How accurate are self-reported height, weight, and BMI among community-dwelling elderly Japanese?:...

ORIGINAL ARTICLE: EPIDEMIOLOGY,CLINICAL PRACTICE AND HEALTH

How accurate are self-reportedheight, weight, and BMI among

community-dwelling elderlyJapanese?: Evidence from a

national population-based studyggi_759 247..256

Vanessa Yong and Yasuhiko Saito

Advanced Research Institute for the Sciences and Humanities (ARISH), Nihon University, Tokyo, Japan

Aim: The aims of this study are to investigate the accuracy of self-reported height,weight, and derived BMI of community-dwelling Japanese aged 70 and older, and to assessthe concordance between report-based and measurement-based BMI categories.

Methods: We compared self-reported height and weight with physical measurementsfrom data from a nationally representative sample (n = 1634).

Results: Self-reported values were strongly correlated with measured values (Pearson’s r:0.92 and 0.89 for men and women, respectively, for height; 0.96 for both sexes for weight;0.93 and 0.91 for men and women, respectively, for BMI). The differences in mean valueswere small. On average, height was overreported by 0.93 cm (SD = 2.48) for men and1.23 cm (SD = 2.84) for women. Weight was underreported by 1.08 kg (SD = 2.55) for menand 0.88 kg (SD = 2.45) for women. BMI was underestimated by 0.68 kg/m2 (SD = 1.16)for men and 0.79 kg/m2 (SD = 1.49) for women. As age increases, height overreportingincreased, particularly among women, but weight underreporting decreased for women.BMI underestimation increased for both sexes. Weighted kappa values showed a reasonablyhigh concordance at 0.715 and 0.670 for men and women, respectively (P < 0.0001).Overweight (BMI 25.0 to 29.9) and obesity (BMI 3 30) prevalence rates were underesti-mated, with better specificity (range: 94.4–100%) than sensitivity (range: 59.3–65.1%).

Conclusion: The accuracy of self-reported height and weight is reasonably high amongelderly Japanese, suggesting that the information can be used in epidemiological surveys.However, caution should be exercised for the oldest age group (age 85+), as the accuracydeclined. Geriatr Gerontol Int 2012; 12: 247–256.

Keywords: body mass index, elderly, height, Japan, weight.

Introduction

Large-scale epidemiological surveys have typically reliedon self-reports of height and weight to derive BMI for

estimates of overweight and obesity prevalence in apopulation. The use of self-reported measures presentsa number of advantages: obtaining the information isnon-invasive and inexpensive, and the measures canbe easily and quickly obtained for a large number ofpeople. However, systematic biases in self-reportsbecause of factors such as social desirability as well asgender and age differences in reporting may exist. Areview of studies that examined the validity of self-reported height and weight showed a general pattern ofoverreporting of height and underreporting of weight,

Accepted for publication 30 August 2011.

Correspondence: Dr Vanessa Yong PhD, Advanced ResearchInstitute for the Sciences and Humanities, Nihon University,12-5 Goban-cho, Chiyoda-ku, Tokyo 102-8251, Japan. Email:[email protected]

Geriatr Gerontol Int 2012; 12: 247–256

© 2011 Japan Geriatrics Society doi: 10.1111/j.1447-0594.2011.00759.x � 247

with a result of an overall underestimation of BMI.1–6

The overreporting of height tends to be more prevalentamong shorter people and men while the underreport-ing of weight is more prevalent among heavier peopleand women.1–7 The accuracy of reporting height andweight has also been found to decrease in older agegroups.8,9

A systematic review of studies compared direct mea-surements with self-reported height, weight, and BMIto assess the accuracy of self-reports; the reviewshowed that most of the studies involve Westernnations, younger age groups (e.g. university students,employees), and special populations (e.g. individuals inweight-loss programs, eating-disorder programs, orpostmenopausal women).10 Until now, there have onlybeen two Japanese studies that have examined thevalidity of self-reported height, weight, and BMI. Thefirst is on a group of employees aged 35–64 yearsin a workplace in Aichi Prefecture,11 and the second ison a sample of 368 Japanese women aged 20–42 yearswho participated in a survey on eating disordersin Fukushima Prefecture.12 Both studies focusedonly on young to middle-aged adults and a specialsubgroup. To our knowledge, there have been nostudies, prior to this one, that have examined the accu-racy of self-reported height, weight, and BMI forelderly Japanese using national population-based data.Given that Japan’s population is aging rapidly, withalmost one-quarter (23%) of the population currentlyover the age of 65 and the concurrent rise in obesitylevels in Japan,13,14 it is imperative to ascertain the accu-racy level of commonly self-reported measures used toderive estimates of obesity prevalence among theelderly.

We compared self-reported height and weight withdirect physical measurements to assess the accuracy ofself-reports. In addition, we calculated BMI based onself-reported height and weight and compared it withmeasurement-based BMI to check for discrepancies andmisclassifications of weight categories. We conductedthe analyses on a national sample of elderly Japaneseaged 70 years and older, stratified by sex and 5-year agegroups.

Methods

Data and sample

Data for this study came from the fifth wave of theNihon University Longitudinal Study of Agingthat was conducted between March and June 2009.The sample for this nationally-representative study ofelderly Japanese aged 65 and over was drawn from the47 prefectures in Japan using a multi-stage probabilitysampling method. The first wave of the face-to-facesurvey was conducted in 1999 with 4997 subjects at

their homes; subsequent waves were conducted in2001, 2003, 2006, and 2009. The sample was refreshedwith new subjects in 2001 and 2003. Details of thissurvey can be found elsewhere and have not beenrepeated here.15,16 We used the fifth wave in 2009 forthis study because height and weight were measuredfor the first time, in addition to the self-reported datafrom the survey questionnaire. In 2009, all subjectswere aged 70 years or older, and only subjects whoprovided information on all four measures – self-reported height, self-reported weight, measured height,and measured weight – were included in the analysis ofthis study. The final analytical sample included 1634elderly Japanese (783 men and 851 women) aged 70 orolder, out of a total of 2570 respondents in the fifthwave of the survey.

Measures

Self-reported height and weight were obtained beforethe measurements were taken and were reported incentimeters and kilograms, respectively, to the nearestmetric unit. To obtain measured height, subjects wererequested to remove their shoes and stand with theirfeet and back directly against a wall or pillar. A self-adhesive note was placed on the wall near the top ofthe subject’s head and a rafter square was placed onthe head parallel to the floor and making a 90° anglewith the wall. After the subject had stepped away, amark was made on the self-adhesive note with apencil. A carpenter’s ruler was used to measure thelength from the floor to the mark on the note. Of the1634 subjects, 101 chose not to remove their shoeswhen their height was measured. The height of these101 subjects’ shoes was also measured (range: 0.5–5.0 cm) and subtracted from the measured height.Measured weight was obtained using a standard TanitaHD-355 digital scale (Tokyo, Japan) that the interview-ers brought with them. Subjects were requested toremove items from their pockets, heavy sweaters, andshoes. Only 19 subjects (1.2% of the sample) chose tokeep their shoes on when they were weighed. Heightand weight were measured to the nearest 0.1 cm and0.1 kg, respectively. These measurements were takenon the same day at the end of the survey interview.Prior to the home visits, the interviewers were trainedto obtain the anthropometric measurements accuratelyand uniformly.

BMI was calculated as weight in kilograms divided bythe square of height in meters (kg/m2). Report-basedBMI were calculated from self-reported height andweight. Measurement-based BMI were calculated frommeasured height and weight.

The Institutional Review Board of the Nihon Uni-versity School of Medicine (Tokyo, Japan) providedethical approval for the study. Subjects also signed

V Yong and Y Saito

248 � © 2011 Japan Geriatrics Society

consent forms to have their anthropometric datacollected.

Statistical analysis

To compare the characteristics of the analytical samplewith those subjects who were excluded from the sample,we computed means and distribution of selected vari-ables for these two groups. We also conducted a t-testfor interval level variables and c2 test for categoricalvariables to test for statistical differences between thetwo groups. The selected variables include age, sex,marital status, level of education, income, number ofactivities of daily living difficulties (physical function),number of words immediately and delayed recalled(cognitive function), and having at least one chronicdisease. We also compared self-reported height, self-reported weight, and self-report-derived BMI betweenthe two groups in order to identify possible bias in theanalytical sample.

To investigate the accuracy of self-reported height,weight, and BMI by sex and age, all analyses were con-ducted separately for men and women and for 5-yearage groups (ages 70–74; 75–79; 80–84; 85+). We useddescriptive statistics (means, standard deviations, andpercentages) to analyze the data. We calculated themean difference between self-reported and measuredheight, weight, and BMI by subtracting the measuredmean values from self-reported mean values to ascertainthe magnitude of the bias in the self-reported data. Apositive difference indicated overreporting and a nega-tive difference indicated underreporting. The differ-ences between self-reported and measured mean valueswere tested for significance using paired t-tests. Pear-son’s correlation coefficient was used to assess the cor-relation between the self-reported and measured values.We further used a decomposition method to ascertainthe extent of BMI differences that are due to weight andheight reporting differences.17

The report-based BMI weight categories were cross-tabulated with measurement-based BMI weight catego-ries to assess the extent of the level of agreement. Inclassifying BMI into weight categories, we used theWorld Health Organization’s international classifica-tion: underweight (BMI < 18.5); normal weight (BMI18.5 to <25); overweight (BMI 25 to <30); and obese(BMI 3 30). We did not further subcategorize the obesegiven the very small proportion of Japanese in this cat-egory. Although other classifications have been pro-posed and intermittently used for Asian populations,18

we used the World Health Organization’s classificationfor consistency and comparability with most existinginternational studies. Weighted kappa statistic was usedto test the degree of agreement between self-reportedand measured BMI weight categories. The prevalenceestimates of weight categories based on self-reported

and measured values were computed, and sensi-tivity tests (accurate identification of cases) and specific-ity tests (accurate identification of non-cases) wereconducted.

The data were weighted to represent the nationalpopulation of Japanese aged 70 years or older in 2009.The statistical analyses were carried out using SAS v.9.1.3 Service Pack 4 (SAS Institute, Cary, NC, USA).

Results

Table 1 shows the selected characteristics of samplepersons who are included in and excluded from theanalyses. As shown, those who are included in theanalyses are younger, more likely to be women, moreeducated, and more economically well-off, have fewerdifficulties with daily living activities, and have bettercognitive function compared to those excluded from theanalyses. All these attributes significantly differ betweenthe two groups. The proportion of those with at leastone chronic disease is not significantly differentbetween the two groups. Differences in the means ofself-reported height and self-report-based BMI aresmall and are not significantly different. However, thedifference in means of self-reported weight between thetwo groups is significantly different.

The age of the subjects ranged from 70 to 95 yearsold, and 51.5% of the sample were women. The meanage for men and women is very close at 77.2 (SD = 5.36)and 77.3 (SD = 5.44) years, respectively.

Table 2 shows the means of measured height, weight,and measurement-based BMI, and the mean differenceswith self-reported values by sex and age groups.The average height of Japanese aged 70 and overis 161.01 cm (SD = 6.32) for men and 147.83 cm(SD = 6.18) for women. For both sexes, measuredheight decreases with age. At ages 70–74, the averagemeasured height is 162.62 cm (SD = 6.20) for men and150.12 cm (SD = 5.61) for women. By age 85, it isreduced to 159.02 cm (SD = 5.07) and 143.83 cm(SD = 4.61), respectively. Both sexes and all ages tendto overreport height relative to measured height, butthe differences are small. The overreporting of height isgreater among women than men, with women onaverage overreporting 1.23 cm (SD = 2.84) and men0.93 cm (SD = 2.48). Height overreporting generallyincreases with age, particularly for women. The averagedifference in overreporting at ages 70–74 is only0.66 cm (SD = 2.10) for men and 0.76 cm (SD = 1.94)for women, but it increases to 1.00 cm (SD = 2.36) formen and 2.26 cm for women by age 85. Pearson’s cor-relation coefficient showed that both measured and self-reported height are very strongly correlated at 0.92 formen (P < 0.0001) and 0.89 for women (P < 0.0001). Thecorrelation drops slightly at older ages but maintainshigh statistical significance.

Self-report derived BMI in Japan

© 2011 Japan Geriatrics Society � 249

The average weight of elderly Japanese is 60.39 kg(SD = 9.50) for men and 51.05 kg (SD = 8.49) forwomen. Similar to measured height, measured weightalso decreases as age increases. For both sexes andacross all age groups, measured weight is greater thanself-reported weight, indicating an underreporting ofweight. The weight underreporting is greater for menthan women at, on average, 1.08 kg (SD = 2.55) for menand 0.88 kg (SD = 2.45) for women. Interestingly, asage increases, women’s self-reported weight generallybecomes more accurate, but the largest differencebetween self-reported and measured weight was forthe oldest age group (age 85+) of men at 1.43 kg(SD = 2.36). Overall, measured and self-reportedweight are very strongly correlated (Pearson’s r = 0.96,P < 0.0001).

From measured height and weight, we computedmeasurement-based BMI. The average BMI is23.27 kg/m2 (SD = 3.24) for Japanese men and23.35 kg/m2 (SD = 3.56) for Japanese women aged 70and over. BMI decreased with age for both sexes. Overall,measurement-based BMI is greater than report-basedBMI, which was underestimated as a result of overreport-ing height and underreporting weight. Nevertheless,the mean BMI difference between report-based andmeasurement-based BMI is very small at 0.68 kg/m2

(SD = 1.16) for men and 0.79 kg/m2 (SD = 1.49) forwomen. The mean difference is larger for women thanfor men at all ages and increases with age for both sexes.Pearson’s correlation coefficient shows a very high cor-relation between measurement-based BMI and report-based BMI (0.93 for men and 0.91 for women).

Table 1 Characteristics of sample subjects included in and excluded from the analyses

Attribute Analyticalsample

Sample persons excludedfrom the analyses

Statisticaltest

Total number of cases 1634 936Mean t-testSelf-reported height 155.31 154.24 n.s.†

Self-reported weight 54.61 52.79 **‡

Self-reported based BMI 22.58 22.33 n.s.Age 77.28 80.85 **Number of ADL difficulties§ 0.16 1.27 **Number of IADL difficulties¶ 0.22 1.66 **Number of words immediately recalled†† 4.72 3.82 **Number of words delayed recalled†† 3.93 2.97 **Distribution c2-testSex 100.00% 100.00% **

Men 48.50% 42.28%Women 51.50% 57.72%

Marital status 100.00% 100.00% **Married 65.32% 52.20%Not married 34.68% 47.80%

Level of education 100.00% 100.00% **Less than high school 47.17% 61.14%High school or more 52.83% 38.86%

Income 100.00% 100.00% **Less than ¥1 million 9.10% 15.04%¥1–1.99 million 18.71% 17.38%¥2–2.99 million 21.23% 14.31%More than ¥3 million 35.64% 20.40%Unknown 15.32% 32.87%

Chronic diseases‡‡ 100.00% 100.00% n.s.Have at least one chronic disease 73.53% 70.43%Does not have chronic diseases 26.47% 29.57%

†Not statistically significant. ‡P < 0.01. §Based on seven activities of daily living (ADL): bathing, dressing, eating, transferring,walking, going outside, and going to the toilet. ¶Based on seven instrumental activities of daily living (IADL): preparing ownmeal, shopping, managing money, making a phone call, doing light house work, using public transportation, and takingmedication. ††Based on 10 Japanese words: dog, knife, train, baseball, cat, pan, airplane, horse, swimming, and bicycle. ‡‡Basedon six diseases: heart diseases, hypertension, cerebrovascular diseases, diabetes, arthritis, and osteoporosis. n.s., not significant.

V Yong and Y Saito


Figures 1 and 2 show the decomposition of the dif-ferences between report-based BMI and measurement-based BMI for men and women, respectively. For men,of the 0.68 kg/m2 difference between self-reported andmeasured BMI (see Table 2), 61.2% is due to differ-ences in weight and 38.8% is due to differencesin height. For women, weight and height account for51.1% and 48.9%, respectively, of the differences inself-reported and measured BMI (0.79 kg/m2). Thus,the inaccuracy in BMI is mainly due to differences inweight. However, as the figures indicate, the relativeimpact of weight decreases as Japanese men and womenget older, and height becomes a more important factorin BMI self-reported and measured differences, espe-cially for women aged 80+. The exception to this trendis men aged 85+.

Table 3a and b show the cross-tabulation of report-based BMI weight categories with measurement-basedBMI weight categories by sex and age groups, respec-tively. We used the World Health Organization interna-tional classification for the weight categories. Thenumbers in bold highlight the level of agreementbetween report-based and measurement-based BMIweight categories. There is a very high level of agree-ment in the obese category (95.2% for men; 97.5%for women), the overweight category (87.1% for men;75.3% for women), and the normal weight category(87.5% for men; 85.7% for women). The level of agree-ment is lower in the underweight category (58.0% formen; 60.7% for women), with the majority of misclas-sified subjects actually in the normal weight category.The weighted kappa statistical test for the level of agree-ment beyond chance shows that the kappa value isreasonably good at 0.715 for men (P < 0.0001) and0.670 for women (P < 0.0001).

As age increases (Table 3b), the level of agreementfor the underweight category generally increases as wellfor both sexes. In particular, for women, underweightreporting tends to be more accurate at older ages thanyounger ages (63.3% at ages 85+ versus 58.7% for ages70–74). The level of agreement for the normal weightcategory declines continuously as age increases formen (91.5%, 86.1%, 86.0%, 80.1%), although itremains generally consistent at about 85% for women.For the overweight category, the level of agreementdeclines for both sexes but increases at ages 85+.Finally, reporting for the obese category is overall veryaccurate with most age groups achieving 100% accu-racy for both sexes, with the exception of men aged75–79 and women aged 80–84. The kappa valuesranged from 0.6070 to 0.7719. Based on the kappavalues, the level of agreement decreased with age forboth sexes and is higher for men than women, exceptfor women aged 85+.

Table 4 shows the prevalence estimates of the weightcategories based on self-reported values and measuredT

able

2M

easu

red

and

self-

repo

rted

heig

ht,w

eigh

t,an

dB

MI

byse

xan

dag

egr

oups

n(%

)M

easu

red

heig

ht,c

m(m

ean

SD)

Hei

ght

diff

eren

ce:

SR–

mea

sure

d(m

ean

SD)

Cor

rela

tion

rM

easu

red

wei

ght,

kg(m

ean

SD)

Wei

ght

diff

eren

ce:

SR–

Mea

sure

d(m

ean

SD)

Cor

rela

tion

rM

easu

red

BM

I,kg

/m2

(mea

nSD

)

BM

Idi

ffer

ence

:SR

–M

easu

red

(mea

nSD

)

Cor

rela

tion

r

Men

783

(48.

5)16

1.01

(6.3

2)0.

93(2

.48)

*0.

92*

60.3

9(9

.50)

-1.0

8(2

.55)

*0.

96*

23.2

7(3

.24)

-0.6

8(1

.16)

*0.

93*

Wom

en85

1(5

1.5)

147.

83(6

.18)

1.23

(2.8

4)*

0.89

*51

.05

(8.4

9)-0

.88

(2.4

5)*

0.96

*23

.35

(3.5

6)-0

.79

(1.4

9)*

0.91

*A

gegr

oup

Men 70

–74

280

(38.

0)16

2.62

(6.2

0)0.

66(2

.10)

*0.

94*

62.1

5(6

.20)

-1.1

0(2

.10)

*0.

98*

23.4

9(6

.20)

-0.6

0(2

.10)

*0.

96*

75–7

923

1(3

2.8)

160.

80(6

.60)

1.01

(2.6

2)*

0.92

*60

.59

(9.9

3)-1

.05

(2.5

4)*

0.97

*23

.40

(3.3

4)-0

.69

(1.2

2)*

0.93

*80

–84

138

(18.

2)15

9.21

(6.3

0)1.

31(2

.97)

*0.

89*

58.0

7(9

.31)

-0.9

1(3

.78)

*0.

92*

22.9

0(3

.37)

-0.7

3(1

.51)

*0.

90*

85+

134

(11.

0)15

9.02

(5.0

7)1.

00(2

.36)

*0.

89*

57.5

5(5

.07)

-1.4

3(2

.36)

*0.

95*

22.7

5(5

.07)

-0.8

3(2

.36)

*0.

92*

Wom

en70

–74

309

(38.

8)15

0.12

(5.6

1)0.

76(1

.94)

*0.

94*

52.9

8(5

.61)

-1.0

5(1

.94)

*0.

97*

23.5

3(5

.61)

-0.7

1(1

.94)

*0.

95*

75–7

921

5(2

8.2)

148.

22(6

.05)

1.07

(2.7

8)*

0.89

*51

.65

(8.1

0)-0

.89

(2.9

5)*

0.93

*23

.52

(3.5

5)-0

.76

(1.6

5)*

0.89

*80

–84

175

(21.

6)14

5.30

(6.3

9)1.

71(3

.82)

*0.

82*

48.6

5(8

.57)

-0.6

0(2

.62)

*0.

95*

23.0

3(3

.64)

-0.8

4(1

.93)

*0.

85*

85+

152

(11.

4)14

3.83

(4.6

1)2.

26(2

.90)

*0.

82*

47.5

2(4

.61)

-0.7

5(2

.90)

*0.

96*

22.9

2(4

.61)

-1.0

7(2

.90)

*0.

91*

*P<

0.00

01.P

erce

ntag

essh

own

for

sam

ple

size

sar

efo

rth

ew

eigh

ted

sam

ples

.Sig

nific

ant

diff

eren

cebe

twee

nSR

and

mea

sure

dva

lues

are

base

don

pair

edt-

test

.SR

,sel

f-re

port

ed.



values. Using measured values as the gold standard, therespective tests were conducted: sensitivity (true posi-tive, accurate identification of cases), specificity (truenegative, accurate identification of non-cases), positivepredictive value (false positive, Type-I error), and nega-tive predictive value (false negative, Type-II error). Forboth sexes, the prevalence of underweight and normalweight were overestimated while the prevalence of over-weight and obesity were underestimated. Based on self-reported values, the prevalence of overweight is 18.1%for both men and women, which are 0.74 and 0.79times that of the measured-based values of 24.3% formen and 23.0% for women. The prevalence of obesitybased on self-reported values and measured values is1.7% versus 2.4% for men and 1.4% versus 4.3% forwomen, respectively. With measured values as the ref-erence, specificity is very high (range: 94.2–100%), andsensitivity is lower (range: 59.3–65.1%).

Discussion

This study investigated the accuracy of self-reportedheight, weight, and BMI by comparing the self-reportedvalues with measured values for an elderly Japanesesample of 783 men and 851 women aged 70 and over.Although the characteristics of the analytical sample aresomewhat different from those who are excluded, the

test result of self-report-based BMI indicates that thereporting patterns those included and excluded arenot significantly different from each other. The report-based and measurement-based BMI were further clas-sified into weight categories and cross-tabulated tocheck for misclassifications of weight status. The resultsshowed a very high correlation between the self-reported and measured values (Pearson’s r: 0.92 and0.89 for men and women, respectively, for height;0.96 for both sexes for weight; 0.93 and 0.91 for menand women, respectively, for BMI). The respectivedifferences in mean values were very small: 0.93 cmand 1.23 cm for men and women (height); 1.08 kgand 0.88 kg for men and women (weight); and0.68 kg/m2 and 0.79 kg/m2 for men and women(BMI), respectively. For both sexes, height tends to beoverreported while weight tends to be underreported.Consequently, report-based BMI is underestimated.Decomposition of the differences between report-basedand measurement-based BMI showed that the differ-ences are due mainly to weight differences, althoughweight’s relative impact decreases with age and heightdifferences become a more important factor. Weightedkappa values (0.7149 and 0.6703 for men and women,respectively; P < 0.0001) attest to the high level ofagreement between report-based and measurement-based BMI weight categories, and high specificity

Figure 1 Decomposition of differencebetween measured and self-reportedBMI for men, by total sample and agegroups.

Figure 2 Decomposition of differencebetween measured and self-reportedBMI for women, by total sample andage groups.

V Yong and Y Saito


Table 3 Report-based versus measurement-based BMI by (a) sex and (b) age groups and sex

Measurement-based BMISelf-reportedbased BMI

Underweight Normal Overweight Obese Weighted(BMI < 18.5) (BMI 18.5–24.9) (BMI 25.0–29.9) (BMI 3 30+) Kappan % n % n % n %

(a)TotalMen

Underweight 40 58.0 27 41.3 1 0.7 0 0.0Normal 5 0.8 485 87.5 70 11.8 0 0.0 0.7149*Overweight 0 0.0 11 8.1 124 87.1 6 4.7Obese 0 0.0 1 4.8 0 0.0 13 95.2

WomenUnderweight 51 60.7 29 38.2 1 1.1 0 0.0Normal 8 1.1 516 85.7 78 13.0 1 0.2 0.6703*Overweight 0 0.0 15 9.4 118 75.3 20 15.3Obese 0 0.0 0 0.0 1 2.5 13 97.5

(b)Men

Ages 70–74Underweight 7 45.0 9 55.0 0 0.0 0 0.0Normal 2 1.3 176 91.5 17 7.2 0 0.0 0.7719*Overweight 0 0.0 4 7.1 58 91.4 1 1.6Obese 0 0.0 0 0.0 0 0.0 6 100.0



Ages 85+Underweight 9 53.8 9 46.2 0 0.0 0 0.0Normal 3 2.4 79 80.1 16 17.5 0 0.0 0.6132*Overweight 0 0.0 1 5.5 14 83.3 2 11.3Obese 0 0.0 0 0.0 0 0.0 1 100.0

WomenAges 70–74

Underweight 17 58.7 11 41.3 0 0.0 0 0.0Normal 3 1.1 179 84.7 26 14.2 0 0.0 0.7113*Overweight 0 0.0 3 5.2 57 84.8 6 10.1Obese 0 0.0 0 0.0 0 0.0 7 100.0



Ages 85+Underweight 12 63.3 7 36.7 0 0.0 0 0.0Normal 0 0.0 93 84.6 19 15.4 0 0.0 0.6639*Overweight 0 0.0 3 15.7 15 79.8 1 4.5Obese 0 0.0 0 0.0 0 0.0 2 100.0

*P < 0.0001. Numbers shown in bold indicate percentage agreement between self-reported and measured BMI. Percentages are weighted.



(range: 94.4–100%) and sensitivity (range: 59.3–65.1%)showed a mostly accurate identification of cases andnon-cases.

The results of our study are generally consistent withprevious studies conducted in Western countries, someof which had the same age subgroups as our studypopulation.1–9 In previous studies, self-reported height,weight, and BMI were similarly strongly correlated withmeasured height, weight, and BMI, and the correlationswere usually more than 0.90.

Height overreporting in other studies ranged from3.06–4.29 cm for men and 2.92–4.50 cm for women,both of which are larger than for the current study.Height overreporting has been attributed to factors such

as social desirability; evidence shows that shorter peopleand men are more likely to overreport their height. Inour study, the overreporting of height increased withage, but was larger for women than men. This could bebecause our study sample consisted of the elderly, forwhom height shrinkage is more pronounced with age,particularly among women who are likely to lose bonedensity at a faster rate than men. Height overreportingcould be associated with unawareness of adult heightshrinkage.19,20

In addition, it is also possible that height overreport-ing could occur among subjects who have back prob-lems (i.e. bent backs) and have difficulties stretchingtheir backs straight, resulting in height measurement

Table 4 Prevalence of measured and self-reported values and test values ofthe diagnosis

Men(n = 783) %

Women(n = 851) %

Prevalence of underweightBased on self-reported values 8.9 9.4Based on measured values 5.7 6.5Test values

Sensitivity (%) 90.6 88.0Specificity (%) 96.0 96.1

Positive predictive value (%) 58.0 60.7Negative predictive value (%) 99.4 99.1

Prevalence of normal weightBased on self-reported values 71.4 71.1Based on measured values 67.7 66.2Test values



Prevalence of overweight (%)Based on self-reported values 18.1 18.1Based on measured values 24.3 23.0Test values



Prevalence of obesity (%)Based on self-reported values 1.7 1.4Based on measured values 2.4 4.3Test values



Definitions: Underweight, BMI < 18.5; Normal, BMI 18.5–24.9; Overweight, BMI25.0–29.9; Obesity, BMI 3 30+.

V Yong and Y Saito


error. A separate analysis of data excluding subjects whohave bent back problem (17% of the analysis sample)showed that the differences between measured and self-reported height and BMI, based on both measured andself-reported data, are smaller than this study’s resultsindicate. Thus, if we took back problems into account,the accuracy of self-reported height and BMI based onself-reports would be higher.

Likewise, weight underreporting was also docu-mented, with women and heavier people more likely tounderreport their weight.1–7 Weight underreporting inprevious studies ranged on average from 0.54–0.76 kgfor men and 0.85–2.50 kg for women. This study hadsimilar findings for women, with underreporting fallingat the lower end of the range, but the discrepancies werelarger for men. In our study, the underreporting ofweight was larger for men than women, and the accu-racy of self-reported weight increased with age forwomen. A possible reason for men underreporting theirweight could be that women weigh themselves moreregularly or frequently. Additionally, the reluctance ofindividuals to reveal their weight may decrease at mucholder ages for women, with the pressure to conform tosocietal norms of body size and image likely to be stron-ger among young Japanese women.11,12 The overallweight underreporting in our study could also be partlya result of the clothes that the subjects were wearing atthe time that they were weighed. According to the data,more than one-third of the subjects (35.6%) werewearing thick sweaters when they were weighed.

Concomitant with the higher correlations for weightthan for height, the accuracy of self-reported weight isgreater than for self-reported height, which could bedue to the fact that it is easier for people to weighthemselves than to measure their own height. Theoverall height overreporting and weight underreportingcould also be based on memories of height and weightmeasurements taken when the subjects were younger –when they were taller and weighed less, as other studieshave indicated.6,21

As a result of height overreporting and weight under-reporting, BMI derived from self-reported height andweight are generally underestimated compared withmeasurement-based BMI. There is some evidence that,depending on whether report-based BMI ormeasurement-based BMI is used, the associationbetween obesity and morbidity may differ to someextent, with the use of report-based BMI typicallyresulting in a stronger association with diseases than ifmeasurement-based BMI were used.22–24 A recent studyalso found that the use of self-reported height andweight biases the BMI-mortality association.25

In addition, we also compared the results of our studywith the two existing Japanese studies of young andmiddle-aged adults. The discrepancies between self-reported and measured values among younger Japanese

were even smaller than in our current study on elderlyJapanese. For Japanese aged 35–64 years in a workplace,men and women, on average, respectively overreportedtheir height by only 0.078 cm and 0.029 cm, and under-reported their weight by 0.034 kg and 0.024 kg.11 Thesecond study among Japanese female factory workersaged 20–42 years found that the mean reported heightand weight were 0.1 cm shorter and 0.2 kg lighter,respectively, compared to the measured values.12 Theextremely small differences in both studies were attrib-uted, in part, to annual health check-ups for employeesin Japan. Apart from this factor, studies that examinedthe effect of age on the validity of self-reported heightand weight have provided evidence that the accuracy ofself-reports decreases at older ages.8,9

To conclude, while a number of studies have exam-ined the validity of self-reported height and weight,there are only a few on the elderly and in an Asiancontext. Our current study attempts to make a contri-bution in this regard. This study has established thatthe self-reports of height, weight, and BMI of elderlyJapanese men and women aged 70 and over are mostlyaccurate, suggesting that they can be used in epide-miological surveys. However, caution must be taken,particularly for the oldest age group (age 85+), wherethe accuracy of self-reports declines. In addition, report-based BMI, and thus the prevalence of overweightand obesity based on self-reported values, tend tobe underestimated on the whole and should be usedprudently.

Acknowledgements

We greatly appreciate the advice from Dr Shiro Horiu-chi of Hunter College (New York, NY, USA) on decom-position method. This research was supported by anAcademic Frontier Grant from the Japanese Ministry ofEducation, Culture, Sports, Science and Technology,2006–2010.

Disclosure statement

The authors declare no conflict of interest.

References

1 Niedhammer I, Bugel I, Bonenfant S, Goldberg M, LeclercA. Validity of self-reported weight and height in the FrenchGAZEL cohort. Int J Obes Relat Metab Disord 2000; 24:1111–1118.

2 Nawaz H, Chan W, Abdulrahman M, Larson D, Katz D.Self-reported weight and height: implications for obesityresearch. Am J Prev Med 2001; 20: 294–298.

3 Brener N, McManus T, Galuska D, Lowry R, Wechsler H.Reliability and validity of self-reported height and weightamong high school students. J Adolesc Health 2003; 32:281–287.



4 Spencer E, Appleby P, Davey G, Key T. Validity of self-reported height and weight in 4808 EPIC-Oxford partici-pants. Public Health Nutr 2002; 5: 561–565.

5 Lawlor D, Bedford C, Taylor M, Ebrahim S. Agreementbetween measured and self-reported weight in olderwomen. Results from the British Women’s Heart andHealth Study. Age Ageing 2002; 31: 169.

6 Gunnell D, Berney L, Holland P et al. How accurately areheight, weight and leg length reported by the elderly, andhow closely are they related to measurements recorded inchildhood? Int J Epidemiol 2000; 29: 456.

7 Villanueva E. The validity of self-reported weight in USadults: a population based cross-sectional study. BMCPublic Health 2001; 1: 11.

8 Kuczmarski MF, Kuczmarski RJ, Najjar M. Effects of ageon validity of self-reported height, weight, and body massindex: findings from the Third National Health and Nutri-tion Examination Survey, 1988–1994. J Am Diet Assoc 2001;101: 28–34.

9 Fillenbaum GG, Kuchibhatla MN, Whitson HE et al.Accuracy of self-reported height and weight in acommunity-based sample of older African Americans andWhites. J Gerontol A Biol Sci Med Sci 2010; 65A: 1123–1129.

10 Gorber S, Tremblay M, Moher D, Gorber B. A comparisonof direct vs. self-report measures for assessing height,weight and body mass index: a systematic review. Obes Rev2007; 8: 307–326.

11 Wada K, Tamakoshi K, Tsunekawa T et al. Validity ofself-reported height and weight in a Japanese workplacepopulation. Int J Obes Relat Metab Disord 2005; 29: 1093–1099.

12 Nakamura K, Hoshino Y, Kodama K, Yamamoto M. Reli-ability of self-reported body height and weight of adultJapanese women. J Biosoc Sci 1999; 31: 555–558.

13 Japan Statistics Bureau. Population Estimates. Tokyo: Min-istry of Internal Affairs and Communications, 2010.

14 McCurry J. Japan battles with obesity. Lancet 2007; 369:451–452.

15 Saito Y. A Survey on Health and Lifestyle. Report Series No. 1(in Japanese). Tokyo: The Nihon University Center forInformation Networking, 2000.

16 NUJLSOA. Nihon University Japanese Longitudinal Studyof Aging. 2004 [Cited 19 Sept 2011.] Available from URL:http://www.usc.edu/dept/gero/CBPH/nujlsoa/.

17 Kitagawa E. Components of a difference between two rates.J Am Stat Assoc 1955; 50: 1168–1194.

18 Kanazawa M, Yoshiike N, Osaka T, Numba Y, Zimmet P,Inoue S. Criteria and classification of obesity in Japanand Asia-Oceania. Asia Pac J Clin Nutr 2002; 11: 732–737.

19 Cline M, Meredith K, Boyer J, Burrows B. Decline of heightwith age in adults in a general population sample: estimat-ing maximum height and distinguishing birth cohorteffects from actual loss of stature with aging. Hum Biol1989; 61: 415.

20 Sorkin J, Muller D, Andres R. Longitudinal change inthe heights of men and women: consequential eftectson body mass index. Epidemiol Rev 1999; 21: 247–260.

21 Stevens J, Keil JE, Waid LR, Gazes PC. Accuracy of current,4-year, and 28-year self-reported body weight in an elderlypopulation. Am J Epidemiol 1990; 132: 1156–1163.

22 Chiolero A, Peytremann-Bridevaux I, Paccaud F. Associa-tions between obesity and health conditions may be over-estimated if self-reported body mass index is used. Obes Rev2007; 8: 373–374.

23 Santillan A, Camargo C. Body mass index and asthmaamong Mexican adults: the effect of using self-reported vsmeasured weight and height. Int J Obes 2003; 27: 1430–1433.

24 Shields M, Gorber S, Tremblay M. Estimates of obesitybased on self-report versus direct measures. Health Rep2008; 19: 1–6.

25 Keith SW, Fontaine KR, Pajewski NM, Mehta T, AllisonDB. Use of self-reported height and weight biases the bodymass index-mortality association. Int J Obes 2010; 35: 401–408.

V Yong and Y Saito


How accurate are self-reported height, weight, and BMI among community-dwelling elderly Japanese?:...

Documents

Transcript of How accurate are self-reported height, weight, and BMI among community-dwelling elderly Japanese?:...