The General Causation (GC) Risk Score: What It Represents and How Praedicat Calculates It

What is the General Causation Risk Score?

• General causation risk is the risk that a court of law will find that a plaintiff’s injury could, in principle, have been caused by exposure to the activities of one or more defendants. We evaluate general causation risk at the level of a Litagion agent-harm hypothesis.
• Praedicat’s patented general causation risk model is a weight-of-evidence model that conceptually mimics how an expert panel of scientific investigators would evaluate a causal hypothesis of injury.
• The GC, or General Causation risk score, is the output of our weight-of-evidence model, which puts the scientific literature on an equal footing and provides a standard means of comparing of risk across Litagion agents.
• The GC risk score measures the extent to which the scientific community accepts a hypothesis that a particular chemical causes a particular bodily injury, i.e., the strength of the evidence that a presumed cause and an observed effect are in fact linked.
• The GC score has a scale that runs from -1 to +1. A score of +1 indicates that the scientific literature overwhelmingly accepts the harm hypothesis. The canonical example of complete acceptance of a harm hypothesis is asbestos and mesothelioma.
• Another current example of a widely accepted litagion agent-harm hypothesis is the bisphenol A and reproductive injury harm hypothesis This is a cause and effect link strongly supported by the scientific literature, and it therefore has a GC score of 0.99.
• A score of -1 indicates that the scientific literature overwhelmingly rejects the hypothesis that exposure to the chemical causes the hypothesized harm. An example of broad rejection of a harm hypothesis is thimerosal and autism.
• A GC score of zero indicates that either there is no literature or the scientific literature is equivocal, i.e., for every study with a positive outcome, there is a study with a negative outcome.

Let’s look at how Praedicat represents the development of the GC score in the software.

• After logging in, you will click on “Litagion Profiles” from the menu on the left, which will bring you to the Litagion Profiles page. Type the name of the Litagion agent you wish to view into the search bar, in this case Bisphenol A, and hit enter.

• From the Bisphenol A summary page, click on the link “Science Risk” to the right, which will bring you to the Science Risk or GC plot.

• The GC (General Causation) Plot displays the GC score over time for a litagion agent in relation to a class of compensable harms and characterizes the extent to which the scientific literature supports the hypothesis that the Litagion® agent causes a particular injury. We track the development of the GC score from the year 2000 (a GC risk line starting after the year 2000 indicates the year of the first identified peer-reviewed article addressing that hypothesis) to the current year and then project the GC score for each litagion agent-harm hypothesis seven years into the future. Praedicat has chosen seven years because our research has found that, on average, it takes seven years from publication of initial articles exploring a hypothesis to public awareness, regulatory response or litigation

• There are a couple of thresholds on the GC plot that are important to note and are indicated by the dotted red lines on the chart below. The first is the 0.50 threshold. For bodily injury
  The GC score for a given litagion agent/harm hypothesis cannot exceed 0.5 without evidence that the litagion agent causes harm in humans.
  Around the 0.65 threshold, there is sufficient scientific evidence to satisfy a Daubert hearing.
  hypotheses, the score for a literature without human evidence cannot increase above a GC Risk of +0.50 or below –0.50.
• You’ll note a big dramatic upward shift in slope for of the dark green risk line representing the BPA/endocrine injury harm hypothesis. This increase is due to an increase in the number of human studies included in the reviewed literature.
• In addition to the 0.50 threshold being significant, the 0.65 threshold is also important to note. This is the threshold at which the scientific evidence is sufficient to satisfy a Daubert hearing. A Daubert hearing is a US legal proceeding where scientific expert testimony is assessed, and it is determined whether the testimony can be admitted in court.

How Praedicat Calculates the GC Risk Score

• We’ll now take a look at how Praedicat calculates the GC score for each harm hypothesis.
• The GC score mirrors the Hill criteria, a group of 9 principles established in 1965 by English epidemiologist Sir Austin Bradford Hill. They are widely used in public health research to establish epidemiologic evidence of a causal relationship between a presumed cause and an observed effect. The Praedicat General Causation Risk Model mimics this to produce the GC scores.
• To calculate the GC score, Praedicat uses both machine learning technology and in-house staff of bio scientists to read more than 35 million peer reviewed scientific article abstracts captured in the U.S. National Library of Medicine’s PubMed database. Pertinent information relating to every harm hypothesis is gathered to create literature metadata.
• At the literature level, the individual studies are aggregated and weighted to reflect the relative weightings that scientists would be expected to place on the studies. For instance, studies that are published in the most well-respected and cited journals are viewed as more authoritative and of higher quality. Praedicat uses an authoritative and external ranking of journals to assign a “quality weight.”
• The outcome of the studies, effect size and study type metadata combine with “quality weight” data to give an article its relative weight in the GC score. Each study can be classified into a category indicating the study’s ability to show a causal link between a chemical and an associated harm.
• When Praedicat calculates a GC score on a full literature, we include all factors to get each paper’s individual weight and then sum them to get the overall weight of the literature.
• While positive results are added, negative results are subtracted.
• If the literature is based entirely on animal evidence, it further restricts the ranges of scores to (-0.5, 0.5), embodying our belief that human evidence is required to sustain a claim in court. Once human evidence exists in the literature, this module changes the output range of the GA score: a completely human literature is restricted to (-0.75, 0.75) and a good mixture of both animal and human literature allows the full (-1.0, 1.0) range to be used.
• This embodies our observation that having corroborating evidence from both animal and human studies raises the likelihood of sustaining a bodily injury claim in court.
• In summary, Praedicat’s GC scoring algorithm weighs the relative importance of the articles comprising a chemical-bodily injury hypothesis using journal quality weightings, study types and effect size, while also taking into account the relative presence of human and animal studies in the literature. The model takes into account how expert panels read and review science while also paying homage to the Hill criteria that establish epidemiological causation.

How Praedicat Calculates Projected GC Risk Scores

• In addition to calculating and assigning the GC scores for the current year, and past years, Praedicat also forecasts the GC scores forward seven years into the future.
• This gives the software user the ability to understand the potential trajectory of the science for each harm hypothesis, while also supplying a measure of the uncertainty around these projections.
• Praedicat has chosen a seven year projection, as research indicated that on average it takes seven years from publication of initial articles exploring a hypothesis to public awareness, regulatory response or litigation.
• Praedicat adopts a two-stage process to forecast the GC scores seven years into the future.
• The first step of the process is to estimate the number of future studies at each point in time.
• The second step is to generate the characteristics associated with those future studies.
• A Monte Carlo simulation is then used to generate synthetic literature metadata for future years (1-7).
• This ‘future’ data is then analyzed in the same way, using the same General Causation Risk Model as is used for the existing literature review that generates current and past GC scores.
• Multiple simulations from this ‘future’ information results in many potential GC score outcomes for a particular harm hypothesis. This range of GC scores creates a distribution around which future GC scores are more or less likely for a single harm hypothesis.
• Praedicat represents this range of potential outcomes within the GC projections in the software. You are able to “show the full GC distribution” either for the Expected Value [the average projected GC value], or for different percentiles of the projection distribution [what the GC score might be in more or less extreme situations either side of the average].

Interpreting and Applying the GC Risk Score

• In practice, the GC score information is very powerful, primarily because it allows for direct comparison of risk across chemicals. Here are some examples of how you might use the GC scoring information in practice:
  • Emerging risk monitoring: You may want to monitor GC score development over time for a pre-defined list of chemicals and identify which chemicals as an insurer you should be most or least concerned about, when looking at them in relation to one another.
  • Horizon scanning: You can also see how the scientific perception of a harm hypothesis has developed over time, and where this might develop seven years into the future. It is possible to keep an eye on emerging chemicals that aren’t an immediate concern right now but which might become a concern in the future.
  • Dynamic watch list creation/curation: It is possible to create internal watch lists around developments in the science, by defining GC thresholds against which to monitor. You might, for example, define a GC threshold which if breached could set off a chain of actions internally to adjust the underwriting approach to litagion agents with potentially concerning trends in the science and better manage aggregations arising from these risks.
  • Product substitution evaluation: You might wish to compare a chemical already in widespread use and known to have harmful effects with a chemical that has been introduced as a substitute and determine if the substitute is indeed less harmful than the existing chemical.