Secondary caries

Secondary caries is a carious lesion at the margin of an existing restoration. On a bitewing it shows as a radiolucency at the restoration's edge, most often the gingival margin: 80% to 90% of clinically diagnosed lesions are reported there. Restorative materials less radiopaque than enamel look dark on the same image and can be misread as the lesion.

Synonyms: recurrent caries; caries around restorations; recurrent decay

What is secondary caries?

It is primary caries in a new address. Mjör and Toffenetti's review put it plainly: secondary caries "is the same as primary caries located at the margin of a restoration," and on the limited literature available it appears to be a localized lesion similar or identical to primary caries, most often localized gingivally on restorations [3]. The formal definition follows the same line — secondary or recurrent caries has been defined as "lesions at the margins of existing restorations," or caries associated with restorations or sealants, abbreviated CARS in the ICCMS criteria [22].

That matters because it sets your expectations. You are not looking for a special radiographic entity with its own rules. You are looking for the same radiolucency you would look for on an unrestored surface, in a place where a radiopaque object is sitting on top of it.

The diagnosis carries more weight than almost any other call you make on a bitewing. Mjör and Gordan describe the clinical diagnosis of secondary caries as the main reason for replacement of all types of directly-placed restorations — and in the same sentence call it "an ill-defined clinical diagnosis both in teaching programs and in general practice" [4]. The finding that most often sends a restoration to be replaced is one the profession has not agreed how to define.

What it looks like

Look gingivally first.

Mjör's practice-based study collected 1,100 sites of clinically diagnosed secondary caries, marked by private practitioners on schematic diagrams of the restoration. Overall, 80% to 90% of those sites were reported gingivally, irrespective of the type of restoration or the restorative material used [2]. Read that figure for what it is: it records where dentists said they found lesions, not where a histological standard confirmed them. It still tells you where to put your eyes.

When Class II restorations were analysed on their own, occlusal secondary caries made up a slightly higher relative proportion on resin composite (16%) than on amalgam (8%) [2]. So the occlusal margin is not empty — it is just the minority report.

The shape is contained. Mjör and Gordan describe secondary caries lesions as usually localized and delineated [4]. A diffuse, gradient wash of grey along a whole margin is not the classic picture; a defined lucency at one point on the margin is.

What it's confused with

A restoration overhang belongs in the same visual neighbourhood, and it is common enough to keep meeting: 96 of 682 filled interdental sites (14.1%) in a study of Swiss Army recruits carried overhanging margins [19], while an older study of patients treated by final-year dental students found 62% of approximal restorations had them [20]. The two figures are not comparable populations, and none of the sources retrieved for this entry states a radiographic discriminator between an overhang and a marginal lucency. The overhang entry covers it.

How reliable is radiographic detection of secondary caries?

Less reliable than the confidence with which it gets called, and the honest answer has a caveat attached that most summaries drop.

Brouwer and colleagues' systematic review and meta-analysis screened 1,179 studies and included 23. Visual assessment (11 studies), radiographic assessment (13 studies) and laser fluorescence (8 studies) had similar sensitivities, in a band from 0.50 to 0.59, with specificities from 0.78 to 0.83 [1]. That band spans the three methods together — it is not radiography's own figure, and no pooled radiography-only sensitivity appears in the review's abstract.

Now the caveat. The same review states that most of the included studies were performed in vitro, on permanent posterior teeth, and had high risk of bias or applicability concerns, and that the analyses seemed to suffer from publication bias [1]. Every accuracy number below is from extracted teeth:

• 296 restored surfaces from a university tooth bank, read against histology: radiographic assessment had sensitivity 0.68, specificity 0.93 and accuracy 0.81 for sound versus any lesion. Visual assessment on the same specimens reached sensitivity 0.80, specificity 0.85 and accuracy 0.83. Radiographic sensitivity fell to 0.61 for amalgam and 0.64 for composite restorations [10].
• 60 extracted molars with Class II composites, read on photostimulable phosphor plates by four observers: digital radiography had sensitivity 0.591 ± 0.093 and specificity 0.891 ± 0.083 [11].
• 74 extracted primary molars with Class I composites: two observers reading radiographs achieved sensitivities of 48.39 and 51.61 with specificity 79.07 [14]. Primary dentition, so this does not transfer to a permanent posterior bitewing.

The pattern across all three is the same: specificity holds up, sensitivity does not. You will miss lesions that are there more often than you will invent lesions that aren't.

What is missing from the literature retrieved for this entry is a chairside figure — an in vivo sensitivity for a dentist reading a real patient's bitewing. The closest thing is a convolutional neural network trained on 425 bitewings from 383 patients across seven general dental practices in the Netherlands, which reached an F1 score of 0.719 for secondary caries and an area under the FROC curve of 0.804 [12]. That is an algorithm's performance on clinical images, not a clinician's.

Why it happens

The intuitive story — the restoration leaks, bacteria track down the gap — does not survive contact with the review literature. Mjör and Toffenetti concluded that narrow gaps, crevices, ditches and "microleakage" do not lead to secondary caries, but that wide voids may [3]. Their hedge is worth keeping: may, on a body of literature they describe as meagre.

What follows from their position that secondary caries is the same disease as primary caries [3] is that it is driven by the same things primary caries is. A review of 34 clinical trials with at least five years of follow-up named secondary caries — "related to the individual caries risk" — and fracture as the main long-term reasons posterior composites fail [15].

What it means clinically

Secondary caries lesions are the main late complication of dental restorations, limiting their life span and generating costs through repeated reintervention [1]. A retrospective study following 683 restorations for up to 33 years recorded 353 failures, a success rate of 48%, with fracture and secondary caries the main reasons [16].

So the call you make on the image feeds directly into a replacement decision — on a diagnosis the literature itself calls ill-defined [4].

How to be sure

• Look before you read. In the in vitro ICDAS-merged comparison, visual assessment out-performed radiographic assessment on sensitivity (0.80 against 0.68) on the same 296 extracted specimens [10]. Mjör and Toffenetti found consistency or hardness, and discoloration of dentin and enamel, to be the best parameters for a difficult clinical diagnosis [3].
• Don't expect a second angle to rescue you. Changing the vertical angle of the beam from 0° to 10° produced no significant difference in accuracy for detecting simulated secondary caries in esthetic restorations across 20 extracted teeth read by 10 observers, with areas under the ROC curve from 0.37 to 0.66 [18].
• Know what the material is doing. Restorative material was found to influence the radiographic diagnosis of secondary caries, while the imaging system — film, photostimulable phosphor or CCD — had no influence, in a study on extracted third molars [6].
• CBCT is not the answer here. On 45 extracted molars, two CBCT units reached accuracies of 0.83 and 0.78 against 0.55 for conventional and 0.49 for digital intraoral radiography [17]. That is an accuracy comparison on extracted teeth. Nothing retrieved for this entry addresses dose, indications or clinical benefit for this purpose.

Common errors

Quoting an in-vitro sensitivity as chairside performance. The most quotable numbers in this field — 0.68, 0.591, 0.83 — all come from extracted teeth read against histology [10][11][17]. A tooth in a wax block is not a tooth in a mouth behind a contact point.

Assuming the material under the restoration is radiopaque. Of 56 restorative materials measured against an aluminium stepwedge, all exceeded the dentin reference value, but 4 of 33 conventional composites and 3 of 16 flowable resin composites had significantly lower radiopacity than enamel [5]. And you cannot check this from the packaging: measured values differed from manufacturers' declared values by up to 1.53 mm eq Al for some materials [5].

Treating the radiopacity question as settled. Two in vitro studies concluded that radiopacity at or slightly above enamel helps the read: materials with greater radiopacity than enamel favoured a true-negative diagnosis [6], and a semi-radiopaque restoration seemed favourable with the best radiopacity slightly exceeding enamel across 49 teeth with gingival-wall lesions and 29 controls read by 11 dentists [7]. A third concluded the opposite — that an increase in a composite's radiopacity could have a subjectively negative influence on secondary caries diagnosis, and that ideal composite radiopacity is closer to dentin [13]. Look at what each measured: the first two counted false positives on sound teeth, the third counted detection of a lesion that was present. High radiopacity protects you from one error and costs you the other.

Calling every cervical lucency. In the meta-analysis of radiographic optical effects, triangular-shaped radiolucencies and the Mach band effect produced false-positive diagnoses of caries or fracture in approximately 13% of observations (60 of 464) — on evidence the authors rate as low to very low certainty, with heterogeneity above 90% [9].