Sir,

Chau et al1 described raised plasma metalloproteinase-9 (MMP-9) in age-related macular degeneration (ARMD), hypothesising that if atherosclerosis and ARMD share a common mechanism, then systemic MMP-9 and MMP-2 may be increased in patients with ARMD. While it is a plausible hypothesis, it is unclear why they have not tested it in a simple cross-sectional analysis of two groups—one with ARMD but no atherosclerosis and another group with atherosclerosis but no ARMD. Instead, each of their three sample groups includes patients with or without a history of symptomatic atherosclerosis (eg, myocardial infarction) as well as risk factors (eg, hypertension). This is pertinent as cardiovascular disease and its risk factors substantially affects MMP levels.2, 3

The authors state ‘no statistically significant differences in clinical data related to atherosclerosis were observed between subjects in the three groups included in the study’. As they offer no statistical test or data in support of this statement, it is unclear to us how they can make this assertion. The authors provide no power calculation to assure the reader that the risks of statistical error types 1 and 2 have been addressed and minimised, given the small numbers in each subject group. We wonder whether in fact the failure to find a difference between data sets is because of a small number error (ie, false negative).

Chau et al1 choose to present their data as SEM, which denies the reader a simple estimate to the distribution of the data—many would argue that the correct measure of variance is SD.4 One of their results is a mean of 740 ng/ml with SE 494 ng/ml. We submit that these data are likely to have a strong non-normal distribution and so convention states that it should be presented as median and interquartile range. We wonder which other indices also have a non-normal distribution, which could be determined by a test of statistical normality (eg, the Anderson–Darling test). The importance of distribution is that it governs the method of analysis. Differences in three or more data sets with a normal distribution should be sought using analysis of variance (ANOVA), as the authors have used. However, if their data contain at least one data set with a non-normal distribution, then the correct analysis would be a test such as the Kruskall–Wallis test. In addition, neither ANOVA nor the Kruskall–Wallis test tell us of differences between individual groups. A post hoc test is necessary to probe for such intergroup differences should have been performed; student's t-test is inappropriate. We will appreciate it if the authors can provide these analyses, as well as a power calculation.

While we do not doubt the accuracy of the raw data derived from the plasma, we have difficulty with their analysis and presentation of the results. They may find a different story with their data, the findings of which could be important.