The statistical relationship is the atomic unit of System; it is extracted from peer-reviewed papers, and connected into the larger System graph so anything in the world can be connected to everything.
Relationships can be searched via directly entering two topics in the global search bar or, you can first navigate to a topic of interest and see all its relations to other topics with the cards. Additionally, relationships are represented as edges on the graph; clicking on any edge will navigate directly to the relationship panel.
Topic panel where relationships can be navigated via the cards and the graph edges
Within each relationship panel, you can first see the established directionality for a relationship using the arrow icons under the header. Depending on the topics, there may be cases when only one direction has been established—i.e. Smoking may be an established risk factor for Coronary Artery Disease (Smoking > Coronary Artery Disease) but Coronary Artery Disease is may not a factor for Smoking (Coronary Artery Disease > Smoking). When there is a correlation established but no direction known, the third icon will be enabled. Use these arrows to toggle between the supporting evidence.
Selected, disabled, and enabled directional buttons
Once the desired direction is selected, you can first explore a short synthesis on the relationship. This synthesis is generated by prompting a Large Language Model (LLM) to produce an overview of the relevant peer-reviewed statistical evidence in the bottom half of the panel. As with all text generated via LLMs, it is possible that it includes unrelated language.
As needed, you can dig in further to the evidence by browsing the evidence clusters in the bottom half of the panel. These clusters are based on similarity between the independent and dependent variables studied. To quickly refine the results, various filters are available, from Publish Date, Journal, Sample Size, Sex, and Age (while not every study reports sex, age, or sample size in a standard manner, the availability of these is dependent on what the System models can extract).
One more level of depth can be achieved by clicking into any cluster of evidence to see each individual piece of evidence, and it’s corresponding study. The forest plot can be used to get a quick understanding of the effect size of each extracted relationship; an empty circle indicates the results were not significant and a circle indicates no sample size was extracted for the corresponding study. As needed, see the legend indicated with the (?) icon in the forest plot header.
Each piece of evidence is annotated and contextualized with the following attributes, visible in the table:
- Sample Size
Strength is a measure of how statistically strong a relationship is in some defined context. The platform computes strength based on this methodology.
The sign of an evidence indicates whether an increase in one metric, would increase (positive) or decrease (negative) the other metric.
The direction of a relationship indicates how information flows between a pair of objects (like metrics). System infers and reports statistical directions (e.g. direction from independent to dependent variables). These should not be interpreted as causal links.
The context and population (subjects) for each study is an important consideration in understanding, comparing, and using the extracted evidence — and evaluating its representativeness. Populations may include parameters like who (e.g. Women aged 18-54), where (e.g. South Korea), when (e.g. May-July 2021).
When possible, we extracted the sample size reported in the study. This helps provide extra context to evaluate the representativeness of extracted evidence. This is the corresponding field to the Sample Size filter.