Lack of Statistical Power

Clearly, hypothesis-testing using data from currently used toxicity test protocols cannot detect effects well at low concentrations. This is due in part to the lack of statistical power, given the number of replicates and the intrinsic laboratory and organismal variability within the experiments. Current assumptions that NOELs are a no-effect or at a safe level are also not warranted. The above studies also indicate that the level of effect that the NOEL represents is highly variable. LOELs are similarly uninformative. 

The only available study on cancer in humans is limited by its lack of statistical power (Siemiatycki et al. 1987). The only chronic animal study is a briefly reported dermal study on possible carcinogenic effects in mice that was limited by the use of a mixture containing Stoddard solvent (90% Stoddard solvent, 7% calcium petroleum sulfonate, and 3% ethylene glycol monobutyl ether) (ERA 1984c). A follow-up to this study, using Stoddard solvent alone, would be useful. Concerns have been raised about the genotoxicity of some individual components of Stoddard solvent. Treatment of V79 Chinese hamster cells by />decane alone did not cause mutagenesis, but in combination with... 

The committee, however, recognizes that there are methodological factors (e.g., length of follow-up, confounding factors, lack of statistical power), practical concerns (e.g., continuity of the research team, tracking of subjects, record retention), and cost considerations that limit the implementation of certain in-market surveillance programs. 

Reproductive Effects. Antispermatogenic effects and possible effects on fertility have been reported in humans occupationally exposed to 1,2-dibromoethane (Heinrichs 1983 Ratcliffe et al. 1987 Ter Haar 1980 Wong et al. 1979). However, many of these studies lacked sufficient statistical power to detect an association between parameters measured and exposure. 

The models developed for the low salt concentrations (0.0 and 0.1 M NaCl) were very different from those for the high salt concentration. The two basic Independent variables used were salt level and the absolute value of the pH minus 4.0. To avoid Implying that the behavior of each functional property on either side of pH 4.0 Is the mirror Image of Its behavior on the other side of pH 4.0, two variables were formed from the absolute value of pH minus 4.0. These were the absolute values of pH minus 4.0 for each pH above 4.0 (values of this variable for observations In which the pH was less than 4.0 were set equal to zero) and the absolute values of pH minus 4.0 for each pH below 4.0 (values of this variable for observations In which pH was greater than 4.0 were set equal to zero). As shown In the table, these basic variables were used In the estimated models along with their squares, their Interactions with salt level (0.0 and 0.1 M NaCl), and salt level to form the Independent variables In the final equations used. Other variables, such as cubic powers of pH minus 4.0, were tried and discarded due to lack of statistical significance In arriving at the final models. 

As reviewed in Kortenkamp et al. (2007), an essential requirement for experimental studies intended to address the issue of mixture effects at doses or concentrations below NOAELs is that NOAELs are estimated for each mixture component by using the same assay system (and endpoint) that is chosen for the mixture study, ideally under identical experimental conditions. Ignoring this requirement can lead to the inadvertent administration of some or all mixture components at doses higher than NOAELs, and would undermine the aim of the experiment. On the other hand, delivery of doses or concentrations smaller than the NOAEL, either by design or by accident, might present problems if the experimental system lacks the statistical power to detect effects. For example, it would be futile to attempt an experiment where 2 agents are combined at 1/100 of their individual NOEL. The resulting mixture effect, if it exists, would be too small to be detectable in most cases, and the experiment would be inconclusive. 

Cohort Studies. To date, a number of cohort studies have been conducted of chemical workers from the U.S. and Europe and of herbicide sprayers from Sweden and Finland (20-28). In general, the studies have not demonstrated a definitive relationship between mortality from any malignant or nonmalignant condition and exposure to chlorophenols or phenoxy herbicides. All of the studies lacked sufficient statistical power due to the small number of individuals in each cohort. 

The amount data corresponding to the response values in 1 above were transformed by the same general family of power transformations until linearity was obtained. The F-test statistic that relates lack of fit and pure error was used as the criterion for linearity. 

To the extent that there was a significant reduction of the 11-deoxycortisol response in PTSD in the absence of an attenuated ACTH response, this would indeed support the idea of a reduced adrenal output. However, the trend for an ACTH response suggests that part of the failure to achieve statistical significance may have also occurred because of Hmited power, particularly given the lack of evidence for increased ambient ACTH levels in PTSD relative to normal controls. Dose-response studies using the higher vs lower dose of metyrapone should certainly be conducted to further address this critical issue. 

Three cohort studies have included workers exposed to ethylene dibromide, but because of their low statistical power and/or lack of information about individual exposures, little can be concluded about the carcinogenicity of this compound in humans. 

Although the results of these studies mostly support the notion that yogurt has immunostimu-latory effects, poor study design, lack of appropriate controls, and short duration of most of the studies limit the value of the conclusions that can be drawn from them. Most early animal and human studies included too few animals or subjects in each group and most did not include statistical analysis. Although more recent studies addressed these points, none provided the statistical basis for the selected number of subjects that is, it seems that no power calculations were performed. 

---