and the patients’ residence. Data from June 1997 to April 2001 were retrieved to cover the entire follow-up period. Statistical Analysis The dataset was divided into a baseline and a follow-up dataset. Linear regression was used to investigate the association between PG 490 temperature and baseline BP, adjusted for gender, age, body mass index, urine protein, smoking behavior 12526815 and drinking behavior. Multilevel modeling was implemented to analyze the association between the temperature and the repeated measurement data of the follow-up dataset. Random effects for the duration of medication and intercept were included in the model. The covariance structure was defined as unstructured, and the estimation method was maximum likelihood. Besides the covariates mentioned above, the baseline BP and the medication duration was also included. Interactions of temperature and other covariates were examined as product terms. To estimate the contribution of temperature to the average change of BP in aggregated 10463589 weeks, the association between the ambient temperature and the average weekly BP was examined by linear regression. The average BP of the subjects who were recruited in the same week was calculated, so was the mean of temperature. Multiple correlation coefficients were used to indicate the proportion of variance that could be explained by ambient temperature; the R2 of medication duration was also investigated. As BP dropped quickly in the first few weeks of benazepril medication and more slowly in the later period, association analyses were conducted only with BP records from the 4th week to the 156th week. To exclude the possible bias from the intake of dihydrochlorothiazide, the analyses were repeated after the 57 patients involved were dropped. For males, this fluctuation was 8.4, 7.0 and 4.4 mmHg, while for females it was 7.2, 5.5 and 4.4 mmHg. The temperature regression coefficients are smaller in the higher BMI group. However, this was not replicated in another confirmation analysis. The drinkers’ DBP fluctuation was estimated to be higher than non-drinkers’, and the difference remained in each of the three years. After these interactions were adjusted, the regression coefficients of daily average temperature were 20.325 and 20.252 respectively, which meant a 9.4/7.3 mmHg increase in BP as the ambient temperature decreased by 29.0uC in a year. Contribution of Temperature to the Weekly Average Continuous variables were described as mean 6 standard deviation. Drinking behavior was recorded as not drinking, drinking,100 g wine per day, or drinking $100 g wine per day. SBP indicates systolic blood pressure; DBP, diastolic blood pressure; BMI, body mass index. temperature range and SBP or DBP. Hence, only daily average temperature was chosen to represent the effect of ambient temperature in the follow-up analyses. Regulators of Blood-pressure Response to Temperature Change The interactions between daily temperature and other factors were also investigated. In the SBP model, the temperature and medication duration interaction, as well as the interaction of temperature and age, was statistically significant. The regression coefficient of the medication-temperature interaction was 0.0016, so under the benazepril therapy, the reaction of SBP to the change of ambient temperature was estimated to decrease by 2.4 mmHg each year. To confirm these interactions, medication duration and age were transformed into ordinal categories and the follow-up dataset wa