Brainscape's Knowledge GenomeTM

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


See full index

Yr 2 - Research methods > Regression analysis > Flashcards

Regression analysis Flashcards

1
Q

What is the most common statistical method used to analyse data from epidemiological studies?

A

Regression analysis

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

what is the purpose of a regression analysis?

A

assess association between exposire and outcome accounting for possible confounders

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

What does correlation measure?

A

Strength of linear association between two variables

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

What is linear regression?

A
  • Nature of association = helps identify which is the outcome and which is the exposure
  • More formal description of association between two variables when outcome and exposure can be identified e.g. birth length and childhood height
  • Depends on explicitly defining line which best describes association
  • Allows estimation of value of outcome per unit change in exposure
How well did you know this?
1
Not at all
2
3
4
5
Perfectly
5
Q

In the linear regression line y = a + bx what does each letter represent?

A

x = exposure

y = outcome

a = y intercept

b = slope of line (= REGRESSION COEFFICIENT)

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

What are the limitations of using regression lines as estimates?

A

They should only be used to ,ale estimates within the range of data on which they were based (we have no information saying that the association is the same outside of this range - it could be totally different!)

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

What should you be aware of with regression lines?

A

UNITS

e.g. make sure that a change of 1 unit in esposure is feasible or not!

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

What can you also do with a linear regression line?

A

Calculate confidence intervals

Perform hypothesiss tests for regression coefficients (null hypothesis is that the coefficient value is 0)

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

What is R2?

A

Correlation coefficient squared

= proportion of variation in outcome explained by variation in exposure

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
10
Q

The regression equation derived for the association between chip consumption and serum cholestrol is calculated to be:

Serum cholestrol (mmol per L) = 4.5 + 1 X (100g portion chips/week)

If a person eats 2 50g portions per week how much higher would ther predicted cholesterol be than someone who ate no chips?

A

1 mmol per L

Because 4.5 is nor,al persons serum cholestrol

the extra serum cholestrol is: 1 X (100g portion chips/week) = 1 X 1 = 1 mmol per L

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

Which of the following statements is false?

Both the correlation coefficient and the regression coefficient:

  1. Take the same value when there is no association between variables
  2. have the same associated p value when the null hypothesis of no association between variables is tested
  3. have the same sign (i.e. both +ve & both -ve)
  4. have units
A

THEY DO NOT BOTH HAVE UNITS

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
12
Q

Correlation vs linear regression:

What do they measure?

What are there unuts?

What is there maximum and minimum?

What is their value when there is no association?

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

Does strength of evidence against null hypothesis depend on the statistic chosen (correlation vs. linear regression)?

A

No

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

Which provides more information correlation or regression coefficient?

A

regression coefficient

How well did you know this?
1
Not at all
2
3
4
5
Perfectly
15
Q

If exposure and outcome can be identified which is more appropriate correlation or regression coefficient?

A

Regression coefficient

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

What are the types of regression modelling (5)?

A

Linear (regression coefficient)

Poisson (rate ratio)

Logistic (odds ratio)

Multinominal logistic (odds ratio)

Ordinal logistic (odds ratio)

n.b. ALL are parametric

17
Q

When does type of regression vary?

A

Depends on form of outcome variable

18
Q

When the outcome is continuous which type of regression is used?

A

Linear

19
Q

When the outcome is Discrete which type of regression is used?

A

Poisson

20
Q

When the outcome is Binary (categorical) which type of regression is used?

A

Logistic

21
Q

When the outcome is Nominal (categorical) which type of regression is used?

A

Multinominal logistic

22
Q

When the outcome is Ordinal (catgorical) which type of regression is used?

A

Ordinal logistic

23
Q

What is survival analysis?

A

Risk of outcome may not be constant over time e.g. recurrence rates of tumours following cancer treatment

24
Q

What do survival analysis methods not mae the assumption of?

A

That rates are cinstant

25
Q

Give examples of survuval analysis methods:

A
  • Kaplan-Meier estimation of survival curves
  • Mantel-Cox estimates of hazard ratios (Log rank test)
  • Cox (proportional hazards) regression
  • Poissons regression
26
Q

How do we adjust for confounders?

A

Use multiple regression models = estimated association between exposure and outcome, adjusted for confounders

27
Q

How should we report results of adjusted analyses?

A

Should present both the unadjusted and adjusted results

= allows assessment of extent of confounding i.e. see how much unadjusted association changes after adjustment

28
Q

If the regression coefficient (95% confidence interval) for the association between an exposure and outcome is .58 (0.23, 0.94) units befor adjusting for confounders and 0.56 (0.22, 0.89) after, how much effect do confounders have on the association between exposure and outcome?

A

Little/none because the regressiom coefficient barely changes!

29
Q

If we have 1 outcome, 1 exposure and 1 confounder (categotical), an interaction between exposure and confounder exist if…?

A

Association betweeen outcome and exposure varies across categories of confounder

e.g.

outcome = birthweight

exposure = gender

confounder = social class

If association between birthweight and gender is different in the different social class groups then there is an interaction between gender and social class

30
Q

What is subgroup analysis/stratified?

A

e.g. look at each social group individual

often no pre-specified hypothesis = multiple testing/ data dredging = small p values found by chance:

in this case anlyses are only meaningful if there is evidence of main effects i.e. esposure-outcome plus confounder-outcome associations; evidence of an interaction -> ALWAYS TREAT FINDINGS WITH CAUTION

31
Q

What is meta analysis?

A

Statistical technique for combining results from >1 independent study which provides precise estimate of exposure-outcome association (based on ALL of the results)

32
Q

In Meta analysis, do you think that the contribution of individual study estimates to the combined estimate of exposure-outcome association:

  1. will be equal from all studies
  2. will be greater for smaller studies
  3. will be greater for larger studies
A
  1. will be greater for larger studies

(weights according to the size of each study are included with smaller studies having a smaller weighting)

33
Q

What does the validity of meta analysis depend on ?

A

The quality of systematic review on which meta analysis is based

caution: issue of publication bias (studies with association are more likely to get published)

34
Q

What are meta analyses often used for?

A

Assessing the clinical effectiveness of healthcare interventions by combining data from >2 randomised controlled trials

35
Q

What is the possible range of values for regression coefficients?

A
  • ∞ to + ∞
36
Q

If the risk of outcome is not likely to be continuous over time which methods can be used?

A

Survival analysis methods

Yr 2 - Research methods (11 decks)

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