The utility of structure-activity relationship (SAR) models for prediction and covariate selection in developmental toxicity: comparative analysis of logistic regression and decision tree models.

Structure-activity relationship (SAR) models can be used to predict the biological activity of potential developmental toxicants whose adverse effects include death, structural abnormalities, altered growth and functional deficiencies in the developing organism. Physico-chemical descriptors of spatial, electronic and lipophilic properties were used to derive SAR models by two modeling approaches, logistic regression and Classification and Regression Tree (CART), using a new developmental database of 293 chemicals (FDA/TERIS). Both single models and ensembles of models (termed bagging) were derived to predict toxicity. Assessment of the empirical distributions of the prediction measures was performed by repeated random partitioning of the data set. Results showed that both the decision tree and logistic regression derived developmental SAR models exhibited modest prediction accuracy. Bagging tended to enhance the prediction accuracy and reduced the variability of prediction measures compared to the single model for CART-based models but not consistently for logistic-based models. Prediction accuracy of single logistic-based models was higher than single CART-based models but bagged CART-based models were more predictive. Descriptor selection in SAR for the understanding of the developmental mechanism was highly dependent on the modeling approach. Although prediction accuracy was similar in the two modeling approaches, there was inconsistency in the model descriptors.

Decision tree SAR models for developmental toxicity based on an FDA/TERIS database.

Humans are exposed to thousands of environmental chemicals for which no developmental toxicity information is available. Structure-activity relationships (SARs) are models that could be used to efficiently predict the biological activity of potential developmental toxicants. However, at this time, no adequate SAR models of developmental toxicity are available for risk assessment. In the present study, a new developmental database was compiled by combining toxicity information from the Teratogen Information System (TERIS) and the Food and Drug Administration (FDA) guidelines. We implemented a decision tree modeling procedure, using Classification and Regression Tree software and a model ensemble approach termed bagging. We then assessed the empirical distributions of the prediction accuracy measures of the single and ensemble-based models, achieved by repeating our modeling experiment many times by repeated random partitioning of the working database. The decision tree developmental SAR models exhibited modest prediction accuracy. Bagging tended to enhance the accuracy of prediction. Also, the model ensemble approach reduced the variability of prediction measures compared to the single model approach. Further research with data derived from animal species- and endpoint-specific components of an extended and refined FDA/TERIS database has the potential to derive SAR models that would be useful in the developmental risk assessment of the thousands of untested chemicals.

Classification of chronic radiation sickness cases using neural networks and classification trees.

Chronic radiation sickness is a deterministic radiation health effect observed among the Mayak Production Association workers in Russia. In this study, unsupervised neural networks were used to cluster hematological measurements in a subset (n = 88) of the Mayak Production Association population while excluding from the analysis the radiation dose and the historical clinical diagnosis. Clusters of observations that had lower average leukocyte and thrombocyte counts were labeled "affected" and those having higher average blood cell counts were labeled "unaffected." The class (cluster) membership for each individual was used subsequently as a dependent variable in a classification tree model in order to identify significant features of the underlying classification model. After re-classification of cases using this method, the results showed a better data separation between the blood cell counts for affected vs. unaffected groups compared to those based on historical classification, and a greater difference between group means for differential blood counts was observed than for the historical diagnosis. The reclassification of diagnostic groups changed the group mean radiation doses. The geometric means (and 95% CL) of cumulative radiation dose equivalent from external exposures, based on the historical diagnosis, are 0.31 (0.0035, 3.4) vs. 1.7 (0.0007, 18) Sv. After clustering and classification tree analyses, the group geometric means were 0.78 (0.0014, 8.6) vs. 1.5 (0.0007, 17) and 0.82 (0.0013, 9.0) vs. 1.4 (0.0008, 16) Sv, using (respectively) whole blood cell counts or differential counts as the independent variables. The approach presented here is useful as a diagnostic aid for both retrospective analyses and in the event of future radiation accidents.

Elevated levels of S-nitrosoalbumin in preeclampsia plasma.

The availability of nitric oxide (NO), which is required for the normal regulation of vascular tone, may be decreased in preeclampsia, thus contributing to the vascular pathogenesis of this pregnancy disorder. Because ascorbate is essential for the decomposition of S-nitrothiols and the release of NO, we speculated that the ascorbate deficiency typical of preeclampsia plasma might result in decreased rates of decomposition of S-nitrosothiols. We tested the hypothesis that total S-nitrosothiol and S-nitrosoalbumin concentrations are increased in preeclampsia plasma, reflecting a decreased release of NO from these major reservoirs of NO. Gestationally matched plasma samples were obtained (before labor or intravenous MgSO(4)) from 21 women with preeclampsia and 21 women with normal pregnancy, and plasma samples were also obtained from 12 nonpregnant women of similar age and body mass index during the follicular phase of the menstrual cycle. All were nonsmokers. The assay included ultraviolet-induced decomposition of S-nitrosothiols to liberate NO captured by a florigenic reagent, 4,5-diaminofluoresceine, to produce diaminofluoresceine-Triazole. Preeclampsia plasma contained significantly higher concentrations of total S-nitrosothiols (11.1+/-2.9 nmol/mL) than normal pregnancy samples (9.4+/-1.5 nmol/mL). Even greater differences were found between preeclampsia plasma and plasma samples from normal pregnancies and nonpregnant women (294+/-110, 186+/-25, and 151+/-25 pmol/mg protein, respectively) when S-nitrosothiol content was expressed per milligram protein. The albumin fraction contained 49.4% of total plasma S-nitrosothiols in the control samples and 53.7% and 56.8% of plasma S-nitrosothiols in normal pregnancy and preeclampsia, respectively. The level of S-nitrosoalbumin was significantly higher in preeclampsia than in normal pregnancy or nonpregnancy plasma (6.3+/-1.4, 5.1+/-0.7, and 4.2+/-1.0 nmol/mL, respectively). The increased concentration of S-nitrosoalbumin in preeclampsia almost completely accounted for the increased levels of S-nitrosothiols in total plasma. Due to combined increases in nitrosothiols and decreases in protein, the preeclampsia plasma concentration of S-nitrosoalbumin was greatly increased on a per milligram of protein basis (271% and 186% compared with normal nonpregnancy and normal pregnancy plasma, respectively). We conclude that S-nitrosoalbumin and total S-nitrosothiol concentrations are significantly increased in preeclampsia plasma and may reflect insufficient release of NO groups in this condition.

Physicochemical and graph theoretical descriptors in developmental toxicity SAR: a comparative study.

Chemical insults to the developing fetus can lead to growth retardation, malformation, death, and functional deficits. The present study seeks to determine if physicochemical and/or graph theoretical parameters can be used to determine a structure-activity relationship (SAR) for developmental toxicity, and if consistency is observed among the selected features. The biological data utilized consists of a diverse series of compounds evaluated within the Chernoff-Kavlock in vivo mouse assay. Physicochemical parameters calculated correspond to electronic, steric, and transport properties. Graph theoretical parameters calculated include the simple, valence, and kappa indices. Both sets of parameters were independently applied to derive SARs in order to compare the quality of the respective models. Multiple random sampling, without replacement, was utilized to obtain ten training/test partitions. Models were built by linear discriminant analysis, decision trees, and neural networks respectively. Comparisons on identical sets of data were carried out to determine if any of the model building procedures had a significant advantage in terms of predictive performance. Furthermore, comparison of the features selected within and across the model building processes led to the determination of model consistency. Our results indicate that consistent features related to developmental toxicity are observed and that both physicochemical and graph theoretical parameters have equal utility.

Deterministic effects from occupational radiation exposures in a cohort of Mayak PA workers: data base description.

Project 2.3 of the Joint Coordinating Committee on Radiation Effects Research (JCCRER) is a study of deterministic health effects among a cohort of Russia nuclear workers. The preliminary study population includes a stratified random sample of 221 radiation workers who were employed in a cohort of 8,055 workers at the Mayak PA facilities for at least one year during the period from 1948 to 1958. High annual doses, approaching 1 Gy per year from external and internal radiation sources, were reported for a significant proportion of the workers in this cohort. The present data set includes 96 cases of chronic radiation sickness (CRS), 14 cases of acute radiation syndrome (ARS) and 13 cases of plutonium pneumosclerosis (PPn). The remainder of the sample consists of "uninjured workers" who had no known history of radiation illness or injury; however, the uninjured workers are not "controls" for radiation exposure. The data base is currently being expanded to 600 individuals sampled from the cohort of workers from 1948 to 1958 to allow a more complete analysis of the deterministic health effects and comparisons with existing health effect models. The final data base will be used with state-of-the-art modeling techniques to determine threshold doses and dose-response relationships for key clinical diagnostic variables.

A robust structure-activity relationship (SAR) model for esters that cause skin irritation in humans.

A structure-activity relationship (SAR) model has been developed to discriminate skin irritant from nonirritant esters. The model is based on the physicochemical properties of 42 esters that were tested in humans for skin irritation. Nineteen physicochemical parameters that represent transport, electronic, and steric properties were calculated for each chemical. Best subsets regression analysis indicated candidate models for further analysis. Regression analyses identified significant models (p < 0.05) that had variables that were also significant (p < 0.05). These candidate models were evaluated using linear discriminant analysis to determine if the irritant esters could be discriminated from nonirritant esters. The stability of the model was evident from the consistency of parameters among ten submodels generated using multiple random sampling of the database. The sensitivity of the ten models, evaluated by "leave-one-out" cross-validation, ranged from 0. 846 to 0.923, with a mean of 0.885 +/- 0.025 (95% CI). The specificity ranged from 0.615 to 0.923, with a mean of 0.738 +/- 0.06 (CI). Compared with nonirritant esters, irritant esters had lower density, lower water solubility, lower sum of partial positive charges, higher Hansen hydrogen bonding parameter, and higher Hansen dispersion parameter. The results indicate that physicochemical features of esters contribute to their ability to cause skin irritation in humans, and that chemical partitioning into the epidermis and intermolecular reactions are likely important components of the response. This model is applicable for prediction of human irritation of esters yet untested.

Development, characterization and application of predictive-toxicology models.

The adoption of SAR techniques for risk assessment purposes requires that the predictive performance of models be characterized and optimized. The development of such methods with respect to CASE/MULTICASE are described. Moreover, the effects of size, informational content, ratio of actives/inactives in the model on predictivity must be determined. Characterized models can provide mechanistic insights: nature of toxicophore, reactivity, receptor binding. Comparison of toxicophores among SAR models allows a determination of mechanistic overlaps (e.g., mutagenicity, toxicity, inhibition of gap junctional intercellular communication vs. carcinogenicity). Methods have been developed to combine SAR submodels and thereby improve predictive performance. Now that predictive toxicology methods are gaining acceptance, the development of Good Laboratory Practices is a further priority, as is the development of graduate programs in Computational Toxicology to adequately train the needed professional.

Issues and findings in the evaluation of occupational risk among women high nickel alloys workers.

BACKGROUND: We present the mortality experience for a cohort of women (n = 2,877) from a large epidemiologic study of production and fabrication high nickel alloys workers (n = 31,165). All the plants were located within the United States and cohort eligibility required some work experience within the period of the late 1940s through the mid 1960s. METHODS: Vital status follow-up was through the end of 1988 and incorporated information from multiple sources. Cause-specific mortality was evaluated by comparing cohort mortality to the general United States female population and to local populations in geographic proximity to the plants. Relative risk estimates were determined for 62 cause of death categories using the Standardized Mortality Ratio (SMR) and were adjusted for age, race, gender, and calendar time by the indirect method using a modified life table technique. RESULTS: Relative risks for all causes (0.98), all cancers (0.90), lung cancer (1.34), and breast cancer (0.96) were nonsignificant when mortality was compared to the US female population. No relationship between mortality and length of time employed in the industry or work area was identified. DISCUSSION: Although there were some difficulties in tracing women due to name changes, comprehensive follow-up was obtained when using multiple sources of information. Our strategy resulted in resolving vital status for over 95% of the women, which is comparable to that of the male cohort. The type of jobs and work activities differed between genders. Females were employed predominantly in two work areas (allocated services, 87%, and grinding, 46%), whereas males were employed in several work areas (allocated services, 76%, grinding, 27%, hot working, 20%, and cold working, 17%). Considerable variation was noted among the study plants with respect to the percent of female production workers in the workforce. Generally, the patterns of relative risks derived for the total cohort and various subgroups are similar across the different comparison populations. Estimated elevated risks are usually lower when cohort mortality is compared to that of local populations. No increased risks were identified for any site-specific cancers or nonmalignant causes of death.

Modeling adverse environmental impacts on the reproductive system.

When priority topics are being established for the study of women's health, it is generally agreed that one important area on which to focus research is reproduction. For example, increasing attention has been directed to environmental exposures that disrupt the endocrine system and alter reproduction. These concerns also suggest the need to give greater attention to the use of animal toxicologic testing to draw inferences about human reproductive risks. Successful reproduction requires multiple simultaneous and sequential processes in both the male and female, and the effect of toxicity on reproduction-related processes is time dependent. Currently, however, the risk assessment approach does not allow for the use of multiple processes or for considering the reproductive process response as a function of time. We discuss several issues in modeling exposure effects on reproductive function for risk assessment and present an overview of approaches for reproductive risk assessment. Recommendations are provided for an effective animal study design for determining reproductive risk that addresses optimization of the duration of dosing, observation of the effects of exposure on validated biomarkers, analysis of several biomarkers for complete characterization of the exposure on the underlying biologic processes, the need for longitudinally observed exposure effects, and a procedure for estimating human reproductive risk from the animal findings. An approach to characterizing reproductive toxicity to estimate the increased fertility risks in a dibromochloropropane (DBCP)-exposed human population is illustrated, using several reproductive biomarkers simultaneously from a longitudinal rabbit inhalation study of DBCP and an interspecies extrapolation method.

Using alternative comparison populations to assess occupation-related mortality risk. Results for the high nickel alloys workers cohort.

The focus of this article is to examine how the choice of comparison group affects the identification and interpretation of cause-specific health risks in occupational cohorts when different external control populations are used. The mortality experience of approximately 31,000 high nickel alloys workers is compared with the total US population and to local populations in geographic proximity to the plants. Generally, the patterns of relative risks derived for the total cohort and various subgroups are similar across the different comparison populations. Estimated elevated risks are usually lower when cohort mortality is compared with that of local populations. An overall significant 13% risk for lung cancer is noted when compared with that of the total US population. However, no significant excess is identified when local populations are used. Subset analysis identified significant excesses of colon cancer among nonwhite males (50%-150%) and kidney cancer among white male workers employed in melting (approximately 100%), irrespective of the comparison population.

Evaluating clinical case report data for SAR modeling of allergic contact dermatitis.

Clinical case reports can be important sources of information for alerting health professionals to the existence of possible health hazards. Isolated case reports, however, are weak evidence of causal relationships between exposure and disease because they do not provide an indication of the frequency of a particular exposure leading to a disease event. A database of chemicals causing allergic contact dermatitis (ACD) was compiled to discern structure-activity relationships. Clinical reports represented a considerable fraction of the data. Multiple Computer Automated Structure Evaluation (MultiCASE) was used to create a structure-activity model to be used in predicting the ACD activity of untested chemicals. We examined how the predictive ability of the model was influenced by including the case report data in the model. In addition, the model was used to predict the activity of chemicals identified from clinical case reports. The following results were obtained: When chemicals which were identified as dermal sensitizers by only one or two case reports were included in the model, the specificity of the model was reduced. Less than one half of these chemicals were predicted to be active by the most highly evidenced model. These chemicals possessed substructures not previously encountered by any of the models. We conclude that chemicals classified as sensitizers based on isolated clinical case reports be excluded from our model of ACD. The approach described here for evaluating activity of chemicals based on sparse evidence should be considered for use with other endpoints of toxicity when data are correspondingly limited.

Radiographic abnormalities in relation to total dust exposure at a bauxite refinery and alumina-based chemical products plant.

A cross-sectional study of 788 male employees of an aluminum production company examined the relationship of radiographic abnormalities to smoking and dust exposure from the mining and refining of bauxite to alumina. Among the aluminas produced were low temperature range transitional forms. The present analyses were limited to nonsmokers and current smokers. Two National Institute of Occupational Safety and Health (NIOSH)-certified "B" readers interpreted the radiographs. The predominant radiographic abnormalities noted were scanty, small, irregular opacities in the lower zones of profusion 0/1 to 1/1. Rounded opacities were rare. Among nonsmokers with low dust exposures, the prevalence of opacities greater than or equal to 1/0 showed no trend with increasing age and duration of exposure, suggesting no relationship between age and prevalence of opacities of Category 1 or more in this cohort (p greater than 0.10). Nonsmokers who had accumulated higher dust exposures showed a trend of increasing prevalence of opacities with increasing duration, suggesting an effect of occupational exposure at higher cumulative exposure levels (p less than 0.05). In most exposure categories, smokers exceeded nonsmokers in their prevalence of opacities greater than or equal to 1/0; the overall prevalence among smokers being 12 and 11% according to Readers A and B, respectively, compared with 4% in nonsmokers (p less than 0.01). In conclusion, 7 to 8% of aluminum workers in this cohort had radiographic findings of scanty, small, irregular opacities, the prevalence of which was increased among smokers (p less than 0.01). There was a moderate increase in the prevalence of opacities with increasing tenure in nonsmokers with high cumulative exposures (p less than 0.05).

Pulmonary function in relation to total dust exposure at a bauxite refinery and alumina-based chemical products plant.

A cross-sectional study of 1,142 male employees at the Arkansas Operations of a large aluminum production company examined the effect on pulmonary function of chronic exposure to total dust produced in the mining and refining of bauxite and the production of alumina chemicals. Never smokers, ex-smokers, and current smokers were analyzed separately. Among never smokers, a pattern of decreasing FEV1 was observed in relation to increasing duration and cumulative total dust exposure. Among never smokers with cumulative total dust exposures of greater than or equal to 100 mg/m3 yr and greater than or equal to 20 yr of exposure, there was a mean reduction from the predicted FEV1 of 0.29 to 0.39 L, in addition to a 3- to 4-fold excess of observed/expected numbers of subjects with FEV1 less than 80% of predicted. These results were observed relative to an external and an internal comparison group. Among current smokers, the deviations from predicted and the excess numbers of subjects with FEV1 less than 80% of predicted were larger in all exposure groups than for the never smokers. However, the quality of the smoking data was inadequate to allow separation of the effects of smoking and dust exposure.

Uncertainty, risk, and the role of epidemiology in public policy development.

Risk taking is a concommitant of life--inescapable beyond the inert, sessile state. Accordingly, risk assessment is implicit for almost all human activities, from daily actions, such as crossing a trafficked street to esoteric ones such as landing on the moon.