Please activate JavaScript!
Please install Adobe Flash Player, click here for download


ISCB 2014 Vienna, Austria • Abstracts - Poster Presentations 119Wednesday, 27th August 2014 • 11:00-11:30 Monday25thAugustTuesday26thAugustThursday28thAugustAuthorIndexPostersWednesday27thAugustSunday24thAugust P3.2.101 Smooth time-dependent ROC curve estimators P Martínez-Camblor1 , JC Pardo-Fernández2 1 OIB-FICYT, Oviedo, Spain, 2 Universidad de Vigo, Vigo, Spain   The ROC curve is a popular graphical method frequently used in order to study the diagnostic capacity of continuous (bio)markers. When the con- sidered outcome is a time-dependent variable, two main extensions have been proposed: the cumulative/dynamic and the incidence/dynamic ROC curves. In both cases, the principal problem for developing appropriate estimators is the estimation of the joint, time-to-event and marker, dis- tribution. As usual, different approximations lead to different estimators. In this work, the authors explore the use of a bivariate kernel density estimator with this goal. The performance of the cumulative/dynamic and the inci- dence/dynamic versions of the time-dependent ROC curve is studied from Monte Carlo simulations. In addition, the influence of the bandwidth se- lection on the obtained results and the use of different indices to measure the global diagnostic capacity are also studied. Finally, some real-world applications are reported. Results suggest that the smooth estimators provide good approximations, in special, when the area under the ROC curve is not too large. As usual, the main handicap of this technique is the impact of the used bandwidth on the obtained estimations. A reasonable empirical rule to choose this parameter is also proposed.   P3.2.103 Seasonality in testing for systemic lupus erythematosis EJ McKinnon1 , M John2 1 Murdoch University, Perth, Australia, 2 Royal Perth Hospital, Perth, Australia   Systemic lupus erythematosis (SLE) is a chronic autoimmune disease that can affect most of the organ systems of the body. The disease generally follows a relapsing and remitting course and is characterized by a range of symptoms including lethargy, fever, rashes and muscle and joint aches. There is no gold standard diagnostic test for SLE, but in practice a positive diagnosis typically comes after clinical assessment combined with a series of laboratory tests. These begin with an initial screen to detect the anti- nuclear antibodies (ANA) that mark disease activity of SLE and other rheu- matic conditions. Choice of follow-up testing to confirm specificity of the antibodies is guided by observed fluorescence patterns, and include those based on detecting antibodies to extractable nuclear antigens (ENA) or anti-double-stranded DNA (anti-dsDNA). Here we explore patterns of seasonality in ANA screening/monitoring, and investigate how they translate to follow-up confirmatory testing. Analysis is based on sequential test results from a large state-wide laboratory da- tabase. For each test type, numbers of results per individual are quite vari- able (range ANA: 1-19; ENA: 1-14; anti-dsDNA: 1-38) and heavily skewed, with only a minority of patients having multiple measures (ANA: 16%, ENA: 10%, anti-dsDNA: 9%). In this presentation we will contrast inferences obtained from several methodological approaches that differ in how they take account of the between-individual variability in testing frequency.   P3.2.117 Clinical factors affecting bias between different eGFR measurements based on the weighted Deming regression AJ Owczarek1 , K Wieczorkowska-Tobis2 , A Skalska3 , A Więcek4 , J Chudek5 1 Division of Statistics, Medical University of Silesia, Sosnowiec, Poland, 2 3 Dept. of Geriatric Med. & Gerontology, Univ. of Med. Sci., Poznan, Poland, 3 6 Dept. of Int. Med. and Geront., Jagiel. Univ. Med. Col., Cracow, Poland, 4 Dept. of Nephr., Endocr. & Met. Dis., Med. Univ. of Silesia, Katowice, Poland, 5 Dept. of Pathophysiology, Medical University of Silesia, Sosnowiec, Poland   Six different eGFR calculation methods were done in 3503 subjects elder than 65. eGFR based on full MDRD formula was chosen as a gold standard. According to intraclass correlation coefficient, through the chronic kidney diseases classes, the most compatible formulas were as follow: CKD-EPI (ICC=0.88), short MDRD (0.76), Cystatine-Creatynine (0.72), Cockroft-Gault (0.57), Hoek (0.52). The highest overestimation occurred in C-G (44.3%), while the lowest in MDRD short formula (0.1%). The highest underestima- tion occurred in Hoek formula (45.2%), while the lowest in the CKD-EPI formula (13.3%). Results of agreement were presented using the agreement chart. Next the bias in corresponding five weighted Deming regression models were calculated, for patients with eGFR lower than 60. It has been shown, that diabetes, hypertension, heart failure, stroke, obesity and immobility were factors that affect the bias effect. Conclusions: It has been shown, that different formulas to calculate esti- mated glomerular filtration rate have different level of bias in the presence of important clinical factors.   P3.2.153 When do latent class models outperform an imperfect gold standard? A problem revisited MR Oliveira1 , A Subtil2 1 Instituto Superior Técnico, Universidade de Lisboa, Lisboa, Portugal, 2 Faculdade de Ciências, Universidade de Lisboa, Lisboa, Portugal   The dynamical biomedical sciences and pharmaceutical industry steadily seek to produce new diagnostic tools. Yet, new tests should only be intro- duced into medical practice after its clinical value is thoroughly evaluated, including the test’s ability to correctly identify diseased and nondiseased patients. Standard performance measures, can be estimated by compari- son with a gold standard test. Since such perfect reference test is frequent- ly unavailable, alternative approaches are needed. An available test perceived as the best one can be used as an imperfect reference test, against which the new test is compared. However, it is known that the imperfect reference, in general, leads to biased estimates. Latent class models (LCM) provide an alternative approach for this prob- lem. A widely used LCM admits a binary latent variable, that indicates the disease status, and manifest binary variables, that express the tests results. This LCM assumes that the test results are independent conditional on the disease state, which may fail in practice and can result in substantial bias. In this work, for the special case of 3 tests, we compare the LCM’s estima- tors of performance measures with alternative estimators. In contrast with simulated comparisons, we take the theoretical viewpoint, based on the estimators analytical forms. In the absence of a gold standard, LCM create a consensual “gold stan- dard”, based on the multiple test results, which can be used to classify patients as diseased or nondiseased. We discuss, from the theoretical perspective, the validity and potential usefulness of this classification as a clinical diagnostic tool.

Pages Overview