I am looking for an approach, be it by simulation or other, that will help me address a question.
As background, a recent systematic review has shown that the mortality rate in large randomised clinical trials (RCTs) of dialysis patients (analysing 79,104 participating patients) was only half of that in the population-based registry data (i.e. the source population of "real world patients" from which they were drawn) to which they were compared. These mortality rates were 8.9 versus 18.6 deaths per 100 patient years, respectively.
I have a population-based -stset- dataset from my own country, with a mortality rate of 15.9 per 100 patient years.
I'm looking determine the rough proportion of patients in my own country for whom the RCT findings would be generalisable. To do this, I was looking to see in a hypothetic manner what proportion of patients in my dataset have the same mortality risk profile as that of the RCT participants in those trials - i.e. they have a mortality rate that is ~50% of the group average, which in my case amounts to an average mortality rate of around 7.6 deaths per 100 patient years).
I was hoping there was a smart way of doing it with a sampling method that produces a subsample that is a reasonable reflection of reality. Obviously, knocking out higher mortality rate patients until the overall - strate - comes down to 7.6 produces weird looking survival curves.
Any advice?
Thanks
MM
HTML Code:
. strate failure _d: dead == 1 analysis time _t: (end_recorddate-origin)/365.25 origin: time rxdate exit on or before: time exitdate id: mrm_patient Estimated rates and lower/upper bounds of 95% confidence intervals (27944 records included in the analysis) +----------------------------------------------+ | D Y Rate Lower Upper | |----------------------------------------------| | 5211 3.3e+04 0.15928 0.15501 0.16366 | +----------------------------------------------+
HTML Code:
. stdes failure _d: dead == 1 analysis time _t: (end_recorddate-origin)/365.25 origin: time rxdate exit on or before: time exitdate id: mrm_patient |-------------- per subject --------------| Category total mean min median max ------------------------------------------------------------------------------ no. of subjects 9387 no. of records 27944 2.976883 1 3 18 (first) entry time .0010083 0 0 2.023272 (final) exit time 3.488286 .0027379 2.740589 19.88501 subjects with gap 30 time on gap if gap 18.354552 .6118184 .0027379 .1943874 5.585216 time at risk 32716.723 3.485323 .0027379 2.737851 19.88501 failures 5211 .5551294 0 1 1 ------------------------------------------------------------------------------
0 Response to Selecting a sub-sample of a survival dataset, such that the mortality rate (-strate-) in the selected sub-sample achieves a certain target
Post a Comment