Body weight is one of the strongest predictors of bone mass in subjects of all age groups,1 and body mass index (BMI) is a well-known predictor of fragility fracture.2, 3 Lean, fat, and bone mass are the three components of body weight. Several epidemiologic studies have reported that both fat body mass and lean body mass may have different relationships with bone density at different stages of life.4–10 Cross-sectional and a few longitudinal studies have shown age-dependent body composition changes with a decline of lean body mass and bone mass and a gain of fat body mass in both sexes.11–13 Beside the effect of age, other determinants of the changes of body composition have been evaluated in few longitudinal studies. Thus, the loss of lean mass was greater with greater baseline lean mass or lesser fat mass,14 partially explaining that men are at greater risk of loss of lean mass than are women.15 The decrease of lean mass and muscle mass leading to sarcopenia with age is associated with the age-related functional decline.16–21 Fat mass has also been shown to have an adverse effect on physical function.22 In those studies, physical function has mostly been assessed during or consequently to the changes of body composition but has rarely been prospectively evaluated as a potential determinant of those changes.23 The decrease of estrogens associated with menopause contributes to the loss of bone mass, to the redistribution of subcutaneous fat to the visceral area, and to an accelerated loss of muscle mass, partly explaining a loss of strength occurring at an earlier age than in men.24–26 Other biochemical determinants of the changes of BC in women have been explored in some cross-sectional studies showing a positive association between skeletal muscle mass with levels of serum-free testosterone27 and 25-hydroxy vitamin D28 and an inverse relationship between fat mass with the levels of bone turnover markers and 25-hydroxy vitamin D.29
However, only a few studies have evaluated prospectively age-related body composition changes and their relationships with the menopausal status and the bone status in a wide age range, ie, both in pre- and postmenopausal women. Moreover, to our knowledge, there are no reports that assessed both biochemical patterns and body composition changes over a long period of time.
Therefore, the aim of our study was to prospectively evaluate, with a 6-year follow-up, age-related body composition changes in premenopausal and postmenopausal French women from the Os des Femmes de Lyon (OFELY) cohort and to explore several clinical, biochemical, and hormonal determinants of those changes.
Materials and Methods
We have studied 757 women from the OFELY cohort who had a first evaluation of body composition on a Hologic QDR 4500 (Hologic, Inc., Waltham, MA, USA) device during the ninth annual follow-up (called baseline for the current analysis). Briefly, OFELY is an ongoing prospective study of the determinants of bone loss in 1039 volunteer women, recruited between February 1992 and December 1993, 31 to 89 years of age, randomly selected from the affiliates of a large health insurance company (Mutuelle Générale de l'Education Nationale) from the Rhône district (ie, Lyon and its surroundings in France), with an annual follow-up.30, 31 From these 757 women, 12 were excluded for the current analysis for the following reasons: alcohol >6 units/day (n = 1), paraplegia (n = 1), current oral corticosteroids >3 months (n = 3), breast cancer treated with tamoxifen (n = 6) or exemestane (n = 1). Among the 745 women analyzed, 719 (96%) had one or more further evaluation of body composition with a mean (SD) follow-up of 72 (19) months. All the exams were performed between February 2000 and December 2008.
Women were categorized into premenopausal (PreM) and postmenopausal (PostM) at baseline. They were considered postmenopausal if they had not been menstruating for at least 1 year. For the longitudinal analysis, PreM women who became PostM women at the last evaluation of body composition were named perimenopausal (PeriM) women. All women completed a written health questionnaire at each visit. The questionnaire included medical history, menopausal status, medication use such as hormone replacement therapy (HRT) for menopause, calcium consumption, alcohol consumption, tobacco use, physical activity, and radiologically confirmed occurrence of low-trauma fractures. Physical activity was expressed by a score calculated from sport or recreation and job and home activities described previously.32 Height and weight were measured with participants wearing indoor clothes and no shoes. Body mass index was calculated as body weight/height2.
Body composition assessment
Body composition was measured annually between the baseline and seventh year of follow-up (except at year 6, which was skipped) with dual-energy X-ray absorptiometry (DXA) using fan beam DXA (Hologic QDR4500A, software version 8.26). Total hip bone mineral density (BMD) was measured at each visit on the same device. Between the recruitment of the OFELY cohort and the ninth annual follow-up (named baseline for the current study), body composition was assessed with another device (QDR 2000), but because the correlation was <0.90 for lean mass between both devices, longitudinal changes were analyzed only on the more recent device.
Fat body mass (FM), total body fat-free mass (FFM), which includes both lean and bone mineral mass, lean body mass (LM), and total body bone mineral content (TBBMC) were measured.
Appendicular skeletal muscle mass (ASM), calculated as the sum of lean mass (nonfat, nonbone) in the arms and legs, was used as an indicator of skeletal muscle mass. The relative skeletal muscle mass index (RASM) was calculated as ASM/height25.2, 16, 17 Height at the age of 25 was used to avoid the effect of height loss with age. It has been shown that the QDR4500A scanners overestimate FFM and underestimate FM compared with the four-compartmental model criterion method. Thus, the recommended corrections were made by multiplying FFM by 0.946 and leaving the total mass unchanged. From those two variables, FM and LM were recalculated. The adjustments were made to individual data for the total body and all the subregions.33–35 Bone status was defined according to the WHO classification from the total hip BMD measured at baseline: normal (T-score > − 1), osteopenia (−2.5 < T-score ≤ −1), and osteoporosis (T score ≤ −2.5).36
For each woman, fasting blood samples were collected before 9:00 a.m. on the day of clinical assessment and kept frozen at −80°C until assayed. Serum osteocalcin (OC), intact N-terminal propeptide of type I collagen (PINP), and β isomerized C-terminal cross-linking of type I collagen (CTX-I) were measured using an automatic test (Elecsys N-MID Osteocalcin, Elecsys P1NP, Elecsys β–Crosslaps; Roche Diagnostics, Mannheim, Germany). Intra- and interassay variations were <8% for all markers. Serum total 17 β-estradiol (E2) was measured by direct radioimmunoassay (CIS Bio International, Gif-sur-Yvette, France). This assay, which shows a limit of detection of 11 pM/L, was validated by Agence Française de Sécurité Sanitaire des Produits de Santé (AFSSAPS) for the determination of very low E2 concentrations, especially in postmenopausal women. Intra- and interassay variations were <10%. Bioavailable E2 was calculated by multiplying total E2 concentration by the percentage of non-sex-hormone-binding-globulin (SHBG)–bound E2. Non-SHBG–bound E2 was determined by differential precipitation of serum proteins after equilibration of the serum with tritiated tracer amounts of E2 (total radioactivity).37 At the end of the reaction, SHBG and E2 bound to it are precipitated using 50% ammonium sulfate. After centrifugation, radioactivity measured in the supernatant represents bioavailable E2 fraction of the total radioactivity.
Serum total testosterone was measured by radioimmunoassay after extraction with a limit of detection of 0.04 nM/L. Intra- and interassay variations were <10%. Free testosterone was calculated using the method described by Södergard.38 Serum intact parathyroid hormone (PTH) was measured by an immunoradiometric assay using two monoclonal antibodies (ELSA-PTH, CIS Bio) with a limit of detection of 0.7 pg/mL. Intra- and interassay variations were <8%. Serum 25-hydrocholecalciferol (25[OH]D) was measured by an automatic competitive two-step chemiluminescence assay that detects both D2 and D3 metabolites (LIAISON 25-OH Vitamin D TOTAL; DiaSorin, Saluggia, Italy). The limit of detection was 4 ng/mL.
Chi-square tests and unpaired nonparametric tests were used to compare baseline characteristics between premenopausal and postmenopausal women. For each woman, changes in body composition were calculated from absolute and relative changes between two successive visits and expressed as percentage per year. To assess the determinants of those changes, we performed mixed model analysis on repeated measures, with dependent variables (body composition parameters), fixed factors (age, menopausal status, follow-up time, bone status, physical activity score, bone markers, hormone levels), and random factors (women). Analyses were performed both with and without interaction between age and each other characteristics effect. An alpha level of 0.05 was regarded as statistically significant. All statistical analyses were performed using Statistical Analysis Software (SAS version 9.1; SAS Institute, Cary, NC, USA).
Baseline cross-sectional analysis
The characteristics of the 745 women at baseline according to their menopausal status are shown in Table 1. As expected, untreated PostM women had greater BMI and FM and lower LM, ASM, RASM, TBBMC, and total hip BMD compared with PreM women. PostM women taking HRT at baseline were younger, had lower BMI and FM, and greater height, FFM, TBBMC, and total hip BMD compared with untreated PostM women. After adjustment for age, PostM women on HRT at baseline still had lower BMI (p < 0.01) and FM (p < 0.002) and greater TBBMC and total hip BMD (p < 0.0001) compared with untreated PostM women. The prevalence of low muscle mass at baseline in untreated PostM—defined with a value of RASM within the lowest decile of values obtained in premenopausal women younger than 50 years at baseline (<4.74 g/m2)—was about twice that of PreM and PostM women on HRT (p = 0.02).
Table 1. Characteristics at Baseline for All Women and According to Their Menopausal Status (Values Are Mean ± SD)
All women (n = 745)
PreM women (n = 145)
p between the 3 subgroups (Kruskal–Wallis)
Untreated (n = 412)
HRT (n = 188)
PreM = premenopausal; PostM = postmenopausal; HRT = hormone replacement therapy; BMI = body mass index; ASM = appendicular skeletal muscle mass; RASM = relative skeletal muscle mass index; BMC = bone mineral content; BMD = bone mineral density.
In the whole group, body composition correlated with age: negatively for FFM (r = −0.21), LM (r = −0.19), ASM (r = −0.23), RASM (r = −0.22), and TBBMC (r = −0.51) (p < 0.0001 for all); positively for BMI and FM (r = 0.18 and 0.16, respectively, p < 0.0001). After controlling for age, both LM and FM were positively correlated with weight (r = 0.83 and 0.91, respectively), BMI (r = 0.60 and 0.90), RASM (r = 0.78 and 0.56), TBBMC (r = 0.53 and 0.20), and total hip BMD (r = 0.39 and 0.32) (p < 0.0001 for all).
Longitudinal analysis of body composition
Twenty-six women who withdrew after baseline and 7 PreM women who began HRT after baseline were excluded for the longitudinal analysis. Moreover, most PostM women on HRT at baseline have stopped their treatment before the end of the follow-up and were also excluded for this analysis. No other exclusion criterion was applied for the longitudinal analysis. Thus, 525 women without HRT were followed during 6.0 ± 1.6 years with two to seven annual evaluations of their body composition. The mean follow-up was 5.5 ± 1.3 years for PreM women at baseline (n = 132) and 6.1 ± 1.6 years for untreated PostM women (n = 393).
As shown in Fig. 1, the evolution of body composition was highly variable at any age. Age was a significant determinant of the changes in FM, LM, RASM, and TBBMC but not of changes in weight. When women were classified into three classes (≤55 years, 55 to 70 years, >70 years), FM significantly increased until 70 years, LM and RASM significantly decreased after 55 years, whereas weight significantly decreased after 70 years. TBBMC decreased at all ages with the greatest loss in women ≤55 years: mean (SEM) −0.66 (0.1)%/year versus −0.27 (0.1)%/year in women 55 to 70 years (p = 0.04) and −0.30 (0.1)%/year in women >70 years (p = 0.05) (Fig. 2). Similar results were obtained for absolute changes of body composition. In the whole group and after controlling for age, changes in weight, FM, LM, and RASM were moderately correlated between them: r = 0.44, p < 0.0001, between changes in weight and FM; r = 0.50, p < 0.0001 between changes in weight and LM; r = 0.26, p < 0.0001 between changes in weight and RASM; r = 0.15, p < 0.0001 between changes in FM and LM; r = 0.10, p < 0.01 between changes in FM and RASM; r = 0.53, p < 0.0001 between changes in LM and RASM. Changes of BC were slightly correlated with changes of total hip BMD after controlling for age, better for TBBMC (r = 0.21, p < 0.0001) than for FM (r = 0.13, p < 0.0001), LM (r = 0.14, p < 0.0001) and RASM (r = 0.12, p < 0.0001).
We performed a subgroup analysis on weight-stable women (change of weight between −0.5 and +0.5%/year, n = 164,) compared with women losing (loss of weight >0.5%, n = 182) or acquiring (gain of weight >0.5%, n = 179) body weight during the 6-year observation period. In weight-stable women, we observed a significant loss of RASM (−0.6%/year, p < 0.001) and a significant gain of FM (+0.9%/year, p < 0.001). In women losing weight, the loss of RASM was greater (−1.04%/year, p < 0.0001) and was associated with a loss of FM (−1.15%/year, p < 0.0001). In women acquiring weight, there was a gain of both RASM (+0.9%/year, p < 0.0001) and FM (+3.8%/year, p < 0.0001).
In PreM and PeriM women (ie, women preM at baseline who became PostM during a mean follow-up of 5.8 ± 1.3 years, mean age 48 ± 4 years, n = 44,), weight, LM, and RASM did not change, whereas a significant increase of FM (p < 0.01) and a decrease of TBBMC (p < 0.001 and p < 0.0001, respectively) were observed. In PostM women, weight did not significantly change and FM increased (p < 0.01), whereas LM, RASM, and TBBMC decreased (p < 0.0001). During the same period of follow-up, total hip BMD significantly decreased in PeriM and in PostM (p < 0.0001) (Fig. 3). After controlling for age, menopausal status was no longer a significant determinant of the changes in FM, LM, and RASM but still was for changes in TBBMC.
Among PostM women, 66 women had total hip BMD in the osteoporotic range, 201 women had osteopenia, and 106 women had normal BMD. At baseline, weight, FM, LM, RASM, and TBBMC were higher in women with normal BMD compared with women with osteopenia or osteoporosis (p < 0.0001). During the follow-up, a significant gain of FM was observed only in women with normal BMD (+0.69%/year, p < 0.01 versus 0) but not in those with osteopenia or osteoporosis. Nevertheless, after adjusting for age, changes in weight, FM, LM, RASM, and TBBMC were not significantly different between the three groups.
The other potential clinical (BMI, physical activity), biochemical (OC, P1NP, sCTX), and hormonal (25[OH]D, PTH, bioavailable estradiol, testosterone) determinants for changes in body composition were analyzed after controlling for age and menopausal status. Partial correlations with body composition at baseline are detailed in Table 2. Despite a strong association between BMI and body composition at baseline (r = 0.81, 0.61, and 0.73 for FM, LM, and RASM, respectively, p < 0.0001 for all), baseline BMI was not associated with changes in body composition except for TBBMC with a greater decrease of TBBMC in women with BMI below the median (−0.66%/year) compared with women with BMI above the median (−0.38%/year) (p = 0.01). At baseline, the level of physical activity was associated with FM and weight that were 13% and 5% higher, respectively, in women having a score below the median compared with women having a higher score (p < 0.001), whereas no association was found with LM, RASM, or TBBMC. The level of physical activity at baseline was not significantly associated with any changes of body composition.
Table 2. Partial Correlations (r) Between Body Composition at Baseline With Physical Activity, BMI, BTM, and Hormones After Controlling for Age and Menopausal Status
BMI = body mass index; BTM = bone turnover marker; FM = fat body mass; LM = lean body mass; RASM = relative skeletal muscle mass index; TBBMC = total body bone mineral content.
Among bone turnover markers (BTM) assessed at baseline, no association was found between both OC and sCTX with any change of body composition, whereas an inverse association was found with baseline FM (r = −0.10, p < 0.01 for both). Conversely, baseline levels of P1NP in the highest quartile were associated with a greater decrease of LM (−0.35%/year) and RASM (−0.61%/year) compared with lower levels (−0.17% and −0.32%, p = 0.07 and 0.02, respectively), whereas no association was found between levels of P1NP with changes of weight, FM, or TBBMC. Among hormonal levels, lower levels of bioavailable estradiol in the first quartile were associated with a greater decrease of TBBMC (−0.62%/year) compared with higher levels (−0.14%/year, p = 0.02) but not with changes of the other variables of body composition. In contrast, a significant positive association between estradiol and baseline FM, LM, and RASM (r = 0.17 to 0.30, p < 0.0001) was observed. Despite a negative association between levels of 25(OH)D with baseline FM, LM, and RASM and a slight positive association between levels of PTH with baseline FM and LM, low levels (lowest quartile) of 25(OH)D and high levels (highest quartile) of PTH were not associated with any change of body composition. Low levels of total testosterone (lowest quartile) were not associated with any change of body composition. In contrast, low levels of free testosterone were associated with the absence of significant change in FM (0.04%, p = 0.9) and a greater decrease of TBBMC (−0.76%/year, p < 0.0001) compared with higher levels (three highest quartiles), which were associated with a gain of FM (+0.56%/year, p = 0.01) and a lower decrease of TBBMC (−0.55%/year, p = 0.01). No association between the levels of free testosterone and change of RASM was found, despite a positive association at baseline.
In our prospective evaluation of body composition changes among French women, we observed an increase in fat mass with age, beginning before the menopause, whereas the loss in lean mass and RASM began after the beginning of bone loss. The loss of RASM was masked in weight-stable women. Age was the main determinant of those changes, whereas menopausal status, bone status, and physical activity did not affect these changes significantly after accounting for age. Among potential biochemical determinants, we found significant slight associations only for P1NP, estradiol, and free testosterone.
At baseline, even if body weight was not significantly different according to menopausal status, FM was higher and LM, RASM, and TBBMC were lower in untreated PostM women compared with PreM women, in agreement with previous cross-sectional studies.7, 39 Among PostM women, those treated with HRT had lower FM compared with untreated PostM women even after controlling for age, which is consistent with previous studies indicating that hormone replacement therapy prevents—at least partially—the postmenopausal increase of weight40 and body FM.41, 42
During a 6-year follow-up, body weight did not significantly increase, and it decreased in women older than 70 years. FM increased in both PreM and PostM women until age 70 years, whereas a significant decrease over time of LM and RASM began after 55 years, ie, after the highest loss of TBBMC observed in perimenopausal women. The last findings are in agreement with cross-sectional studies showing that bone loss begins before lean mass loss.43 In the elderly, a cross-sectional analysis of 7518 healthy French women 75 years or older from the EPIDOS study showed a continued loss of weight, total fat mass, and total bone mineral mass even over 85 years of age, whereas the decrease in RASM mass was observed until 85 years and was reduced thereafter.44 Only a few longitudinal studies of body composition have been conducted, especially in the elderly15, 19, 45 or in small samples of women,12, 46 showing—as in our study—a loss of lean mass and a gain of fat over time in healthy women with an attenuation of the increase in fat in the oldest women.23 As shown in our study and consistently in literature, the loss of muscle mass is masked by weight stability, which is owing in part to an increased fat mass.46 In addition, the amount of muscle mass loss is often greater with weight loss than the gain of muscle mass with weight gain, leading to sarcopenia in older adults.19
Our study showed that age was the major determinant of the changes of body composition, which was not significantly affected by menopausal status after accounting for age, except for the changes in bone mass. In a cross-sectional study,43 64 postmenopausal women aged 50 to 53 years and 59 age-matched premenopausal women did not differ in weight, fat body mass, and lean body mass, whereas BMD was significantly lower in postmenopausal women, suggesting that lean and fat tissues are less sensitive to hypogonadism than bone tissue is. Our findings that lower levels of estradiol compared with higher levels were associated with a greater decrease of TBBMC but not with changes of the other variables of body composition are consistent with that result.
FM, LM, and RASM were higher in PostM women with normal BMD compared with those with low BMD. FM is a well-known determinant of both bone mineral density and the rate of change of bone density in postmenopausal women, with a protective effect of obesity on bone loss.1, 7, 47 Moreover, changes in fat mass are predictive of the change in BMD.10, 48 Lean mass is another determinant of BMD, but changes in BMD are less related to changes in lean mass than fat mass in postmenopausal women.6, 48–50 In a cross-sectional study, Walsh and colleagues showed that the prevalence of sarcopenia—defined as a relative skeletal mass index <5.45 kg/m2—was twice as high in women with osteoporosis as in those with osteopenia.51 To our knowledge, no prospective study analyzed changes of fat or lean mass regarding the bone status. In our study, the bone status was not associated with the changes of body composition after taking age into account, suggesting that the differences of body composition are inherent to bone status and precede bone loss.
As expected, we found a strong association between BMI and body composition at baseline. In contrast, BMI was not associated with changes of body composition except for TBBMC with a greater decrease in women with low BMI compared with women with BMI above the median. Prospective studies have shown a protective effect of physical activity in preventing weight gain in middle-aged persons.52, 53 That is in agreement with our findings showing that both fat mass and weight were lower in women having a higher level of physical activity. Nevertheless, in our study, the level of physical activity at baseline was not associated with changes of body composition. Although resistance-type exercise is effective in augmenting muscle mass in short-term intervention studies, there is less evidence that daily physical activity can attenuate age-related declines in muscle mass. The few prospective studies on the effect of physical activity on the further changes of body composition are controversial. A 9-year longitudinal study of 78 weight-stable women aged 60 ± 8 years showed that baseline age and level of physical activity were inversely and independently associated with changes in fat mass but not with changes in fat-free mass.23 In another longitudinal study conducted in 66 women older than 65 years, physical activity at baseline had no effect on changes over 3 years of fat mass, fat-free mass, or appendicular muscle mass.54 Finally, the results from a recent prospective population-based study of community-dwelling adults aged 50 to 79 years indicate that habitual levels of pedometer-determined ambulatory activity are weakly associated with lower declines over 2.6 years in leg lean mass and not associated with changes in fat mass in women.55 As the authors suggested, the lack of association with changes of fat mass contrasting with the negative association in cross-sectional analysis could be partly owing to a constant association between fat mass and physical activity over time. This constant association with time should be even more applicable for BMI. Other explanations for the discrepancies between studies are the variability of the assessment of physical activity and the lack of validated questionnaire. Moreover, the beneficial effect of physical activity on muscle function (strength) that was not evaluated in our study may occur independently of measurable changes in muscle mass.
Our findings are in agreement with cross-sectional studies conducted in postmenopausal women and showing that bone markers of formation and resorption are inversely related to fat mass.29 Moreover, there is biochemical evidence that weight change can impact bone turnover. Thus, fasting for 3 days significantly reduces indices of osteoblast activity.56 Conversely, our prospective study is the first having analyzed BTM as a potential determinant for changes of body composition. We found that high levels of P1NP at baseline were associated with a greater decrease of LM and RASM over 6 years, whereas no association was found between levels of BTM with changes of weight, FM, or TBBMC. Those findings suggest an association between the increase in bone formation and the loss of muscle mass that could be partly explained by the relationship between bone and muscle.7 Alternatively, this association may not be causal, and greater levels of P1NP may be associated with the causal factor. That should be further evaluated because no previous study reported this.
With hormones, association with changes of the body composition was found only between estradiol with TBBMC and between free testosterone with both TBBMC and FM but not with RASM changes. For most of the clinical, biochemical, and hormonal potential determinants that were analyzed, their association with body composition at baseline was greater than that observed with their changes. That is probably the result of both the high interindividual variability of those changes and, as suggested before, a constant association with time. A better approach should be to prospectively analyze those determinants in women with incident sarcopenia that need larger studies.
Our study has strengths and limitations. The strengths are a significant duration of follow-up in well-characterized healthy women and a rigorous statistical analysis on repeated measures. One limitation is that women who were followed until the ninth follow-up for the current study were younger at the inclusion in the OFELY study compared with those who did not (54 years ± 11 versus 58 years ± 11, p < 0.0001). Nevertheless, after adjusting for age, there was no significant difference for weight, BMI, and total hip BMD. The longitudinal analysis of body composition was performed in relatively healthy, ambulatory women who could come at later visits, allowing a possible underestimation of the loss of lean mass and RASM compared with women with poorer health status who would have stopped their follow-up or not participated to this cohort study. Moreover, the DXA measurement of muscle mass may not be sufficiently sensitive to detect age-related composition changes, such as selective atrophy of type II fibers, increased interstitial fat and fibrous tissues, or functional changes such as muscle denervation that could affect strength independent of mass.57, 58 Finally, other potential determinants were not assessed in our study such as protein intake, which has been shown to be associated with a lower loss of muscle mass.59
In conclusion, in this 6-year prospective follow-up, we observed a high interindividual variability in body composition changes. Age was the main determinant of these changes, whereas menopausal status and hormone concentrations did not affect these changes significantly.
All authors state that they have no conflicts of interest.
The authors thank Annick Bourgeaud, Sylviane Ailloud, and Wafaa Wirane for excellent technical assistance.
Authors' roles: Conception and design: ESR, CKG, FM, RC. Acquisition of data: ESR, CKG, FM. Analysis and interpretation of data: ESR, CKG, FM, BC. Participation in drafting manuscript or revising it critically for important intellectual content: ESR, CKG, FM, BC, RC. Approving final version of manuscript: ESR, CKG, FM, BC, RC. ESR accepts responsibility for the integrity of the data analysis.