Transcript

Tooth Cementum Annulation for Age Estimation: ResultsFrom a Large Known-Age Validation StudyUrsula Wittwer-Backofen,1,2* Jutta Gampe,1 and James W. Vaupel1

1Max Planck Institute for Demographic Research, Rostock, Germany2Department of Human Genetics and Anthropology, University of Freiburg, Freiburg, Germany

ABSTRACT Recent research indicates that tooth-ce-mentum annulations (TCA) may be used more reliablythan other morphological or histological traits of the adultskeleton to estimate age. Until now, however, confidenceintervals for age estimated by this method have not beenavailable for paleodemographic and forensic applications.The present study addresses this problem. Based on alarge known-age sample, age estimates by TCA were con-ducted in a blind study involving 363 teeth. Tooth-rootcross sections were made using a refined preparation tech-nique. Improved digital graphic procedures and enhance-ment strategies were used to produce digital images witha specially adapted software package. This resulted in

high concordance between the TCA age estimates andchronological age. Assessment of the method’s accuracy,as expressed by 95% confidence intervals, showed thaterror bounds for age estimates do not exceed 2.5 years. Sexdifferences, intraindividual correlations, and the effects ofperiodontal disease were studied. None of these indicatorshad a quantitative effect on the number of TCA bandswhen the proposed methodological standard was followed.We conclude that the TCA technique is a reliable methodfor estimating a subject’s age from cementum annulations.Am J Phys Anthropol 123:119–129, 2004.© 2004 Wiley-Liss, Inc.

The reconstruction of mortality patterns in pastpopulations is necessary for paleodemographic anal-yses. The reliability of mortality reconstruction de-pends on individual sex and age estimates of theskeleton as a biological source of information. Fordecades, osteologists and paleodemographers havestrived to improve methods for determining age andsex. The importance of this research was recentlyhighlighted in a new approach to paleodemographydescribed by Hoppa and Vaupel (2002).

Almost all established macroscopic methods forage estimation in the skeleton are problematic(Buikstra and Ubelaker, 1994; Jackes, 2001). This isbecause only changes in biological age can be ob-served in skeletons. High interindividual variabilityresults in error margins that may reach 7 years, atbest, for ages after skeletal growth is complete(Buikstra and Ubelaker, 1994; Jackes, 2001;Kemkes-Grottenthaler, 2002). The problem intensi-fies at older ages, as individual variability of age-dependent changes in the skeleton increases. Thus,methodological problems increase with the age ofthe person.

It is clear, then, that an age-estimation method isneeded that is less sensitive to continuous and non-quantified age-dependent changes in the skeleton.

An alternative method, based on counting the in-cremental lines seen in tooth-root cementum, hasshown promise. We hypothesize that these incre-mental lines in the tooth cementum can be used as amore reliable age marker than other morphologicalor histological traits in the human skeleton. Thishypothesis is based on the biological factors of the

tooth-cementum annulations (TCA) formationknown so far.

Cementum is the calcified tissue that surroundsthe dentine and forms the attachment site for theperiodontal fibers that link the tooth to the alveolarbone. In cementum formation, hypermineralizedlayers of extracellular matrix alternate with lessmineralized layers. The first layer of acellular ce-mentum is produced before the tooth erupts, andfurther layers are added during and after eruption.Cementum layers consist primarily of uncalcifieddense bundles of collagen fibrils. These bundles laterbecome mineralized by hydroxyapatite crystals,whose changing orientations may be responsible forthe optical effect of alternating dark and translucentlayers. A biological explanation for these alternatinglayers was given by Lieberman (1994) and Schroder(2000). They suggested that the dark lines are stopphases of mineralization during continuing growthof the fibroblasts, leading to a change in mineralcrystal orientation. This pattern is visible under the

Grant sponsor: Max Planck Institute for Demographic Research.

*Correspondence to: Prof. Dr. Ursula Wittwer-Backofen, Depart-ment of Human Genetics and Anthropology, University Clinics ofFreiburg, Breisacher Str. 33, D-79106 Freiburg, Germany.E-mail: [email protected]

Received 8 May 2002; accepted 24 February 2003.

DOI 10.1002/ajpa.10303Published online 9 June 2003 in Wiley InterScience (www.

interscience.wiley.com).

AMERICAN JOURNAL OF PHYSICAL ANTHROPOLOGY 123:119–129 (2004)

© 2004 WILEY-LISS, INC.

microscope as a series of alternating light and darklines or bands.

Along the axis of the tooth root there are two zonesof different cementum types: the acellular cemen-tum, that grows close to the cervical part of the root,and the cellular cementum, which mainly covers theapical part of the tooth root. In the present study, wefocus on the acellular cementum, predominantlyseen in the middle third of the root. It was shownthat each pair of lines corresponds to 1 year of lifeand constitutes a biological record that can be usedto estimate the age of an individual (e.g., Lieberman,1994; Kagerer and Grupe, 2001). By adding the for-mation age of the tooth root to the number of TCA,the age at death or tooth extraction can be esti-mated.

Variations in cementogenesis that change the ap-pearance of lines may be induced by different fac-tors, including, for example, biomechanical forces,nutrition, hormonal cycle, or ecological conditionssuch as temperature, ultraviolet light, humidity, al-titude, or pollution (Lieberman, 1994; Halberg et al.,1983; Kagerer and Grupe, 2001).

The appearance of cementum lines, observed inmore than 50 different mammal species all over theworld, has been said to reflect the natural metabolicrhythm of seasonal changes (Laws, 1952; Geiger,1993; Grue and Jensen, 1979; Kay et al., 1984).Seasonal rhythms in cementum annulations, as ob-served in the alternating dark and light bands, canbe explained by the metabolism of the parat hor-mone, which is responsible for the regulation of thecalcium blood level, interacting with vitamin Dwhich regulates the resorption of calcium. Thus,both hormones and vitamins may interact to pro-duce a circannual rhythm by a complex mechanismof environmental and physiochemical “synchroniz-ers” (Halberg et al., 1983). Many questions remainregarding the mechanisms of tooth cementum annu-lation and its influencing factors, particularly con-cerning the interpretation of seasonal increments(Stott et al., 1980; Lieberman, 1994; Cipriano, 2001).

The first use of cementum in human age estima-tion began with measurements of width of the totalcementum layer, rather than with counts of incre-mental lines (Gustafson, 1950). In the early 1980s,the study of three human teeth showed that the TCAmethod could be applied to human teeth as it hadbeen to other mammals (Stott et al., 1982). Furthertechnical improvements (Naylor et al., 1985) led tothe suggestion that TCA is superior to other tooth-based methods of age estimation in the adult skele-ton (Gustafson, 1950, 1955; Azaz et al., 1974; Phil-ipsen and Jablonski, 1992).

Initially, the TCA method was applied to freshlyextracted teeth, but Großkopf (1989, 1990) showedthat the method was also applicable to historicalskeletons and cremations. This was confirmed byothers (e.g., Charles et al., 1986; Condon et al., 1986;Lipsinic et al., 1986; Geuser et al., 1999) and ex-tended to forensic cases (Jankauskas et al., 2001).

These findings add further support to the idea thatthe number of incremental lines is a stable property,even under circumstances where other characteris-tics of the lines (e.g., width, degree of mineraliza-tion) have been altered by environmental or physi-ological perturbations (e.g., Karger and Grupe,2001). It was on the basis of these kinds of resultsthat the TCA method was recently recommended asa reliable technique for age estimation in adultsusing skeletal materials (for an overview of TCAapplications in humans, see Wittwer-Backofen andBuba, 2002). In subadults, however, the use of mi-crostructural analysis of enamel and dentine leadsto even greater accuracy, often within days of thetrue chronological age (Antoine et al., 2000).

However, some problems remain regarding thefull application of the TCA method. For example, thesmall samples used in previous studies limited theestablishment of statistical parameters needed forpractical paleodemographic and forensic applica-tions. In addition, the question of whether dentaldisease, particularly periodontal disease, has an im-pact on TCA is still open. Großkopf et al. (1996), forexample, found no impact of periodontal disease onTCA, whereas Kagerer and Grupe (2001) reportedthat the latter pathology reduced or arrested annualcementum formation.

In the present study, we assess the TCA methodin a sufficiently large, well-characterized tooth sam-ple. Our purpose is to outline perspectives and lim-itations of the TCA method by calculating confidenceintervals, intraindividual correlations, and the im-pact of periodontal disease and tooth type on thenumber of TCA incremental lines.

We also focus on the graphic enhancement of in-cremental lines as a basis for reproducible results.

DATA AND METHODS

The sample

Our sample consists of 433 freshly extracted per-manent teeth collected from several dentists anddental clinics in Germany. This sample is differentfrom the one used in a previous study (Wittwer-Backofen and Buba, 2002): that previous samplewas used as a training sample for our observers.Maxillary and mandibular incisors, canines, andpremolars are included in the study. In all cases,tooth extractions were performed as part of essentialclinical care. In addition to the extraction date of thetooth and reason for extraction, the records containthe patient’s date of birth, sex, and ethnicity. Among433 teeth, there are 70 teeth from 63 individuals(16.2%) that could not be counted, either because ofthe poor quality of the tissue following preparationor because significant irregularities in the cemen-tum incremental lines were present. These teethwere removed from the sample in an initial phase.

Among the excluded teeth was a significantlygreater number of more maxillary than mandibularteeth, which were eliminated from the sample under

120 U. WITTWER-BACKOFEN ET AL.

consideration because of irregular histology. Inthese cases, a wave-like course of cementum lineswas observed. The cementum lines seemed to besuperimposed on each other. Neighboring areasformed broad cementum bands with no regularlines. This rendered counting impossible. Theseteeth were removed from the sample after micro-scopic image scanning. Second premolars were thepredominant tooth type to be excluded for this rea-son (Table 1). Teeth were also excluded if the con-trast of cementum lines was too low. The suitabilityof teeth for counting was based on judgments ofimage quality: 16.2% of the teeth in our sampleprovided unsuitable images for TCA analysis, and363 countable teeth remained in the sample andwere included in the statistical procedures.

Teeth from both men and women were included inthe sample, but there are more teeth from men thanfrom women (Table 1). Patient’s ages ranged from12–96 years, with approximately 85% of the teethbeing from individuals older than 35 years. The agedistribution for men and women was quite similarup to the median, but the upper 50% of the data aredispersed more widely for women (Fig. 1).

The reasons for tooth extraction differed system-atically with age, as expected. There were only slightdifferences in reasons for tooth extraction betweenthe sexes.

Teeth were categorized according to their reasonfor extraction. The five categories were: 1) dentalcaries, 2) periodontal disease, 3) orthodontic care, 4)odonto-prosthetics, and 5) multiple pathologies. In

practice, most teeth suffered from periodontal dis-ease, dental caries, or both. Orthodontic therapywas a reason for extraction only for juveniles. Theseteeth, as well as those extracted for odonto-pros-thetic reasons, can be regarded as the only healthyteeth in our sample. The presence of periodontaldisease, however, did not result in exclusion fromthe sample. This was because we were interested inthe possible impact of pathologies on TCA. This is animportant issue in the analysis of historic teeth,since many such samples come from individualswith significant dental disease, especially dentalcaries and periodontal pathologies.

Among the 363 teeth in the sample, 229 originatedfrom multiple extractions of 77 individuals. In mostcases, 2 teeth per individual were available, al-though as many as 9 teeth were extracted from asingle individual (Fig. 2). These multiple extractionsprovided the basis for establishing intraindividualcorrelations in TCA.

Preparation technique

Several tests indicated that the choice of opti-mized techniques for preparation, staining, micro-scope use, and counting have a significant impact on

TABLE 1. Number of teeth in analysis

Tooth by type (FDI code)Total analyzed

(n)Males

(n)Females

(n)

Excluded

n %

Maxilla11.21 (central incisor) 45 34 11 11 19.612.22 (lateral incisor) 54 39 15 8 12.913.23 (canine) 22 14 8 13 37.114.24 (first premolar) 22 11 11 6 21.415.25 (second premolar) 19 13 6 14 42.4Subtotal 162 111 51 52Mandibula31.41 (central incisor) 56 34 22 3 5.132.42 (lateral incisor) 62 28 34 3 4.633.43 (canine) 14 6 8 0 0.034.44 (first premolar) 43 29 14 0 0.035.45 (second premolar) 26 18 8 12 31.6Subtotal 201 115 86 18Total 363 226 137 70 16.2

Fig. 1. Age distribution in years by sex.

Fig. 2. Frequency distribution of multiple extractions per pa-tient.

TOOTH CEMENTUM ANNULATION 121

the results (Renz et al., 1997). Different methods ofpreparation and of image presentation for countingmay be partly responsible for the considerable vari-ation in the reliability of TCA for age estimationthat has been reported.

To estimate the intensity of periodontal regres-sion, fresh teeth were stained in 1% Fuchsin watersolution, and the maximum distance between thecemento-enamel junction and the stained soft-tissuemargin was measured with a microcalliper on foursurfaces of the tooth root (labial, lingual, mesial, anddistal). A mean value was calculated from thesemeasurements and used in further analysis as an“index of periodontal regression.” The greater thisindex value, the greater the degree of loss of peri-odontal attachment (i.e., more advanced periodontaldisease).

After measurements were made, 70–80-�m non-decalcified transverse sections were prepared fromthe middle third of the root, using a Leica 1600microtome with a diamond coated blade (for a de-tailed description of technical features, see Leica,2003). This was done after embedding the toothcrown into a block of epoxy resin, a technique de-signed to hold the tooth steady. Embedding the toothroot was not necessary for freshly extracted teeth,however. Next, the apical third of the root was re-moved, and at least three subsequent sections werecut at low speed. The unstained sections werecleaned in alcohol and mounted on slides for furtheranalysis.

Microscopic analysis

The sections were examined in bright-field trans-mitted light at 200–400� magnification, using aLeica DMRXA microscope. Images were scanned us-ing an 824 � 1,026 pixel resolution digital camera(Leica DC 250), and the image was viewed on alarge-scale monitor.

The most significant technical improvements,compared to previously used methods (Wittwer-Backofen and Buba 2002), were the use of real-timeimages and the digital image enhancement andcounting routines. Real-time images (6–7 images/sec) allowed the complete cementum band of eachforsection to be scanned, facilitating the search foran optimal focus and providing immediate qualitycontrol. At least three images per tooth were ac-quired and stored in a graphic database (Softwarepackage: Imagic 1000, Imagic Company Switzer-land, distributed by Leica).

Images were enhanced by contrast improvementand adjusted either through the grey-scale grada-tion or embossing procedures. Cutting artifactsproduced by the microtome were eliminated by asoftware macro routine via a fast Fourier trans-formation (Software package: Qwin, by Leica). Adetailed description of the imaging system can befound in Leica (2003).

The resulting images showed distinct alternatingdark-and-light annulation lines between the ce-

mento-dentin junction and the periodontal ligament(Figs. 3, 4). Counting the dark lines was done man-ually at the monitor, based on digitally enhancedimages. The dark lines were counted using the “mea-surement overlay.” This is a software tool that al-lows the researcher to mark each line by mouse-click, which is then summed, reducing the risk ofhuman error.

Age estimation

The counting results are based on up to four dif-ferent images of one tooth. All counting was con-ducted by one observer (U.W.-B.). Where there wasdiscrepancy between different counts, the mean

Fig. 3. Acellular tooth cementum of a lower canine, male, age87.8 years. TCA age estimation of 88 years resulted from 78incremental lines counted, added by mean eruption age of 10.0years. Topmost arrow indicates first cementum line, followed byan arrowhead each 10 lines toward outer tooth margin. Lastgrouped area consists of 8 lines.

Fig. 4. Acellular tooth cementum of an upper first premolar,male, age 15.0 years. Each cementum line is marked by anarrowhead, numbered by order of formation. Line number 6 isemerging, resulting in a TCA age estimation of 15.5–16.5 years,including a mean eruption age of 10.5 years.

122 U. WITTWER-BACKOFEN ET AL.

value was used for statistical analysis. To assess theimpact of tooth image quality on possible countingerrors, images were classified by the observer on athree-category quality scale (1 � poor quality, withlow contrast, lots of artifacts, and irregular wavepatterns of cementum lines; 2 � moderate quality,with typically visible contrast, scattered artifacts,and regular lines; 3 � best quality, with almost noartifacts, good contrast, and well-defined regularlines).

To obtain an estimate of a patient’s age at toothextraction, the cementum annulation counts wereadded to the tooth- and sex-specific mean eruptionage with its worldwide variability (Adler, 1967;Schumacher et al., 1990). It is not completely clearwhen exactly during tooth formation the cementumannulation process starts, especially in the midsec-tion of the root. Practical considerations led to theuse of mean eruption age to determine the age atformation of the first cementum ring. Tooth eruptiondata, based on large samples, are available for manypopulations worldwide and especially for Germany,where our sample stems from. Furthermore, an ad-ditive bias, which is similar for different tooth types,can be accounted for by the statistical calibrationprocedure.

The deviation between chronological age at ex-traction and TCA-estimated age is the sum of tworandom components: 1) individual variability intooth eruption age, and 2) measurement errors aris-ing from the counting procedure. These componentscan be assumed to act independently. Whether themeasurement error can be modelled via traditionalassumptions, and whether the error variance maybe assumed to be constant over the whole age range,were the concerns of the present study.

RESULTS

Figure 5 summarizes the results of the countingand age-estimation procedure, separately for malesand females. The top row shows a strong linearrelationship between the true and the estimatedages (correlation coefficient r � 0.970 for men, andr � 0.978 for women). There are a few scatteredoutliers, some of which deviate markedly from trueage. A deviation of more than 5 years was observedin 2.2% of the cases. Residual plots of estimated agesminus true ages are shown in the bottom row ofFigure 5. Observations for which estimated age islarger than actual age result in positive differences,whereas negative values indicate age-underestima-tion. The two broken lines give a corridor of �3 yearsas a rough assessment of the general quality ofresults. Neither systematic over- nor underestima-tion is evident over a wide range of ages. The onlyexception is for ages above 70, where a slight ten-dency of underestimation seems to be present forwomen. The extreme outliers are scattered over theage range, and do not cluster at certain ages.

Results for different tooth types are given in Fig-ure 6. Minor upward bias is apparent in both men

and women for the maxillary canines, and to a lesserextent for mandibular second premolars. However,no tooth position is prone to extraordinary inaccu-racies.

Figure 7 shows the distribution of age-estimationerrors as a function of image-quality index. In con-trast to what might have been expected, the reliabil-ity of results is not affected by image quality. In allthree index categories, the deviations center at zeroand are distributed symmetrically. The most ex-treme outliers are seen in category 2 (intermediateimage quality), and not in category 1 (images of thepoorest quality).

Figure 8 shows the deviations of estimated fromtrue ages for increasing periodontal attachmentloss. The data were divided into four groups of equalsize, split at the quartiles of the measured index ofperiodontal regression (see Preparation Technique).As the data show, and contrary to some studies (e.g.,Kagerer and Grupe, 2001), the degree of periodontaldecline did not affect the accuracy of TCA. Devia-tions from true age were no different for teeth show-ing minimum periodontal decline compared to thosewith maximum loss.

Multiple teeth extracted from the same individualallowed intraindividual variability of age-estimationresults to be analyzed. To investigate whether teethfrom one patient showed some common bias, Figure9 gives the deviations between true and estimatedage for the 77 patients with more than one toothavailable. For a majority of individuals (44 out of77), the estimated ages cluster around the actualage. In 17 cases, age is underestimated for all teeth

Fig. 5. Top row: Estimated age vs. true age at extraction formales (left) and females (right). Thin broken line indicates iden-tical values. Correlation is rm � 0.970 for males and rf � 0.978 forfemales (P � 0.075). Bottom row: Deviations between estimatedand actual ages are plotted over true ages. Positive values repre-sent overestimation; negative values stand for underestimation.Broken lines indicate an interval of �3 years.

TOOTH CEMENTUM ANNULATION 123

of a single patient, and in 16 cases the age of theindividual is overestimated for all teeth. Three ex-treme outliers stem from individuals with multipleextractions; however, for all these individuals show-ing such an outlier result, the other teeth gave re-sults within the 2s confidence band limits. There isno consistent high error for all teeth of a singleindividual.

Calibration of age-estimation method

To estimate the unknown chronological age fromTCA counts on future observations, a statistical cal-ibration method is needed. This method has to pro-vide both a formula for transforming a TCA countinto an age estimate, and then for assessing theprecision of this age estimate. If the exact number ofTCA lines for an individual as well as the exact ageat tooth eruption were known, then the correct agewould be the sum of these two numbers. In practice,however, these two unknowns have to be replaced byappropriate estimates.

The exact number of lines will be replaced by thecounted number of lines, while the unknown age ateruption will be substituted by the sex-specific andtooth type- specific mean age at eruption. Thus, de-viations between the true age at extraction (ordeath) and the estimated TCA age stem from two

Fig. 8. Deviations vs. age for four stages of periodontal de-cline. Cut-points for four groups are quartiles (Q25, Q50 � Me-dian, Q75) of periodontal index. Remote outliers are clipped.

Fig. 9. Deviations of TCA age from true age for those 77individuals from whom multiple extractions were available. In-dividuals are ranked such that those for whom age is consistentlyunderestimated are at left, and those for whom age is overesti-mated are at right of scale. Individuals with both positive andnegative deviations are randomly arranged in middle. Note thatvertical axis is clipped for better visibility.

Fig. 6. Differences be-tween true and estimatedage in years by tooth posi-tion, for males (E) and fe-males (�). Top row: Maxilla.Bottom row: Mandibula (toothtype after FDI code; see Ta-ble 1). Note that vertical axisis clipped for better visibility.

Fig. 7. Deviations by categorical quality index (1 � low, 2 �moderate, 3 � good quality; for explanations, see text). Box-and-whiskers plots show central half of data as boxes with median-line inserted. Whiskers extend in both directions to last observa-tion, which is within a distance of 1.5 times the interquartilerange from first or third quartile. Values beyond whiskers areplotted as single dashes, and indicate outliers.

124 U. WITTWER-BACKOFEN ET AL.

different and independent sources of variation. Onesource of variation is the measurement error of thecounting procedure; the other is the variation of theindividual eruption-ages around their mean value.Multiple extractions from several individuals allowfor an estimation of both variance components fromthe sample.

If yij denotes the estimated TCA-age based ontooth j from individual i in the sample, then yij � Cij� ��, where Cij is the counted number of TCA-lines,and �� is the appropriately matched mean age-at-eruption. The exploratory analysis of the sampledata above supports a linear relationship betweenthe estimated age yij and the actual age xi whichthen leads to the following model:

yij � � � � xi � di � eij.

Here � and � are fixed regression coefficients forwhich we would expect that the slope parameter � 1, if the counting procedure is accurate. Both di andeij are zero-mean random variables, where eij is themeasurement error of the counting technique, andall the eij are assumed to be independently distrib-uted as N(0,2). The di represent the interindividualvariability of the tooth eruption times, and are in-dependent between different individuals and inde-pendent from the measurement errors eij. Whereasobservations on two different individuals are uncor-related, age-estimates on the same patient share acommon component and are thus correlated. IfVar(di) � �2, then Var(yij) � �2 � 2 and Cov(yij,yik)� �2. The intraindividual correlation is thereforeCor(yij,yik) � �2/(�2 � 2). The random-effects di areassumed to be normally distributed.

Modelling the individual-specific characteristicsby a common random effect di takes into accountpotential correlations between observations. This issupported by observed covariation in eruption timesbetween different teeth, which was recently docu-mented by Parner et al. (2001).

Parameters in the above model were estimated byrestricted maximum likelihood (REML), as providedby S-Plus (Pinheiro and Bates, 2000). The samplewas not divided by sex because preliminary analysisrevealed no reason to treat males and females sep-arately. In the presence of severe outliers, one wouldusually perform a regression analysis by using arobust procedure such as least-median-of-squares oran appropriate M-estimator (Rousseeuw and Leroy,1987). These methods take into account that outliersmay have an undue influence on the estimation re-sults and implicitly weigh down their impact. Aslikelihood-based approaches for variance componentmodels are not easily amenable to robust proce-dures, those eight observations with the largest de-viation between actual and TCA age were excludedbefore applying the REML procedure (McCullochand Searle, 2001). The results from the estimatedmodel are given in Table 2, while the intraindividualresiduals, their normal-QQ-plot, and a normal-QQ-

plot for random effects are shown in Figure 10. Allplots confirm the validity of the model assumptions.

The estimated slope parameter � equals one al-most exactly, which together with the small stan-dard error of the estimate indicates that an increasein age by 1 year is translated into a 0.98-year in-crease in estimated TCA age. From these estimates,a total variance �2 � 2 � 1.3996 and a rather smallintraindividual correlation of �2/(�2 � 2) � 0.0788results.

For a new TCA-age y0, the unknown true age �1is estimated by inverting the regression equation,

x0 �y0 � �

�.To determine error bounds for this es-

timate, an inverse prediction interval for a givenlevel (1 � �) can be derived from the limits of theprediction interval for y0, based on the obtainedvariance estimates. Formally, this is achieved bysolving the following quadratic equation for x0:

y0 � � � � x0)2

Var(y0)� t2

where t is the (1 � �2)-quantile of the tn�2-distribu-tion. A detailed derivation of this equation can befound in Brown (1993). The interval width dependson the position of x0 relative to the age values in thecalibration sample. Table 3 presents the calibrationresults for selected values of y0 and a level of (1 � �) �100 � 95%. For all ages in the relevant range, theerror bounds, defined by the limits of this 95% con-fidence interval, cover values which deviate by nomore than 2.5 years from the central point estimatex0.

DISCUSSION

In the present study, the TCA method for ageestimation provided accurate results for a largesample. The sample size allowed us to conduct adetailed investigation of the impact of different in-dividual parameters such as sex, tooth position, andpresence of periodontal disease. The large samplesize also allowed for a thorough statistical analysisof the results.

The results of our study demonstrate that TCAanalysis may serve as a powerful method for post-mortem age estimation. The accuracy is reflected inthe error bounds obtained from the statistical anal-ysis which, based on a 95% confidence level, are no

TABLE 2. Mixed-effects model, estimating results

ValueStandard

error df t-value P-value

Fixed effects� 0.9493 0.2530 205 3.7520 0.0002� 0.9839 0.0043 148 227.2584 �0.0001

Value 95% confidence interval

Random effects� 0.3320 (0.0884, 1.2479) 1.1355 (0.9824, 1.3125)

TOOTH CEMENTUM ANNULATION 125

more than �2.5 years. Given the individual variabil-ity of ages at tooth eruption which are reported tohave (depending on tooth type and sex) a standarddeviation between 0.6–1.6 years (Adler, 1967; afterSchumacher et al., 1990), the remaining countingerror is small.

There are two explanations for the reduction ofcounting errors in this study. Firstly, only thoseimages suitable for counting were submitted to TCAanalysis. Secondly, and more important, the techni-cal equipment available for the present study wassignificantly better than that used in previous stud-ies. In a previous study (Wittwer-Backofen andBuba, 2002), the observers either had to count di-rectly while looking through the microscope (a tiringprocedure), or had to use simple photographs. Threefeatures were crucial to the improvements: 1) real-time image control that allows for selection of themost appropriate detail of the cementum band on

the computer screen; 2) the digital image enhance-ment procedures; and 3) a software module thatallows lines to be marked on the screen. In combi-nation, these features eliminated many of the draw-backs that former studies had to cope with.

The omission of teeth of which the images werejudged inappropriate for counting deserves closerconsideration. The original sample of 433 teeth in-cluded all maxillary and mandibular single-rootteeth. The teeth excluded were predominantly sec-ond premolars, which in fact were diagnosed as sin-gle-root teeth even when they showed a tendencytowards root bifurcations. In these cases, cementumlines followed a “wave pattern” where the cementumband partly surrounds artifacts or overlays itself inundulations. They are detectable at a first quickcheck of the microscopic images and must be ex-cluded from the sample, as they do not mirror reg-ular yearly annulation lines. They look like a mul-tiple number of interrupted line fragments, andmight account for the “doubling cases” reported inprevious studies. A second reason for omission wasbad image quality resulting from poor quality of thetissue after preparation. If (even with the advancedtechnology at hand) no satisfactory contrast of thelines can be adjusted, then omitting the specimen isthe only sensible thing to do. It may be argued that

Fig. 10. Top row: Intraindividual residuals and residual normal QQ-plot for linear calibration model. Results are displayedseparately for males and females; however, model was estimated jointly. Bottom row: Normal QQ-plot for random effects.

TABLE 3. Calibration results for selected TCA ages

y0 x0

95% inverse prediction interval

Lower limit Upper limit

40 39.69 37.25 42.1360 60.02 57.54 62.5080 80.34 77.83 82.86

126 U. WITTWER-BACKOFEN ET AL.

the high level of accuracy obtained in the presentstudy is a consequence of excluding 16.2% of theoriginal sample. This is true in certain respects.However, what our study shows is that if obviouslyinappropriate specimens are excluded, then TCAanalysis produces convincing results. We want tostress once again that teeth were excluded fromfurther processing solely because their images werenot of sufficient quality and not because of anycounting results: All images assessed to be count-able were actually counted and remained in thestatistical analysis.

We tested sex-specific subsamples to quantify thepossible influences due to different male and femalephysiological conditions. Former studies showedhigher error rates for teeth extracted from females.Precisely how cementum synthesis and mineraliza-tion are regulated to produce annual incrementallines is unknown, but almost certainly the cemen-tum pattern results from the action of specific hor-mones and growth factors. Women experiencegreater fluxes in steroid hormone levels (at men-arche, pregnancy, childbirth, and menopause) andcalcium homeostasis (at pregnancy) than men,which could conceivably alter cementum synthesisand mineralization. However, our data do not showsuch an impact, as TCA errors occur equally for bothsexes. The number of excluded cases correspondedexactly to the ratio of men’s and women’s teeth inour study. In addition, the reliability of TCA ageestimations in females does not depend on age. Thecorrelation coefficient between true age and esti-mated age is r � 0.982 (n � 49) for women below age50, and r � 0.966 (n � 87) for women above age 50,which is a minor difference (P � 0.06).

Therefore, our results suggest that the quantita-tive process of cementum annulations is not influ-enced by menopausal disorders of the metabolic sys-tem. However, as suggested by Kagerer and Grupe(2001), the quality (e.g., width, apparent degree ofmineralization) of the lines may be altered in womenby such events as pregnancy and childbirth. This isan important issue of incremental line quality, andis the focus of ongoing studies in our laboratory.

In previous studies, the reliability of TCA ageestimations in both sexes proved to be age-depen-dent. With higher age, inaccuracy increased for bothmales and females. This led to recommendations ofan age-limited applicability of the TCA method forage estimation (Solheim, 1990; Kvaal and Solheim,1995; Stein and Corcoran, 1994; Lipsinic et al.,1986). Usually this was interpreted as a metabolicdisorder of higher age with the influence of periodon-tal regression, dental caries, or other individualcharacters cumulating over age.

The present study does not support these obser-vations, since we found no age-dependent errorrates. This may be due to some of the following facts.By using the scanning procedure on real-time im-ages on the computer screen, we were able to detectthe most suitable areas for further analysis, some-

thing that might have been missed by traditionalocular microscopy. Carefully applied image en-hancement procedures may filter artifacts and im-prove the visibility of cementum lines, potentiallyresulting in less erroneous counting, as supported bymeasurement and counting software tools (Soft-ware: Imagic 1000 measurement module). This al-lows for counts with the aid of the computer monitorin contrast to microscopic ocular counts, which aremore tedious the more lines have to be counted. As aconsequence of eyestrain, reliability directly de-pends on the number of counted lines. Therefore, weconclude that previous inaccuracies might havebeen induced by less favorable technical conditions,and that problems of high-age (under)estimationscan be solved by using the technical methods de-scribed in this study.

One of the more contentious discussions associ-ated with TCA is related to the effect of periodontaldisease. As the cementum annulations are interre-lated with the Sharpey fibers, it is plausible thatwith increasing periodontal decline, the Sharpey fi-bers may lose their functional significance and decayowing to the reducing alveolar bone. An arrest of thecementum annulation process might occur. We paidspecial attention to a quantified measurement ofperiodontal decline. In our sample, however, we didnot observe any correlation (r � �0.075) betweenthe deviation of TCA age from true age and degree ofperiodontal decline. The outliers were distributedover the whole range of the periodontal decline. Thismeans that even in the quarter with the highestperiodontal regressions, the cementum annulationscontinue. From our study, we conclude that the ac-curacy of the TCA age estimation is independent ofperiodontal disease, a finding that supports the re-sults of Großkopf et al. (1996) rather than those ofKagerer and Grupe (2001). This provides a strongargument for the application of the TCA method inarchaeological skeletal samples in which most of theindividuals suffered from extreme dental disease.

To analyze reliability by tooth type, we had to splitour sample into rather small subgroups. However,as no sex differences in TCA estimation error wereobserved, different tooth types were not analyzedseparately by sex. The results of our study do notsuggest the selection of a specific tooth type as afavorable tooth type for age estimation by cementumannulations. All teeth in our sample from the max-illa as well as from the mandible provided results ofcomparable reliability. We do suggest avoiding sec-ond premolars, because they showed irregular struc-tures of the cementum band more often than anyother tooth type. If this is not considered, “doublingcases” might result. In these cases, exactly twice thenumber of expected lines are counted. This problem,which can affect up to 15% of a sample, was men-tioned in previous studies as one of the main limi-tations for TCA applications (Jacobshagen, 1999). Ithas been argued that the doubling of cementumlines is due to distinctive individual characteristics

TOOTH CEMENTUM ANNULATION 127

of the biomineralization process, which result in theapposition of two dark and light bands per year overthe whole life span. In our sample, we observed nooutliers that showed a doubled number of cementumbands (Fig. 5).

Deviations of TCA age estimation from true agemay be caused by distinctive individual featuressuch as an extremely early or late tooth eruption ordisorders of the calcium metabolism. Both these rea-sons may lead to alterations affecting the wholedentition, which are detectable by comparison ofresults obtained from different teeth of the sameindividual. Our results, based on 77 individuals withmultiple extractions, revealed an intraindividual ac-curacy which did not differ from that of independentsingle teeth. The individuals whose results all liebelow (22%) or above (21%) the estimated age mayrepresent early and late types of tooth eruption.

However, our sample includes 9 cases (2.5%) inwhich TCA age deviates by 5 years or more fromtrue age. We carefully checked these cases for anypeculiarities. Concerning the other known traits, wefound no regularities in this subsample. Four caseswith severe underestimation were from patients be-tween ages 60–68 years, whereas five teeth withTCA overestimation were distributed over a broaderage range. Three cases with counts too low by 10years or more stemmed from multiple extractions.In all these cases, the corresponding results of theother one or two teeth of the same individual pro-vided reliable results within the 2.5-year errorrange. This leads to the interpretation that there isno systematic factor influencing the process of ce-mentum apposition rhythm of alternating patternsas observed under the microscope. A check of theseextreme outliers with the aspartic acid racemizationmethod of Ritz et al. (1993) did not provide usefulresults, as the teeth were treated with alkaline so-lutions during the maceration process, which prob-ably caused protein degradation (Ritz-Timme, per-sonal communication).

We cannot rule out that these cases resulted froma mix-up of teeth at the dentist’s office, during prep-aration or during data processing. Although 2.5% isonly a small percentage of the whole sample, wesuggest that two teeth should be used for TCA anal-ysis in forensic cases, and TCA should be combinedwith the method of root translucency of Lamendin etal. (1992) until the reasons for erroneous results aredetected.

CONCLUSIONS AND PERSPECTIVES

Our results suggest that there is no statisticallysignificant influence of sex, age, periodontal disease,or tooth type on the estimation quality of the TCAmethod, if the described preparation and analysisstandard is followed. The obtained confidence bandprovides error estimates of less than 2.5 years.These results were obtained in a contemporaryknown-age sample. A subsequent study needs toshow whether the same accuracy can be achieved in

a historical known-age collection exposed to differ-ent environmental conditions than in a modern pop-ulation.

The results of the current study extend to severalaspects of historical and forensic anthropology. Onthe one hand, individual age estimation is improvedby smaller confidence intervals, thus providing indi-vidual ages with high probability in narrow ageranges. The application of TCA age estimation im-proves individual age estimation and even makesage estimation possible in cases of poorly preservedskeletal fragments. For a successful application inhistorical populations, the influence of living condi-tions that severely affect the calcium metabolism inthe human body still has to be tested, includingsevere malnutrition or specific diseases. This will bethe aim of a forthcoming paper based on a historicalknown-age sample.

In paleodemographic applications, it is importantto obtain accurate individual age estimates, espe-cially for the oldest individuals in the population,because the highest ages achieved and the numberof people reaching these ages significantly affect thecalculation of different mortality parameters (Woodet al., 2002). If the chronological age of the oldestindividuals in a historical population can be accu-rately determined by the TCA method, this will al-low us to better estimate life expectancy and thedistribution of life spans within the population un-der study, and to compare historical populationswith respect to their proportion of elderly individu-als.

Finally, we recommend the application of TCA ageestimation for forensic identification, since it maylead to reliable results even in poorly preserved deadbodies, and it may serve as a valuable aid for iden-tification.

ACKNOWLEDGMENTS

We thank the staff of the Max Planck Institute forDemographic Research (Rostock, Germany) for gen-erous support. The project would not have been pos-sible without the committed help of the laboratorystaff: Alexandra Alt, Helene Buba, Uta Cleven,Katja Krause, Kristin Lubcke, Heike Petzold,Doreen Pick, Kirstin Ringkowski, and MatthiasVoigt. We also gratefully acknowledge the hospital-ity of the Institute of Anthropology (University ofGießen, Gießen, Germany), which hosted the MaxPlanck “Tooth Lab” Team during the beginningstages of this project. We thank Arnold Kahn formany comments and for critical reading of themanuscript, and Sara Grainger for valuable edito-rial help. We also thank four anonymous reviewersfor helpful comments and suggestions on the manu-script, which led to significant improvements.

LITERATURE CITED

Adler P. 1967. Die Chroudogie der Gebissentwicklung. In: HarndtE, Weyers H, editors. Zahn-, Mund- und Kieferheilkunde imKindesalter. Berlin: Die Quintessent. p 38–74.

128 U. WITTWER-BACKOFEN ET AL.

Antoine D, Dean C, Hillson S. 2000. The periodicity of incremen-tal structures in dental enamel based on the developing denti-tion of post-medieval known-age children. In: Mayhall JT,Heikkinen T, editors. Dental morphology 1998. Oulu: OuluUniversity Press. p 102–111.

Azaz B, Ulmansky M, Moshev R, Sela J. 1974. Correlation be-tween age and thickness of cementum in impacted teeth. OralSurg 28:691–694.

Brown PJ. 1993. Measurement, regression, and calibration. Ox-ford: Oxford University Press.

Buikstra JE, Ubelaker DH. 1994. Standards for the data collec-tion from human skeletal remains. Research series 44. Fayet-teville: Arkansas Archeological Survey.

Charles KD, Condon K, Cheverud JM, Buikstra JE. 1986. Cemen-tum annulation and age determination in Homo sapiens. I. Toothvariability and observer error. Am J Phys Anthropol 71:311–320.

Cipriano A. 2001. Cold stress in captive great apes recorded inincremental lines of dental cementum. Folia Primatol (Basel)73:21–31.

Condon K, Charles KD, Cheverud JM, Buikstra JE. 1986. Cemen-tum annulation and age determination in Homo sapiens. II.Estimates and accuracy. Am J Phys Anthropol 71:321–330.

Geiger G. 1993. Vergleich verschiedener Methoden der Alters-beurteilung anhand von Zahnen und anderen morphologischenMerkmalen mit dem Lebensalter vorwiegend altersmarkierterWildtiere der Ordnungen Artiodactyla und Carnivora. Giessen:Habilitationsschrift Giessen.

Geuser G, Bondioli L, Capussi E, Cipriano A, Grupe G, Savore C,Macchiarelli R. 1999. Dental cementum annulations and age atdeath estimates. In: Bondioli L, Macchiarelli R, editors. Oste-odontal biology of the people of Portus Romae (Necropolis ofIsola Sacra, 2nd–3rd cent. AD). II. Digital archives of humanpaleobiology. Milan: E-LISA Sas.

Großkopf B. 1989. Incremental lines in prehistoric crematedteeth. A technical note. Z Morphol Anthropol 77:309–311.

Großkopf B. 1990. Individualaltersbestimmung mit Hilfe von Zu-wachsringen im Zement bodengelagerter menschlicher Zahne.Z Rechtsmed 103:351–259.

Großkopf B, Denden JM, Kruger W. 1996. Untersuchungen zurZementapposition bei Paradontitis marginalis profunda. DtschZahn Z 51:295–297.

Grue H, Jensen B. 1979. Review of the formation of incrementallines in tooth cementum of terrestrial mammals. Dan RevGame Biol 11:1–48.

Gustafson G. 1950. Age determinatin of teeth. J Am Dent Assoc41:45–54.

Gustafson G. 1955. Altersbestimmung an Zahnen. Dtsch Zahn Z25:1763–1768.

Halberg F, Lagoguey M, Reinberg A. 1983. Human circannualrhythms over a broad spectrum of physiological processes. IntJ Chronobiol 8:225–268.

Hoppa RD, Vaupel JW, editors. 2002. Paleodemography age dis-tributions from skeletal samples. Cambridge studies in biolog-ical and evolutionary anthropology 31. Cambridge: CambridgeUniversity Press.

Jackes M. 2001. Building the bases for paleodemographic analy-sis: adult age determination. In: Katzenberg MA, Saunders SR,editors. Biological anthropology of the human skeleton. NewYork: John Wiley and Sons, Inc. p 417–466.

Jacobshagen B. 1999. Prazisierte Altersschatzungen nach demZementzonenverfahren anhand der Zahne einer mittelneo-lithischen Skelettpopulation aus Trebur, Kreis Gross-Gerau.Vergleich mit den Ergebnissen der kombinierten Methode. In:Spatz H, editor. Das mittelneolithische Graberfeld von Trebur,Kreis Gross-Gerau. Materialien zur Vor- und Fruhgeschichtevon Hessen IG. Wiesbaden: Selbstuerlag des Laudesamtes furDenkmalpflege Hessen. p 333–348.

Jankauskas R, Barakauskas S, Bojarun R. 2001. Incrementallines of dental cementum in biological age estimation. Homo52:59–71.

Kagerer P, Grupe G. 2001. Age-at-death diagnosis and determi-nation of life-history parameters by incremental lines in hu-man dental cementum as an identification aid. Forensic Sci Int118:75–82.

Kay RF, Rasmussen DT, Beard KC. 1984. Cementum annuluscounts provide a means for age determination in Macaca mu-latta (Primates, Anthropoidea). Folia Primatol (Basel) 42:85–95.

Kemkes-Grottenthaler A. 2002. Aging through the ages: historicalperspectives on age indicator methods. In: Hoppa RD, Vaupel JW,editors. Paleodemography. Age distributions from skeletal sam-ples. Cambridge studies in biological and evolutionary anthropol-ogy 31. Cambridge: Cambridge University Press. p 48–72.

Kvaal SI, Solheim T. 1995. Incremental lines in human dentalcementum in relation to age. Eur J Oral Sci 103:225–230.

Lamendin H, Baccino E, Humbert JF, Tavernier JC, Nossin-tchouk RM, Zerilli A. 1992. A simple technique for age estima-tion in adult corpses: the two criteria dental method. J ForensicSci 37:1373–1379.

Laws RM. 1952. A new method of age determination for mam-mals. Nature 169:972–973.

Leica. 2003. www.leica-microsystems.com/website/Products.nsf/ProductIntermediate

Lieberman DE. 1994. The biological basis for seasonal incrementsin dental cementum and their applications to archaeologicalresearch. J Archaeol Sci 21:525–539.

Lipsinic FE, Paunovich E, Houston DG, Robison SF. 1986. Cor-relation of age and incremental lines in the cementum of hu-man teeth. J Forensic Sci 31:982–989.

McCulloch CE, Searle SR. 2001. Generalized, linear, and mixedmodels. New York: Wiley.

Naylor JW, Miller GW, Stokes GN, Stott GG. 1985. Cementumannulation enhancement: a technique for age determination inman. Am J Phys Anthropol 68:197–200.

Parner ET, Heidmann JM, Kjær I, Væth M, Poulsen S. 2001.Biological interpretation of the covariation of eruption times ofpermanent teeth. Research report 2001-3. Aarhus: Departmentof Biostatistics, University of Aarhus.

Philipsen HP, Jablonski NG. 1992. Age estimation from the struc-ture of adult human teeth: review from the literature. ForensicSci Int 54:23–28.

Pinheiro JC, Bates DM. 2000. Mixed-effects models in S andS-Plus. New York: Springer-Verlag.

Renz H, Schaefer V, Duschner H, Radlinski RJ.1997. Incrementallines in root cementum of human teeth: an approach to theirultrastructural nature by microscopy. Adv Dent Res 11:472–474.

Ritz S, Schutz HW, Peper C. 1993. Postmortem estimation of ageat death based on aspartic acid racemization in dentin: itsapplicability for root dentin. Int J Leg Med 105:289–293.

Rousseeuw PJ, Leroy AM. 1987. Robust regression and outlierdetection. New York: Wiley.

Schroder HE. 2000. Orale Strukturbildung: Entwicklungsge-schichte, Struktur und Funktion normaler Hartgewebe undWeichgewebe der Mundhohle und des Kiefergelenkes. Berlin:Thieme-Verlag.

Schumacher GH, Schmidt H, Boring H, Richter W. 1990. Anato-mie und Biochemie der Zahne. Stuttgart: Gustav Fischer Verlag.

Solheim T. 1990. Dental cementum apposition as an indicator ofage. Scand J Dent Res 98:510–519.

Stein TJ, Corcoran JF. 1994. Paradicular cementum deposition asa criterion for age determination in human beings. Oral SurgOral Med Oral Pathol 77:266–270.

Stott GG, Sis RF, Levy BM. 1980. Cemental annulation as an agecriterion in the common marmoset (Callithrix jacchus). J MedPrimatol 9:274–285.

Stott GG, Sis RF, Levy BM. 1982. Cemental annulation as an agecriterion in forensic dentistry. J Dent Res 61:814–817.

Wittwer-Backofen U, Buba H. 2002. Age estimation by toothcementum annulation: perspectives of a new validation study.In: Hoppa RD, Vaupel JW, editors. Paleodemography. Age dis-tributions from skeletal samples. Cambridge studies in biolog-ical and evolutionary anthropology 31. Cambridge: CambridgeUniversity Press. p 107–128.

Wood JW, Holman DJ, O’Connor K, Ferrell RJ. 2002. Mortalitymodels for paleodemography. In: In: Hoppa RD, Vaupel JW,editors. Paleodemography. Age distributions from skeletalsamples. Cambridge studies in biological and evolutionary an-thropology 31. Cambridge: Cambridge University Press. p 129–168.

TOOTH CEMENTUM ANNULATION 129