A new chemotherapeutic agent (Peloruside A) was developed from a marine sponge that lives in the Pelorus Sound of New Zealand. Because this agent binds to a different protein site than that used by the taxoid drugs, it may synergize with these agents. Synergistic drug combinations are promising treatments for a variety of diseases, not only for their enhanced efficacy, but also for potential reductions in toxicity. Statistical assessment of synergy, however, is complex and debates rage over the correct methodology. Developing appropriate methodology for a particular application should include consideration of the sources of variability inherent in the experimental design (such as plate-to-plate and well-to-well variability in in-vitro studies). We developed a new method that extends the simple model-free design of Laska et al. to incorporate plate-to-plate and well-to-well variability, while improving the power and robustness of the method (Kelly et al. 2012; Wilmes et al. 2007).
A name-brand manufacturer of golf balls was involved in trademark litigation with a company that recovered used balls from golf courses, restored them and marketed them, claiming that they were “as good as brand-new name-brand balls”. We designed an experiment that would compare the performance of new and refurbished balls, while controlling for sources of variability in performance, such as player ability, wind, temperature, and humidity. A statistical analysis of the experiments showed that the new balls performed better than the refurbished balls and these results were presented in deposition. Additionally, we re-analyzed data gathered by the opposition which also showed the superiority of the new balls (contrary to the opposition’s claims).
Beginning in 1991, New Zealand experienced an epidemic of meningococcal disease; by the end of 2003 more than 5293 cases and 216 deaths had been reported. A “tailor-made” vaccine against the predominant strain was developed by Chiron Vaccines and a nationwide immunization program began in 2004. The staggered vaccine rollout program allowed an epidemiological analysis of its efficacy. Disease rates in vaccinated and unvaccinated areas were compared, adjusting for potential confounding variables such as disease progression over time, socio-economic status, seasonality and geographic region. This analysis showed the vaccine to be effective (Kelly et al. 2007; Kelly et al. 2008)
A medical device company developed a new continuous vital signs monitor. Although an existing ANSI standard explicitly described how to statistically demonstrate accuracy for this type of monitor at a single time point, there was no standard for continuous measurements. We developed statistical methodology to demonstrate accuracy for continuous measurements, wrote the Statistical Analysis Plans, and conducted the data analysis for a FDA 510K submission.
A recent epidemiological study reported an association between residential proximity to ultramafic rock, considered a proxy for naturally-occurring asbestos, and mesothelioma. If this association is true, it would have broad impacts on public health and land-use policies. Inference of a causal relationship is problematic in observational studies; statistically significant associations may result when confounding variables are not properly controlled for in data analyses. We investigated the effects of misclassification in occupational and residential exposure, in the presence of correlations (Kelsh et al. 2009).
How intense human-generated noise (such as from military sonar and explosive device testing) affects marine mammals is of great concern, since it is thought that these sounds may be the source of strandings and beachings, as well as other injuries. We investigated the effects of intense underwater sounds on nervous system activation and immune function in a white whale and a bottlenose dolphin (Romano et al. 2004).
In ecology, the factors influencing mating and breeding success are generally assessed in uncontrolled observational studies with replicated measurements and thus these analyses may be subject to the regression fallacy. The regression fallacy occurs when a researcher attributes changes over time to an intervention or other causal effect, when in fact these changes are explained by the statistical phenomenon known as regression to the mean. We illustrate how to correctly analyze this type of data and avoid committing the regression fallacy (Kelly and Price 2006). This paper was reviewed in Discover magazine (Why Even Tiger Has Off Days, March 2006).
We analyzed survey transect data collected at wind farms to estimate the number of bird and bat fatalities per year. The estimated number of fatalities was corrected for the probability of detection and the probability of scavenging. The bootstrap technique was used to construct interval estimates in addition to the point estimates.
We measured the vocal responses and movements of radio-tagged Florida black rails in response to vocalization tape-playback. We found the probability of response varied significantly with the bird distance from the tape, the bird gender, the nesting status and the time of year. This heterogeneity of response probabilities could have significant impacts on population estimates, if not properly accounted for.