**** pe vs. pd ****

* Use Smoking-Mortality as a real example in NHIS Data

* Ages 50-84 attained age
* Ages 40-84 baseline age


* PAF from cigarette smoking
* Confounders are gender, race/ethnicity, and age
* Poor Health and Institutionalization (i.e., collider bias / survey selection bias)


use "F:\Private\ryma0788\obesity_mortality\NHIS\nhis_healthy.dta", clear

drop if agecalc < 40
drop if age_e >= 85
drop if age_e < 50
drop if hispanic == 1
drop if nhother == 1
drop if nhblack == .
drop age_e

replace yexpose = .25 if yexpose == 0

gen age_e = agecalc+yexpose

drop if mortwt == 0

* Smoking Status
* Never
* Former
* Current

gen nsmoker = .
replace nsmoker = 0 if smoke1 == 20
replace nsmoker = 0 if smoke1 >= 30 & smoke1 <= 32
replace nsmoker = 0 if smoke1 == 92
replace nsmoker = 1 if smoke1 == 10
replace nsmoker = 1 if smoke2 == 30

gen fsmoker = .
replace fsmoker = 0 if smoke1 == 10
replace fsmoker = 0 if smoke1 == 20
replace fsmoker = 1 if smoke1 >= 30 & smoke1 <= 32
replace fsmoker = 1 if smoke1 == 92
replace fsmoker = 1 if smoke2 == 20
replace fsmoker = 1 if smoke2 == 40

gen smoker = .
replace smoker = 0 if smoke1 == 10
replace smoker = 0 if smoke1 >= 30 & smoke1 <= 32
replace smoker = 0 if smoke1 == 92
replace smoker = 1 if smoke1 == 20
replace smoker = 1 if smoke2 >= 10 & smoke2 <= 13

replace nsmoker = 0 if smoker == 1
replace fsmoker = 0 if smoker == 1

drop if smoker == .

gen men = 0
replace men = 1 if sex == 1

gen cage = agecalc - 50
gen cage2 = cage*cage

sum birthyr

gen cbirth = birthyr - 1940
gen cbirth2 = cbirth*cbirth



* Likelihood of Smoking Confounded by Sex and Race/Ethnicity

gen bmen = nhblack*men

logit smoker cage cage2 cbirth cbirth2 men nhblack bmen if fsmoker==0 & age_e > 50, or

* White Men have 65% higher odds of smoking than never smoking compared to White Women
* Black Women have 4% lower odds of smoking than never smoking compared to White Women
* Black Men have 120% higher odds of smoking than never smoking compared to White Women

logit fsmoker cage cage2 cbirth cbirth2 men nhblack bmen if smoker==0 & age_e > 50, or

* White Men have 138% higher odds of former smoking than never smoking compared to White Women
* Black Women have 35% lower odds of former smoking than never smoking compared to White Women
* Black Men have 63% higher odds of former smoking than never smoking compared to White Women


gen status = .
replace status = 1 if nsmoker==1
replace status = 2 if fsmoker==1
replace status = 3 if smoker ==1


gen group = .
replace group = 1 if nhwhite==1 & men == 1
replace group = 2 if nhblack==1 & men == 1
replace group = 3 if nhwhite==1 & men == 0
replace group = 4 if nhblack==1 & men == 0


gen id = _n

stset age_e [pweight = mortwt], failure(mort==1) enter(time agecalc) id(id)

sts graph if age_e > 50, by(status) noorigin xscale(range(40 85))

tab group _d if age_e > 50



* Split Ages into 10yr Blocks

stsplit agecat, at(50,60,70,80)

tab agecat _d 

drop if agecat == 0

gen smkage = smoker*cage
gen smkage2 = smoker*cage2

gen fsmkage = fsmoker*cage
gen fsmkage2 = fsmoker*cage2



* pe

tab status 

* pe by age

tab agecat status 

* Confounder-specific pe with agecat

bysort group: tab agecat status  


* pd

tab status if _d == 1  

* pd by age

tab agecat status if _d == 1  

* Confounder-specific pd with agecat

bysort group: tab agecat status if _d == 1  




* Observed RRRs

tab agecat _d if nsmoker==1 
tab agecat _d if fsmoker==1 
tab agecat _d if smoker==1 
 


gen s50 = smoker*(agecat == 50)
gen s60 = smoker*(agecat == 60)
gen s70 = smoker*(agecat == 70)
gen s80 = smoker*(agecat == 80)

gen fs50 = fsmoker*(agecat == 50)
gen fs60 = fsmoker*(agecat == 60)
gen fs70 = fsmoker*(agecat == 70)
gen fs80 = fsmoker*(agecat == 80)


**** Model 1: Baseline ****
* no age included *

cloglog _d i.status

predict base_hat
tabulate status, sum(base_hat)means

**** Model 2: Age-specific ****

cloglog _d i.status#i.agecat

predict age_hat
tabulate agecat status, sum(age_hat)means


**** Model 3: Confounder ****

cloglog _d i.status#i.agecat men nhblack

**** Model 3: Confounder-specific **** 

cloglog _d i.status#i.agecat if men==1 & nhwhite==1
cloglog _d i.status#i.agecat if men==1 & nhblack==1
cloglog _d i.status#i.agecat if men==0 & nhwhite==1
cloglog _d i.status#i.agecat if men==0 & nhblack==1


* Proportion of Deaths by Race/Eth & Gender
tab group _d

* Get Proportion of TOTAL deaths (not just within-group)
tab agecat _d if group==1
tab agecat _d if group==2
tab agecat _d if group==3
tab agecat _d if group==4


**** Model 4: Age-related Bias for pd ****

xi: cloglog _d i.status*i.agecat cbirth cbirth2 cage cage2 smkage smkage2 men nhblack 

**** Model 3: Age-related Bias, Confounder-specific ****

cloglog _d i.status#i.agecat cbirth cbirth2 cage cage2 smkage smkage2  if men==1 & nhwhite==1
cloglog _d i.status#i.agecat cbirth cbirth2 cage cage2 smkage smkage2  if men==1 & nhblack==1
cloglog _d i.status#i.agecat cbirth cbirth2 cage cage2 smkage smkage2  if men==0 & nhwhite==1
cloglog _d i.status#i.agecat cbirth cbirth2 cage cage2 smkage smkage2  if men==0 & nhblack==1

***** Frailty Model *****

* REs for 5yr age groupings on Age-at-Baseline

gen a4550 = 0
replace a4550 = 1 if agecalc < 50
gen a5055 = 0
replace a5055 = 1 if agecalc >= 50 & agecalc < 55
gen a5560 = 0
replace a5560 = 1 if agecalc >= 55 & agecalc < 60
gen a6065 = 0
replace a6065 = 1 if agecalc >= 60 & agecalc < 65
gen a6570 = 0
replace a6570 = 1 if agecalc >= 65 & agecalc < 70
gen a7075 = 0
replace a7075 = 1 if agecalc >= 70 & agecalc < 75
gen a7580 = 0
replace a7580 = 1 if agecalc >= 75 & agecalc < 80
gen a8085 = 0
replace a8085 = 1 if agecalc >= 80

gen bagecat = .
replace bagecat = 1 if a4550 == 1
replace bagecat = 2 if a5055 == 1
replace bagecat = 3 if a5560 == 1
replace bagecat = 4 if a6065 == 1
replace bagecat = 5 if a6570 == 1
replace bagecat = 6 if a7075 == 1
replace bagecat = 7 if a7580 == 1
replace bagecat = 8 if a8085 == 1

* Cohort of Entry (5yr Age) stratified 

cloglog _d i.status#i.agecat if bagecat == 1
cloglog _d i.status#i.agecat if bagecat == 2
cloglog _d i.status#i.agecat if bagecat == 3
cloglog _d i.status#i.agecat if bagecat == 4
cloglog _d i.status#i.agecat if bagecat == 5
cloglog _d i.status#i.agecat if bagecat == 6
cloglog _d i.status#i.agecat if bagecat == 7



* Estimate Shared Frailty for 5-year cohorts of survey age

mecloglog _d i.agecat cbirth cbirth2 men nhblack if smoker ==1 || bagecat: 
mecloglog _d fsmoker#i.agecat cbirth cbirth2 men nhblack if smoker ==0 || bagecat: , intpoints(4)


**** Strat by Confounders ****

mecloglog _d i.agecat cbirth cbirth2  if smoker ==1 & nhwhite==1 & men==1 || bagecat: ,  intmethod(mcaghermite)
mecloglog _d fsmoker#i.agecat cbirth cbirth2  if smoker ==0 & nhwhite==1 & men==1 || bagecat: ,  intmethod(mcaghermite) intpoints(5)

mecloglog _d i.agecat cbirth cbirth2  if smoker ==1 & nhblack==1 & men==1 || bagecat: , intmethod(mcaghermite) intpoints(5)
mecloglog _d fsmoker#i.agecat cbirth cbirth2  if smoker ==0 & nhblack==1 & men==1 || bagecat: , intmethod(mcaghermite) intpoints(4)

mecloglog _d i.agecat cbirth cbirth2 if smoker ==1 & nhwhite==1 & men==0 || bagecat: , intmethod(mcaghermite) 
mecloglog _d fsmoker#i.agecat cbirth cbirth2  if smoker ==0 & nhwhite==1 & men==0 || bagecat: , intmethod(mcaghermite) intpoints(5)

mecloglog _d i.agecat cbirth cbirth2  if smoker ==1 & nhblack==1 & men==0 || bagecat: , intmethod(mcaghermite) intpoints(4)
mecloglog _d fsmoker#i.agecat cbirth cbirth2  if smoker ==0 & nhblack==1 & men==0 || bagecat: , intmethod(mcaghermite) intpoints(3)