Faculty Bibliography

Two hundred one cases with permanent bilateral high-frequency hearing loss and 208 controls with normal hearing were identified through the New York City hearing screening program. Aircraft noise exposure was estimated for the residences of cases and controls from Noise Exposure Forecast contour maps of the New York City airports. The noise exposure estimates had a correlation on 0.89 with noise level measurement made in the area. Birth certificates and questionnaires were used to provide additional information on other sources of noise exposure and potential confounders. The results showed a positive, but not statistically significant association between aircraft noise exposure and the risk of high-frequency hearing loss.

This paper proposes the use of group sequential methods to calculate sample sizes for individually matched case-control study designs. A table is presented in which the average sample size required for a group sequential (i.e., multistage) matched pair design is compared to that of the conventional matched pair fixed sample size plan for the usual constant relative risk situation. The table shows that group sequential designs are in general more efficient than fixed sample size plans. Computer simulations showed that group sequential methods yield the appropriate type I and type II error rates not only for matching on a one-to-one basis, but also more generally with multiple matched controls per case. Further simulation studies indicated that there may be only a small loss of power when the matching variable(s) is associated with the probability of exposure but not with the disease. This is shown for both the multistage and fixed sample tests.

The percent of students reading below grade level from 1972 to 1976 was regressed on racial, socioeconomic, educational, and noise level variables for all elementary schools in Brooklyn and Queens, New York. Schools were assigned noise exposure scores based on Noise Exposure Forecast contours for New York City airports. The correlations between these noise scores and a variety of noise level metrics ranged from 0.74 to 0.97. The regression coefficients adjusted for confounding factors, indicated that an additional 3.6% of the students in the noisiest schools read at least 1 yr below grade level with 95% confidence limits from 1.5 to 5.8%. The dose-response relationship indicated that the percent reading below grade level increased as noise level increased.

This paper proposes the use of group sequential methods of calculate sample sizes for cohort and case-control study designs. The methods are based upon the theory of repeated significance tests, one of several sequential methods currently available. Group sequential methods permit repeated significance testing of relative risks obtained from periodically accumulated data while maintaining the required overall level of significance. Tables are presented for cohort and case-control studies in which the average sample size required for a group sequential design is compared to that of the conventionl fixed sample size plan for the usual constant relative risk situation. The tables show that group sequential designs are in general more efficient than fixed sample size plans for cohort and case-control studies. Computer simulations showed that group sequential methods can be employed when stratified data analyses are to be used and when the sample sizes of the two study groups are unequal.

Breast cancer incidence data were analyzed from three populations of women exposed to ionizing radiation: survivors of the Hiroshima and Nagasaki atomic bombs, patients in Massachusetts tuberculosis sanitoria who were exposed to multiple chest fluoroscopies, and patients treated by X-rays for acute postpartum mastitis in Rochester, New York. Parallel analyses by radiation dose, age at exposure, and time after exposure suggested that risk of radiation-induced cancer increased approximately linearly with increasing dose and was heavily dependent on age at exposure; however, the risk was otherwise remarkably similar among the three population, at least for age 10-40 years at exposure, and followed the same temporal pattern of occurrence as did breast cancer incidence in nonexposed women of similar ages