Re: st: Carry forward an observation within a time frame

[David Kantor <kantor.d@att.net>]

Dear Benigno,

At 10:45 AM 10/9/2013, you wrote:
> Dear David:
>
> Thank you for this detailed response. But my problem is that I need 
> to carry forward to a given row with a missing CD4 the closest in 
> time CD4 value, as long as it is within the previous 4 months (the 
> last 120 days is enough precision for my purposes).
>
> Correct me if I'm thinking about this wrong, but with your approach, 
> a value could be propagated down even if the "seed" for that value 
> occurred more than 4 months earlier. I have added a few rows 
> (artificially) to illustrate. Look at row 10. It occurs within 4 
> month of row[_n-1], but I don't want it filled out with that value, 
> because the last time we actually had a non-missing value was in row 
> 3, which is more than 4 months prior to row 10.
>
> The approach I was trying was to try to define a "spell" as one that 
> begins with a missing CD4 AND for which the row above has a 
> non-missing value (easily done). But implementing the solution from 
> there has proven challenging.
>
>
>         patid   date            CD4     desired
> 1.      1007    5-May-55        .       .
> 2.      1007    1-Jan-00        .       .
> 3.      1007    3-Apr-02        5       5
> 4.      1007    8-Apr-02        .       5
> 5.      1007    11-Apr-02       .       5
> 6.      1007    13-May-02       .       5
> 7.      1007    30-May-02       .       5
> 8.      1007    30 Jun-02       .       5
> 9.      1007    30-Jul-02       .       5
> 10.     1007    30-Aug-02       .       .
> [...]

Indeed, that is a slightly more-involved problem than I understood; I 
didn't read your original question with adequate attention. You don't 
want to carry a value too long; it should expire after 120 days.
Try this:
gen int seeddate = date if ~mi(CD4)
by patid (date): carryforward seeddate, replace
by patid (date): carryforward CD4 if date-seeddate < 120, replace /* 
or gen(newvar) */

HTH
--David

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/