BACKGROUND: The relationship of risk factors to outcomes has traditionally been assessed by measures of association such as odds ratio or hazard ratio and their statistical significance from an adjusted model. However, a strong, highly significant association does not guarantee a gain in stratification capacity. Using recently developed model performance indices, we evaluated the incremental discriminatory power of individual and neighborhood socioeconomic status (SES) measures after myocardial infarction (MI). METHODS: Consecutive patients aged ≤65 years (N=1178) discharged from 8 hospitals in central Israel after incident MI in 1992 to 1993 were followed-up through 2005. A basic model (demographic variables, traditional cardiovascular risk factors, and disease severity indicators) was compared with an extended model including SES measures (education, income, employment, living with a steady partner, and neighborhood SES) in terms of Harrell c statistic, integrated discrimination improvement (IDI), and net reclassification improvement (NRI). RESULTS: During the 13-year follow-up, 326 (28%) patients died. Cox proportional hazards models showed that all SES measures were significantly and independently associated with mortality. Furthermore, compared with the basic model, the extended model yielded substantial gains (all P<0.001) in c statistic (0.723 to 0.757), NRI (15.2%), IDI (5.9%), and relative IDI (32%). Improvement was observed both for sensitivity (classification of events) and specificity (classification of nonevents). CONCLUSIONS: This study illustrates the additional insights that can be gained from considering the IDI and NRI measures of model performance and suggests that, among community patients with incident MI, incorporating SES measures into a clinical-based model substantially improves long-term mortality risk prediction.
BACKGROUND: The relationship of risk factors to outcomes has traditionally been assessed by measures of association such as odds ratio or hazard ratio and their statistical significance from an adjusted model. However, a strong, highly significant association does not guarantee a gain in stratification capacity. Using recently developed model performance indices, we evaluated the incremental discriminatory power of individual and neighborhood socioeconomic status (SES) measures after myocardial infarction (MI). METHODS: Consecutive patients aged ≤65 years (N=1178) discharged from 8 hospitals in central Israel after incident MI in 1992 to 1993 were followed-up through 2005. A basic model (demographic variables, traditional cardiovascular risk factors, and disease severity indicators) was compared with an extended model including SES measures (education, income, employment, living with a steady partner, and neighborhood SES) in terms of Harrell c statistic, integrated discrimination improvement (IDI), and net reclassification improvement (NRI). RESULTS: During the 13-year follow-up, 326 (28%) patients died. Cox proportional hazards models showed that all SES measures were significantly and independently associated with mortality. Furthermore, compared with the basic model, the extended model yielded substantial gains (all P<0.001) in c statistic (0.723 to 0.757), NRI (15.2%), IDI (5.9%), and relative IDI (32%). Improvement was observed both for sensitivity (classification of events) and specificity (classification of nonevents). CONCLUSIONS: This study illustrates the additional insights that can be gained from considering the IDI and NRI measures of model performance and suggests that, among community patients with incident MI, incorporating SES measures into a clinical-based model substantially improves long-term mortality risk prediction.
Authors: John M Brooks; Yuexin Tang; Cole G Chapman; Elizabeth A Cook; Elizabeth A Chrischilles Journal: J Clin Epidemiol Date: 2013-08 Impact factor: 6.437
Authors: Merete Osler; Eva Prescott; Ida Kim Wium-Andersen; Else Helene Ibfelt; Martin Balslev Jørgensen; Per Kragh Andersen; Terese Sara Høj Jørgensen; Marie Kim Wium-Andersen; Solvej Mårtensson Journal: PLoS One Date: 2015-10-29 Impact factor: 3.240
Authors: Marquita W Lewis; Yulia Khodneva; Nicole Redmond; Raegan W Durant; Suzanne E Judd; Larrell L Wilkinson; Virginia J Howard; Monika M Safford Journal: BMC Public Health Date: 2015-12-29 Impact factor: 3.295
Authors: Inge Kirchberger; Christa Meisinger; Hildegard Golüke; Margit Heier; Bernhard Kuch; Annette Peters; Philip A Quinones; Wolfgang von Scheidt; Andreas Mielck Journal: Int J Equity Health Date: 2014-02-19