The analyses were structured according to a two-by-four design
to compare the two diets (control and DASH) during the four periods
(the run-in period and three intervention periods). The baseline
blood pressure used for the analyses was the average of the
measurements taken during the screening and run-in periods, and
the blood pressure used for the end of each intervention period was
the average of the last five measurements. A unified generalizedestimating-
equation
18
model with an exchangeable covariance matrix
was used for all primary analyses. Blood pressure was the outcome.
The base-line blood pressure, the clinical center, and the
cohort were represented in the model as fixed effects, whereas the
intervention periods were included as random effects to allow for
within-person correlation among blood-pressure measurements.
The model included indicators of the cohort, the clinical center, and
the carryover effect from the previous intervention. Results were
similar with and without carryover in the model. Indicators for the
subgroups specified in the study protocol (hypertensive status,
race, and sex) and for the relevant interactions with the effects of
the diet assignments and sodium levels were included in the subgroup
analyses.