Article type
Year
Abstract
Introduction: Some examples have shown how the availability of individual patient data in systematic reviews could modify the treatment effect estimate. The enhancement of data reliability is an obvious advantage of this approach. We present here some examples of other advantages, linked with the possibility of multivariate or survival modelling.
Methods/Results: We collected data from 11 randomised controlled trials that assessed antihypertensive drug interventions on cardiovascular morbi-mortality. The corresponding data base concerns 53,799 hypertensive people included between 1972 and 1990, cumulates more than 270,000 person years of follow-up, during which 5,487 major clinical events have been observed. Example l: We explored if treatment effect, expressed as hazard ratio, was constant over time. Significant variations were observed, which were in opposite directions than what was expected a priori: the treatment effect significantly increased with time in stroke, whereas it decreased in coronary events. Example 2: We are currently exploring if baseline characteristics of patients significantly changed the relative risk due to treatment. Preliminary results do not suggest important or significant changes. Two important consequences can be expected: first , absolute risk reduction can be directly extrapolated from applying relative risk to the untreated risk predicted for the person to treat; and second, treatment may be beneficial to high risk people, irrespective of their baseline blood pressure. Example 3: We explored to which extent the blood pressure reduction explained the stroke risk reduction. Surprisingly, the first approach we used suggested that blood pressure reduction explained only a small part of the treatment effect, what is in contradiction with results obtained on summarised tabulated data. Example 4: We assessed the prognostic value of baseline characteristics of individuals, on the same population as this in whom the treatment effect has been assessed. In addition to the classical ones, glomerular filtration rate and height were found to be independent cardiovascular risk factors, and should be incorporated in cardiovascular risk scoring algorithms. Exploring heterogeneity between trials allowed us to assess the generalizability of results.
Discussion: In addition to the improvement of data quality and the development of collaboration, obtaining individual data in a systematic review of clinical trials allowed us both to suggest that the usual therapeutic model may not be valid, and to develop hypotheses to decrease the distance between treatment target population and the population of treatment dissemination.
Methods/Results: We collected data from 11 randomised controlled trials that assessed antihypertensive drug interventions on cardiovascular morbi-mortality. The corresponding data base concerns 53,799 hypertensive people included between 1972 and 1990, cumulates more than 270,000 person years of follow-up, during which 5,487 major clinical events have been observed. Example l: We explored if treatment effect, expressed as hazard ratio, was constant over time. Significant variations were observed, which were in opposite directions than what was expected a priori: the treatment effect significantly increased with time in stroke, whereas it decreased in coronary events. Example 2: We are currently exploring if baseline characteristics of patients significantly changed the relative risk due to treatment. Preliminary results do not suggest important or significant changes. Two important consequences can be expected: first , absolute risk reduction can be directly extrapolated from applying relative risk to the untreated risk predicted for the person to treat; and second, treatment may be beneficial to high risk people, irrespective of their baseline blood pressure. Example 3: We explored to which extent the blood pressure reduction explained the stroke risk reduction. Surprisingly, the first approach we used suggested that blood pressure reduction explained only a small part of the treatment effect, what is in contradiction with results obtained on summarised tabulated data. Example 4: We assessed the prognostic value of baseline characteristics of individuals, on the same population as this in whom the treatment effect has been assessed. In addition to the classical ones, glomerular filtration rate and height were found to be independent cardiovascular risk factors, and should be incorporated in cardiovascular risk scoring algorithms. Exploring heterogeneity between trials allowed us to assess the generalizability of results.
Discussion: In addition to the improvement of data quality and the development of collaboration, obtaining individual data in a systematic review of clinical trials allowed us both to suggest that the usual therapeutic model may not be valid, and to develop hypotheses to decrease the distance between treatment target population and the population of treatment dissemination.