StatsDirect—Statistical Software for Medical Research in the 21st CenturyBMJ 2000; 321 doi: https://doi.org/10.1136/bmj.321.7275.1536 (Published 16 December 2000) Cite this as: BMJ 2000;321:1536
- Nick Freemantle, reader in epidemiology and biostatistics
Iain E Buchan
CamCode, £99 single user academic, £179 single user commercial, £49 students and developing world
Clinicians involved in medical research are increasingly numerate and statistically sophisticated. Recently, two colleagues hounded me. They wanted to see confidence intervals for every number in a paper describing a randomised trial that we had worked on together. I found myself in a strange position—(1);the confidence intervals that they wanted were neither straightforward nor ultimately appropriate, but it was pleasing to be asked for them after years of emphasising the importance of estimation rather than hypothesis testing.
In part, this increased sophistication is due to the availability of accessible and intuitive statistical packages. StatsDirect is an addition that can be used by those without the questionable luxury of a PhD in statistics and a spare year or two to learn a new programming language.
StatsDirect performs a range of functions, including sample size calculations, exact 95% confidence intervals for odds ratios, numbers needed to treat, non-parametric analyses, and survival analysis methods. A host of useful features, such as t tests from summary data, make it valuable for reviewing research.
Its greatest value is its functionality—(1);one colleague used it to perform a meta-analysis in a matter of minutes. Statistical novices will probably find the package straightforward to use. It also does a good job of keeping up to date with new methods. After recent work on analysis of cost data in clinical trials published in the BMJ, the program's author agreed to add non-parametric bootstrap confidence intervals as a feature for a future release of the package.
Its main disadvantage is that it does not provide a programming language and a range of model specifications, a limitation particularly important for generalised linear modelling. There are, however, other packages that can address this requirement, but surprisingly few can make statistical analyses available to a broad range of researchers, as this package does.
Competing interests NF provided advice to the author on the development of meta-analysis methods. Neither he nor his department received any financial reward for this work beyond a complementary copy of the package.