JJOD 2426 1–10

journal of dentistry xxx (2015) xxx–xxx

Available online at www.sciencedirect.com

ScienceDirect

journal homepage: www.intl.elsevierhealth.com/journals/jden

1
2
3 Review

4 Radiographic caries detection: A systematic review

5 and meta-analysis

6 Q1 Falk Schwendicke *, Markus Tzschoppe, Sebastian Paris

7 Department of Operative and Preventive Dentistry, Charité–Universitätsmedizin Berlin, Germany

article info abstract

Article history: Objectives: This systematic review aimed at evaluating the accuracy of radiographic caries
Received 20 January 2015 detection for different lesions at different locations.
Received in revised form Data: Studies reporting on the accuracy (sensitivity/specificity) of radiographic detection of
13 February 2015 natural primary caries lesions under clinical or in vitro conditions were included. Risk of bias
Accepted 15 February 2015 was assessed using QUADAS-2. Pooled sensitivity, specificity and diagnostic odds ratios
Available online xxx (DORs) were calculated using random-effects meta-analysis. Analyses were performed
separately for occlusal and proximal lesions, with further discrimination between any kind
Keywords: of lesions, dentine lesions, and cavitated lesions.
Bitewing Sources: Electronic databases (Medline, Embase, Cochrane Central) and grey literature were
Caries systematically searched, complemented by cross-referencing from bibliographies.
Dental Study selection: From 947 identified articles, 442 were analyzed full-text. 117 studies (13,375
Radiography teeth, 19,108 surfaces) were included, the majority of them reporting on permanent teeth
and having high risk of bias. The detection of any kind (i.e. also initial) lesions had low
sensitivities (pooled DOR [95% CI]: 0.24 [0.21/0.26] to 0.42 [0.31/0.34]), but moderate to high
specificities (0.70 [0.76/0.84] to 0.97 [0.95/0.98]). For dentine lesions, sensitivities were higher
(from 0.36 [0.24/0.49] for proximal to 0.56 [0.53/0.59] for occlusal lesions), and specificities
ranged between 0.87 [0.85/0.89] and 0.95 [0.94/0.96]. No studies reported on cavitated
occlusal lesions, whilst for cavitated proximal lesions, sensitivities increased above 0.60,
whilst sensitivities remained high (above 0.90).
Conclusions: Radiographic caries detection is highly accurate for cavitated proximal lesions,
and seems also suitable to detect dentine caries lesions. For detecting initial lesions, more
sensitive methods could be considered in population with high caries risk and prevalence.
Clinical significance: Radiographic caries detection is especially suitable for detecting more
advanced caries lesions, and has limited risks for false positive diagnoses. For groups with
high caries risk and prevalence, alternative detection methods with higher sensitivity for
initial lesions might be considered.
# 2015 Published by Elsevier Ltd.
10
89
11
12

* Corresponding author at: Charité Centre for Dental Medicine, Department for Operative and Preventive Dentistry, Aßmannshauser Str.
4-6, 14197 Berlin, Germany. Tel.: +49 30 450 62556; fax: +49 30 450 7562 556.
E-mail addresses: falk.schwendicke@charite.de (F. Schwendicke), markus.tzschoppe@charite.de (M. Tzschoppe),
sebastian.paris@charite.de (S. Paris).
http://dx.doi.org/10.1016/j.jdent.2015.02.009
0300-5712/# 2015 Published by Elsevier Ltd.

Please cite this article in press as: Schwendicke F, et al. Radiographic caries detection: A systematic review and meta-analysis. Journal of
Dentistry (2015), http://dx.doi.org/10.1016/j.jdent.2015.02.009
 JJOD 2426 1–10

2 journal of dentistry xxx (2015) xxx–xxx

13 2.1. Eligibility criteria 68

1. Introduction
We included clinical or in vitro studies reporting on the 69
14 Q2 Screening for dental caries lesions is presumably one of the following items: 70
71
15 most frequent exercises in general dental practice, aiming to
16 detect early caries lesions to then provide non- or micro- - Participants: Humans with primary caries lesions (clinical 72
73
17 invasive treatments and thereby prevent more invasive, studies), or human teeth with primary natural caries lesions 73
74
18 expensive restorative therapy.1–3 The generally used visual- (in vitro studies), which were submitted to caries detection 74
75
19 tactile detection means have only limited accuracy for via radiographic means. Studies investigating teeth with 75
76
20 detecting non-cavitated lesions, especially on proximal secondary lesions/lesions adjacent to restorations, or teeth 76
77
21 surfaces.1 Dentists thus regularly use additional caries with artificially induced lesions were excluded. We per- 77
78
22 detection aids, with (bitewing) radiography likely being the formed separate analyses to account for the potential 78
79
23 most frequent one.3 Alternative methods comprise, for difference between clinical and in vitro settings. Neither 79
80
24 example, fluorescence-aided caries detection, and have been caries prevalence in the population nor lesion depth were 80
81
25 found highly sensitive and, to a lesser degree, specific. These used to exclude studies. Studies should have allowed 81
82
26 methods might eventually be advantageous compared with separate analysis of caries detection in occlusal and 82
83
27 radiography, as they avoid potential harm by ionizing proximal surfaces. 83
84
28 radiation and could be easier to perform with less technical - Index test: Intraoral radiography, i.e. bitewing or peri-apical 84
86
85
29 efforts.4 radiographs. We did not attempt to differentiate between 85
87
30 To assess the accuracy of caries detection methods, the both, as most studies did not clearly report which radio- 86
88
31 yielded proportion of false or true positively or negatively graphic technique was used. 87
89
32 diagnosed lesions, as compared with a ‘‘gold standard’’ like - Reference test: The caries status of the examined surface 88
91
90
33 histologic assessment of dental hard tissues, is assessed, and needed to be assessed using a reference test, i.e. a ‘‘gold 89
92
34 sensitivity (true positives per all diseased surfaces) and specificity standard’’. Used standards were first categorized as (a) 90
93
35 (true negatives per all health sound surfaces) are calculated. By destructive or (b) non-destructive (visual-tactile assessment 91
94
36 evaluating both values, dentists can assess the risks of over- without [occlusal] or with [proximal] tooth separation). 92
95
37 and under-treatment associated with different detection Destructive methods were further divided into histologic, 93
96
38 methods, and can decide which method might be most microradiographic or operative assessment, with the latter 94
97
39 appropriate in which population.5 The latter is relevant when typically performed in clinical studies when invasively 95
98
40 considering the increasing polarization of caries prevalence accessing the lesion. It should be noted that the latter might 96
99
41 between populations, with only few individuals bearing most lead to under-estimation of false negative findings (Wood- 97
100
42 lesions, and large groups of young or adolescent patients ward, 1996). 98
101
43 having only few or no caries lesions at all.6,7 There is, however, - Outcomes: Reported data should allow to construct a 99
103
102
44 great variability in the reported accuracy (sensitivity and diagnostic 2  2 table, i.e. to determine true positive, false 100
104
45 specificity) between studies,1 which might be due to underly- negative, false positive and true negative detections. Studies 101
105
46 ing clinical (lesion depth, dentition, surface location) or which only reported receiver operating characteristics (ROC) 102
106
47 methodological (technical standard of the radiographic index curves (see below) were not included. 103
107
48 test, used reference test, number and experience of exam- 104
108
49 iners) heterogeneity. Systematically compiling the available 2.2. Information sources 105
109
50 data of studies which investigated the accuracy of radio- 106
51 graphic caries detection should allow to assess this accuracy Electronic databases (Medline via PubMed, Embase via Ovid, 110
52 and its potential modification by various confounders. Cochrane Central) were screened for articles published until 111
53 The present review aimed to systematically appraise September 2014, without any further restriction regarding 112
54 studies investigating the accuracy of radiographic caries publication date or language. Diagnostic reviews were 113
55 detection and to meta-analyze them, with separate analyses additionally searched via the Medion database, which was 114
56 for occlusal and proximal lesions. Moreover, we aimed at designed specifically for that purpose (http://www. 115
57 assessing the effects of histologic lesion stage (initial/any mediondatabase.nl). Grey literature was retrieved via open- 116
58 lesions, only dentine lesions, only cavitated lesions). The latter grey.eu. Cross-referencing was performed using the bibliog- 117
59 is of importance, as these stages guide the allocation of raphies of full-text articles.9 118
60 treatments, i.e. act as thresholds for clinical decision making.
61 Last, we investigated if the accuracy of radiographic caries 2.3. Search and study selection 119
62 detection differs between primary or permanent teeth, if it has
63 improved recently due to technical advances of radiographic Electronic searches used a specificity-optimised strategy 120
64 methods, or if it varies greatly even within studies (i.e. based on recommendations for diagnostic reviews9 and a 121
65 between examiners or applied radiographic techniques). similar review.4 A three-pronged approach combining the 122
condition (caries OR carious OR decay), the detection method 123
66 (bitewing OR radiograph OR radiography OR xray OR x-ray OR 124
2. Materials and methods roentgen OR radiology OR radiologic OR radiographic) and the 125
outcomes (detection OR roc OR sensitivity OR specificity OR 126
67 Reporting of this review follows the PRISMA guideline.8 predictive value OR receiver) using Boolean operators, without 127

Please cite this article in press as: Schwendicke F, et al. Radiographic caries detection: A systematic review and meta-analysis. Journal of
Dentistry (2015), http://dx.doi.org/10.1016/j.jdent.2015.02.009
 JJOD 2426 1–10

journal of dentistry xxx (2015) xxx–xxx 3

128 any restrictions or controlled vocabularies (MeSH), was were further evaluated regarding their applicability for the 184
129 employed. Title and abstract of identified studies were review question. Studies which had low risk of bias or low 185
130 independently screened by two calibrated reviewers (MT, concerns regarding applicability in all domains were rated as 186
131 FS) for potential eligibility. Potentially eligible articles were ‘‘low risk’’, all others as ‘‘high or unclear risk’’. 187
132 assessed in full-text by two independent reviewers, and
133 inclusion decided by both reviewers. If required, consensus 2.6. Summary measures and data synthesis 188
134 was obtained by discussion or consultation of a third reviewer
135 (SP). The unit of analysis was the assessed (occlusal or proximal) 189
tooth surface. For summary measures, sensitivity, specificity 190
136 2.4. Data collection and items and diagnostic odds ratio (DOR) were calculated from the 191
primary dataset (mean accuracy). The DOR is the ratio of the 192
137 Data from eligible studies was extracted by one reviewer (MT) odds of a positive test result in the diseased relative to the odds 193
138 using piloted electronic spreadsheets. Data extraction was of positive test result in the non-diseased group,12 and ranges 194
139 first independently repeated by a second reviewer (FS) for a from zero (the worst possible diagnostic value) to infinity, with 195
140 random sample of 25% of studies (30 articles). As no values of one indicating random accuracy and values above 196
141 discrepancies were found, the second reviewer then contin- one indicating useful test performance.12 We used DORs, as 197
142 ued to verify the extracted data, consensus was achieved by they are independent from prevalence and summarize 198
143 discussion. accuracy data in one estimate. 199
144 Data items followed guidelines outlined by the Cochrane Data synthesis was performed separately for different 200
145 Collaboration10 and another source.9 The following items were surfaces (occlusal/proximal) and cut-offs (any kind of lesions, 201
146 collected: study (year, country, setting – clinical or in vitro), lesions extending into dentine, cavitated lesions) using the 202
147 participants, dentition (primary or permanent teeth), lesion DerSimonian Laird estimator, i.e. random-effects meta-anal- 203
148 location (occlusal or proximal), index test (analogue/digital ysis. Within the main analysis (mean reported accuracy in 204
149 radiography, or both/unclear), reference test (visually-tactile each study), we further separately synthesized studies 205
150 or histological/microradiographical/by excavation), lesion performed in clinical studies and those performed in vitro. 206
151 stage as verified by the reference test (any kind of lesions Summary receiver operator characteristic (sROC) curves were 207
152 [both enamel and dentinal lesions, as many studies did not used to present synthesized data, allowing to graphically 208
153 report separately on these lesions], dentine lesions, cavitated display potential sensitivity-specificity trade-offs at different 209
154 lesions), and outcome data (prevalence, accuracy, sensitivity, cut-offs. By integrating the area under the ROC curve (AUC), 210
155 specificity, reported true/false positives/negatives). Note that the diagnostic test performance was summarized, with AUC 211
156 lesion stage (extension) was determined by the used reference values of 1 and 0 indicating best and worst test accuracies, 212
157 tests, which differed between studies. Moreover, for cavitated whilst an AUC of 0.5 indicates a test with random accuracy.13 213
158 lesions, most studies did not only assess the detection of these Moreover, the cut-off point where sensitivity and specificity 214
159 lesions, but also the accuracy of deciding if the lesion was are equal (Q*) was calculated.14 95% confidence intervals (CI) 215
160 cavitated or not. were used as error estimates for DOR and sROC, whilst 216
161 Outcome data was extracted in three datasets: mean, best standard errors (SE) were used for AUC and Q*. 217
162 and worst accuracy. Note that these datasets were built on Cochran’s Q and I2-statistics15 were used to evaluate 218
163 accuracy data reported for different examiners (often with potential heterogeneity between studies, with heterogeneity 219
164 different experiences) within each study, but also by using data being assessed via DORs. Funnel plot analysis and one-sided 220
165 reported for different radiographic detection or evaluation Egger’ regression test were performed to assess small study 221
166 systems etc. Separate extraction and analysis of these effects or publication bias.10,16 To calculate adjusted DORs 222
167 datasets was performed to assess the potential variability of accounting for possible publication bias, we imputed DOR 223
168 radiographic caries detection within each study and to avoid values not missing at random17 and re-assessed pooled DORs. 224
169 extracting data only for a specific (e.g. the first reported or the Meta-regression analyses were used to explore the influence 225
170 best) examiner, which introduces bias. of potential confounders (dentition, year of publication, 226
reference test, risk of bias/applicability concerns) on diagnos- 227
171 2.5. Risk of bias assessment tic accuracy (DORs) using the unrestricted maximum-likeli- 228
hood method. Both uni- and multivariate regression analyses 229
172 Risk of bias was assessed using the QUADAS-2 tool,11 with the were performed. Data was analyzed using MetaDisc 1.4 230
173 following domains being recorded: selection (no inappropriate (Unidad de Bioestadistica Clı́nica del Hospital Ramón y Cajal, 231
174 exclusions, no case-control design, random or consecutive Madrid, Spain) and Comprehensive Meta-Analysis 2.2.64 232
175 inclusion), index test (assessment blinded for and indepen- (Biostat, Englewood, USA). 233
176 dent of reference test, defined cut-offs), reference test (valid
177 reference test, assessment independent from index test), flow 234
178 and timing (sufficient time between index and reference, all 3. Results
179 lesions submitted to same reference and all included in 235
180 analysis). Selection (patients/lesions match review question, 3.1. Results of the search
181 realistic prevalence and lesion spectrum), index test (test,
182 conduct and interpretation match review question) and Using electronic databases, 947 studies were found to be 236
183 reference test (identified condition matches review question) possibly eligible. No ongoing trials were identified. 442 studies 237

Please cite this article in press as: Schwendicke F, et al. Radiographic caries detection: A systematic review and meta-analysis. Journal of
Dentistry (2015), http://dx.doi.org/10.1016/j.jdent.2015.02.009
 JJOD 2426 1–10

4 journal of dentistry xxx (2015) xxx–xxx

238 were analyzed full-text, and 14 further studies retrieved by remaining studies either used both or did not clearly state 260
239 cross-referencing and screened for eligibility. In total, 456 what system was used (Table S2). 261
240 studies were investigated full-text (Fig. 1). 339 studies were Risk of bias or limited applicability was found low in 23 and 262
241 excluded; reasons for exclusion at the full-text-stage can be high in 94 studies. Most frequently, risk of bias was high for the 263
242 found within the appendix (Table S1). Grey literature was not domain of study flow/timing, as most studies performed the 264
243 found eligible. 117 studies were included in the review and index and reference test too soon after each other or did not 265
244 meta-analysis (Table S2). report on this at all (59%). Further bias or concerns stemmed 266
from the used reference standard and risk of selection bias, 267
245 3.2. Included studies and risk of bias oftentimes associated with the prevalence and spectrum of 268
investigated lesions (Table S2). 269
246 Included studies were conducted between 1975 and 2013. All
247 studies were published in English. Overall, 13,375 teeth with 3.3. Data synthesis 270
248 19,108 surfaces were examined. 24 and 93 studies were
249 performed in vivo and in vitro, respectively, with a total of In the primary analysis (mean accuracy dataset), the pooled 271
250 1755 reported patients being subjects of clinical examinations. sensitivity for detecting any kind of occlusal caries lesions was 272
251 54 and 55 studies reported on caries detection on occlusal and 0.35 (95% CI: 0.31/0.40) and 0.41 (0.39/0.44) in clinical and 273
252 proximal surfaces, respectively; 8 studies reported on both. in vitro studies, respectively, whilst the pooled specificity was 274
253 Ninety-two studies investigated caries detection in permanent 0.78 (0.73/0.83) and 0.70 (0.76/0.84). For detecting any kind of 275
254 teeth, 23 in primary teeth, 2 studies reported on caries proximal lesions, sensitivity in clinical and in vitro studies was 276
255 detection in both dentitions. As reference test, 106 studies 0.24 (0.21/0.26) and 0.42 (0.31/0.34), respectively, and specificity 277
256 performed some kind of destructive validation (histologic or was 0.97 (0.95/0.98) and 0.89 (0.88/0.90). Diagnostic Odds Ratios 278
257 microradiographic analysis or excavation); 11 studies per- (DOR) all significantly exceeded 1 (indicating a useful test) and 279
258 formed only visual-tactile validation. 67 and 25 studies used were higher in proximal than occlusal lesions, whilst 280
259 analogue and digital radiographic detection systems; the heterogeneity was moderate (I2 > 50–67%). The area under 281

Fig. 1 – Screening process. Different domains (boxes) of the search sequence were combined using the Boolean operator
‘AND’ (example for Medline search). Searching all databases yielded 947 records (after deduplication). From 456 articles
screened as full-text, 117 studies were included.

Please cite this article in press as: Schwendicke F, et al. Radiographic caries detection: A systematic review and meta-analysis. Journal of
Dentistry (2015), http://dx.doi.org/10.1016/j.jdent.2015.02.009
 JJOD 2426 1–10

journal of dentistry xxx (2015) xxx–xxx 5

282 the ROC curves (AUC) ranged between 0.66 and 0.79, and the

Table 1 – Sensitivity, specificity and accuracy of radiographic caries detection. Separate analysis were performed for any kind of lesions, dentine lesions, and cavitated
caries lesions, either on occlusal or proximal surfaces. Data was further discriminated into reports from clinical and in vitro studies. Pooled sensitivities, specificities and
diagnostic odds ratios (DOR) and their 95% confidence intervals (in parentheses) are given. Heterogeneity was assessed based on DOR and reported as I2 and Cochran’s Q

Q = 26.2 ( p < 0.001)

Q = 113 ( p < 0.001)
Q = 105 ( p < 0.001)
Q = 36 ( p < 0.01)

Q = 41 ( p < 0.05)
283 threshold at which sensitivity and specificity were equal (Q*)

Heterogeneity
284 was between 0.62 and 0.73 (Fig. 1a). The total number of
285 studies used for these analyses ranged from 4 to 40 (Table 1).
286 When evaluating the accuracy of detecting dentine lesions,
287 sensitivities increased compared to the previous analysis, and

I2 = 50%;
I2 = 66%;
I2 = 58%;
I2 = 44%;

I2 = 81%;
288 ranged from 0.36 (0.24/0.49) for proximal lesions to 0.56 (0.53/
289 0.59) for occlusal lesions. Specificity ranged between 0.87 (0.85/

n/a
290 0.89) and 0.95 (0.94/0.96) for detecting dentinal lesions

16.0 (11.5/22.4)

(0.87/0.93) 15.2 (4.81/48.0)

291 (Table 1). DOR values were high (up to 26.1 [12.0/56.7] for

8.3 (6.1/11.3)
6.1 (4.6/8.22
2.4 (1.4/3.9)
292 detecting occlusal lesions under clinical settings), whilst

DOR
293 heterogeneity was variable (I2 = 0–76%). For all but the analysis
of proximal lesion detection in vitro (2 studies), AUC values

In vitro
294

n/a
295 ranged between 0.81 and 0.93, and Q* was between 0.74 and

(0.76/0.84)
(0.88/0.90)
(0.85/0.89)
(0.94/0.96)
296 0.77 (Fig. 1b). The total number of studies used for these

Specificity
297 analyses ranged from 2 to 45 (Table 1).
298 No studies reported on the detection of cavitated occlusal
lesions. For proximal lesions, all except one study18 did not

0.80
0.89
0.87
0.95

(0.53/0.68) 0.90
299

n/a
300 only report on the accuracy of detecting these lesions, but also

(0.39/0.44)
(0.41/0.45)
(0.53/0.59)
(0.41/0.48)
on the accuracy of the decision for or against surface

Sensitivity
301
302 cavitation. All of these studies used radiographic dentine
303 extension as threshold for this decision. The sensitivity for

( p: level of statistical significance of Q-statistics). n/a not available, i.e. less than 3 studies reported on this.

0.41
0.43
0.56
0.45

0.61
n/a
304 detecting cavitated lesions was higher than that for all or
305 dentinal lesions, with sensitivity values above 0.60, whilst

studies
No. of
306 specificities remained high (above 0.90). DOR values were

n/a
19
40
45
24

6
307 consequently high (15.2 [4.8/48.0) and 56.3 [14.0/225] under
308 in vitro and clinical settings, respectively). Heterogeneity was

I2 = 76%, Q = 59 ( p < 0.001)

I2 = 91%; Q = 57 ( p < 0.001)

I2 = 56%; Q = 6.8 ( p > 0.05)
I2 = 67%; Q = 13 ( p < 0.05)
high (I2 > 81%). AUC values ranged between 0.93 and 0.96, and

I2 = 0%, Q = 0.5 ( p > 0.05)

309

Heterogeneity
310 Q* was between 0.86 and 0.91 (Fig. 1c). Six studies were used for
311 Q3 both these analyses (Table 1). In all analyses, pooled
312 sensitivities were lower than pooled specificities (Fig. 2).

313 3.4. Publication bias

n/a
314 Publication bias was assessed using funnel plots for all
26.1 (12.0/56.7)

56.3 (14.0/225)
315 described subgroup analyses (Fig. 3a and b) and Egger’s
2.6 (0.80/8.2)

7.5 (3.4/16.5)
11.3 (3.9/32)

316 regression test. Especially studies performed under in vitro

DOR

317 settings were found prone for publication bias. Imputation of

Clinical

318 potentially missing studies did not significantly change

n/a

319 calculated DOR values except for one analysis (detection of

(0.73/0.83)
(0.95/0.98)
(0.94/0.96)
(0.89/0.97)

(0.97/0.99

320 any kind of lesions in a clinical setting). Egger’s test largely

Specificity

321 confirmed these findings (Fig. 3).

0.78
0.97
0.95
0.94

(0.59/0.70) 0.98

3.5. Variability and meta-regression

n/a

322
(0.31/0.40)
(0.21/0.26)
(0.52/0.59)
(0.24/0.49)
Sensitivity

323 To further assess the variability of accuracy of radiographic

324 detection, we compared the best and worst reported set of
325 sensitivity and specificity values and the respective DOR and
0.35
0.24
0.56
0.36

0.64
n/a

326 heterogeneity (Table S3). For all comparisons, both sensitivity

327 and specificity were higher in the best compared with the
studies
No. of

328 worst accuracy, with overall accuracy (DOR) being 23–123%

n/a
15
5
4

329 higher under the best compared to the worst diagnostic

330 circumstances (e.g. most versus least experienced examiner,
Proximal

Proximal

Proximal
Surface

Occlusal

Occlusal

Occlusal

331 best versus worst detection/evaluation system), indicating

332 large variability of results within studies. Heterogeneities
333 remained similar in these separate analyses of different
of lesions

334 reported accuracies compared to the main analysis.

Cavitated
Any kind

lesions

lesions
Lesions

Dentine

335 Uni- and multivariate meta-regression analyses were

336 performed to assess the impact of potential confounders
337 (year of publication, risk of bias or applicability concerns [low

Please cite this article in press as: Schwendicke F, et al. Radiographic caries detection: A systematic review and meta-analysis. Journal of
Dentistry (2015), http://dx.doi.org/10.1016/j.jdent.2015.02.009
377
376
375
374
373
372
371
370
369
368
367
366
365
364
363
362
361
360
JJOD 2426 1–10

6 journal of dentistry xxx (2015) xxx–xxx

338

specificity were plotted against each other. The area under the curve AUC (in parentheses: SE) and the cut-off, at which sensitivity and specificity are equal (Q*) are shown.
AUC values of 1 and 0 indicate the best and worst possible accuracy, values of 0.5 indicate random accuracy. We separately analyzed radiographic detection of (a) any kind
Fig. 2 – Receiver operating characteristics (ROC) curves. Symmetric ROCs were plotted, the three plots indicate pooled ROC and confidence intervals. Sensitivity and 1-
339
340

caries lesions, (b) dentine lesions, and (c) cavitated lesions. We further discriminated occlusal from proximal lesions, and studies performed clinically and in vitro.
341
340
342
341
340
343
342
341
340
344
343
342
341
340
345
344
343
342
341
340
346
345
344
343
342
341
340
347
346
345
344
343
342
341
340
348
347
346
345
344
343
342
341
340
349
348
347
346
345
344
343
342
341
340
350
349
348
347
346
345
344
343
342
341
340
351
350
349
348
347
346
345
344
343
342
341
340
352
351
350
349
348
347
346
345
344
343
342
341
340
353
352
351
350
349
348
347
346
345
344
343
342
341
340
354
353
352
351
350
349
348
347
346
345
344
343
342
341
340
355
354
353
352
351
350
349
348
347
346
345
344
343
342
341
340
356
355
354
353
352
351
350
349
348
347
346
345
344
343
342
341
340
357
356
355
354
353
352
351
350
349
348
347
346
345
344
343
342
341
340
358
357
356
355
354
353
352
351
350
349
348
347
346
345
344
343
342
341
358
357
356
355
354
353
352
351
350
349
348
347
346
345
344
343
342
341
359
358
357
356
355
354
353
352
351
350
349
348
347
346
345
344
343
342
359
358
357
356
355
354
353
352
351
350
349
348
347
346
345
344
343
342
360
359
358
357
356
355
354
353
352
351
350
349
348
347
346
345
344
343
360
359
358
357
356
355
354
353
352
351
350
349
348
347
346
345
344
343
361
360
359
358
357
356
355
354
353
352
351
350
349
348
347
346
345
344
361
360
359
358
357
356
355
354
353
352
351
350
349
348
347
346
345
344
362
361
360
359
358
357
356
355
354
353
352
351
350
349
348
347
346
345
362
361
360
359
358
357
356
355
354
353
352
351
350
349
348
347
346
345
363
362
361
360
359
358
357
356
355
354
353
352
351
350
349
348
347
346
363
362
361
360
359
358
357
356
355
354
353
352
351
350
349
348
347
346
364
363
362
361
360
359
358
357
356
355
354
353
352
351
350
349
348
347
364
363
362
361
360
359
358
357
356
355
354
353
352
351
350
349
348
347
365
364
363
362
361
360
359
358
357
356
355
354
353
352
351
350
349
348
365
364
363
362
361
360
359
358
357
356
355
354
353
352
351
350
349
348
366
365
364
363
362
361
360
359
358
357
356
355
354
353
352
351
350
349
366
365
364
363
362
361
360
359
358
357
356
355
354
353
352
351
350
349
367
366
365
364
363
362
361
360
359
358
357
356
355
354
353
352
351
350
367
366
365
364
363
362
361
360
359
358
357
356
355
354
353
352
351
350
368
367
366
365
364
363
362
361
360
359
358
357
356
355
354
353
352
351
368
367
366
365
364
363
362
361
360
359
358
357
356
355
354
353
352
351
369
368
367
366
365
364
363
362
361
360
359
358
357
356
355
354
353
352
369
368
367
366
365
364
363
362
361
360
359
358
357
356
355
354
353
352
370
369
368
367
366
365
364
363
362
361
360
359
358
357
356
355
354
353
370
369
368
367
366
365
364
363
362
361
360
359
358
357
356
355
354
353
371
370
369
368
367
366
365
364
363
362
361
360
359
358
357
356
355
354
371
370
369
368
367
366
365
364
363
362
361
360
359
358
357
356
355
354
372
371
370
369
368
367
366
365
364
363
362
361
360
359
358
357
356
355
372
371
370
369
368
367
366
365
364
363
362
361
360
359
358
357
356
355
373
372
371
370
369
368
367
366
365
364
363
362
361
360
359
358
357
356
373
372
371
370
369
368
367
366
365
364
363
362
361
360
359
358
357
356
374
373
372
371
370
369
368
367
366
365
364
363
362
361
360
359
358
357
374
373
372
371
370
369
368
367
366
365
364
363
362
361
360
359
358
357
375
374
373
372
371
370
369
368
367
366
365
364
363
362
361
360
359
358
375
374
373
372
371
370
369
368
367
366
365
364
363
362
361
360
359
358
376
375
374
373
372
371
370
369
368
367
366
365
364
363
362
361
360
359
376
375
374
373
372
371
370
369
368
367
366
365
364
363
362
361
360
359
377
376
375
374
373
372
371
370
369
368
367
366
365
364
363
362
361
360
377
376
375
374
373
372
371
370
369
368
367
366
365
364
363
362
361
360

Please cite this article in press as: Schwendicke F, et al. Radiographic caries detection: A systematic review and meta-analysis. Journal of
Dentistry (2015), http://dx.doi.org/10.1016/j.jdent.2015.02.009
 JJOD 2426 1–10

journal of dentistry xxx (2015) xxx–xxx 7

338 or high/unclear], dentition [primary versus permanent teeth]. fluorescence-based methods in both permanent and primary 395
339 Only the year of publication significantly modified the DOR for teeth. For proximal surfaces of permanent teeth, the accuracy of 396
340 detecting cavitated lesions in proximal surfaces in vitro (R2 radiographic methods is supported by significantly more 397
341 [95% CI]: 0.09 ( 0.13/ 0.05). As mentioned, all except one studies than fluorescence-based detection (only two studies 398
342 study had used the radiographic lesion extension into dentine were included in the most recent review on this matter), whilst 399
343 as threshold to test if cavitation occurred or not. As the for primary teeth, a similar number of studies investigated both 400
344 accuracy of this threshold might have changed over time due radiographic and fluorescence-based detection of proximal 401
345 to changing lesion activity, all studies using this threshold lesions. Again, fluorescence-based techniques seem more 402
346 were submitted to meta-regression as well, with year of sensitive, but significantly less specific, with lower overall 403
347 publication being used as effect modifier. In this case, accuracy. For the threshold of surface cavitation, which greatly 404
348 however, meta-regression did not find a statistically signifi- affects treatment allocation, no systematically compiled data 405
349 cant effect (Fig. 4). for fluorescence-based methods is available, whilst our pooled 406
DOR and AUC values indicate very high accuracy of radiogra- 407
350 phy. When compared to visual-tactile caries detection, radiog- 408
4. Discussion raphy seems superior for detecting enamel, dentine or 409
minimally cavitated proximal lesions, whilst obviously, visu- 410
351 Besides visual–tactile screening for dental caries, radiographic al-tactile detection is well suited for detecting cavitated occlusal 411
352 caries detection is the most frequently used diagnostic tool in or smooth-surface lesions and deep proximal lesions.1 412
353 general dental practice.3 Given that radiography is associated In summary, radiographic caries detection is especially 413
354 with potentially harmful ionizing radiation, and considering suitable for detecting dentine lesions and cavitated proximal 414
355 that an increasing number of alternative techniques are lesions, whilst it is less sensitive, but highly specific for 415
356 available for detecting both initial and advanced occlusal or detecting initial caries lesions. Detecting initial lesions is 416
357 proximal lesions, the accuracy of radiographic caries detection required to allocate non- or micro-invasive treatments (remi- 417
358 needs to be critically appraised to inform both clinicians and neralisation, caries sealing or infiltration) before cavitation 418
359 researchers. It is further relevant if this accuracy differs occurs. Whilst fluorescence-based methods are better suited for 419
360 between lesion stages and locations, as this might guide detecting these lesions, they have lower specificity: this, in turn, 420
361 specific application of this detection methods, or if studies would be unproblematic if the resulting over-treatment would 421
362 performed under different settings report different accuracies, always be non- or micro-, but not invasive therapies.5 Given that 422
363 as this would have an impact on future study design. Given the the majority of dentists do, however, provide restorative 423
364 number of published studies and the variability of their interventions even to lesions radiographically confined to 424
365 findings, a systematic review and meta-analysis was per- enamel,20–24 specificity might be prioritized over sensitivity 425
366 formed to compile and synthesize the available data and thus for detecting initial caries lesions. This is the more true when 426
367 address some of these questions. considering the general decrease in caries prevalence, with only 427
368 The present study is the first to meta-analyze systemati- few (oftentimes socioeconomically) disadvantaged groups 428
369 cally retrieved accuracy data for radiographic caries detection. having the majority of lesions, and many (young) patients 429
370 We found radiography to have limited sensitivity for detecting having no lesions at all6,7,25: the positive predictive value of less 430
371 any kind, i.e. both initial and dentinal, caries lesions on specific methods like fluorescence-based detection will be low 431
372 occlusal and, even more so, proximal surfaces. In contrast, in these low-prevalence populations. For dentine caries lesions 432
373 specificity of radiographic detection was generally high on on occlusal surfaces, radiography seems suitable for detecting 433
374 proximal surfaces, and moderate for occlusal surfaces. For these oftentimes ‘‘hidden’’ lesions.26 Valuable information was 434
375 more advanced lesions extending into dentine, pooled gained by assessing those studies which evaluated the 435
376 sensitivities were moderate on proximal and occlusal sur- diagnostic value of radiographs for deciding cavitation status 436
377 faces. Radiographic detection seems especially suitable for of proximal lesions. As all but one study described this decision 437
378 detecting cavitated caries lesions on proximal surfaces; as being dependent on radiographic lesion extension into dentine, 438
379 both sensitivity and specificity were high and the resulting we performed a meta-regression to assess if the diagnostic 439
380 accuracy was excellent. These findings are in agreement with value of this threshold changed over time. This is relevant as 440
381 previous review results.1,19 When comparing results from lesion activities might have decreased along with the wide- 441
382 studies performed in in vitro versus clinical settings, the spread use of fluorides, with lesions cavitating at later stages 442
383 former might over-estimate sensitivity and under-estimate today than in the past. However, the trend identified by our 443
384 specificity. Moreover, these studies seem more prone to small- analysis remained non-significant. In conclusion, caution is 444
385 study effects or publication bias. required when interpreting the reported probabilities of dentine 445
386 The accuracy of radiographic caries detection methods lesions being cavitated, and in many low risk populations, tooth 446
387 needs to be discussed in comparison with that of alternatives separation and visual–tactile lesion assessment should be 447
388 like fluorescence-aided detection. For occlusal surfaces, the considered before allocating invasive therapies.19,27,28 448
389 detection of initial lesions (which is not the same, but There are a number of further confounders modifying the 449
390 comparable to our assessment of ‘‘all kind of lesions’’, as accuracy of radiographic caries detection methods, and we 450
391 initial lesions are included here), fluorescence-based caries have evaluated several of them using subgroup or meta- 451
392 detection showed significantly higher accuracy, sensitivity and regression analysis. First, most included studies evaluated 452
393 specificity.4 For more advanced occlusal lesions, radiographic detection methods in permanent teeth, and whilst we did not 453
394 detection seems less sensitive, but more specific than find significant accuracy differences between dentitions, this 454

Please cite this article in press as: Schwendicke F, et al. Radiographic caries detection: A systematic review and meta-analysis. Journal of
Dentistry (2015), http://dx.doi.org/10.1016/j.jdent.2015.02.009
493
492
491
490
489
488
487
486
485
484
483
482
481
480
479
478
477
476
JJOD 2426 1–10

8 journal of dentistry xxx (2015) xxx–xxx

455
456
455
457
456
455
458
457
456
455
459
458
457
456
455
460
459
458
457
456
455
461
460
459
458
457
456
455
462
461
460
459
458
457
456
455
463
462
461
460
459
458
457
456
455
464
463
462
461
460
459
458
457
456
455
465
464
463
462
461
460
459
458
457
456
455
466
465
464
463
462
461
460
459
458
457
456
455
467
466
465
464
463
462
461
460
459
458
457
456
455
468
467
466
465
464
463
462
461
460
459
458
457
456
455
469
468
467
466
465
464
463
462
461
460
459
458
457
456
455
470
469
468
467
466
465
464
463
462
461
460
459
458
457
456
455
471
470
469
468
467
466
465
464
463
462
461
460
459
458
457
456
455
472
471
470
469
468
467
466
465
464
463
462
461
460
459
458
457
456
455
473
472
471
470
469
468
467
466
465
464
463
462
461
460
459
458
457
456
473
472
471
470
469
468
467
466
465
464
463
462
461
460
459
458
457
456
474
473
472
471
470
469
468
467
466
465
464
463
462
461
460
459
458
457
474
473
472
471
470
469
468
467
466
465
464
463
462
461
460
459
458
457
475
474
473
472
471
470
469
468
467
466
465
464
463
462
461
460
459
458
475
474
473
472
471
470
469
468
467
466
465
464
463
462
461
460
459
458
476
475
474
473
472
471
470
469
468
467
466
465
464
463
462
461
460
459
476
475
474
473
472
471
470
469
468
467
466
465
464
463
462
461
460
459
477
476
475
474
473
472
471
470
469
468
467
466
465
464
463
462
461
460
477
476
475
474
473
472
471
470
469
468
467
466
465
464
463
462
461
460
478
477
476
475
474
473
472
471
470
469
468
467
466
465
464
463
462
461
478
477
476
475
474
473
472
471
470
469
468
467
466
465
464
463
462
461
479
478
477
476
475
474
473
472
471
470
469
468
467
466
465
464
463
462
479
478
477
476
475
474
473
472
471
470
469
468
467
466
465
464
463
462
480
479
478
477
476
475
474
473
472
471
470
469
468
467
466
465
464
463
480
479
478
477
476
475
474
473
472
471
470
469
468
467
466
465
464
463
481
480
479
478
477
476
475
474
473
472
471
470
469
468
467
466
465
464
481
480
479
478
477
476
475
474
473
472
471
470
469
468
467
466
465
464
482
481
480
479
478
477
476
475
474
473
472
471
470
469
468
467
466
465
482
481
480
479
478
477
476
475
474
473
472
471
470
469
468
467
466
465
483
482
481
480
479
478
477
476
475
474
473
472
471
470
469
468
467
466
483
482
481
480
479
478
477
476
475
474
473
472
471
470
469
468
467
466
484
483
482
481
480
479
478
477
476
475
474
473
472
471
470
469
468
467
484
483
482
481
480
479
478
477
476
475
474
473
472
471
470
469
468
467
485
484
483
482
481
480
479
478
477
476
475
474
473
472
471
470
469
468
485
484
483
482
481
480
479
478
477
476
475
474
473
472
471
470
469
468
486
485
484
483
482
481
480
479
478
477
476
475
474
473
472
471
470
469
486
485
484
483
482
481
480
479
478
477
476
475
474
473
472
471
470
469
487
486
485
484
483
482
481
480
479
478
477
476
475
474
473
472
471
470
487
486
485
484
483
482
481
480
479
478
477
476
475
474
473
472
471
470
488
487
486
485
484
483
482
481
480
479
478
477
476
475
474
473
472
471
488
487
486
485
484
483
482
481
480
479
478
477
476
475
474
473
472
471
489
488
487
486
485
484
483
482
481
480
479
478
477
476
475
474
473
472
489
488
487
486
485
484
483
482
481
480
479
478
477
476
475
474
473
472
490
489
488
487
486
485
484
483
482
481
480
479
478
477
476
475
474
473
490
489
488
487
486
485
484
483
482
481
480
479
478
477
476
475
474
473
491
490
489
488
487
486
485
484
483
482
481
480
479
478
477
476
475
474
491
490
489
488
487
486
485
484
483
482
481
480
479
478
477
476
475
474
492
491
490
489
488
487
486
485
484
483
482
481
480
479
478
477
476
475 Fig. 3 – Funnel plots. Calculated diagnostic odds ratios (DOR) were transformed (logDOR) and plotted on the x-axis against
492
491
490
489
488
487
486
485
484
483
482
481
480
479
478
477
476
475 their standard error (on the y-axis). An asymmetric funnel indicates potential publication or small study bias. We
493
492
491
490
489
488
487
486
485
484
483
482
481
480
479
478
477
476 separately analyzed the accuracy of radiographic caries detection of any kind of lesions, dentine lesions, and cavitated
493
492
491
490
489
488
487
486
485
484
483
482
481
480
479
478
477
476

Please cite this article in press as: Schwendicke F, et al. Radiographic caries detection: A systematic review and meta-analysis. Journal of
Dentistry (2015), http://dx.doi.org/10.1016/j.jdent.2015.02.009
 JJOD 2426 1–10

journal of dentistry xxx (2015) xxx–xxx 9

criteria as well, which greatly affect the measured accuracy. 489

Again, whilst our meta-regression analysis did not find the 490
validation method to significantly affect accuracy, this might 491
be a lack of statistical power rather than a lack of effect. 492
The focus on criterion validity, i.e. accuracy compared with 493
a gold standard, has been criticized before,29 as on the one 494
hand, no consensus has been established regarding such 495
‘‘gold’’ standard, which on the other hand might be irrelevant 496
to patients anyway: predictive validity and, more importantly, 497
the health benefit resulting from the use of a test might be 498
Fig. 4 – Meta-regression analysis. For detecting cavitated more relevant to patients than accuracy itself.5,29 In this sense, 499
proximal caries lesions, the diagnostic odds ratio (DOR) the focus of this review on criterion validity is certainly a 500
was regressed on the year of publication to assess a limitation. A further limitation is the quality of reporting of the 501
potential change in the diagnostic value of the used validation process itself, and the generally low quality of many 502
threshold (lesions radiographically extending into included studies certainly affects the strength of our conclu- 503
dentine). Meta-regression could not confirm such change sions: the examination of unrepresentative populations of 504
(p > 0.05). The slope of the regression line (R2) and 95% teeth or problematic study flow and timing (no blinding, too 505
confidence intervals are shown. In vitro studies are short times between index and reference test, only one 506
additionally indicated by black rings. examiner) could impact on the accuracy results. Clustering of 507
lesions in one mouth remains another limitation (but is reality 508
in clinical dentistry as well), and could lead to calculated 509
455 should not be confused with a proven lack of effect, but rather confidence intervals being artificially narrow. The difference 510
456 a potentially lack of statistical power. Second, the used of used detection and evaluation systems is a limitation, as we 511
457 technical system of radiographic caries detection and evalua- were unable to account for these in subgroup analysis given 512
458 tion might affect accuracy. We used year of publication as their variety and incomplete reporting. The number of studies 513
459 surrogate to assess this confounder, assuming that more used for different subgroup analyses varied greatly, too, and 514
460 recent studies used different (digital, not analogue) detection most studies were performed in vitro rather than clinically. 515
461 systems. However, the effect of publication year on accuracy This is understandable, as validation remains a severe 516
462 might be confounded itself by more recent studies employing problem clinically, with validation via excavation potentially 517
463 more rigorous study designs (reduced risk of bias, different neglecting false negative results, whilst other means are 518
464 settings, lesion spectra, or reference tests). We used both uni- problematic (like visual–tactile assessment) and potentially 519
465 and multivariate meta-regression for evaluating this, with risk unavailable (histological assessment is often only performed 520
466 of bias and year of publication being both used as independent for premolars or 3rd molars scheduled for extraction, i.e. 521
467 variables. The absence of significant effects in most analyses unrepresentative teeth). In this regard, it should be highlight- 522
468 seems to confirm previous reports on newer systems not ed that this and a previous review did not find radiographic 523
469 necessarily yielding higher accuracies.19 However, these accuracy to greatly differ between clinical and in vitro 524
470 meta-regression findings could also be due to stricter study studies.1 Last, this review itself introduces limitations. Our 525
471 designs or more representative lesion prevalence in more search might not have been comprehensive enough, especial- 526
472 recent than older studies. ly as accuracy studies are inadequately indexed.30 Data 527
473 Differences in study design, index tests, or experience and extraction for accuracy studies is prone to introduce bias 528
474 number of examiners have been used before to explain the given the inadequate reporting in most studies; this bias might 529
475 reported heterogeneity between studies,1 which we confirm as have been aggravated by the performed confirmative extrac- 530
476 well. This heterogeneity, in turn, indicates a potentially tion. Last, we did not focus on secondary, but only primary 531
477 lacking reliability as well. We assessed this reliability by caries. Given that secondary caries or caries adjacent to 532
478 synthesizing the best and worst accuracy datasets reported restorations is a main justification for providing invasive 533
479 within each study, and found accuracy (DOR) to vary up to interventions, further research in this direction is required. 534
480 123%. It should be highlighted that this might be due to different In conclusion, current evidence finds radiographic caries 535
481 degrees of examiners’ experience, which is known to affect accuracy,1 detection relatively insensitive, but highly specific for detect- 536
482 but also due to different detection systems (analogue versus ing initial occlusal and proximal lesions. Sensitivity, specificity 537
483 radiographic etc.) being used. As reliability is another important and the overall accuracy are significantly higher for lesions 538
484 parameter when evaluating a detection method, future extending into dentine or, for proximal lesions, those with 539
485 studies should use standardized designs which reduce cavitated surfaces. The majority of included studies had high 540
486 between-study variability, and should report the accuracies risk of bias or their findings were not fully applicable in general 541
487 of different examiners (with different experiences) in more detail. dental practice, oftentimes due to their in vitro setting. 542
488 Moreover, study design standards should include validation However, within the limitations of our statistical evaluation, 543

lesions. Lesions were further divided into occlusal (a) and proximal (b) lesions. For 3 analyses, less than 3 studies were
available and construction of funnel plots impossible (n/a). White circles: Published DORs, black circles: imputed DORs.
White diamond: Calculated pooled DOR based on published studies, black diamond: pooled DOR including imputed values.

Please cite this article in press as: Schwendicke F, et al. Radiographic caries detection: A systematic review and meta-analysis. Journal of
Dentistry (2015), http://dx.doi.org/10.1016/j.jdent.2015.02.009
 JJOD 2426 1–10

10 journal of dentistry xxx (2015) xxx–xxx

544 this did not seem to generally affect accuracy measures. 11. Whiting P, Rutjes A, Westwood M, Mallett S, Deeks J, 600
545 Future studies should not aim at merely expanding this data, Reitsma J, et al. QUADAS-2: a revised tool for the quality 601
assessment of diagnostic accuracy studies. Annals of Internal 602
546 but should employ more standardized designs in more
Medicine 2011;155:529–36. 603
547 representative study populations. Blinding of examiners should
12. Glas AS, Lijmer JG, Prins MH, Bonsel GJ, Bossuyt PMM. The 604
548 be sought and maintained, with sufficiently long period between diagnostic odds ratio: a single indicator of test performance. 605
549 index and reference test conductance. The inclusion of different Journal of Clinical Epidemiology 2003;56:1129–35. 606
550 examiners with different degrees of experience and the explicit 13. Hanley JA, McNeil BJ. The meaning and use of the area 607
551 reporting of their results is recommendable. The used cut-offs should under a receiver operating characteristic (ROC) curve. 608
552 be described in more detail, and studies should aim at standardizing Radiology 1982;143:29–36. 609
14. Walter SD. Properties of the summary receiver operating 610
553 these cut-offs as well as the used reference test. In this sense, core
characteristic (SROC) curve for diagnostic test data. Statistics 611
554 outcome sets for diagnostic accuracy studies for caries detection might in Medicine 2002;21:1237–56. 612
555 be desirable. Last, both clinicians and researchers should focus 15. Higgins JPT, Thompson SG. Quantifying heterogeneity in a 613
556 not only on accuracy (as measured against a potentially meta-analysis. Statistics in Medicine 2002;21:1539–58. 614
557 unavoidable or irrelevant gold standard), but the resulting 16. Egger M, Smith GD, Schneider M, Minder C. Bias in meta- 615
558 health outcomes in different populations. analysis detected by a simple, graphical test. British Medical 616
Journal 1997;315:629–34. 617
559
17. Duval S, Tweedie R. Trim and fill: a simple funnel- 618
560 plot-based method of testing and adjusting for publication 619
Acknowledgements bias in meta-analysis. Biometrics 2000;56:455–63. 620
18. Sansare K, Singh D, Sontakke S, Karjodkar F, Saxena V, 621
561 This study was funded by the authors’ institutions. Frydenberg M, et al. Should cavitation in proximal surfaces 622
be reported in cone beam computed tomography 623
examination? Caries Research 2014;48:208–13. 624
562
19. Wenzel A. Bitewing and digital bitewing radiography for 625
Appendix A. Supplementary data
detection of caries lesions. Journal Dental Research 626
2004;83:C72–5. 627
563 Supplementary data associated with this article can be 20. Gomez J, Ellwood RP, Martignon S, Pretty IA. Dentists’ 628
564 found, in the online version, at http://dx.doi.org/10.1016/j. perspectives on caries-related treatment decisions. 629
566
565 jdent.2015.02.009. Community Dental Health 2014;31:91–8. 630
21. Gordan VV, Garvan CW, Heft MW, Fellows JL, Qvist V, Rindal 631
DB, et al. Restorative treatment thresholds for interproximal 632
567 references primary caries based on radiographic images: findings from 633
the Dental Practice-Based Research Network. General 634
Dentistry 2009;57. 654–63; quiz 64–6, 595, 680. 635
568 1. Bader J, Shugars D, Bonito A. Systematic reviews of selected 22. Kakudate N, Sumida F, Matsumoto Y, Manabe K, Yokoyama 636
569 dental caries diagnostic and management methods. Journal Y, Gilbert GH, et al. Restorative treatment thresholds for 637
570 of Dental Education 2001;65:960–8. proximal caries in dental PBRN. Journal Dental Research 638
571 2. Gordan VV, Garvan CW, Richman JS, Fellows JL, Rindal DB, 2012;91:1202–8. 639
572 Qvist V, et al. How dentists diagnose and treat defective 23. Traebert J, Wesolowski CI, de Lacerda JT, Marcenes W. 640
573 restorations: evidence from the dental practice-based Thresholds of restorative decision in dental caries 641
574 research network. Operative Dentistry 2009;34:664–73. treatment among dentists from small Brazilian cities. Oral 642
575 3. Rindal DB, Gordan VV, Litaker MS, Bader JD, Fellows JL, Qvist Health & Preventive Dentistry 2007;5:131–5. 643
576 V, et al. Methods dentists use to diagnose primary caries 24. Vidnes-Kopperud S, Tveit AB, Espelid I. Changes in the 644
577 lesions prior to restorative treatment: findings from The treatment concept for approximal caries from 1983 to 2009 645
578 Dental PBRN. Journal of Dentistry 2010;38:1027–32. in Norway. Caries Research 2011;45:113–20. 646
579 4. Gimenez T, Braga MM, Raggio DP, Deery C, Ricketts DN, 25. Schwendicke F, Dörfer C, Schlattmann P, Foster Page L, 647
580 Mendes FM. Fluorescence-based methods for detecting Thomson M, Paris S. Socioeconomic inequality and caries: a 648
581 caries lesions: systematic review, meta-analysis and systematic review and meta-analysis. Journal Dental Research 649
582 sources of heterogeneity. PLOS ONE 2013;8:e60421. 2015;94:10–8. 650
583 5. Schwendicke F, Stolpe M, Meyer-Lueckel H, Paris S. 26. Ricketts D, Kidd E, Weerheijm K, de Soet H. Hidden caries: 651
584 Detecting and treating occlusal caries lesions: a cost- what is it? Does it exist?. Does it matter? International Dental 652
585 effectiveness analysis. Journal Dental Research 2014. http:// Journal 1997;47:259–65. 653
586 dx.doi.org/10.1177/0022034514561260. 27. Ratledge DK, Kidd EA, Beighton D. A clinical and 654
587 6. Marthaler T, O’Mullane D, Vrbic V. The prevalence of dental microbiological study of approximal carious lesions, 655
588 caries in Europe 1990–95. Caries Research 1996;30:237–55. Part 1: the relationship between cavitation, radiographic 656
589 7. Marthaler TM. Changes in dental caries 1953–2003. Caries lesion depth, the site-specific gingival index and the 657
590 Research 2004;38:173–81. level of infection of the dentine. Caries Research 2001;35:3–7. 658
591 8. Moher D, Liberati A, Tetzlaff J, Altman D. Preferred reporting 28. Wenzel A, Hintze H, Horsted-Bindslev P. Accuracy of 659
592 items for systematic reviews and meta-analyses: the radiographic detection of residual caries in connection with 660
593 PRISMA statement. PLoS Medicine 2009;6:e1000097. tunnel restorations. Caries Research 1998;32:17–22. 661
594 9. Deville W, Bezemer P, Bouter L. Publications on diagnostic 29. Baelum V, Heidmann J, Nyvad B. Dental caries paradigms in 662
595 test evaluation in family medicine journals: an optimal diagnosis and diagnostic research. European Journal of Oral 663
596 search strategy. Journal of Clinical Epidemiology 2000;53:65–9. Sciences 2006;114:263–77. 664
597 10. Higgins JPT, Green S, editors. Cochrane Handbook for 30. Centre for Reviews Dissemination. Systematic Reviews: 665
598 Systematic Reviews of Interventions. Version 5.10 (updated March CRD’s guidance for undertaking reviews in health care. 666
599 2011). The Cochrane Collaboration; 2011. York: York Publishing Services; 2009. 667

Please cite this article in press as: Schwendicke F, et al. Radiographic caries detection: A systematic review and meta-analysis. Journal of
Dentistry (2015), http://dx.doi.org/10.1016/j.jdent.2015.02.009