Machine learning can improve prediction of depression in Generation Scotland

Research output: Contribution to conferencePaperpeer-review


In this study we have evaluated the prognostics value of a wide range of clinical, demographic and genomic variables in the classification of depression recurrence and outcome in the Generation Scotland cohort; using a set of 14 machine learning methods. In total, 21,476 study participants under took the structured clinical interview for the Diagnostic and Statistical Manual of Mental Disorders (SCID); following data editing and quality control 475 single and 960 recurrent case episodes, and 1,130 cases and 5,043 controls were available for inclusion in the analysis. The recurrence and outcome of depression could be discriminated based on clinical and demographic
variables alone with up to 71% and 77% accuracy respectively. The inclusion of different genomic data sources did not augment the classification accuracy of depression outcome, and only increased the prognostic value for depression recurrence by 2% in two the methods applied.
Original languageEnglish
Publication statusPublished - 23 Nov 2015
EventSackler PhD Conference 2015 - Yudowitz Lecture Theatre, Wolfson Medical School Building, University of Glasgow, Glasgow, United Kingdom
Duration: 23 Nov 2015 → …


ConferenceSackler PhD Conference 2015
CountryUnited Kingdom
Period23/11/15 → …


  • Major Depressive Disorder
  • Prediction
  • Machine learning
  • Generation Scotland: the Scottish family health study
  • Area Under Curve
  • Environmental factors
  • Genomic factors
  • Clincal factors
  • Socio-economic factors
  • Demographic factors
  • Recurrence
  • Lifetime
  • Cardiovascular disease risk factors
  • Polygenic risk scores
  • Co-morbid disorders

Fingerprint Dive into the research topics of 'Machine learning can improve prediction of depression in Generation Scotland'. Together they form a unique fingerprint.

Cite this