Quantitative Reasoning

Design of Studies: Looking Forward

Diabetes Risk
Female
Male
Column total

rate(diabetes | male) =

Diabetic
72,000
52,000
124,000
52,000
208,000

risk(diabetes | male) =
rate(male | diabetes) =

sensible.

52,000
124,000

Healthy
144,000
156,000
300,000

Row total
216,000
208,000
424,000

= 0.25, or 25%.
52,000
208,000

= 0.25

0.42. risk(male

| diabetes) not so

Risk Ratio (RR)

Female (F)
Male (M)
Column total

risk(D|F) =

Diabetic (D)
72,000
52,000
124,000

72,000
216,000

Healthy
144,000
156,000
300,000

Row total
216,000
208,000
424,000

0.33 > 0.25 = risk(D|M)

risk(D|F)
risk(D|M)

0.33
0.25

Risk ratio =

RR = 1: no association.

1.33. Also called relative risk.

Population vs Sample
Most observational studies deal with a sample.

Carefully designed studies allow accurate estimation

of population risks and population risk ratio (RR).

Two strategies:

From each exposure group. Eg: females and males.

From each disease group. Eg: diabetic and healthy.

Study 1: the sample table

Female
Male
Column total

Diabetic
378
53
431

Healthy
702
155
857

Row total
1,080*
208**
1,288

*: 5 in 1000 females sampled. **: 1 in 1000 males sampled.

1
3

Expected no. of diabetic females = 1,080 = 360. Actual = 378.

1
4

Expected number of diabetic males = 208 = 52. Actual = 53.

Numbers in black fluctuate with repeated samplings.

Study 1: estimation
Diabetic
378
53
431

Female*
Male**
Column total

Healthy
702
155
857

Row total
1,080
208
1,288

*: 5 in 1000 females sampled. **: 1 in 1000 males sampled.

Estimated risk(D|F) =

378
1,080

0.35. 0.33 = population risk(D|F)

53
208

0.25. 0.25 = population risk(D|M)

Estimated risk(D|M) =
Estimated RR

0.35
0.25

1.37. 1.33 = population RR

Study 2
Diabetic
364
256
620*

Female
Male
Column total

Healthy
142
158
300**

*: 5 in 1000 diabetic sampled. **: 1 in 1000 healthy sampled.

364
506

256
414

0.72 0.33.

0.72
0.62

1.16 1.33

Estimation does not work. What can be done?

0.62 0.25

Row total
506
414
920

Cohort and Case-control Studies

A
Study

Samples from

Advantage

Cohort
Exposure groups Risks and RR can be estimated
(study 1)
CaseDisease groups Good for rare diseases
control
(study 2)

Summary

A risk is like a rate, but more specific. Confounding

still relevant.

Risk ratio, or relative risk: measures association.

A cohort study permits accurate estimation of

population risks and RR with random samples. Not
possible with a case-control study.

Final remarks on sampling

Randomised experiments: subjects need not resemble
population. Extrapolation to population is an issue.

Observational studies: both extrapolation and

confounding are important issues.

Cohort studies rely on random samples for accurate

estimation of risks.

Optional: Prospective and Retrospective

Prospective: goes forward in time, like most cohort

studies.

Retrospective: looks back in time, includes all casecontrol studies.

A historical cohort study is very rare.