Randomization

Subjects are randomed at the level of the individual. This avoids some of the pt demographic imbalances that were seen in WATCH by blocking patients by age and gender within physician. Patients will be categorized by age (20-34, 35-54, 55-70) and gender into strata.  Randomization will occur by order of entry into the age-gender-physician stratum, with the first patient randomly assigned to one of the conditions, and the subsequent patient in the stratum assigned the alternative condition. This pattern of condition assignment will continue as patients accrue. Assignment will be made using the random number generator in Lotus Notes.

Statistical Power

Measures used in this study and certain other aspects of design are similar to those employed in the original WATCH study.  Hence, variance component estimates from the WATCH were used in designing this study.  We anticipate that the patient population enrolled in this study will be similar to that in the WATCH study.  Assuming a 75% rate of complete data collection, the study will have 98% power to detect a difference between conditions in the change over time in LDL-C of 5 mg/dl based on a two-sided test. To detect an effect as large as that observed between conditions I and III in WATCH we will have 99% power.  For the primary dietary endpoint,  this study is designed to detect a difference of 1% energy from SFA between subjects in the two conditions with 91% power, also based on a two-sided test.  This power estimate is based on a single 24HR administered at the one-year time point.  Because of the large intra-person variability associated with a single 24HR (65), analyzing a change score based on the difference between the one-year and baseline values would result in lower power to detect an effect than would result from using the single cross-sectional measurement.  Because we are randomizing at the level of the individual and blocking by age, gender, and physician (see D.11.1.2.), it is very unlikely that an imbalance in baseline dietary intake by randomization condition would countervail the effect of increased variance resulting from the use of a baseline 24HR.

The overall effect size on which we base our sample size and statistical power is approximately that which we observed in condition III– condition I of WATCH. It is approximately 40-50% of the effect observed in attendees of >3 GNI sessions in that study. (2) In the present study we expect that as few as ¼ of the pts in cond II will receive the entire GNI (see attendance estimates for other more recent studies in C.2.4. and C.2.5.). If the effect size is as large than that seen in WATCH, i.e., 15.4 mg/dl LDL-C,  this will result in an observed reduction of 3.9 mg/dl averaged over the entire group. Though there are few empirical data to guide us, we estimate that the combination of pt tracking with programmed lipid measurements, pt-centered telephone counseling by the DLMC, and other interventional materials (videotapes, newsletters) will produce an effect as large as that of condition III in WATCH (i.e., 6.9 mg/dl.) The net effect size is a reduction of 9.8 mg/dl.

The study has sufficient power to also allow for adjustments due to multiple comparisons in post-hoc analyses of study data. In particular, we are especially interested in analyses looking at the effect of gender and level of education on our primary outcomes.

Table D.11.4.  Statistical Power for various sample sizes comparing difference scores and simple differences post-test scores by condition for SFA intake (% energy) and LDL-C (also see Appendix Q)

                *Model of choice for this analysis