Machine Learning Assessment of Melanoma Primary Phenotype

  • Research type

    Research Study

  • Full title

    Machine Learning Assessment of Melanoma Primary Phenotype to Predict Sentinel Node Status

  • IRAS ID

    287112

  • Contact name

    Marc Moncrieff

  • Contact email

    marc.moncrieff@nnuh.nhs.uk

  • Sponsor organisation

    Norfolk & Norwich University Hospital Foundation NHS Trust

  • Duration of Study in the UK

    1 years, 2 months, 30 days

  • Research summary

    Summary of Research

    Cutaneous melanoma has a poor prognosis in advance disease making accurate stratification for risk of progression important. The single most important prognostic variable is the node status which is obtained by undertaking a sentinel lymph node biopsy (SLNB). This informs the treating team that the melanoma has the biological potential to metastasise and is an indication for additional treatments such as targeted drug therapy or immunotherapy. Ultimately 80% of all patients undergoing a SLNB have a negative biopsy and do not derive any therapeutic benefit from the procedure.
    Much research effort is currently being dedicated to develop more sophisticated markers for high-risk melanoma, thereby avoiding an unnecessary operation for both low-risk patients, who can simply be observed, and high-risk patients, who can commence systemic drug treatments without the delay the surgical procedure entails.

    Summary of Results

    With the advent of adjuvant systemic therapy for micrometastatic stage III (those patients with a positive staging sentinel node biopsy) melanoma, there has been a recent renewed interest in novel diagnostic techniques, in addition to algorithms for predicting sentinel node (SN) status. We undertook a machine-learning analysis (also know as artificial intelligence) to produce predictive models of sentinel node (SN) status, based on primary tumour characteristics.
    We conducted a single-centre, retrospective cohort analysis and identified 1254 sets of patient records appropriate for inclusion.
    Using robust machine learning modelling (artificial intelligence methods), we were unable to identify subgroups of patients where the analysis performs accurately, whether weighted to identifying sentinel node positive or sentinel node negative patients and the models generated performed no better than traditional statistical methodology. These data indicate the stochastic behaviour of the metastatic process to the SN, questioning the validity of SN prediction based on primary tumour characteristics alone.

  • REC name

    West of Scotland REC 4

  • REC reference

    20/WS/0102

  • Date of REC Opinion

    16 Jul 2020

  • REC opinion

    Favourable Opinion