Comparison of machine learning approaches and statistical methods for estimation of airborne chemical exposures

Accurate quantitative estimation of exposure using retrospective data has been one of the most challenging tasks in the exposure assessment field. To improve these estimates, some models have been developed using published exposure databases with their corresponding exposure determinants. These models are designed to be applied to reported exposure determinants obtained from study subjects or exposure levels assigned by an industrial hygienist, so quantitative exposure estimates can be obtained. ^ In an effort to improve the prediction accuracy and generalizability of these models, and taking into account that the limitations encountered in previous studies might be due to limitations in the applicability of traditional statistical methods and concepts, the use of computer science- derived data analysis methods, predominantly machine learning approaches, were proposed and explored in this study. ^ The goal of this study was to develop a set of models using decision trees/ensemble and neural networks methods to predict occupational outcomes based on literature-derived databases, and compare, using cross-validation and data splitting techniques, the resulting prediction capacity to that of traditional regression models. Two cases were addressed: the categorical case, where the exposure level was measured as an exposure rating following the American Industrial Hygiene Association guidelines and the continuous case, where the result of the exposure is expressed as a concentration value. Previously developed literature-based exposure databases for 1,1,1 trichloroethane, methylene dichloride and, trichloroethylene were used. ^ When compared to regression estimations, results showed better accuracy of decision trees/ensemble techniques for the categorical case while neural networks were better for estimation of continuous exposure values. Overrepresentation of classes and overfitting were the main causes for poor neural network performance and accuracy. Estimations based on literature-based databases using machine learning techniques might provide an advantage when they are applied to other methodologies that combine `expert inputs' with current exposure measurements, like the Bayesian Decision Analysis tool. The use of machine learning techniques to more accurately estimate exposures from literature-based exposure databases might represent the starting point for the independence from the expert judgment.^

Use of short term test systems for the prediction of the hazard represented by potential chemical carcinogens

It has been hypothesized that results from the short term bioassays will ultimately provide information that will be useful for human health hazard assessment. Although toxicologic test systems have become increasingly refined, to date, no investigator has been able to provide qualitative or quantitative methods which would support the use of short term tests in this capacity.^ Historically, the validity of the short term tests have been assessed using the framework of the epidemiologic/medical screens. In this context, the results of the carcinogen (long term) bioassay is generally used as the standard. However, this approach is widely recognized as being biased and, because it employs qualitative data, cannot be used in the setting of priorities. In contrast, the goal of this research was to address the problem of evaluating the utility of the short term tests for hazard assessment using an alternative method of investigation.^ Chemical carcinogens were selected from the list of carcinogens published by the International Agency for Research on Carcinogens (IARC). Tumorigenicity and mutagenicity data on fifty-two chemicals were obtained from the Registry of Toxic Effects of Chemical Substances (RTECS) and were analyzed using a relative potency approach. The relative potency framework allows for the standardization of data "relative" to a reference compound. To avoid any bias associated with the choice of the reference compound, fourteen different compounds were used.^ The data were evaluated in a format which allowed for a comparison of the ranking of the mutagenic relative potencies of the compounds (as estimated using short term data) vs. the ranking of the tumorigenic relative potencies (as estimated from the chronic bioassays). The results were statistically significant (p $<$.05) for data standardized to thirteen of the fourteen reference compounds. Although this was a preliminary investigation, it offers evidence that the short term test systems may be of utility in ranking the hazards represented by chemicals which may be human carcinogens. ^

THE SHORT- AND LONG-TERM EFFECTS OF A HANDBOOK ON THE ANTIMICROBIAL PRESCRIBING PATTERNS OF HOSPITAL PHYSICIANS

This study evaluates the effect of a specially designed, physician-oriented handbook of antimicrobial use on the prescribing patterns of a group of fifty doctors at a university hospital. Data were evaluated over a peroid of one-and-one-half years, before and after the distribution of the handbook. For the purposes of this study, antimicrobial therapy was classified: (1) inappropriate if it violated one of a number of recognized principles of antimicrobial therapy, (2) appropriate if it agreed with specific recommendations or alternatives given in the distributed reference handbook, and (3) acceptable if it was neither inappropriate nor appropriate as defined by the handbook. An initial survey of antimicrobial prescribing patterns was made. Five months later the handbook was distributed and a two-week orientation program, consisting of the distribution and promotion of the problem-oriented, pocket-size handbook of appropriate antimicrobial therapy, was conducted. The handbook, which was developed by the authors and reviewed and approved by a panel of infectious disease specialists, presented guidelines for appropriate and efficacious usage of antimicrobial agents as most currently accepted in common clinical infections. Subsequent surveys were then conducted two weeks, three months, and six months after distribution of the handbook. A statistically significant difference (p < 0.01) in antimicrobial prescribing patterns was noted between the survey conducted two weeks after the introduction of the handbook and the other surveys. In this survey, while therapy classified inappropriate decreased from 44% to 28%, therapy appropriate as recommended increased from 31% to 53%. The findings of this study demonstrate that the introduction and promotion of the handbook decreases abuse and increases proper use of antimicrobial therapy, although the effect is sustainable for only a short duration--no longer than three months. These results indicate the need for a vigorous, updated program to achieve and maintain current appropriate antibotic therapy in clinical medicine. ^