MMAD-Risk: Multivariate Mixed survival analysis for the prediction of Age-dependent Disease Risks from plasma proteomes

MMAD-Risk is a statistical model for age-dependent disease risk prediction from plasma proteomes. It is a multivariate model, i.e. it considers multiple diseases at once and is able to identify shared disease mechanisms and the comorbidity structure between diseases.

Setup

MMAD-Risk is designed to run on the UKB-RAP. In order to reproduce the results, you first have to extract the following data from UK Biobank using Table exporter:

• The plasma protein concentrations (field 30900, instance 0)
• Diagnoses - ICD10 (field 41270)
• Type of Cancer (field 40006)
• Unterlying (primary) cause of death: ICD10 (field 40001)
• Contributory (secondary) causes of death: ICD10 (field 40002)
• Date of first in-patient diagnosis: ICD10 (field 41280)
• Date of cancer diagnosis (field 40005)
• Date of death (field 40000)
• Year of birth (field 34)
• Month of birth (field 52)
• Sex (field 31)
• BMI (field 21001, instance 0)
• Date of blood sample collection (field 3166, instance 0)

The code necessary to run MMAD-Risk on UK Biobank is contained in the folder UK Biobank. The notebook analysis.ipynb handles data preprocessing, running MMAD-Risk on the full 3000-protein panel and comparing it to the Cox-model, the multivariate fixed model, the univariate mixed model and logistic regression. The notebook analysis_smaller_P.ipynb was to perform the greedy backwards selection of the 10-protein panel and to compare the performance of MMAD-Risk to the Cox model and the multivariate fixed model on this data set.

Local analysis of results

Parts of the analysis of the results, such as the comparison of c-indices and the clustering of diseases, was conducted locally. The folder Data contains aggregated and fully anonymized data that is necessary to run these analyses, such as the c-indeces for each diseases for each model. The code used for the local analysis of UK Biobank results is contained in the folder local.

Data availability

This research has been conducted using the UK Biobank Resource under application number 532367. The UK Biobank data can be accessed by a procedure described here.