Identification and correction of confounders in a statistical analysis
AssigneePharsight Corporation
InventorEdmund G. Lee
Life of this patent
Estimating the term…
Abstract
Systems and methods are disclosed for identifying and/or compensating for relationships between influencers and outcomes. First, a metadata model of information that characterizes relationships between influencers and outcomes is built. A query is accepted from a user, and an outcome of interest is determined based on the query. The model is subsequently relied on to flag influencers that might have an impact on an outcome of interest. The impact of these influencers on the outcome of interest is then analyzed with respect to specific conditions. When the impact is sufficiently large, the user is notified or an adjustment is made to the user's query.
Claims33
A computer implemented method of identifying a confounder, the method comprising the steps of:
- building a model of information that characterizes relationships under many different conditions between influencers and outcomes;
- storing metadata in a computer readable storage medium as a part of the model, the metadata specifying,
- whether a relationship exists between an influencer and an outcome, and
- a characteristic of that relationship, including whether the influencer represents a confounder of the outcome;
- inputting to a user interface a query that specifies a set of conditions;
- determining a relationship between the set of conditions specified in the query and a particular outcome that is represented in the model;
- identifying, based on the metadata of the model, at least one potential influencer of the particular outcome, and whether the potential influencer represents a confounder of the particular outcome;
- and
- if the potential influencer represents a confounder of the particular outcome, reporting the confounder to the user interface.
A computer-implemented method of identifying relationships between influencers and outcomes under a particular set of conditions, the method comprising the steps of:
- building a model of information that characterizes relationships under many different conditions between influencers and outcomes;
- storing metadata in a computer readable storage medium as a part of the model, the metadata specifying,
- whether a relationship exists between an influencer and an outcome, and
- a characteristic of that relationship, including whether the influencer represents a confounder of the outcome;
- inputting to a user interface a query that specifies a set of conditions;
- determining a relationship between the set of conditions specified in the query and a particular influencer that is represented in the model;
- determining a relationship between the set of conditions specified in the query and a particular outcome that is represented in the model;
- identifying, based on the metadata of the model, at least one potential confounder of the particular outcome, wherein the potential confounder is unaccounted for by the query;
- and
- reporting the confounder to the user interface.
A computer-implemented method of identifying relationships between influencers and outcomes under a particular set of conditions, the method comprising the steps of:
- building a model of information that characterizes relationships under many different conditions between a plurality of nodes, wherein at least some of the nodes represent influencers and at least some of the nodes represent outcomes;
- storing metadata in a computer readable storage medium as a part of the model, the metadata specifying,
- whether a relationship exists between an influencer and an outcome, and
- a characteristic of that relationship, including whether the influencer represents a confounder of the outcome,
- inputting to a user interface a query that relates to at least one of the nodes;
- determining a relationship between the query and the nodes of the model;
- and
- identifying, based on the metadata of the model, a potential influencer of an outcome associated with the query, and whether the potential influencer represents a confounder of the outcome;
- and
- if the potential influencer represents a confounder of the outcome, reporting the confounder to the user interface.
Dependencies are inferred from the claim text.