TY - JOUR
T1 - How do statistical properties influence findings of tracking (maintenance) in epidemiologic studies? An example of research in tracking of obesity
AU - Wang, Youfa
AU - Wang, Xiaofei
PY - 2003
Y1 - 2003
N2 - There is great interest in studying the tracking (maintenance) of health conditions and risk factors over the life span. Tracking is often defined as the maintenance of a distribution position (e.g., quintile or percentile) in a study population over time. This study investigated how statistical properties might influence research findings of tracking with a special attention on the tracking of extreme ranking. Our results show that when repeated measures over time were positively correlated, the probability of tracking in extreme rankings was greater than other rankings and this was closely influenced by the overall correlation (r) and by the categorization. For example, when r = 0.4, 38% remained in the bottom and upper quintile (Q1, Q5) respectively, while only 22% remained in the middle quintile (Q3); when r = 0.8, the figure became 65% vs. 32%. When r = 0.4 and 0.8, 19 and 50% remained in the upper 95th percentile (or under the 5th percentile), respectively. Our real data show that children in the upper body mass index (=weight(kg)/height(M)2) quintile were more likely to maintain their ranking (54%) than others (about 30%), but not significantly higher than the expected (47%, p > 0.05). In conclusion, the overall correlation should be considered when studying tracking. Our proposed methods and predicted probabilities of tracking can help test whether one's observed tracking patterns are different from the statistically predicted ones.
AB - There is great interest in studying the tracking (maintenance) of health conditions and risk factors over the life span. Tracking is often defined as the maintenance of a distribution position (e.g., quintile or percentile) in a study population over time. This study investigated how statistical properties might influence research findings of tracking with a special attention on the tracking of extreme ranking. Our results show that when repeated measures over time were positively correlated, the probability of tracking in extreme rankings was greater than other rankings and this was closely influenced by the overall correlation (r) and by the categorization. For example, when r = 0.4, 38% remained in the bottom and upper quintile (Q1, Q5) respectively, while only 22% remained in the middle quintile (Q3); when r = 0.8, the figure became 65% vs. 32%. When r = 0.4 and 0.8, 19 and 50% remained in the upper 95th percentile (or under the 5th percentile), respectively. Our real data show that children in the upper body mass index (=weight(kg)/height(M)2) quintile were more likely to maintain their ranking (54%) than others (about 30%), but not significantly higher than the expected (47%, p > 0.05). In conclusion, the overall correlation should be considered when studying tracking. Our proposed methods and predicted probabilities of tracking can help test whether one's observed tracking patterns are different from the statistically predicted ones.
KW - Epidemiologic study
KW - Maintenance
KW - Obesity
KW - Statistical method
KW - Tracking
UR - http://www.scopus.com/inward/record.url?scp=0242269334&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=0242269334&partnerID=8YFLogxK
U2 - 10.1023/A:1026196310041
DO - 10.1023/A:1026196310041
M3 - Article
C2 - 14620937
AN - SCOPUS:0242269334
SN - 0393-2990
VL - 18
SP - 1037
EP - 1045
JO - European Journal of Epidemiology
JF - European Journal of Epidemiology
IS - 11
ER -