Accuracy of SIAscopy for pigmented skin lesions encountered in primary care: development and validation of a new diagnostic algorithm

Abstract

Background: Diagnosing pigmented skin lesions in general practice is challenging. SIAscopy has been shown to increase diagnostic accuracy for melanoma in referred populations. We aimed to develop and validate a scoring system for SIAscopic diagnosis of pigmented lesions in primary care.

Methods: This study was conducted in two consecutive settings in the UK and Australia, and occurred in three stages: 1) Development of the primary care scoring algorithm (PCSA) on a sub-set of lesions from the UK sample; 2) Validation of the PCSA on a different sub-set of lesions from the same UK sample; 3) Validation of the PCSA on a new set of lesions from an Australian primary care population. Patients presenting with a pigmented lesion were recruited from 6 general practices in the UK and 2 primary care skin cancer clinics in Australia. The following data were obtained for each lesion: clinical history; SIAscan; digital photograph; and digital dermoscopy. SIAscans were interpreted by an expert and validated against histopathology where possible, or expert clinical review of all available data for each lesion.

Results: A total of 858 patients with 1,211 lesions were recruited. Most lesions were benign naevi (64.8%) or seborrhoeic keratoses (22.1%); 1.2% were melanoma. The original SIAscopic diagnostic algorithm did not perform well because of the higher prevalence of seborrhoeic keratoses and haemangiomas seen in primary care. A primary care scoring algorithm (PCSA) was developed to account for this. In the UK sample the PCSA had the following characteristics for the diagnosis of 'suspicious': sensitivity 0.50 (0.18-0.81); specificity 0.84 (0.78-0.88); PPV 0.09 (0.03-0.22); NPV 0.98 (0.95-0.99). In the Australian sample the PCSA had the following characteristics for the diagnosis of 'suspicious': sensitivity 0.44 (0.32-0.58); specificity 0.95 (0.93-0.97); PPV 0.52 (0.38-0.66); NPV 0.95 (0.92-0.96). In an analysis of lesions for which histological diagnosis was available (n = 111), the PCSA had a significantly greater Area Under the Curve than the 7-point checklist for the diagnosis of melanoma (0.83; 95% CI 0.71-0.95 versus 0.61; 95% CI 0.44-0.78; p = 0.02 for difference).

Conclusions: The PCSA could have a useful role in improving primary care management of pigmented skin lesions. Further work is needed to develop and validate the PCSA in other primary care populations and to evaluate the cost-effectiveness of GP management of pigmented lesions using SIAscopy.

Figure 1

Example SIAscans, with labelled features, of a) benign naevus b) malignant melanoma c) seborrhoeic keratosis d) haemangioma.

Figure 2

ROC curves to show development of Primary Care Scoring Algorithm; ROC for diagnosis of melanoma after removal of lesions classified on SIAscopy as haemangioma or seborrhoeic keratosis.

Figure 3

Primary Care Scoring Algorithm (PCSA).

Figure 4

Comparative ROC curves for the Primary Care Scoring Algorithm and 7-point checklist for lesions with a histological diagnosis (n = 111; 10 melanoma and 2 lentigo maligna).