Society is becoming increasingly concerned about problems related to fertility and pregnancy, birth defects, and developmental abnormalities. Consequently, regulatory agencies such as the U.S. Environmental Protection Agency (EPA) and the Food and Drug Administration (FDA) have placed an increased priority on protecting the public from drugs, chemicals and other environmental exposures that may contribute to these risks.  Because human data are generally limited, data from controlled animal experiments are generally used as the basis for regulation.  This work is motivated by data collected from studies with a segment II design that involve exposing pregnant animals (rats, mice or rabbits) during the period of major organogenesis and structural development. Dose levels for the Segment II design consist of a control group and $3$ or $4$ dose groups, each with $20$ to $30$ pregnant dams.  The dams are sacrificed just prior to normal delivery, at which time the uterus is removed and examined for resorptions and fetal deaths.  The viable fetuses are examined carefully for many different types of malformations, which are commonly classified into three broad categories: external malformations are those visible by naked eye, for instance missing limbs; skeletal malformations might include missing or malformed bones; visceral malformations affect internal organs such as the heart, the brain, the lungs etc. Each specific malformation is typically recorded as a dichotomous variable (present/absent) and fetuses may have several types.  The data, presented in this work, investigate the effects of di (2-ethyhexyl)-phtalate (DEHP) in mice and of ethylene-glycol (EG) in rats.  The DEHP study is concerned about the possible toxic effects of phtalic acid esters.  These are used extensively as plasticizers for numerous plastic devices.  Due to their presence in human and animal tissues, considerable concern has been developed as to their possible toxic effects.  EG is a high-volume industrial chemical with diverse applications.  It may represent little hazard to human health in normal industrial handling, except ehnt used at elevated temperatures.  However, accidental or intentional ingestion is toxic and may result in death.
The analysis of such data raises a number of challenges.  Models that try to approximate the complex data generating mechanism of a developmental toxicity study, should take into account the litter effect and the number of viable fetuses, malformation indicators, weight and clustering, as a function of exposure. Furthermore, the size of the litter may be related to outcomes among live fetuses.  Finally, one may have to deal with outcomes of a mixed continuous-discrete nature.  Scientific interest may be in inference about the dose effect, on implications of model misspecification, on assessment of model fit, etc. Several modelling families exist, such as marginal, conditional, and random-effects models. Advantages and disadvantages are discussed. To overcome computational limitations of classical likelihood approaches, pseudo-likelihood is proposed as an attractive alternative.

GEERT MOLENBERGHS
Biostatistics, Center for Statistics
Limburgs Universitair Centrum
Universitaire Campus, Building D
B-3590 Diepenbeek, Belgium
geert.molenberghs@luc.ac.be
 

In 1999, the U.S.Environmental Protection Agency initiated (1) a national study of fish tissue contaminants in lakes and reservoirs and (2) a biological monitoring study of streams and rivers in twelve western states.  Each study requires the development of a survey design to meet the study objectives.  For the national lake study, a list frame of waterbodies greater than 1 hectare is available.  The frame provides information on the lake surface area and its geographic location, in the form of a geographic information system (GIS) coverage.  However, the frame includes waterbodies that do not meet the definition of the target population.   The frame includes 270,761 waterbodies.  For the western stream study, a "list" frame of stream channels is available as a GIS coverage and provides information of the stream Strahler order and type of stream (perennial/intermittent, natural/manmade).  Again the frame includes stream channels that do not meet the definition of the target population and that have miscoded stream types.  This paper develops the survey designs for these studies and discusses how an underlying discrete global grid can be used to control the spatial distribution of the sample.  The survey designs do not use finite population sampling theory, but a continuous population in a bounded area theory that parallels it.  The spatially-restricted design enables the concept of a systematic sample to be implemented while maintaining the ability to obtain design-based estimates and variance estimates.

ANTHONY R. OLSEN
U.S. Environmental Protection Agemcy
National Health and Environmental Effects Research Laboratory
Western Ecology Division
200 SW 35th Street
Corvallis, Oregon 97333, USA
tolsen@mail.cor.epa.gov