256 



Where diseases had joint multiple PARs for different substances (smoking and alcohol), the 

 alcohol PAR was weighted by a factor of 0.5. 



For the Medicare population, we used a prevalence of 9% for heavy drinking based 

 on studies of drinking in the hospitalized elderly population. The prevalence of drug use was 

 obtained from the 1991 National Household Drug Survey, and smoking from the 1987 National 

 Medical Expenditures Survey. For most smoking-related diseases, we calculated PARs using 

 prevalences for current and former smokers with their respective relative risks (see TABLE 4 

 in Results). However, for malignant neoplastic disease, some scientific evidence suggests that 

 once smokers have reached a certain threshold of smoking (more than 1 cigarettes per day 

 for more than 35 years), their relative risk is not diminished by cessation.'" Smoking over a 

 long period of time may have an irreversible oncogenic effect which is not altered by 

 quitting.'"^ For the Medicare population over 65 we found that 32.7% of former smokers 

 meet the criterion of having smoked more than 10 cigarettes per day for more than 35 years. 

 For this reason, we considered this subset of former smokers to be equivalent to current 

 smokers in the PAR calculations for cancer. 



Once PARs were computed for all diseases and conditions, 1CD-9-CM codes were 

 matched to the general diagnostic categories used in much of the epidemiologic literature. For 

 example, the lung cancer category included lCD-9 codes 162. 2-. 9 (malignant neoplasms of the 

 bronchus and lung). However, if the ICD-9 codes were not specifically identified in the 

 original study, with the assistance of a medical coder and several physician consultants, we 

 selected ICD codes that fell into the general disease classification and then matched the PARs 

 for that disease category with the associated ICD-9 codes (see Appendix II). 



14 



