Document VooL6k4EbyzKDno9jbV79D88

responsible for the Bohr effect in the Hb A molecule. The main aspect of this investigation consists ofthe role ofj?146His, an amino add residue previouriy shown to contribute significantly to the alkaline Bohr effect in Hb A. We have found that, in the absence ofphosphate ions and at low chloride ion concentrations, the pK value ofj3146His does not change significantly upon ligation; thus, under these experimental conditions, the contribution of/S146His to the alkaline Bohr effect is negligible. We have concluded that the molecular mechanism ofthe Bohr effect is not only the result ofthe quaternary and tertiary structure changes brought about in tile Hb A molecule upon ligation, but it also involves specific interactions ofone or several oxygen-linked groups with the solvent molecules or ions. For the case ofthe Hb S molecule, we have found that three histidine residues in CO-Hb S and six histidine residues in deoxy-Hb S have pK values and proton intrinsic chemical shifts which are different from the corresponding ones in Hb A. These findings indicate that the surface conformation ofthe Hb S molecule is different from that ofthe Hb A molecule. The molecular mechanism ofthe polymerization process of Hb S has been investigated by measuring the longitudinal relaxation rates (Tj-1) of the C2 protons ofthe eleven observahle.histidine.rcsidues in deoxy-Hb S and deoxy-Hb A as a function ofhemoglobin concentration and temperature. We have found that, under certain experimental conditions, several surface histidine C2 protons have increased Tj-1 values in deoxy-Hb S as compared to the corresponding ones in deoxy-Hb A. When the Hb S solution approaches gelation .these differences in Tj-t values between deoxy- Hb S and deoxy-Hb A are enhanced. This result suggests that the regions on the surface ofthe Hb S molecule on which these histidine residues are situated are involved in the early stages ofthe polymerization process. We propose that the increasedTf1 values ofseveral histidine resonances in deoxy-Hb S result from the formation in the Hb S solution ofsmall pregelation aggregates prior to the main polymerization. The histidine residues for which the differences in Tf*values are tire largest are viewed as hang located at or near the "contact" areas between the Hb S molecules within the pregelation aggregate. BIOSTATISTICS A MULTIVARIATE ANALYSIS OF THE INCIDENCE OF ELEVATED BLOOD LEAD LEVEL IN CHILDREN Order No. *015388 Al -Ra w i, Zey ad Ra s h a d , Ph .D. University of'Pittsburgh, 1979. 170pp. In the presented dissertation a major effort has been made to determine the environmental andsorio-economic variables that increase the probability ofa child between the ages ofone and six having an elevated blood lead level (EBL). It has been shown, given that environmental lead sources exist inareas, a child's socio-economic variables will influence the probability ofingestion ofabnormal quantities of lead. Eight socio economic variables have been considered which have been found to have some association with the dependent variable - the Proportion ofEBL These variables are: Xj = The proportion ofhouse units built before 1940; X2 = The proportion ofpopulation below poverty level; X3 -- The proportion ofchildren < 6 years old living with one parent families; X4 = The proportion ofpopuIation < 6 years old; X5 e= The proportion ofblacks among children < 6 years old; Xt = The proportion ofchildren 3 - 6 years old not attending pre-school; X7 = The average family size for children < 6 years old; and = The proportion ofpopulation > 25 years old having < 10 years schooling. The 1970 census data were used to calculate these variables by census tract in Pittsburgh, Philadelphia, and Memphis. Analysis ofvariance studies were performed to see ifthe census tracts were homogeneous with regard to their mean blood lead level in the three dries. It was found that they were not A principal component analysis was performed to develop a model as a linear combination ofthese sodo-economic variables which utilizes appropriate weights for these variables - variables with small variance which provide more information receive large weights and variables with large variance which provide little information receive small weights. When the model developed from Pittsburgh data was applied to the data from two other dties, the percent ofexplained variation of EBL was 40.22% in Philadelphia and 32.77% in Memphis. For biological data, these percentages ofexplained variation would be considered large especially if this is done by using non-biological independent variables. Other multivariate statistical procedures such as Fisher's discriminant ' function analysis and the multivariate logistic method, were performed to investigate and identify the census tracts with high risk ofEBLas well as providing more information on the roost important sodo-economic variables associated with the EBL The results ofthese analyses indicate that three sodo-economic variables appear to make a significant contribution to the variation in the proportion ofEBLin children. These variables are (Xj) the proportion ofhouse units built before 1940, (Xj) theproportion of population below poverty level, and (X$) the proportion ofblack children. It is suggested that these three variables be given high priority whenever a screening program is planned for elevated blood lead level in children. MEASUREMENT ERROR IN LOGISTIC REGRESSION BY DISCRIMINANT ANALYSIS WITH APPLICATIONS TO THE EPIDEMIOLOGY OF CORONARY HEART DISEASE Order No. 8014607 Ba w o l , Ric h ar d Da v id , Ph .D. University ofCalifornia, Berkeley, 1979. 131pp. The effectofmeasurement error on logistic regression has been examined, primarily under normality assumptions when coeffidents can be estimated by discriminant analysis and in the context ofa prospective epidemiological study ofcoronary heart disease (CHD). The statistical problem is that of relating a dicotomous response to continuous variables ; when the continuous variables are measured with error. This occurs when i the presence or absence ofCHD during a follow-up period is related to baseline risk variables which are subject to intra-individual variation and ; errors in measurement procedures. Two general questions are studied in thj dissertation: (1) How do observable relationships compare to true relationships? (2) How can true relationships be estimated from observations measured with error? Literature on measurement error in regression analysis is reviewed to : provide a source ofideas and a frame ofreference. Ofparticular interest art the attenuation effects for regression coeffidents, and maximum.likelihood estimation for functional and structural relationships with measurement error. The coeffident ofreliability (variance oftrue/variance ofobservable) can be estimated from variance components or by the correlation between repeated measurements. The equivalence between these two types of estimators is established for each ofthree error models. When these estimators are applied to a diet reliability sample, differences between the error models appear to be inconsequential Attenuation effects ofmeasurement error on logistic regression are. formulatedby relating discriminant coeffidents for observables to those fot true values. The primary effect is to attenuate a discriminant coeffident by t ffictor equal to the coeffident ofreliability. A secondary effect which operates through correlated covariates is studied for a bivariate discriminant. Data from the Western Collaborative Group Study (WCGS) indicate the impact ofmeasurement error in a prospective epidemiological study ofCHD. Two alternative models are also considered. The "fiat slope syndrome" I of negative correlation between measurement errors and true values may \ further attenuate a discriminant coeffident A ln-linear dicotomous respons model with measurement error again involves an attenuation factor equal t the coeffident of reliability. Estimation and hypothesis testing for true univariate logistic regression by discriminant analysis of observations with error are ittvestigated by deriving maximum likelihood estimators and their standard errors. The general idea is to adjust the discriminant coeffident to remove die effect of measurement error by estimating intra-individualvariation from replicate observations. Loss ofpower due to measurement error and compensating Sample Size requirements are determined from normal theory. WCGS data are analyzed again, this time to estimate discriminant coeffidents from replicate observations and tojudge the effect ofmeasurement error on the significance and standard error ofestimates. Finally, the results ofthis dissertation are summarized, with particular regard to epidemiological implications and directions for further study. Investigators are urged to report measurement reliability, to improve reliability by using averages ofreplicates, and to consider eventually adjusting risk assodations for.measurement error. Further study is needed to develop multivariateadjustment methods for general logistic regression with measurement error. N33786 Dissertation Abstracts International VoL 41 No. 01 July 1980 TEH 0531908