Artificial intelligence and machine learning algorithms for early detection of skin cancer in community and primary care settings: a systematic review.

Skin cancers occur commonly worldwide. The prognosis and disease burden are highly dependent on the cancer type and disease stage at diagnosis. We systematically reviewed studies on artificial intelligence and machine learning (AI/ML) algorithms that aim to facilitate the early diagnosis of skin cancers, focusing on their application in primary and community care settings. We searched MEDLINE, Embase, Scopus, and Web of Science (from Jan 1, 2000, to Aug 9, 2021) for all studies providing evidence on applying AI/ML algorithms to the early diagnosis of skin cancer, including all study designs and languages. The primary outcome was diagnostic accuracy of the algorithms for skin cancers. The secondary outcomes included an overview of AI/ML methods, evaluation approaches, cost-effectiveness, and acceptability to patients and clinicians. We identified 14 224 studies. Only two studies used data from clinical settings with a low prevalence of skin cancers. We reported data from all 272 studies that could be relevant in primary care. The primary outcomes showed reasonable mean diagnostic accuracy for melanoma (89·5% [range 59·7-100%]), squamous cell carcinoma (85·3% [71·0-97·8%]), and basal cell carcinoma (87·6% [70·0-99·7%]). The secondary outcomes showed a heterogeneity of AI/ML methods and study designs, with high amounts of incomplete reporting (eg, patient demographics and methods of data collection). Few studies used data on populations with a low prevalence of skin cancers to train and test their algorithms; therefore, the widespread adoption into community and primary care practice cannot currently be recommended until efficacy in these populations is shown. We did not identify any health economic, patient, or clinician acceptability data for any of the included studies. We propose a methodological checklist for use in the development of new AI/ML algorithms to detect skin cancer, to facilitate their design, evaluation, and implementation.

Construct validity of the anglicised FACE-Q skin cancer module.

OBJECTIVES: The FACE-Q Skin Cancer module is a patient-reported outcome measure (PROM) for facial skin cancer. It has been anglicised for the UK population and undergone psychometric testing using classical test theory. In this study, further evaluation of construct validity using Rasch measurement theory and hypothesis testing was performed. METHODS: Patients were prospectively recruited to the Patient-Reported Outcome Measures In Skin Cancer Reconstruction (PROMISCR) study and asked to complete the anglicised FACE-Q Skin Cancer module. The scalability and unidimensionality of the data were assessed with a Mokken analysis prior to Rasch analysis. Response thresholds, targeting, fit statistics, local dependency, and internal consistency were examined for all items and subscales. Four a priori hypotheses were tested to evaluate the convergent and divergent validity. We additionally hypothesised that the median 'cancer worry' score would be lower in post-operative than pre-operative patients. RESULTS: 239 patients self-completed the questionnaire between August 2017 and May 2019. Of the ten subscales assessed, five showed relative fit to the Rasch model. Unidimensionality was present for all five subscales, with most demonstrating ordered item thresholds and appropriate fit statistics. Two items in the 'cancer worry' subscale had either disordered or very close response thresholds. Subscales of the FACE-Q Skin Cancer module demonstrated convergent and divergent validity with relevant Skin Cancer Index comparators (p < 0.001). Median 'cancer worry' was lower in post-operative patients (44 vs 39, p < 0.001). CONCLUSION: The anglicised FACE-Q Skin Cancer module shows psychometric validity through hypothesis testing, and both classical and modern test theory.

Observational study to estimate the proportion of surgical site infection following excision of ulcerated skin tumours (OASIS study).

BACKGROUND: Ulceration is a recognized risk factor for surgical site infection (SSI); however, the proportion of patients developing SSI after excision of an ulcerated skin cancer is unknown. AIM: To determine the proportion of participants with SSI after surgical excision of an ulcerated skin cancer. A secondary aim was to assess feasibility outcomes to inform the design of a randomized controlled trial to investigate the benefits and harms of perioperative antibiotics following excision of ulcerated tumours. METHODS: This was a multicentre, prospective, observational study of patients undergoing excision of an ulcerated skin cancer between March 2019 and March 2020. Prior to surgical excision, surface swabs of the ulcerated tumours of participants recruited from one centre were undertaken to determine organism growth. At 4 weeks after surgery, all participants were e-mailed or posted the Wound Healing Questionnaire (WHQ) to determine whether they had developed SSI. RESULTS: In total, 148 participants were recruited 105 (70.9%) males; mean ± SD age 77.1 ± 12.3 years. Primary outcome data were available for 116 (78.4%) participants, of whom 35 (30.2%) were identified as having an SSI using the WHQ with a cutoff score of 8, and 47 (40.5%) were identified with a cutoff score of 6. Using the modified WHQ in participants with wounds left to heal by secondary intention, 33 (28.4%) and 43 (37.1%) were identified to have SSI respectively. CONCLUSION: This prospective evaluation of SSI identified with the WHQ following excision of ulcerated skin cancers demonstrated a high proportion with SSI. The WHQ was acceptable to patients; however, further evaluation is required to ensure validity in assessing skin wounds.

Total body photography for the diagnosis of cutaneous melanoma in adults: a systematic review and meta-analysis.

BACKGROUND: Early detection of melanoma is essential to reduce mortality. Total body photography (TBP) can facilitate the detection of melanoma in high-risk individuals. However, the accuracy of TBP in diagnosing melanoma remains unclear. OBJECTIVES: To determine the diagnostic accuracy of TBP for the detection of melanoma in adults. METHODS: MEDLINE, Embase, Cochrane and Centre for Reviews databases were searched from inception to 26 May 2020. Studies that included TBP for diagnosing melanoma with at least one follow-up appointment were eligible for inclusion in the systematic review if they provided data to calculate at least one diagnostic accuracy measure. Two authors independently screened articles, extracted data and assessed quality. Disagreements were resolved by a third reviewer. RESULTS: In total, 10 studies were included, comprising 41 703 patients who underwent TBP and 6203 biopsies. Melanoma in situ (MIS) was diagnosed in 315 (5·1%) lesions and invasive melanoma was diagnosed in 187 (3·0%) lesions biopsied. Summary estimates for TBP in diagnosing melanoma were calculated as follows: mean percentage of biopsies positive for MIS or melanoma was 15% [95% confidence interval (CI) 10-21], number needed to biopsy (NNB) was 8·6 (range 2·3-19·6), naevus : melanoma ratio was 7·6 (range 1·3-18·6), and MIS : melanoma ratio was 1·7 (1·0-3·5). Regression analysis showed a negative correlation between NNB and MIS : melanoma ratio. CONCLUSIONS: Available data regarding the diagnostic accuracy of TBP are heterogeneous, owing to variability in the risk profile of cohorts and TBP protocols. Best current estimates suggest that TBP for diagnosing melanoma has an acceptable NNB in high-risk patients. However, prospective diagnostic test accuracy studies are needed to gauge the diagnostic accuracy of TBP.

A 10-year review of surgical management of dermatofibrosarcoma protuberans.

BACKGROUND: Dermatofibrosarcoma protuberans (DFSP) is a rare skin cancer. Standard treatment in the UK is either wide local excision (WLE) or Mohs micrographic surgery (MMS). It is unclear which approach has the lower recurrence rate. OBJECTIVES: We undertook a retrospective comparative review of surgical management of DFSP in the UK National Health Service in order to define (i) current surgical practice for primary and recurrent DFSP, (ii) local recurrence rates for primary DFSP and (iii) survival outcomes for DFSP. METHODS: A retrospective clinical case-note review of patients with histologically confirmed DFSP (January 2004 to December 2013) who have undergone surgical treatment. RESULTS: The surgical management of 483 primary and 64 recurrent DFSP in 11 plastic surgery and 15 dermatology departments was analysed. Almost 75% of primary DFSP (n = 362) were treated with WLE and 20% (n = 97) with MMS. For recurrent DFSP, 69% (n = 44) and 23% (n = 15) of patients underwent WLE and MMS, respectively. Recurrent primary DFSP occurred in six patients after WLE and none after MMS. The median follow-up time was 25·5 months (interquartile range 6·8-45·1) for new and 19·8 (IQR 4·5-44·5) for recurrent DFSP [Correction added on 1 Feb 2021, after first online publication: 4.8 years (interquartile range 3.5-5.8) was incorrect], with eight reported deaths during the follow-up analysis period (one confirmed to be DFSP related). CONCLUSIONS: WLE was the most common surgical modality used to treat DFSP across the UK. The local recurrence rate was very low, occurring only after WLE. Although a prospective randomized controlled trial may provide more definitive outcomes, in the absence of a clearly superior surgical modality, treatment decisions should be based on patient preference, clinical expertise and cost.

Development of a core outcome set for cutaneous squamous cell carcinoma trials: identification of core domains and outcomes.

BACKGROUND: The lack of uniformity in the outcomes reported in clinical studies of the treatment of cutaneous squamous cell carcinoma (cSCC) complicates efforts to compare treatment effectiveness across trials. OBJECTIVES: To develop a core outcome set (COS), a minimum set of agreed-upon outcomes to be measured in all clinical trials of a given disease or outcome, for the treatment of cSCC. METHODS: One hundred and nine outcomes were identified via a systematic literature review and interviews with 28 stakeholders. After consolidation of this long list, 55 candidate outcomes were rated by 19 physician and 10 patient stakeholders, in two rounds of Delphi exercises. Outcomes scored 'critically important' (score of 7, 8 or 9) by ≥ 70% of patients and ≥ 70% of physicians were provisionally included. At the consensus meeting, after discussion and voting of 44 international experts and patients, the provisional list was reduced to a final core set, for which consensus was achieved among all meeting participants. RESULTS: A core set of seven outcomes was finalized at the consensus meeting: (i) serious or persistent adverse events, (ii) patient-reported quality of life, (iii) complete response, (iv) partial response, (v) recurrence-free survival, (vi) progression-free survival and (vii) disease-specific survival. CONCLUSIONS: In order to increase the comparability of results across trials and to reduce selective reporting bias, cSCC researchers should consider reporting these core outcomes. Further work needs to be performed to identify the measures that should be reported for each of these outcomes.

Should we be imaging lymph nodes at initial diagnosis of early-stage mycosis fungoides? Results from the PROspective Cutaneous Lymphoma International Prognostic Index (PROCLIPI) international study.

BACKGROUND: Early-stage mycosis fungoides (MF) includes involvement of dermatopathic lymph nodes (LNs) or early lymphomatous LNs. There is a lack of unanimity among current guidelines regarding the indications for initial staging imaging in early-stage presentation of MF in the absence of enlarged palpable LNs. OBJECTIVES: To investigate how often imaging is performed in patients with early-stage presentation of MF, to assess the yield of LN imaging, and to determine what disease characteristics promoted imaging. METHODS: A review of clinicopathologically confirmed newly diagnosed patients with cutaneous patch/plaque (T1/T2) MF from PROspective Cutaneous Lymphoma International Prognostic Index (PROCLIPI) data. RESULTS: PROCLIPI enrolled 375 patients with stage T1/T2 MF: 304 with classical MF and 71 with folliculotropic MF. Imaging was performed in 169 patients (45%): 83 with computed tomography, 18 with positron emission tomography-computed tomography and 68 with ultrasound. Only nine of these (5%) had palpable enlarged (≥ 15 mm) LNs, with an over-representation of plaques, irrespectively of the 10% body surface area cutoff that distinguishes T1 from T2. Folliculotropic MF was not more frequently imaged than classical MF. Radiologically enlarged LNs (≥ 15 mm) were detected in 30 patients (18%); only seven had clinical lymphadenopathy. On multivariate analysis, plaque presentation was the sole parameter significantly associated with radiologically enlarged LNs. Imaging of only clinically enlarged LNs upstaged 4% of patients (seven of 169) to at least IIA, whereas nonselective imaging upstaged another 14% (24 of 169). LN biopsy, performed in eight of 30 patients, identified N3 (extensive lymphomatous involvement) in two and N1 (dermatopathic changes) in six. CONCLUSIONS: Physical examination was a poor determinant of LN enlargement or involvement. Presence of plaques was associated with a significant increase in identification of enlarged or involved LNs in patients with early-stage presentation of MF, which may be important when deciding who to image. Imaging increases the detection rate of stage IIA MF, and identifies rare cases of extensive lymphomatous nodes, upstaging them to advanced-stage IVA2.

Characteristics associated with significantly worse quality of life in mycosis fungoides/Sézary syndrome from the Prospective Cutaneous Lymphoma International Prognostic Index (PROCLIPI) study.

BACKGROUND: Mycosis fungoides (MF) and Sézary Syndrome (SS) are the most common cutaneous T-cell lymphomas. MF/SS is accompanied by considerable morbidity from pain, itching and disfigurement. AIM: To identify factors associated with poorer health-related quality of life (HRQoL) in patients newly diagnosed with MF/SS. METHODS: Patients enrolled into Prospective Cutaneous Lymphoma International Prognostic Index (PROCLIPI; an international observational study in MF/SS) had their HRQoL assessed using the Skindex-29 questionnaire. Skindex-29 scores were analysed in relation to patient- and disease-specific characteristics. RESULTS: The study population consisted of 237 patients [60·3% male; median age 60 years, (interquartile range 49-70)], of whom 179 had early MF and 58 had advanced MF/SS. In univariate analysis, HRQoL, as measured by Skindex-29, was worse in women, SS, late-stage MF, those with elevated lactate dehydrogenase, alopecia, high modified Severity Weighted Assessment Tool and confluent erythema. Linear regression models only identified female gender (ß = 8·61; P = 0·003) and alopecia (ß = 9·71, P = 0·02) as independent predictors of worse global HRQoL. Item-level analysis showed that the severe impairment in symptoms [odds ratio (OR) 2·14, 95% confidence interval (CI) 1·19-3·89] and emotions (OR 1·88, 95% CI 1·09-3·27) subscale scores seen in women was caused by more burning/stinging, pruritus, irritation and greater feelings of depression, shame, embarrassment and annoyance with their diagnosis of MF/SS. CONCLUSIONS: HRQoL is significantly more impaired in newly diagnosed women with MF/SS and in those with alopecia. As Skindex-29 does not include existential questions on cancer, which may cause additional worry and distress, a comprehensive validated cutaneous T-cell lymphoma-specific questionnaire is urgently needed to more accurately assess disease-specific HRQoL in these patients. What's already known about this topic? Cross-sectional studies of mixed populations of known and newly diagnosed patients with mycosis fungoides (MF)/Sézary syndrome (SS) have shown significant impairment in health-related quality of life (HRQoL). Previous studies on assessing gender-specific differences in HRQoL in MF/SS are conflicting. More advanced-stage disease and pruritus is associated with poorer HRQoL in patients with MF/SS. What does this study add? This is the first prospective study to investigate HRQoL in a homogenous group of newly diagnosed patients with MF/SS. In patients newly diagnosed with MF/SS, HRQoL is worse in women and in those with alopecia and confluent erythema. MF/SS diagnosis has a multidimensional impact on patient HRQoL, including a large burden of cutaneous symptoms, as well as a negative impact on emotional well-being.

Use of the Skin Cancer Quality of Life Impact Tool (SCQOLIT) - a feasibility study in non-melanoma skin cancer.

BACKGROUND: Incidence of non-melanoma skin cancers (NMSCs) is increasing and can significantly impact on quality of life (QOL), yet there are few studies evaluating patient-reported outcome measures (PROMs) in NMSC populations. We undertook a prospective feasibility study to evaluate a skin cancer-specific PROM, the Skin Cancer Quality of Life Impact Tool (SCQOLIT), in patients with a new diagnosis of NMSC. OBJECTIVES: (i) To establish acceptability of SCQOLIT in dermatology clinics, (ii) a descriptive analysis of SCQOLIT scores in NMSC. METHODS: Patients with histologically confirmed NMSC completed SCQOLIT, EQ-5D and a transition item. Questionnaires were completed at baseline and 3 months for group 1 ('low-risk' NMSC) and group 2 ('high-risk' NMSC) with additional questionnaires at 6-9 months for group 2. Patients participated in structured interviews. Clinician experience was captured through staff evaluation forms and a focus group. Acceptability and psychometric properties of SCQOLIT were assessed. RESULTS: Overall, 318 patients consented to participate. Mean SCQOLIT score at baseline was 5.33, with 2.6% of patients scoring ≥20. No ceiling effects were observed, whilst 13.9% scored 0. Validity was demonstrated against EQ-5D. Cronbach's alpha 0.84 demonstrated internal consistency. Thirteen patients were interviewed and thought SCQOLIT was comprehensive, captured impact on health-related QOL and helped express their needs to clinicians. Most clinicians found SCQOLIT 'very useful' or 'useful to some extent' in facilitating discussions. CONCLUSIONS: This feasibility study demonstrates that SCQOLIT is acceptable to patients and staff in dermatology skin cancer clinics. The psychometric properties of SCQOLIT confirm its utility in NMSC populations.

Patient preferences for topical treatment of actinic keratoses: a discrete-choice experiment.

BACKGROUND: The treatment of actinic keratosis (AK) is a potentially effective strategy for the prevention of cutaneous squamous cell carcinoma (cSCC). However, the patient perspective on potential benefits of AK treatment in terms of skin cancer reduction has received little attention to date. OBJECTIVES: (i) To investigate patient preferences for topical treatments for AK using a discrete-choice experiment (DCE); (ii) to evaluate patient willingness to trade between clinical benefit and medical burden. METHODS: The DCE was conducted as part of a study to establish the feasibility of a phase III randomized controlled trial evaluating the prevention of cSCC using currently available topical interventions. Preferences were elicited by asking patients to make a series of choices between treatment alternatives with different hypothetical combinations of attribute levels. Willingness to trade between treatment attributes was estimated using a flexible-choice model that allows for the heterogeneity of patient preferences. RESULTS: A total of 109 patients with AK completed the DCE. The majority of patients who expressed valid preferences were willing to accept some reduction in both prophylactic and cosmetic efficacy to reduce the burden of the treatment regimen, the severity of skin reaction and other adverse effects. Patients may reject treatment if the perceived therapeutic benefit is outweighed by the subjective burden of treatment. CONCLUSIONS: Evidence of significant variation in the perceived utility of treatments across patients highlights the importance of taking individual patient preferences into account to improve AK treatment acceptability and adherence.

Cutaneous hyperpigmentation and familial gastrointestinal stromal tumour associated with KIT mutation.

Gastrointestinal stromal tumours (GISTs) are mesenchymal tumours arising in the gastrointestinal tract. Early detection, before metastasis occurs, is important as complete surgical excision achieves cure. Approximately 85% of GISTs are associated with mutations in the KIT gene, and although the majority of GISTs are sporadic, familial GISTs have been identified. Several families with multiple GIST tumours have also been described with various cutaneous findings including hyperpigmentation, multiple lentigines, vitiligo and urticaria pigmentosa. We discuss a 6-year-old boy who presented with an unusual pattern of hyperpigmentation in association with a family history of GIST. A causative KIT mutation was identified in DNA from the pigmented skin and from the resected GIST, and the patient was referred to the Paediatric Gastroenterology department for GIST screening. The term 'GIST cutaneous hyperpigmentation disease' has been suggested previously for the association of familial GIST with cutaneous hyperpigmentation caused by a germline KIT mutation.

Do perioperative antibiotics reduce the risk of surgical-site infections following excision of ulcerated skin cancers? A Critically Appraised Topic.

AIM: To review the efficacy of perioperative antibiotics in reducing the risk of surgical-site infections (SSIs) following excision of ulcerated skin cancers. SETTING AND DESIGN: Study selection, data extraction and analysis were carried out independently by four authors. Only randomized controlled trials (RCTs) reported in the English language were included. INCLUDED STUDIES: RCTs in the English language in which patients received perioperative topical, intralesional or oral antibiotics for dermatological surgery, including Mohs micrographic surgery in general practice, dermatology or plastic surgery departments, were included. OUTCOME: The proportion of participants developing SSI following excision of skin lesions. RESULTS: Thirteen RCTs were identified from our literature search of PubMed and Embase, which evaluated SSI following use of topical (n = 5), oral (n = 3), intramuscular (n = 2), intravenous (n = 1) and intralesional antibiotics (n = 2) in dermatological surgery. Two RCTs specifically investigated SSIs in ulcerated skin cancer excisions; one RCT investigated the SSI rate following surgical treatment specifically for ulcerated skin cancers in individuals randomized to topical antibiotics vs. oral cephalexin; and one RCT compared intravenous cefazolin with no antibiotic, demonstrating significant reduction in SSI rates for ulcerated tumours (P = 0·04). CONCLUSIONS: The heterogeneity of the RCTs included in this study makes it difficult to make a direct comparison of the outcomes measured. High-quality evidence demonstrating a beneficial effect of the use of perioperative antibiotics to prevent SSI following excision of ulcerated skin cancers is lacking. In the absence of an evidence base, we propose that a well-designed multicentre RCT could evaluate the effect of perioperative antibiotics following excision of ulcerated tumours, and potentially reduce inappropriate antibiotic prescription.

Invasive melanoma in vivo can be distinguished from basal cell carcinoma, benign naevi and healthy skin by canine olfaction: a proof-of-principle study of differential volatile organic compound emission.

BACKGROUND: Volatile organic compounds (VOCs) are continuously released by the body during normal metabolic processes, but their profiles change in the presence of cancer. Robust evidence that invasive melanoma in vivo emits a characteristic VOC signature is lacking. OBJECTIVES: To conduct a canine olfactory, proof-of-principle study to investigate whether VOCs from invasive melanoma are distinguishable from those of basal cell carcinoma (BCC), benign naevi and healthy skin in vivo. METHODS: After a 13-month training period, the dog's ability to discriminate melanoma was evaluated in 20 double-blind tests, each requiring selection of one melanoma sample from nine controls (three each of BCC, naevi and healthy skin; all samples new to the dog). RESULTS: The dog correctly selected the melanoma sample on nine (45%) occasions (95% confidence interval 0·23-0·68) vs. 10% expected by chance alone. A one-sided exact binomial test gave a P-value of < 0·01, supporting the hypothesis that samples were not chosen at random but that some degree of VOC signal from the melanoma samples significantly increased the probability of their detection. Use of a discrete-choice model confirmed melanoma as the most influential of the recorded medical/personal covariates in determining the dog's choice of sample. Accuracy rates based on familiar samples during training were not a reliable indicator of the dog's ability to distinguish melanoma, when confronted with new, unknown samples. CONCLUSIONS: Invasive melanoma in vivo releases odorous VOCs distinct from those of BCC, benign naevi and healthy skin, adding to the evidence that the volatile metabolome of melanoma contains diagnostically useful biomarkers.