We’ll find that needle – but only if it’s there

Target Analytics Advantage

Building on a rich academic research foundation involving top institutions and scientists in the areas of semi-parametric models and causality, including the work of the company's founder, Dr. Mark van der Laan at UC Berkeley, the Target Analytics product line utilizes novel, proprietary methodology of causal inference and targeted maximum likelihood analysis to improve precision and reliability over best existing tools. Target Analytics super-learning algorithms automatically arrive at the best combination of input models based on the cross-validation within each particular dataset, eliminating the need for high levels of domain, statistical, and programming expertise.

TargetDiscovery

For each potential causal factor in a given list, TargetDiscovery performs a separate targeted maximum likelihood estimate of the effect on the outcome, controlling for confounding variables.
In addition, it provides a reliable measure of the signal-to-noise ratio in the dataset. For example, consider a study to evaluate the effect of genetic markers on progression free survival in cancer patients. For each genetic marker, TargetDiscovery will assess its effect on progression free survival, controlling for the effect of other genetic markers as well as the effect of other patient characteristics such as age, gender, stage of the disease, etc. The reliability of this assessment is measured by its signal to noise ratio mapped into a p-value and confidence interval. Critically, TargetDiscovery lowers the rate of false positives by adjusting the p-values for the number of factors evaluated in this study.

In several studies, TargetDiscovery has shown significantly higher precision and reliability than such state-of-the-art current approaches as least-angle, stepwise and other modern regression analyses, random forest, and neural networks.

TargetImpact

TargetImpact performs automated targeted maximum likelihood estimate of the causal effect of time-point treatment interventions in longitudinal observational as well as randomized studies.
TargetImpact builds on the framework of counterfactuals and the corresponding in-depth statistical theory and methodology for marginal structural models. Traditionally, causal effect assessment analyses have been performed by highly trained statisticians who are also domain experts. TargetImpact is the first product to automate this process, while preserving the ability to fine-tune analysis parameters.
For example, consider a comparative post-market analysis of a newly introduced HIV-drug against a previously available treatment. The object of the study is to assess the adverse effects of the new HIV drug across several subgroups of patients.
A marginal structural model defines (and models) a causal effect of treating the patients with this new HIV-drug versus the other drug as the difference in two rates of adverse events in the two hypothetical (i.e., counterfactual) worlds in which everybody gets treated with the new drug or old drug, respectively.

In our study the data set would consist of treatment histories (with the old and new drug), and the observed adverse outcomes over time across a diverse population of patients. TargetImpact allows one to specify and model a causal effect question of interest. For example, one might wish to assess the effect of each drug over a 3 month period on 1 year survival, or an immediate effect of a single dose on the immune system. In addition, the user of TargetImpact will be able to model how this effect of the two HIV drugs changes in response to potential effect modifiers such as the baseline health of the patient.