Applied Epidemiologic Analysis. Patricia Cohen, Ph.D. Henian Chen, M.D., Ph. D. Teaching Assistants Julie Kranick Sylvia Taylor Chelsea Morroni Judith Weissman. Lecture 12 Multiple time point assessments in epidemiology: purposes, questions, and analyses. GOALS:
Patricia Cohen, Ph.D.
Henian Chen, M.D., Ph. D.
Julie Kranick Sylvia Taylor
Chelsea Morroni Judith Weissman
To understand some of the reasons for longitudinal analysis in epidemiology
To outline the circumstances in which a survival analytic method, a Poisson regression method, a repeated measure ANOVA method, a GEE method, or a multilevel method may be selected
To understand the connection between multilevel methods for clustered data and for longitudinal data in terms of the kinds of questions that may be answered.
Data: Time to some event such as disease onset or diagnosis
Analytic choices depend on
Data include person-time for both cases and referent participants (time to censoring)
A. Designs in which measurement of exposures, including timing, are retrospective
B. Nested case-control studies with prospective measurement of exposure and confounders and matched timing for cases and controls
C. Cohort studies in which data on exposures are gathered prospectively and disease outcomes are assessed in subsequent follow-up
One method of attempting to control at least some of these problems:
The biologically plausible period between the exposure and disease needs to be taken into account.
When studying the relationship between variables that may change over fairly short time periods, it is useful to include more than two measurement points.
1. To inform about average changes in variables over time/ trials/ age
2. To improve inferences about the direction and magnitude of influences of one variable or set of variables on another by establishing sequence
3. To inform about individual or group differences in change over time and variables that relate to these differences
Dal S, Labarthe DR, Grunbaum JA, Harrist RB, Mueller WH. 2002. American Journal of Epidemiology, 156, 720-729.
Goals of study:
To examine growth patterns of obesity indices of 678 children studied at 4 month intervals for 4 years, beginning ages 8,11, and 14.
To compare these patterns for males and females and for Black and White Children.
Lawson CC, LeMasters GK, Levin LS, Liu JH. 2002. Pregnancy hormone metabolite patterns, pregnancy symptoms, and coffee consumption. American Journal of Epidemiology, 156, 428-437.
Goals of study
To determine relationships between:
1. Pregnancy hormone metabolites (PHM) & concurrent coffee consumption (CC)
2. PHM and concurrent pregnancy symptoms (PS)
3. Pregnancy symptoms and coffee consumption
4. Pre-pregnancy coffee consumption and PHM
Analyses to answer study questions regarding relationships between:
Pregnancy hormone metabolites (PHM) & concurrent coffee consumption (CC)
Other study findings determining relationships between:
PHM and concurrent pregnancy symptoms (PS) Weekly changes in these variables were strongly related in multilevel models.
Nausea and appetite loss were not related to current coffee consumption, but more vomiting was associated with less coffee consumption.
Pre-pregnancy coffee consumption was related to lower mean level of one of the pregnancy hormones (a not anticipated finding).
Modeling changes in CD4+ T- Lymphocyte counts after the start of highly active antiretroviral therapy and the relation with risk of opportunistic infections
Binquet C, Chene G, Jacqmin-Gadda H, Journot V, Saves M, Lacoste D, Dabis, F et al., American Journal of Epidemiology,
Question: Is there a minimal duration of CD4+ cell count increase before its impact on opportunistic infection?
Design: 553 HIV patients treated with at least one protease inhibitor and studied at least twice over the next months with regard to CD4+ counts and opportunistic infection.
If a time series is based on aggregated data we may think of it as comparable to ecological analyses with data only at the clustered level.
If data are available both at the individual and aggregate level we may still focus on the effects over time, using multilevel analyses.
Note that in contrast to the previous graph, these rates are per 1000, and differences are smaller, numbers being partly due to decline in number of children.
Conclusion: Age effects were larger among older children in the earliest vaccinations, but were relatively equal among age groups thereafter.
Pertussis is a major cause of childhood morbidity and mortality globally. Even among those who did not receive booster shots there was a decline in mortality if there were subsequent infections.
There also appeared to be a herd immunity effect of vaccination, although the level receiving even one dose was never much over 40%.
This study has data only at the aggregate level, and is thus rather like ecological studies, without the ability to examine individual differences (e.g. in number of vaccination doses effects or in the duration over which immunity effects associated with specific numbers is effective).