Combining Evidence: The Art of Data Integration in Toxicology
Why Single-Source Answers Fail in Toxicology
Toxicology rarely offers the comfort of a single, unambiguous answer. A laboratory assay may report a compound as cytotoxic at a given concentration, while a computational model predicts negligible risk and an epidemiological record shows no adverse effects in exposed populations. Each source is legitimate, yet each is partial. The art of data integration begins with accepting that no single evidence stream is sufficient for a reliable risk judgement.
At ASPIS, we treat laboratory results, in silico predictions and historical records as complementary witnesses rather than competing authorities. Their disagreements are not noise to be silenced; they are signals that point to uncertainty, context dependence or gaps in study design.
The Three Pillars of Integrated Evidence
Integrated assessment rests on three pillars, each contributing a distinct kind of knowledge. Understanding what each pillar can and cannot deliver is the first step toward sound combination.
- Laboratory results provide mechanistic detail and dose-response data under controlled conditions.
- Computer models extend coverage to untested substances and predict properties at scale.
- Historical records anchor predictions in real-world exposure, susceptible populations and long-term outcomes.
None of these pillars stands alone. A model trained without laboratory grounding may extrapolate unreliably, while a historical record without mechanistic insight cannot explain why an effect appeared or disappeared.
Practical Steps for Merging Evidence Streams
Combining evidence is a structured process, not an informal average. The following sequence helps teams move from isolated findings to a defensible judgement.
- Define the question and the decision context before collecting any evidence.
- Assess the quality and relevance of each source independently.
- Map findings onto a common dose or exposure metric where possible.
- Identify convergence, contradiction and knowledge gaps explicitly.
- Weight sources according to reliability, relevance and uncertainty.
Transparency at every step is essential. When a judgement rests on multiple streams, decision-makers must be able to see how each stream influenced the outcome.
Weighting, Uncertainty and Transparency
Weighting is where integration becomes an art as much as a science. A well-validated model may deserve substantial weight for a data-poor substance, while a historical cohort with strong exposure data may dominate for a well-studied chemical. These choices should be documented and open to challenge.
Evidence is not combined by averaging numbers; it is combined by reasoning about what each source can and cannot tell us.
Uncertainty must therefore be reported alongside the integrated conclusion. A result that hides its weakest link is less trustworthy than one that names it openly.
From Integration to Reliable Risk Judgements
When laboratory, computational and historical evidence are merged thoughtfully, the resulting judgements are more robust, more transparent and more useful for regulators, industry and the public. The table below summarises the contribution of each stream.
| Evidence stream | Primary strength | Main limitation |
|---|---|---|
| Laboratory results | Mechanistic and dose-response detail | Limited coverage, controlled conditions |
| Computer models | Scale and prediction for untested substances | Depends on training data quality |
| Historical records | Real-world exposure and long-term outcomes | Confounding and incomplete exposure data |
Integration does not eliminate uncertainty, but it distributes it honestly across sources. That honesty is what makes combined evidence a stronger foundation for protecting human health and the environment.
