- Email: [email protected]
The relative accuracy and reliability of histological aging methods
181
Forensic Science International, 15 (1980) 181 - 190 @ Elsevier Sequoia S.A., Lausanne - Printed in the Netherlands
THE RELATIVE ACCURACY AGING METHODS
AND
RELIABILITY
OF HISTOLOGICAL
SAMUEL D. STOUT and SARAH J. GEHLERT Department
of Anthropology,
(Received October 1,1979;
University
of Missouri,
Columbia,
Missouri 65211
(U.S.A.)
accepted December 10,1979)
Introduction Several histological methods for estimating the age of skeletal remains are available [ 1 - 31. These methods are especially useful in physical anthropology and forensic medicine, since they can provide greater accuracy of age estimation for adult skeletal remains than do traditional gross morphological methods. Histological aging methods utilize various combinations of histomorphological features such as intact osteons, fragmentary o&eons, nonHaversian bone, and primary vascular canals. They vary with respect to the number of fields read, the field size, and anatomical location. Predicting formulas are based upon specific sample populations for which age and health status are known. A requisite to the broad application of histological aging methods, especially to such diverse samples as those from archeological populations, is to test the accuracy of each method on population samples other than those from which the parameters were derived. This is also the only way in which the relative accuracy of various methods can be determined. An earlier study [4] compared only Kerley’s femoral intact osteon predicting formula to that of Ahlqvist and Damsten [2] . The present study compares the reliabilities and accuracies among all Haversian age-predicting parameters developed by Kerley [ 11 and Ahlqvist and Damsten [2] .
Materials
and methods
Midshaft sections were taken from the femurs, tibias, and fibulas of each of 13 individuals, obtained from dissecting room collections. Sex, age, and cause of death (except one) were available for all individuals (see Table 1). None of the samples included in the study were reported to have suffered from primary bone disease. All midshaft sections were embedded in epoxide, cut on a Buehler Isomet saw, and ground to a thickness of approximately 100 pm. Sections were mounted, and then read according to the methods of Kerley [l] and Ahlqvist and Damsten [ 21 . Only Haversian bone parameters were used.
182 TABLE
1
Sex, age. and cause of death of present Sex
Age
sample
Cause of death
(years) F M F F M M M M F F M F M
13 33 39 40 46 51 60 67 72 81 82 90 102
Brain stem tumor Cardiac arrest Hemorrhage Gunshot wound Electrocution Carbon monoxide poisoning Cardiopulmonary arrest Myocardial infarction Carcinoma of breast Cerebral hemorrhage Unknown Arteriosclerotic cardiovascular Pulmonary embolus
disc
For each specimen microscopic fields were located and read independently by two observers. For Kerley’s method, an ausJena scope combined with matched eyepieces fitted with a Merz [5] counting reticule was used. Our grid size measured with a stage micrometer is 0.970 mm2. Using Kerley’s reported field size of 2.06 mm2 [6] a correction factor of 2.1 was calculated, and all counts of intact and fragmentary osteons were multiplied by this factor. Ahlqvist and Damsten’s [2] method was carried out using a Nikon L-Ke scope fitted with a lOOsquare grid. Since our grid measured 1 mm2, as did Ahlqvist and Damsten’s, a field-size correction factor was not necessary. Ahlqvist and Damsten’s original predicting formula [2] and the revised prediction formulas for Kerley’s methods [6] were used in this study. It has been reported [4] that the accuracy of particular histological aging methods is not the same for samples with different age ranges. We therefore performed our analysis on subsamples with age ranges of 13 - 51 and 60 - 102 years, as well as on the sample as a whole. The relative accuracies of the various histological aging methods were determined by comparing the average standard deviations of the differences in years (mean differences) between predicted and known age using each of the bone sampling sites (femur, tibia, fibula) and each parameter (intact osteon count, osteon fragment count and percentage Haversian bone) where applicable. Mean differences between ages predicted by the observers were used as measures of interobserver error, and served as measures of reliability among the methods. The method of choice should be both reliable and accurate. The statistical significance of interobserver differences for each parameter and between predicted and known ages for each parameter was tested by a t-test.
183
Kerley and Ubelaker
Ahlqvist and Damsten
Fig. 1. Histogram for the total sample (13 - lo&year age range) showing: (1) the differences between predicted and known ages, and (2) the differences between the age estimates of the two observers, for each of Kerley’s predicting formulas and profile method, as well as for the modified method of Ahlqvist and Damsten. Results for age estimates based upon averaging ages from Kerley’s predicting formulas are included.
Results No statistical significance was found between observer estimates at the p = 0.05 level of confidence. Differences between each observer estimate and known age were also not significant at the p = 0.05 level. Figures 1 - 3 present the mean differences (average standard deviations) between known and predicted ages, and between observers.
Ftbulat fragments
Fibular intacfs
Ttblal fragments
Tibia1 l”MCtS
Femoral fragments
Femoral mtactr
ProfIle
P
w
I
0)
I
m
I
I
(D
I
I I
0‘
Mean difference m
I
2
(standard
I
E;
deviation
I
w‘
I
p’
of differences)
I
;;
in years z
I
I
y‘ I
6 0”
1
Ci
I
E rp
185
Kerley and Ubelaker
Ahlqvist and Damsten
Fig. 3. Histogram for the 60 - 102-year age range showing: (1) the differences between predicted and known ages, and (2) the differences between the age estimates of the two observers, for each of Kerley’s predicting formulas and profile method, as well as for the modified method of Ahlqvist and Damsten. Results for age estimates based upon averaging ages from Kerley’s formulas are included.
For the total sample, averaging the predicted ages by each of Kerley’s regression formulas (mean regression) produced the greatest accuracy, and reliability, followed by femoral osteon fragments. Accuracy for the remaining methods decreased in the following order: femoral intact osteons, fibular osteon fragments, Kerley’s profile method, tibial intact osteons, Ahlqvist and Damsten’s method, fibular intact osteons, and tibial osteon fragments (see Table 2). Reliability of the remaining methods, in decreasing order, is as follows: femoral intact osteons, Kerley’s profile method, tibial intact osteons, Ahlqvist and Damsten’s method, tibial osteon fragments, fibular osteon fragments, and fibular intact osteons (see Fig. 1).
186 TABLE
2
Relative accuracies for histological aging methods (Oi = intact osteons; Or = osteon fragments; A & D = modified method of Ahlqvist and Damsten; profile = Kerley’s profile method; mean regression = age arrived at by averaging the predicted ages by each of Kerley’s regression formulas) Present Total
sample range (13
- 102 years)
Mean regression Femur Or Femur Oi Fibula Or Profile Tibia Oi A&D Fibula Or Tibia Of *Relative
accuracies
13 - 51-year
age range
Femur Of Femur Or Mean regression Profile Fibula Of Tibia Oi A & D modified method Fibula Oi Tibia Or based upon
60 - 102-year range
age
Fibula Of Femur Or A&D Mean regression Tibia Oi Profile Fibula Ot
Kerley and Ubelaker*+t
Fibula Of Femur Of Fibula Of Tibia Or Femur Oi Tibia Oi
Femur Oi Tibia Of
mean-square residual values of Kerley and Ubelaker and profile methods unavailable.
[ 61.
‘Comparative data for mean regression
When only the 13 - 51-year age range is considered, femoral intact osteons and fragmented osteons for the femur are the most accurate, followed by mean regression, Kerley’s profile method, fibular osteon fragments, tibial intact osteons, Ahlqvist and Damsten’s method, fibular intact osteons, and tibial osteon fragments (see Table 2). Reliability in decreasing order is: mean regression, Kerley’s profile method, Ahlqvist and Damsten’s method, femoral intact osteons, femoral osteon fragments, tibial osteon fragments, tibial intact o&eons, fibular o&eon fragments, and fibular intact osteons (see Fig. 2). Accuracy for the 60 - 102-year age range was greatest for fibular osteon fragments and femoral osteon fragments, followed by Ahlqvist and Damsten’s method, mean regression, tibial intact osteons, Kerley’s profile method, fibular intact osteons, femoral intact o&eons, and tibial osteon fragments (see Table 2). Reliability decreased as follows: mean regression, femoral osteon fragments, femoral intact o&eons, fibular intact osteons, tibial intact o&eons, Ahlqvist and Damsten’s method, Kerley’s profile method, tibial osteon fragments, and fibular osteon fragments (see Fig. 3).
Discussion This paper evaluates the relative accuracy and reliability of several histological parameters of bone currently used to predict age at death [ 1, 21. An earlier comparison of these methods by Bouvier and Ubelaker [4] con-
187
eluded that Kerley’s method was the more accurate of the two. In their study, however, only the regression formulas for femoral intact osteons and percentage Haversian bone were used. Also, since the Bouvier and Ubelaker study used femoral sections from Kerley’s original sample, it might be argued that greater accuracy by the Kerley method would be expected, since predicting formulas are more accurate for the samples from which the parameters were derived. When methods which use more than one bone to predict age are used, errors from spatial variance and/or incoherence are reduced by increasing the amount of bone read [ 71. The use of Kerley’s profile method is preferable, therefore, to any “single bone” regression formula for predicting age. When using the profile method, however, problems arise from non-overlapping of the age-range estimates for each of the bone sampling sites [ 1, 81. If estimates from different bones do not overlap, a single estimate cannot be made. In the present study, averaging ages determined by each of Kerley’s six regression formulas (mean regression) produced the most accurate and reliable age estimates for our sample as a whole. Using the mean of ages estimated by Kerley’s six formulas also avoids the problem of non-overlapping predicted age ranges encountered with Kerley’s profile method. In the 13 - 51-year age subsample, mean regression and Kerley’s femoral intact osteon and osteon fragment equations were the most accurate predictors of age. Of these three, femoral osteon fragments were the most accurate, but mean regression had the least interobserver error. However, the mean differences from known age for the three methods were quite similar, + 7.38, + 6.28 and + 6.06, respectively, and interobserver error for mean regression was half that found for femoral intact osteons or osteon fragments. For skeletal samples that fall within this age range, mean regression would, therefore, be the method of choice. In the 60 - 102-year age range, estimates based upon femoral osteon fragments were most accurate. Fibular osteon fragments were of comparable accuracy, but considerable inter-observer error occurred with their use. Ahlqvist and Damsten’s method was slightly less accurate than Kerley’s fibular osteon fragment formula, but since it predicted age with a mean difference of 10 years, and exhibited much less interobserver error, it is preferable to fibular osteon fragments. In this older age range, mean regression improved reliability, but its accuracy was substantially less than the three methods noted above. When the entire age range is considered, mean regression is probably the best method with which to age unknown skeletal remains. Further study is needed to determine the number and/or combinations of bones for which greater accuracy is achieved by mean regression, since complete skeletons are infrequently found. In a recent revision of Kerley’s method, Kerley and Ubelaker [6] reported finding greater accuracy for fibular fragments. Other bones, in order of decreasing accuracy, were femoral osteon fragments, fibular intact osteons,
188
tibial o&eon fragments, femoral intact o&eons, and tibial intact osteons. For our total sample, greatest accuracy of prediction among these parameters was for femoral osteon fragments, followed by femoral intact osteons, fibular osteon fragments, tibial intact osteons, fibular intact osteons, and tibial o&on fragments. Table 2 compares our findings with relative accuracies reported by Kerley and Ubelaker [6]. In agreement with the findings of Bouvier and Ubelaker [4] , the Ahlqvist and Damsten modified method was found to be the most accurate for samples of an older age range. Unlike the findings of Bouvier and Ubelaker [4], however, Kerley’s fibular age-predicting formulas were of greater accuracy for an older age range, rather than being biased towards the younger ages. Greatest accuracy for Kerley’s femoral predicting formulas is found for our entire sample and for a younger subsample. Poor age prediction by Kerley’s femoral formulas for the 60 - 102 age subsample may be due to the fact that of the three samples used in our study these two are most similar to Kerley’s original population sample. It should be pointed out that since residuals are apparently not normally distributed (see Figs. 1 - 3), especially for older ages, third-order predicting equations can introduce error. Kerley’s [6, 81 revised formulas that are third-order equations proved to be of greater accuracy for the original sample, but may reduce accuracy when applied to different samples. The population samples from which parameters for some predicting formulas were derived may be more or less similar to a given sample than others. Several additional factors might influence results among populations and investigators. Reliability and accuracy require the precise location of standard fields. Kerley provides a detailed description of his standard microscopic locations only for the femur. Precise field location is difficult to determine in the tibia and fibula due to their asymmetric shape. Additional problems in field location arise because the gross morphology of fibular midshafts is highly variable. Ongoing research in our laboratory suggests that considerable field-to-field histomorphometric variation also exists in the tibia. Field-to-field variation is also considerable in those bones that are undergoing cortical drift. A general problem in the use of the various available histological aging methods relates to precise definition of parameters. For example, Kerley [l] defines an intact osteon as one that is easily distinguishable over 80% of its area, and has a complete Haversian canal. For Ortner [9], an intact osteon is one in which the Haversian canal is complete. Wu et al. [7] define intact o&eons as “completely unremodeled osteons”. Stout and Teitelbaum [lo] define an intact osteon as one in which at least 90% of the perimeter of the Haversian canal exhibits no evidence of remodeling. In the present study, definitions for each method were adhered to strictly. This study originally intended to include Singh and Gunberg’s [3] method. Our error in age estimation by this method, however, ranged from 12 to 49 years. This high error may have resulted in part from several factors. Our bone sections were not decalcified and were approximately 100 pm
189
thick, whereas Singh and Gunberg largely employed 10 pm thick, stained decalcified sections. Quantitative histology has been shown to be influenced by section thickness [ 111. The Singh and Gunberg method involves reading only two randomly chosen microscopic fields from the periosteal third of the anterior midshaft of the femur and tibia. Since considerable topographical variation in histomorphology occurs in cortical bone, considerable error can result from inadequate sampling of the histomorphology of a bone. It was therefore decided not to include this method. It should be pointed out that this study is a relatively rigorous test of histological aging methods. Bone specimens from dissecting rooms are poor representatives of real populations, and although cause of death was usually known, this often does not accurately reflect the health of individuals. It should also be noted that none of the original studies included individuals as old as our lO%-year-old. Ahlqvist and Damsten’s method, since it is based upon percentage of Haversian bone, cannot estimate an age above 97.9 and 94.1 years for their equations 1 and 2, respectively. Kerley’s femoral o&eon fragment predictory equation was also found to be unusable for our 102year-old individual. Since individuals of this age are rarely encountered, these problems are of minor concern. It is concluded from this comparative study that, of the age-associated histological parameters of bone examined, the average of age estimates produced from Kerley’s predicting formulas (mean regression) is the overall most accurate and reliable for aging unknown samples. Summary
and conclusions
Histological aging methods of Kerley and of Ahlqvist and Damsten were applied to bone samples from thirteen individuals of known age at death. Relative accuracy and reliability were determined for six of Kerley’s predicting formulas for the femur, fibula and tibia, and his profile method, Ahlqvist and Damsten’s femoral predicting formulas, and age determined by averaging ages predicted by Kerley’s six formulas. Averaging age estimates by Kerley’s six formulas (mean regression) was found to produce the overall greatest accuracy and reliability. Dividing the sample into two age groups (13 - 51 and 60 - 102 years) altered the results only slightly. Kerley’s femoral intact osteon formula produced the greatest accuracy for individuals in the younger age category, while his fibular osteon fragment formula was most accurate for older ages. Mean regression produced the greatest reliability for all age classes. Based upon both accuracy and reliability, averaging age predictions by Kerley’s regression formulas appears to be the method of choice for broad application of histological aging. Acknowledgements Bone specimens used in this study were made available through the courtesy of Dr. Roy Peterson, Department of Anatomy and Neurobiology,
190
Washington University School of Medicine, and Dr. William R. Goodge, Department of Anatomy, University of Missouri-Columbia Medical School. This research was funded in part by the Wenner-Gren Foundation grant number 3279, and by a University of Missouri Research Council grant.
References E. R.
Kerley,
The microscopic determination of age in human bone. Am. J. Phys. 23 (1965) 149 - 164. J. Ahlqvist and 0. Damsten, A modification of Kerley’s method for the microscopic determination of age in human bone. J. Forensic Sci., 14 (1969) 205 - 213. I. J. Singh and D. L. Gunberg, Estimation of age at death in human males from quantitative histology of bone fragments. Am. J. Phys. Anthropol., 33 (1970) 373 - 382. M. Bouvier and D. H. Ubelaker, A comparison of two methods for the microscopic 46 (1977) 391 - 394. determination of age at death. Am. J. Phys. Anthropol., W. A. Merz and R. K. Schenk, Quantitative structural analysis of human cancellous bone. Acta Anat., 75 (1970) 54 - 66. E. R. Kerley and D. H. Ubelaker, Revision in the microscopic method of estimating age at death in human cortical bone. Am. J. Phys. Anthropol., 49 (1978) 545 - 546. K. Wu, K. E. Schubeck, H. M. Frost and A. Villanueva, Haversian bone formation rates determined by a new method in a mastodon, and in human diabetes mellitus and osteoporosis. Calcif. Tissue Res., 6 (1970) 204 - 219. D. H. Ubelaker, Problems of the microscopic determination of age at death. Paper presented at Annual Meeting of the American Academy of Forensic Science, San Diego, 1977. D. J. Ortner, The effects of aging and disease on the micromorphology of human compact bone. Ph.D. Dissertation, University of Kansas, Lawrence, Kansas, 1970. S. D. Stout and S. L. Teitelbaum, Histomorphometric determination of formation rates of archaeological bone. Calcif. Tissue Res., 21 (1976) 163 - 169. H. M. Frost, Microscopy: Depth of focus, optical sectioning and integrating eyepiece measurement. Henry Ford Hosp. Med. Bull., 10 (1962) 267 - 285.
Anthropol.,
8
9 10
11
Forensic Science International, 15 (1980) 181 - 190 @ Elsevier Sequoia S.A., Lausanne - Printed in the Netherlands
THE RELATIVE ACCURACY AGING METHODS
AND
RELIABILITY
OF HISTOLOGICAL
SAMUEL D. STOUT and SARAH J. GEHLERT Department
of Anthropology,
(Received October 1,1979;
University
of Missouri,
Columbia,
Missouri 65211
(U.S.A.)
accepted December 10,1979)
Introduction Several histological methods for estimating the age of skeletal remains are available [ 1 - 31. These methods are especially useful in physical anthropology and forensic medicine, since they can provide greater accuracy of age estimation for adult skeletal remains than do traditional gross morphological methods. Histological aging methods utilize various combinations of histomorphological features such as intact osteons, fragmentary o&eons, nonHaversian bone, and primary vascular canals. They vary with respect to the number of fields read, the field size, and anatomical location. Predicting formulas are based upon specific sample populations for which age and health status are known. A requisite to the broad application of histological aging methods, especially to such diverse samples as those from archeological populations, is to test the accuracy of each method on population samples other than those from which the parameters were derived. This is also the only way in which the relative accuracy of various methods can be determined. An earlier study [4] compared only Kerley’s femoral intact osteon predicting formula to that of Ahlqvist and Damsten [2] . The present study compares the reliabilities and accuracies among all Haversian age-predicting parameters developed by Kerley [ 11 and Ahlqvist and Damsten [2] .
Materials
and methods
Midshaft sections were taken from the femurs, tibias, and fibulas of each of 13 individuals, obtained from dissecting room collections. Sex, age, and cause of death (except one) were available for all individuals (see Table 1). None of the samples included in the study were reported to have suffered from primary bone disease. All midshaft sections were embedded in epoxide, cut on a Buehler Isomet saw, and ground to a thickness of approximately 100 pm. Sections were mounted, and then read according to the methods of Kerley [l] and Ahlqvist and Damsten [ 21 . Only Haversian bone parameters were used.
182 TABLE
1
Sex, age. and cause of death of present Sex
Age
sample
Cause of death
(years) F M F F M M M M F F M F M
13 33 39 40 46 51 60 67 72 81 82 90 102
Brain stem tumor Cardiac arrest Hemorrhage Gunshot wound Electrocution Carbon monoxide poisoning Cardiopulmonary arrest Myocardial infarction Carcinoma of breast Cerebral hemorrhage Unknown Arteriosclerotic cardiovascular Pulmonary embolus
disc
For each specimen microscopic fields were located and read independently by two observers. For Kerley’s method, an ausJena scope combined with matched eyepieces fitted with a Merz [5] counting reticule was used. Our grid size measured with a stage micrometer is 0.970 mm2. Using Kerley’s reported field size of 2.06 mm2 [6] a correction factor of 2.1 was calculated, and all counts of intact and fragmentary osteons were multiplied by this factor. Ahlqvist and Damsten’s [2] method was carried out using a Nikon L-Ke scope fitted with a lOOsquare grid. Since our grid measured 1 mm2, as did Ahlqvist and Damsten’s, a field-size correction factor was not necessary. Ahlqvist and Damsten’s original predicting formula [2] and the revised prediction formulas for Kerley’s methods [6] were used in this study. It has been reported [4] that the accuracy of particular histological aging methods is not the same for samples with different age ranges. We therefore performed our analysis on subsamples with age ranges of 13 - 51 and 60 - 102 years, as well as on the sample as a whole. The relative accuracies of the various histological aging methods were determined by comparing the average standard deviations of the differences in years (mean differences) between predicted and known age using each of the bone sampling sites (femur, tibia, fibula) and each parameter (intact osteon count, osteon fragment count and percentage Haversian bone) where applicable. Mean differences between ages predicted by the observers were used as measures of interobserver error, and served as measures of reliability among the methods. The method of choice should be both reliable and accurate. The statistical significance of interobserver differences for each parameter and between predicted and known ages for each parameter was tested by a t-test.
183
Kerley and Ubelaker
Ahlqvist and Damsten
Fig. 1. Histogram for the total sample (13 - lo&year age range) showing: (1) the differences between predicted and known ages, and (2) the differences between the age estimates of the two observers, for each of Kerley’s predicting formulas and profile method, as well as for the modified method of Ahlqvist and Damsten. Results for age estimates based upon averaging ages from Kerley’s predicting formulas are included.
Results No statistical significance was found between observer estimates at the p = 0.05 level of confidence. Differences between each observer estimate and known age were also not significant at the p = 0.05 level. Figures 1 - 3 present the mean differences (average standard deviations) between known and predicted ages, and between observers.
Ftbulat fragments
Fibular intacfs
Ttblal fragments
Tibia1 l”MCtS
Femoral fragments
Femoral mtactr
ProfIle
P
w
I
0)
I
m
I
I
(D
I
I I
0‘
Mean difference m
I
2
(standard
I
E;
deviation
I
w‘
I
p’
of differences)
I
;;
in years z
I
I
y‘ I
6 0”
1
Ci
I
E rp
185
Kerley and Ubelaker
Ahlqvist and Damsten
Fig. 3. Histogram for the 60 - 102-year age range showing: (1) the differences between predicted and known ages, and (2) the differences between the age estimates of the two observers, for each of Kerley’s predicting formulas and profile method, as well as for the modified method of Ahlqvist and Damsten. Results for age estimates based upon averaging ages from Kerley’s formulas are included.
For the total sample, averaging the predicted ages by each of Kerley’s regression formulas (mean regression) produced the greatest accuracy, and reliability, followed by femoral osteon fragments. Accuracy for the remaining methods decreased in the following order: femoral intact osteons, fibular osteon fragments, Kerley’s profile method, tibial intact osteons, Ahlqvist and Damsten’s method, fibular intact osteons, and tibial osteon fragments (see Table 2). Reliability of the remaining methods, in decreasing order, is as follows: femoral intact osteons, Kerley’s profile method, tibial intact osteons, Ahlqvist and Damsten’s method, tibial osteon fragments, fibular osteon fragments, and fibular intact osteons (see Fig. 1).
186 TABLE
2
Relative accuracies for histological aging methods (Oi = intact osteons; Or = osteon fragments; A & D = modified method of Ahlqvist and Damsten; profile = Kerley’s profile method; mean regression = age arrived at by averaging the predicted ages by each of Kerley’s regression formulas) Present Total
sample range (13
- 102 years)
Mean regression Femur Or Femur Oi Fibula Or Profile Tibia Oi A&D Fibula Or Tibia Of *Relative
accuracies
13 - 51-year
age range
Femur Of Femur Or Mean regression Profile Fibula Of Tibia Oi A & D modified method Fibula Oi Tibia Or based upon
60 - 102-year range
age
Fibula Of Femur Or A&D Mean regression Tibia Oi Profile Fibula Ot
Kerley and Ubelaker*+t
Fibula Of Femur Of Fibula Of Tibia Or Femur Oi Tibia Oi
Femur Oi Tibia Of
mean-square residual values of Kerley and Ubelaker and profile methods unavailable.
[ 61.
‘Comparative data for mean regression
When only the 13 - 51-year age range is considered, femoral intact osteons and fragmented osteons for the femur are the most accurate, followed by mean regression, Kerley’s profile method, fibular osteon fragments, tibial intact osteons, Ahlqvist and Damsten’s method, fibular intact osteons, and tibial osteon fragments (see Table 2). Reliability in decreasing order is: mean regression, Kerley’s profile method, Ahlqvist and Damsten’s method, femoral intact osteons, femoral osteon fragments, tibial osteon fragments, tibial intact o&eons, fibular o&eon fragments, and fibular intact osteons (see Fig. 2). Accuracy for the 60 - 102-year age range was greatest for fibular osteon fragments and femoral osteon fragments, followed by Ahlqvist and Damsten’s method, mean regression, tibial intact osteons, Kerley’s profile method, fibular intact osteons, femoral intact o&eons, and tibial osteon fragments (see Table 2). Reliability decreased as follows: mean regression, femoral osteon fragments, femoral intact o&eons, fibular intact osteons, tibial intact o&eons, Ahlqvist and Damsten’s method, Kerley’s profile method, tibial osteon fragments, and fibular osteon fragments (see Fig. 3).
Discussion This paper evaluates the relative accuracy and reliability of several histological parameters of bone currently used to predict age at death [ 1, 21. An earlier comparison of these methods by Bouvier and Ubelaker [4] con-
187
eluded that Kerley’s method was the more accurate of the two. In their study, however, only the regression formulas for femoral intact osteons and percentage Haversian bone were used. Also, since the Bouvier and Ubelaker study used femoral sections from Kerley’s original sample, it might be argued that greater accuracy by the Kerley method would be expected, since predicting formulas are more accurate for the samples from which the parameters were derived. When methods which use more than one bone to predict age are used, errors from spatial variance and/or incoherence are reduced by increasing the amount of bone read [ 71. The use of Kerley’s profile method is preferable, therefore, to any “single bone” regression formula for predicting age. When using the profile method, however, problems arise from non-overlapping of the age-range estimates for each of the bone sampling sites [ 1, 81. If estimates from different bones do not overlap, a single estimate cannot be made. In the present study, averaging ages determined by each of Kerley’s six regression formulas (mean regression) produced the most accurate and reliable age estimates for our sample as a whole. Using the mean of ages estimated by Kerley’s six formulas also avoids the problem of non-overlapping predicted age ranges encountered with Kerley’s profile method. In the 13 - 51-year age subsample, mean regression and Kerley’s femoral intact osteon and osteon fragment equations were the most accurate predictors of age. Of these three, femoral osteon fragments were the most accurate, but mean regression had the least interobserver error. However, the mean differences from known age for the three methods were quite similar, + 7.38, + 6.28 and + 6.06, respectively, and interobserver error for mean regression was half that found for femoral intact osteons or osteon fragments. For skeletal samples that fall within this age range, mean regression would, therefore, be the method of choice. In the 60 - 102-year age range, estimates based upon femoral osteon fragments were most accurate. Fibular osteon fragments were of comparable accuracy, but considerable inter-observer error occurred with their use. Ahlqvist and Damsten’s method was slightly less accurate than Kerley’s fibular osteon fragment formula, but since it predicted age with a mean difference of 10 years, and exhibited much less interobserver error, it is preferable to fibular osteon fragments. In this older age range, mean regression improved reliability, but its accuracy was substantially less than the three methods noted above. When the entire age range is considered, mean regression is probably the best method with which to age unknown skeletal remains. Further study is needed to determine the number and/or combinations of bones for which greater accuracy is achieved by mean regression, since complete skeletons are infrequently found. In a recent revision of Kerley’s method, Kerley and Ubelaker [6] reported finding greater accuracy for fibular fragments. Other bones, in order of decreasing accuracy, were femoral osteon fragments, fibular intact osteons,
188
tibial o&eon fragments, femoral intact o&eons, and tibial intact osteons. For our total sample, greatest accuracy of prediction among these parameters was for femoral osteon fragments, followed by femoral intact osteons, fibular osteon fragments, tibial intact osteons, fibular intact osteons, and tibial o&on fragments. Table 2 compares our findings with relative accuracies reported by Kerley and Ubelaker [6]. In agreement with the findings of Bouvier and Ubelaker [4] , the Ahlqvist and Damsten modified method was found to be the most accurate for samples of an older age range. Unlike the findings of Bouvier and Ubelaker [4], however, Kerley’s fibular age-predicting formulas were of greater accuracy for an older age range, rather than being biased towards the younger ages. Greatest accuracy for Kerley’s femoral predicting formulas is found for our entire sample and for a younger subsample. Poor age prediction by Kerley’s femoral formulas for the 60 - 102 age subsample may be due to the fact that of the three samples used in our study these two are most similar to Kerley’s original population sample. It should be pointed out that since residuals are apparently not normally distributed (see Figs. 1 - 3), especially for older ages, third-order predicting equations can introduce error. Kerley’s [6, 81 revised formulas that are third-order equations proved to be of greater accuracy for the original sample, but may reduce accuracy when applied to different samples. The population samples from which parameters for some predicting formulas were derived may be more or less similar to a given sample than others. Several additional factors might influence results among populations and investigators. Reliability and accuracy require the precise location of standard fields. Kerley provides a detailed description of his standard microscopic locations only for the femur. Precise field location is difficult to determine in the tibia and fibula due to their asymmetric shape. Additional problems in field location arise because the gross morphology of fibular midshafts is highly variable. Ongoing research in our laboratory suggests that considerable field-to-field histomorphometric variation also exists in the tibia. Field-to-field variation is also considerable in those bones that are undergoing cortical drift. A general problem in the use of the various available histological aging methods relates to precise definition of parameters. For example, Kerley [l] defines an intact osteon as one that is easily distinguishable over 80% of its area, and has a complete Haversian canal. For Ortner [9], an intact osteon is one in which the Haversian canal is complete. Wu et al. [7] define intact o&eons as “completely unremodeled osteons”. Stout and Teitelbaum [lo] define an intact osteon as one in which at least 90% of the perimeter of the Haversian canal exhibits no evidence of remodeling. In the present study, definitions for each method were adhered to strictly. This study originally intended to include Singh and Gunberg’s [3] method. Our error in age estimation by this method, however, ranged from 12 to 49 years. This high error may have resulted in part from several factors. Our bone sections were not decalcified and were approximately 100 pm
189
thick, whereas Singh and Gunberg largely employed 10 pm thick, stained decalcified sections. Quantitative histology has been shown to be influenced by section thickness [ 111. The Singh and Gunberg method involves reading only two randomly chosen microscopic fields from the periosteal third of the anterior midshaft of the femur and tibia. Since considerable topographical variation in histomorphology occurs in cortical bone, considerable error can result from inadequate sampling of the histomorphology of a bone. It was therefore decided not to include this method. It should be pointed out that this study is a relatively rigorous test of histological aging methods. Bone specimens from dissecting rooms are poor representatives of real populations, and although cause of death was usually known, this often does not accurately reflect the health of individuals. It should also be noted that none of the original studies included individuals as old as our lO%-year-old. Ahlqvist and Damsten’s method, since it is based upon percentage of Haversian bone, cannot estimate an age above 97.9 and 94.1 years for their equations 1 and 2, respectively. Kerley’s femoral o&eon fragment predictory equation was also found to be unusable for our 102year-old individual. Since individuals of this age are rarely encountered, these problems are of minor concern. It is concluded from this comparative study that, of the age-associated histological parameters of bone examined, the average of age estimates produced from Kerley’s predicting formulas (mean regression) is the overall most accurate and reliable for aging unknown samples. Summary
and conclusions
Histological aging methods of Kerley and of Ahlqvist and Damsten were applied to bone samples from thirteen individuals of known age at death. Relative accuracy and reliability were determined for six of Kerley’s predicting formulas for the femur, fibula and tibia, and his profile method, Ahlqvist and Damsten’s femoral predicting formulas, and age determined by averaging ages predicted by Kerley’s six formulas. Averaging age estimates by Kerley’s six formulas (mean regression) was found to produce the overall greatest accuracy and reliability. Dividing the sample into two age groups (13 - 51 and 60 - 102 years) altered the results only slightly. Kerley’s femoral intact osteon formula produced the greatest accuracy for individuals in the younger age category, while his fibular osteon fragment formula was most accurate for older ages. Mean regression produced the greatest reliability for all age classes. Based upon both accuracy and reliability, averaging age predictions by Kerley’s regression formulas appears to be the method of choice for broad application of histological aging. Acknowledgements Bone specimens used in this study were made available through the courtesy of Dr. Roy Peterson, Department of Anatomy and Neurobiology,
190
Washington University School of Medicine, and Dr. William R. Goodge, Department of Anatomy, University of Missouri-Columbia Medical School. This research was funded in part by the Wenner-Gren Foundation grant number 3279, and by a University of Missouri Research Council grant.
References E. R.
Kerley,
The microscopic determination of age in human bone. Am. J. Phys. 23 (1965) 149 - 164. J. Ahlqvist and 0. Damsten, A modification of Kerley’s method for the microscopic determination of age in human bone. J. Forensic Sci., 14 (1969) 205 - 213. I. J. Singh and D. L. Gunberg, Estimation of age at death in human males from quantitative histology of bone fragments. Am. J. Phys. Anthropol., 33 (1970) 373 - 382. M. Bouvier and D. H. Ubelaker, A comparison of two methods for the microscopic 46 (1977) 391 - 394. determination of age at death. Am. J. Phys. Anthropol., W. A. Merz and R. K. Schenk, Quantitative structural analysis of human cancellous bone. Acta Anat., 75 (1970) 54 - 66. E. R. Kerley and D. H. Ubelaker, Revision in the microscopic method of estimating age at death in human cortical bone. Am. J. Phys. Anthropol., 49 (1978) 545 - 546. K. Wu, K. E. Schubeck, H. M. Frost and A. Villanueva, Haversian bone formation rates determined by a new method in a mastodon, and in human diabetes mellitus and osteoporosis. Calcif. Tissue Res., 6 (1970) 204 - 219. D. H. Ubelaker, Problems of the microscopic determination of age at death. Paper presented at Annual Meeting of the American Academy of Forensic Science, San Diego, 1977. D. J. Ortner, The effects of aging and disease on the micromorphology of human compact bone. Ph.D. Dissertation, University of Kansas, Lawrence, Kansas, 1970. S. D. Stout and S. L. Teitelbaum, Histomorphometric determination of formation rates of archaeological bone. Calcif. Tissue Res., 21 (1976) 163 - 169. H. M. Frost, Microscopy: Depth of focus, optical sectioning and integrating eyepiece measurement. Henry Ford Hosp. Med. Bull., 10 (1962) 267 - 285.
Anthropol.,
8
9 10
11