0:00:02i'm l o'clock or like to tell you about work the colleagues and i have
0:00:05done to identify when large medical data bases are suitable places to conduct comparative effectiveness
0:00:12research
0:00:14the introduce some graphics
0:00:16a user capital sigma to designate a provider it's as a nation side because doctors
0:00:21add up data and come to decisions
0:00:24a question mark a device a patient
0:00:28in american cancer patient comes to something that he wants fix
0:00:31a doctor brings experience in training
0:00:34the doctor elicits information
0:00:37the doctor uses all the available information
0:00:41the country treatment decisions
0:00:43adaptation can elsewhere a different doctor mitre brought different training and experience to bear because
0:00:49to a different decision
0:00:51another doctor might have made yet another decision
0:00:54and so on
0:00:55let's imagine for simplicity that all transcribers choose one of two treatments
0:01:00the same patient stands an equal chance of getting treatment a or treatment be
0:01:05depending only on which doctor you happens to visit treating community is in the corpus
0:01:11our analysis screens population medical data for a coke voice
0:01:14we propose that where we find it we found the suitable place for doing comparative
0:01:18affection of this research
0:01:22these graphs illustrate the technique with community acquired pneumonia
0:01:25the x-axis or patients according to their probability of getting the treatment shown in that
0:01:30this is called a preference score
0:01:32the x-axis to pick seventy patients there are at each preference score level
0:01:36the two lines each crap represent the patient preference scores for the two three bits
0:01:40exam
0:01:42the left and graph comparison use only fluxes in is it true motion at different
0:01:46preference scores
0:01:47at the midpoint of preference we have patients room that reading communities in perfect a
0:01:51corpus
0:01:52how to either side
0:01:54we can skip channels about the vehicle points
0:01:56so it should what lots of getting one treatment or the other
0:02:00among patients receive closer mizuno marks of lexus and fewer fell into this range of
0:02:05wealth of the corpus
0:02:07if not patient characteristics predict treatment
0:02:10then the to treatment groups will be alike
0:02:13here's a situation that leave a fluxes and is it from us in patients
0:02:17the two groups are very similar
0:02:19okay treatment failure
0:02:21the patients receive leave of losses and had fewer failures
0:02:24they did better
0:02:26this is finding deserve ad hoc research we believe that that's
0:02:31our tools identified one treatment decision for which at least one prescribe or community seems
0:02:36to be of divided opinion
0:02:38i difference of twenty percent improvement failure rates would be important of true
0:02:43the tools shown that so this population is suitable for comparative effectiveness research in that
0:02:48many patient characteristics are balanced and so we're not confounded
0:02:53but patients are more the totality of the recorded characteristics
0:02:58two patients to appear to be similar with respect to the recorded characteristics differ with
0:03:02respect to their unrecorded characteristics
0:03:06proper research requires that all important covariance the measured accounted for
0:03:12keepers of large medical databases can easily apply our tool to identify situations of the
0:03:16parrot empirical a corpus
0:03:19we can now when compared of effectiveness research might be successful
0:03:24and when we should point
0:03:27thank you for your attention