REPRODUCIBILITY OF RESULTS IN THE GENETICS OF PREDISPOSITIONS AND THEIR PREDICTIVE VALUES
Poor reproducibility and low predictive values of the results in the genetics of predispositions become a systemic problem. Results of the
statistical quality control of genetic tests in the study should be supported with not only integral indices such as odds ratios (OR), but with the
post-test (posterior) predictive probabilities (PPV and NPV) and likelihood ratios (LR[+] and LR[-]). Usefulness of predictiveness graphs
for visualization of the relationships between the prevalence as a pretest (prior) probability of disease and predictive valuesPPV and NPV as
posttest (posterior) probabilities is demonstrated. Predictive capabilities of widely used genetic, observational, instrumentaland immunological
diagnostic tests are discussed. Several examples of such tests are presented and it is shown that despite of their high statistical significance they
are not able to provide clinically important association between the disease and biomarker. The predictive power of the vast majority of genetic
markers (given very wide confidence intervals due to small sample sizes) differs little from the population prevalence of the disease. Extremely
rare the odds ratios in the studies on the genetics of dispositions exceed practically critical OR = 5. As a result, in most cases, recommendations
of medical geneticists are based on clinically negligible (though statistically significant) recognizablity and predictability of genetic markers.
Keywords:
genetics of predisposition, genetic association, reproducibility, predictive values, Bayesian graphs