| Abstract |
INTRODUCTION :
Every participant in ophthalmological studies usually contributes two eyes in analysis and the measurements from both eyes are usually highly correlated. The aim of the study was to provide a summary of appropriate statistical techniques, for a variety of continuous and binary ophthalmic measurements. Also to compare different approaches using analysis of continuous and discrete outcomes.
METHODOLOGY :
A review of 78 articles, published in Investigative Ophthalmology and Visual Science (IOVS) has been set out during the first trimester of 2007, in order to indicate the analytical approaches. The main objective was to examine if the authors of these publications have used the most appropriate statistical approach. A sum up of the appropriate methods has been made, based on the case study, the measurement and the type of the variable.
Finally, in order to achieve the practical implementation, real data were deduced out of two studies, continuous and binary data respectively, based on dependent and independent observations among ophthalmological measurements, investigating the different types of approaches and to compare the results. Specifically, the analysis of continuous outcomes concluded measurements of Pattern VEP with data from 28 right and 35 left eyes (63 eyes) and measurements of Pattern ERG with data of 30 right and 37 left eyes (67 eyes). The analysis of binary data included visual acuity data from high school students (301 Greek students (602 eyes) and 161 students (322 eyes) from Bulgaria.
ΣΤΑΤΙΣΤΙΚΗ ΕΠΑΓΩΓΗ ΜΕ ΟΦΘΑΛΜΟΛΟΓΙΚΑ ΔΕΔΟΜΕΝΑ
ΚΑΡΑΚΩΣΤΑ ΑΝΝΑ 13
RESULTS :
The use of two eyes in analysis is due to the design of the study (binocular diagnosis, outcome variable on individual, paired data) in 12,8% of the cases. In studies were it is not necessary to use both eyes for the analysis (77,6%), most researchers prefer to examine only one eye per person (79,5%) for the statistical analysis, while the studies using data from both eyes (20,5%) rarely take account of inter-eye correlation (13,3%). In 9,6% of the cases the analysis is qualitative - not statistical analysis - or the researchers do not explain the analytical approach.
The analysis of quantitative data included small sample of observations and missing data. The correlation between the two eyes was found to be strong for VEPRATIO measurements (ΙCC=67%) and small for ERGRATIO measurements (ICC=29%). The use of right and left eyes separately, showed that the estimation of the Pvalue was different and thus, led to opposite results about the differences among the groups. Visual acuity data had strong inter-eye correlation (ICC=0,7062 and Effective number of eyes ==1,17). The analysis of both quantitative and qualitative data which concluded data from both eyes without regard on correlation, led to underestimation of the Pvalue in comparison to the method which takes into account the intraclass correlation. ˆe
CONCLUSIONS :
It seems, that in most ophthalmological studies the researchers prefer to use only one eye for the analysis. When a selection of only one eye per person is made, the sample must be large and it is more appropriate to analyze right and left eyes separately. In cases were the analysis uses the averaging results, the analysis is valid as long as there is an appropriate number of observations and there is no missing information. When the sample is small, the use of two eyes is more effective.
Furthermore, it is important to estimate the intraclass correlation before the choice of the analysis is made, in order to avoid the incorrect analysis. When the analysis includes data from two eyes per person and there is a correlation between eyes, specialized techniques must be used which will take into account the inter-eye
ΣΤΑΤΙΣΤΙΚΗ ΕΠΑΓΩΓΗ ΜΕ ΟΦΘΑΛΜΟΛΟΓΙΚΑ ΔΕΔΟΜΕΝΑ
ΚΑΡΑΚΩΣΤΑ ΑΝΝΑ 14
correlation and avoid the underestimation of the real Pvalue or the probability of the Type I error (rejection of H0 when it is real).
|
Search
