D. J. Caplan, J. Kolker, E. M. Rivera & R. E. Walton
Department of Dental Ecology, University of North Carolina, Chapel Hill, NC, USA.
Departments of Operative Dentistry and Endodontics, University of Iowa, Iowa City, IA, USA.

Introduction.
As part of treatment planning, dentists frequently advise when endodontically involved teeth should be extracted rather than root canal treated (RCT) and restored. The decision depends on several factors, including periodontal status and structural integrity of the tooth, the desires of the patient, and the importance of the tooth in the overall treatment plan. Yet if a patient were to ask, ‘How long will I be able to keep the tooth if I elect to undergo root canal therapy?’ or ‘What factors affect my keeping the tooth in the long run, should I choose to save it?’, the provider’s response would be based heavily on anecdotal evidence. Because this treatment planning decision is common and consequences can be time-consuming and costly, there should be scientific evidence to support claims of longevity for RCT teeth. In addition, variables associated with tooth survival should be identified.
Few published articles address survival of RCT teeth (Meeuwissen & Eschen 1983, Sjögren et al. 1990, Jaoui et al. 1995). Variables related to loss of RCT teeth are even less understood; most studies have not employed multivariate regression analyses. One investigation (Eckerbom et al. 1992) reported that, amongst patients seen in a Swedish hospital dental clinic, RCT teeth with preoperative periapical periodontitis, a root filling >2 mm from the radiographic apex, or a screw-type post were more likely lost over a 5–7-year period than were RCT teeth without these features. Caplan & Weintraub (1997) reported that amongst patients enrolled continuously for 8 years in a single dental health maintenance organization in the Pacific North-West of the USA, five factors were predictive of RCT tooth loss over a 6–8-year period: older age, history of facial injury, missing non-third molars at the time of endodontic access, higher plaque level, and fewer than two proximal contacts (PCs) on the RCT tooth at access.
In the Caplan & Weintraub (1997) study, RCT teeth with two PCs at access were almost three times less likely to be extracted during follow-up than RCT teeth missing only a mesial and/or distal contact at access. In reaching their conclusion, the authors employed a case-control design and logistic regression analysis to develop predictive models of tooth loss; the number of PCs at access was the most predictive of all factors analyzed. The present investigation built upon that study by using a retrospective cohort design and proportional hazards regression to develop explanatory models of loss of RCT teeth. The objective was to test the hypothesis that the number of PCs at access is associated with improved survival of RCT teeth, controlling for important preaccess, endodontic and restorative factors.

Materials and methods.
The protocol was approved by the Committee for the Protection of Human Subjects at the University of Iowa College of Dentistry (COD), USA. Study data were collected in 1997 from existing records at the COD.
An existing treatment database identified all permanent teeth receiving root fillings between 1 July 1985 (the database’s first date of operation) and 31 December 1987. This interval allowed identification of an adequate number of cases with the longest potential follow-up. The list was restricted to patients with at least one visit to the COD in each two-year interval from 1985 to 1986 through 1995–96, resulting in 1089 teeth from 734 patients. Teeth were grouped by patient, and patients were listed in random order. A simple random sample of teeth was chosen by selecting the first 400 teeth on the list (corresponding to the first 280 patients).

Table 1. Collected variables and data sources

From computerized databases.
Age (years).
Sex.
a. Insurance/Medicaid billed for root canal therapy.
a. Provider type (for root canal therapy).
a. Clinic.

From treatment notes.
Tooth arch.
Tooth type.
Pulpal diagnosis.
a. Endodontic complication.
a. Time from access to obturation (days).
a. Time from obturation to foundation (days).
a. Intracoronal restoration.
a,b. Crown status (crowned initially after obturation versus crowned at access versus never crowned).
a. Endodontic retreatment/apicoectomy/root amputation.

From patient health questionnaire.
Heart/hypertension medication.
1 medication of any kind.
History of excessive bleeding.
Diabetes.
Interested in keeping teeth.
Ever had orthodontic treatment.
History of injury to mouth or jaws.
Grind or clench teeth.
Gums bleed when brushing teeth.

From preaccess periapical radiograph.
Tooth arch.
Tooth type.
Number of proximal contacts.
Caries.
Periapical lesion on any root.

From immediate postobturation periapical radiograph.
a. Root filling past apex of any root.
a. Root filling >2 mm short of apex in any root.
a. Root filling with lateral/apical voids in any root.
a. Value determined after endodontic access.
(other variables ascertainable prior to access).
b. Crown status at access from preaccess periapical radiograph; crown status after obturation from treatment notes.

Patient records, radiographs, and computerized databases were examined to ascertain variables with potential relationships to tooth loss (Table 1) and to verify that study inclusion criteria were satisfied. A total of 146 teeth did not satisfy study inclusion criteria, including those undergoing endodontic retreatment ( n = 50), those with no documentation of a permanent restoration after obturation ( n = 17), those with no recorded date of access ( n = 11), and those with a miscoded procedure or access date in the database ( n = 4). Teeth that were bridge abutments at access ( n = 28) were excluded because assignment of a value for the main exposure variable was not straightforward. Teeth that became bridge or overdenture abutments after obturation ( n = 30) were also excluded because the number of PCs at access did not represent their postobturation status with respect to stability or occlusion. Finally, third molars ( n = 6) were excluded, since they could not have two PCs. An additional 33 teeth could not be assigned a value for the main exposure variable due to missing radiographs ( n = 28) or charts ( n = 5). Data from the remaining 221 teeth (180 patients) were analyzed.
Proximal contacts were considered absent if the adjacent tooth was a root tip, missing or impacted. The outcome was ‘time to tooth loss’, with follow-up beginning on the date of access. For teeth that subsequently were extracted, follow-up ended on the extraction date. For teeth that were not extracted, follow-up ended on the data collection date. The most recent radiograph of the tooth space was examined to verify extraction (or lack thereof) as documented in the treatment notes.
Range checks were performed for each variable and data sources were re-examined to confirm values of potential outliers. Kaplan–Meier survival estimates for the RCT teeth were generated for the explanatory variables (Kleinbaum 1996), and crude associations between explanatory variables and tooth survival were evaluated via the Log-Rank test (Kleinbaum 1996) using SAS Version 6.12 for Windows (Cary, NC, USA). Multivariate Cox proportional hazards (PH) models (Kleinbaum 1996) then were developed to generate estimates of the association of interest whilst controlling for confounding factors. Because patients could contribute multiple teeth to the dataset, SUDAAN Release 7.11 for Windows (Research Triangle Institute, RTP, NC, USA) was used to obtain appropriate variance estimates (Caplan et al. 1999).
To be eligible for inclusion in multivariate models, covariates were required to have a moderately strong bivariate relationship with tooth survival (Log-Rank P < 0.20), no greater than a 90/10 split in their univariate frequency distributions, and no more than 5% missing values. Mean values were imputed for missing values of eligible covariates. Interaction terms were not tested, and inclusion of time-dependent covariates was not necessary, based on visual verification of PH assumptions using SAS log (–log survival) curves.
In developing the PH models, the first step was to obtain an unadjusted hazard ratio (HR) from a model containing only the main exposure variable. Next, the single covariate that most affected the parameter estimate of interest ( ) was added to the previous model, but only if it changed the main exposure by at least 10%. This process was repeated until the addition of no single covariate elicited a change in the main exposure of at least 10%, controlling for other variables in the model. At this point, included covariates were given the opportunity to be removed from the model, one at a time, if their removal did not elicit a change in the main exposure from the previous model by at least 10%. Allowing covariates to be removed from regression models permitted different combinations of covariates to be included. The goal was to generate the most parsimonious model for which the main exposure approximated that obtained from a model containing all eligible covariates.

The disposition of RCT teeth amongst the 180 patients is presented in Table 2. Subjects had up to four RCT teeth, with 148 (82%) of the subjects contributing only one tooth. Of these 148 subjects, 31 (21%) lost that one tooth during the follow-up period. Amongst the 32 subjects who had two or more RCT teeth, 18 (56%) lost at least one tooth. Overall, 55 (25%) of the 221 RCT teeth were extracted during the follow-up period.
To preserve the stability of the regression models, and because only 55 teeth were extracted during follow-up, models were allowed to contain a maximum of 10 variables (Hosmer & Lemeshow 1989). For each of the 10 factors selected for multivariate analyses, Table 3 presents bivariate results in the form of Kaplan–Meier 5- and 10-year survival estimates. Not controlling for confounding variables, RCT teeth with more PCs at access had significantly better survival than teeth with fewer PCs.

Table 2. Number of patients, by number of RCT teeth included and extracted during follow-up.


Table 3. Bivariate relationships between explanatory variables and RCT tooth survival.


Table 4. Estimates generated during model-building process.


Table 5. Final multivariate Cox proportional hazards regression model for RCT tooth survival.

Table 4 presents s, HRs, and 95% confidence intervals (CIs) for the main exposure variable during each stage of model-building, with the main exposure variable dichotomized as ‘two’ versus ‘zero or one’ PCs. Model 1 contains only the main exposure variable and provides an unadjusted HR of 3.9, implying that RCT teeth with zero or one PC at access were lost at a rate almost four times that of teeth with two PCs. In model 2, the covariate ‘crown status’ was added because its inclusion changed the of the main exposure variable by at least 10% and to a greater degree than any other covariate; this addition results in an adjusted HR of 4.8. In model 3, addition of the covariate ‘tooth type’ decreases the adjusted HR to 3.7, and in model 4, addition of the covariate ‘caries’ reduces the adjusted HR to 3.1. At this point, the addition of no single covariate elicits a change of at least 10% in the of interest, and the removal of no single covariate elicits a change less than 10%, so model 4 is deemed the final model. For comparison, model 5 shows the main exposure with all eligible covariates included. The main exposure is similar for models 4 and 5, indicating minimal confounding of the relationship of interest by the other eligible covariates.
The final model is shown in Table 5. RCT teeth with zero or one PC at access were lost at a rate 3.1 times that of RCT teeth with two PCs (95% CI: 1.9–5.1), controlling for crown status, tooth type, and caries. The model also shows that: (i) teeth that were not crowned were lost at a greater rate than teeth crowned at access, which in turn were lost at a greater rate than teeth crowned after obturation; (ii) second molars were lost at a greater rate; and (iii) teeth that were carious at access were lost at a greater rate.

This hypothesis-testing study aimed to determine whether survival of RCT teeth was influenced by the number of PCs at access. Results are consistent with the previous investigation linking fewer PCs at access with loss of RCT teeth (Caplan & Weintraub 1997). Several explanations for this association are plausible. First, adjacent teeth help distribute occlusal forces over a wider span, potentially reducing the probability of tooth fracture. Secondly, RCT teeth adjacent to an edentulous space are more likely to serve as removable partial denture abutments, which may provide additional mechanical stress. Thirdly, patients with existing edentulous spaces may have a greater history of oral disease, place lesser value on keeping their teeth, or have more financial limitations than other patients, and thus may be less interested in saving a compromised RCT tooth. Previous tooth loss has been shown to predict future tooth loss (Burt et al. 1990, Eklund & Burt 1994), but these studies have not considered the endodontic status of the extracted teeth. One study that did consider endodontic status (Caplan & Weintraub 1997) reported a statistically significant relationship between previous tooth loss and loss of RCT teeth.
The goal of the present analysis was to ascertain the least biased estimate of the association between number of PCs at access and tooth survival. Thus, the enhanced survival of teeth with crowns should not imply that RCT teeth will have better survival if crowns are placed after obturation because the multivariate models did not necessarily include factors that might be related to crown placement. The decision not to crown an RCT tooth might depend on the tooth’s expected life span, e.g. providers may choose a less involved, less expensive restoration because of poor periodontal health. Nevertheless, better survival amongst teeth crowned after obturation is reasonable, since:

1. the tooth’s endodontic landmarks would be preserved at access (reducing the probability of perforation);
2. root canal treatment must have been considered ‘predictably successful enough’ to warrant crown placement;
3. the RCT tooth would be protected from propagation of existing fractures;
4. patients electing to crown a newly RCT tooth might be more dentally educated or financially secure, characteristics associated with tooth survival.

Also intuitive is poorer survival amongst second molars (Table 5), which generally are considered more difficult to treat endodontically due to anatomy and compromised visibility and access. They are more difficult to retreat and might be extracted more frequently, especially if non-functional or in a nonaesthetic area. Molars tend to be extracted more frequently than other teeth (Marcus et al. 1996, Hujoel et al. 1998); a 45-year study of Norwegian men found the mortality rate of second molars to be 1.9 times higher than that of first molars (Hujoel et al. 1998).
The increased mortality amongst RCT teeth with radiographic evidence of caries at access is not surprising. These teeth as a group might be expected to have less sound tooth structure than non-carious teeth, especially interproximally and toward the root surfaces. Further, their periodontal condition might well be worse due to carious extension. One would expect these features to be associated with subsequent tooth loss regardless of endodontic status.
Endodontic success has been addressed previously, and generally is determined using a combination of clinical and radiographic evaluation (Safavi et al. 1987, Sjögren et al. 1990, Smith et al. 1993, Ray & Trope 1995). However, endodontic success may or may not be correlated with tooth loss. RCT teeth can be lost for nonendodontic reasons such as periodontal disease and nonrestorable fracture, whilst many teeth considered to be endodontic failures are not extracted. Loss of RCT teeth, whilst not necessarily related to endodontic failure, may represent a more meaningful outcome to patients, especially if the decision to undergo endodontic therapy was based on expected tooth longevity.
As with all retrospective studies, data quality was dependent upon the accuracy and consistency of existing information. Certain variables likely related to tooth loss were unavailable, such as patients’ socio-economic status (SES). We chose not to estimate the degree of bone loss from periapical radiographs, since films were not standardized in terms of angulation. Also, most RCT teeth presumably had good periodontal support at access or they probably would have been extracted rather than endodontically treated.
The findings are generalizable to other patient populations, but with certain caveats. First, the sample included only patients with at least one dental visit every 2 years during an interval of 10–12 years. This restriction aimed to provide a sample whose dental services during followup were provided solely at the COD. In insured populations, continuous enrollment might serve this purpose (Caplan & Weintraub 1997), but regardless of the research setting, dental treatment might have been obtained elsewhere. Further, the study design did not permit direct assessment of patients’ motivation for maintaining a healthy dentition. However, restricting the sample to those patients with consistent utilization of dental services could be considered a proxy variable for ‘high level of dental education’. A relatively high level of motivation is assumed because the sample included only individuals who presumably elected to save their teeth via root canal treatment rather than opt for extraction.
The study population was treated at one university dental school. These patients might receive different procedures than they would in other settings due to varying philosophies of care and school requirements of students. Additionally, many patients undoubtedly seek treatment at the COD because of reduced fees, and thus may have a lower SES. This would affect our study’s primary conclusion only if the relationship between the number of PCs and tooth survival were influenced by SES, which is not likely; the observed adjusted HR of 3.1 is consistent with the odds ratio of 2.7 found in a group of dentally insured patients (Caplan & Weintraub 1997).
Finally, the sample was restricted to teeth undergoing initial root canal treatment because retreated teeth might have a systematically different survival experience. Survival of retreated teeth might be worse, considering that they already have failed or were destined to fail, which could be associated with future loss. In contrast, survival of retreated teeth might be better, since these patients probably represent a more motivated or higher SES group. The present bivariate analysis supports the latter concept. Of the 221 teeth undergoing initial root canal therapy, 13% subsequently underwent nonsurgical retreatment, root-end resection (apicoectomy), or root resection (root amputation) during the follow-up period; these teeth had better survival than those not subsequently retreated (Table 3).

References.

Bray KK, Williams K (1997) An evidence-based approach to clinical practice guidelines in periodontics. Current Opinion in Periodontology 4 , 96-103.
Burt BA, Ismail AI, Morrison EC, Beltran ED (1990) Risk factors for tooth loss over a 28-year period. Journal of Dental Research 69 , 1126-30.
Caplan DJ, Slade GD, Gansky SA (1999) Complex sampling: implications for data analysis. Journal of Public Health Dentistry 59 , 52-9.
Caplan DJ, Weintraub JA (1997) Factors related to loss of root canal filled teeth. Journal of Public Health Dentistry 57 , 31- 9.
Dodson TB (1997) Evidence-based medicine: its role in the modern practice and teaching of dentistry. Oral Surgery, Oral Medicine, Oral Pathology, Oral Radiology and Endodontics 83 , 192-7.
Eckerbom M, Magnusson T, Martinsson T (1992) Reasons for and incidence of tooth mortality in a Swedish population. Endodontics and Dental Traumatology 8 , 230-4.
Eklund SA, Burt BA (1994) Risk factors for total tooth loss in the United States; longitudinal analysis of national data. Journal of Public Health Dentistry 54 , 5-14.
Hosmer D, Lemeshow S (1989) Applied Logistic Regression. New York: John Wiley & Sons.
Hujoel PP, L?e H, Anerud A, Boysen H, Leroux BG (1998) Fortyfive- year tooth survival probabilities among men in Oslo, Norway [published erratum appears in Journal of Dental Research 1999; 78(3), 815]. Journal of Dental Research 77 , 2020-7.
Jaoui L, Machtou P, Ouhayoun JP (1995) Long-term evaluation of endodontic and periodontal treatment. International Endodontic Journal 28 , 249-54.
Klein SP, Bohannan HM, Bell RM, Disney JA, Foch CB, Graves RC (1985) The cost and effectiveness of school-based preventive dental care. American Journal of Public Health 75 , 382-91.
Kleinbaum D (1996) Survival Analysis: A Self-Learning Text . New York: Springer.
Marcus SE, Drury TF, Brown LJ, Zion GR (1996) Tooth retention and tooth loss in the permanent dentition of adults: United States. 1988-91. Journal of Dental Research 75 , 684-95.
Maryniuk GA, Haywood VB (1990) Placement of cast restorations over direct pulp capping procedures: a decision analytic approach. Journal of the American Dental Association 120 , 183-7.
Meeuwissen R, Eschen S (1983) Twenty years of endodontic treatment. Journal of Endodontics 9 , 390-3.
Ray HA, Trope M (1995) Periapical status of endodontically treated teeth in relation to the technical quality of the root filling and the coronal restoration. International Endodontic Journal 28 , 12-8.
Safavi KE, Dowden WE, Langeland K (1987) Influence of delayed coronal permanent restoration on endodontic prognosis. Endodontics and Dental Traumatology 3 , 187-91.
Sjogren U, Hogglund B, Sundqvist G, Wing K (1990) Factors affecting the long-term results of endodontic treatment. Journal of Endodontics 16 , 498-504.
Smith CS, Setchell DJ, Harty FJ (1993) Factors influencing the success of conventional root canal therapy - a five-year retrospective study. International Endodontic Journal 26 , 321-33.