Characteristics of the participants and dental surfaces included
Initially, 252 children were included from March 19, 2014 to November 25, 2015. From these, 216 were followed-up until 24 months (attrition rate of 14.3%).
One hundred and six children from the visual inspection group and 110 from the radiographic method group finished the study (p =0.108, by chi-square test) in the mail trial. Full data was published elsewhere, including the flow chart and participants’ data at baseline [35].
Since only occlusal and proximal surfaces of deciduous molars that did not receive any type of treatment, or were submitted to non-operative or restorative treatment at the baseline were considered for this study’s purpose, we analyzed 4,383 dental surfaces (66.7% of proximal and 33.3% of occlusal surfaces) in 1,461 molars, being 720 first (49.3%) and 741 second (50.7%) deciduous molars of 216 children. One hundred and eight (50.5%) were male and 106 (49.5%) female, 107 (50.0%) were 3 or 4 years old and 107 (50.0%) were 5 or 6 years old. Moreover, 97 (45.3%) children had a dmf-s from 0 to 3, and 117 (54.7%) children presented a dmf-s higher than 3.
Caries detection and treatment conducted considering all types of treatments
Figure 2 shows the decision tree with the diagnosis and subsequent dental treatment performed using, firstly, visual inspection alone and then adding the radiographic assessment. Both methods were coincident for most surfaces (almost 70% of all surfaces). Considering sound surfaces, initial and more advanced caries lesions in the assessment, the radiographic method underestimated the diagnosis and treatment decision made by visual inspection in around 25% of the surfaces, and overestimated the treatment decision in 4.5% of the surfaces (Figure 2). Underestimation occurred mainly in surfaces with initial caries lesions according to the visual inspection (indicative of non-operative treatment), but for which the radiographs did not present any radiolucency (Figure 2, pathway b).
We found only 121 dental surfaces (less than 3% of all surfaces examined) classified as sound (an indication of non-local treatment) by visual inspection, but with a radiolucency in enamel or dentin in the radiographs (Figure 2, pathway a, and Figure 3). This situation has been advocated as one of the advantages of taking bitewings in the clinical practice, since non-operative treatments could be performed to avoid caries lesion progression and cavitation. From these 121 surfaces, non-operative treatment was performed in 41 surfaces, while 64 surfaces did not receive any type of treatment (16 surfaces were restored). Besides the low occurrence of this situation, the frequency of failures (cavitation during the follow-up) of the untreated surfaces was 18.8%, while failures in the surfaces submitted to non-operative treatment occurred in 19.5% (OR = 1.05; 95%CI = 0.35 to 3.09) (Figure 3).
From the 1,357 surfaces presenting initial caries lesions detected by visual inspection, radiographs did not show radiolucencies in 1,079 dental surfaces (Figure 4). Extensive caries lesions detected by visual inspection were observed in 548 dental surfaces, and from these, only in 42 surfaces the radiographic examination did not present radiolucencies reaching the dentin (Figure 5).
Caries detection related to the decision for operative treatment
The decision tree on Figure 6 is related to the diagnosis made by different strategies in relation to the decision of operative treatment performed on occlusal and proximal surfaces of deciduous molars. This emphasis was given due to the assertion that radiographs are useful to detect caries lesions missed by visual inspection, which is the main reason to justify the use of the radiographs as a protocol for caries diagnosis in all children. Here, the vast majority of surfaces presented coincident results between both methods (more than 96%) (Figure 6) when considering the indication for operative treatment in dental surfaces of deciduous molars. Discordances were observed in only 3.7% of the assessed surfaces. Radiographs indicated operative treatment more frequently than visual inspection in around 2.8% of cases (Figure 6, pathway b). In around 1.0%, however, bitewings did not show radiolucency in surfaces classified as decayed visually (Figure 6, pathway a).
For the results that would be reached with the use of the simultaneous association of methods, from non-operative with visual inspection to operative treatment with the radiographic method, changes in treatment decision would occur only in 121 surfaces (Figure 6, pathway b). From these, 65 were not restored, and 23 failed (35.4%). Other 56 surfaces were restored, and 22 restorations (39.3%) needed to be replaced during the follow-up. No significant differences were observed when comparing these frequencies (OR=1.10; 95%CI=0.12 to 10.43).
On the other hand, if the results from both diagnostic methods had been considered as a sequential association, where bitewings would be used to confirm a positive result obtained by visual inspection, 42 dental surfaces (39 in the true-positive branch and 3 in the false-positive branch) would not have been indicated for operative treatment due to a negative result on the radiographic assessment. From these, 41 surfaces were restored and the failure rate was 19.5% (8 replaced restorations during the follow-up) (Figure 6, pathway a).
Occurrence of false-positive results
Figure 6 also shows the occurrence of false-positive results obtained by the use of both caries detection methods. A false-positive result was recorded when a dental surface was submitted to operative treatment, but when the absence of carious soft dentin after opening was observed. Consequently, all surfaces classified as false-positive results were restored.
We observed a total of 45 surfaces with false-positive results (1.02% considering all included surfaces). From these surfaces, 25 (55.6% of all false-positives) were diagnosed as positive for both methods (Figure 6, pathway a). In 3 surfaces (6.7% from the false-positives), the decision for operative treatment was reached only by visual inspection (Figure 6, pathway a), and in 17 surfaces (37.8%) the result was positive only with radiographic method (Figure 6, pathway b).
Evidence of overdiagnosis
The occurrence of overdiagnosis was estimated by assessing surfaces indicated for operative treatment by both detection methods, that were not restored and did not progress during the follow-up. Since the main clinical trial was designed to compare the simultaneous association of visual inspection with radiographic examination versus the visual inspection performed alone, only 4 dental surfaces, positively diagnosed at this threshold by visual inspection, were not restored due to issues in following the outlined treatment plan. Three of these surfaces progressed in the subsequent two years (Figure 6, pathway a), indicating a low probability of overdiagnosis with visual inspection.
On the other hand, 65 dental surfaces indicated for operative treatment by the radiographic assessment were not restored (Figure 6, pathway b). From these, 42 surfaces (64.6%) did not require operative interventions during the follow-up (Figure 6, pathway b), which can be seen as an estimative of the radiographic method´s overdiagnosis on dental surfaces of deciduous molars with non-obvious clinical signs of caries lesions.
Factors associated with the necessity of new operative interventions during the follow-up
According to the multilevel logistic regression analysis, when caries lesions, indicated for operative interventions, were detected with the radiographic method and missed by visual inspection, the occurrence of failures during the follow-up was significantly higher than when detected by both methods (Table 1). On the other hand, when the same type of lesions was detected by visual inspection, but not confirmed by the radiographic assessment, the occurrence of new operative treatments during the follow-up was similar to that when detected by both methods (Table 1). Evidently, when both methods were coincident in classifying a dental surface as sound (no intervention needed), the occurrence of new treatments was significantly lower in comparison to that when both methods detected lesions requiring restorations. These trends were observed in the univariate analysis, as well as in the multiple analysis adjusted by possible confounding variables. Moreover, when a possible mediator in the multiple model was added, same trend was observed (Table 1).
A higher frequency of new operative interventions would be expected for cases where caries lesions, supposedly requiring operative treatment, were not detected by visual inspection but presented radiolucency reaching the dentin on radiographs. Part of these findings could have been related to defective restorations, and/or the incidence of new caries lesions missed at visual inspection. Therefore, and to evaluate if a restoration performed at baseline could exert any influence on the occurrence of new interventions during the follow-up in such cases, a mediation analysis was conducted. A significant and direct effect between the result obtained by the diagnostic strategy and the occurrence of new operative interventions was observed (Figure 7a). This effect, however, remained significant after the inclusion of the mediator in the model, and the Sobel test indicated that the mediation effect of performing a restoration at the baseline was not statistically significant (Figure 7b). This fact was probably due to the similar failure rates found for non-restored and restored surfaces: 35.4% of non-restored surfaces needed a restoration during the follow-up, and 39.3% of restorations needed to be replaced, respectively.
Evidence of lead-time bias
Despite the similar failure rates when comparing non-restored and restored dental surfaces with caries lesions detected only by radiographic assessment, the method could indicate the occurrence of lead time bias. To test such occurrence, a survival analysis using Cox regression for multiple-failure-time was conducted considering the 121 surfaces with caries lesions detected only by the radiographic method. When t0 was set at children´s birth date, a higher probability of failures for dental surfaces restored at the beginning of the study was observed (HR = 9.92; 95% CI = 5.78 to 17.02, p < 0.001). This trend can be clearly observed in figure 8. Thirty-four surfaces, restored at the beginning did not present failures throughout the study. However, other 18, also restored at the beginning of the study failed 22 times within the follow-up period. In relation to non-restored surfaces at the beginning of the study, 42 surfaces remained with no obvious cavities after 24 months (overdiagnosis made by the radiographic method). In addition, other 23 restored surfaces failed in 12 occasions (Figure 8). This analysis reflects the occurrence of lead-time bias when therapeutic decisions are taken based on the simultaneous association of both caries detection strategies.
When radiographs brought real benefits for caries diagnosis in preschool children
Despite the previously described issues, the diagnosis process made by associating visual inspection and radiographic findings could have presented some benefits.
Eight teeth were submitted to endodontic treatment during the follow-up. From these, in 5 teeth, caries lesions requiring operative intervention were detected by both methods and consequently restored at the baseline. One tooth was restored based on the radiographic assessment, this restoration failed and the tooth was subsequently submitted to endodontic treatment. The remaining two teeth presented caries progression reaching the pulp and needed endodontic treatment. In both cases, the caries lesions on proximal surfaces were overlooked by visual inspection, but the radiolucency was radiographically present. Therefore, these two teeth (corresponding to 6 dental surfaces) would benefit from the diagnosis made by the simultaneous association of both methods: visual and radiographic.
Moreover, 10 teeth were extracted due to caries related reasons. Two of them were extracted as a consequence of a failed endodontic treatment (already considered in the previous paragraph). Five were restored as indicated by both methods. For the remaining three, the presence of caries lesions was not observed by both methods. Thus, the radiographic method would not have a therapeutic impact compared to the visual inspection performed alone.
The radiographic method would also be beneficial for caries diagnosis in preschool children if the visual inspection presented false-positive results not confirmed by radiographs. This situation actually occurred in 3 surfaces (Figure 6, pathway a). In other dental surfaces, an operative treatment decision was reached following a positive result (true positive) clinically observed but absent in the radiographic assessment. Eight of these surfaces required restoration replacements within the follow-up period (Figure 6, pathway a).
The real benefits of the radiographic method could be observed in 17 dental surfaces (0.39% of all 4,383 surfaces) when considering the abovementioned possible scenarios.
Other aspects could also be considered as advantages of the radiographic method, although such benefits are not too evident. In 23 dental surfaces, no operative treatment was decided after visual inspection, while the radiographs presented positive results (Figure 6, pathway b). However, these surfaces presented new caries lesions within the follow-up (17 surfaces) or were submitted to endodontic treatment, as described previously (6 surfaces). The benefit in these 17 surfaces is not too clear because this occurrence could be characterized as lead-time bias, as stated before.
Another situation concerns the restored surfaces that presented a positive result with the radiographic method but a negative through visual inspection. In this sample, 19 surfaces with these characteristics did not failed. This is not a clear benefit since part of these lesions could be cases of overdiagnosis.
Therefore, considering an optimistic estimative of the benefits of the radiographs for caries detection in deciduous molars, a total of 53 (6 + 3 + 8 + 17 + 19) dental surfaces (1.21% of all surfaces examined) possibly would have benefited from the radiographic method used in association with the visual inspection.
Although proximal surfaces have been pointed out as the type of surface that would have more benefits with the use of radiographs, a similar trend was observed compared to the total sample. At proximal and occlusal surfaces, possible benefits of the radiographic method were also observed in 1,21% and 1.81% of these surfaces, respectively. The decision trees related to the operative treatment divided by proximal and occlusal surfaces are presented as supplemental material (supplementary appendixes A and B).