Corneal inflammatory events with daily silicone hydrogel lens wear.

PURPOSE: This study aimed to determine the probability and risk factors for developing a corneal inflammatory event (CIE) during daily wear of lotrafilcon A silicone hydrogel contact lenses. METHODS: Eligible participants (n = 218) were fit with lotrafilcon A lenses for daily wear and followed up for 12 months. Participants were randomized to either a polyhexamethylene biguanide-preserved multipurpose solution or a one-step peroxide disinfection system. The main exposures of interest were bacterial contamination of lenses, cases, lid margins, and ocular surface. Kaplan-Meier (KM) plots were used to estimate the cumulative unadjusted probability of remaining free from a CIE, and multivariate Cox proportional hazards regression was used to model the hazard of experiencing a CIE. RESULTS: The KM unadjusted cumulative probability of remaining free from a CIE for both lens care groups combined was 92.3% (95% confidence interval [CI], 88.1 to 96.5%). There was one participant with microbial keratitis, five participants with asymptomatic infiltrates, and seven participants with contact lens peripheral ulcers, providing KM survival estimates of 92.8% (95% CI, 88.6 to 96.9%) and 98.1% (95% CI, 95.8 to 100.0%) for remaining free from noninfectious and symptomatic CIEs, respectively. The presence of substantial (>100 colony-forming units) coagulase-negative staphylococci bioburden on lid margins was associated with about a five-fold increased risk for the development of a CIE (p = 0.04). CONCLUSIONS: The probability of experiencing a CIE during daily wear of lotrafilcon A contact lenses is low, and symptomatic CIEs are rare. Patient factors, such as high levels of bacterial bioburden on lid margins, contribute to the development of noninfectious CIEs during daily wear of silicone hydrogel lenses.

Alzheimer's disease and smoking: bias in cohort studies.

The discrepancy between cohort and case-control studies regarding the association between smoking and Alzheimer's disease (AD) has been attributed to the competing risk of early mortality of smokers. A simulation study was conducted to show that the bias favoring smokers acts also on cohort studies. In the model, individuals {grow older} and have smoking habits according to published year-age-gender-specific patterns, with morbidity and mortality according to their demographic and smoking profiles. Those individuals dying of smoking-related causes ("phantoms") remain at risk of AD and of death from other causes. Three scenarios were considered: no association of AD and smoking, increased risk for smokers, and decreased risk for smokers. For each simulation of a cohort study, two incidence density ratios (IDR) were computed: one including the phantoms that developed AD (thus ignoring smoking-related deaths) and another excluding them (thus mimicking real-life studies). For all scenarios, the simulations show that smoking-related death creates a bias, resulting in smokers having an understated risk of AD compared to non-smokers. The speculation that the conflicting results of case-control and cohort studies are solely due to the increased mortality in smokers thus appears unjustified. Other factors must also be considered to explain the discrepancy in results.

CLINICAL/MEDICAL OUTCOME PREDICTION BY NEURAL NETWORKS WITH STATISTICAL ENHANCEMENT.

Neural networks offer a powerful new approach to information processing through their ability to generalize from a specific training data set. The success of this approach has raised interesting new possibilities of incorporating statistical methodology in order to enhance their predictive ability. This paper reports on two complementary methods of prediction. one using neural networks and the other using traditional statistical methods. The two methods are compared on the basis of their prediction applied to standardized developmental infant outcome measures using preselected infant and maternal variables measured at birth. Three neural network algorithms were employed. In our study, no one network outperformed the other two consistently. The neural networks provided significantly better results than the regression model in terms of variation and prediction of extreme outcomes. Finally we demonstrated that selection of relevant input variables through statistical means can produce a reduced network structure with no loss in predictive ability.