J. Dent. 1990; 18: 130-l 36
130
Accuracy in radiographic diagnosis: practitioners and dental caries* P. A. Mileman
Dutch
and L. T. van der Weelet
Department of Radiology, The Academic University of Groningen, The Netherlands
Centre for Dentistry
in Amsterdam
(ACTA) and jcomputer
Centre,
ABSTRACT The effective diagnostic accuracy of dentists is the missing link in models for calculating the optimal frequency for radiographic recall. The purpose of this study was to describe the accuracy of Dutch general dental practitioners using bitewing radiographs for the diagnosis and treatment of approximal caries in dentine. Practitioners from a 10 per cent (n = 444) national random sample were sent a set of simulated bitewings, a patient description and a diagnostic response form. The radiographs showed tooth surfaces with (n = 45) and without (n = 60) dentine caries. Microradiography of the sectioned teeth was used as a diagnostic reference standard. The participants were asked to diagnose dentine caries using a five-point certainty scale and decide if restorative treatment was indicated. The response was 62 per cent (n = 276). The area under a mean receiver operating characteristic curve for the practitioners was 88 per cent of the area representing optimal accuracy. Diagnostic accuracy, however, varied considerably between dentists. For caries considered to be ‘probably’ in the dentine mean sensitivity was 54 per cent (s.d. 14 per cent) and specificity was 97 per cent (s.d. 5 per cent). The pooled restorative treatment decisions of the dentists had a mean sensitivity of 62 per cent and a specificity of96 per cent. More surfaces were considered in need of treatment than were diagnosed as ‘probably’ having dentine caries. Treatment decisions were more accurate than diagnostic ability might indicate. Paradoxically this may be because errors in visual ability were partly compensated for by ‘over treatment’ decisions. KEY WORDS:
Dental caries, Radiography,
J. Dent. 1990; 18: 130-l
Diagnosis
36 (Received 22 August
1989;
reviewed
20 October
1989;
accepted
19 February
1990) Correspondence should be addressed to: Dr P. A. Mileman, Department Dentistry in Amsterdam (ACTA), Louwesweg 1, 1066 EA Amsterdam,
INTRODUCTION The use of bitewing radiographs is an accepted diagnostic test for approximal caries (Mileman et al.. 1988). However, the merits of their use or their substitution by a technique such as fibreoptic transillumination or xeroradiography require a rational basis (White et al., 1984; MejAre et al.. 1985; Sidi and Naylor, 1988). This could be in terms of their efficacy or optimal diagnostic properties when used by experts, or the efficiency of their performance in their routine application in the general practice setting (Committee for Evaluating Medical Technologies in Clinical Use, 1985). *Part of the material used in this manuscript has been submitted in fulfilment of the requirements for a PhD degree of the University of Groningen, The Netherlands. 0 1990 Butterworth-Heinemann 0300-5712/90/030130-07
Ltd.
of Dental Radiology, Academic Centre for The Netherlands.
The diagnostic accuracy of bitewing radiographs when used for the diagnosis of approximal caries and subsequent therapeutic decision making is the missing link in models for calculating the optimal diagnostic strategy or frequency of radiographic recall (Pliskin et al., 1985; Shwartz et al.. 1986; Mileman et al., 1988). Previously only the efficacy of the radiographic system has been described when using small selected groups of experts or volunteers (White et al., 1984; Mejare et al., 1985; Okano et al., 1985; Douglass et al., 1986; Espelid, 1987; Sidi and Naylor, 1988). This might give a misleading indication of the frequency with which radiographs should be taken and the radiographic depth at which treatment should be initiated. The resultant piecemeal suggestions for changes in patterns of dental practice could have deleterious rather than beneficial effects on patient oral health.
Mileman
and van der Weele: Accuracy
The purpose of this study was to describe the accuracy with which Dutch general dental practitioners could use bitewing radiographs for the diagnosis of caries in dentine. In addition, diagnostic agreement between the practitioners, expert observers and a more objective microradiographic diagnostic ‘gold standard was tested. It is assumed, only for the purpose of analysing dentist treatment decision-making behaviour, that dentine caries diagnosed either by microradiography or visually on dental radiographs represents clinical cavitation (BackerDirks et al., 1961; Purdell-Lewis et al., 1974; Mejare et al.. 1985; Espelid and Tveit, 1986; Downer, 1989) which should be treated restoratively.
METHOD In 1984 a 10 per cent (n = 444) random sample of Dutch dental practitioners was sent a two-wave questionnaire based on an analytical decision-making approach to radiographic caries diagnosis and restorative treatment decision making. In the second wave, the practitioners who had responded to the first questionnaire were sent a duplicated set of simulated bitewing radiographs, a patient description and a diagnostic response form*. The overall response was 62 per cent (n = 276). Further details of the method have been reported elsewhere (Mileman, 1985; Mileman et al., 1988). Teeth Human molar and premolar teeth, both sound and with natural carious lesions, were used as the test objects. These teeth had been extracted during a 3 month period mainly for orthodontic reasons by local dentists. The teeth were stored immediately after extraction in 10 per cent buffered neutral formalin until used. They were mounted so as to reconstruct a quadrant of the natural dentition. The mounted, cleaned teeth, preselected for distribution of surface conditions and absence of anatomical anomalies and fillings, were then radiographed. Seventy-five per cent of surfaces (n = 79) were sound or had initial carious lesions with a cavity less than 0.5 mm in diameter, 2.5per cent (n = 26) showed signs of carious attack and had a cavity > 0.5 mm (Purdell-Lewis et al., 1974). After radiographic examination, the teeth were fixed to the sectioning table of a cooled circular saw (Wetzlar 1600, Leitz. Wetzlar, FRG) and cut mesiodistallyin planoparallel sections through the largest part of the lesion or surface defect and including the pulp. These sections were then microradiographed using the materials and a general method previously described (Josselin de Jong and Ten Bosch, 1985). A tracing of the relative mineral content at maximum lesion depth was made using a densitometer so that the lesion depth could be measured. *Both questionnaires
are available from the authors on request.
in radiographic
caries diagnosis
131
X-ray generator A General Electric 1000 X-ray machine (General Electric, Millwaukee, WI, USA) was used. It was operated at 65 kVp, 15 mA and had 2.5 mm of aluminium equivalent tiltration. The exposure time used was 1.5 s. The half-value layer was equivalent to 2.7 mm of aluminium. The focus to film distance of 48 cm gave a 9.0 cm field diameter. A 4 mm thick circular piece of aluminium was placed at the corner of each of the films, where the mean film density was 1.3. Radiographs The 105 surfaces of the premolar and molar teeth were radiographed together as a ‘quadrant of the jaw’, in a bitewing mode, using Kodak Ultra Speed Film (Kodak, Rochester, NY, USA). Perspex, 3 cm thick, was interposed between the teeth and the X-ray source to simulate the natural absorbtion of the soft tissues. The long-cone of the X-ray machine was orientated perpendicular to the tooth/ film combination, held in close apposition by a specially constructed device, using a right-angled triangle measure. Duplication
of the radiographs
The original films were duplicated onto single sheets of Kodak Duplicating Film (Kodak) using a Kodak Blue Ray Mark II Duplicator and Kodak X-omat MSA-N Processor with freshly made Kodak Chemicalium at 35 “C. Strict quality control of the procedure was carried out so that all the duplicate radiographs had film density values within 0.05 of each other and these were on average only 0.1 less than the original ‘bitewing’ radiographs. Patient
description
The dentists were given a description used previously of a 15-year-old girl patient: ‘She is motivated, brushes her teeth twice a day and attends the dentist twice a year for inspections. She uses a fluoride toothpaste and has reasonable oral hygiene’ (Mileman, 1985). Response
form
The practitioners were asked to diagnose for each surface the presence of caries in the outer half of dentine. They used the following Receiver Operating Characteristic (ROC) scoring system for lesion presence (McNeil et al.. 1975; White et al., 1984; Mileman, 1985; Espelid, 1987): 0, definitely absent; 1, probably absent; 2, uncertain; 3, probably present; 4, definitely present. Subsequently they were asked to decide if immediate restorative treatment was necessary. Validation
of the practitioners’
judgements
Two ‘gold standard diagnoses were used for validation. The teeth used were sectioned and examined for dentine caries using microradiography; 45 of the 105 surfaces were
132
J. Dent. 1990;
18: No. 3
Table 1. Quantitative measures of diagnostic accuracy for the sample of dentists derived from two types of standard diagnosis Measures of dentists’ diagnostic accuracy for caries probably present* in the outer half of dentine Dichotomized standard diagnosis Microradiography (‘sound’/dentine caries) Expert observers for the radiographs (‘sound’/dentine
% Sensitivity
Kappa
% Specificity
Mean
s.d.
Mean
s.d.
Mean
s.d.
54
14
97
5
0.53
0.14
7g
,7
95
5
0.75
0.13
caries) *Diagnostic confidence levels 3 + 4. n = 276 dentists.
diagnosed as having dentine caries. The duplicated radiographs were viewed and judged by two experienced examiners from an independent research institution to provide a validated norm for the visual presence and depth of caries. These examiners had trained to 95 percent agreement before viewing the radiographs using a standardized method (Backer-Dirks et al., 1961). They found 28 surfaces with and 77 without visible dentine caries. Statistical
analysis
Using a decision matrix for the presence or absence of caries in the outer half of dentine enabled the diagnostic measures of sensitivity, specificity, true positives (TP), false positives (FP), false negatives (FN) (McNeil et al., 1975) and Kappa (Landis and Koch, 1977) for diagnostic agreement to be calculated. The dental practitioners’ scores used were the combined rating scores indicating probable lesion presence (3 + 4) or lesion absence (0 + 1 + 2) and the decision whether or not to restoratively treat the tooth surface in question. The chi-square test at a 5 per cent level of significance was used to test for differences between the various diagnostic measures. In a separate ROC analysis the full range of the pooled rating scores was used together with a previously tested and published computer programme for the measurement of diagnostic accuracy (Swets and Pickett, 1982).
79 per cent which means that they missed 21 per cent of the dentine lesions, even when using their most lenient criterion. A graphic display of the range of diagnostic agreement is presented in Fig 1. When looking at the values for Kappa these can be assumed to represent ‘slight’ (0.0) to ‘substantial’ (0.8) diagnostic agreement (Landis and Koch, 1977). A comparison of the frequency with which the surveyed dentists indicated that surfaces were probably carious and the frequency with which they considered surfaces require restoring is set out in Fig. 2. The frequency histogram of fillings needed, shown in Fig. 2, is to the right of that for surfaces probably with outer dentine caries. The mean (26) of the latter is lower than that of the former (30), indicating that dentists in general use an even more lenient criterion than ‘probably carious’ in the outer dentine or restore surfaces with radiolucencies confined to enamel. (‘Lenient’ is used here to mean deciding to initiate restorative treatment with a low degree of certainty that caries in the dentine is present. ‘Strict’ 30%
I
07 20%
;; .-
7
s ‘;;
s 10%
RESULTS In Table I the mean results for the measures of diagnostic quality-sensitivity, specificity and Kappa-are presented using as a ‘gold standard’ microradiographic outer dentine caries of the actual tooth from which the dental radiograph was made. Also included in Table I is a comparison with the judgement of the expert observer standard. The latter provides an indication of the general ability of the practitioners to read radiographs in the same way as this standard. When compared with the expert observers the dentists had a mean diagnostic sensitivity of
E $ e L m
0% o.o*
0.2 Kappa
Slight-Agreement
0.4 outer
0.6
dentine
0.8
caries
-Substantial
Fig. 7. Variation between Dutch dental practitioners in diagnostic agreement for caries probably in the outer half of dentine (confidence levels 3 + 4) using microradiography as a validating norm. Mean = 0.53, s.d. = 0.14, range = -0.05-0.80. (*Dentistwith Kappavalueof - 0.05added to frequency with Kappa scores between 0.0 and 0.1.)
Mileman
30 Mean
and van der Weele: Accuracy
Filling needed
30
in radiographic
12
caries diagnosis
25 30 34 (median)
47
133
60
Frequency distribution of dentists
I
0
10
20
I
I
30
40
1
50 b
0 10
20 30 40 50 Surfaces (out of 105)
60
’
l--l
60
No. of fillings indicated
Fig. 3. Frequency distribution diagram of Dutch dental practitioners according to the number of surfaces judged to require restorative treatment. The number of surfaces with at least outer dentine caries according to the expert observers (28) and microradiography (45) is superimposed. ci, 1.5 X interquartile range; El, interquartile range; *, extreme values. Mean no. of surfaces with diagnostic errors 4 I
False positives False negatives I
l2r a
4 0 4
a 12 16
Mean
26 *Score
3+4
Probably carious in outer dentine
Fig. 2. Variation in surfaces requiring restorative compared with probably* carious surfaces.
treatment,
describes the opposite tendency.) These frequency distributions can be compared with the number of surfaces judged to have dentine caries according to the expert observer (n = 28) or themicroradiography norms (n = 45). This has been done by superimposing the values of the norms on to the frequency distribution of the dentists in this and subsequent figures. A measure of the variation in the distribution of the judgement of the dentists about a central tendency is presented in Fig. 3. This diagram (‘box-plot’) of the frequency distribution of dentists according to the number of fillings indicated shows the values between which 50 per cent of the dentists would have prescribed fillings. Extreme maximum values are shown falling outside the rectangle representing one and a half times the interquartile range. Because as a result of diagnostic errors some extensive dentine caries might have been missed (false-negative decisions, FN) and some sound surfaces or surfaces with initial caries may have been mistakenly indicated in need of treatment (false-positive decisions, FP). an error analysis of the decisions made per surface has been carried out. This is presented in Fig. 4. These results show, for example. that almost all of the dentists decided that
Fig. 4. Histogram of the mean frequency of diagnostic errors made by 276 dentists for (a) restorative treatment decisions and (b) the diagnosis ‘probably dentine caries’. These diagnoses were compared with the microradiography and expert observer ‘gold standard’ diagnoses for dentine caries. ?,Treatment ? decisions using microradiography; E%treatment decisions using expert observers; 8, diagnosis ‘probably dental caries’ from radiograph using microradiography.
fewer surfaces were in need of treatment than were indicated, by microradiography, as having outer dentine caries. A maximum of 19 surfaces were ‘overtreated’ and 37 ‘undertreated’by one dentist or another. On average, 2.4 surfaces were ‘overtreated’ whilst 17.2 of the 45 carious surfaces were missed or ‘undertreated’. A similar average frequency of errors occurred for the diagnosis by the practitioners of surfaces with ‘probably dentine caries’. False-positive errors outnumber false-negative errors when compared with the expert observers’ norm by approximately two to one The results of paired comparisons between the agreement of the dentists’ diagnosis ‘probably carious’ in the dentine and their decisions to treat surfaces or the judgement of the respective diagnostic standards are shown in Table II. The results show that the best diagnostic agreement (P < 0.005) occurs between the dentist’s decision that a filling is needed and his own perception of whether dentine caries was visible on the dental radiographs or not. This is an indication that selfcancelling errors may occur both through limitations in the radiographic technique and due to perceptual error.
134
J. Dent. 1990;
18: No. 3
Tab/e II. Mean relative diagnostic agreement for dentine caries between the various diagnostic measures Kappa (s.d.) Mean Dentists ‘probably carious’ microradiography (a) Filling ‘needed’/microradiography (b) Filling ‘needed’/expert observers Filling ‘needed’/dentine (c) caries as perceived by dentists (d)
0.53
(0.14)*
0.60 0.77
(0.1 1 )* (0.1 l)*
0.85
(0.13)*
Differences between the measures of agreement (a), (b). (c) and (d) have been tested on the original data using the chi-square test. All the measures of agreement differed significantly. x2 > 150, d.f. = 6, * = P < 0.005.
decisions are more accurate than diagnostic ability might indicate. This may mean that dentists subconsciously adapt their restorative criteria to compensate for their inability to diagnose dentine caries depth correctly. The radiographic diagnostic accuracy of practitioners as an indication of the efficiency of radiography as a diagnostic system is best presented in the form of an ROC curve since there is always a relationship between sensitivity and specificity which varies depending on the stringency of the diagnostic criteria used @wets and Pickett, 1980). Data for the varying degrees of diagnostic confidence of the 276 dentists are pooled in Fig. 5. This figure shows the mean relationship between sensitivity and specificity for caries in the outer half of dentine using microradiography as the ‘gold standard’ diagnosis. It Treatment
ROC graph
with
s .Y 2
40
Definitely
ov 0
carious /
microradiography
/ I
I
20 % False
as norm
40
positive
I
60
80
(l-specificity)
fig. 5. Receiver operating characteristic (ROC) curve for the accuracy when using bitewing radiographs of dentine caries recognition. The data of the 276 Dutch dental practitioners has been pooled and microradiography has been used as the ‘gold standard’ diagnosis. *Restorative treatment decision judgement superimposed: TP 61.7 per cent, FP 4.1 per cent.
should be born in mind that as sensitivity rises with increasing leniency of the decision-making cut-off point, so does the proportion of false-positive diagnoses (i.e. specificity declines). The plot points in this figure are, for definitely carious TP 39.5 per cent, FP 1.2 per cent, for probably carious TP 53.5 per cent, FP 3.4 per cent, for uncertain if carious TP 64.7 per cent, FP 7.3 per cent and for probably not carious TP 73.0 per cent, FP 12.4per cent. The resultant area under this ROC curve was AZ 0.88. Mean restorative treatment criterion has a position of TP 61.7 per cent, FP 4.1 per cent and is superimposed on the figure.
DISCUSSION Caveats that must accompany the interpretation of the results of this study include: the adequacy of the response; the quality of the duplicated radiographs; the appropriateness of caries in the outer half of the dentine according to the microradiographic technique as a standard for analysing diagnostic accuracy and treatment decision-making behaviour, and the representativeness of the sample of carious teeth. This was the second part of a two-wave questionnaire study. The first part used a computer-generated 10 per cent random sample of dentists and the subsequent response did not differ in age structure from the total population of dentists (Mileman et al., 1988). This group of respondents was surveyed on the second wave, resulting in a final response of 62 per cent of the total sample. It has been suggested that in postal surveys of dentists the results with a response of45 per cent or more do not differ significantly from those obtained using the total sample (Houland et al., 1980). This means that the response in this study can be considered satisfactory. The quality of duplicated radiographs was strictly controlled so that it is extremely unlikely to be the cause of variation between the practitioners. It is unlikely that the present study underestimates diagnostic variation or overestimates diagnostic accuracy in clinical practice, since in general practice the quality of bitewing radiographs is often inadequate (Mannyet al., 1980;McDonald, 1983, Nysether and Hansen, 1983). The difference in density between the duplicate and the original radiographs was small. Psychophysical studies (Erales and Manson-Hing, 1979; Price, 1980) of the differences between duplicate and original radiographs have concluded that the duplication of radiographs is ‘not likely to materially effect the clinical performance’ whilst the observers used ‘rated all duplicates superior to the original radiograph’ (Erales and Manson-Hing, 1979). Microradiography has been chosen as a ‘gold standard for radiographic caries diagnosis in the past (PurdellLewis et al., 1974; White et al., 1984; Josselin de Jong and Ten Bosch, 1985). It offers in one scale an objective measure of the degree of lesion progression towards the pulp and the associated chance of surface cavitation (Purdell-Lewis et al., 1974). Surface cavitation is assumed
Mileman
and van der Weele: Accuracy
to be synonymous with lesion irreversibility and is therefore the histological stage at which restorative treatment should probably be initiated. Most surfaces (94 per cent) with microradiographic dentine caries demonstrate lesion cavitation (Purdell-Lewis et al., 1974). It seems reasonable therefore to use dentine caries diagnosed microradiographically at least as a ‘benchmark’ for analysing dentist restorative treatment decision-making behaviour. The use of radiographic dentine caries according to the expert observers as a threshold from which to measure restorative treatment decision-making behaviour has support (Espelid and Tveit, 1986) although it is also to some degree subjective (Tulloch et al., 1988) and depends on caries prevalence. The selection and prevalence of lesions to form the test battery for a disease with low prevalence is a compromise between having enough cases of disease to provide reasonably reliable results for sensitivity and having enough cases without disease to reliably measure specificity, without being too onerous for the test participants. We were concerned essentially to choose a selection of disease cases within which the practitioners’ treatment decision range would be likely to fall. The average accuracy for the diagnosis of dentine caries was substantial (AZ = 0.88). Although it provides evidence for the systematic underdiagnosis of lesions on radiographs, it compares favourably with other estimates of diagnostic accuracy using selected observers (Ruttimann, 1987) and indicates that the practitioners were able to comprehend the task they had been set. Variation between practitioners in diagnostic accuracy as measured by the ‘Kappa’ statistic (Fig. 1) was however large. Not all the dental practitioners would wait until a lesion was in dentine before deciding to carry out restorative treatment. The variation in treatment decision making, combined with the mean ‘overtreatment’ of surfaces with dentine caries (Fig. 4) can be explained by the fact that about a quarter of the dentists could be considered to have criteria leading to an abnormally high frequency of prescribing restorative treatment need (Fig. 3). The suggestion that these criteria have been adopted in part to compensate for underdiagnosis (Fig. 5 average FN derived from the ROC curve was 38.8 per cent) requires further investigation. This study demonstrates that it cannot be assumed that the radiographic method of caries diagnosis currently used by dental practitioners can achieve optimal accuracy as a diagnostic test. Even the ‘experts’ used in this study could not diagnose seven of the 45 lesions in dentine at any degree of certainty. On average the practitioners underestimated dentine caries presence when compared with a microradiography or expert ‘gold standard’ (Table Z.Fig. 4). It is interesting to note however that their decisions to place restorations had generally less errors (Fig. 4) and showed significantly greater agreement with the ‘gold standards’ than their perception of carious lesion depth (Table II). This may
in radiographic
caries diagnosis
135
mean that in dental practice treatment criteria are partly derived from diagnostic ability. The majority of these dentists report having treatment criteria in the enamel (Mileman, 1985). However they may in fact treat dentine lesions which they perceive as being confined to the enamel. If this were so then recommendations to use stricter treatment criteria in the dentine might lead to an unacceptably high frequency of dentine lesions being left untreated despite a radiographic examination. The current risk of overtreatment should not be underestimated. For example, with an approximal dentine caries prevalence of say 5 per cent for the patient we have described (Grondahl et al., 1977) ‘over-treatment’decisions would be made just as often as true positive treatment decisions using the mean restorative treatment criterion fromFig. 5 (TP 61.7 per cent, FP 4.1 per cent: a likelihood ratio of 15 for the test). This is an indication that, at the time of this study, some dentists had a treatment philosophy which could lead in practice to the unnecessary restorative treatment of carious lesions confined to the enamel (DHSS, 1986). On the other hand basing recommendations concerning restorative treatment criteria on the diagnostic accuracy of experts (Mileman et al., 1988; Tulloch et al., 1988) should be done with some caution. False-positive treatment decisions might be reduced at the expense of an unacceptable rise in undertreatment with all its consequences. For this reason it is suggested that restorative treatment decision making and radiographic caries diagnosis processes should be taught in an integrated form in graduate and postgraduate dental courses (DHSS, 1986). Acknowledgements The authors would like to thank Dr D. J. Purdell-Lewis, Dr J. Bouma, Dr R. Liem, Dr A. C. M. van de Poel, Mr L. J. van Rijn and The Dutch Dental Association as well as the dentists who participated in the study.
References
Committee for Evaluating
Medical Technologies in Clinical Use, Institute of Medicine (1985) Assessing Medical Technologies. Washington, National Academy Press. pp. 70-175. Backer-Dirks O., Houwink B. and Kwant G. W. (1961) The results of 6% years of artificial fluoridation of drinking water in The Netherlands. Arch. Oral Sol. 5, 284-300. DHSS (1986) Report of the Committee of Enquiry into Unnecessary Dental Treatment. London. HMSO. Douglass C. W.. Valachovic R. W.. Wijesinha A. et al. (1986) Clinical efficacy of dental radiography in the detection of dental caries and periodontal diseases. Oral Surg. Oral Med. Oral Pathol. 62, 330-339. Downer M. C. (1989) Validation of methods used in dental caries diagnosis. Int. Dent. J. 39, 241-246. Erales F. A. and Manson-Hing L. R. (1979) A study of the quality of duplicated radiographs. Oral Surg. Oral Med. Oral Pafbol. 47, 98-104. Espelid I. (1987) Radiographic Diagnoses and Treatment Decisions on Approximal Caries. Bergen, University of Bergen.
136
J. Dent.
1990; 18: No. 3
Espelid I. and Tveit A. B. (1986) Clinical and radiographic assessment of approximal carious lesions. Acta Odon tel. Stand. 44, 31-37. Griindahl H-G., Hollender L.. Malmcrona E. et al. (1977) Dental caries and restorations in teenagers: II. A longitudinal radiographic study of the caries increment of proximal surfaces among urban teenagers in Sweden. Swed. Dent J. 1, 51-57. Houland E. S.. Romberg E. and Moreland E. F. (1980) Nonresponse bias to mail survey questionnaires within a professional population. J. Dent. Educ. 44,270-274. Josselin de Jong E. and Ten Bosch J. J. (1985) Error analysis of the microradiographic determination of mineral content in mineralized tissue slices. Phys. Med. Biol. 30, 1067-1075. Landis J. R. and Koch G. G. (1977) The measurement of observer agreement for categorical data. Biometrics 33, 157-174. Manny E. F., Carlson K. C., McClean P. M. et al. (1980) An Overview of Dental Radiology (Monograph Series). Rockville, Maryland, National Institute of Health Care Technology. McDonald S. P. (1983) A method to reduce interproximal overlapping and improve reproducibility of bitewing radiographs for clinical trials. Community Dent Oral Epidemiol. l&289-295. McNeil B. J.. Keeler E. and Adelstein S. J. (1975) Primer on certain elements of medical decision making. N. Engl. J. Med. 293, 211-215. Mejare I., GrBndahl H-G., Carlstedt K. et al. (1985) Accuracy at radiography and probing for the diagnosis of proximal caries. &and. J. Dent Res. 93, 178-184. Mileman P. A. (1985) Radiographic Caries Diagnosis and Restorative Treatment Decision Making. PhD Thesis, University of Groningen. Mileman P. A., van der Weele L. T., van de Poe1 A. C. M. et al. (1988) Dutch dentists’ decisions to take bitewing radiographs. Community Dent. Oral Epidemiol. 16, 368-373.
Nysether S. and Hansen B. F. (1983) Errors on bitewing radiographs. Community Dent. Oral Epidemiol. 11, 286-288. Okano T.. Huang H-J. and Nakamura T. (1985) Diagnostic accuracy on detection of proximal enamel lesions in nonscreen radiographic performance. Oral Surg. Oral Med. Oral Pathol. 59, 543-547. Pliskin J. S., Shwartz M., Grijndahl H-G. et al. (1985) Incorporating individual patient preferences in scheduling bitewing radiographs. Methods Inf: Med. 24, 213-217. Price C. (1980) A densitometric evaluation of two radiographic duplicating films under differing conditions of exposure and processing. Oral Surg. Oral Med. Oral Pathol. 50, 190- 194. Purdell-Lewis D. J., Groeneveld A, Pot T. J. et al. (1974) Proximal carious surfaces-a comparison of visual, radiographical and micro radiographic appearance. Ned. Tijdschr. Tandheelkd. 81; Nether&&s_ Dtf
130
Accuracy in radiographic diagnosis: practitioners and dental caries* P. A. Mileman
Dutch
and L. T. van der Weelet
Department of Radiology, The Academic University of Groningen, The Netherlands
Centre for Dentistry
in Amsterdam
(ACTA) and jcomputer
Centre,
ABSTRACT The effective diagnostic accuracy of dentists is the missing link in models for calculating the optimal frequency for radiographic recall. The purpose of this study was to describe the accuracy of Dutch general dental practitioners using bitewing radiographs for the diagnosis and treatment of approximal caries in dentine. Practitioners from a 10 per cent (n = 444) national random sample were sent a set of simulated bitewings, a patient description and a diagnostic response form. The radiographs showed tooth surfaces with (n = 45) and without (n = 60) dentine caries. Microradiography of the sectioned teeth was used as a diagnostic reference standard. The participants were asked to diagnose dentine caries using a five-point certainty scale and decide if restorative treatment was indicated. The response was 62 per cent (n = 276). The area under a mean receiver operating characteristic curve for the practitioners was 88 per cent of the area representing optimal accuracy. Diagnostic accuracy, however, varied considerably between dentists. For caries considered to be ‘probably’ in the dentine mean sensitivity was 54 per cent (s.d. 14 per cent) and specificity was 97 per cent (s.d. 5 per cent). The pooled restorative treatment decisions of the dentists had a mean sensitivity of 62 per cent and a specificity of96 per cent. More surfaces were considered in need of treatment than were diagnosed as ‘probably’ having dentine caries. Treatment decisions were more accurate than diagnostic ability might indicate. Paradoxically this may be because errors in visual ability were partly compensated for by ‘over treatment’ decisions. KEY WORDS:
Dental caries, Radiography,
J. Dent. 1990; 18: 130-l
Diagnosis
36 (Received 22 August
1989;
reviewed
20 October
1989;
accepted
19 February
1990) Correspondence should be addressed to: Dr P. A. Mileman, Department Dentistry in Amsterdam (ACTA), Louwesweg 1, 1066 EA Amsterdam,
INTRODUCTION The use of bitewing radiographs is an accepted diagnostic test for approximal caries (Mileman et al.. 1988). However, the merits of their use or their substitution by a technique such as fibreoptic transillumination or xeroradiography require a rational basis (White et al., 1984; MejAre et al.. 1985; Sidi and Naylor, 1988). This could be in terms of their efficacy or optimal diagnostic properties when used by experts, or the efficiency of their performance in their routine application in the general practice setting (Committee for Evaluating Medical Technologies in Clinical Use, 1985). *Part of the material used in this manuscript has been submitted in fulfilment of the requirements for a PhD degree of the University of Groningen, The Netherlands. 0 1990 Butterworth-Heinemann 0300-5712/90/030130-07
Ltd.
of Dental Radiology, Academic Centre for The Netherlands.
The diagnostic accuracy of bitewing radiographs when used for the diagnosis of approximal caries and subsequent therapeutic decision making is the missing link in models for calculating the optimal diagnostic strategy or frequency of radiographic recall (Pliskin et al., 1985; Shwartz et al.. 1986; Mileman et al., 1988). Previously only the efficacy of the radiographic system has been described when using small selected groups of experts or volunteers (White et al., 1984; Mejare et al., 1985; Okano et al., 1985; Douglass et al., 1986; Espelid, 1987; Sidi and Naylor, 1988). This might give a misleading indication of the frequency with which radiographs should be taken and the radiographic depth at which treatment should be initiated. The resultant piecemeal suggestions for changes in patterns of dental practice could have deleterious rather than beneficial effects on patient oral health.
Mileman
and van der Weele: Accuracy
The purpose of this study was to describe the accuracy with which Dutch general dental practitioners could use bitewing radiographs for the diagnosis of caries in dentine. In addition, diagnostic agreement between the practitioners, expert observers and a more objective microradiographic diagnostic ‘gold standard was tested. It is assumed, only for the purpose of analysing dentist treatment decision-making behaviour, that dentine caries diagnosed either by microradiography or visually on dental radiographs represents clinical cavitation (BackerDirks et al., 1961; Purdell-Lewis et al., 1974; Mejare et al.. 1985; Espelid and Tveit, 1986; Downer, 1989) which should be treated restoratively.
METHOD In 1984 a 10 per cent (n = 444) random sample of Dutch dental practitioners was sent a two-wave questionnaire based on an analytical decision-making approach to radiographic caries diagnosis and restorative treatment decision making. In the second wave, the practitioners who had responded to the first questionnaire were sent a duplicated set of simulated bitewing radiographs, a patient description and a diagnostic response form*. The overall response was 62 per cent (n = 276). Further details of the method have been reported elsewhere (Mileman, 1985; Mileman et al., 1988). Teeth Human molar and premolar teeth, both sound and with natural carious lesions, were used as the test objects. These teeth had been extracted during a 3 month period mainly for orthodontic reasons by local dentists. The teeth were stored immediately after extraction in 10 per cent buffered neutral formalin until used. They were mounted so as to reconstruct a quadrant of the natural dentition. The mounted, cleaned teeth, preselected for distribution of surface conditions and absence of anatomical anomalies and fillings, were then radiographed. Seventy-five per cent of surfaces (n = 79) were sound or had initial carious lesions with a cavity less than 0.5 mm in diameter, 2.5per cent (n = 26) showed signs of carious attack and had a cavity > 0.5 mm (Purdell-Lewis et al., 1974). After radiographic examination, the teeth were fixed to the sectioning table of a cooled circular saw (Wetzlar 1600, Leitz. Wetzlar, FRG) and cut mesiodistallyin planoparallel sections through the largest part of the lesion or surface defect and including the pulp. These sections were then microradiographed using the materials and a general method previously described (Josselin de Jong and Ten Bosch, 1985). A tracing of the relative mineral content at maximum lesion depth was made using a densitometer so that the lesion depth could be measured. *Both questionnaires
are available from the authors on request.
in radiographic
caries diagnosis
131
X-ray generator A General Electric 1000 X-ray machine (General Electric, Millwaukee, WI, USA) was used. It was operated at 65 kVp, 15 mA and had 2.5 mm of aluminium equivalent tiltration. The exposure time used was 1.5 s. The half-value layer was equivalent to 2.7 mm of aluminium. The focus to film distance of 48 cm gave a 9.0 cm field diameter. A 4 mm thick circular piece of aluminium was placed at the corner of each of the films, where the mean film density was 1.3. Radiographs The 105 surfaces of the premolar and molar teeth were radiographed together as a ‘quadrant of the jaw’, in a bitewing mode, using Kodak Ultra Speed Film (Kodak, Rochester, NY, USA). Perspex, 3 cm thick, was interposed between the teeth and the X-ray source to simulate the natural absorbtion of the soft tissues. The long-cone of the X-ray machine was orientated perpendicular to the tooth/ film combination, held in close apposition by a specially constructed device, using a right-angled triangle measure. Duplication
of the radiographs
The original films were duplicated onto single sheets of Kodak Duplicating Film (Kodak) using a Kodak Blue Ray Mark II Duplicator and Kodak X-omat MSA-N Processor with freshly made Kodak Chemicalium at 35 “C. Strict quality control of the procedure was carried out so that all the duplicate radiographs had film density values within 0.05 of each other and these were on average only 0.1 less than the original ‘bitewing’ radiographs. Patient
description
The dentists were given a description used previously of a 15-year-old girl patient: ‘She is motivated, brushes her teeth twice a day and attends the dentist twice a year for inspections. She uses a fluoride toothpaste and has reasonable oral hygiene’ (Mileman, 1985). Response
form
The practitioners were asked to diagnose for each surface the presence of caries in the outer half of dentine. They used the following Receiver Operating Characteristic (ROC) scoring system for lesion presence (McNeil et al.. 1975; White et al., 1984; Mileman, 1985; Espelid, 1987): 0, definitely absent; 1, probably absent; 2, uncertain; 3, probably present; 4, definitely present. Subsequently they were asked to decide if immediate restorative treatment was necessary. Validation
of the practitioners’
judgements
Two ‘gold standard diagnoses were used for validation. The teeth used were sectioned and examined for dentine caries using microradiography; 45 of the 105 surfaces were
132
J. Dent. 1990;
18: No. 3
Table 1. Quantitative measures of diagnostic accuracy for the sample of dentists derived from two types of standard diagnosis Measures of dentists’ diagnostic accuracy for caries probably present* in the outer half of dentine Dichotomized standard diagnosis Microradiography (‘sound’/dentine caries) Expert observers for the radiographs (‘sound’/dentine
% Sensitivity
Kappa
% Specificity
Mean
s.d.
Mean
s.d.
Mean
s.d.
54
14
97
5
0.53
0.14
7g
,7
95
5
0.75
0.13
caries) *Diagnostic confidence levels 3 + 4. n = 276 dentists.
diagnosed as having dentine caries. The duplicated radiographs were viewed and judged by two experienced examiners from an independent research institution to provide a validated norm for the visual presence and depth of caries. These examiners had trained to 95 percent agreement before viewing the radiographs using a standardized method (Backer-Dirks et al., 1961). They found 28 surfaces with and 77 without visible dentine caries. Statistical
analysis
Using a decision matrix for the presence or absence of caries in the outer half of dentine enabled the diagnostic measures of sensitivity, specificity, true positives (TP), false positives (FP), false negatives (FN) (McNeil et al., 1975) and Kappa (Landis and Koch, 1977) for diagnostic agreement to be calculated. The dental practitioners’ scores used were the combined rating scores indicating probable lesion presence (3 + 4) or lesion absence (0 + 1 + 2) and the decision whether or not to restoratively treat the tooth surface in question. The chi-square test at a 5 per cent level of significance was used to test for differences between the various diagnostic measures. In a separate ROC analysis the full range of the pooled rating scores was used together with a previously tested and published computer programme for the measurement of diagnostic accuracy (Swets and Pickett, 1982).
79 per cent which means that they missed 21 per cent of the dentine lesions, even when using their most lenient criterion. A graphic display of the range of diagnostic agreement is presented in Fig 1. When looking at the values for Kappa these can be assumed to represent ‘slight’ (0.0) to ‘substantial’ (0.8) diagnostic agreement (Landis and Koch, 1977). A comparison of the frequency with which the surveyed dentists indicated that surfaces were probably carious and the frequency with which they considered surfaces require restoring is set out in Fig. 2. The frequency histogram of fillings needed, shown in Fig. 2, is to the right of that for surfaces probably with outer dentine caries. The mean (26) of the latter is lower than that of the former (30), indicating that dentists in general use an even more lenient criterion than ‘probably carious’ in the outer dentine or restore surfaces with radiolucencies confined to enamel. (‘Lenient’ is used here to mean deciding to initiate restorative treatment with a low degree of certainty that caries in the dentine is present. ‘Strict’ 30%
I
07 20%
;; .-
7
s ‘;;
s 10%
RESULTS In Table I the mean results for the measures of diagnostic quality-sensitivity, specificity and Kappa-are presented using as a ‘gold standard’ microradiographic outer dentine caries of the actual tooth from which the dental radiograph was made. Also included in Table I is a comparison with the judgement of the expert observer standard. The latter provides an indication of the general ability of the practitioners to read radiographs in the same way as this standard. When compared with the expert observers the dentists had a mean diagnostic sensitivity of
E $ e L m
0% o.o*
0.2 Kappa
Slight-Agreement
0.4 outer
0.6
dentine
0.8
caries
-Substantial
Fig. 7. Variation between Dutch dental practitioners in diagnostic agreement for caries probably in the outer half of dentine (confidence levels 3 + 4) using microradiography as a validating norm. Mean = 0.53, s.d. = 0.14, range = -0.05-0.80. (*Dentistwith Kappavalueof - 0.05added to frequency with Kappa scores between 0.0 and 0.1.)
Mileman
30 Mean
and van der Weele: Accuracy
Filling needed
30
in radiographic
12
caries diagnosis
25 30 34 (median)
47
133
60
Frequency distribution of dentists
I
0
10
20
I
I
30
40
1
50 b
0 10
20 30 40 50 Surfaces (out of 105)
60
’
l--l
60
No. of fillings indicated
Fig. 3. Frequency distribution diagram of Dutch dental practitioners according to the number of surfaces judged to require restorative treatment. The number of surfaces with at least outer dentine caries according to the expert observers (28) and microradiography (45) is superimposed. ci, 1.5 X interquartile range; El, interquartile range; *, extreme values. Mean no. of surfaces with diagnostic errors 4 I
False positives False negatives I
l2r a
4 0 4
a 12 16
Mean
26 *Score
3+4
Probably carious in outer dentine
Fig. 2. Variation in surfaces requiring restorative compared with probably* carious surfaces.
treatment,
describes the opposite tendency.) These frequency distributions can be compared with the number of surfaces judged to have dentine caries according to the expert observer (n = 28) or themicroradiography norms (n = 45). This has been done by superimposing the values of the norms on to the frequency distribution of the dentists in this and subsequent figures. A measure of the variation in the distribution of the judgement of the dentists about a central tendency is presented in Fig. 3. This diagram (‘box-plot’) of the frequency distribution of dentists according to the number of fillings indicated shows the values between which 50 per cent of the dentists would have prescribed fillings. Extreme maximum values are shown falling outside the rectangle representing one and a half times the interquartile range. Because as a result of diagnostic errors some extensive dentine caries might have been missed (false-negative decisions, FN) and some sound surfaces or surfaces with initial caries may have been mistakenly indicated in need of treatment (false-positive decisions, FP). an error analysis of the decisions made per surface has been carried out. This is presented in Fig. 4. These results show, for example. that almost all of the dentists decided that
Fig. 4. Histogram of the mean frequency of diagnostic errors made by 276 dentists for (a) restorative treatment decisions and (b) the diagnosis ‘probably dentine caries’. These diagnoses were compared with the microradiography and expert observer ‘gold standard’ diagnoses for dentine caries. ?,Treatment ? decisions using microradiography; E%treatment decisions using expert observers; 8, diagnosis ‘probably dental caries’ from radiograph using microradiography.
fewer surfaces were in need of treatment than were indicated, by microradiography, as having outer dentine caries. A maximum of 19 surfaces were ‘overtreated’ and 37 ‘undertreated’by one dentist or another. On average, 2.4 surfaces were ‘overtreated’ whilst 17.2 of the 45 carious surfaces were missed or ‘undertreated’. A similar average frequency of errors occurred for the diagnosis by the practitioners of surfaces with ‘probably dentine caries’. False-positive errors outnumber false-negative errors when compared with the expert observers’ norm by approximately two to one The results of paired comparisons between the agreement of the dentists’ diagnosis ‘probably carious’ in the dentine and their decisions to treat surfaces or the judgement of the respective diagnostic standards are shown in Table II. The results show that the best diagnostic agreement (P < 0.005) occurs between the dentist’s decision that a filling is needed and his own perception of whether dentine caries was visible on the dental radiographs or not. This is an indication that selfcancelling errors may occur both through limitations in the radiographic technique and due to perceptual error.
134
J. Dent. 1990;
18: No. 3
Tab/e II. Mean relative diagnostic agreement for dentine caries between the various diagnostic measures Kappa (s.d.) Mean Dentists ‘probably carious’ microradiography (a) Filling ‘needed’/microradiography (b) Filling ‘needed’/expert observers Filling ‘needed’/dentine (c) caries as perceived by dentists (d)
0.53
(0.14)*
0.60 0.77
(0.1 1 )* (0.1 l)*
0.85
(0.13)*
Differences between the measures of agreement (a), (b). (c) and (d) have been tested on the original data using the chi-square test. All the measures of agreement differed significantly. x2 > 150, d.f. = 6, * = P < 0.005.
decisions are more accurate than diagnostic ability might indicate. This may mean that dentists subconsciously adapt their restorative criteria to compensate for their inability to diagnose dentine caries depth correctly. The radiographic diagnostic accuracy of practitioners as an indication of the efficiency of radiography as a diagnostic system is best presented in the form of an ROC curve since there is always a relationship between sensitivity and specificity which varies depending on the stringency of the diagnostic criteria used @wets and Pickett, 1980). Data for the varying degrees of diagnostic confidence of the 276 dentists are pooled in Fig. 5. This figure shows the mean relationship between sensitivity and specificity for caries in the outer half of dentine using microradiography as the ‘gold standard’ diagnosis. It Treatment
ROC graph
with
s .Y 2
40
Definitely
ov 0
carious /
microradiography
/ I
I
20 % False
as norm
40
positive
I
60
80
(l-specificity)
fig. 5. Receiver operating characteristic (ROC) curve for the accuracy when using bitewing radiographs of dentine caries recognition. The data of the 276 Dutch dental practitioners has been pooled and microradiography has been used as the ‘gold standard’ diagnosis. *Restorative treatment decision judgement superimposed: TP 61.7 per cent, FP 4.1 per cent.
should be born in mind that as sensitivity rises with increasing leniency of the decision-making cut-off point, so does the proportion of false-positive diagnoses (i.e. specificity declines). The plot points in this figure are, for definitely carious TP 39.5 per cent, FP 1.2 per cent, for probably carious TP 53.5 per cent, FP 3.4 per cent, for uncertain if carious TP 64.7 per cent, FP 7.3 per cent and for probably not carious TP 73.0 per cent, FP 12.4per cent. The resultant area under this ROC curve was AZ 0.88. Mean restorative treatment criterion has a position of TP 61.7 per cent, FP 4.1 per cent and is superimposed on the figure.
DISCUSSION Caveats that must accompany the interpretation of the results of this study include: the adequacy of the response; the quality of the duplicated radiographs; the appropriateness of caries in the outer half of the dentine according to the microradiographic technique as a standard for analysing diagnostic accuracy and treatment decision-making behaviour, and the representativeness of the sample of carious teeth. This was the second part of a two-wave questionnaire study. The first part used a computer-generated 10 per cent random sample of dentists and the subsequent response did not differ in age structure from the total population of dentists (Mileman et al., 1988). This group of respondents was surveyed on the second wave, resulting in a final response of 62 per cent of the total sample. It has been suggested that in postal surveys of dentists the results with a response of45 per cent or more do not differ significantly from those obtained using the total sample (Houland et al., 1980). This means that the response in this study can be considered satisfactory. The quality of duplicated radiographs was strictly controlled so that it is extremely unlikely to be the cause of variation between the practitioners. It is unlikely that the present study underestimates diagnostic variation or overestimates diagnostic accuracy in clinical practice, since in general practice the quality of bitewing radiographs is often inadequate (Mannyet al., 1980;McDonald, 1983, Nysether and Hansen, 1983). The difference in density between the duplicate and the original radiographs was small. Psychophysical studies (Erales and Manson-Hing, 1979; Price, 1980) of the differences between duplicate and original radiographs have concluded that the duplication of radiographs is ‘not likely to materially effect the clinical performance’ whilst the observers used ‘rated all duplicates superior to the original radiograph’ (Erales and Manson-Hing, 1979). Microradiography has been chosen as a ‘gold standard for radiographic caries diagnosis in the past (PurdellLewis et al., 1974; White et al., 1984; Josselin de Jong and Ten Bosch, 1985). It offers in one scale an objective measure of the degree of lesion progression towards the pulp and the associated chance of surface cavitation (Purdell-Lewis et al., 1974). Surface cavitation is assumed
Mileman
and van der Weele: Accuracy
to be synonymous with lesion irreversibility and is therefore the histological stage at which restorative treatment should probably be initiated. Most surfaces (94 per cent) with microradiographic dentine caries demonstrate lesion cavitation (Purdell-Lewis et al., 1974). It seems reasonable therefore to use dentine caries diagnosed microradiographically at least as a ‘benchmark’ for analysing dentist restorative treatment decision-making behaviour. The use of radiographic dentine caries according to the expert observers as a threshold from which to measure restorative treatment decision-making behaviour has support (Espelid and Tveit, 1986) although it is also to some degree subjective (Tulloch et al., 1988) and depends on caries prevalence. The selection and prevalence of lesions to form the test battery for a disease with low prevalence is a compromise between having enough cases of disease to provide reasonably reliable results for sensitivity and having enough cases without disease to reliably measure specificity, without being too onerous for the test participants. We were concerned essentially to choose a selection of disease cases within which the practitioners’ treatment decision range would be likely to fall. The average accuracy for the diagnosis of dentine caries was substantial (AZ = 0.88). Although it provides evidence for the systematic underdiagnosis of lesions on radiographs, it compares favourably with other estimates of diagnostic accuracy using selected observers (Ruttimann, 1987) and indicates that the practitioners were able to comprehend the task they had been set. Variation between practitioners in diagnostic accuracy as measured by the ‘Kappa’ statistic (Fig. 1) was however large. Not all the dental practitioners would wait until a lesion was in dentine before deciding to carry out restorative treatment. The variation in treatment decision making, combined with the mean ‘overtreatment’ of surfaces with dentine caries (Fig. 4) can be explained by the fact that about a quarter of the dentists could be considered to have criteria leading to an abnormally high frequency of prescribing restorative treatment need (Fig. 3). The suggestion that these criteria have been adopted in part to compensate for underdiagnosis (Fig. 5 average FN derived from the ROC curve was 38.8 per cent) requires further investigation. This study demonstrates that it cannot be assumed that the radiographic method of caries diagnosis currently used by dental practitioners can achieve optimal accuracy as a diagnostic test. Even the ‘experts’ used in this study could not diagnose seven of the 45 lesions in dentine at any degree of certainty. On average the practitioners underestimated dentine caries presence when compared with a microradiography or expert ‘gold standard’ (Table Z.Fig. 4). It is interesting to note however that their decisions to place restorations had generally less errors (Fig. 4) and showed significantly greater agreement with the ‘gold standards’ than their perception of carious lesion depth (Table II). This may
in radiographic
caries diagnosis
135
mean that in dental practice treatment criteria are partly derived from diagnostic ability. The majority of these dentists report having treatment criteria in the enamel (Mileman, 1985). However they may in fact treat dentine lesions which they perceive as being confined to the enamel. If this were so then recommendations to use stricter treatment criteria in the dentine might lead to an unacceptably high frequency of dentine lesions being left untreated despite a radiographic examination. The current risk of overtreatment should not be underestimated. For example, with an approximal dentine caries prevalence of say 5 per cent for the patient we have described (Grondahl et al., 1977) ‘over-treatment’decisions would be made just as often as true positive treatment decisions using the mean restorative treatment criterion fromFig. 5 (TP 61.7 per cent, FP 4.1 per cent: a likelihood ratio of 15 for the test). This is an indication that, at the time of this study, some dentists had a treatment philosophy which could lead in practice to the unnecessary restorative treatment of carious lesions confined to the enamel (DHSS, 1986). On the other hand basing recommendations concerning restorative treatment criteria on the diagnostic accuracy of experts (Mileman et al., 1988; Tulloch et al., 1988) should be done with some caution. False-positive treatment decisions might be reduced at the expense of an unacceptable rise in undertreatment with all its consequences. For this reason it is suggested that restorative treatment decision making and radiographic caries diagnosis processes should be taught in an integrated form in graduate and postgraduate dental courses (DHSS, 1986). Acknowledgements The authors would like to thank Dr D. J. Purdell-Lewis, Dr J. Bouma, Dr R. Liem, Dr A. C. M. van de Poel, Mr L. J. van Rijn and The Dutch Dental Association as well as the dentists who participated in the study.
References
Committee for Evaluating
Medical Technologies in Clinical Use, Institute of Medicine (1985) Assessing Medical Technologies. Washington, National Academy Press. pp. 70-175. Backer-Dirks O., Houwink B. and Kwant G. W. (1961) The results of 6% years of artificial fluoridation of drinking water in The Netherlands. Arch. Oral Sol. 5, 284-300. DHSS (1986) Report of the Committee of Enquiry into Unnecessary Dental Treatment. London. HMSO. Douglass C. W.. Valachovic R. W.. Wijesinha A. et al. (1986) Clinical efficacy of dental radiography in the detection of dental caries and periodontal diseases. Oral Surg. Oral Med. Oral Pathol. 62, 330-339. Downer M. C. (1989) Validation of methods used in dental caries diagnosis. Int. Dent. J. 39, 241-246. Erales F. A. and Manson-Hing L. R. (1979) A study of the quality of duplicated radiographs. Oral Surg. Oral Med. Oral Pafbol. 47, 98-104. Espelid I. (1987) Radiographic Diagnoses and Treatment Decisions on Approximal Caries. Bergen, University of Bergen.
136
J. Dent.
1990; 18: No. 3
Espelid I. and Tveit A. B. (1986) Clinical and radiographic assessment of approximal carious lesions. Acta Odon tel. Stand. 44, 31-37. Griindahl H-G., Hollender L.. Malmcrona E. et al. (1977) Dental caries and restorations in teenagers: II. A longitudinal radiographic study of the caries increment of proximal surfaces among urban teenagers in Sweden. Swed. Dent J. 1, 51-57. Houland E. S.. Romberg E. and Moreland E. F. (1980) Nonresponse bias to mail survey questionnaires within a professional population. J. Dent. Educ. 44,270-274. Josselin de Jong E. and Ten Bosch J. J. (1985) Error analysis of the microradiographic determination of mineral content in mineralized tissue slices. Phys. Med. Biol. 30, 1067-1075. Landis J. R. and Koch G. G. (1977) The measurement of observer agreement for categorical data. Biometrics 33, 157-174. Manny E. F., Carlson K. C., McClean P. M. et al. (1980) An Overview of Dental Radiology (Monograph Series). Rockville, Maryland, National Institute of Health Care Technology. McDonald S. P. (1983) A method to reduce interproximal overlapping and improve reproducibility of bitewing radiographs for clinical trials. Community Dent Oral Epidemiol. l&289-295. McNeil B. J.. Keeler E. and Adelstein S. J. (1975) Primer on certain elements of medical decision making. N. Engl. J. Med. 293, 211-215. Mejare I., GrBndahl H-G., Carlstedt K. et al. (1985) Accuracy at radiography and probing for the diagnosis of proximal caries. &and. J. Dent Res. 93, 178-184. Mileman P. A. (1985) Radiographic Caries Diagnosis and Restorative Treatment Decision Making. PhD Thesis, University of Groningen. Mileman P. A., van der Weele L. T., van de Poe1 A. C. M. et al. (1988) Dutch dentists’ decisions to take bitewing radiographs. Community Dent. Oral Epidemiol. 16, 368-373.
Nysether S. and Hansen B. F. (1983) Errors on bitewing radiographs. Community Dent. Oral Epidemiol. 11, 286-288. Okano T.. Huang H-J. and Nakamura T. (1985) Diagnostic accuracy on detection of proximal enamel lesions in nonscreen radiographic performance. Oral Surg. Oral Med. Oral Pathol. 59, 543-547. Pliskin J. S., Shwartz M., Grijndahl H-G. et al. (1985) Incorporating individual patient preferences in scheduling bitewing radiographs. Methods Inf: Med. 24, 213-217. Price C. (1980) A densitometric evaluation of two radiographic duplicating films under differing conditions of exposure and processing. Oral Surg. Oral Med. Oral Pathol. 50, 190- 194. Purdell-Lewis D. J., Groeneveld A, Pot T. J. et al. (1974) Proximal carious surfaces-a comparison of visual, radiographical and micro radiographic appearance. Ned. Tijdschr. Tandheelkd. 81; Nether&&s_ Dtf
