Observational Studies of Disease

1
Observational Studies of Disease

• Descriptive (incidence, prevalence)
• Analytic (associate characteristics of population
  with risk of disease)
• Population experience can be studied with group
  level or individual level data
• Studies using group level data are called
  ecological studies

2
Studies Using Group Data (1)

• Disadvantages of group data/ecological studies
• Most group data measures are made on individuals
  but not under investigators control
• Do not know if persons with given characteristic
  are those at higher risk of disease the
  ecological fallacy
• Confounding problem for all observational studies
  but greatest with group data lack of
  investigator control in measuring confounding
  variables

3
Studies Using Group Data (2)

• Advantages of group data/ecological studies
• Inexpensive secondary data already collected
  (vital statistics, disease registries, HMOs,
  etc)
• Rapid test of hypothesis
• Idea that ecological studies are
  hypothesis-generating doesnt reflect their usual
  purpose
• If hypothesized risk factor is associated with
  disease, it may well be seen in group level data
• Can overcome threshold problem exposure is so
  universal that effect is difficult to detect in
  one setting

4
Studies Using Group Data (3)

• Advantages of group data/ecological studies
• Some disease transmission dynamics can only be
  studied at group level (eg, herd immunity and
  infectious disease transmission
• Allows global measures of group characteristics
  (e.g., type of health care system)
• Allows tests of area-level interventions (eg,
  closing of a public hospital)

5
Studies Using Group Data (3)

• All ecological studies are not created equal
• Typical study relates a disease rate across
  different geographic areas with an aggregrate
  measure of a characteristic of the individuals in
  the areas (eg, average alcohol consumption) or
  with a global measure of a characteristic of the
  area (eg, climate). No attempt is made to
  control for confounding.
• Quality of secondary data varies widely
• Length of time secondary data has been collected
  varies (relevant to looking at an association in
  different time periods)

6
Strategies used to strengthen ecological studies
(1)

• There are several strategies that can strengthen
  inferences from ecological studies
• Multiple kinds of comparisons to strengthen
  inference of association eg, across geographic
  areas and over different time periods
• Example Valerie Berals study showing inverse
  association between average family size and
  ovarian cancer mortality using comparisons among
  different birth cohorts, different countries, and
  different social and ethnic groups (Lancet, 1978)

7
Strategies used to strengthen ecological studies
(2)

• Small Area analysis Used in health services
  research to investigate variation within small
  geographic areas
• Reduce confounding by comparing small areas from
  a larger area thought to be fairly homogeneous on
  potential confounders (SES, disease prevalence)
• Example Wennbergs study of variation in rates
  of surgical procedures in 6 areas of Vermont with
  similar disease prevalence (Medical Care, 1987)

8
Strategies used to strengthen ecological studies
(3)

• Mixed studies that collect data on individuals
  but use secondary group data for rare outcomes
• Doesnt avoid ecological fallacy but reduces
  confounding by key measures at individual level
• Using group data may make study feasible that
  would be otherwise prohibitively expensive
• Example Bindmans study of health care access
  and rates of preventable hospitalizations in
  California medical service areas (JAMA, 1995)

9
Studies Using Individual Data

• Unifying concept Characterize morbidity and
  mortality in a defined population during a
  defined period of time
• Defined population Study Base morbidity and
  mortality experience of a cohort of individuals
  over time
• Cohort studies, case-control studies, and
  cross-sectional studies are best understood
  within the framework of a common study base

10
Study Base

• Establish by
• Assembling an explicit cohort from
• Sample of larger population of interest
• Sample of persons with and without an exposure of
  interest
• Identifying cases of a specific disease and
  defining population that gave rise to the cases
• Defines the cohort within which the cases
  occurred
• Study designs differ in how they sample disease
  experience of the study base

11
Cohort Study

• Easiest design to understand because it
  explicitly defines the study base as a cohort
• Measures individual characteristics before
  disease occurrence fulfilling the temporal order
  required for cause and effect (but is not the
  only study design that can do this).
• Provides conceptual basis for understanding
  sampling strategies of case-control, case-cohort,
  and cross-sectional designs

12
Cohort Study
X
L
Subjects dying or lost to follow-up
X
X
X
D
L
L
D
X
X
D
D
D
Subjects followed until end of study
D
D
D
D
D
Begin
End
Time of Follow-up
X dead L lost D disease
13
Types of Cohort Studies

• Fixed (closed) versus dynamic (open) cohort
• Fixed All subjects identified at baseline in
  study
• Dynamic (open) Additional subjects taken during
  follow-up subjects enter at different times
• Fixed versus dynamic exposure measurement
• Groups of individuals with and without exposure
  of interest do not change during follow-up.
  Sometimes assembled and followed as two separate
  cohorts of exposed and unexposed.
• Dynamic exposure may vary during follow-up (eg,
  individuals stop or start a behavior or exposure
  is defined by an accumulation of years of
  exposure)

14
Measuring exposures that can vary over time in a
cohort study

• Simple cohort study with exposure status fixed at
  baseline calculates risk by number of cases of
  disease among exposed and unexposed subjects
• More complex cohort studies allow individuals to
  change from exposed to unexposed and therefore
  have to calculate disease occurrence on basis of
  both number of persons and length of time exposed
  (called person-time)
• Diseases with long incubation periods, such as
  cancer, require lag time to be taken into account
  in relating exposure to disease occurrence

15
Threats to Validity of a Cohort Study

• Ascertainment of disease outcome
• Length of follow-up
• Time between ascertainment of status (visits,
  follow-up interviews, medical record checks,
  etc.)
• Subtlety of disease onset (case definition)
• Secondary data sources for outcomes (eg,
  registries)
• Subjects lost during follow-up
• Key issue is whether losses are related to
  exposure and disease outcome
• If disease incidence is important outcome, losses
  may bias results even if not related to exposures

16
Threats to Validity of a Cohort Study (2)

• Long follow-up time biggest threat to validity of
  cohort studies
• Difficult to retain cohort and ascertain all
  outcomes
• Bias from loss to follow-up is analogous to bias
  in case-control study based on prevalent cases
• Large size and expense of cohort may require
  compromise in measurements
• Can be complicated to measure dynamic exposures
  and allow for incubation periods

17
Common paradigm of study design presents time the
study is undertaken as key to design but
neglects time measurements taken and concept of
study base
Past Present
Future
Cross-sectional Classify exposure and disease at
one time
Cohort Classify by exposure
Classify by disease
Case-control Classify by disease
Classify by exposure
18
Timing of Study and Measurements

• Prospective versus retrospective study
  terminology not always clear about when
  measurements were made
• Exposure and disease measurements may be
  concurrent, non-concurrent, or both with respect
  to the experience of the study base
• Study may be carried out concurrently, or
  non-currently, or both with respect to the
  experience of the study base

19
Timing of Study and Measurements

• Some authors designate case-control studies as
  retrospective studies inaccurate since cohort
  studies can also be retrospective
• Distinction between when measurement of exposure
  was made and study recorded it
• Key issue for causality is measuring exposure
  before disease--that is not design dependent

20
Timing of measurement of exposures and disease
with respect to timing of study
A
CHD
Diet Exercise
Study begins and makes measurements
B
CHD
Diet Exercise
Study begins and records measurements made
previously in medical record
C
CHD
Diet Exercise
Study begins, asks subjects to recall
information in the past
21
Timing of Study and Measurements (2)

• Schematic A is a prospective or concurrent cohort
  study
• Schematic B could be either retrospective cohort
  or case-control using medical records
• Schematic C is most often a case-control study
  but could be a retrospective cohort that uses
  recall to ascertain exposure
• Mixed designs are also possible with some
  measures concurrent with study and some measures
  non-concurrent

22
Cross-sectional Study

• In context of a cohort, a cross-sectional sample
  is equivalent to sampling those with prevalent
  disease and those without at one point in time in
  the follow-up
• A comparison of exposure in those with and
  without disease is equivalent to a case-control
  study using prevalent cases and concurrent
  controls

23
Cross-sectional Study in Context of a Cohort
Cross-sectional sample of cohort (population) at
one point in time Equivalent to sampling
prevalent cases and concurrent controls
Possible source of bias Missing potential
subjects
D
D
D
D
C
C
D
D
C
D
Subjects in Cross-sectional Study
C
D
C
D
C
D
C
D disease case C control (no disease)
24
Cross-sectional Study in a Dynamic Population
Cross-sectional sample of a dynamic population
differs from sampling in fixed cohort setting.
Persons enter as well as leave the population.
Disease sampling is still of prevalent cases.
Persons entering the population
D
Subjects in cross-sectional study
D
D
D
Persons leaving the population
D
D
D
D
D
D
D
D disease case
25
Cross-sectional Study (2)

• Using cross-sectional study to identify a cohort
  can provide a representative sample (prevalent
  cases of disease usually excluded)
• Repeated cross-sectional studies of a population
  can provide important information on trends a
  cohort might miss
• Although major weakness is problem of temporal
  order (cause and effect), timing of exposure and
  disease can sometimes be determined
  retrospectively

26
Case-Control Study Designs (1)

• Best conceptualized as occurring within a cohort
  study
• Variations on the case-control design come from
  how the cases and the controls are sampled
• From the point of view of design, case-control
  studies can be just as valid as cohort studies
• Threats to validity come from greater difficulty
  in defining and sampling the study base and in
  measuring exposure prior to disease

27
Case-Control Studies Case-based sampling

• In context of a cohort, case-based sampling
  identifies all cases of disease during the
  follow-up period and samples individuals disease
  disease free at the time of study (end of
  follow-up in the cohort context)
• Unbiased sample of cases but possibly biased
  sample of controls
• Requires rare disease assumption for odds ratio
  to estimate relative risk

28
Case-Control Study with Case-Based Sampling
Sampling within a Cohort Study Ascertaining all
cases and sampling controls from subjects
disease free at end of follow-up
Possible bias Potential controls not in study
at end of follow-up
D
D
D
C
D
C
C
D
Subjects in Case-Control Study
C
D
C
D
C
D
C
D
C
C
D
C
D disease case C control (no disease)
29
Case-Control Studies Case-Based Sampling (2)

• Case-based sampling is most common case-control
  design outside setting of explicit cohort
• The study base that gave rise to the cases is
  often not defined
• Examples of study bases
• Cases from population disease registry study
  base is the population covered by the registry
• Cases from HMO study base is plan members
• Hospital cases study base is persons who would
  have been admitted to hospital with the disease

30
Nested Case-control Studies

• Nested case-control studies occur within a
  defined cohort and sample controls from the risk
  set of persons at risk in the cohort at the
  occurrence of each case (called incidence-density
  sampling)
• Controls may become cases at some point later in
  follow-up (true of any study design if everyone
  is not followed until death)

31
Nested Case-Control (Incidence Density Sampling)
Sampling within a cohort Including all cases and
sampling controls from subjects disease free at
the time each case is diagnosed
Cases 10 Ds Controls 10 Cs Formed from 9
risk sets
D
C
D
C
D
C
C
D
C
Subjects in Case-Control Study
D
D
C
C
D
D
C
D
D
C
C
Risk Set 1
Risk Set 2
Etc.
Risk Set 9
32
Nested Case-control Studies (2)

• In example, 10 cases occur at nine points in time
  (2 cases occur at same time) giving rise to 9
  risk sets
• One control for each case is selected in each
  risk set, so 2 controls selected in risk set with
  2 cases
• One of the controls, selected in the second risk
  set, becomes a case at the fourth risk set

33
Nested Case-control Studies (3)

• Nested used by some authors to mean any
  case-control study conducted within a cohort
  study used here to mean incidence-density
  sampling design
• Outside of prior cohort study, incidence sampling
  of the study base giving rise to the cases
  produces same nested design
• Example Identify cases as they occur from cancer
  disease registry for S.F. county and obtain
  controls from random sample of county at time
  each case occurs

34
Nested Case-control Studies (4)

• Avoids potential biases of prevalent controls or
  prevalent cases
• Incidence density sampling gives unbiased
  estimate of ratio of disease rates in exposed and
  unexposed subjects
• Controls for secular (calendar time) trends since
  cases and controls are matched on calendar time

35
Case-Cohort Studies

• Alternative design to nested case-control
  study--in context of a cohort selects all cases
  and takes random sample of the cohort baseline
  for controls
• Like the nested design, some persons selected as
  controls may become cases
• Like the nested design, can be extended outside
  setting of cohort study to a study base

36
Case-Cohort Study
Sampling within a cohort Including all cases and
sampling controls from all subjects at baseline
of cohort
Study subjects
C
D
D
C
D
C
D
C
D
Controls in Case-Cohort Study
C
C
D
Cases in Case-Cohort Study
C
D
D
C
D
C
D
C
D disease case C control (no disease)
37
Case-Cohort Studies (2)

• Taking random sample of cohort at baseline gives
  estimate of prevalence of exposure in the cohort
  and allows calculation of attributable risk
• Controls are not linked to timing of disease
  occurrence so not matched to cases on calendar
  time
• A single baseline control group can be used for
  more than one disease outcome

38
Case-Cohort Studies (3)

• No necessity to screen out silent cases of
  disease from the control group
• Same sub-cohort can be used for future period of
  extended cohort follow-up
• Gives unbiased estimate of relative risk

39
Choosing a Study Design

• What has already been done?
• If no research, a rapid and inexpensive
  ecological study may be useful
• If several case-control studies have already been
  done, what would yours contribute?
• Is it worth repeating a cohort study that has
  been done in a one population in a different
  population (eg, in women rather than in men)?

40
Choosing a Study Design (2)

• Cohort study decisions
• Need to represent a larger population?
• Not necessarily relevant to biological question
  of relative disease risk in exposed and unexposed
• May be important to generalizing findings
• Larger cohort versus longer follow-up
• If disease rate is constant, same number of
  outcome events by more subjects rather than more
  follow-up
• Shorter follow-up limits potential usefulness of
  cohort to examine other research questions
• Shorter follow-up desirable if rapid answer to
  research question is a high priority

41
Choosing a Study Design Case-cohort versus
nested case-control

• Nested case-control somewhat more statistically
  efficient in cohorts with long follow-up and
  substantial censoring
• Analysis is more familiar and available for
  nested case-control
• Power of nested case-control requires only
  estimate of number of cases and controls
  case-cohort requires information on whole cohort
  and drop out rates

42
Choosing a Study Design Case-cohort versus
nested case-control (2)

• Case-cohort can use same controls for multiple
  disease outcomes
• Case-cohort allows direct modeling of disease
  incidence in exposed and unexposed
• Case-cohort allows multiple time scales (age,
  calendar time) nested case-control only one
• Nested case-control allows more efficient
  collection of time dependent exposures

43
Choosing a Study Design Case-cohort versus
nested case-control (3)

• Case-cohort can use same controls for a future
  period of additional cohort follow-up
• Case-cohort can use controls for other purposes
  (such as monitoring compliance)
• Controls can be selected more rapidly in
  case-cohort nested case-control may require
  control selection at end of study for late cases