KEvidence Risk Assessment Workbench

Codex-aligned risk assessment: problem formulation → hazard identification → hazard characterisation → exposure assessment → risk characterisation
Prototype • supports the scientific risk-assessment step of risk analysis, not risk management decisions

Problem formulation

Problem formulation is the practical first step of a modern risk assessment — it is not one of the four formal Codex steps, but it frames all of them. Define the question, scope, population, endpoint, product/use scenario, data needs, and decision context. KEvidence carries this structured context into hazard identification, hazard characterisation, exposure assessment, and risk characterisation.

Select an output purpose before starting. This workbench organizes evidence for expert review across the four Codex risk-assessment steps; it does not produce risk-management decisions, EFSA conclusions, health-based guidance values, ADIs, ARfDs, or acceptability decisions. AOP structure comes from local AOP-Wiki-derived data; NAM PODs, exposure estimates, IVIVE factors, and thresholds are user-supplied or heuristic unless explicitly connected to curated sources.

1. Hazard identification

Hazard identification asks: can this agent cause harm? Search candidate Adverse Outcome Pathways by chemical, key event, adverse outcome, or AOP ID, and select the pathway that best represents a plausible route from molecular initiating event to adverse outcome. Identifying a hazard is not yet a risk conclusion — that requires exposure (steps 3–4).

2. Hazard characterisation

Hazard characterisation asks: what is the nature and severity of the harm, and at what dose? Review the selected pathway and its evidence, gather reference points and health-based guidance values (NOAEL/BMD, ADI/TDI/ARfD) from OpenFoodTox, and establish a dose-response point of departure (POD) from NAM data. The POD you set here becomes the hazard side of risk characterisation (step 4).

No AOP selected yet. Identify a hazard pathway in step 1.

Pathway events

KER evidence & quantitative understanding

Toxicological reference points — OpenFoodTox

Query EFSA OpenFoodTox for substance identity, reference points, reference values, and study documents that describe the nature, severity, and dose of the hazard. Use a local SQLite index built from the official Excel export, or a local IUCLID 6 import.

Run scripts/import_openfoodtox.py --download-latest --db data/openfoodtox.db to download and index automatically. Institutions can alternatively import i6z dossiers into IUCLID 6 and configure the local IUCLID REST API URL.
No OpenFoodTox query run yet.

Dose-response point of departure (POD)

Record the NAM / dose-response point of departure that characterises the hazard. This POD is carried into risk characterisation (step 4) and compared against the exposure estimate from step 3.

A POD describes the hazard's dose-response; it is not itself a risk. Confirm KE mapping, assay quality, hit call, and units before use.

Browse EPA ToxCast/Tox21/ToxRefDB PODs

Search a locally indexed EPA bioactivity/toxicity export for AC50 or POD candidates for the current chemical. Records are ranked higher when assay or endpoint text overlaps the selected AOP events.

No EPA bioactivity search run yet.

3. Exposure assessment

Exposure assessment asks: who is exposed, how much, how often, and by which route? For food safety this is usually dietary exposure — concentration in food/feed × consumption pattern × population group × use conditions. Record the scenario and the resulting exposure estimate that will be integrated in risk characterisation (step 4).

The exposure estimate is user-supplied unless connected to a curated dietary-exposure model. Exposure is what turns an intrinsic hazard into a risk — without it, no risk conclusion can be drawn.

4. Risk characterisation

Risk characterisation is the integration step: given the hazard (step 2) and the exposure (step 3), what is the risk? KEvidence compares the dose-response POD with the exposure estimate to produce a screening margin (BER / MOE) and a hazard-quotient-style ratio. Hazard is intrinsic potential for harm; risk is that hazard under actual exposure conditions.

This is a prototype screening calculator, not a validated regulatory threshold framework. Screening margins support prioritization and hypothesis refinement unless validated sources and program-specific thresholds are configured.
No risk characterisation run yet. Set a POD in step 2 and an exposure estimate in step 3.

Uncertainty analysis

After risk characterisation, analyse uncertainty. Run the evidence-to-decision engine to summarize hazard hypotheses, confidence across the four steps, key uncertainties, critical data gaps, and next recommended data collection.

No analysis run yet.

Scientific opinion / advice

Assemble a structured draft opinion from problem formulation, hazard identification and characterisation, exposure assessment, risk characterisation, and uncertainty analysis. This scientific advice is the output handed to risk managers — risk assessment informs, but does not make, the risk-management decision. Use the assistant to rewrite or explain sections.

Quick user manual

A one-page guide to the Codex risk-assessment workflow. Risk assessment is the scientific part of risk analysis (which also covers risk management and risk communication). Problem formulation frames the four Codex steps below; keep every generated statement traceable to source data and expert judgement.

Problem formulation — define the question.Enter the substance/stressor, product domain, use scenario, endpoint of concern, and decision context. This pre-step frames all four Codex steps.
Hazard identification — can it cause harm?Search AOPs by chemical, key event, adverse outcome, or AOP ID, and select the pathway representing a plausible route from MIE to adverse outcome.
Hazard characterisation — nature, severity, dose.Review KER evidence, pull reference points/HBGVs from OpenFoodTox, and set a dose-response point of departure (POD) mapped to a key event.
Exposure assessment — who, how much, how often, route.Record concentration in food × consumption × population × use conditions, and the resulting exposure estimate.
Risk characterisation — integrate hazard × exposure.Compare the POD with the exposure estimate to obtain a screening margin (BER/MOE). Interpret as prioritisation support, not a final regulatory conclusion.
Uncertainty & scientific opinion.Run the uncertainty & data-gap analysis, then build a draft scientific opinion that separates curated evidence, user-entered data, assumptions, and gaps for handoff to risk managers.
Tip: required fields are the values needed to make the next tab meaningful. If uncertain, record assumptions in the draft caveat.

Sample data for testing

Use these starter files and public repositories to test the workflow before you have project-specific data. Confirm licenses, versions, units, and applicability before using any value in a regulatory submission.

NeedUse thisWhat to do in KEvidence
NAM POD / AC50 exampleDownload local CSV template
EPA ToxCast/Tox21 data portal
EPA invitroDB on Figshare		Copy assay, POD type, POD value/unit, and mapped event ID into the hazard characterisation step (POD). Use the EPA bioactivity import script for larger local searches.
OpenFoodTox reference pointsOpenFoodTox 3.0 on Zenodo
Earlier OpenFoodTox Zenodo record		Download the Excel export, build the local SQLite index, then query the hazard characterisation step by substance name, CAS, EC number, or synonym.
AOP structure and event IDsAOP-Wiki quarterly downloads
AOP-Wiki RDF on Zenodo		Use event IDs to map NAM assays to MIE/KE/AO records during hazard identification and characterisation.
Exposure / IVIVE test valuesDownload local CSV templatePaste the exposure value/unit into the exposure assessment step and the optional IVIVE factor into the hazard characterisation POD, documenting assumptions in the scientific opinion.

Questions?

Suggest improvements, report missing data sources, or propose new risk-assessment workflow features. You can open the GitHub repository directly or send a feature request by email.

Open GitHub repository
This form opens your email client and sends to kevidence@devalier.com. No request is stored by KEvidence unless you send the email.