Background
Planar bone scintigraphy (BS) has been recommended for the assessment of bone metastasis in newly diagnosed prostate cancer for decades [
1,
2]. The reading of BS can be a timely task, and it may be biased by the extent and nature of the clinical data, the access to prior imaging results, and the experience of the readers [
3]. Computer-aided analysis is an objective method, which may improve the classification of M status and thus optimize patient management. EXINI Bone
BSI is a Food and Drug Administration-approved software for the analysis of planar BS with solid technical and clinical documentation [
4‐
10].
The bone scan index (BSI), the cumulated tumour load of malignant lesions in the entire skeleton are the most used endpoints with this software, which has been assessed in a number of clinical scenarios. The BSI has been shown to carry solid prognostic value for estimating the outcome and survival in metastatic, hormone-naïve and castration-resistant prostate cancer [
9‐
15] and to be valuable in the assessment of treatment response to anti-cancer treatments [
11,
16,
17]. The diagnostic value of the BSI for the staging of newly diagnosed prostate cancer has not been assessed so far.
The purpose of this study was to compare the diagnostic value of the BSI versus a panel of trained nuclear medicine experts for the classification of metastatic disease in a large prospective cohort of patients with newly diagnosed prostate cancer.
Discussion
A large number of publications have reported on the use of the computer-aided analysis of planar bone scans [
22‐
24]. Commercially available software, such as EXINI Bone
BSI, has become an established modality in both clinical settings and clinical trials for the objective and quantitative assessment of bone involvement in prostate cancer patients [
4‐
8]. The clinical value of lesional data, in particular the BSI, has been investigated in many trials in different clinical scenarios, including in the prognosis of prostate cancer [
11‐
15] and in the assessment of the treatment response to anti-cancer treatments [
11,
16,
25]. To the best of our knowledge, no previous reports have focused on the diagnostic role of the BSI in the staging of newly diagnosed prostate cancer. The key findings of our study were that EXINI Bone
BSI declared M1 disease on a large proportion of patients who were classified by experts as M0 or missed M1 disease, depending on the BSI cut-off values. These findings were confirmed in patients with a true reference.
Different BSI values have been proposed to indicate metastatic disease. A BSI > 0 has been proposed by several groups as an indicator of malignant involvement of the skeleton [
8,
12]. A BSI value of > 1 has been proposed as a limit to differentiate between impaired outcomes in prostate cancer patients without apparent metastasis [
11], minimal malignant bone involvement [
8], and an indicator of poor prognosis [
13]. We observed that a BSI > 0 correctly identified M1 patients who were identified by experts but classified a large proportion of M0 patients with metastasis. In contrast, a BSI > 1 only identified bone metastasis in approximately 60% of the patients who were diagnosed as having M1 disease by experts. Thus, neither cut-off value is applicable in clinical practice.
Even though BS is used in clinical medicine to classify the presence or absence of metastatic disease, consensus or expert reading is not a true reference test. We applied supplementary imaging and/or the biochemical response after radical prostatectomy to reach a true reference. A large proportion of patients could not be classified as true M0 or M1 due to confounding variables that interfered with correct classification at the time of the BS. This includes an androgen deprivation therapy, radiation therapy with androgen blockade, watchful waiting/active surveillance, and a lack of confirmative imaging. This is a drawback of the non-interventional design, but it reflects daily practice. Still, among more than 140 patients with a true reference, we found a low specificity with a BSI > 0 and a low sensitivity with a BSI > 1, similar to the findings of agreement with the experts.
The findings of several malignant lesions in patients who were classified as M0 by experts or true M0 may raise concern regarding the use of hot spot analysis for the assessment of treatment response in metastatic prostate cancer. The number of new lesions is a key factor among the Prostate Cancer Working Group criteria for disease progression in metastatic, castration-resistant prostate cancer [
26]. Kaboteh et al. used EXINI Bone
BSI to calculate new lesions and BSI values in patients who were treated with chemotherapy [
27]. In comparison to an increase in the BSI, the appearance of two new lesions shown with EXINI Bone
BSI did not correlate with overall survival. We consider it premature to use the number of malignant lesions from computer-assisted analysis in clinical decision-making based on the existing documentation.
The tested version of the software came with options to adjust settings based on the pre-test risk of metastasis. Patient-specific settings, based on EAU risk classification, did not improve the diagnostic performance of the software. These findings are in line with our recent study on the use of a patient-specific setting for the diagnostic properties of EXINI Bone
BSI on a patient level [
19]. This function has been removed in recent versions of the software.
The endpoints of this report, the analyses of lesional data, were different from the patient classification outcomes reported previously for the same population [
19]. Two issues favour publication of the lesional data in addition to the patient classification outcomes that were previously reported: first, lesional analyses, in particular the BSI, are the most frequently reported outcomes in studies with EXINI Bone
BSI; second, no clear association exists for the lesional analysis and patient classification. A total of 231 of 342 patients in this study were classified as definitive non-malignant (‘normal’) by software on a patient outcome classification, but 127 of these 231 ‘normal’ patients had at least one malignant lesion (and thus a BSI > 0). Sixty-seven patients had only one malignant lesion, 60 patients had more than one lesion, and four patients had 10 or more lesions. Thus, we find it relevant to report our data with the lesional data reported separately [
28].
We manually corrected for artefacts, which the software wrongly classified as malignant lesions in approximately 15% of the cases. Therefore, the presented results represent a combination of software capabilities plus manual corrections. Without the adjustment of artefacts, many more lesions were malignant and the BSI values would be higher, likely resulting in more misclassifications of M0 patients. Finally, it should be emphasized that the present findings are valid for DICOM file data only; it has previously been shown that the diagnostic performance of EXINI Bone
BSI varied with the file format [
19].
Acknowledgements
We thank EXINI, Lund, Sweden, for the installation and guidance on the use of EXINI BoneBSI free of charge at the involved trial sites for the duration of the study. We thank Helle H. Eriksen, Aalborg University Hospital Statistical Consultancy Group, for assistance with the statistical analyses.