constantly forgotten

14
09-11 H.S. 1 Constantly forgotten Hein Stigum Presentation, data and programs at: http://folk.uio.no/heins/ Talks, Constantly forgotten

Upload: hubik38

Post on 23-Dec-2015

235 views

Category:

Documents


0 download

DESCRIPTION

Constantly Forgotten

TRANSCRIPT

09-11 1 09-11 H.S. 1

Constantly forgotten

Hein Stigum

Presentation, data and programs at:

http://folk.uio.no/heins/ Talks, Constantly forgotten

09-11 2 09-11 H.S. 2

Agenda

•  Example

•  Concepts – Prevalence, risk and odds

•  Methods – Regression models

•  RD •  RR •  OR

09-11 3 09-11 H.S. 3

Smoking among 10th graders

Adjusted Adjusted

Variable N % p-valueOdds Ratio

Confidence interval

All 10785 14.5 Sex <.001

Boy 5045 8.7 1.0 Girl 5740 19.5 3.0 (2.7 - 3.4)

Parents marital status <.001Living together 7165 10.4 1.0 Single 3564 22.5 2.3 (2.1 - 2.6)

Educational plans <.001Academic 5157 9.9 1.0 Secondary 3 years 540 15.4 1.7 (1.3 - 2.2)Secondary 1 year 41 22.0 2.4 (1.9 - 3.1)Vocational 2701 23.4 2.8 (2.5 - 3.2)

Family economy <.001Well off 936 14.2 1.0 Good 9381 14.0 1.3 (1.0 - 1.6)Short of mony 331 27.8 1.7 (1.3 - 2.2)

?

09-11 4 09-11 H.S. 4

Constant term, sex

sexββy 1o +=

010

2030

40

0 1, Boys 2, Girls

1sexββy1-sexsex1 generate

1o +=

=0

1020

3040

0 1, Boys 2, Girls

Constant term, intercept

09-11 5 09-11 H.S. 5

Constant term, age

ageββy 1o +=

30ageββy30geaage30 generate

1o +=

−=-10

010

2030

40

0 20 25 30 35 40Age

-10

010

2030

40

0 20 25 30 35 40Age

09-11 6 09-11 H.S. 6

Prevalence and risk

•  Prevalence – Risk of having disease

•  Incidence proportion – Risk of getting disease

09-11 7 09-11 H.S. 7

Generalized linear models, GLM

•  Smoking as outcome –  y=0/1, x = covariates –  E(y|x) = P(y=1|x) = p –  family: y|x~Binomial

–  identity RD –  link: log RR –  logit OR

09-11 8

RD versus RR and OR

0 1 ∞

RR, OR

0 1

RD

-1

09-11 9

Linear binomial model, RD

09-11 H.S. 9

21

21

RDRDβpagesexββp

o

o

++=

++= βVariable RD

Confidence interval

Risk at reference 0.000 Sex

Boy 0 Girl (0.107 , 0.132)

Parents marital status Living together 0 Single 0.104 (0.088 , 0.119)

Educational plans Academic 0 Secondary 3 years 0.057 (0.027 , 0.088) Secondary 1 year 0.107 (0.069 , 0.145) Vocational 0.130 (0.113 , 0.147)

Family economy Well off 0

Short of mony 0.086 (0.038 , 0.134)

Girl, with single parents, academic plans and well off: Prevalence= 0.000+0.120+0.104=0.223 =22.3%

09-11 10

Variable RR Confidence

interval Risk at reference 0.043 Sex

Boy 1 Girl 2.45 (2.21 , 2.72)

Parents marital status Living together 1 Single 1.94 (1.77 , 2.12)

Educational plans Academic 1 Secondary 3 years 1.55 (1.26 , 1.90) Secondary 1 year 2.04 (1.68 , 2.46) Vocational 2.26 (2.05 , 2.48)

Family economy Well off 1

Short of mony 1.45 (1.21 , 1.74)

Log binomial model, RR

09-11 H.S. 10

21

21

21

)log(

RRRReep

agesexββp

o

o

β

agesexββo

⋅⋅=

=

++=++ β

β

Girl, with single parents, academic plans and well off: Prevalence= 0.043 * 2.45 * 1.94=0.203 =20.3%

09-11 11 09-11 H.S. 11

Odds and probability

NotDiseaseDisease

ppOdds =−

=1

P Odds1 % 1.01 %

10 % 11.11 %30 % 42.86 %

OddsOddsp+

=1

09-11 12 09-11 H.S. 12

Disease frequency depicted

Existing Cases

Healthy

New Cases

Healthy*time

050

100

time

a

t

09-11

Variable OR interval Odds at reference 0.039 Sex

Boy 1 Girl 3.01 (2.65 , 3.41)

Parents marital status Living together 1 Single 2.30 (2.05 , 2.58)

Educational plans Academic 1 Secondary 3 years 1.70 (1.31 , 2.19) Secondary 1 year 2.44 (1.90 , 3.13) Vocational 2.82 (2.49 , 3.19)

Family economy Well off 1

Short of mony 1.71 (1.31 , 2.24)

Logistic model, OR

09-11 H.S. 13

21

21

21

)1

log(

OROReeodds

agesexββpp

o

o

β

agesexββ

o

⋅⋅=

=

++=−

++ β

β

Girl, with single parents, academic plans and well off: Prevalence odds= 0.039 * 3.01 * 2.30=0.269 Prevalence=21.2%

09-11 14 09-11 H.S. 14

Summing up

•  Reporting constant -  Increases information a lot!

•  Technical –  0 must be part of the range

– Not for traditional Case Control