AMERICAN JOURNAL OF PHYSICAL ANTHROPOLOGY 00:00–00 (2014)

The Influence of Body Size on Adult Skeletal Age Estimation Methods Catherine E. Merritt* Department of Anthropology, University of Toronto, Toronto, ON M5S 2S2, Canada KEY WORDS age estimation; transition analysis; body size; body mass index (BMI); stature; body mass ABSTRACT Accurate age estimations are essential to archaeological and forensic analyses. However, reliability for adult skeletal age estimations is poor, especially for individuals over the age of 40 years. This is the first study to show that body size influences skeletal _ ¸can et al., Lovejoy et al., Buckage estimation. The Is berry and Chamberlain, and Suchey-Brooks age methods were tested on 764 adult skeletons from the HamannTodd and William Bass Collections. Statures ranged from 1.30 to 1.93 m and body masses ranged from 24.0 to 99.8 kg. Transition analysis was used to evaluate the differences in the age estimations. For all four methods, the smallest individuals have the lowest ages at transition and the largest individuals have the highest ages at transition. Short and light individuals are consistently underaged, while tall and heavy individuals are consistently overaged. When femoral length and femoral head

When human skeletal remains are found in prehistoric, historic, or forensic contexts, establishing age at death is an important step to reconstruct life histories, build demographic profiles, and identify victims of mass disasters, genocides, and homicides. The most common methods used to estimate age at death are based on changes to dental and skeletal features; these indicators provide relatively accurate and precise age estimations for juveniles who died during the period of growth and development (Crowder and Austin, 2005; Cameriere and Ferrante, 2008; Cardoso, 2008 a,b; Lee et al., 2008; Meinl et al., 2008), but are less accurate and precise for adult age estimations (Saunders et al., 1992; Bednarek and Sliwka, 2004; Martrille et al., 2007; Storey, 2007; Dorandeu et al., 2008; Konigsberg et al., 2008; Meinl et al., 2008). Most biological studies of whole organisms recognize the importance of controlling for body size (i.e. body mass index [BMI], stature, body mass) when analyzing and interpreting data (Stillwell et al., 2011). However, in biological anthropology, age estimation studies have not systematically considered body size as a variable that could influence our standards. Understanding how the skeletal aging process differs among individuals of varying BMIs, statures, and body masses may help explain some of the unattributed variance observed in skeletal age estimations. For example, analyses of past populations show a higher than expected young adult mortality, i.e., their adult ages at death are between 30 and 35 years of age (Storey, 2007; Roksandic and Armstrong, 2011). Previous research has shown that most past populations of Homo sapiens, as well as the majority of our hominin ancestors, were shorter and lighter than the Ó 2014 WILEY PERIODICALS, INC.

diameter are compared with the log-age model, results show the same trend as the known stature and body mass measurements. The skeletal remains of underweight individuals have fewer age markers while those of obese individuals have increased surface degeneration and osteophytic lipping. Tissue type and mechanical loading have been shown to affect bone turnover rates, and may explain the differing patterns of skeletal aging. From an archaeological perspective, the underaging of light, short individuals suggests the need to revisit the current research consensus on the young mortality rates of past populations. From a forensic perspective, understanding the influence of body size will impact efforts to identify victims of mass disasters, genocides, and homicides. Am J Phys Anthropol 000:000–000, 2014. VC 2014 Wiley Periodicals, Inc.

populations our current age standards were created on (see Table 1). Our current age standards were developed on late 19th century/early 20th century North American skeletal collections (i.e. Lovejoy et al. auricular surface method) and forensic collections from the 1970s and _ ¸can et al. fourth rib method, Suchey-Brooks 1980s (i.e. Is pubic symphysis method), both of which are composed of individuals that are generally taller and heavier than individuals of past populations. Researchers apply these standards to archaeological collections of smaller, lighter individuals with the understanding that there are differences in the life histories, lifestyles, and environments of these individuals, but without consideration for the ways in which body size affects skeletal aging. Similarly, forensic anthropologists apply these age estimation methods to modern human populations that are generally taller and heavier than the populations used to create the standards. With obesity rates in Canada and the Grant sponsor: Social Sciences and Humanities Research Council of Canada Doctoral Fellowship; Grant number: 752–2010-2124. *Correspondence to: Catherine E. Merritt, Department of Anthropology, University of Toronto,19 Russell Street, Toronto, ON M5S 2S2, Canada. E-mail: [email protected] Received 11 August 2014; revised 15 September 2014; accepted 15 September 2014 DOI: 10.1002/ajpa.22626 Published online 00 Month 2014 in Wiley Online Library (wileyonlinelibrary.com).

2

C.E. MERRITT TABLE 1. Stature and body mass estimations of hominin species

Species

Stature (cm)

Body mass (kg)

Australopithecus afarensis x 128.0 Std. Dev. 37.0 N 9 Australopithecus africanus x 126.5 Std. Dev. – N 12 Paranthropus robustus x 121.0 Std. Dev. – N 4 Paranthropus bosei x 130.5 Std. Dev. – N 2 Australopithecus sediba x 130.0 Std. Dev. – N 2 H. habilis x 141.0 Std. Dev. – N 5 Homo erectus x 170.0 Std. Dev. – N 2 Homo neanderthalensis x 162.4 Std. Dev. 3.6 N 17 Homo floresiensis x 106.0 Std. Dev. – N 2 Homo sapiens (Paleolithic European) x 163.6 Std. Dev. 6.0 N 41 Homo sapiens (modern European) x 164.4 Std. Dev. 5.6 N 482 Homo sapiens (modern Sub-Saharan Africa) x 158.5 Std. Dev. 7.8 N 109

– – 9 35.5 – 12 36.0 – 4 41.5 – 2 44.2 – 2 47.5 – 5 68.0 – 2 71.0 10.5 12 30.0 – 2 63.4 6.6 17 63.8 6.2 213 51.5 7.5 78

Stature is estimated from long bones, body mass is predicted by femoral head diameter or bi-iliac breadth (adapted from McHenry, 1992, 2005; Kappelman, 1996; Helmuth, 1998; Holliday, 2002; Ruff, 2010).

USA steadily increasing, understanding the ways in which modern populations age, and the ways in which skeletal aging is affected by body size, is important. Controlling for variables such as life history, lifestyle, and environment are not easy, but developing a factor to account for BMI, stature, and body mass, which can be measured and calculated from the skeleton, may be a realistic goal. The current study applies transition analysis to BMI (calculated using the recorded stature and body mass), stature (recorded in the skeletal collection), body mass (recorded in the skeletal collection), femoral length (measured as a proxy for unknown stature), and femoral head diameter (measured as a proxy for unknown body American Journal of Physical Anthropology

mass) for four age estimation methods. The Buckberry and Chamberlain method is further assessed using ANOVAs to test for differences among the body size groups for the individual traits recorded (i.e. surface texture, transverse organization, porosity, and apical lipping).

MATERIALS AND METHODS Materials There are several human skeletal collections with known age and stature at death; however, there are few established skeletal collections with known body mass at death. For this project, the Hamann-Todd Collection curated at the Cleveland Museum of Natural History in Cleveland, Ohio, and the William Bass Collection curated at the University of Tennessee in Knoxville, Tennessee, were used. The Hamann-Todd Collection was established in 1839 and skeletons were collected until 1938 by Carl August Hamann and T. Wingate Todd. The collection consists of approximately 3,100 human skeletons with both males and females represented. Ages at death range from fetal to 105 years, and most are of European or African-American descent. The William Bass Collection was established in 1981 by William Bass as a donor program in Knoxville, Tennessee, and the Anthropology Research Facility at the University of Tennessee continues to accept donors. Currently, the collection houses approximately 1,000 skeletons, both males and females, with birth years ranging from 1892 to 2011. Most individuals were born after 1940 and most are of European or African-American ancestry.

Sample selection Before visiting each collection, skeletons were age and size selected to ensure representation from a broad spectrum of stature and body mass categories. Age, sex, ancestry, stature, body mass, and cause of death were known for most individuals in both collections. Individuals with unknown statures, body masses, and causes of death were eliminated from the collection samples. Most of the skeletons in the Hamann-Todd collection are individuals with positively identified known ages, but some of the skeletons were derived from individuals where age was determined through soft tissue markers (Meindl et al., 1990). For this project, cases where ages at death were estimated from soft tissue markers were not used. Individuals were then eliminated from the sample based on the following criteria: 1. Skeletons of individuals who had a stature greater than 1.95 m (6.4 ft) and body mass greater than 102.1 kg (225 lbs), due to the small sample size, or; 2. Skeletons of individuals who had a stature of less than 1.60 m (5.4 ft) and body mass greater than 79.35 kg (175 lbs), due to the small sample size, or; 3. Skeletons of individuals who died of cancer, tuberculosis, or any other wasting disease, or; 4. Skeletons of individuals who had a stature greater than 1.64 m (5.4 ft) and body mass less than 45.35 kg (100 lbs), as the cause of death for the majority of these individuals was likely to be a wasting disease (i.e. cancer, tuberculosis, etc.). Skeletal inventories were not available before visits to each collection; therefore, inventories were conducted on site at the Hamann-Todd and William Bass Collections

3

INFLUENCE OF BODY SIZE ON AGE ESTIMATION TABLE 2. Sample distribution by sex and ancestry

Males Females European ancestry African ancestry Total

N

Mean age

Average stature (m/ft)

Stature range (m/ft)

Average body mass (kg/lbs)

Body mass range (kg/lbs)

540 224 550 214 764

52.6 51.2 54.1 45.0 51.5

1.74/5.7 1.60/5.3 1.69/5.6 1.72/5.6 1.70/5.6

1.30–1.93/4.3–6.3 1.30–1.87/4.3–6.1 1.30–1.93/4.3–6.3 1.50–1.92/4.9–6.3 1.30–1.93/4.3–6.3

68.9/152 57.7/127 66.7/147 63.1/139 65.5/145

32.2–99.8/71–220 24.0–99.8/53–220 26.3–99.8/58–220 24.0–99.8/53–220 24.0–99.8/53–220

TABLE 3. Total number of males, females, and individuals of European and African ancestry

European ancestry African ancestry Total

Males

Females

Total

407 133 540

143 81 224

550 214 764

Fig. 2. Ancestry distribution of the sample by age category.

Fig. 1. Sex distribution of the sample by age category.

before data collection. The descriptive statistics of the sample are in Tables 2 and 3 and the sex and ancestry distribution of the sample by age category are illustrated in Figures 1 and 2. BMI was calculated for each individual using the formula body mass (kg)/stature (m)2, where an individual’s body mass is divided by stature squared in order to estimate body fat (Keys, 1972). A potentially problematic area with BMI is ancestry. BMI was developed on individuals of European ancestry from North America, and its applicability to other populations has been questioned. Studies have found that BMI calculators do not work as well for individuals of African, Asian, and Pacific Island ancestries (Jackson et al., 2002; Rush et al., 2009); however, there are no recognized standards available to account for these variations. The WHO conducted an international expert consultation in 2002 to review the applicability of the BMI calculator to other populations, and specifically populations from Asia and the Pacific Islands; they concluded that, based on the incidence of type 2 diabetes and cardiovascular disease

in each BMI group, the current standards in use are applicable to all populations (WHO Expert Consultation, 2004). A second problematic aspect to BMI is that it does not always accurately reflect body fat composition. Since the BMI formula depends only upon stature and body mass, the assumptions made about the distribution between lean mass and fat tissue are only an approximation. BMI generally overestimates body fat on those with more lean body mass and underestimates excess body fat on those with less lean body mass (Romero-Corral et al., 2008). A study by Romero-Corral et al. (2008) examined 13,601 individuals from the United States’ Third National Health and Nutrition Examination Survey and reported that obesity (BMI >30) was present in 21% of men and 31% of women. Using body fat percentages (BF%), obesity was found to be significantly higher in women than men (62% BF% for women and 50% BF% for men). Despite this under-representation of obese individuals using BMI compared with BF%, BMI values in the intermediate BMI range of 20 to 30 were found to be associated with a wide range of body fat percentages. For men with a BMI of 25, about 20% have a BF% below 20% and about 10% have BF% above 30%. BMI is particularly inaccurate for athletes and individuals who regularly exercise, as higher muscle mass tends to classify these individuals in the overweight BMI category, even though their BF% frequently falls between 10 and 15%, which is below that of a sedentary person who has a normal BMI (Ode et al., 2007). Body American Journal of Physical Anthropology

4

C.E. MERRITT

Fig. 3. Distribution of sample by BMI for stature and body mass.

Fig. 5. Body mass distribution of the sample by age at death.

_ ¸can et al. stages TABLE 4. Sample distribution across Is Phase I II III IV V VI VII VIII

N

Mean age

Std. dev.

Median

Range

11 18 38 105 234 201 92 62

28.27 37.00 45.92 42.92 50.79 54.64 58.41 66.53

14.34 13.78 18.23 14.42 14.26 13.17 13.08 11.52

25 35 40 38 50 54 60 68

19–70 19–73 19–87 22–78 20–85 25–85 20–85 34–89

TABLE 5. Sample distribution across Lovejoy et al. stages Phase

Fig. 4. Stature distribution of the sample by age at death.

composition for athletes is often better calculated using measures of body fat, as determined by skin-fold measurements. Alternative methods to measure obesity, such as the body volume index and body frame size (Frisancho, 1984; Heuberger et al., 2007), have been proposed. Unfortunately these alternative measures to estimate body size are not available for use on skeletal remains. This study uses BMI, but also analyzes each stature and body mass independently. Figure 3 illustrates the demarcations for each BMI group and the number of individuals in each group. The normal-sized BMI category has more than double the number of individuals than the other BMI categories, as would be expected in a normally distributed population. Figures 4 and 5 show the distributions of stature and body mass by age. BMI and body mass, and to a lesser extent stature, are difficult to ascertain from the skeletal record. Femoral length and femoral head diameter measurements are often used in linear regression formulae to calculate stature and body mass, but this approach has large margins of error (Ruff et al., 1991, 2012; McHenry, 1992; American Journal of Physical Anthropology

I II III IV V VI VII VIII

N

Mean age

1 23 71 120 138 160 194 87

23.00 25.61 31.99 39.96 47.72 56.91 63.66 69.17

Std. dev. – 3.99 7.49 6.94 9.74 11.11 10.28 10.45

Median

Range

23 25 32 40 48 55 65 71

– 19–33 19–66 24–65 26–79 30–86 30–85 38–89

Grine et al., 1995). For this reason, body mass and BMI are rarely calculated in archaeological and forensic contexts. This study is the first to use femoral length and femoral head diameter measurements as a proxy for stature and body mass without applying them in a linear regression formula. Maximum femoral length and maximum femoral head diameter were collected on a subset of the sample (N 5 706) and tested in the cumulative probit model as body size variables. The femoral measurements may be more useful to biological anthropologists since soft tissues are rarely recovered to calculate body mass in a meaningful way.

Methods Four age estimation methods were applied to 764 individuals. The fourth rib age estimation method was _ ¸can et al. (1984, 1985), and the pelvic age developed by Is

5

INFLUENCE OF BODY SIZE ON AGE ESTIMATION TABLE 6. Sample distribution across Buckberry and Chamberlain stages Phase I II III IV V VII VIII

N

Mean age

Std. dev.

Median

Range

5 44 73 136 204 197 134

29.20 32.50 39.30 45.95 53.19 57.71 64.62

11.21 8.57 11.48 13.84 12.59 13.25 12.76

25 32 39 45 53 59 65

19–44 20–64 19–68 20–79 22–81 30–87 25–89

TABLE 7. Sample distribution across Suchey-Brooks stages Phase I II III IV V VI

N

Mean age

Std. dev.

Median

Range

8 22 51 191 327 194

21.25 27.09 33.98 40.40 55.82 66.82

2.66 6.23 6.70 9.15 11.30 10.55

20 25 32 39 55 68

19–25 20–45 22–52 20–69 53–85 49–89

TABLE 8. Descriptive statistics for the sample age-at-transition distributions in years from one phase to the next using each of the four age estimation methods Phase

Mean age

_ ¸can et al. method Is I/II-III 15.38 III-IV 22.07 IV-V 33.18 VI-VI 53.59 VI-VII 81.08 VII-VIII 109.22 Buckberry and Chamberlain method I/II-III 18.19 III-IV 27.83 IV-V 41.09 V-VI 58.64 VI-VII 80.73 Lovejoy et al. method I/II-III 21.71 III-IV 31.35 IV-V 40.39 V-VI 49.52 VI-VII 60.41 VII-VIII 81.49 Suchey-Brooks method I-II 19.49 II-III 22.89 III-IV 28.18 IV-V 44.08 V-VI 67.90

Std. err.

Std. dev.

7.84 4.11 1.70 1.38 2.82 4.00

22.89 24.81 22.21 28.01 26.54 22.01

2.61 1.72 1.18 1.21 2.69

15.74 18.13 19.98 23.71 24.88

1.40 0.77 0.63 0.66 0.78 2.22

6.62 7.72 9.21 11.38 13.46 17.56

1.15 1.23 1.13 0.64 1.06

3.37 6.88 9.99 10.20 15.42

estimation methods were developed by Lovejoy et al. (1985), Buckberry and Chamberlain (2002), and Suchey and Brooks (1990). Age assessments were performed on each skeleton without knowledge of the age, ancestry, BMI, stature, body mass, or cause of death for the specimen under consideration; the author had a research assistant for data collection at the Hamann-Todd Collection to help with this process. Sex was known for all _ ¸can et al. and Suchey-Brooks methindividuals as the Is ods have sex-specific criteria. All methods were applied as described by the authors. Casts for the SucheyBrooks method were used, and color image slides of the auricular surface for the Lovejoy et al. method were

_ ¸can et al. method using TABLE 9. Body size statistics for the Is the cumulative probit model Intercepts Model 1: log age and BMI Log age coefficient 5 21.125 BMI coefficient 5 0.130 Phase I/II-III III-IV IV-V V-VI VI-VII VII-VIII Model 2: log age and stature Log age coefficient 5 1.818 Stature coefficient 5 6.814 Phase I/II-III III-IV IV-V V-VI VI-VII VII-VIII Model 3: log age and body mass Log age coefficient 5 21.252 Body mass coefficient 5 0.037 Phase I/II-III III-IV IV-V V-VI VI-VII VII-VIII Model 4: log age femoral length Log age coefficient 5 1.602 Femoral length coefficient 5 0.024 Phase I/II-III III-IV IV-V V-VI VI-VII VII-VIII Model 5: log age femoral head diameter Log age coefficient 5 21.457 Femoral head diameter coefficient 5 20.027 Phase I/II-III III-IV IV-V V-VI VI-VII VII-VIII

2.925 3.690 4.559 5.585 6.471 7.115

25.355 24.572 23.685 22.641 21.744 21.102

3.777 4.567 5.467 6.530 7.442 8.107

26.261 25.788 25.072 24.088 23.194 22.550

5.543 6.022 6.747 7.730 8.620 9.258

used to score phases. In all cases, the authors’ specifications with regard to element side were followed. When scoring the pubic symphysis and auricular surface, the methods were applied to the left side; in cases where the left side was unavailable, the right side was used. When scoring the fourth ribs, the ribs were seriated by the _ ¸can et al. (1984, 1985) method was author, and the Is applied to the right fourth ribs; if the right fourth rib was unavailable or damaged, the right third or fifth ribs were used (Loth et al., 1994). Each individual was assigned to an age phase, as specified by each method. Tables 4 to 7 show the descriptive statistics and sample distributions of each method for each phase. For the _ ¸can et al., Lovejoy et al., and Buckberry and ChamberIs lain methods, Phases I and II were combined as there were small sample sizes in Phase I. American Journal of Physical Anthropology

6

C.E. MERRITT

_ ¸can et al. method. The ages-at-transition (a) for BMI, (b) for stature, (c) for body mass, Fig. 6. (a–e) Ages-at-transition for the Is (d) for femoral length, and (e) for femoral head diameter.

American Journal of Physical Anthropology

INFLUENCE OF BODY SIZE ON AGE ESTIMATION

7

Fig. 6. Continued

Maximum femoral length and maximum femoral head diameter were measured as described by Buikstra and Ubelaker (1994). Maximum femoral length was measured using a Paleo-Tech Lightweight Field Osteometric Board and maximum femoral head diameter was measured using 150 mm digital Muzika calipers. When assessing the Buckberry and Chamberlain method, individual traits are scored in order to create a composite score. These trait scores were analyzed separately in groups—for BMI, individuals were analyzed by BMI group, and for the other body size traits, individuals were separated into four approximately equal groups using a cluster analysis in order to assess differences in the scoring of traits between the lightest/shortest and heaviest/tallest groups.

Statistical analyses A cumulative probit model was used to calculate the mean, standard deviation, and standard error of the ages-of-transition for each age estimation phase. As described by Boldsen et al. and Konigsberg et al. else-

where (Boldsen et al., 2002; Konigsberg et al., 2008), a cumulative probit model, or transition analysis, is an estimation procedure that calculates the probability of the timing (i.e. age) of the transition from one phase to another. The cumulative probit model for this study was created using the age estimation phase assessed for each _ ¸can et al., individuals as the dependent variable (i.e. Is Lovejoy et al., Buckberry and Chamberlain, SucheyBrooks phases), and the independent variables were logage and one of the body size variables (i.e. BMI, stature, body mass, femoral length, femoral head diameter). A likelihood ratio test was used to assess significant differences between the log-age models and the log-age and body size variable models. For the Buckberry and Chamberlain method, individual trait scores such as transverse organization and apical lipping were assessed to ascertain whether there were different patterns among the body size groups using a one-way ANOVA and Tukey’s honest significant difference (HSD) post hoc test. Tukey’s HSD was used for the post hoc test as it is a more conservative test to control for type 1 errors, and it American Journal of Physical Anthropology

8

C.E. MERRITT TABLE 10. Body size statistics for the Lovejoy et al. method using the cumulative probit model Intercepts

Model 1: log age and BMI Log age coefficient 5 25.124 BMI coefficient 5 20.148 Phase I/II-III III-IV IV-V V-VI VI-VII VII-VIII Model 2: log age and stature Log age coefficient 5 25.298 Stature coefficient 5 23.320 Phase I/II-III III-IV IV-V V-VI VI-VII VII-VIII Model 3: log age and body mass Log age coefficient 5 25.214 Body mass coefficient 5 20.058 Phase I/II-III III-IV IV-V V-VI VI-VII Model 4: log age femoral length Log age coefficient 5 26.338 Femoral length coefficient 5 20.021 Phase I/II-III III-IV IV-V V-VI VI-VII VII-VIII Model 5: log age femoral head diameter Log age coefficient 5 27.573 Femoral head diameter coefficient 5 20.305 Phase I/II-III III-IV IV-V V-VI VI-VII VII-VIII

_ ¸can et al. method Is

16.774 18.224 19.342 20.238 21.112 22.375

19.511 20.970 22.103 23.009 23.896 25.165

17.367 18.845 19.975 20.880 21.758 23.024

23.090 24.532 25.674 26.556 27.429 28.703

27.651 29.121 30.273 31.157 32.026 33.298

produces more reliable results for uneven sample sizes. The cumulative probit analyses were performed with the program “R” (http://www.r_project.org) and the descriptive statistics and one-way ANOVAs were performed using SPSS 21.0 (IBM Corp, 2012).

RESULTS Table 8 shows the descriptive statistics for the ages-oftransition for each age estimation method generated from the cumulative probit model without any body size _ ¸can et al. and Buckberry and factors considered. The Is Chamberlain methods show the most variability among the ages-of-transition, with high standard errors and high standard deviations for the predicted probability ages-of-transition between each phase. The Lovejoy et al. and Suchey-Brooks methods display the expected American Journal of Physical Anthropology

lower standard errors and lower standard deviations in the lower age phases and higher standard errors and higher standard deviations for the higher age phases.

The likelihood ratio test between the log-age model and the BMI/log-age model shows significant differences between the transition ages for each phase (v2(2) 5 34.533, P < 0.001). As BMI increases, the transition age for each phase also increases. This trend is similar for stature (v2(2) 5 52.711, P < 0.001), body mass (v2(2) 5 80.672, P < 0.001), femoral length (v2(2) 5 17.790, P < 0.001), and femoral head diameter (v2(2) 5 16.704, P 5 0.002). Table 9 shows the statistics from the log-age/ body size cumulative probit regression applied to the _ ¸can et al. method. Figure 6a–e shows sample for the Is the age-at-transition distributions for all five body size variables. The three lines in each figure represent the predicted transition distributions for individuals in each body size category; for example, in Figure 6a the three lines represent the predicted transition distributions for individuals with BMIs of 19.22, 22.46, and 25.72, in Figure 6b the three lines represent the predicted transition distributions for individuals with statures of 1.62 m, 1.71 m, and 1.78 m, and so on.

Lovejoy et al. Method The likelihood ratio test between the log-age model and the BMI/log-age model shows no significant differences between the transition ages for each phase for the two models (v2(2) 5 3.829, P 5 0.147). When stature and body mass are considered separately, there are significant differences between the log-age and body size models [stature (v2(2) 5 16.316, P < 0.001), body mass (v2(2) 5 13.773, P 5 0.001)], with taller and heavier individuals having higher ages-of-transition. Similarly, when femoral length and femoral head diameter are compared with the log-age model, both are significant and show a similar trend [femoral length (v2(2) 5 14.384, P < 0.001), and femoral head diameter (v2(2) 5 17.156, P < 0.001)]. Table 10 shows the statistics from the log-age/body size cumulative probit regression applied to the sample for the Lovejoy et al. method, and Figure 7a–e illustrates the age-at-transition distributions for all five body size variables.

Buckberry and Chamberlain method The likelihood ratio test between the log-age model and the BMI/log-age model shows significant differences between the transition ages for each phase (v2(2) 5 14.354, P < 0.001). As BMI increases, the transition age for each phase also increases. This trend is similar for stature (v2(2) 5 13.803, P 5 0.001), body mass (v2(2) 5 25.686, P < 0.001), femoral length (v2(2) 5 19.113, P < 0.001), and femoral head diameter (v2(2) 5 9.333, P 5 0.009). Table 11 shows the statistics from the log-age/ body size cumulative probit regression applied to the sample for the Buckberry and Chamberlain method, and Figure 8a–e illustrate the age-at-transition distributions for all five body size variables.

Suchey-Brooks method The likelihood ratio test between the log-age model and the BMI/log-age model shows significant differences between the transition ages for each phase

INFLUENCE OF BODY SIZE ON AGE ESTIMATION

9

Fig. 7. (a–e) Ages-at-transition for the Lovejoy et al. method. The ages-at-transition (a) for BMI, (b) for stature, (c) for body mass, (d) for femoral length, and (e) for femoral head diameter.

American Journal of Physical Anthropology

10

C.E. MERRITT

Fig. 7. Continued

(v2(2) 5 17.106, P < 0.001). As BMI increases, the transition age for each phase also increases. This trend is similar for stature (v2(2) 5 19.896, P < 0.001), body mass (v2(2) 5 38.669, P < 0.001), femoral length (v2(2) 5 18.263, P < 0.001), femoral head diameter (v2(2) 5 11.264, P 5 0.004). Table 12 shows the statistics from the logage/body size cumulative probit regression applied to the sample for the Suchey-Brooks method, and Figure 9a–e show the age-at-transition distributions for all five body size variables.

Additional observations For the Buckberry and Chamberlain auricular surface method, individual traits such as transverse organization, surface texture, porosity, and apical activity were assessed and scored separately. Further analyses using one-way ANOVAs were conducted on each of these features to ascertain whether there are different patterns among BMI, stature, body mass, femoral length, and femoral head diameter groups for these individual traits. American Journal of Physical Anthropology

For this test, a cluster analysis was used to divide stature, body mass, femoral length, and femoral head diameter into four equal groups. See Tables 13 and 14 for the descriptive statistics for each group. There are five features of the auricular surface scored using the Buckberry and Chamberlain method: transverse organization on a scale of 1 to 5, surface texture on a scale of 1 to 5, microporosity on a scale of 1 to 3, macroporosity on a scale of 1 to 3, and apical activity on a scale of 1 to 3. The lowest scores categorize early maturation stages while the highest scores categorize progressive degenerative changes. When separated by BMI, one-way ANOVA tests show there are statistically significant differences among the BMI groups for three of the five traits: surface texture, microporosity, and apical activity (Table 15). Tukey’s HSD post hoc tests show that obese individuals have significantly higher scores for surface texture compared with the individuals in the other BMI groups and obese individuals have significantly higher microporosity scores compared with normal size and overweight individuals. Underweight

INFLUENCE OF BODY SIZE ON AGE ESTIMATION TABLE 11. Body size statistics for the Buckberry and Chamberlain method using the cumulative probit model Intercepts Model 1: log age and BMI Log age coefficient 5 21.530 BMI coefficient 5 0.118 Phase I/II-III III-IV IV-V V-VI VI-VII Model 2: log age and stature Log age coefficient 5 23.698 Stature coefficient 5 24.252 Phase I/II-III III-IV IV-V V-VI VI-VII Model 3: log age and body mass Log age coefficient 5 22.085 Body mass coefficient 5 0.007 Phase I/II-III III-IV IV-V V-VI VI-VII Model 4: log age femoral length Log age coefficient 5 23.742 Femoral length coefficient 5 20.018 Phase I/II-III III-IV IV-V V-VI VI-VII Model 5: log age femoral head diameter Log age coefficient 5 25.666 Femoral head diameter coefficient 5 20.311 Phase I/II-III III-IV IV-V V-VI VI-VII

4.604 5.305 6.061 6.900 7.800

14.620 15.334 16.096 16.930 17.820

6.945 7.657 8.422 9.265 10.168

14.876 15.586 16.365 17.190 18.055

21.058 21.768 22.543 23.360 24.215

individuals have significantly lower apical activity compared with all BMI groups, obese individuals have significantly higher apical activity scores compared with all BMI groups, and normal-sized and overweight individuals’ apical activity scores are significantly different from underweight and obese individuals. Although not significant in the ANOVA, there are differences in the macroporosity scores; underweight individuals have significantly higher macroporosity scores compared with obese individuals (see Table 16). When separated by stature, one-way ANOVA tests show there are statistically significant differences among stature groups for four of the five traits: surface texture, microporosity, macroporosity, and apical activity (Table 17). Tukey’s HSD post hoc tests show that individuals in Group 2 have significantly higher surface texture scores compared with individuals in Group 1; individuals in Group 3 have significantly lower microporosity scores compared with individuals in Group 4; individuals in Group 1 have significantly higher macroporosity scores

11

compared with individuals in all other stature groups; and individuals in Group 1 also have significantly lower apical activity scores compared with individuals in all other stature groups (see Table 18). When separated by body mass, one-way ANOVA tests show there are statistically significant differences among stature groups for three of the five traits: surface texture, macroporosity, and apical activity (Table 19). Tukey’s HSD post hoc tests show that individuals in Group 1 have significantly lower surface texture scores compared with individuals in Groups 3 and 4; individuals in Group 1 have significantly higher macroporosity scores compared with individuals in Groups 3 and 4; individuals in Groups 1 and 2 have significantly lower apical activity scores compared with individuals in Groups 3 and 4. Although not statistically significant in the ANOVA, there are differences for transverse organization; individuals in Group 2 have significantly lower scores compared with individuals in Group 4 (see Table 20). When separated by maximum femoral length, one-way ANOVA tests show there are statistically significant differences for two of the five traits: surface texture and apical activity (Table 21). Tukey’s HSD post hoc tests show that individuals in Group 1 have significantly lower surface texture scores compared with individuals in Groups 2 and 4, and individuals in Group 1 have significantly lower apical activity scores compared with individuals in all other groups, while individuals in Group 4 have the highest apical activity scores (see Table 22). When separated by maximum femoral head diameter, one-way ANOVA tests show there are statistically significant differences for two of the five traits: surface texture and apical activity (Table 23). Tukey’s HSD post hoc tests show that individuals in Group 1 have significantly lower surface texture scores compared with individuals in all other groups, and individuals in Group 1 also have significantly lower apical activity scores compared with individuals in all other groups, while individuals in Group 4 have significantly higher apical activity scores compared with individuals in Groups 1 and 2 (see Table 24).

DISCUSSION Three important patterns emerge from this data. First, individuals in the shorter and lighter body size groups have lower ages-at-transition compared with taller and heavier individuals, while taller and heavier individuals have higher ages-at-transition compared with all groups; second, maximum femoral length and femoral head diameter can be used a proxies for stature and body mass; and third, there are differences among the individual scoring features observed among individuals of differing body sizes.

Factors that affect skeletal aging Light and short individuals in this study show lower ages-at-transition compared with all other groups, suggesting that underweight individuals are subject to a decreased rate of skeletal aging. Conversely, heavy and tall individuals show higher ages-at-transition, suggesting that obesity accelerates skeletal aging. There are a variety of factors that may contribute to the differences in age estimations among BMI, stature, and body mass groups. The main ones are discussed below: bone American Journal of Physical Anthropology

12

C.E. MERRITT

Fig. 8. (a–e): Ages-at-transition for the Buckberry and Chamberlain method. The ages-at-transition (a) for BMI, (b) for stature, (c) for body mass, (d) for femoral length, and (e) for femoral head diameter.

American Journal of Physical Anthropology

INFLUENCE OF BODY SIZE ON AGE ESTIMATION

13

Fig. 8. Continued

remodeling rates and bone mineral density (BMD), nutrition, physical activity and occupation, and hormone levels. Bone remodeling rates and bone mineral density (BMD). As an individual ages, the rates of bone removal outpace the rates of bone formation (Cao et al., 2005, 2007; Pietschmann et al., 2007), especially in postmenopausal women who have decreased levels of estrogen, which is important in bone formation (Kaptoge et al., 2003). Studies have shown that older bone might be preferentially selected for osteoclast resorption, and compounded with the decrease in bone matrix deposited by osteoblasts, the rate of bone removal occurs faster than bone deposition as an individual ages (Lee et al., 2005; Szulc and Seeman, 2009). These changes in bone remodeling rates influence BMD scores; most individuals reach their peak BMD around the age of 30 years, and after this time BMD begins to decrease in a predictable fashion (Lekamwasam et al., 2009). These changes at

the microscopic level also translate to changes at the macroscopic level. Between the ages of 30 and 40 years, when BMD is beginning to decline, the bone surface is beginning to show signs of aging—striae become dense and granular, billowing becomes smooth, microporosity begins to show on the surface—and these changes are evaluated by macroscopic age estimation methods. As an individual continues to age, and the rates of bone remodeling continue to outpace the rates of bone formation, BMD values decrease, and the surface of the bone reflects these changes—dense granularity becomes course, smooth surfaces become roughened and concave, macroporosity begins to replace microporosity, osteophytic activity occurs, and bone becomes more fragile. As demonstrated by this study, these surface changes are also influenced by body size, with lightweight individuals showing a slower pace of remodeling compared with average-sized and heavier individuals. When looking at the individual scoring traits of the Buckberry and Chamberlain method, increased surface texture and apical activity in heavier individuals are driving the age American Journal of Physical Anthropology

14

C.E. MERRITT

TABLE 12. Body size statistics for the Suchey-Brooks method using the cumulative probit model Intercept Model 1: log age and BMI Log age coefficient 5 24.704 BMI coefficient 5 20.117 Phase I/II-III III-IV IV-V V-VI VI-VII Model 2: log age and stature Log age coefficient 5 24.275 Stature coefficient 5 21.546 Phase I/II-III III-IV IV-V V-VI VI-VII Model 3: log age and body mass Log age coefficient 5 24.907 Body mass coefficient 5 20.050 Phase I/II-III III-IV IV-V V-VI VI-VII Model 4: log age femoral length Log age coefficient 5 24.181 Femoral length coefficient 5 20.003 Phase I/II-III III-IV IV-V V-VI VI-VII Model 5: log age femoral head diameter Log age coefficient 5 26.929 Femoral head diameter coefficient 5 20.266 Phase I/II-III III-IV IV-V V-VI VI-VII

14.865 15.825 16.900 18.574 20.453

15.262 16.214 17.295 18.979 20.861

15.860 16.829 17.941 19.662 21.571

13.904 15.007 15.983 17.623 19.526

24.508 25.635 26.645 28.306 30.208

estimations of obese individuals higher, while lower surface texture and apical activity scores are under-aging lighter and shorter individuals. Bone remodeling and body size. When considering the rates of bone remodeling as measured by BMD on individuals of differing body sizes, underweight individuals generally display lower BMD than individuals who are within the normal, overweight, and obese BMI range, or heavier body masses, and have an increased risk for osteoporosis (Shah et al., 2011). Examining the ways in which under-nutrition and obesity affect bone remodeling and BMD may help explain the differences in skeletal aging observed in this study. Undernutrition and BMD. The combination of poor nutrition and low amounts of fat tissue has been implicated in lower bone turnover rates for underweight individuals (Coin et al., 2000, 2008). This low bone American Journal of Physical Anthropology

remodeling rate compensates for the lack of nutrients to support bone growth, but only to a certain extent. A lifetime of poor nutrition and low body mass will ultimately lead to low BMD, and increased risk of osteoporosis in older individuals (Coin et al., 2008). A study by Fern andez-Garcıa et al. (2009) comparing “healthy” females with low body mass with “unhealthy” females with low body mass (i.e. females with anorexia nervosa) found no significant differences among BMD scores between the two groups, with both exhibiting lower than expected BMD compared with the control group of normal-sized females, indicating that nutrition may not be as important as body mass in skeletal health. The current study has shown that low body mass is associated with the underaging of individuals, suggesting the protective effect of low remodeling rates in underweight individuals is also seen on the bone surface and influences skeletal age estimations. Obesity and BMD. When considering the rates of bone remodeling measured by BMD on obese individuals, the pattern of BMD is less clear. Some studies have shown that obesity increases BMD, while other studies have shown that obesity decreases BMD. Researchers who argue that obesity increases BMD have shown that as body mass increases, there is increased mechanical stress on bone, resulting in decreased bone resorption and higher BMD (Felson et al., 1993; Looker et al., 2001; Lim et al., 2004). Researchers who argue that obesity decreases BMD do so based on the following evidence: First, obesity has been shown to decrease bone formation (i.e. osteoblastogenesis) while increasing fat formation (i.e. adipogenesis). Typically, mechanical loading promotes osteoblast differentiation and inhibits adipose formation; however, in obese individuals, the gene that promotes this action is downregulated, leading to a decrease in osteoblast differentiation and an increase in adipocyte differentiation (Cao, 2011). Second, obesity has also been shown to increase bone resorption through the upregulation of proinflammatory genes, which stimulate osteoclast activity. There has also been evidence that bone marrow adipocytes might promote osteoclastogenesis (Cao, 2011). Third, obesity is associated with higher levels of leptin and decreased levels of adiponectin. Higher levels in leptin in obese animals have been found to be a negative regulator of bone mass, and decreased levels of adiponectin increase osteoclastogenesis and bone resorption (Cao, 2011). The results from this study show that individuals in the obese BMI group and heavy body mass group do have accelerated aging compared with individuals in the underweight BMI group and light body mass group, suggesting support for the second of these two observations—heavier individuals display increased rates of bone remodeling. Research on the ways in which visceral fat and muscle mass influence bone’s remodeling might provide one explanation for the accelerated aging of individuals in the obese BMI group and heavy body mass group compared with individuals in the underweight BMI group and light body mass group. Two studies have shown that differences in the ratio of visceral fat to lean muscle exhibit varied rates of bone loss (Gnudi et al., 2007; Bredella et al., 2011). Individuals with high body mass and more lean muscle, which decreases the pace of bone remodeling, had higher BMD values, and individuals with high body mass and more visceral fat, which increases the pace of bone remodeling, had lower BMD

INFLUENCE OF BODY SIZE ON AGE ESTIMATION

15

Fig. 9. (a–e): Ages-at-transition for the Suchey-Brooks method. The ages-at-transition (a) for BMI, (b) for stature, (c) for body mass, (d) for femoral length, and (e) for femoral head diameter.

American Journal of Physical Anthropology

16

C.E. MERRITT

Fig. 9. Continued

TABLE 13. Distribution of the combined sample by BMI, including number of individuals in each category, average recorded stature and body mass, and stature and body mass ranges BMI Underweight Normal Overweight Obese

N

Mean age (yrs)

Average stature (m/ft)

Stature range (m/ft)

Average body mass (kg/lbs)

Body mass range (kg/lbs)

144 420 181 51

54.0 51.9 50.7 56.6

1.68/5.5 1.70/5.6 1.70/5.6 1.67/5.5

1.46–1.92/4.8–6.3 1.35–1.93/4.2–6.3 1.40–1.90/4.5–6.2 1.30–1.87/4.3–6.1

46.5/103 63.6/140 78.1/172 90.7/200

24.0–83.9/53–185 42.2–97.5/93–215 51.3–99.8/113–220 63.5–99.8/140–220

values. These results suggest that the accelerated skeletal aging of obese individuals observed in this study may be more dependent on the distribution and type of tissue (i.e. fat vs. muscle) rather than body mass or BMI group. Physical activity and occupation. Another potential explanation for the decelerated aging of lighter individuAmerican Journal of Physical Anthropology

als observed in this study is physical activity. Most of the underweight individuals used in this study are from the Hamann-Todd collection, which is predominantly comprised of European male migrant workers. The occupations recorded or found in census data for some individuals suggest that most worked labor intensive jobs (for example, bricklayer, iron worker, etc.: personal communication, Lyman Jellema, Cleveland Museum of Natural History). These labor-intensive activities suggest

17

INFLUENCE OF BODY SIZE ON AGE ESTIMATION TABLE 14. Descriptive statistics for each body size group from the cluster analysis Stature

Group 1 2 3 4

N 198 229 230 139

Mean age (yrs) 54.28 51.60 51.07 48.26

Mean stature (m) 1.55 1.67 1.76 1.85

Stature range (m) 1.20–1.60 1.61–1.72 1.73–1.80 1.81–1.94

Std. dev. 0.0533 0.0328 0.0259 0.0277

Body mass

Group

N 166 240 202 156

Mean body mass (kg) 45.59 59.61 72.13 87.96

Body mass range (kg) 24.04–54.37 54.38–66.10 66.11–80.20 80.21–99.79

Std. dev.

1 2 3 4

Mean age (yrs) 54.72 50.97 49.54 51.42

Group

N 173 209 193 134

Mean femoral length (mm) 408.71 445.11 473.76 502.91

Femoral length range (mm) 331–429 430–460 461–489 490–552

Std. dev.

1 2 3 4

Mean age (yrs) 52.25 52.68 49.89 50.17

Group

N 129 189 182 209

Mean femoral head diameter (mm) 40.63 44.56 48.04 51.58

Femoral head diameter range (mm) 36–42 43–46 47–49 50–62

Std. dev.

1 2 3 4

Mean age (yrs) 50.54 50.94 48.04 51.58

Femoral length

Femoral head diameter

6.484 3.560 3.840 5.802

16.53 8.74 7.82 12.29

1.36 1.73 0.778 1.75

TABLE 15. One-way ANOVA results for the Buckberry and Chamberlain method individual scoring features for BMI groups

Transverse organization Between groups Within groups Total Surface texture Between groups Within groups Total Microporosity Between groups Within groups Total Macroporosity Between groups Within groups Total Apical activity Between groups Within groups Total

Sum of squares

df

Mean square

F

3.795 641.960 645.755

3 758 761

1.265 0.847

1.493

0.215

25.537 919.924 945.461

3 758 761

8.512 1.214

7.014