BinastelDiscuss a project

Selected work

Selected analytical work

These research examples demonstrate the statistical, machine-learning and computational approaches that underpin Binastel’s consulting work. They are presented as research experience rather than commercial client case studies.

Research example 01

Comparing the effectiveness of two rheumatoid arthritis treatments

Analysed a nationwide, prospective real-world cohort to compare treatment response between abatacept and adalimumab and examine whether genetic differences altered their relative effectiveness.

Evidence and methods

  • 342 patients included after quality control
  • Clinical treatment-response outcomes
  • Genetic markers and drug-level information
  • Multivariable statistical modelling
  • Adjustment for relevant patient and treatment factors
  • Testing of treatment, genetic and interaction effects
  • Careful interpretation of negative as well as positive findings

What the analysis showed

The analysis found no evidence that the relative effectiveness of the treatments differed between the genetic groups examined. The work demonstrates rigorous comparative analysis, integration of heterogeneous clinical and genetic data, and the ability to test commercially or scientifically important claims without overstating the evidence.

  • Comparative effectiveness
  • Clinical data analysis
  • Statistical modelling
  • Genetic stratification
  • Uncertainty and validation
Read the peer-reviewed publication
Research example 02

Identifying treatment-response biomarkers using machine learning

Used whole-blood transcriptomic data, machine-learning classifiers and targeted statistical analysis to investigate biomarkers associated with treatment response in rheumatoid arthritis.

Evidence and methods

  • Cohort of 100 patients
  • 97 transcriptomic samples passing quality control
  • High-dimensional RNA-sequencing data
  • Differential-expression analysis
  • Random-forest classification
  • Predictive performance with AUC values of up to 0.86
  • Network and survival analysis
  • Identification of MZB1 as a potential treatment-response biomarker

What the analysis showed

The work combined machine-learning data mining with statistical and biological validation to identify interpretable candidate biomarkers. The reported signatures require independent replication and should not be presented as clinically validated tests.

  • Machine learning
  • Biomarker discovery
  • High-dimensional data
  • Feature identification
  • Network analysis
  • Reproducible scientific computing
Read the peer-reviewed publication

Start a conversation

Need the same rigour applied to a commercial or technical question?

Discuss the decision, available evidence and what a useful, proportionate analysis would need to establish.

Discuss a project