Article
Predictive model for chronic venous insufficiency
Search Medline for
Authors
Published: | September 20, 2011 |
---|
Outline
Text
Background: Venous disorders like varicosis and chronic venous insufficiency (CVI) are among the most common diseases in western populations. Little population-based data on incidence of venous diseases is available. Our study investigates the ability of factors, like age, sex, bmi, phlebological symptoms, lifestyle etc. to predict the incidence of CVI during a 7-year follow-up.
Material and methods: Data are derived from the Bonn Vein Study, a population-based cohort study (2000-2008) which included 3,072 participants between 18 and 79 years, residing in Bonn and two rural adjacent townships. All participants filled in a questionnaire regarding socio-economic status, lifestyle, medical history and quality of life. Trained examiners conducted an interview concerning the phlebological history and performed a clinical examination including duplex-sonography for reflux. CVI was defined by CEAP clinical stages C3 to C6. The development of the predictive model included a univariate analysis to select factors significantly associated with CVI, followed by a backward selection based on AIC. Since an independent data set was not available for validation and investigation of generalizability, a five-fold cross-validation was conducted. The validation results, such as area under the curve, serve as a measurement for the goodness of prediction in practice.
Results: Response at follow-up after a mean of 6.6 years was 84.6%. 1,617 participants showed no signs of CVI at baseline. At follow-up, 216 cases of CVI were diagnosed. For CVI age, bmi, systolic blood pressure, childbearing, chronic diseases, sitting/standing at work, a reflux in superficial veins, varicosis, corona phlebectatica paraplantaris and phlebothrombotical diseases were identified as predictive factors. The prediction of incidence was satisfying with an area under the curve of 0.773. Results from cross-validation yielded to a smaller area under the curve of 0.750. An additional analysis, examining the role of reflux for prediciton, showed, that screening for reflux did not improve the prediction model for CVI.
Discussion: This is the first attempt to develop a predicition model for the incidence of CVI. The predictive model performed moderately well in this data set; however a validation in an independent sample is necessary before any recommendations for clinical applications can be made.