Background

Basal cell carcinoma (BCC) is the most common human cancer. Facial BCCs most commonly occur on the nose and the management of these lesions is particularly complex, given the functional and complex implications of treatment. Multidisciplinary team (MDT) meetings are routinely held to integrate expertise from dermatologists, surgeons, oncologists, radiologists, pathologists and allied health professionals. The aim of this research was to develop a supervised machine-learning algorithm to predict MDT recommendations for nasal BCC to potentially reduce MDT caseload, provide automatic decision support and permit data audit in a health service context.

Methods

The study population included all consecutive patients who were discussed at skin cancer-specialised MDT (SSMDT) with a diagnosis of nasal BCC between January 1, 2015 and December 31, 2015. We conducted analyses for gender, age, anatomical location, histological subtype, tumour size, tumour recurrence, anticoagulation, pacemaker, immunosuppressants and therapeutic modalities (Mohs surgery, conventional excision or radiotherapy). We used S-statistic computing language to develop a supervised machine-learning algorithm.

Results

We found that 37.5% of patients could be reliably predicted to be triaged to Mohs micrographic surgery (MMS), based on tumour location and age. Similarly, the choice of conventional treatment (surgical excision or radiotherapy) by the MDT could be reliably predicted based on the patient’s age, tumour phenotype and lesion size. Accordingly, the algorithm reliably predicted the MDT decision outcome of 45.1% of nasal BCCs.

Conclusions

Our study suggests that the machine-learning approach is a potentially useful tool for predicting MDT decisions for MMS vs conventional surgery or radiotherapy for a significant group of patients. We suggest that utilising this algorithm gives the MDT more time to consider more complex patients, where multiple factors, including recurrence, financial costs and cosmetic outcome, contribute to the final decision, but cannot be reliably predicted to determine that outcome. This approach has the potential to reduce the burden and improve the efficiency of the specialist skin MDT and, in turn, improve patient care, reduce waiting times and reduce the financial burden. Such an algorithm would need to be updated regularly to take into account any changes in patient referral patterns, treatment options or local clinical expertise.

Clinical Trial Registration

lPLAS_20-21_A08.

中文翻译：

---

预测基底细胞癌管理中多学科团队治疗建议的机器学习算法

背景

基底细胞癌（BCC）是最常见的人类癌症。面部 BCC 最常发生在鼻子上，考虑到治疗的功能性和复杂性，这些病变的管理特别复杂。定期举行多学科团队 (MDT) 会议，以整合来自皮肤科医生、外科医生、肿瘤科医生、放射科医生、病理学家和专职医疗人员的专业知识。本研究的目的是开发一种有监督的机器学习算法来预测针对鼻 BCC 的 MDT 建议，以潜在地减少 MDT 病例量，提供自动决策支持并允许在卫生服务环境中进行数据审计。

方法

研究人群包括在 2015 年 1 月 1 日至 2015 年 12 月 31 日期间在皮肤癌专科 MDT (SSMDT) 讨论并诊断为鼻 BCC 的所有连续患者。我们对性别、年龄、解剖位置、组织学亚型、肿瘤大小、肿瘤复发、抗凝、起搏器、免疫抑制剂和治疗方式（莫氏手术、常规切除或放疗）。我们使用S统计计算语言来开发一种有监督的机器学习算法。

结果

我们发现，根据肿瘤位置和年龄，可以可靠地预测 37.5% 的患者将接受莫氏显微手术 (MMS)。同样，可以根据患者的年龄、肿瘤表型和病灶大小可靠地预测 MDT 对常规治疗（手术切除或放疗）的选择。因此，该算法可靠地预测了 45.1% 的鼻 BCC 的 MDT 决策结果。

结论

我们的研究表明，机器学习方法是一种潜在有用的工具，可用于预测 MMS 的 MDT 决策与大量患者的常规手术或放射治疗。我们建议使用该算法让 MDT 有更多时间考虑更复杂的患者，其中包括复发、财务成本和美容结果在内的多种因素有助于最终决定，但无法可靠地预测以确定该结果。这种方法有可能减轻专家皮肤 MDT 的负担和提高效率，进而改善患者护理、减少等待时间并减轻经济负担。这种算法需要定期更新，以考虑患者转诊模式、治疗方案或当地临床专业知识的任何变化。

临床试验注册

lPLAS_20-21_A08。