Am I correct in assuming that "conditioned on surviving to age 65" is what you mean by left-censored?   If so then "the estimated probability of surviving to e.g. age 70  conditioned on surviving to age 65" is, iirc, is the very definition of the 5-year hazard rate (or actually 1 minus the 5-year hazard rate) for 65 to 70 period, which I'm guessing can be produced by LIFETEST.  And even if LIFETEST doesn't produce something literally called hazard-rate, isn't what you want equivalent to the probability of surviving to 70 divided by the probability of surviving to 65?  Those values would certainly be generated by LIFETEST, no?   It's been a long time since I was exposed to survival analysis, so I admit to being a bit insecure about these statements.     ... View more

So how do expect to include the PersonID that have neither Y or N for the smoke variable? What years to expect the N for smoke variable to represent? If we treat all of them as years 1940 to 2009 we may well be including years that individual was not alive. Or including smoking in the cradle...   If your example data is representative of your full 100,000+ records you appear to have years of information for about 11% of the records.   If you had a more complete data set I might do something like this to get one record per year: data example; input personid $ sex $ smoke $ startyear endyear; datalines; 1 F Y 1960 1969 1 F N 1970 1995 2 M Y 1945 1986 3 M Y 1954 1973 3 M N 1974 1998 4 F Y 1958 1988 5 M Y 1961 1989 ; run; data want; set example; do year=startyear to endyear; output; end; drop startyear endyear; run; proc freq data=want; tables year*smoke; run; Which would allow a number of different types of analysis. You could also get some graphs with data summarized on year for counts or percentages of interest.   ... View more