Although vertical cup-to-disc ratio is a well-recognized parameter in the prediction of OAG risk, the accuracy of prediction based solely on this parameter is poor owing to disc appearance in preclinical and early glaucomatous damage overlapping with the normal range of this trait. Predictive accuracy

for the individual patient should be improved by the inclusion of other variables, including genetics. With the genetics tools available CT99021 datasheet at this time, discriminatory power above and beyond that achievable with clinical risk factors is minimal; however, ongoing research uncovering the genetic basis of OAG is likely to lead to better risk prediction models. Neural networks allow an alternative approach to estimating the usefulness of clinical and genetic variables in predicting incident glaucoma. Input variables that are predictive of incident glaucoma naturally benefit the performance of the network. However, we see that those variables of trivial or no predictive value negatively affect the performance of the network: their inclusion necessarily makes the network structure more complex, which will lead to increased noise in the network. Neural networks are therefore helpful in distinguishing those patient characteristics that might help the clinician to predict

glaucoma incidence and those that will merely overload him or her with unhelpful information. This approach could easily be expanded to larger datasets where specific combinations of variables that are particularly beneficial might become apparent. The matching of age Tanespimycin (an important OAG risk factor) between cases and controls in the neural network analysis resulted in the TMCO1 SNP, rs4656461, becoming the highest-ranked genetic variable. This is consistent with a previously reported finding of the association of this SNP with age of onset of OAG. 20 Each of the associated SNPs in the logistic regression model also inhibitors contributed positively in the neural network. Thus, the combination of IOP, disc parameters, and genotype at-risk SNPs could improve the accuracy of OAG risk prediction, which in turn will inform early treatment

decisions for those most likely to develop of this blinding disease. The authors have completed and submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest and report the following: P. Mitchell received funding from Novartis (Frenchs Forest, NSW, Australia), Bayer (Pymble, NSW, Australia), and Abbott (Pymble, NSW, Australia); A. Lee from MSD products, Alcon (Frenchs Forest, NSW, Australia), and Allergan (Gordon, NSW, Australia); and A. White from Alcon (Frenchs Forest, NSW, Australia) and Allergan (Gordon, NSW, Australia); all for consultancy and lectures unrelated to the current project. K.P. Burdon is funded by a National Health and Medical Research Council (NHMRC) of Australia (Canberra, ACT), Career Development Fellowship (595944), J.J.