Brainscape's Knowledge GenomeTM

Browse over 1 million classes created by top students, professors, publishers, and experts.


See full index

Principles of Research > L14 to L16 - OS, Assoc and causation > Flashcards

L14 to L16 - OS, Assoc and causation Flashcards

1
Q

Describe Ecologic (Correlational) Study

A

Using Ppn as unit of observation, Disease rate and E measured in ppn, and relation is examined using aggregated data

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
2
Q

Advantage and Disadvantage of Ecologic Study

A

Good:

  • Cheap
  • Easy to conduct (no need collect more data)
  • Can generate hypothesis

Bad:

  • Cannot link E&O to Individuals (only groups)
  • High risk of confounding (CFing)
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
3
Q

One problem with ecologic study

A

ECOLOGIC FALLACY: applying results to individuals

- Not true as results can only apply on populations since data is aggregated or group-level

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
4
Q

Describe the design of Case report and Case Series

A
  • Case Report: Detailed report of single case
  • Case Series: Profile of series of patients
  • With respect to factors that could be related to O
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
5
Q

Strength and limitation of case report and case series

A

Strength: Hyp generating
Limitation: No comparison/control group

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
6
Q

Using a case report or case series, how is causality attempted to be established?

A

Challenge - Dechallenge - Rechallenge

i. e. Administer, withdraw, re-administer
(note: if ADR too serious, it is NOT ethical to rechallenge, or even challenge in the first place)

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
7
Q

Describe the design of cross-sectional studies (XSS)

A

An analytical OS where E&O of individuals assessed simultaneously, giving information on PREVALENCE

(i.e. a “snap-shot in time”)

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
8
Q

Strength and limitation of XSS

A

Strengths:

  • Efficient (no waiting required)
  • Many O and E can be assessed
  • No loss to f/u (no f/u involved)

Limitation: Unclear temporal relation btwn E and O
- Hence difficult to establish cause-effect since you don’t know which cause which, and what came first

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
9
Q

Describe the design of Cohort study (CohS). State the two subsets of CohS.

A

Design: Select subjects based on EXPOSURE, then follow to O. The difference between E and uE are analysed for BOTH CohS

  1. Prospective CohS: O not occured. Subjects followed through time and identified via f/u.
  2. Retrospective CohS: O occured. Subjects are ID-ed via interview or records. No need f/u
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
10
Q

What must be considered when selecting nE group for CohS?

A
  1. nE and E should be similar to reduce CFing

2. Collect potential baseline differences that could affect O and account for it

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
11
Q

Strengths and limitation of cohort study

A

Good:

  1. Clear temporal sequence between E and O
  2. Several O assoc with E can be studied
  3. Best for rare E
  4. Can measure O incidence (O over period of time)

Bad:

  1. Bad for rare O (rq large n)
  2. If O takes long to develop, study is inefficient (time-consuming, costly)
  3. Loss to f/u (LTFU): can form bias
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
12
Q

Advantage and disadvantage of rCohS compared to pCohS

A

rCohS:
- More efficient
BUT
- Less control over quality and quantity of data quality of data hence greater potential for bias

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
13
Q

Describe the design of case-control study (CCS)

A

Analytical, retrospective study where subjects are selected based on OUTCOME, and their E compared to find predictors of O

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
14
Q

A CCS is to be carried out where the cases have BPH. State the ppn in which you will sample out and use as control. Explain your choice

A

Ppn: Male without BPH

Reason:

  • Ppn must be AT RISK of developing O
  • Female no risk of getting BPH hence not suitable as control
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
15
Q

One important point of selecting case and controls respectively for CCS

A

Case: Selected INDEPENDENTLY of E (i.e. if case, just select, don’t care exposure)

Control: Sample independently of E and represents ppn at risk of becoming cases

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
16
Q

Strengths of CCS

A
  1. Best for rare O
  2. Efficient for O that takes long time to develop
  3. Study multiple E/risk factors for single O
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
17
Q

Limitations of CCS

A
  1. Not for rare E (use CohS instead)
  2. Selection of ctrl is difficult
  3. Data on E may be difficult to get
  4. Cannot estimate rate of O, but can get relative measure (odds ratio)
  5. Increased bias as O is already known when assessing exposure (solution: blind the interviewers)
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
18
Q

Describe the design of Nested CCS (nCCS)

A

CCS nested within CohS

  1. Coh is identified
  2. Within Coh, identify subjects with O during f/u
  3. Those with O are now cases for CCS
  4. Within Coh, select sample from the rest of the Coh. This is the control
  5. Assess E/risk factors that predicts for O
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
19
Q

Strengths and limitations of nCCS

A

Good:

  • Efficient as data has already been collected
  • Decreased bias compared to CCS as subjects are drawn from SAME ppn
  • All other good of CohS

Bad:

  • Resh qns cannot change, hence cannot store materials for later analysis on a sample of study subjects
  • If entire Coh data available at no additional hassle, might as well use whole Coh
20
Q

Distinguish between Risk Ratio (RiR), Rate Ratio (RaR) and Odds Ratio (OR) in terms of:

  • Which studies are they used for?
  • How they are calculated
A
  1. RiR:
    - CohS, when subjects are followed for same duration of time
    - Formula: Cuin E/Cuin uE
  2. RaR:
    - CohS, when subejcts have different lengths of f/u
    - Formula: Incidence rate in E/ Incidence rate in uE
    - Unit of time when each subject is observed = person-time (e.g. person years)
  3. OR:
    - CCS, since RR cannot be calculated directly (less commonly for XSS and CohS)
    - Formula: Odds that case was E/ Odds that control was E
    (odds of an event = no. events/no. of non-events)
21
Q

Mathematically, why is OR a good estimate of the risk ratio when O is rare?

A

Supposed 2x2 table a,b,c,d
a &laquo_space;b and c &laquo_space;d
- Formula for RR:
[a/(a+b)]/[c/(c+d)] ≈ a/b x d/c = ad/bc = OR

Hence OR is a good estimate of the RiR when O is rare

22
Q

RiR and RaR are classified as

A

Relative risk

23
Q

For multiple E groups (more than 2 groups), how to get ratio?

A

Use the Control group, or least E group as the reference group (denominator for equation)

24
Q

What information does confidence interval (CI) give? How does it give the information you mentioned?

A
  1. Statistical significance: whether it includes 1 in its range
  2. Precision of a point estimate: The width of the interval
25
Q

What is the width of CI affected by?

A
  1. n: large n = narrow CI = more precise RR or OR
  2. SD: higher = increased width
  3. Confidence level = 1-a
    - High confidence = increased width
26
Q

Why must wide CI be interpreted with caution?

A

If too wide, it may mask statistical significance. A larger study may narrow the range and show statsig

27
Q

General steps that establishes causation from E to O

A
  1. Is assoc observed btwn E and O?
  2. If yes, is assoc valid? Anything else affect the findings?
  3. If nothing from 2, is the observed association causal?
28
Q

Three alternative explanation that can cause E and O to be observed, aside from causation

A
  1. Chance
  2. Bias
  3. Confounding (CFing)
29
Q

What factors affect the random CHANCE of mistaking an assoc being observed between E and O?

A
  1. Sample size: Smaller sample = larger chance variability
  2. p-value: probability for more extreme result to occur by chance if H0 is true. If p-value small, observation is less likely by chance
  3. 95%CI: Narrower = less likely observed by chance
30
Q

Define Bias. What are the two major type of bias?

A
  • A SYSTEMATIC error in design/conduct/analysis of study, resulting in inaccurate estimate of E effect on O. Once introduced, cannot be fixed

Types:

  1. Selection bias
  2. Information bias (Obv bias)
31
Q

Cohort study investigating effects of health screening on overall health outcome, in which exposed group includes individuals who voluntarily participate in annual health screening, & unexposed group includes individuals who do not participate in annual health screening. Those who participate in annual health screening may differ in other important ways (e.g. more health conscious, having family)

What kind of bias is observed here?

A

Selection Bias (due to an underlying difference between groups selected)

32
Q

A CCS uses patients with arthritis as a control group to evaluate potential protection against colorectal cancer with use of NSAIDs. One investigator stated that the OR is underestimated.

State the type of bias that may have caused the underestimation of OR and explain why.

A

Selection Bias, due to selecting ctrl group that is skewed in the outcome of interest

  • Patients with arthritis consume NSAID as chronic medication. Hence control has increased exposure
  • Since OR = ad/bc, b&raquo_space; d, OR is underestimated
33
Q

How does info bias arise Describe the two subtypes of information bias

A

Information bias arise from incorrect determination of E, O or both, causing misclassification of subjects hence inaccurate results

Subtypes:

  1. Reporting bias (subjects reluctant to report their past truthfully)
  2. Recall bias (subjects cannot recall properly)
34
Q

Types of misclassification from information bias and its effect on results

A
  1. Differential misclass: rate of misclass different in different study groups
    - Effect: bias in EITHER direction (leads to apparent assoc or no assoc when there was)
  2. Non-differential misclass: rate of misclass in study groups are the same
    - Dilute towards null value (towards no assoc)
35
Q

A patient was unable to recall accurately the number of eggs he ate the last week when interviewed, and was wrongly misclassified into the control group.

State the type of bias present here

A

Information (recall) bias

36
Q

A 15yo girl has a history in smoking. However out of fear of embarrassment, the girl decides to report herself as a non-smoker (never smoked before) in a survey.

State the type of bias present

A

Information (reporting) bias

37
Q

State the three conditions to be met for a variable to be identified as a CF?

A
  1. Association with E
  2. Risk factor for O
  3. NOT intermediary step in the causal pathway from E to O
38
Q

In a study between E: use of cigarettes and O: COPD, it was found that high proportion of cigarette users have a larger mean mass of mucus in the windpipe compared to non-smokers.

Is the mean mass of mucus a CF in the study?

A
  1. Associated with E: Y
  2. Risk factor for O: Y
  3. Intermediary step in causal pathway from E to O: Y

Smoking may cause mucus buildup, which hence leads to COPD

Hence not all conditions are met, mean mass of mucus is NOT a CF in the study

39
Q

In a study between E: use of cigarettes and O: COPD, it was found that high proportion of cigarette users are older in age compared to non-cigarette users.

Is age a CF in the study?

A
  1. Associated with E: maybe (minimum age to smoke)
  2. Risk factor for O: Y (established risk)
  3. Intermediary step in causal pathway: N
    - Smoking does not lead to old age.
    - Hence, age IS a CF
40
Q

In terms of study design, how can CF be controlled?

A

Restriction:

  1. Properly set I/E criteria for subjects (e.g. limited to only non-smokers for the study)
  2. Matching case to control in which baseline chracteristics are similar (e.g. both are smokers)
41
Q

In terms of Data analysis, how can CF be controlled?

A
  1. Stratified analysis: grouping and analysing similar subjects together (e.g. smokers together)
  2. MultiVARIABLE analysis: Use models that examines one E on O while the other variables are controlled (E.g. MLoR)
42
Q

An association between many E and a single O shows the following data for one of the exposure in a CCS at sig. level = 0.05

  • SLoR: expB = 2.08 (1.80 - 2.39)
  • MLoR: expB = 2.13 (1.97 - 2.80)
    both are statsig (P<0.05)

Formulate a conclusion for the study

A

At a significance level of 0.05, subjects who are exposed has 2.13 times (1.97 - 2.80) the odds of developing the O compared to subjects are are not exposed, after controlling for all other exposures (p<0.05).

(note: to control for CF, always interpret assoc based on adjusted RR or OR, which can be found using MLoR)

43
Q

What is one thing that you can do for CF that you can’t do for bias?

A

CF: can be adjusted for in analysis, provided that info on confounders are available

Bias cannot be eliminated in analysis: once introduced into study, it cannot be modified or removed.

44
Q

If controlling for CF in data analysis, why calculate the crude RR/OR then calculate adjusted RR/OR?

A

Compare the unadjusted and adjusted values. If there is a small difference between them, chances are the factors controlled for in MLoR are not CF

45
Q

State the Bradford Hill Criteria to establish causation

A

ACCESS PTB:

  • Analogy: Assoc established for similar E?
  • Coherence: findings make sense with current data/knowledge?
  • Consistency: similar studies have same findings? (in different ppn)
  • Experiment: RCT/in vivo supports causal relationship
  • Specificity of assoc (nt impt)
  • Strength of assoc: stronger assoc = more likly causal
  • Plausibility: Biologic plausibility: assoc make sense based on biologic knowledge?
  • Temporality: E should be before O
  • Biological gradient (for dose-response): if dose-response relation established, then its strong evi for causal relationship

Principles of Research (9 decks)

Brainscape helps you reach your goals faster, through stronger study habits.
© 2024 Bold Learning Solutions. Terms and Conditions