Cambridge Isotope Laboratories Inc, Andover, MA) was administered orally to each subject regardless of body weight. After the subjects took the complete dose, the dose bottle was weighed again to check the exact amount taken. Postdose saliva samples were collected after administration of the dose on days 1, 2, 3, 4, 13, and 14. The samples were centrifuged for 5 min at 11 500 x g, and the supernatant fluid was kept at –20 °C until analyzed.
Enrichment of the saliva samples was measured with a Fourier transformed infrared spectrophotometer (FTIR; Shimadzu 8300; Shimadzu, Vienna, Austria) equipped with an automatic sample shuttle and a pair of matched calcium fluoride sample cells with a 0.1-mm path length. The precision of the measurements was tested and validated with an isotope ratio mass spectrometer (18, 19). This FTIR equipment was available at the University of Dakar and was previously used to measure deuterium saliva enrichment in infant mother pairs for breast-milk output (25, 26). Before the saliva measurement was made, the calibration procedure involved preparation of the D2O calibrator by dilution of D2O with deionized water. The enrichment of our calibrator was confirmed by the isotope ratio mass spectrometer in the Medical Research Council's Human Nutrition Research (Cambridge, United Kingdom). In a previous study, measurement of the serial dilution (range: 110-990 ppm) of our calibrator with our FTIR equipment showed a coefficient of correlation of 0.99 and a root means square error (RMSE) of 7.5 ppm, reflecting good precision of our FTIR method. For analysis, the pre- and postdose samples were simultaneously loaded into the instrument and automatically positioned in the light beam. This minimizes any interference due to the absorption of atmospheric carbon dioxide in the sample chamber. The infrared spectra were measured in the range 2300-2800 cm–1. The slope intercept multipoint procedure was used to determine the water pool size (27). A curve fitting to multipoint data set developed by the Medical Research Council's Human Nutrition Research was used. Each postdose saliva sample was measured in duplicate. The CV of the duplicates ranged between 0.0% and 2.4% (
Four kinds of equations were tested in our African sample, depending on the ethnic origin of the population in which they were developed. First, the equations developed in white subjects (11, 13, 16, 22) were retained because their validity was also tested in black subjects or because, in the same study, equations in blacks were also developed for comparison purposes (11, 13, 16). Second, we tested prediction equations developed in predominantly white populations (2, 29-32). In these cases, the ethnic origin of the US population was not reported, and it was assumed that the sample was a mixed US population of whites and blacks. Third, we retained equations from studies in which it was specified that the samples comprised whites and blacks in the United States (12, 13). Finally, we tested equations established in African Americans (11, 13, 16) and one prediction equation that was very recently established in Africans (17). In each case, whenever possible (ie, when provided by the authors), 2 types of equations developed in the same sample were tested: 1) the one-predictor equation, which has only the impedance index as an explanatory variable, and 2) the final equation retained from the regression analyses with added anthropometric characteristics. In such cases, the 2 equations are identified by the same letter, followed by a different number. All of the equations are presented in Table 2. The equations used impedance and anthropometrics dimensions that can be easily measured in the field. Some detailed information about their initial development and validation is also given in Table 2. RMSE, often expressed as the SEE, was reported from the articles cited in the reference and indicated whether these predictive equations fit well to the data from which they were generated, ie, their precision was assessed. Some articles expressed the TBW in liters and the others in kilograms. Because the conversion factor between volume and weight of TBW is 0.9937, ie, the density of the water at 36 °C, we considered this difference to be negligible and compared the TBW results regardless of the units used.
View this table:
TABLE 2. Published bioelectrical impedance analysis (BIA)–based prediction equations (dummy code) tested for the prediction of total body water (TBW) in the African women in the current study1
Statistical analysis
The statistical software used for data entry and validation was EPI-INFO (Centers for Disease Control and Prevention, Atlanta, GA). Statistical analyses were performed by using the SAS system for WINDOWS (release 8.0; SAS Institute Inc, Cary, NC). Values are expressed as means and SDs. First type error risk was set at 0.05 for all analyses.
The intrasubject reliability of the BIA measurement was tested by performing duplicates. The intrasubject differences were calculated as absolute values. The technical error of measurement [(intrasubject difference)2/2 x number of duplicates], and the percentage reliability (technical error of measurement x 100/overall mean of the measurements) were calculated.
BIA equations from the literature were cross-validated on our sample of African women. Pearson's correlation coefficients were used to study the relation between the measured and predicted values of TBW. The difference between measured and predicted values (bias) of TBW was tested against zero (paired t test). Error was determined as the SD of the bias. The dependency of the bias on the mean of measured and predicted values was tested by using correlation analysis (33, 34). The performance of the prediction equations, when applied to a sample independent of the one used to construct the equation (cross-validation), was measured by pure error (35). Pure error is calculated as the square root of the sum of squared differences between the observed and the predicted values divided by the number of subjects in the cross-validation sample. It summarizes the differences between observed and predicted values and is mathematically the same as the SEE often used for precision. The smaller the pure error, the greater the accuracy of the equation. There is no criterion value for the pure error that indicates successful cross-validation, but the pure error should be similar to the precision (ie, RMSE value) of the equation from its validation in the population from which it is derived.
RESULTS
The intrasubject reliability of whole-body BIA measurements was estimated as part of a larger study in 137 women, including the 36 subjects in the present study. Absolute values of the difference between replicates of R50 ranged from 0 to 5.2
View this table:
TABLE 3. Cross-validation of the published bioelectrical impedance analysis (BIA)–based equations for the prediction of total body water in the African women in the current study1
Only one equation (F) showed a significant correlation (R = –0.56, P = 0.0004) between the bias and the mean of measured and predicted TBW values; for all other equations the correlation was not significant (P > 0.10). Some anthropometric (age, BMI, and relative leg length), impedance (impedance index and ECW/TBW ratio index), and hydration (TBW/body weight and TBW/height2) characteristics of the subjects were tested for their correlation with the bias (detailed data not shown). A significant correlation of the bias was found with age for only 2 equations (B1 and C), with BMI for 2 other equations (F and H) and with relative leg length for 4 equations (F, G1, H, and O1). For 10 equations (B1, E, F, G2, H, I1, I2, N, O1, and O2) the bias was related to the impedance index, and for 4 equations (F, H, J1, and L1) it was related to the ECW/TBW ratio index. For 8 equations (A1, A2, B2, C, G3, J1, L1, and M) we observed a significant correlation between the bias and the TBW index, and for 15 equations (A1, B1, C, D, F, G1, G3, H, I1, J1, K, L1, M, N, and O1) we observed a significant correlation between the bias and the ratio of TBW to body weight.
Of the 7 comparisons we made between single- and multiple-predictor equations developed in a single sample, 6 (A, B, G, J, L, and O) showed significantly different biases (Table 3). Interestingly, for all of the 8 single-predictor equations, the bias was not related to the ratio of TBW to body weight, whereas in all the other equations where body weight was an additional predictor, the bias was related to the ratio of TBW to body weight.
DISCUSSION
It is necessary to know whether published predictive equations for body composition are usable in Africa, where new equations are likely to be rare. In the present study we examined cross-validation of BIA-based equations as a function of the ethnic origin of the sample in which they were developed for prediction of TBW, which is directly related to body impedance. Using TBW measured by deuterium oxide dilution as the frame of reference, we examined the validity in African women of TBW estimates with 23 published prediction equations derived from 16 different samples. Moreover, whenever possible, the results of a cross-validation of the same equations performed by other authors in black subjects were considered (Table 4).
View this table:
TABLE 4. Review of articles that tested the validity of published bioelectrical impedance analysis (BIA)–based equations for the prediction of total body water that were used in the current study
Equations resulting in marked bias (C, M, and O2) were developed in samples that probably included considerable differences in mean age or in range (Table 2) compared with those in our sample (Table 1). Equations C and M were developed in older subjects, and the inclusion of age among their predictor variables apparently did not correct for the effect of differences in age between samples, even if equation M was established in blacks. Equations O1 and O2 were the only equations developed in African subjects, but in a sample that included children. Such a wide range of age, also observed for equations I1 and I2, probably compromised the opportunity to really test these equations. Unfortunately, equations C, I, M, and O have not been tested by other authors in Africans so it is presently difficult to know whether a difference in age across samples is the main cause of the bias. The inaccuracy of equation E could be explained by the fact that the original sample comprised only males; however, the other equation developed in males only (H) showed a lower bias value. Moreover, when cross-validated by other authors, equation E underestimated TBW in African American women in the same way as in our sample, whereas it performed well in white women (Table 4).
Equations G1, G2, and L2 can be retained as usable in Africans, notably L2 as it was established in blacks, in a larger sample than G, and its bias showed no correlation with characteristics of the subjects. Moreover, when cross-validated by other authors, equation G1 was just as valid in Africans as in African Americans and, surprisingly, was not as valid in whites (Table 4). When cross-validated in black subjects in other studies, equations A1, B1, B2, and F (derived from whites) showed weak bias, as in our sample, whereas equation D was twice as accurate as in our study (Table 4). When tested by other authors in 2 African groups, equation N (derived from African Americans) showed a bias similar to that in our sample (Table 3). When tested in the same 2 samples of Africans, equation D (derived from whites) surprisingly showed better accuracy when compared with equation N (derived from blacks) (Table 4). Conversely, in our sample of African women, equation D was less accurate than was equation N (Table 3).
Total-body impedance is largely determined by the impedance values of the arms and the legs (37, 38), whereas most body water is located in the trunk. Whole-body impedance makes assumptions in the calculation of TBW, namely a constant ratio between TBW in the arms and legs compared with that of the total body. The relative extremity length (to body height) differs in blacks and whites (7, 39, 40) and may influence the relation between body impedance and TBW (39). In our study, the sitting height/height value was 0.50 (ie, relative leg length of 0.50), which is lower than the typical mean European value of 0.52 (40). This finding indicates that our sample was likely to have a higher value for relative leg length than that for whites. However, the hypothesis of the influence of differences in body build across ethnic groups on the validity of BIA-based predictions (4, 5) seems to be incorrect in the case of TBW assessment because in the present study, and in a few studies by other authors, the validity appeared to be independent of the ethnic origin of the equation. One explanation could be that this hypothesis was generally reported for percentage body fat or fat-free mass estimation, whereas TBW requires less indirect evaluation and fewer assumptions than does the assessment of lean or fat compartments. Indeed, in the present study, on only 3 occasions did we find a significant correlation between the bias and the relative leg length; in one case, this may have resulted because the original sample included children who likely had a different relative leg length to adults.
The prediction of TBW from an impedance index could be improved by additional independent predictors such as body weight, age, and sex, which could correct for differences across groups. When Deurenberg et al (14) applied equations B1 and B2 in a sample of Ethiopian women, the single- and multiple-predictor equations both led to an overestimation of 1 kg (Table 4), whereas in our sample the bias of equations B1 and B2 differed significantly.
For the purpose of cross-validation, we compared results obtained with different BIA devices or dilution techniques, which may explain differences in TBW estimates. The Xitron 4000B BIA device we used showed no significant differences from the RJL 101 device (41) used for producing 20 of the 23 equations tested. One equation (H) was established by using the Danninger TVI-10 BIA device (Danninger Medical, Columbus, OH) for which we found an outstanding correlation coefficient (R = 0.99) when compared with the Xitron 4000B in a separate study on 142 women (data not shown). Two equations (B1 and B2) were produced by using Dietosystem (Milano, Italy) BIA. Twenty equations used the deuterium dilution technique (with extrapolation to zero intercept or one sample at 3–4 h dilution time), as we did in our sample, and 3 equations (C, H, and M) used the tritium dilution technique. TBW can be measured by using these 2 different techniques with a precision and accuracy of 1-2% (20), and they do not differ by >1–2% (32). Therefore, it is unlikely that methodologic differences in TBW techniques between studies played an important role in the observed error.
The hypothesis of better validity for an equation derived from black subjects when cross-validated in Africans, compared with equations from whites, is clearly inconsistent with the results from the large panel of equations we tested in African women. Our findings confirm the recent statement that BIA prediction equations (for TBW) differ between African Americans and whites, but it is not clear why (16). The assumptions of differences in body build as reasons for invalidity across ethnic groups appeared to be nonvalid in the case of TBW assessment. Because the development of new prediction equations is a difficult and expensive process in low-income countries, we suggest focusing research efforts on cross-validating published equations. The absence of a clear trend in cross-validation among equations according to their ethnic origin should encourage further exploration of the causes of the lack of validity. Meanwhile, 3 equations tested (G1, G2, and L2) and developed >10 y ago could be used to assess TBW in Africans. Alternatively, for longitudinal or internal comparative purposes, crude parameters such as the impedance index or the ECW/TBW ratio index could be used.
ACKNOWLEDGMENTS
We are indebted to the women who participated in this study.
FD and BM initiated the research project for this study and helped write the manuscript. AG was in charge of the study in Senegal and was the principal investigator. AD was in charge of recruiting the subjects, collecting data from isotopic dilution, and writing the manuscript and participated in the data management and analysis. ASC and SW helped treat the saliva samples, conduct the FTIR spectroscopy measurements, and write the manuscript. AG and YS were responsible for the data management and analysis and helped write the manuscript. The authors confirm that there was no conflict of interest because the study received no external sponsorship.
REFERENCES