Date of Award

1995

Degree Type

Thesis

Degree Name

Master of Science in Statistics

Department

Computer Science and Statistics

First Advisor

R. Choundary Hanumara

Abstract

Various statistical methods are utilized in the determination of reliable cholesterol measurements via the Reflotron desk-top analyzer. The analysis presented in this research concerns data from the Worksites, Occupational Nurses, and Cholesterol Change Program, a study consisting of volunteers at various worksites in Southern Massachusetts and Rhode Island.

Accuracy and precision tests are utilized to determine reliability of cholesterol measurements. The initial lot testing of reagent strips gives estimates of bias resulting from Reflotron cholesterol measurement. Regression analysis techniques are employed to adjust for this bias. The accuracy testing failed to fall within the guidelines set by the National Cholesterol Education Program (bias<=5% ). A negative bias exists between the Reflotron and reference laboratory values when measuring total blood cholesterol. However, lots ofreagent strips were chosen to proceed to the next round of testing since resulting cholesterol measurements can be adjusted for bias.

The second phase of testing involves analysis of experimental precision. The design of the experiment entails all apparent sources of variation inherent to the cholesterol measurement process by Reflotron determination. The precision testing was successful in meeting the NCEP guidelines (CV<=5%), and information concerning the various random components was collected. Estimates of variance components are utilized in the construction of tolerance intervals to be used for quality control procedures in the field experiment consisting of worksite-wide cholesterol screening. A standardized method of reporting accuracy termed the maximum percent error is introduced and employed to compare various trials of the experiment, along with the calculation of coefficients of variation. Approximate confidence intervals are constructed for each variance component and for the overall variability in the experimental model. This gives a range of variability associated with the process, and an idea of which components contribute the greatest source of variation to the experimental model.

The final calculations are for the adjustment of cholesterol measurements in the field experiment. The purpose is to utilize the quality control levels to determine the estimated true cholesterol values of participants in the study. Prediction intervals are then constructed to determine a range of values within which an individual's true cholesterol level falls. Thus, if misclassification of individual true cholesterol measurements occurs, the prediction interval gives information concerning the risk category to which an individual is likely to belong. Remedial measures are recommended for future testing.

Share

COinS