Abstract

We used the health screening data of some rural and urban residents to examine the cross-sectional association between leukocyte count and hypertension. The 206 male and 203 female rural residents were selected by multi-stage cluster sampling method in Kyungsan-Kun area of Kyungbuk province in 1985 and 600 urban residents were selected by the same sampling method as the rural residents in Daegu city of the same province in 1986 compatible with age-sex distribution of Daegu city of 1985 census, but of whom 384 actually responded. The rest of 600 were replaced by age and sex with those who were members of the medical insurance plan visiting the health management department of the university hospital to get the biannual preventive medical checkups. Excluded in the analysis were those having hypertensive history, diseases and extreme outlying values of the screening tests, leaving 373 rural and 571 urban residents. Leukocyte count was measured with ELT-8 Laser shadow method and the unit cells/mm3. Blood pressures were determined with an aneroid sphygmomanometer with pre-standardized method and hypertensives were defined as those showing systolic blood pressure more than 140 mmHg and / or diastolic blood pressure more than 90 mmHg. Total residents pooled (N=944) showed a significant difference between hypertensives and normotensives (6965.93+/-1997.01 vs 6490.61+/-1941.32, P=0.00) and in rural residents was noted the similar significant difference (P=0.03). None of significant differences were noted in any stratum stratified by residency and sex. Compared to the lowest quintile of WBC, 2/5 quintile showed odds ratio 0.99 (95% CI 0.90-2.21), 4/5 quintile 1.76 (95% CI 1.14-2.72), and highest quintile 1.80 (1.15-2.82) in the total residents. Likelihood ratio test for linear trend for in indicated a significant trend (x2 trend=5.53, df=1, P<0.05). There were no other significant odds ratios compared to the lowest quintile of WBC in strata stratified by residency and sex. The odds ratios in total residents which had showed significant odds ratios became nonsignificant and of reduced magnitude after controlling age, frequency of smoking and drinking with multiple logistic regression. In each stratum, it changed magnitudes of odds ratios slightly and unstably. None of the trend tests showed any significant trend. These results suggest that the Friedman et al's finding of association between leukocyte count and hypertension may be due to an statistical type I error result in from the data dredging in an exploratory study, in which more than 800 variables were screened as possible predictors of hypertension.

• Keywords: Leukocyte counthypertensionconfounding