Editing Openai/69101fc6-f990-800a-9638-7716259af6c4 (section)

=== how do we establish safety (beyond “no news”)? ===
# Mechanism & off-target profiling (preclinical). In vitro/in silico screens (hERG, CYPs, receptor panels), animal toxicology, and modeling (PBPK) map plausible hazards before humans ever see an ingredient. Regulators then set reference points like NOAEL/BMD and derive margins of exposure/acceptable intakes. Modern guidance actually prefers benchmark dose (BMD) over old NOAEL because it uses the whole dose–response curve. EFSA Journal<ref>{{cite web|title=EFSA Journal|url=https://efsa.onlinelibrary.wiley.com/doi/10.2903/j.efsa.2022.7584|publisher=EFSA Journal|access-date=2025-11-09}}</ref>US EPA<ref>{{cite web|title=US EPA|url=https://www.epa.gov/sites/default/files/2015-01/documents/benchmark_dose_guidance.pdf|publisher=US EPA|access-date=2025-11-09}}</ref>
# Human trials (premarket efficacy & safety). RCTs characterize common adverse events, but they’re underpowered for rare/long-latency harms. FDA’s current benefit–risk framework makes this explicit (uncertainty is part of the decision). U.S. Food and Drug Administration<ref>{{cite web|title=U.S. Food and Drug Administration|url=https://www.fda.gov/regulatory-information/search-fda-guidance-documents/benefit-risk-assessment-new-drug-and-biological-products|publisher=U.S. Food and Drug Administration|access-date=2025-11-09}}</ref>
# Post-market active surveillance. Instead of waiting for case reports, regulators now query huge real-world databases (claims/EHRs) via systems like FDA Sentinel; spontaneous reports flow into FAERS for signal detection. These exist precisely to catch harms that trials miss. U.S. Food and Drug Administration<ref>{{cite web|title=U.S. Food and Drug Administration|url=https://www.fda.gov/safety/fdas-sentinel-initiative|publisher=U.S. Food and Drug Administration|access-date=2025-11-09}}</ref>Sentinel Initiative<ref>{{cite web|title=Sentinel Initiative|url=https://www.sentinelinitiative.org/|publisher=Sentinel Initiative|access-date=2025-11-09}}</ref>OpenFDA<ref>{{cite web|title=OpenFDA|url=https://open.fda.gov/data/faers/|publisher=open.fda.gov|access-date=2025-11-09}}</ref>
# Modern observational methods (causal “safety science”). To move beyond simple observation, analysts use target-trial emulation, active-comparator new-user designs, self-controlled case series, and negative-control outcomes/exposures to detect/correct bias and estimate real-world harm rates. (See OHDSI/LEGEND examples.) PubMed<ref>{{cite web|title=PubMed|url=https://pubmed.ncbi.nlm.nih.gov/35680274/|publisher=pubmed.ncbi.nlm.nih.gov|access-date=2025-11-09}}</ref>ohdsi-studies.github.io<ref>{{cite web|title=ohdsi-studies.github.io|url=https://ohdsi-studies.github.io/LegendT2dm/Protocol|publisher=ohdsi-studies.github.io|access-date=2025-11-09}}</ref>ohdsi.org<ref>{{cite web|title=ohdsi.org|url=https://www.ohdsi.org/wp-content/uploads/2019/09/ERICA-VOSS_POSTER_Assessing-Negative-Control-Exposure-Outcome-Pair-Selection-Strategies-on-a-Replication-Study_2019Symposium.pdf|publisher=ohdsi.org|access-date=2025-11-09}}</ref>OHDSI<ref>{{cite web|title=OHDSI|url=https://ohdsi.github.io/TheBookOfOhdsi/MethodValidity.html|publisher=ohdsi.github.io|access-date=2025-11-09}}</ref>
# Triangulation with genetics. Mendelian randomization uses genetic proxies for long-term exposures (e.g., lifelong higher nutrient levels) to suggest whether very long-run risks exist when trials are infeasible. It’s not perfect, but it’s more than “we haven’t seen anything yet.” PMC<ref>{{cite web|title=PMC|url=https://pmc.ncbi.nlm.nih.gov/articles/PMC4170722/|publisher=pmc.ncbi.nlm.nih.gov|access-date=2025-11-09}}</ref>
# Standards for judging certainty. Frameworks like GRADE force you to rate certainty (risk of bias, inconsistency, imprecision, etc.). A low-certainty signal of harm can still outweigh a tiny, uncertain benefit. PMC<ref>{{cite web|title=PMC|url=https://pmc.ncbi.nlm.nih.gov/articles/PMC6542664/|publisher=pmc.ncbi.nlm.nih.gov|access-date=2025-11-09}}</ref>CDC<ref>{{cite web|title=CDC|url=https://www.cdc.gov/acip-grade-handbook/hcp/chapter-7-grade-criteria-determining-certainty-of-evidence/index.html|publisher=cdc.gov|access-date=2025-11-09}}</ref>training.cochrane.org<ref>{{cite web|title=training.cochrane.org|url=https://training.cochrane.org/grade-approach|publisher=training.cochrane.org|access-date=2025-11-09}}</ref>