J Periodontol • February 2005

279

* Center for Pharmacogenomics and Complex Disease Research, New Jersey DentalSchool, UMDNJ, Newark, NJ.

† Department of Preventive Sciences, University of Minnesota School of Dentistry,Minneapolis, MN.

‡ Clinical Research Center for Periodontal Diseases, School of Dentistry, VirginiaCommonwealth University, Richmond, VA.

Traditionally, descriptions of peri-odontal disease status and diagnoseshave been based on observation of

sites with periodontal attachment loss (AL)upon probing, severity of the observedperiodontal lesions, and the age of thepatient at time of examination or at theassumed onset of disease.1-9 Until recently,periodontal diseases without systemic eti-ology have been broadly divided into adultperiodontitis and early-onset periodontitis,the distinction based largely upon the ageof onset of attachment loss after age 30 orbefore age 25, respectively. Early-onsetperiodontitis was subcategorized as local-ized when it involved primarily first molarsand incisors or generalized otherwise. Thegeneralized early-onset form was even fur-ther subdivided according to the estimatedage of onset as well as laboratory-basedcharacterization of the microflora and hostresponse.5,9 In 1999, the American Acad-emy of Periodontology revised the systemfor categorizing periodontal diseases byomitting references to age of onset.10 Thepresent system relies upon the relativerates of disease progression.9,10 Formspreviously designated as “early-onset” arenow categorized as “aggressive periodon-titis” (AgP), while cases previously diag-nosed as “adult periodontitis” are nowcalled “chronic periodontitis,” without ref-erence to specific criteria for the actual ageof disease onset.

In applying currently accepted meth-ods of diagnosis, there are no universally

Quantitative Measures of AggressivePeriodontitis Show Substantial Heritabilityand Consistency With TraditionalDiagnosesScott R. Diehl,* Tianxia Wu,* Bryan S. Michalowicz,† Carol N. Brooks,‡ Joseph V. Califano,‡John A. Burmeister,‡ and Harvey A. Schenkein‡

Background: Aggressive periodontitis (AgP) research nearlyalways classifies subjects into traditional discrete categories oflocalized or generalized, based upon degree of attachment loss(AL) and types of affected teeth. Since AL is continuous andquantitative, however, useful information is lost. We developedquantitative measures of AgP, compared these to traditional meth-ods, and estimated heritabilities in families.

Methods: We examined 237 healthy, 169 localized AgP, and204 generalized AgP subjects. We used the site of maximumAL of each tooth to calculate means for each subject for differ-ent groups of teeth. We also applied principal components analy-sis (PCA) to condense variation among 28 teeth into threeorthogonal (uncorrelated) variables. We used discriminant func-tion analysis (DFA) to evaluate how well the quantitative mea-sures match with traditional classifications. Quantitative traitheritabilities were estimated by variance components.

Results: PCA clustered first molars, incisors, and the otherteeth into three groups. DFA showed that quantitative measuresclassified subjects consistent with traditional methods (87% to94% agreement). Heritabilities ranged from 13.7% (P = 0.10) to30.0% (P = 0.008) for quantitative measures, with highest valuesobtained for first molars. A combination of the principal compo-nent variables most heavily weighted on first molars and incisorsgave the best model of disease susceptibility, with good separa-tion of healthy versus diseased subjects, independent of diseaseextent or severity.

Conclusions: Quantitative measures may provide improvedprecision and power for many kinds of periodontal research. Ourfinding of significant heritability supports their use in gene map-ping studies of AgP susceptibility. J Periodontol 2005;76:279-288.

KEY WORDSGene mapping; hereditary diseases; periodontal diseases;periodontitis, aggressive.

30101.qxd 3/10/05 10:41 AM Page 279

280

Quantitative Measures of Periodontitis Volume 76 • Number 2

accepted thresholds for determining whether a partic-ular patient has periodontitis or a subtype or level ofseverity of disease. Research studies of periodontaldiseases often set threshold criteria for patient entry;for example, in clinical trials of new therapeutic agents,homogeneity in patient groups may be desirable, sosubjects with mild or very severe levels of disease maybe excluded. However, in epidemiological surveys andfamily studies, it is necessary to assign a periodontalstatus to all members of a given cohort or family and,therefore, to apply specific rules for assigning eachsubject into a unique diagnostic group.11

To objectively diagnose disease, investigators havesometimes developed numerical cutoffs,12 for exam-ple, “the percent of teeth or percent of sites with atleast 3 mm attachment loss.” Such variables may bevery useful for describing the extent of disease, butcutoff is arbitrary, especially when only a single valueis considered. Furthermore, this approach does nottake into account the localized or generalized natureof AgP. To address this concern, in previous researchwe developed more complex numerical algorithms thattake into account which teeth are affected at differentseverity thresholds of AL.11 Ultimately, however, westill classified our subjects as either healthy or affectedby localized or generalized disease.

Discrete classification approaches ignore the factthat the severity and extent of periodontal disease actu-ally range across a continuum. A subject just belowand a subject just above the cutoff value in our algo-rithm may have identical biological mechanisms andenvironmental risk factors causing their disease, butone subject would be classified as healthy and theother as diseased. Alternatively, one subject may bejust barely above the threshold for classification asaffected and another very severely affected and farabove the threshold. Unfortunately, our algorithm usingonly discrete classification would lump these togetheras affected, and subsequent analyses would treat themas if they were clinically identical. By contrast, in stud-ies measuring blood pressure to assess risk of hyper-tension, or measuring blood lipids to estimate risk ofcoronary artery disease, investigators rarely convertcontinuous measurements into simple categories oflow and high. Instead, they utilize all of the informationcontained in the continuous range of measurement byapplying methods of statistical analysis appropriatefor quantitative data.

Here, we report the development and evaluation ofquantitative measures of periodontitis in a study samplerecruited for the study of AgP. We tested simple statis-tics such as means of AL in all teeth or in subsets ofteeth such as incisors and first molars. We also evalua-ted multivariate statistical methods that condense ALmeasures from all teeth into three complex variableswhile retaining most of the information of the 28 orig-

inal measures. We assessed the validity of alternativequantitative measures by observing how well diag-nostic classifications of subjects based on these mea-sures match those of traditional methods and byanalyzing families to estimate the genetic basis (heri-tability) of the measures.

MATERIALS AND METHODSClinical Study Sample and Traditional DiagnosesWe recruited and examined 959 subjects referred to theClinical Research Center for Periodontal Disease fromSchool of Dentistry Clinics at Virginia CommonwealthUniversity or by practicing clinicians in the Richmond,Virginia, area and surrounding counties. The protocolwas reviewed and approved by an Institutional ReviewBoard at Virginia Commonwealth University andinformed consent was obtained from all study sub-jects. Our study sample included AgP affected pro-bands and their periodontally healthy and AgP affectedrelatives.

Examiners calibrated to provide uniform measuresassessed AL (in mm) at mesio-facial, disto-facial, mid-buccal, and mid-lingual locations on each tooth. Sub-jects were classified as healthy or as having eitherlocalized AgP, or generalized AgP according to the fol-lowing criteria.

Healthy periodontium. Subjects of any age with noevidence of attachment loss (AL) of ≥3 mm at morethan one site, or pockets greater than 3 mm; i.e., nodetectable periodontitis.

Chronic periodontitis. Subjects age 25 or older with≥2 mm AL on more than one tooth. For subjects youn-ger than 35 years of age, a diagnosis of chronic perio-dontitis could be made if the AL was commensuratewith plaque levels or other local contributing etiologi-cal factors, less severe than in localized or generalizedAgP cases (described below), or clearly developed dur-ing adulthood. Furthermore, the disease must not haveaffected only first molars and incisors, or been relatedsolely to trauma, endodontic disorders, or other locallydeterminable etiology other than periodontitis.

Localized AgP. For probands, disease onset frompuberty up to age 30, with ≥4 mm AL on at least twopermanent first molars and incisors (at least one firstmolar must have been affected) and no more than twoteeth which were not first molars or incisors that wereaffected by ≥5 mm AL or more. By these criteria, a sub-ject with AL of 4 mm at all sites would not qualify asa localized AgP proband and would be excluded fromour study (but no such cases have been encounteredto date). For family members of probands, AL at firstmolars and incisors may have been less than 4 mm butnot less than 2 mm.

Generalized AgP. For probands, disease onset up toage 35, with eight or more teeth having ≥5 mm of AL,at least three of which were not first molars or incisors.

30101.qxd 3/10/05 10:41 AM Page 280

281

J Periodontol • February 2005 Diehl, Wu, Michalowicz, et al.

For family members of probands, AL at multiple sitesmay have been less than 5 mm but not less than 3 mm.In addition, the pattern of AL must have been moreexcessive than would be likely for chronic periodontitis.

Family members and probands were excluded ifthey had diabetes or any other systemic disorders asso-ciated with periodontitis. Additional details about therecruitment of our clinical sample and our methods ofdiagnosing periodontitis have been reported else-where.11 To identify current tobacco users, serum cotin-ine was analyzed by double antibody radioimmunoassaymethods as described previously.13 Cotinine measureswere strongly bi-modal, with most subjects having con-centrations well above or well below 75 ng/ml. There-fore, a cutoff point of >75 ng/ml was used to classifysubjects as smokers.

Quantitative Measures of PeriodontitisOf 959 subjects initially considered, 572 had data froma single clinical examination and 387 had data frommultiple exams. For subjects with multiple exams, weapplied the following rules to select which examinationto use in this analysis. We first chose the examina-tion(s) at which the subject had been assigned themost severe form of AgP (generalized selected overlocalized over healthy). For subjects then assigned aslocalized or generalized AgP, the first examination atwhich this diagnosis was assigned was selected. Forsubjects assigned as chronic periodontitis or as healthy,the most recent examination within that diagnosis cat-egory was selected.

AL values over 10 mm were truncated at 10 mm.Third molars were not included in the analyses. Forsubjects with missing teeth, previous examinationswere used to impute the missing AL data. For eachsubject, the previous examination for which the sub-ject had the most sites measured was chosen forimputing. After imputing by previous exam, 31 subjectswith less than 10 teeth with AL data and four subjectswith other missing or inconsistent data were excluded.If a tooth was missing at the subject’s baseline exam,AL for that tooth was imputed as the mean maximumAL on the remaining teeth of that type; i.e., first molars,second molars, first and second premolars, cuspids, orincisors. Eighteen subjects were excluded because theydid not have teeth with AL measures in one of thetooth groups. Finally, because we focused on AgP, weexcluded 296 subjects in our families who were lessthan 13 or greater than 35 years old at the time of theselected examination or who were assigned a diagno-sis of chronic periodontitis. This yielded our final dataset of 610: 237 periodontally healthy subjects, 169subjects classified with localized, and 204 with gener-alized AgP.

Preliminary analyses using all four AL measuresfrom all 28 teeth (112 measures total) indicated that

this was too great a number of variables for our sam-ple size of 610 subjects. Therefore, we selected themaximum AL for each tooth as our primary unit ofanalysis, thus reducing the number of variables to 28.Using these measures of maximum AL per tooth, wecalculated five simple means for each subject: all 28teeth, incisors and first molars combined, all otherteeth combined (cuspids, premolars, and secondmolars), incisors alone, and first molars alone. The val-ues of these means were plotted to compare their dis-tributions in periodontally healthy subjects versussubjects affected by localized or generalized AgP.

We next applied principal components analysis tothe 28 variables using a statistical software program,§

setting the initial factor method to principal compo-nents.14 Three eigenvalues >1 cumulatively accountedfor 78% of the variance in the original 28 variablesand, therefore, were retained in the analysis. After vari-max orthogonal rotation, three rotated principal com-ponents weights were calculated for each of the 28teeth. Three rotated principal components scores werethen calculated for each subject using the scores com-mand. Scores are standardized to have a mean of 0.0and standard deviation of 1.0. The values of pairs ofscores were plotted to compare their distributions inperiodontally healthy versus subjects affected by local-ized or generalized AgP.

Discriminate function analysis was conducted toassess how well the quantitative measures match withtraditional disease classifications.15 Overlap of thequantitative measures between the diagnostic groups(healthy, localized AgP, or generalized AgP) was asses-sed by cross-validation. Each subject was classifiedinto one of the three groups using a discriminant func-tion that was computed from the other observations,excluding only the subject being classified. Age atexamination was not incorporated into these analy-ses because we sought to assess robustness of thequantitative measures alone. We wanted to avoid pro-ducing a more favorable match with the traditionaldiagnoses by relying on the well-established age ofonset patterns associated with this disease.

Quantitative Genetic AnalysesVariance components analyses were conducted to esti-mate the heritability of quantitative measures of AgP.To perform these analyses, we used release 1.7.3 of theSequential Oligogenic Linkage Analysis Routines(SOLAR) software package.16 This method is robust todeviations from normality as long as skewness andkurtosis are both less than 1.0. The means of maximumAL per tooth were consistent with this requirement,and so were analyzed without transformation. Therotated principal component scores, however, exhib-

§ Procedure FACTOR, SAS Software Version 6.12, SAS, Cary, NC.

30101.qxd 3/10/05 10:41 AM Page 281

282

Quantitative Measures of Periodontitis Volume 76 • Number 2

ited kurtosis of 1.5 to 2.0. After square-roottransformation, kurtosis was less than 0.27and variance component analyses were per-formed using the transformed scores. Covari-ates evaluated included age, age squared,gender, race (African-American [black] orCaucasian [white]), and smoking (coded asa binary “yes” or “no” trait). Heritability analy-ses were limited to a subset of 288 study sub-jects who were members of 17 Caucasianand 43 African American families where ALdata were available on multiple relatives.These included one family with 21 localizedor generalized AgP-affected, one family with11 affected, five families with five to nineaffected, eleven families with four affected,14 families with three affected, 20 familieswith two affected, and eight families with onlyone affected member. AL data sufficient forour quantitative analyses were available for112 periodontally healthy subjects, 71 subjectsclassified as localized AgP, and 105 general-ized AgP.

RESULTSFigure 1 shows the mean maximum AL pertooth for all 28 teeth for healthy, localizedAgP, and generalized AgP subjects. Eventhough this simple statistic ignores the dis-tinction between incisors, first molars, andother types of teeth, overlap among thesethree groups is limited. For example, mostgeneralized AgP subjects have means >4 mmand only overlap with localized AgP subjectsin the 2 to 4 mm range. Most localized AgPsubjects, in turn, have means less than2 mm, and only a quarter of the localizedAgP subjects overlap with healthy subjects inthe 0 to 1 mm range.

When we examine mean maximum ALper tooth separately for the combination ofincisors and first molars versus all other teeth(Fig. 2), we observed substantial overlapbetween healthy and localized AgP subjectsin the 0.5 to 1.5 mm range for incisor andfirst molars. We also found several individu-als classified as healthy even though theirmean maximum AL per tooth for bothincisors and first molars and other teeth were fairlyhigh (2 mm). Most localized and generalized groups didnot overlap but rather were separated, as expected,by differences in AL at teeth other than incisors andfirst molars. However, there are about a dozen gener-alized subjects with mean maximum AL per toothscores that place them well within the localized AgPcluster.

We found the mean maximum AL/tooth was muchgreater in first molars than incisors for both localizedand generalized subjects; i.e., most values were abovethe diagonal; see Figure 3. In further analyses (datanot shown), we determined that this was because ahigher proportion of first molars had AL ≥4 mm, andnot because AL was more severe at these teeth. Therewas substantial overlap of localized and generalized

Figure 1.Healthy, localized and generalized AgP subjects show limited overlap in their meanmaximum attachment loss per tooth. Each category on the x-axis defines subjectswith mean AL greater than the first number but less than or equal to the secondnumber shown, except the first group with exactly 0.0 mm and the last group withgreater than 8 mm mean AL.

Figure 2.Localized AgP subjects show substantial overlap with generalized subjects in meanmaximum attachment loss per tooth for the combination of incisors and first molars,but little overlap for other tooth types.

30101.qxd 3/10/05 10:41 AM Page 282

283

J Periodontol • February 2005 Diehl, Wu, Michalowicz, et al.

subjects in mean maximum AL/tooth at first molars.However, generalized subjects had greater AL thanlocalized subjects at incisors (Fig. 3).

Table 1 shows results of the rotated principal com-ponents analysis (PCA) of the maximum AL valuesfor each of the 28 teeth. The intent of PCA is to con-dense the measures of 28 teeth into fewer summarymeasures and to describe the simpler measures. Sincethe factor pattern shown in Table 1 can be interpretedas correlations, we see that the first principal compo-nent (second column in the Table) correlates moststrongly with attachment loss of cuspids, premolarsand second molars (illustrated by boxes around thecomponents for these teeth). For instance, the firstPCA correlates 0.74 with the maximum attachmentloss of the first premolar in the upper left quadrant.These three groups nearly perfectly match the toothtypes recognized by clinicians as important for distin-guishing localized versus generalized AgP. That is, thefirst rotated principal component is correlated with themean/max AL on all cuspids, premolars, and secondmolars; the second rotated principal component (thirdcolumn) is correlated with the incisors (with the excep-tion of one cuspid that has nearly equal weights for thefirst and second components), and the third (fourthcolumn) PCA is correlated with the mean/max ALobtained from the first molars. These three rotatedprincipal components account for 33.8%, 23.3%, and20.7%, respectively, a total of 77.8%, of the variance inthe original 28 individual tooth measurements. Thus,

instead of using 28 teeth-specific measures,our analyses may use the reduced set of onlythree PCA scores.

The rotated principal components for eachtooth were used to derive rotated principalcomponents scores for each subject as a lin-ear combination of each column in Table 1.For simplicity, scores for each principal com-ponent were designated as the CP1P2M2component, the incisor component, and thefirst molar component. However, it should benoted that because the weights for all com-ponents teeth are greater than zero, eachcomponent score is actually a linear combi-nation of the means of maximum AL of allteeth. Because of orthogonal rotation, thethree component scores are uncorrelated.

Figure 4 shows the distribution of theincisor and first molar components scoresfor healthy, localized AgP, and generalizedAgP subjects. Here we see a much wider dis-tribution of quantitative values in the healthysubjects than was observed in any of thesimple means shown in Figures 1, 2, or 3.Furthermore, healthy subjects group togetherin a cluster that is very distinct from AgP

subjects, while there is nearly complete overlap of thescores of localized and generalized subjects. A line hasbeen drawn to illustrate the separation of the two clus-ters of healthy versus diseased subjects. The distribu-tion of the incisor and the CP1P2M2 components, bycontrast (Fig. 5), clusters together the healthy andlocalized subjects in a region of space distinct fromthe scores of generalized AgP subjects. A line againillustrates the separation of the two clusters.

Results in Table 2 show that disease classificationbased on the three rotated principal components scoresis highly consistent with traditional methods. Discrim-inate function analysis classified 97% of healthy sub-jects, 91.7% of localized AgP subjects, and 92.7% ofgeneralized AgP subjects into the same group asassigned by our traditional method of diagnosis. Onlyone subject fell midway between two groups and wasassigned “unclassified.” Using the simple mean max-imum AL per tooth averaged across all teeth (Fig. 1),we still could correctly classify 91.4% of healthy sub-jects, 72.7% of localized subjects, and 92.9% general-ized subjects. Using two variables, means for incisorsand first molars (combined) and means for other teeth(Fig. 2), we classified healthy subjects 95.1%, local-ized subjects 88.9%, and generalized subjects 93.4%consistent with our traditional method.

Variance components analyses of eight quantitativemeasures of AgP in families with one or more AgP-affected member are presented in Table 3. Estimatesof heritability ranged from 13.7% (P = 0.103) to 30.0%

Figure 3.Both localized and generalized AgP subjects show much greater mean attachmentloss per tooth for first molars than for incisors.AL of incisors is greater in generalizedsubjects.

30101.qxd 3/10/05 10:41 AM Page 283

284

Quantitative Measures of Periodontitis Volume 76 • Number 2

Table 1.

Principal Component Analysis of MaximumAttachment Loss per Tooth

Rotated Principal Components(% of variance explained)

CP1P2M2 Incisors First MolarsTooth (33.8) (23.3) (20.7)

First premolar (upper left) 0.74 0.25 0.40

First premolar (upper right) 0.74 0.36 0.34

Cuspid (upper left) 0.74 0.38 0.27

First premolar (lower left) 0.73 0.41 0.26

Cuspid (upper right) 0.72 0.43 0.20

First premolar (lower right) 0.71 0.43 0.21

Second premolar (upper left) 0.69 0.27 0.47

Second premolar (lower left) 0.68 0.37 0.37

Second premolar (upper right) 0.68 0.30 0.41

Second molar (lower left) 0.67 0.34 0.45

Second premolar (lower right) 0.66 0.36 0.38

Cuspid (lower left) 0.66 0.48 0.20

Second molar (lower right) 0.65 0.30 0.49

Second molar (upper left) 0.64 0.27 0.55

Second molar (upper right) 0.64 0.31 0.50

Central incisor (lower left) 0.31 0.79 0.35

Central incisor (lower right) 0.30 0.76 0.39

Lateral incisor (lower right) 0.42 0.75 0.32

Lateral incisor (lower left) 0.45 0.74 0.32

Central incisor (upper left) 0.42 0.65 0.39

Cuspid (lower right) 0.62 0.63 0.17

Central incisor (upper right) 0.37 0.60 0.43

Lateral incisor (upper left) 0.55 0.58 0.34

Lateral incisor (upper right) 0.54 0.56 0.35

First molar (upper left) 0.32 0.35 0.81

First molar (lower left) 0.33 0.33 0.80

First molar (lower right) 0.33 0.35 0.79

First molar (upper right) 0.35 0.36 0.77

Figure 4.The combination of the two principal components most stronglyweighted on incisors and first molars provides wide separation ofhealthy from both localized and generalized AgP subjects as indicatedby the diagonal line.

Figure 5.The combination of the two principal components most stronglyweighted on incisors and all other teeth (except first molars) providesgood separation of generalized AgP from both healthy and localizedAgP subjects as indicated by the diagonal line.

30101.qxd 3/10/05 10:41 AM Page 284

285

J Periodontol • February 2005 Diehl, Wu, Michalowicz, et al.

(P = 0.008). AL at first molars tended to show thehighest heritability, both for the simple mean maximumAL per tooth (26.0%) and for the first molar componentscores (30.0%). As expected, age and age-squared werevery highly significant covariates, with the exceptionof the first molar component scores. Gender was notsignificant for any variable and race was only significantlyassociated with the simple mean variable for firstmolars and for the incisor component scores. Smokingwas significantly associated with all of these quantita-tive measures except the incisor component scores.Combined, the covariates accounted for 15% to 30%of the variance, with the exception of the first molarcomponent score, where less than 3% of the variancecould be attributed to these variables.

DISCUSSIONThe methods utilized in this analysis result in the cre-ation of a series of simple (e.g., mean AL) and com-plex (principal components) variables, each of which

can be used to quantitatively describe the periodontalcondition of individual AgP patients or healthy sub-jects in AgP families. The results indicate that quanti-tative variables that utilize AL information from all teethmay be superior to traditional definitions of AgP forobjective descriptions of patients’ diagnostic status,and that complex variables that assign different levelsof importance to certain teeth may be the preferreddescriptors of periodontal status.

We have shown that the classification of subjectsinto discrete groups of healthy versus localized or gen-eralized AgP obscures a wide range of continuous vari-ation in AL among different types of teeth (Figs. 1, 2,and 3). Nevertheless, simply using means of the maxi-mum AL per tooth for all teeth (Fig. 1) or various toothsubgroups (Figs. 2 and 3) can reliably classify sub-jects according to their disease status. Our quantitativedata show that a greater proportion of first molars thanincisors are affected (Fig. 3). Whether AL occurs atincisors or at first molars is usually not emphasized in

traditional systems for classifying AgP.9,11

The fact that our quantitative methods dis-tinguish these may provide better resolutionof the disease process.

Most of the variation in quantitative mea-sures based on these simple mean scoresdistinguishes levels of severity of disease.Healthy subjects cluster very close togetherat low levels of AL in these mean score dis-tributions, while subjects classified as local-ized or generalized AgP are spread out overa much wider range. These quantitativemeasures based on these simple means donot primarily reflect “susceptibility” to dis-ease but are more like indicators of disease

Table 3.

Genetic and Environmental Effects on Maximum Attachment Loss per Tooth in VarianceComponents Models

First Molars Incisors Incisors CP1P2M2 Incisor First Molar CP1P2M2All Teeth Only Only and First Molars Teeth Component Component Component

Heritability (%)(P value) 16.2 (0.048) 26.0 (0.020) 18.5 (0.032) 20.1 (0.029) 13.7 (0.103) 22.8 (0.056) 30.0 (0.008) 21.3 (0.017)

Covariate P value:Age <0.001 <0.001 <0.001 <0.001 <0.001 <0.001 0.506 <0.001Age-squared <0.001 <0.001 <0.001 <0.001 <0.001 0.084 0.327 <0.001Gender 0.266 0.113 0.196 0.168 0.408 0.071 0.428 0.867Race 0.249 0.037 0.560 0.250 0.252 0.004 0.163 0.123Smoking 0.006 0.047 0.002 0.009 0.006 0.319 0.032 0.010

Proportion of variance 0.297 0.258 0.239 0.256 0.315 0.150 0.026 0.298due to all covariates

Table 2.

Discriminant Function Classification Using RotatedPrincipal Components Scores

Discriminant Function Classification (N subjects, %)TraditionalClassification Healthy Localized AgP Generalized AgP Unclassified

Healthy 230 (97.0) 4 (1.7) 2 (0.9) 1 (0.4)

Localized AgP 10 (5.9) 155 (91.7) 4 (2.4) 0 (0.0)

Generalized AgP 0 (0.0) 15 (7.4) 189 (92.7) 0 (0.0)

30101.qxd 3/10/05 10:41 AM Page 285

286

Quantitative Measures of Periodontitis Volume 76 • Number 2

severity (magnitude of AL at affected sites) and extent(number and types of teeth affected).

In contrast, when we applied multivariate principalcomponents analysis to these data, we obtained threeindependent variables that provide a much betterreflection of disease susceptibility (Fig. 4). Two of therotated principal components have greatest weightsfor first molars and incisors (Table 1), the tooth typeslong recognized as important in early-onset disease9,11

However, these complex variables have substantialcontributions from other teeth also, and this makestheir distributions very different from those of the sim-ple mean scores of first molars and incisors. Thesetwo rotated principal components scores show a muchwider dispersion of periodontally healthy subjects.There is also very substantial overlap of the scores oflocalized and generalized AgP subjects. The line drawnin Figure 4 shows how the combination of these twovariables divides the Cartesian space of these twoquantitative scores (with very few exceptions) intoregions of health versus disease.

This contrast is precisely what is needed for quanti-tative studies of disease susceptibility. For genetic stud-ies, our finding that these quantitative measures of AgPhave substantial heritability up to 30% (Table 3) isequally important. Unless the measures of disease showevidence of heritability (increased risk in relatives), it isimpossible to identify genetic factors that influence sus-ceptibility by linkage or linkage disequilibrium (associ-ation) approaches.17,18 It should be noted that withoutdata from twins or molecular markers, non-geneticcauses of similarity among close relatives, such as famil-ial transmission of virulent microbes,19 could be respon-sible for the significant heritabilities observed in thesequantitative measures of AgP. Future gene mappinganalyses should be able to resolve these alternativehypotheses.

The first rotated principal component score is weightedmost heavily on cuspids, premolars and second molars(Table 1). As expected, this variable distinguishes local-ized from generalized forms of AgP (Fig. 5). Since italso has significant heritability of 21.3% (Table 3), stud-ies focusing on this variable may provide an approachfor investigating the genetic and environmental causesof disease severity and extent.

We did not specifically test for heritability of local-ized versus generalized disease. To do so, would haverequired additional analyses in which only localizedand healthy subjects or only generalized and healthysubjects were included. Although chronic periodonti-tis subjects within our families were not included inthe current analyses which focused on AgP, thequantitative methods developed here can also beapplied to this type of disease.

Because the rotated principal component scoresare complex combinations of weights from all 28 teeth,

they are much less intuitively easy to grasp and under-stand from a clinical perspective than the simplemeans of maximum AL for groups of teeth. This maybe considered a disadvantage for use in research or insome future application to clinical practice. However,in addition to their potential advantages of reflectingdisease susceptibility or disease severity (dependingon which variables are chosen), and having signifi-cant heritability, another advantage of these complexvariables is that they match closely with traditionalmethods of classification of periodontal disease in dis-crete categories 97% of the time (Table 2). Thus, thiscomplex statistical analysis, despite being “masked”to clinical traditions and experience, arrived at gen-eral patterns and relationships very similar to theseprecedents.

We noticed, however, that the rotated principalcomponent scores exhibited surprising patterns of asso-ciation with some of the covariates. Smoking was asso-ciated with all quantitative measures except the incisorprincipal component, race was strongly associated onlywith this incisor principal component, and age wasstrongly associated with all quantitative measuresexcept the first molar principal component (Table 3).Apparently, the “masked” process that the statisticalalgorithm applied to generate these complex variablesled to the grouping of the smoking, race, and age-related variance into only two of these variables. Alter-natively, we could have included age, smoking, andthe other covariates along with the 28 AL variables thatwere used to calculate the principal components. Wedecided not to do this because it would have madeinterpretation of the resulting component variables evenmore difficult to interpret and to relate to the diseaseprocess.

Our approach has some similarity to the extent andseverity index (ESI) reported previously.20 Both meth-ods aim to identify quantitative measures that reflectboth the extent of disease; i.e., the distribution amongdifferent teeth in the mouth and the severity of diseasewhere it is present. The “extent” measure of the ESIused a cutoff point of 1 mm and calculated the percent-age of sites examined that exceed this value. Becauseof our concerns about the arbitrariness of using only asingle cutoff point as mentioned earlier, we used meanAL per tooth measures rather than percentages. Weinvestigated the distribution of disease by focusing onspecific tooth types for the means, and by assigning dif-ferent weights to each tooth in our principal componentsanalysis. The “severity” measure of the ESI is alsobased on mean AL, but includes only sites with AL>1 mm and can include multiple sites of a single tooth.In contrast, we used only one measurement per tooth,the tooth’s maximum AL site, and included all teeth inour mean scores, even when 0 mm of AL was found.The ESI was designed for epidemiological studies where

30101.qxd 3/10/05 10:41 AM Page 286

287

J Periodontol • February 2005 Diehl, Wu, Michalowicz, et al.

only a subset of teeth are measured. No attempt wasmade in previous reports applying the ESI to imputeany missing values as we did for our analysis. Anotherquantitative method for assigning AgP diagnoses hasbeen proposed recently that can be extended to sub-jects not available for examination by using case his-tories and previously gathered clinical evidence ofdisease.21 Although the authors recognize that diag-noses cannot be as certain without examinations, instudies of families for genetic research, or for otherpurposes, it may be better to make an educated guessabout an individual’s phenotype than to simply codeunexamined family members as missing.

Previously reported segregation analyses of AgPsuggested that most of the heritable variation in sus-ceptibility to AgP could be attributed to a single geneof “major” effect.22-25 However, many investigatorsnow consider the complex disease model17,18,26 moreplausible for periodontal diseases.8,11,27 The variancecomponents method we used assumes that the quan-titative measures of AgP are controlled by multiplegenes and environmental factors; i.e., it is a complexdisease.16,28 Our finding of heritabilities of only 30%means that identifying susceptibility genes for AgPmay be much more difficult than mapping the singlegene of major effect that had been suggested by seg-regation analyses of this disease. However, genes ofmoderate effect have been identified for other com-plex disorders including HIV susceptibility; thrombosis;hemochromatosis; diabetes; colon, breast, and othercancers; and Alzheimer’s, Graves, and Crohn’s dis-eases,18 and new genomic technologies are advanc-ing rapidly to expand our capabilities.29

Nearly 30 years ago, Stratford characterized an idealperiodontal disease index as one that is simple, accu-rate, quantitative, reproducible, objective, and amen-able to statistical analysis.30 The quantitative measuresdeveloped and evaluated in this study represent ourattempt to address this challenge today. While theiruse today is limited to research studies, in the futureit may be beneficial to bring such quantitativeapproaches into clinical practice.

ACKNOWLEDGMENTSWe thank our study families for their cooperation inthis research, and express our appreciation for techni-cal assistance and administrative support provided byJ. Francis, D. Ruggles, P. Ober, A. Miller-Chisholm(National Institute of Dental and Craniofacial Research,Bethesda, Maryland), K. Lake, M. Poland, and D.Williams (Virginia Commonwealth University [VCU],Richmond, Virginia). We thank A.M. Best (VCU) foradvice on multivariate statistical methods; J. Blangero,C. Peterson, and T. Dyer (Southwest Foundation forBiomedical Research, San Antonio, Texas) for guid-ance in running the program and interpreting the results;

and A.M. Best, R.P. Erickson (University of ArizonaHealth Science Center, Tucson, Arizona), and D.H. Fine(New Jersey Dental School, University of Medicine andDentistry of New Jersey, Newark, New Jersey) for com-ments on the manuscript. This study was supportedby Public Health Service grant DE-13102 from theNational Institute of Dental and Craniofacial Research,and resources provided by the University of Medicineand Dentistry of New Jersey, Newark, New Jersey andVirginia Commonwealth University, Richmond, Virginia.

REFERENCES1. Lilienthal B, Amerena V, Gregory G. An epidemiological

study of chronic periodontal disease. Arch Oral Biol1965;10:553-566.

2. Glickman I. Periodontal disease. N Engl J Med 1971;284:1071-1077.

3. Polson AM, Goodson JM. Periodontal diagnosis. Currentstatus and future needs. J Periodontol 1985;56:25-34.

4. Löe H, Brown LJ. Early onset periodontitis in the UnitedStates of America. J Periodontol 1991;62:608-616.

5. Schenkein HA, Van Dyke TE. Early-onset periodontitis:Systemic aspects of etiology and pathogenesis. Perio-dontology 2000 1994;6:7-25.

6. Van Dyke TE, Tohme ZN. Periodontal diagnosis: Evalu-ation of current concepts and future needs. J Int AcadPeriodontol 2000;2:71-78.

7. Irfan UM, Dawson DV, Bissada NF. Epidemiology ofperiodontal disease: A review and clinical perspectives.J Int Acad Periodontol 2001;3:14-21.

8. Schenkein HA. Finding genetic risk factors for perio-dontal diseases: Is the climb worth the view? Periodon-tol 2000 2002;30:79-90.

9. Armitage GC. Classifying periodontal diseases–a long-standing dilemma. Periodontol 2000 2002;30:9-23.

10. Armitage GC. Development of a classification systemfor periodontal diseases and conditions. Ann Periodon-tol 1999;4:1-6.

11. Diehl SR, Wang Y, Brooks CN, et al. Linkage disequilib-rium of interleukin-1 genetic polymorphisms with early-onset periodontitis. J Periodontol 1999;70:418-430.

12. Michalowicz BS, Diehl SR, Gunsolley JC, et al. Evidenceof a substantial genetic basis for risk of adult periodon-titis. J Periodontol 2000;71:1699-1707.

13. Quinn SM, Zhang JB, Gunsolley JC, Schenkein HA, TewJG. The influence of smoking and race on adult perio-dontitis and serum IgG2 levels. J Periodontol 1998;69:171-177.

14. Stevenson J, Bryant T, Greenhorn D, Deakin J, Smith T.Development of factor-score-based models to explainand predict maximal box-lifting performance. Ergonom-ics 1995;38:292-302.

15. Schmidt T, Vogt T, Surmann JP, Kossler W. Glass clas-sification by linear discriminant analysis of LA-ICP-MSdata. Pharmazie 2002;57:396-398.

16. Almasy L, Blangero J. Multipoint quantitative-trait link-age analysis in general pedigrees. Am J Hum Genet 1998;62:1198-1211.

17. Diehl SR, White PS. Cambridge Healthtech Institute’sThird Annual Conference on human genetic variation.Pharmacogenomics 2001;2:79-84.

18. Botstein D, Risch N. Discovering genotypes underlyinghuman phenotypes: Past successes for Mendelian dis-ease, future approaches for complex disease. Nat Genet2003;33(Suppl.):228-237.

30101.qxd 3/10/05 10:41 AM Page 287

288

Quantitative Measures of Periodontitis Volume 76 • Number 2

19. Irfan UM, Dawson DV, Bissada NF. Assessment of famil-ial patterns of microbial infection in periodontitis. J Perio-dontol 1999;70:1406-1418.

20. Carlos JP, Wolfe MD, Kingman A. The extent and severityindex: A simple method for use in epidemiologic stud-ies of periodontal disease. J Clin Periodontol 1986;13:500-505.

21. Hodge PJ, Teague PW, Wright AF, Kinane DF. Clinical andgenetic analysis of a large North European Caucasianfamily affected by early-onset periodontitis. J Dent Res2000;79:857-863.

22. Boughman JA, Beaty TH, Yang P, Goodman SB, WootenRK, Suzuki JB. Problems of genetic model testingin early onset periodontitis. J Periodontol 1988;59:332-337.

23. Long JC, Nance WE, Waring P, Burmeister JA, RanneyRR. Early onset periodontitis: A comparison and evalu-ation of two proposed modes of inheritance. Genet Epi-demiol 1987;4:13-24.

24. Saxen L, Nevanlinna HR. Autosomal recessive inheri-tance of juvenile periodontitis: Test of a hypothesis. ClinGenet 1984;25:332-335.

25. Marazita ML, Burmeister JA, Gunsolley JC, Koertge TE,Lake K, Schenkein HA. Evidence for autosomal domi-nant inheritance and race-specific heterogeneity in early-onset periodontitis. J Periodontol 1994;65:623-630.

26. Badano JL, Katsanis N. Beyond Mendel: An evolvingview of human genetic disease transmission. Nat RevGenet 2992l3:779-789.

27. Kornman K, Duff G, Reilly P. Comment on: A criticalassessment of interleukin-1 (IL-1) genotyping when usedin a genetic susceptibility test for severe chronic perio-dontitis. Greenstein G, Hart TC (2002;73:231-247). J Periodontol 2002;73:1553-1556.

28. Blangero J, Williams JT, Almasy L. Variance compo-nent methods for detecting complex trait loci. Adv Genet2001;42:151-181.

29. Bansal A, van den Boom D, Kammerer S, et al. Associ-ation testing by DNA pooling: An effective initial screen.Proc Natl Acad Sci (USA) 2002;99:16871-16874.

30. Stratford JM. Methods of assessing gingival and perio-dontal disease: A review. Int J Epidemiol 1975;4:79-86.

Correspondence: Dr. Scott R. Diehl, Center for Pharmaco-genomics and Complex Disease Research, New Jersey DentalSchool, UMDNJ, 185 South Orange Ave., MSB C-636, Newark,NJ 07101-1709. Fax: 973/972-0993; e-mail: diehlsr@umdnj.edu.

Accepted for publication June 8, 2004.

30101.qxd 3/10/05 10:41 AM Page 288