Skip to content
LUMING洛铭BIOMEDICAL DATA SCIENCE CONSULTING
简体中文Free consultation
简体中文Free consultation

SERVICE 03

Statistical Analysis and Predictive Modeling

Survival, GEE, Lasso, Random Forest and SVM models with cross-validation and external validation. Analysis plans, model code and nomograms.

When you need this

  • You need a statistical analysis plan before the study starts.
  • You have a set of clinical and molecular variables and want a risk stratification tool that clinicians or your R&D team will actually use.
  • Reviewers have asked for external validation or a more rigorous model assessment.
  • You have repeated measurements on the same patients and need methods that handle them properly.

METHODS

What we do

Statistical methods

  • Survival analysis: for time-to-event data such as progression-free survival.
  • Logistic regression: for yes/no outcomes such as benefit or response.
  • GEE (generalized estimating equations): for repeated measurements on the same patient.
  • GLM/Lasso: Lasso selects the variables that matter when there are many candidates, and reduces overfitting.

Machine learning

Random Forest, SVM and other machine learning methods. We only deliver models whose behavior we can explain. Models are assessed with cross-validation and external validation, and the code and validation results ship with the report.

Multivariable survival forest plot

Hazard ratios with 95% confidence intervals for age, stage, mutation status and performance status.

Illustrative forest plot of hazard ratios with 95% confidence intervals for age, stage, mutation status and performance status, with a reference line at HR = 1. Values are illustrative.
ILLUSTRATIVE · Not real client data

Nomograms

A nomogram is a scoring chart. Age, stage, mutation status and other variables each convert to points; the total maps to a predicted risk or survival probability. A clinician can use it with a pen, no software required.

DELIVERABLES

Deliverables

  • Statistical analysis plan
  • Model code and validation
  • Nomogram or risk stratification tool

QUESTIONS WE HEAR

Questions R&D teams bring us

"How should we write the statistical analysis plan for this study, and how do we set the primary endpoint and covariates?"

"We have dozens of clinical and molecular variables. Can they become a usable risk stratification tool?"

"Reviewers want external validation. How do we design it with the data we have?"

Start with one question

Tell us your study design, endpoints and available variables. In a free 30-minute consultation, we'll outline a modeling approach and tell you what the data can support.

Book a free 30-minute consultation