Purchase this article with an account.
Hui-Chun T. Hsu, Kathryn S. Kennedy, Dale Usner, Dale J. Kennedy, Richard Abelson; Evaluation and Selection of Statistical Analyses for Diary Data in Dry Eye Studies. Invest. Ophthalmol. Vis. Sci. 2012;53(14):2366.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Patient symptom diaries are a commonly used method to collect efficacy data in clinical trials such as those for dry eye treatments. Typically, patients are asked to report symptom severity several times per day over the course of a study that may last weeks or even months. There are several possible ways to analyze data from these diary entries, but to date, systematic comparisons of different analytical methods are lacking. To compare a number of different methods, 500 sets of random samples were generated and different analyses were performed and compared based on the simulated data.
The diary data were randomly selected from a multivariate normal distribution with parameters selected from the previous studies for each treatment based on a compound symmetry correlation structure. For each simulation set, one week of diary data was generated for 50 subjects per treatment (active and placebo), assuming a 0.5 treatment mean difference on a scale of 0-5 with a mean standard deviation of 1. A higher score indicates more severity. Three t-test methods and three mixed models were used to analyze the treatment effects. T-test methods: (1) a two-sample t-test for daily diary score where the majority of days show significance; (2) a two-sample t-test for the overall mean score; and (3) a two-sample t-test for area under the curve (AUC). The mixed models accounted for repeated measures within each subject with different pre-specified covariance matrix structures: (4) unstructured; (5) compound symmetry; and (6) autoregressive of order 1. The percentage of times where the results indicated significant treatment differences were summarized for comparison.
A mixed model with autoregressive covariance matrix structure (6) yielded the highest percentage of significance, 85%, whereas a two-sample t-test for each day with the majority of days being significant (1) yielded the lowest rate of 70%. All the other methods had similar significance rates around 78%.
All of the above statistical analyses are valid for diary analysis. However, the mixed model approach is recommended because it can capture the trend of the curve and requires less data manipulation compared to mean score or AUC approaches. It also has the same/more power to detect treatment differences.
This PDF is available to Subscribers Only