Material and methods

A detailed report of the study procedures, including enrollment and data collection, has been provided in a previous publication. In brief, the study used a convenience sample of participants between the age of 19 and 85 years enrolled by dentists in the National Dental Practice-based Research Network. To be eligible, participants were required to have at least 1 single, vital posterior tooth with at least 1 observable external crack. Participating dentists were asked to select 1 of these teeth in each patient and to characterize this tooth for 20 eligible participants or as many as they could enroll in 8 weeks, whichever came first. The study was reviewed and approved by the institutional review board of the lead investigators (T.J.H., J.L.F.), as well as by the various institutional review boards that oversee the 6 regions of the network. Informed consent was obtained for all participants enrolled into the study. There were 2 phases to enrollment: a pilot phase with 183 patients from 12 practices from April through July 2014, and a main launch phase that occurred from October 2014 through April 2015. The study instruments were evaluated in the pilot phase to address understandability, coverage, and ease of form completion and were revised for full-study implementation based on feedback from pilot practitioners.

Dentists and their designated practice personnel were trained in data collection using a training manual developed and approved by the study principal investigators (PIs). Data including the presence and type of pain, as well as data on treatment recommendations for participant teeth, were collected at the patient, tooth, and crack levels. Data forms are publicly available at http://nationaldentalpbrn.org/study-results/cracked-tooth-registry.php . Confirmation of tooth vitality of enrolled teeth was with cold (for example, refrigerant or ice), although other methods were used such as air, air-water spray, or electric pulp testing. Spontaneous pain information was obtained from patient report; sensitivity to cold was ascertained using refrigerant, ice, or air-water spray; and pain upon biting was verified by having the patient occlude on a device or instrument placed on the occlusal surface of the cracked tooth. To help patients discriminate between pain (an increased response to the cold or bite assessment) and an ordinary response, dentists were asked to also perform these tests on an unaffected (for example, the contralateral) tooth. Practitioners indicated reason(s) why they recommended the study teeth for treatment from a list of 9 options (with the instruction to check all that apply, plus the option to write in an additional reason). If a practitioner recommended a tooth for restoration, they were asked to specify restoration type (intracoronal, partial crown, or complete crown), placement technique (direct or indirect), and adhesive bonding (yes or no).

Frequencies according to treatment recommendation were obtained, and odds ratios (ORs) were calculated for recommendations of treatment versus monitoring. As the validity of a statistical test depends on independent observations and the model and as the test must reflect the correlation structure of the data to yield valid estimates of variance and valid statistical tests, patients within a practice represent clusters that are often correlated to the outcome being studied. Clustering typically reduces precision of estimation, yielding lower statistical power and wider confidence intervals than studies of equal sample size but without clusters. In a univariable fashion, each patient-, tooth-, and crack-level characteristic was entered into a logistic regression model that used generalized estimating equations (GEEs) adjusted for clustering of patients within the practice and implemented using the SAS procedure for generalized models, with an exchangeable compound symmetric correlation matrix (PROC GENMOD in SAS with the CORR=EXCH option). This approach specified a model in which observations on individual patients seen by a particular practitioner are allowed to be correlated, whereas those from different practitioners are assumed to be independent. This approach removed variability caused by differences among practitioners from the tests for association between the predictor variable and the outcome variable and so uses the appropriate estimate of standard error for statistical tests.

To identify independent associations for recommending that a study tooth be restored versus monitored, all characteristics with P <.05 after adjusting only for clustering of patients within the practice were entered into a full model, followed by backward elimination to remove all variables for which P was ≥.05, using GEE to adjust for clustering. After fitting the final model, all interaction terms were tested for significance at the 5% level. To assess the robustness of findings, regressions were repeated, comparing all definitive treatment recommendations (extraction, endodontics, and restorations) to monitoring only. All ORs and P values were adjusted for clustering of patients treated by the same practitioner with GEE. All analyses were performed using statistical software (SAS v9.4; SAS Institute Inc).

Results

A total of 2858 patients with a posterior cracked tooth were enrolled by 209 practitioners. The mean/median was 14.8/15 patients per practice, and the range was 1 to 20. The distribution of the characteristics that study dentists took into consideration when deciding whether to restore versus monitor a cracked tooth is presented in Figure 1 .

Distribution of factors in dentists’ decision to restore or monitor cracked teeth (N=2836).

A total of 1040 teeth (36%) were recommended for the following treatments: restoration only (998; 96%), endodontic treatment only (9; 0.1%), endodontic treatment and restoration (20; 2%), extraction (2; 0.2%), or noninvasive treatment (for example, occlusal device, desensitizing [11; 1%]). The disposition of the 1018 cracked teeth recommended for restoration is as follows—type of restoration: 357 (35%) intracoronal, 34 (3%) partial crown, and 623 (61%) complete crown; type of placement: direct 452 (45%) and indirect 562 (55%); bonded: yes 624 (62%) and no 380 (38%) ( Fig. 2 ).

Restorative treatment recommendations by restoration type, placement type, and bonding (N=1018).

Virtually all those teeth recommended for indirect placement were to receive complete crowns (N=534; 95%), and the majority of those teeth recommended for direct placement were to receive an intracoronal restoration (N=355; 79%). Similarly, the majority of restorations not recommended for bonding were complete crowns (N=315; 83%). Approximately equal numbers of intracoronal restorations and complete crowns were recommended as bonded restorations.