‘Tailored’ risk model developed for stroke
MedWire News: Researchers have devised a model to estimate the 10-year risk for developing stroke using data from nine European countries and incorporating competing risks.
The stepwise model showed good accuracy even in its simplest form, with more clinical information, meaning it could be "tailored for different countries and different settings according to data and resources available," say Anders Borglykke (Glostrup University Hospital, Denmark) and colleagues.
The study, reported in the journal Heart, used baseline and follow-up data on fatal and nonfatal stroke from 16 centers in nine different European countries from the MOnica Risk, Genetics, Archiving, and Monograph (MORGAM) collaboration.
The researchers developed a Cox regression model in a stepwise manner, first using only data from questionnaires, then adding data from physical examinations, and finally including data from blood samples.
The model combined age as an underlying time scale with the competing risk of dying without preceding stroke events - an advantage that makes it "possible to estimate the risk of an event given that you do not die before the event," Borglykke et al explain.
During 1,176,296 years of observation, 2928 incident fatal and nonfatal stroke events were registered. The model showed good calibration and accuracy of prediction, as measured by the Hosmer-Lemeshow test and the Brier score, respectively.
Discriminative ability, measured as the area under the receiver operating characteristic curve (AUC), varied between country- and gender-specific models.
The lowest AUC was seen in the model containing only data from blood samples (0.77 men; 0.75 women) and highest in the model containing all three steps (0.79 men; 0.80 women).
When looking at the country-specific models, a high AUC of around 0.8 or higher was seen for Finland, Sweden, Russia, and Italy, and all the country-specific models had acceptable discriminative ability, with France and Lithuania as the lowest with an AUC of around 0.7.
The researchers comment: "This is of real public health importance since it is possible to develop simple models with acceptable performance and use these in low resource healthcare settings."
MedWire (www.medwire-news.md) is an independent clinical news service provided by Current Medicine Group, a trading division of Springer Healthcare Limited. © Springer Healthcare Ltd; 2010
By Andrew Czyzewski