Abstract
We evaluated quasi-healthy cohorts (model cohorts), derived from clinical data, to determine how well they simulated control cohorts. Control cohorts comprised individuals extracted from a public checkup database in Japan, under the condition that their values for 3 basic laboratory tests fall within specific reference ranges (3Ts condition). Model cohorts comprised outpatients, extracted from a clinical database at a hospital, under the 3Ts condition or under the condition that their values for 4 laboratory tests fall within specific reference ranges (4Ts condition). Because even a patient with a serious illness, such as cancer, may present with normal values on basic laboratory tests, one additional condition was added: the duration (1 or 3 months; 1M or 3M) during which patients were not hospitalized after their first laboratory test. For evaluations, cohorts were specified by age and sex. The 4Ts + 3M condition was the most effective condition, under which model cohorts were used to successfully simulate age-dependent changes and sex differences in laboratory test values for control cohorts. Therefore, by properly setting the conditions for extracting quasi-healthy individuals, we can derive cohorts from clinical data to simulate various types of cohorts. Although some issues with the proposed method remain to be solved, this approach presents new possibilities for using clinical data for cohort studies.