/* 

   Predict EQ-5D from SF-12 summary scores
    
   September 3, 2015	

   This code uses the parameters presented in Table 2 of 

   Perraillon MC, Ya-Chen Tina Shih, Ronald A. Thisted. "Predicting the EQ-5D Preference Index        from the SF-12 Health Survey: A finite Mixture Approach.", Medical Decision Making, October        2015 vol. 35 no. 7 888-901.

   Inputs: age (in years), and SF-12 MCS and PCS scores. 

*/ 



* Center covariates
gen mcsc = mcs - 50
gen pcsc = pcs - 50 
gen agec = age - 65 

* Prediction of latent normal
gen u1 = .2308311 -.0020312*mcsc -.0045566*pcsc
gen u2 = .6449207 -.0046078*mcsc -.0095772*pcsc

* Censored predictions
gen u1c = normal(u1/.0547989)*( ///
          u1+.0547989*normalden(u1/.0547989)/normal(u1/.0547989)) 
gen u2c = normal(u2/.1521847)*( ///
          u2+.1521847*normalden(u2/.1521847)/normal(u2/.1521847)) 
		  
* Probabilities
gen m1 =  1.563787  -.1483147*mcsc  -.2152055*pcsc + .0263548*agec  
gen m2 = -1.954959 -.2490964*mcsc -.3208357 *pcsc + .0212149 *agec 		  

gen p1 = exp(m1) / (1+exp(m1)+ exp(m2)) 
gen p2 = exp(m2) / (1+exp(m1)+ exp(m2))
gen p3 = 1 - p1 - p2

* Classify
egen     _pmax = rowmax(p1 p2 p3)
gen      _class = 0 if _pmax == p3
replace  _class = 1 if _pmax == p1
replace  _class = 2 if _pmax == p2

gen     _eqhat = 0  if  _class==0
replace _eqhat = u1c if _class==1
replace _eqhat = u2c if _class==2

gen eq5d = 1 - _eqhat

drop u1 u2 u1c u2c m1 m2 p1 p2 p3 _*