Fancy math models that bend at different points give better estimates of safe chemical exposure levels from real-world human data than older, simpler models — especially when studying how PFAS chemicals affect the immune system.

What we've found so far suggests that piecewise linear models may offer more accurate estimates of safe dose levels for PFAS based on human immune effects compared to hockey-stick or K-power models. Our analysis of the available research shows a consistent lean toward more flexible, mathematically complex models when interpreting real-world data. We looked at 33.0 assertions from studies that evaluated how different dose-response models perform in estimating safe exposure levels for PFAS, particularly in relation to immune system impacts in humans. All 33.0 support the idea that models allowing for changes in slope at different dose levels—piecewise linear models—perform better than older models with fixed thresholds or smooth curves like the hockey-stick or K-power models [1]. These simpler models assume a clear cutoff point or a gradual, predictable change in effect, which may not reflect how PFAS actually behave in the human body. In contrast, piecewise linear models can capture shifts in response at various exposure levels, which appears to improve their fit with observed human data [1]. This flexibility may help account for biological complexity, such as differing immune responses at low versus high doses. Since all the evidence we’ve reviewed supports this approach—and none contradicts it—the current analysis leans toward piecewise linear models being more suitable for this specific use. However, we emphasize that this is based on what we’ve reviewed so far. More studies could change or refine this picture over time. We don’t claim these models are definitively better in all contexts, only that the evidence we’ve seen so far points in that direction. Practical takeaway: When setting safe limits for PFAS, using models that can adapt to changing response patterns in human data might give more realistic estimates than older, rigid models.