M2 Medical Epidemiology

1
M2 Medical Epidemiology

• How to Fairly Compare Disease Frequencies Between
  Groups

2
How to Fairly Compare Disease Frequencies Between
Groups

• Simple epidemiologic indices review/summary
• Interpreting epidemiologic comparisons overview
• chance
• bias
• confounding
• Adjustment of epidemiologic indices for
  confounding
• direct
• indirect

3
Simple epidemiologic indices review/summary

• Questions
• What fraction of a group has the condition now?
• What fraction of the community carries the
  condition at any one time?
• What is the endemic level of the condition,
  relative to the size of a community?
• What fraction of UI students has hay fever now?

4
Simple epidemiologic indices review/summary
Answer

• Point prevalence, or prevalence for short.
• A dimensionless proportion.
• Sometimes erroneously called prevalence rate

5
Simple epidemiologic indices
review/summary Questions

• What is the cumulative risk (probability) of
  developing a condition at least once during a
  fixed time period?
• What fraction of a group can we predict will have
  developed a condition over a given time period,
  or during an epidemic?
• Why must I take this medicine, doctor? What are
  my chances of a heart attack in the next ten
  years, if I don't?

6
Simple epidemiologic indices review/summary
Answers

• Cumulative incidence
• A dimensionless proportion
• Called the attack rate when describing infectious
  disease outbreaks,
• e.g., The attack rate in the county during the
  West Branch hepatitis outbreak was estimated as
  6.565 cases/1000 population.
• One women in 11 (9) is expected to develop
  breast cancer during her lifetime. 

7
Simple epidemiologic indices
review/summary Questions

• How strong is the process causing new cases?
• How many new cases occur per person per unit
  time, or other unit of experience (e.g., per
  passenger-trip, per passenger-mile traveled)?
• How many new cases of esophageal cancer occur in
  Illinois/1000 population per year?
• How many ruptured spleens occur from automotive
  accidents in Illinois, per million person-miles
  traveled?
• How many new HIV infections occur per 1000 acts
  of vaginal intercourse? Of anal intercourse?

8
Simple epidemiologic indices review/summaryAnsw
er

• Incidence density (rate, dimension new cases
  per unit of experience, such as person-year,
  passenger-mile, sexual acts)
• e.g. 5 new cases per 1000 persons per year
• 5 new cases per 1000 person-years
• .005 new cases per person per year
• Units e.g.
• New cases / persons x years
• New cases / million passenger-miles
• New cases / 100 sexual acts

9
Simple epidemiologic indices
review/summary Examples

• Mortality rate The death density, i.e. the
  incidence density of death.
• For political units in which records are kept
  routinely and where the population size may be
  constantly changing, often calculated using the
  mid-year population as denominator.
• The mid-year population approximates the total
  person-years exposure in the population for the
  full year.

10
(No Transcript)
11
Simple epidemiologic indices
review/summary Examples

• Case-Fatality rate The cumulative incidence of
  death due to a disease, during the course of the
  disease.
• i.e. the fraction of cases which result in death
  from the illness.
• Equivalently, the chance of dying from a case of
  the disease.

12
Case Fatality Rate

• The cumulative incidence of death due to a
  disease, during the course of the disease.
• i.e. the fraction of cases which result in death
  from the illness.
• Equivalently, the chance of dying from a case of
  the disease.
• Number of deaths from a specific disease/number
  of cases of the disease.
• Usually overestimates. Why?

13
Simple epidemiologic indices review/summary
14
Simple epidemiologic indices review/summary

• the rate (incidence density) of 52 per 100
  person-years or, equivalently, 1 per 100
  person-weeks.

Incidence density (ID) vs. Cumulative incidence
(CI) Question In a population of 100
persons, deaths occur at the rate (incidence
density) of 52 per 100 person-years or,
equivalently, 1 per 100 person-weeks. After
one year of this, what proportion of the 100
people will have died?
15
Simple epidemiologic indices review/summary

• Answer 41,
• not 52

16
Simple epidemiologic indices review/summary

• For all factors stable,
• P ID x MD
• where
• P Prevalence
• MD Mean Duration

17
Example

• If incidence is 12 new cases per 1000
  person-years.
• And duration of illness is 6 months.
• What is the average prevalence?
• 6 per thousand

18
Simple epidemiologic indices review/summary
Relative Risk RReither CUMULATIVE INCIDENCE
RATIO CIR CI1/CI0 or INCIDENCE DENSITY RATIO
IDR ID1/ID0
19
Association a statistical feature of
comparisons(s), with six possible explanations

• Causation, with exposure promoting disease
• Chance
• Bias 2 categories
• Selection Bias
• Measurement bias
• Confounding variable(s)
• Causation, with disease promoting appearance of
  the exposure
• Always ask are there plausible alternative
  explanations for the data? 

20
Chance

• due to random variation from sampling or
  measurement
• addressed using
• statistical tests of hypotheses (p-values)
• confidence intervals
• power analyses

21
Bias. 2 types

• Selection, the way you selected subjects for the
  study biased your results.
• Measurement, the way you measured variables in
  your subjects biased the results.

22
Selection bias

• Bias from the use of a non-representative group
  as the basis of generalization to a broader
  population of subjects or patients.
• For instance, a common bias of this type appears
  when
• the prognosis of patients newly diagnosed with a
  given disease is inferred from the study of
  hospitalized patients with this disease at a
  major referral center,
• and
• the disease in question has a broad spectrum
  behavior. 

23
Selection bias

• More commonly
• We have 2 groups
• Exposed and unexposed
• We compare them with regards to an outcome.
• But the way we selected the 2 groups causes
  differences in the outcome that have nothing to
  do with the exposure.
• Example if we used hospitalized smokers as the
  exposed and healthy volunteer non-smokers as the
  unexposed.

24
Selection Bias (Admission Rate -- Berkson)
25
Selection Bias (Berkson)
Necropsies
26
More Selection Biases

• Whenever we compare a group of patients who use a
  drug to those who dont in a non experimental
  observational study (cohort, not randomized).
• The 2 groups differ in many respects.
• One of the most important respects is that the
  patients on the drug have a reason to be on it
  (indication). The others dont. Called Bias by
  indication.

27
Bias by indication

• For example calcium channel blockers have 2
  indications hypertension and coronary disease.
• If you compare hypertensive patients who are on
  Ca blockers to those who are on other agents (not
  randomized, totally at the discretion of their
  doctors), we would find

28
Bias by indication

• Patient on Ca blockers have higher prevalence of
  CAD
• Also higher prevalence of risk factors for CAD
• So if you do an observational study of
  hypertensive patients, comparing the outcome in
  those on Ca blockers to those on other agents,
  you may find

29
Bias by indication

• That patients on Ca blockers have much worse
  outcomes.
• This is bias by indication.
• You can adjust and correct for preexisting heart
  disease and for risk factors, but may not be
  enough.

30
Bias by indication

• If you compare hypertensive patients who are on
  minoxidil or hydralazine to those on other agents
  you find
• That patients on those agents have higher BP
• Is it because they dont work as well ?
• No, the opposite. They are reserved for those
  with severe resistant hypertension.
• That is the indication for those agents.

31
Survivor Treatment Bias

• Patients who received statin during admission for
  MI had much lower in-hospital mortality.
• Statin?
• The ones who died are different.
• Some died very soon after admission (no statin).

32
Competing Medical Issues Bias

• Some were so sick that they were treated with
  multiple drugs, modalities, ICU etc.
• No statin

33
Bias by contraindication

• If you compare hypertensive patients who are on
  beta blockers to those on other agents you find
  that they have better outcomes.
• That does not mean they are better for you. No,
  this comparison is biased by contraindication.
• Beta blockers are contraindicated in severe COPD,
  CHF, PVD etc.

34
Measurement bias

• Systematic or non-uniform failure of a
  measurement process to accurately represent the
  measurement target, e.g.
• different approaches to questioning, when
  determining past exposures in a case-control
  study.
• more complete medical history and physical
  examination of subjects who have been exposed to
  an agent suspected of causing a disease than of
  those who haven't been exposed to the agent. 

35

• Measurement Bias -- Recall Bias

36
Measurement Bias
Family information biasThe flow of family
information about exposures and illnesses is
stimulated by and directed to a new case in its
midst.
37
Measurement Bias
38
Measurement Bias -- Family Information
39
Avoid confounding

• Confounding refers to distortion of the true
  biologic relation between an exposure and a
  disease outcome of interest, due to a research
  design and analysis that fail to properly account
  for additional variables associated with both.
  Such variables are referred to as confounders or,
  less formally, as lurking variables. 

40
Confounding
41
Confounding
42
Confounding
43
Direct Rate Adjustment
44
Age specific mortality rate
45
Direct Rate Adjustment
46
Direct Rate Adjustment
47
Direct Rate Adjustment
48
Direct Rate Adjustment
49
Direct Rate Adjustment
50
Direct Rate Adjustment
51
Direct Rate Adjustment
52
Indirect Rate Adjustment

• Calculate Expected Deaths
• ?
• Divide Observed Deaths by Expected Deaths (O/E)
• ?
• SMR (Standardized Mortality Ratio)

53
Indirect Rate Adjustment

• Calculate SMR standardized mortality ratio.
• SMR Observed mortality / Expected mortality
• To Calculate that you need to calculate expected
  mortality.

54
Indirect Rate Adjustment
55
Indirect Rate Adjustment
56
Indirect Rate Adjustment

• Calculate Expected Deaths
• ?
• Divide Observed Deaths by Expected Deaths (O/E)
• ?
• SMR (Standardized Mortality Ratio)

57
Indirect Rate Adjustment

• STANDARDIZED MORTALITY RATIO (SMR)
• OBSERVED DEATHS/EXPECTED DEATH
• 54/73.2 74

58
Indirect Rate Adjustment
59
Indirect Rate Adjustment

• STANDARDIZED MORTALITY RATIO (SMR)
• OBSERVED DEATHS/EXPECTED DEATHS
• 22/38 58

60
Proportional Mortality

• The 4 leading causes of death in Chamapign County
  are.
• CAD is the leading cause being responsible for
  32 of all deaths in the County in 2002.

61
(No Transcript)
62
Proportional Mortality

• Number of deaths from a specific cause/ Total
  number of deaths in same time

63
Proportional Mortality RatioPMR

• Proportional Mortality Ratio
• Proportion of deaths from specified cause
  /Proportion of deaths from specified cause in
  comparison population

64
Proportional Mortality RatioPMR

• CAD is responsible for 32 of all deaths in the
  County in 2002. (Compared to 40 in the State of
  Illinois)
• PMR 32/40 32/40 0.8
• Is that good or bad ?

65
PMR

• Relative frequency of other causes of death can
  affect the PMR for the cause of interest
• An epidemic of a fatal disease in your population
  will decrease PMR for all other causes
• Low mortality from a very common cause (CAD for
  example) in your population will increase PMR for
  all other causes

66
PMR

• Fast, easy, cheap
• Can be calculated when all you have is death
  certificates
• Dont need information on demography of
  population.
• Leading Causes of Death

67
(No Transcript)
68
How does one decide whether to present a set of
data using crude, adjusted, or category-specific
indices?
If possible, use crude indices only to produce a
quick picture of the magnitude of a problem in a
population, for the purpose of establishing a
prima facie need for public health and/or medical
services, and as a first-cut at estimating the
resources needed.
69
How does one decide whether to present a set of
data using crude, adjusted, or category-specific
indices?Use category-specific indices when you
wish to focus attention on the problem in one or
a few population subgroups, when space is
available to give a detailed presentation in
order to communicate the fullest understanding of
the data, and especially if specific indices vary
between two populations being compared in a
different manner in different population
subgroups (e.g. effects are modified by age, sex
or race).
70
How does one decide whether to present a set of
data using crude, adjusted, or category-specific
indices?

• Use adjusted rates when
• you wish to avoid possible confounding,
• but do not have the space to present the full
  schedules of specific indices, or your audience
  does not have the patience for that,
• Avoid adjusted rates when
• there variable being adjusted out is an effect
  modifier, that is, the relationship between
  groups being compared changes from stratum to
  stratum -- more later on this.

71
How does one decide whether to present a set of
data using crude, adjusted, or category-specific
indices? Note that

• crude indices require one only to know the
  numerator cases and the denominator (population
  size or exposure-time) of each total population
  to be compared
• indirect adjustment requires knowledge of only
  the numerator cases from the total populations
  and the (joint) distributions of confounder(s) in
  the populations to be compared
• direct adjustment and specific rates require
  knowledge of both the numerator cases and the
  corresponding denominators within levels of the
  confounding variable(s), for all populations
  under comparison.

72
Note that

• A directly adjusted rate of a single community
  means nothing by itself. It is only used to
  compare different communities and only if all of
  them are adjusted to the same standard
  population.
• SMR of a single community IS useful. It does by
  itself compare 2 populations.