idiopathic osteosclerosis, condensing osteitis and osteosarcoma - as the subjects for this study, because of the similarity of their radiographic appearance1·2.
1
U T T E R W O R T H
Dentomaxillofac.
Ε I Ν Ε Μ Α Ν Ν
Radiol.,
Vol. 24, N o . 3, pp. 155-159, 1995
Copyright © 1995 Elsevier Science Ltd. for the I A D M F R Printed in G r e a t Britain. All rights reserved
0250-832Χ(94)00006-9
0250-832X/95 $10.00 + 0.00
Does radiographic feature recognition contribute to dentists' diagnosis of pathology? S.E. Stheeman, P.A. Mileman, M.A. van't Hof* and P.F. van der Stelt Department of Oral Radiology, Academic Centre for Dentistry Amsterdam, Amsterdam and ' D e p a r t m e n t of Medical Informatics and Epidemiology, Catholic University of Nijmegen, Nijmegen, The Netherlands
Received 14 July 1994 and in final form 2 November 1994 Objectives. To measure the degree to which dentists can accurately identify radiographic features and use their findings in the diagnosis of three lesions. Methods. Fourteen radiographic examples were used in this study and five features derived from the literature: border sharpness, border radiopacity, lesion shape, lesion radiopacity and root contact. A 'silver standard' for the presence of the features in the radiographic images of the lesions was established by expert observers; and a diagnostic 'gold standard' based on the histopathology of the lesions was available. Dentists scored each image according to these features and gave their diagnosis. Results. When dentists made different subjective diagnoses for a case, significant differences were found in the extent to which features were reported to be present. The degrees of presence of border sharpness (21%) and border radiopacity (46%) measured using the 'silver standard' were significantly related to the diagnosis of osteosarcoma. Conclusions. Radiographic information is not used to its full extent and improvement in the recognition of features and the cognitive use of this radiographic information in the diagnostic process could improve diagnostic accuracy. Keywords: radiography, dental; decision making; diagnosis; diagnosis, oral Dentomaxillofac. Radiol., Vol. 24, No. 3,155-159,1995
Dentists obtain periapical radiographs as a diagnostic aid in the management of dental disease. Diagnosis is complicated by the fact that a number of lesions have a similar radiographic appearance at first sight 1 . Despite this apparent similarity, lesions can differ extensively in their aetiology, treatment and prognosis 1 . Textbooks of oral radiology use certain radiographic features to describe their appearance 1 · 2 , e.g. an osteosarcoma is said to be "typically . . . unicentric and the borders are ill defined, suggestive of its malignant nature" 1 . This use of verbal descriptions of radiographic features assumes that the dentist can accurately recognize such features on a radiograph 3 and also that they will use these features in making a diagnosis 4 in a manner that is based on knowledge of the relationship between the features and the likelihood of a diagnosis 5 · 6 . Both these f a c e t s can b e e x p e c t e d t o i n f l u e n c e d i a g n o s t i c accuracy 7 · 8 . Interobserver variation in the interpretation of radiographs is considerable 9 . Variation in diagnosis has in part been attributed to variation in the interpretation of the meaning of commonly used terminology in the textbooks. For example, the interpretation of probabi-
listic terminology describing the strength of the relationship between features and disease has been shown to vary considerably from one dentist to another 1 0 . In the medical literature it has been suggested that recognition of the features themselves is a main reason, for instance, for the variation in the radiographic diagnosis of lung disease 1 1 ; these authors recommended that feature recognition as a source of diagnostic error should be the subject of further research. We carried out this study because the importance of feature recognition is assumed to be central to radiographic diagnosis. We used features described in the oral radiology literature for three periapical bone lesions with fundamentally different treatment and prognosis 1 . We intended: 1. to measure the degree to which dentists are able to accurately identify the degree to which five radiographic features frequently mentioned in the literature on oral radiographic diagnosis were present; 2. to test the importance of the recognition of five radiographic features to dentists' subjective diagnosis of three periapical bone lesions with substantially different implications for treatment and prognosis; and Dentomaxillofac. Radiol., Vol. 24, No. 3, 1995
155
Radiographic feature use in diagnosis of pathology: S.E. Stheeman et al. 3. to test the relationship between radiographic features and the validated diagnosis of three periapical bone lesions.
Material and methods We chose three types of lesions found in the mandible idiopathic osteosclerosis, condensing osteitis and osteosarcoma - as the subjects for this study, because of the similarity of their radiographic appearance 1 · 2 . We selected 14 periapical radiographs, each containing an image of one of these three lesions. For each case a 'gold standard diagnosis' was established. We used five radiographic features found in the literature on oral radiology to describe the three types of lesions. A 'radiographic feature silver standard' was established for each feature, which expressed the degree to which it was present on each radiograph. Fifty-five general dental practitioners (GDPs) participating in postgraduate dental education took part in the study. They were given a short lecture on the three lesions that had been selected and a general description of a patient. We asked them to view and diagnose the 14 radiographs and to assess the degree of presence of each of the features, by means of a short questionnaire for each radiograph.
The radiographs and the gold standard diagnoses We selected the lesions idiopathic osteosclerosis (10), condensing osteitis (CO) and osteosarcoma (OS) for this study because they have fundamentally different consequences for treatment and prognosis despite their similar radiographic appearance 1 · 212 " 14 . The radiographs used were obtained from the archives of the Department of Oral Radiology, University of Göteborg, Sweden. The set of radiographs comprised two cases of idiopathic osteosclerosis, six of condensing osteitis and two of chondroblastic osteogenic sarcoma all certified by histopathology (Gröndahl H-G., personal communication, 1993). The histopathology was used as the gold standard (GS) for the radiographs. Four additional radiographs were obtained from the Department of Oral Radiology, Academic Centre for Dentistry Amsterdam, The Netherlands. The GS for these four radiographs was a consensus diagnosis reached through a Delphi agreement procedure 15 by a team of expert oral radiologists (hereafter termed the Delphi panel) in a manner reported in an earlier study 16 . The Delphi panel reached full agreement on the diagnosis of idiopathic osteosclerosis for all four radiographs after three rounds of anonymous written correspondence.
The radiographic features and the radiographic feature silver standard We selected five radiographic features from the literature on oral radiology which had been used to describe 156
Dentomaxillofac. Radiol., Vol. 24, No. 3, 1995
the three types of lesions we had chosen 1 . The features were: 1. 2. 3. 4. 5.
border sharpness (0-100% well defined); border radiopacity (0-100% radiopaque); lesion shape (0-100% smooth); lesion radiopacity (0-100% radiopaque); root contact (contact or certainly no contact).
In order to reach consensus on the degree of presence of Features 1 to 4, the Delphi panel assessed the radiographs in a second Delphi consensus procedure. For each radiograph the panel reported the degree to which the feature was present on visual analog scales by placing a cross on a line 10 cm long representing a scale from zero to 100% 17 . The data were collected and 25% and 75% quartiles calculated. In order to encourage convergence these were marked on a new set of visual analog scales and then returned to the Delphi panel who judged the presence of the features again. For each feature the median of these assessments was used in the radiographic feature silver standard (RFSS). The panel's opinions were also elicited for the feature 'root contact', summarized and returned anonymously to its members. They were then asked to rejudge their original responses when they did not fully agree, and to give reasons for their decisions, as well as to comment on the judgements of their fellow panel members. After these results had been collected, the Delphi panel was again asked to reconsider the few cases where disagreement remained. The third round response was considered final and was used to complete the RFSS. In three cases of 5 * 14 possible assessments a slight difference of opinion among panel members remained and the judgement of the majority was used.
Interview format Dentists may interpret the meaning of phrases very differently 10 . Before the investigation, therefore, we gave a short lecture to the participating GDPs to introduce the standard terminology to be used to identify the lesions. A general description of each lesion was given and, at the same time, a radiographic example shown on a slide. Both descriptions and radiographic examples had been obtained from the literature 1 . Diagnostic variation could be caused by different expectations of the extent to which a lesion might be present in a particular patient. To avoid this, the dentists were given a general description of the patient: "A healthy adult has recently moved house. They visit a new dentist for a check-up but have no dental complaints". The GDPs viewed the periapical radiographs on a viewbox and completed a short questionnaire for each radiograph. From this the following information was obtained: 1. the diagnosis; 2. the degree of presence of four radiographic features, and
Radiographic feature use in diagnosis of pathology: S.E. Stheeman et al. 03 Ο c
