Assessment of the effect on statistical power of regression model misspecification by using techniques of mathematical statistics and simulation study

Objectives. This paper seeks to assess the effect on statistical power of regression model misspecification in a variety of situations. ^ Methods and results. The effect of misspecification in regression can be approximated by evaluating the correlation between the correct specification and the misspecification of the outcome variable (Harris 2010).In this paper, three misspecified models (linear, categorical and fractional polynomial) were considered. In the first section, the mathematical method of calculating the correlation between correct and misspecified models with simple mathematical forms was derived and demonstrated. In the second section, data from the National Health and Nutrition Examination Survey (NHANES 2007-2008) were used to examine such correlations. Our study shows that comparing to linear or categorical models, the fractional polynomial models, with the higher correlations, provided a better approximation of the true relationship, which was illustrated by LOESS regression. In the third section, we present the results of simulation studies that demonstrate overall misspecification in regression can produce marked decreases in power with small sample sizes. However, the categorical model had greatest power, ranging from 0.877 to 0.936 depending on sample size and outcome variable used. The power of fractional polynomial model was close to that of linear model, which ranged from 0.69 to 0.83, and appeared to be affected by the increased degrees of freedom of this model.^ Conclusion. Correlations between alternative model specifications can be used to provide a good approximation of the effect on statistical power of misspecification when the sample size is large. When model specifications have known simple mathematical forms, such correlations can be calculated mathematically. Actual public health data from NHANES 2007-2008 were used as examples to demonstrate the situations with unknown or complex correct model specification. Simulation of power for misspecified models confirmed the results based on correlation methods but also illustrated the effect of model degrees of freedom on power.^

MEASUREMENT OF TUMOR BLOOD FLOW FOLLOWING NEUTRON IRRADIATION (ACTIVATION)

Clinical oncologists and cancer researchers benefit from information on the vascularization or non-vascularization of solid tumors because of blood flow's influence on three popular treatment types: hyperthermia therapy, radiotherapy, and chemotherapy. The objective of this research is the development of a clinically useful tumor blood flow measurement technique. The designed technique is sensitive, has good spatial resolution, in non-invasive and presents no risk to the patient beyond his usual treatment (measurements will be subsequent only to normal patient treatment).^ Tumor blood flow was determined by measuring the washout of positron emitting isotopes created through neutron therapy treatment. In order to do this, several technical and scientific questions were addressed first. These questions were: (1) What isotopes are created in tumor tissue when it is irradiated in a neutron therapy beam and how much of each isotope is expected? (2) What are the chemical states of the isotopes that are potentially useful for blood flow measurements and will those chemical states allow these or other isotopes to be washed out of the tumor? (3) How should isotope washout by blood flow be modeled in order to most effectively use the data? These questions have been answered through both theoretical calculation and measurement.^ The first question was answered through the measurement of macroscopic cross sections for the predominant nuclear reactions in the body. These results correlate well with an independent mathematical prediction of tissue activation and measurements of mouse spleen neutron activation. The second question was addressed by performing cell suspension and protein precipitation techniques on neutron activated mouse spleens. The third and final question was answered by using first physical principles to develop a model mimicking the blood flow system and measurement technique.^ In a final set of experiments, the above were applied to flow models and animals. The ultimate aim of this project is to apply its methodology to neutron therapy patients. ^

Respiratory rhythm generation: A modeling study of the respiratory central pattern generator and the phrenic motor neuron

The respiratory central pattern generator is a collection of medullary neurons that generates the rhythm of respiration. The respiratory central pattern generator feeds phrenic motor neurons, which, in turn, drive the main muscle of respiration, the diaphragm. The purpose of this thesis is to understand the neural control of respiration through mathematical models of the respiratory central pattern generator and phrenic motor neurons. ^ We first designed and validated a Hodgkin-Huxley type model that mimics the behavior of phrenic motor neurons under a wide range of electrical and pharmacological perturbations. This model was constrained physiological data from the literature. Next, we designed and validated a model of the respiratory central pattern generator by connecting four Hodgkin-Huxley type models of medullary respiratory neurons in a mutually inhibitory network. This network was in turn driven by a simple model of an endogenously bursting neuron, which acted as the pacemaker for the respiratory central pattern generator. Finally, the respiratory central pattern generator and phrenic motor neuron models were connected and their interactions studied. ^ Our study of the models has provided a number of insights into the behavior of the respiratory central pattern generator and phrenic motor neurons. These include the suggestion of a role for the T-type and N-type calcium channels during single spikes and repetitive firing in phrenic motor neurons, as well as a better understanding of network properties underlying respiratory rhythm generation. We also utilized an existing model of lung mechanics to study the interactions between the respiratory central pattern generator and ventilation. ^

The assumptions and effects of conservative procedures in low -dose risk assessment for setting safety standards

Conservative procedures in low-dose risk assessment are used to set safety standards for known or suspected carcinogens. However, the assumptions upon which the methods are based and the effects of these methods are not well understood.^ To minimize the number of false-negatives and to reduce the cost of bioassays, animals are given very high doses of potential carcinogens. Results must then be extrapolated to much smaller doses to set safety standards for risks such as one per million. There are a number of competing methods that add a conservative safety factor into these calculations.^ A method of quantifying the conservatism of these methods was described and tested on eight procedures used in setting low-dose safety standards. The results using these procedures were compared by computer simulation and by the use of data from a large scale animal study.^ The method consisted of determining a "true safe dose" (tsd) according to an assumed underlying model. If one assumed that Y = the probability of cancer = P(d), a known mathematical function of the dose, then by setting Y to some predetermined acceptable risk, one can solve for d, the model's "true safe dose".^ Simulations were generated, assuming a binomial distribution, for an artificial bioassay. The eight procedures were then used to determine a "virtual safe dose" (vsd) that estimates the tsd, assuming a risk of one per million. A ratio R = ((tsd-vsd)/vsd) was calculated for each "experiment" (simulation). The mean R of 500 simulations and the probability R $<$ 0 was used to measure the over and under conservatism of each procedure.^ The eight procedures included Weil's method, Hoel's method, the Mantel-Byran method, the improved Mantel-Byran, Gross's method, fitting a one-hit model, Crump's procedure, and applying Rai and Van Ryzin's method to a Weibull model.^ None of the procedures performed uniformly well for all types of dose-response curves. When the data were linear, the one-hit model, Hoel's method, or the Gross-Mantel method worked reasonably well. However, when the data were non-linear, these same methods were overly conservative. Crump's procedure and the Weibull model performed better in these situations. ^