658

Prognosis Versus Actual Outcome.II. The Effectiveness of ClinicalParameters in Developing anAccurate PrognosisMichael K. McGuire* and Martha E. Nunnf

The assignment of prognosis is one of the most important functions undertaken inclinical practice, yet there is little evidence to support the current decision-makingprocess which is based on an outdated model of disease etiology and progression.This study evaluated 100 treated periodontal patients (2,484 teeth) under maintenancecare for 5 years, with 38 of these patients followed for 8 years, to determine therelationship of assigned prognoses to the clinical criteria commonly used in the de-velopment of prognosis. The method of generalized estimating equations (GEE) forcorrelated data was utilized to determine the relationship of each clinical factor to theassignment of initial prognosis, improvement in prognosis at 5 years, and worseningin prognosis at 5 years. A multiple linear regression model was constructed for pre-dicting initial prognosis based on initial clinical data. Increased probing depth, moresevere furcation involvement, greater mobility, unsatisfactory crown-to-root ratio, mal-positioned teeth, and teeth used as fixed abutments resulted in worse initial prognoses.The coefficients from this model were able to predict accurately the 5-year and 8-yearprognoses 81% of the time. When teeth with "good" prognoses were excluded, thepredictive accuracy dropped approximately 50%. Multiple logistic regression modelsindicated that improvement in prognoses and worsening in prognoses were bothstrongly associated with initial probing depth, initial furcation involvement, initialtooth malposition, and smoking when adjusted for initial prognosis. In addition, goodhygiene was found to increase the probability of improvement in prognosis whileinitial mobility was found to decrease the likelihood of improvement in prognosis.Neither of these factors was found to be significant in worsening of prognosis. Smok-ing decreased the likelihood of improvement by 60% and doubled the likelihood ofworsening in prognosis at 5 years. The results of this study indicate that some clinicalfactors used in the assignment of prognoses are clearly associated with changes inclinical condition over time. The data also demonstrated that the traditional approachfor assigning prognoses is ineffective for teeth with an initial prognosis of less thangood. Since most periodontally involved teeth are compromised, further work shouldinclude the development of a more effective method for assigning prognoses that isbased on clear, objective clinical criteria. J Periodontol 1996;67:658-665.Key Words: Tooth survival; decision making; periodontal diseases/diagnosis; prog-nosis; treatment outcome; tooth loss; risk factors; dental models.

Very few studies have evaluated the process for the as-signment of an accurate prognosis. Recently, interest inthe subject has increased, but close inspection of the lit-erature reveals that researchers focus more on risk factors

Private practice, Houston, TX; Department of Periodontics, Universityof Texas, Dental Branch, Houston; and Health Science Center, San An-tonio.'University of Washington, School of Dentistry, Department of DentalPublic Health Sciences and Department of Biostatistics, Seattle, WA.

than on prognostic factors.1 "9 "Risk factors" are definedas those patient characteristics associated with the devel-opment of the disease in the first place; "prognostic fac-tors" are defined as those characteristics that may predictthe outcome once the disease is present but do not actu-ally cause it.10 Although some models will allow one toevaluate risk factors in an epidemiological sense, the realchallenge lies in the development of a model that willpredict outcome on an individual patient basis.11

Volume 67Number 7 MCGUIRE, NUNN 659

Table 1. Commonly Taught Clinical Factors Used in AssigningPrognosisIndividual Tooth Prognosis

Percentage of bone lossDeepest probing depth (in mm)Horizontal or vertical bone lossDeepest furcation involvement: 0, 1, 2, 3Mobility: 0, 1, 2, 3Crown-to-root ratio: favorable or unfavorableRoot form: favorable or unfavorableCaries or pulpal involvement: yes or noTooth malposition: yes or noFixed or removable abutment: yes or no

Overall PrognosisAgeSignificant medical history (smoker and/or diabetic):

yes or noFamily history of periodontal disease (mother, father, sibling):

yes or no and whomHygiene: good, fair, poorCompliant: yes or noMaintenance interval": 2 months; 2 months alternate; 3 months;

3 months alternateParafunctional habit with biteguardParafunctional habit without biteguard

Even though the assignment of prognosis is one of themost important functions undertaken in clinical practice,no clear guidelines have yet been established. In fact,practitioners have handed down their particular approachfor assigning prognosis from one generation to the nextwithout investigating evidence to support the decision-making process. Most periodontal textbooks routinely in-clude a chapter on prognosis which reflects the belief thatthe practitioner must consider many factors about thetooth, dentition, and individual (Table 1) to arrive at aprognosis, but how one arrives at that prognosis is notclear. The process is further confounded by therapists em-phasizing factors based on their idiosyncratic judgmentand past experience. As important as clinical intuition andjudgment are, the evidence-based decision-making pro-cess has demonstrated that subjective variables may resultin the clinician's overestimating the efficacy of a partic-ular approach. The ability to assign a correct prognosiscan be improved through the incorporation of unbiasedquantitative data.'213

The first paper in this series14 evaluated 100 treatedperiodontal patients (2,484 teeth) under maintenance carefor 5 to 8 years to determine the accuracy of assignedprognosis based on the commonly taught clinical criteriafound in Table 1. The results indicated that the ultimatefate of teeth initially labeled as hopeless varied substan-tially and, even though the average prognosis of teethstudied at each interval remained relatively stable overtime, individual prognosis categories and individual toothprognoses changed frequently. Possible reasons for theseshifts were discussed. In conclusion, it was found thatprojections utilizing the factors listed in Table 1 were in-

Table 2. Definitions of Various Prognoses Good Prognosis (one or more of the following): Control of the eti-

ologic factors and adequate periodontal support as measured clinicallyand radiographically to assure the

,

tooth would be relatively easy tomaintain by the patient and clinician assuming proper maintenance.

Fair Prognosis (one or more of the following): Approximately 25%attachment loss as measured clinically and radiographically and/orClass I furcation involvement. The location and depth of the furcationwould allow proper maintenance with good patient compliance.

Poor Prognosis (one or more of the following): 50% attachment losswith Class II furcations. The location and depth of the furcationswould allow proper maintenance, but with difficulty.

Questionable Prognosis (one or more of the following): Greater than50% attachment loss resulting in a poor crown-to-root ratio. Poor rootform. Class II furcations not easily accessible to maintenance care orClass III furcations. 2+ mobility or greater. Significant root proximity.

Hopeless Prognosis: Inadequate attachment to maintain the tooth. Ex-traction performed or suggested.

effective in predicting outcomes other than good, and thatprognoses tended to be more accurate for single rootedteeth than for multirooted teeth. These results were re-cently confirmed in a study reported by Ghiai and Bis-sada.15

The purpose of the present study is to further evaluatethe data derived from the longitudinal investigation usedin the previous article, to explore the relationship of eachclinical factor in Table 1 to their prognosis assignmentand, if possible, to determine which clinical parametersare the most important in developing an accurate prog-nosis.

MATERIALS AND METHODSAs reported earlier, 100 consecutive patients with at least5 years of maintenance care were selected from one clin-ician's appointment book over a 2-month period.14 All hadbeen initially diagnosed as having chronic generalizedmoderate to severe adult Periodontitis and were treated bythe same clinician. Patients in the study were under main-tenance regimens of 2- or 3-month intervals with the ma-jority under a 3-month interval and reflected many of thecharacteristics observed in "well maintained" patients.16Additional information regarding the study population,therapy, and assignment of prognoses can be found in theinitial report.14 The prognostic criteria used in this seriesof studies are described in Table 2.

Determining the Actual OutcomeTeeth lost during the initial active phase of periodontaltherapy were documented, along with the prognosis as-signed each tooth following active therapy and prior tomaintenance care. The same set of criteria were used forassigning prognoses at 5 and 8 years. Subsequent prog-noses were determined by charted clinical data accumu-lated between initial and 5 years and 5 years and 8 years,rather than on information recorded only at the 5-year and8-year examinations. A more accurate projection of prog-nosis was intended by this method. All assessments were

660 PROGNOSIS FOR PERIODONTAL OUTCOME: EFFECTIVENESS OF CLINICAL PARAMETERSJ Periodontol

July 1996

blind to previous assessments and conducted by the sameexaminer. The prognoses initially, at 5 years, and at 8years were then compared.Statistical MethodsStatistical analyses were conducted using SAS and S Plusstatistical software packages. Prognosis was used to fa-cilitate the exploration of which clinical factors impact atooth in such a way as to change its prognosis over aperiod of time. Prognosis in this setting could be consid-ered to be a measure of the tooth's overall clinical ap-pearance at a particular point in time. To evaluate therelationship of each clinical factor (Table 1) to the as-signment of prognosis, a series of simple linear regres-sions were performed using the method of generalizedestimating equations (GEE) as outlined by Zeger and Li-ang.17 This method was utilized since the data were cor-related and ordinary least squares evaluations would beinvalid. Prognoses were given numerical values from 1 to5 for the initial prognoses and 1 to 6 for later prognoseswith a score of 1 representing a good prognosis and ascore of 6 representing a hopeless tooth that was extractedduring the maintenance phase. The initial prognoses wereregressed on each of the clinical factors individually.Clinical factors that were considered included age at thebeginning of the study; history of smoking; history ofdiabetes; patient compliance; history of parafunctionalhabit; use of a biteguard; percent overall bone loss; deep-est probing depth; type of bone loss (horizontal or verti-cal); degree of furcation involvement (0 for none up to 3for the most involved); mobility (0 for no mobility up to3 for the greatest mobility); crown-to-root ratio (satisfac-tory or unsatisfactory); root formation (satisfactory or un-satisfactory); endodontic involvement; presence of caries;root proximity (satisfactory or unsatisfactory); tooth mal-position (satisfactory or unsatisfactory); fixed abutment;removable abutment; oral hygiene (good, fair, poor), andoverall family history of periodontal disease (0 for nohistory, 1 for history in sibling, mother, or father). In allcases an exchangeable correlation structure was assumed.

To discover which clinical factors were related to im-proved prognosis (and by interference, improved condi-tion), only teeth that had a prognosis of less than goodwere included. The response variable considered was adichotomous variable that was assigned 1 if the tooth im-proved in prognosis and 0 otherwise. To evaluate whatclinical factors were related to worse prognosis (and byinference, worse condition), only teeth that had a prog-nosis of better than questionable were fitted in the model.The response variable considered was a dichotomousvariable that was assigned 1 if the tooth worsened inprognosis and 0 otherwise. The method of GEE was uti-lized to fit a series of simple logistic regressions to de-termine which clinical factors were significantly associ-ated with improved prognoses and worsening prognoses.

The clinical factors mentioned above were considered foranalysis. Improvement in prognosis and worsening inprognosis were measured from initial to 5-year prognoses.

Multiple linear regression and logistic regression mod-els were constructed including all clinical factors thatwere significantly correlated (a s 0.10) with initial prog-nosis, improved prognosis at 5 years, and worseningprognosis at 5 years for the respective models. Insignifi-cant variables were then dropped one at a time until allvariables left in the model were significant at the 0.10level of significance. The method of GEE was also usedto fit the multiple regression models.

Multiple logistic regressions for improved prognosisand worsening prognosis incorporated initial prognoses.Indicator variables for the different levels of initial prog-nosis .were used (poor, questionable, and hopeless for theimproved prognosis model and fair and poor for the wors-ening prognosis model). A forward step-wise approachwas utilized with indicator variables for the prognosis en-tered initially and variables added one at a time until asignificance level of 0.10 was achieved for everyvariable in the model.

To assess the validity of the multiple logistic regressionmodels, specificity and sensitivity were calculated. Spec-ificity is the proportion of teeth that did not improve thatwere correctly predicted not to improve by the model.Sensitivity is the proportion of teeth that improved thatwere correctly predicted to improve by the model. Spec-ificity and sensitivity were calculated with different pre-dicted probabilities used to categorize predicted improve-ment and predicted worsening in prognosis in order tojointly maximize sensitivity and specificity based on thedata. Sensitivity and specificity were calculated for im-provement and worsening at 5 years to obtain an upperestimate of the accuracy of the constructed models. Tofurther test the validity of the models, parameter estimatesobtained from the models for improved prognoses at 5years and worsening prognoses at 5 years were then usedto calculate sensitivity and specificity of improvement at8 years and worsening at 8 years based on 5-year data.

RESULTS

The Relationship of Clinical Factors to PrognosisFactors that were significantly correlated (a < 0.05) withinitial prognosis included history of periodontal diseasein a sibling, history of parafunctional habit, percent over-all bone loss, deepest probing depth, presence and sever-ity of furcation involvement, tooth mobility, unfavorablecrown-to-root ratio, unfavorable root formation, toothmalposition, and fixed prosthesis abutment. Unfavorableroot proximity was also marginally significant (P value= 0.054). Those with a history of periodontal disease ina sibling had proportionally more teeth with worse prog-noses when compared to those with no history of peri-

Volume 67Number 7 MCGUIRE, NUNN 661

Table 3. Multiple Regression on Initial Prognosis Table 4.5 Years

Initial Clinical Factors Affecting Change in Prognosis at

Parameter EstimateStandard

Error Robust Value

InterceptProbing depthFurcationCrown/root ratioMobilityMalposedFixed abutment

0.6150.1430.1920.1080.2680.3320.125

0.05310.01360.03690.04530.07880.17950.0693

11.5710.495.202.373.401.851.81

662 PROGNOSIS FOR PERIODONTAL OUTCOME: EFFECTIVENESS OF CLINICAL PARAMETERSJ Periodontol

July 1996

Table 5. Multiple Logistic Regression on Improved Prognosis at 5Years Based on Initial Clinical Factors

Clinical Factor EstimateStandard

Error ValueOddsRatio

InterceptPoor*Questionable*Hopeless*Probing depth*Furcation*Mobility*SmokingGood hygieneMalposed*

2.5652.0052.4590.944

-0.312-0.401-0.340-0.873

0.918-1.121

0.35680.33860.58260.60320.06090.10830.14960.33140.35720.5598

Volume 67Number 7 MCGUIRE, NUNN 663

should also be undertaken to better understand their re-lationships to tooth loss.

As previously reported, the study population had manyof the characteristics of Hirschfeld and Wasserman's"well-maintained" group.16 The only exception was that40% of our group had parafunctional habits and of those,only 40% wore biteguards. The oral hygiene pattern forthe group was rather typical for a periodontal maintenancepatient, with the great majority of patients having fair oralhygiene (20% good, 66% fair, 14% poor). Almost one-half of the patients (47%) were on a 3-month alternatingrecall; one fourth (25%) were seen exclusively in the per-iodontist's office on a 3-month interval; the remainingpatients were split between a 2-month recall exclusivelyin the periodontist's office (17%) and a 2-month alternat-ing recall (10%). The study population was found to bevery compliant, probably the result of the previously de-scribed selection process. In order to be included in thestudy, all patients had to be in maintenance for at least 5years and, presumably, most of the non-compliant pa-tients had dropped out by that time. Therefore, extrapo-lation of data from this study would be most valid whenapplied to other "well-maintained" periodontal mainte-nance patients.

The first paper in this series14 concluded that an accu-rate prognosis was more difficult for clinicians to assignfor teeth with an initial prognosis of less than good andsuggested that most good prognoses tend to remain good.This current report not only reinforces the earlier conclu-sion, but also demonstrates that overall accuracy of themodel at 5 years dropped by nearly 50% when teeth with"good" prognoses were excluded. At 8 years, the accu-racy dropped even more when "good" prognoses wereexcluded, from 81% to 35%. Clearly, these data challengethe effectiveness of the traditional approach to the as-signment of prognosis. A coin toss would be an easierand more accurate way for the clinician to assign a prog-nosis under traditional guidelines, if the initial prognosisis less than good. Surprisingly, the model was slightlymore accurate for assigning prognosis to molars than fornon-molars when teeth with "good" prognoses were ex-cluded. This unexpected result perhaps can be explainedby the proportionately greater number of anterior teethexcluded because of their good prognoses. In addition,more clinical factors such as "furcation involvement" ap-ply to molars than anterior teeth when the traditional ap-proach is used.

The previous analysis14 revealed that prognoses oftenchanged over time, and changed most frequently in teeththat had less than "good" prognoses initially. Statisticalanalysis revealed that several factors were significantlyrelated, either positively or conversely, to the probabilityof improvement in prognoses at 5 years. In particular,increased probing depth initially, more severe furcationinvolvement initially, more mobility initially, and mal-

position of a tooth initially were all associated with adecreased probability of improvement. Initial hopelessprognoses did not significantly differ from initial fairprognoses in terms of the likelihood of improvement.Ironically, teeth with poor prognoses initially were almostseven and a half times as likely to improve when com-pared with teeth with fair prognoses initially, while teethwith questionable prognoses initially were nearly 12 timesas likely to improve in prognoses when compared to teethwith fair prognoses initially. With the exception of teethwith hopeless prognoses, it appears that teeth with worseprognoses initially are more likely to improve. It was alsoshown that smoking decreased the likelihood of improve-ment by 60%. Good hygiene was found to increase thelikelihood of improvement by about two and a half timesthat of teeth that had fair or poor hygiene (which did notdiffer significantly from each other in terms of improve-ment in prognoses).

Probability of worsening in prognoses was also eval-uated using a multiple logistic regression model. In-creased initial probing depth, more severe initial furcationinvolvement, malposition of a tooth initially, initial un-satisfactory root form, and initial endodontic involve-ment, history of smoking and diabetes were all associatedwith the probability that the prognosis of a tooth wouldworsen at 5 years. Fair prognoses initially and poor prog-noses initially had a substantially lower probability ofworsening in prognoses at 5 years when compared toteeth with good prognoses initially. Specifically, teethwith fair prognoses initially were only 17% as likely toworsen in prognosis compared with teeth with good prog-noses initially, and teeth with poor prognoses initiallywere only 2% as likely to worsen in prognosis when com-pared to teeth with initial good prognoses. In other words,teeth that are already compromised tend not to get muchworse under maintenance care. Smoking and diabetesboth doubled the likelihood of worsening in prognosis at5 years. Interestingly, hygiene level and mobility did notappear to significantly affect the probability of worseningin prognosis at 5 years, possibly because teeth with mo-bility and exposed to poor hygiene were initially assigneda worse prognosis. It is also possible that the professionalmaintenance that the patients received overcame any neg-ative effects of poor personal oral hygiene.

Interesting observations can be made between the datareported in this communication and a paper recently pub-lished by Ghiai and Bissada.15 Although their study pop-ulation seems to closely parallel the one in this paper,comparisons should be viewed cautiously because of dif-ferences between the two studies in both the definition ofprognoses and which clinical factors were evaluated. Tak-ing that into account, they also found (when evaluatingall teeth) that it was more difficult to accurately predictprognoses for posterior teeth than anterior teeth. Whenevaluating their success of assigning an accurate prog-

664 PROGNOSIS FOR PERIODONTAL OUTCOME: EFFECTIVENESS OF CLINICAL PARAMETERSJ Periodontol

July 1996

nosis to all teeth in their study over 5 to 13 years, theyfound that their model was 38% successful. (They did notevaluate the success of their model on teeth whose initialprognoses was less than good.) In their study, the onlyfactors that showed statistical significance in their effortsto develop a predictive model were mobility, oral hy-giene, and alveolar bone score. Mobility was the mostimportant clinical parameter in predicting prognosis re-gardless of tooth type. Wheeler et al. also found that mo-bility was the most important clinical parameter whencombined with other microbiologie and immunologicalrisk factors for predicting alveolar bone loss in elderlypopulations.18 Wang et al. reported in 1994 that teeth withfurcation involvement and mobility had significantlymore attachment loss during the maintenance period,which one might assume equates with a worsening prog-nosis.19 With the exception of the probability of worsen-ing in prognosis, our models also demonstrated the im-portance of mobility in the development of prognosis.Perhaps the reason mobility is such an important clinicalparameter is that this one indicator provides a good over-view of many other parameters such as the amount ofattachment loss, the stability of the occlusion, and pos-sibly the presence of parafunctional habits.

A number of recent studies20-23 have implicated smok-ing as an important risk factor in periodontal disease. Asmentioned earlier, our study indicated that smoking de-creased the likelihood of improvement of prognoses by60% and it doubled the likelihood of worsening in prog-nosis at 5 years.

The data from this study seem to indicate that the uti-lization of all of the factors listed in Table 1 to assign aprognosis may not be necessary. It appears that cliniciansshould weigh certain criteria (increased probing depth,more severe furcation involvement, greater mobility, un-satisfactory crown to root ratio, malposed teeth, smoking,and teeth used as fixed abutments) more heavily in theassignment of a prognosis when evaluating the populationas a whole. The problem, however, with some of thesesignificant clinical variables (increased probing depth,more severe furcation involvement, and greater mobility)is that they are a reflection of the progression of the dis-ease process and not useful predictors for the assignmentof prognosis because the variables themselves do not existuntil the downward shift in prognosis occurs. On the otherhand, the other four significant clinical factors (unsatis-factory crown-to-root ratio, malposed teeth, smoking, andteeth used as fixed abutments) are available when the as-signment of prognosis occurs and the data would seem toindicate that more weight should be placed on their pres-ence than the other commonly taught clinical parametersin Table 1.

Further work needs to be accomplished to validate theeffectiveness of the predictive models outlined in this pa-per, especially as they relate to teeth whose initial prog-

noses is less than "good." In addition, studies need to beconducted to determine which, if any, microbial or im-munological factors may also be important in the assign-ment of an accurate prognosis, especially in non-adultforms of the disease.

The goal of this paper was to evaluate the effectivenessof commonly taught clinical parameters in the develop-ment of an accurate prognosis. The summation of evi-dence from this study demonstrates that it is possible topredict prognoses using clinical parameters alone and,furthermore, that some clinical factors appear to be moreimportant than others in the assignment of prognosiswhen the entire population is evaluated. The data indicatethat a model which includes initial probing depth, furca-tion involvement, mobility, unsatisfactory crown-to-rootratio, malposed position of the tooth, and utilization ofthe tooth as a fixed abutment is surprisingly accurate inpredicting actual prognoses at 5 (81%) and 8 years (81%).It appears, however, that this same approach for the as-signment of prognosis is ineffective for teeth with an ini-tial prognosis of less than "good." This is disappointingbecause the reality of clinical practice is that even afterour best therapeutic efforts, many teeth that have sufferedthe ravages of periodontal disease cannot be brought backto a pristine condition and thus have less than a "good"prognosis. For these teeth, more effective methods for theassignment of prognoses are needed.

REFERENCES1. Leknes KN, Lie T, Selvig KA. Root grooves: A risk factor in peri-

odontal attachment loss. J Periodontol 1994;65:859-863.2. Beck JD. Methods of assessing risk for Periodontitis and developing

multifactorial models. J Periodontol 1994;65:468-478.3. Michalowicz BS. Genetic and heritable risk factors in periodontal

disease. J Periodontol 1994;65:479-488.4. Newman MG, Kornman KS, Holtzman S. Association of clinical

risk factors with treatment outcomes. J Periodontol 1994;65:489-497.

5. Wolff LE Dahlen G, Aeppli D. Bacteria as risk factors for Perio-dontitis. J Periodontol 1994;65:498-510.

6. Lamster IB, Smith QT, Celenti RS, Singer RE, Grbic JT. Develop-ment of a risk profile for periodontal disease: Microbial and hostresponse factors. J Periodontol 1994;65:511-520.

7. Hart TC, Shapira L, Van Dyke TE. Neutrophil defects as risk factorsfor periodontal disease. J Periodontol 1994;65:521-529.

8. Oliver RC, Tervonen T. DiabetesA risk factor for Periodontitis inadults? J Periodontol 1994;65:530-538.

9. Bakdash B. Oral hygiene and compliance as risk factors in Perio-dontitis. J Periodontol 1994;65:539-544.

10. Laupacis A, Wells G, Richardson S, Tuguell P. Users' guide to themedical literature: How to use an article on prognosis. JAMA1994;272:234-237.

11. Kornman KS. Risk modeling for periodontal disease. Proceedingsof the Conference on Risk Assessment in Dentistry. University ofNorth Carolina, June 2-3, 1989. Chapel Hill, NC: UNC Press;1990.

12. McGuire MK, Newman MG. Evidence-based periodontal treatment.I. A strategy for clinical decisions, lnt J Periodontics RestorativeDent 1995;15:3-15.

13. Newman MG, McGuire MK. Evidence-based periodontal treatment

Volume 67Number 7 MCGUIRE, NUNN 665

II. Predictable regenerative treatment. Int J Periodontics RestorativeDent 1995;15:116-127.

14. McGuire MK. Prognosis versus actual outcome: A long-term surveyof 100 treated periodontal patients under maintenance care. J Per-iodontol 1991;62:51-58.

15. Ghiai S, Bissada NF. The reliability of various periodontal param-eters for predicting the outcome of periodontally treated teeth. JDent Res 1994;73(Spe. Issue): 164(Abstr. 499).

16. Hirschfeld L, Wasserman B. A long-term survey of tooth loss in600 treated periodontal patients, J Periodontal 1978;49:225-237.

17. Liang K, Zeger S. Longitudinal data analysis using generalized lin-ear models. Biometrika 1986;73:13-22.

18. Wheeler TT, MeArthur WP, Magnusson I, et al. Modeling the rela-tionship between clinical, microbiologie, and immunologie param-eters and alveolar bone levels in an elderly population, J Periodontol1994;65:68-78.

19. Wang HL, Burgett FG, Shyr Y, Ramfjord S. The clinical influenceof molar furcation involvement and mobility on further clinical peri-odontal attachment loss. J Periodontol 1994;65:25-29.

20. Holm G. Smoking as an additional risk factor for tooth loss. J Per-iodontol 1994;65:996-1001.

21. Bergstrom J, Preber H. Tobacco user as a risk factor. J Periodontol1994;65:545-550.

22. Grossi SG, Zambn JJ, Ho AW, et al. Assessment of risk for peri-odontal disease. I. Risk indicators for attachment loss. J Periodontol1994;65:260-267.

23. Tonetti MS, Pini-Prato G, Cortellini P. Effect of cigarette smokingon periodontal healing after GTR in intrabony defects. A prelimi-nary retrospective study. J Clin Periodontol 1995;22:229-234.

Send reprint requests to: Dr. Michael K. McGuire, Suite 102, 3400South Gessner, Houston, TX 77063.

Accepted for publication December 27, 1995.