T-Space at The University of Toronto Libraries >
School of Graduate Studies - Theses >
Please use this identifier to cite or link to this item:
|Title: ||Methods to Predict Individualized Combined Benefit/Harm Patient Profiles for Warfarin|
|Authors: ||Pereira, Jennifer|
|Advisor: ||Holbrook, Anne|
|Department: ||Pharmaceutical Sciences|
|Issue Date: ||26-Feb-2009|
|Abstract: ||Warfarin has well-proven benefit (stroke prevention) but an associated increase in harm (major bleeding) in patients with atrial fibrillation (AF). Current clinical prediction rules (CPRs) are limited in that stroke CPRs predict only the probabilities of “stroke” and “no stroke” and bleeding CPRs predict only “bleed” and “no bleed” despite the fact that outcomes actually include combinations of these four groups. The study objective was to evaluate methods to create a CPR that calculates individual patient probabilities of warfarin’s four combined benefit/harm outcome groups: i) no stroke/no bleed; ii) no stroke/bleed; iii) stroke/no bleed; iv) stroke/bleed.
Methods: Patient-level data were analyzed from a randomized controlled trial database (n=9,155) and an observational anticoagulant clinic database (n=5,475) from start of trial or time of AF diagnosis respectively (baseline), until end of follow-up. Patients were stratified into the four groups based on their outcomes during follow-up. Due to high mortality in both datasets, death was included as an outcome. Decision tree modeling and polytomous logistic regression (PLR) were conducted to identify baseline patient factors predicting each outcome group.
Results: Based on a literature review of recent high quality RCTs, benefit and harm are reported separately and not at a more individualized level than subgroup analysis. In this individualized combined benefit/harm analysis, both PLR and decision tree modeling identified predictors of no stroke/no bleed, no stroke/bleed, stroke/no bleed and death without a prior stroke or bleed. PLR results predicted probabilities of combined benefit/harm outcomes for every patient but required detailed computation. However, results could potentially be converted into automated form for ease of use. Decision trees provided a visual algorithm approach to risk assessment but did not i) predict the probability of warfarin’s combined benefit/harm outcomes based on all predictors simultaneously, ii) predict the probability of these outcomes for every patient or iii) provide statistical parameters of predictive value (odds ratios).
Conclusions: The PLR technique could be used to predict patient probabilities of combined benefit/harm outcomes with warfarin. The study results require validation, preferably prospectively, in other cohorts. If validated, this approach should be tested to determine if it aids patient decision-making.|
|Appears in Collections:||Doctoral|
Leslie L. Dan Faculty of Pharmacy - Doctoral theses
Items in T-Space are protected by copyright, with all rights reserved, unless otherwise indicated.