Russian Escorts Delhi | 9711199171 | all area service available
Ecological study design multiple group study and statistical analysis
1. Presented by : Sirjana Tiwari
MPH(PHSM)
Pokhara University
2017
Study Designs in Ecologic Studies: Multiple Group Study and
Statistical Analysis
1
2. Correlation studies
Examine association between exposure and disease measures on
the population level
Unit of measurement = population groups
Eg: Behavioral Risk Factor surveillance system
2
3. Variables in an ecologic analysis
aggregate measures
environmental measures
global measures
3
4. variables
Aggregate measures are summaries (e.g. means or proportions) of
observations derived from individuals in each group
e.g. the proportion of smokers
the proportion of median family income.
Environmental measures are physical characteristics of the place in
which members of each group live or work
e.g. air-pollution level or hours of sunlight
Global measures are attributes of groups or places for which there is no
distinct analogue at the individual level.
e.g. population density
level of social disorganization
the existence of a specific law
4
5. STUDY DESIGNS
1. The groups of an ecologic study may be identified
a. by place (multiple-group design),
b. by time (time-trend design)
c. by combination of place and time (mixed design).
2. The method of exposure measurement
a. an ecologic design is called exploratory if the primary exposure of
potential interest is not measured,
b. analytic if the primary exposure variable is measured and included
in the analysis. In practice, this dimension is a continuum, since
most ecologic studies are not conducted to test a single hypothesis
5
6. TYPE OF ECOLOGIC STUDY DESIGNS
6
Based upon Groups of an
Ecologic study
Based upon the Method of exposure
measurement in Ecological study
a. by place (multiple-group
design)
Exploratory Analytical/Etiological
b. by time (time-trend design) Exploratory Analytical/Etiological
c. by combination of place and
time (mixed design)
Exploratory Analytical/Etiological
7. Multiple-Group Study(by place)
a. Exploratory study
We compare the rate of disease among many regions during the same
period
For example, the age-adjusted cancer mortality rates in the U.S. by county for
the period 1950-69 for oral cancers, found a striking difference in geographic
patterns by-sex:
Among men, the mortality rates were greatest in the urban Northeast,
among women, the rates were greatest in the South east.
7
8. Example: An Multiple group studies of obesity
and income inequality
8
The relation between male obesity and
income inequality in 21 rich countries.
The relation between female obesity and income
inequality in 21 rich countries.
9. example
It also involve the comparison of rates between migrants and their offspring
and residents of their countries of emi gration and immigration.
For example, if US immigrants from Japan have rates of a disease similar to US
whites but much lower than Japanese residents, the difference may be due to
environmental or behavioral risk factors operating during adulthood.
However, the interpretation of results from these studies is often limited by
differences between countries in the classification and detection of disease or
cause of death
9
10. Multiple-Group Study(by place)
a. Exploratory study
mapping
A simple comparison of rates across regions is done in Mapping.
First, regions with smaller numbers of observed cases show greater
variability in the estimated rate; thus the most extreme rates tend to be
observed for those regions with the fewest cases.
Second, nearby regions tend to have more similar rates than do distant
regions (i.e. autocorrelation) because unmeasured risk factors tend to cluster in
space.
Statistical methods for dealing with both problems have been developed by
fitting the data to an autoregressive spatial model and using empirical Bayes
techniques to estimate the smoothed rate for each region
10
11. Multiple-Group Study(by place)
b. Analytical study
In this type of study, we assess the ecologic association between the average
exposure level or prevalence and the rate of disease among many groups.
The unit of analysis is a geopolitical region.
For example, Hatch & Susser (29) examined the association between
background gamma radiation and the incidence of child hood cancers
between 1975 and 1985 in the region surrounding a nuclear power plant.
The authors found positive associations between radiation level and the
incidence of leukemia (an expected finding) as well as solid tumors (an
unexpected finding).
11
12. Data analysis in this type of multiple-group study usually involves fitting
the data to a mathematical model.ie.
linear relative rate
exponential relative rate
12
13. Statistical analysis
1. 1. Apply regression analysis to the group-level data, modeling disease rate as a
function of exposure prevalence. Several forms of regression can be used for
this purpose, as described below.
2. Use the fitted regression model to predict the disease rate for a population in
which everyone is exposed. Call expose rate R 1. Similarly, predict the rate for
a population in which nobody is exposed, and call unexposed rate R 0.
3. Estimate relative risk as R 1/R 0 and attributable risk as R 1 − R 0.
13
14. Look at overall association between exposure and outcome
measure for ecological units being compared
Scatter plot
Pearson correlation coefficient (r)-overall measure of
correlation
Measures degree of linear relationship between the
continuous variable
-1<r<+1
14
15. Regression analysis
Figure : Power of the repeated measurements multilevel regression analysis as a function of the
magnitude of the regression coefficient, n=60 persons with 40 repeated measurements15
16. 16
In multiple-group ecologic studies, we regress the group-specific
disease rates (Y) on the group-specific exposure prevalence's (X).
(Note that, throughout this chapter, uppercase letters are used to
represent ecologic variables and their estimated regression
coefficients; lowercase letters are used to represent individual-level
variables and their estimated regression coefficients
17. Linear model
17
The most common model form for analyzing ecologic data is the
linear model. Ordinary least-squares methods can be used to
produce the following prediction equation: Ŷ = B0 + B1X, where B0
and B1 are the estimated intercept and slope. An estimate of the
biologic effect of the exposure (at the individual level) can be
derived from the regression results
The predicted disease rate (Ŷx = 1) in a group that is entirely
exposed is B0+B1(1) = B0+B1, and the predicted rate (Ŷx = 0) in a
group that is entirely unexposed is B0+B1(0) = B0. Therefore, the
estimated rate difference is B0+B1 - B0 = B1, and the estimated rate
ratio is (B0 + B1)/B0 = 1 + B1/B0.
18. 18
Alternatively, fitting a log-linear (exponential) model to the data
yields the following prediction equation: ln(Ŷ) = B0 + B1X or Ŷ =
exp(B0+B1X). Applying the same method as used for linear models,
the estimated rate ratio is Ŷx = 1/Ŷ x = 0 = exp (B1).
19. Ordinary least-squares linear regression
19
Durkheim's (1951) examination of religion and suicide in four
groups of Prussian provinces between 1883 and 1890.
The groups were formed by ranking 13 provinces according to the
proportion (X) of the population that was Protestant.
Using ordinary least-squares linear regression, we estimate the
suicide rate (Ŷ, events per 105 person-years) in each group to be
3.66 + 24.0(X).
Therefore, the estimated rate ratio, comparing Protestants with
other religions, is 1 + (24.0/3.66) = 7.6. Note in Figure that the fit of
the linear model appears to be excellent (R2 = 0.97)
20. 20
Suicide rate (Y, events per 105 person-years) by proportion Protestant (X)
for four groups of Prussian provinces, 1883–1890. The four observed
points (X, Y) the fitted line is based on unweighted least-squares
regression.
21. Empirical Bayes method
21
It is the one of the most important application of conditional
probability which gives the information about the occurrence of one
event to predict the probability of another events.
Ie.
Empirical Bayes model is use to calculate sensitivity, specificity,
predictive value positive and predictive value negative
22. Spatial autocorrelation
22
spatial autocorrelation test is designed for of long-term, large-scale
ecological datasets
This test foster significant future advances in understanding
population regulation, metapopulation dynamics and other areas of
population ecology.
23. 23
Fig: Synchrony in relative mean density of mountain hares (Lepus timidus) among 11 provinces in
Finland measured over 39 years (1946–1984) plotted against the distance between the
geographical centers of the provinces. The total number of comparisons is 55, corresponding to
the number of pairwise comparisons possible between 11 sets of data. Note that synchrony
declines with increasing distance between provinces (r 5 –0.57)
24. Mathematical model
24
Mathematical models developed to predict and control behavior in applied
settings, and they have guided research in other areas of psychology.
A good mathematical model can provide a common framework for
understanding what might otherwise appear to be diverse and unrelated
behavioral phenomena.
It is important for those who develop mathematical models of behavior to find
ways (such as verbal analogies, pictorial representations, or concrete examples)
to communicate the key premises of their models to non specialists.
25. 25
Fig: the percentage of JEAB articles that presented at least one
equation to describe the relation between an independent variable
and a dependent variable
26. Reference
http://hinarilogin.research4life.org/uniquesigwww.oxfordscholarship.com/unique
sig0/view/10.1093/acprof:oso/9780195150780.001.0001/acprof-
9780195150780-chapter-12
Morgenstern, H. (1995). "Ecologic studies in epidemiology: concepts, principles,
and methods." Annu Rev Public Health 16: 61-81.
Bogers RP, Bolte JFB, Houtveen JH, et al. Design of an ecological momentary
assessment study of exposure to radiofrequency electromagnetic fields and
non-specific physical symptoms. BMJ Open 2013;3:e002933.
doi:10.1136/bmjopen-2013-002933
Rothman KJ. 1 986. Modem Epidemiology, pp. 41-49, 82-94. Boston: Little,
Brown & Co.
J Epidemiol Community Health 2005;59:670–674. doi:
10.1136/jech.2004.028795
Umesh raj aryal and Yogesh man shrestha. Biostatistics for medical science.2nd
edition.Makalu Publication.P(95-6).
Mazur, J. E. (2006). Mathematical Models and the Experimental Analysis of
26
Editor's Notes
J Epidemiol Community Health. 2005 August; 59(8): 670–674
Statistical methods for dealing with both problems have been developed by fitting the data to an autoregressive spatial model and using empirical Bayes techniques to estimate the smoothed rate for each region.
The degree of spatial autocorrelation or clustering can be measured to reflect environmental effects on the rate of disease.
The empirical Bayes approach can also be applied to data from analytic multiple-group studies (described below) by including covariates in the model.
Hatch M (Division of Epidemiology, School of Public Health, Columbia University, New York, 10032, USA) and Susser M. Background gamma radiation and childhood cancers within ten miles of a US nuclear plant. International Journal of Epidemiology 1990; 19: 546–552.
In light of some recent reports concerning childhood leukaemia near nuclear installations, we examined rates of cancer in children in relation to background gamma ray exposure. Data from a national monitoring programme around nuclear facilities were used to map the distribution of background gamma radiation for 69 small geographical subunits (average population 2300) within ten miles of one US nuclear plant. An association was found for incidence of childhood cancers as a whole (odds ratio (OR) = 2.4; 95% confidence limits (CL) 1.2, 4.6). For leukaemias specifically, the odds ratio was also elevated but confidence limits were very wide (OR = 2.4; 95% CL 0.5, 12.9). Analyses adjusting for sociodemographic characteristics of study tracts (population density and income) gave similar results; data on other risk factors were unavailable.
Conventional risk models would not predict a detectable increase in childhood cancer from background gamma radiation, particularly not an increase of this magnitude. The large effect for solid tumours as well as leukaemias is also somewhat counter to expectation. Since a priori the association we observed was unlikely, it is important to know if similar trends in childhood cancer with background radiation are seen in other areas before rejecting chance or bias as an explanation for the result.
For example, Prentice & Sheppard proposed a linear relative rate model using iteratively reweighted least-squares procedures to estimate the model parameters.
Prentice & Thomas also considered an exponential relative rate model, which, they argue, may be more parsimonious than the linear-form model for specifying covariates. These methods can be applied to data aggregated by place and/or time .
Figure describe Ecological Momentary Assessment (EMA) study to determine whether non-specific physical symptoms in persons who self-report to be sensitive to radiofrequency electromagnetic fields (RF EMF) can be explained by objectively measured exposure to Radiofrequency ecological momentary assessment, or by psychological measures such as perceived exposure and mood.
Figure illustrates how the power of the statistical analysis differs according to the magnitude of the regression coefficient. It can be seen that at a power of 80%, the detectable regression coefficient is slightly over 1.5, which corresponds to an increase of 1.5 on the sum of 10 symptoms (range 0–40) at an increase of 1 in perceived exposure (range 0–4).
(e.g. population density of snowshoe hares at sites spread throughout the northern hemisphere)
. As a first step, it is often desirable to modify the raw data. Common procedures include log-transformation to reduce the correlation between the mean and the variance, and calculation of residuals from a linear regression to avoid patterns caused by large-scale global trends instead of the regional processes of direct interest9. An example of the latter problem would be if densities increased at all sites during the time the data are collected. Pairwise correlations based on the annual raw census data would then all be positive because of the long-term trend. Standardizing the data by using residuals from a linear regression eliminates this long-term trend and focuses the analysis on more regional processes.
Mathematical models developed in basic behavioral research have been used to predict and control behavior in applied settings, and they have guided research in other areas of psychology. A good mathematical model can provide a common framework for understanding what might otherwise appear to be diverse and unrelated behavioral phenomena. Because psychologists vary in their quantitative skills and in their tolerance for mathematical equations, it is important for those who develop mathematical models of behavior to find ways (such as verbal analogies, pictorial representations, or concrete examples) to communicate the key premises of their models to non specialists.
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1472627/
journal of the Experimental Analysis of Behavior