**The title, authors, and abstract for this completion report are provided below.  For a copy of the completion report, please contact the GLFC via e-mail or via telephone at 734-662-3209**




Improving the Accuracy and Precision of Predictions of TFM-Niclosamide Concentrations for Treatment of Sea Lamprey Spawning Tributaries



Steve Gutreuter1, Bradley P. Carlin2, Michael A. Boogaard1, Laura A. Hatfield2




November 2009




1 U.S. Geological Survey, Upper Midwest Environmental Sciences Center, 2630 Fanta Reed Road, La Crosse, WI 54650


2 Division of Biostatistics, University of Minnesota, A427 Mayo Building, Minneapolis, MN 55455





The toxicity of formulations of 3-trifluoromethyl-4-nitrophenol (TFM) and 2,5-dichloro-4-nitrosalicylanilide (niclosamide) to sea lamprey Petromyzon marinus ammocoetes is affected by pH and alkalinity. No historic laboratory studies have simultaneously varied TFM-niclosamide formulations and controlled both pH and alkalinity. We filled that data gap with a new toxicity study. We then developed and evaluated Bayesian hierarchical models combining data from two published, two unpublished toxicity studies and our new data for the prediction of LC99.9 of variable TFM-niclosamide formulations. Bayesian hierarchical models provide the natural means to combine data from multiple studies. Among the alternatives we evaluated, an Empirical Bayes (EB) hierarchical model yielded precise and unbiased predictions of LC99.9, and out-performed the conventional two-stage modeling strategy that was used to develop the historic stream-treatment charts. Unlike the conventional two-stage method, our hierarchical EB model yielded valid measures of uncertainty for our estimates of LC99.9, including 95% probability intervals. We produced revised stream treatment charts based on the EB model that include those probability intervals. 


We also used data from streamside toxicity tests from Canada and the U.S., 1993-2006, to evaluate the ability of laboratory data and models to predict LC99.9 on data that were not used to develop the model. The historic streamside toxicity tests exhibited a strong seasonal pattern in the apparent sensitivity of sea lampreys to the toxicant mixtures. Our analyses indicated that the EB model (and revised treatment charts) predicted streamside LC99.9 estimates well during May-June and October, but underestimated streamside LC99.9 during the intervening months.


A similar seasonal pattern was not evident in the laboratory data, indicating that the apparent seasonal effect is more likely to be caused by a latent variable associated with the physiological status of sea lampreys rather than a direct effect of time. Although the historic and current laboratory data were not sufficient to accurately predict the seasonal pattern in the streamside tests, our Bayesian hierarchical modeling indicates that accurate and precise model-based prediction of stream treatment requirements is possible if the cause of the latent seasonal effect can be identified, measured in situ and incorporated into a predictive model.