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

ISCB2014_abstract_book

120 ISCB 2014 Vienna, Austria • Abstracts - Poster PresentationsWednesday, 27th August 2014 • 11:00-11:30 Monday25thAugustTuesday26thAugustThursday28thAugustAuthorIndexPostersWednesday27thAugustSunday24thAugust P3.2.154 Bayesian latent class models for the evaluation of diagnostic tests in multiple populations A Subtil1,2 , PZ Bermudez1,2 , L Gonçalves2,3 1 Faculdade de Ciências, Universidade de Lisboa, Lisboa, Portugal, 2 CEAUL, Lisboa, Portugal, 3 IHMT, Universidade Nova de Lisboa, Lisboa, Portugal   The evaluation of a diagnostic test’s ability to correctly discern between diseased and non-diseased individuals is crucial to establish the test’s clinical relevance and practical utility. In many situations, the true disease state of the individuals is unknown, because it is not possible to apply a perfect reference test (gold standard). In such cases, latent class models are often used to estimate diagnostic tests performance measures, such as sensitivity and specificity, as well as the disease’s prevalence. In this work, we look into situations where, under the absence of a gold standard, multiple diagnostic tests are applied to multiple subpopula- tions, admitting that dependencies between these subpopulations may exist. Plausible dependencies between prevalences may arise, for in- stance, between parents and offspring. Distinguishing population subgroups allows differentiating prevalences and the tests’ performance measures, and thereby further detailing the case under study. While stratified sampling naturally defines subpopula- tions, it may also be acceptable to artificially construct populations with a practical meaning, when an appropriate sampling scheme is missing. A bayesian approach can be particularly relevant in complex settings, with multiple populations and covariates. Furthermore, it allows for the introduction of prior information, such as experts opinions or findings from previous studies, which may improve the inferences and avoid non- identifiability. We explore and compare alternative bayesian latent class models with dif- ferent strategies to model dependencies between subpopulations.   P3.3 Analysis of electronic health records P3.3.49 A validation algorithm for detecting dose increase from longitudinal data of psychotropic drug users, using Monte Carlo simulation F Feuillet1,2 , C Victorri-Vigneau1,3 , J-B Hardouin1,2 , V Sébille1,2 1 EA4275, University of Nantes, Nantes, France, 2 Nantes University Hospital - Biometric Department, Nantes, France, 3 CEIP, Nantes University Hospital, Nantes, France   Introduction: Several methods have been recently developed from French National Insurance Health System (IHS) database concerning prob- lems in drug use (misuse, addiction). No valid indicators are available to characterize dose increase which could reveal drug inefficiency or com- pulsive use. Objective: To validate properties of an algorithm for detecting dose in- crease from longitudinal data (sensitivity, specificity). Methods: Moving average method was applied for detecting dose in- crease. Several steps were defined for the algorithm: 1) defining a refer- ence dose 2) calculating average doses for each drug delivery 3) compar- ing each dose with the reference dose, according to a defined detection threshold depending on dosage of the study drug. Monte Carlo simulations were used to vary different parameters. Population parameters: study duration, increase rate and increase dura- tion. Algorithm parameters: moving average method (one to four-period) and detection threshold. Results: 1 million patients per dataset were simulated. For a dataset with duration of 12 months and an increase rate of 50%, algorithm with two- period moving average method achieves a sensitivity of 76% and a speci- ficity of 94%. When study duration increases, specificity decreases and sensitivity increases. The one-period moving average method is sensitive (90%) but less specific (64%). Conversely, the four-period moving average method is not very sensitive (10%) but highly specific (100%). A high de- tection threshold results in a poor sensitivity of the algorithm. Conclusion: This algorithm for detecting dose increase has good proper- ties. The two-period moving average method optimizes properties of the algorithm. Usage recommendations may be proposed based on study ob- jectives (population, pharmacological class, potential drug dependence...). P3.3.96 How much of socioeconomic differences in breast cancer patient survival can be explained by stage at diagnosis and treatment? R Li1 , R Daniel1 , B Rachet1 1 London School of Hygiene & Tropical Medicine, London, United Kingdom   Socioeconomic inequalities in breast cancer survival persist in England. The main contributing factors could be presentation at different stages and variation in access to treatment. Information on 36,793 women diagnosed with breast cancer during 2000- 2007 was routinely collected by an English population-based cancer reg- istry. Surgical treatment information from Hospital Episode Statistics was dichotomised into “major” versus “minor or no procedures”. A deprivation category was allocated according to each patient’s area of residence at diagnosis. We estimated the proportion of the effect of deprivation on short-term survival mediated by stage and by treatment using G-computation pro- cedures. Single stochastic imputation was incorporated to handle missing stage (8%). Net survival differed between the most affluent and most deprived pa- tients at one year (97% vs 94%), and at five years (86% vs 76%) after di- agnosis. Adverse stage distribution was associated with more deprived patients (p<0.01). The more advanced the stage at diagnosis, the less likely the pa- tient was to receive major surgical treatment (p<0.01). The most deprived patients were almost three times more likely to die within six months after diagnosis than the most affluent (OR: 2.77 [2.17- 3.53]). One third of this excess mortality was mediated by adverse stage distribution whilst none was mediated through differential surgical treat- ment. Our results showed that the effort to advance the diagnoses is important, but would reduce the socio-economic inequalities in cancer survival only by a third. We did not have reliable information on comorbidity, which could be another mediator on the causal pathway.   P3.3.152 Adjustment for hidden confounding in the analysis of pneumococcal vaccination effectiveness using electronic health records AJ Streeter1 , A Ble1 , J Foster1 , D Melzer1 , WE Henley1 1 University of Exeter Medical School, Exeter, United Kingdom   Vaccination against pneumococcal infection is currently recommended for adults aged over 65y in the UK. However the practical and ethical dif- ficulties in conducting trials in this age group limit the evidence for this policy. Observational studies based on analysis of routinely collected pa- tient records provide an alternative source of information for evaluating effectiveness of the vaccine in the population, away from the ideal envi- ronment of the clinical trial. A major challenge in adopting this approach is addressing the potential for bias due to hidden confounding. We used a quasi-experimental approach to estimate vaccine effectiveness

Pages Overview