The proliferation of powerful statistical software for personal computers has seen a similar growth in the number of statistical techniques employed in the medical literature. Here we review common problems in the application of these techniques. Basic statistical terminology and techniques are briefly reviewed. Errors in the design and conduct of scientific trials can be grouped into four broad areas: 1. Experimental design and conduct of the trial; 2. Analysis of the data; 3. Interpretation of results, and 4. Presentation of the results. Common problems in trial design and conduct are examined in terms of appropriate sample and control selection (including sample size), collection of data and retrospective study designs. Numerous common analytical errors are presented with more appropriate alternatives. Common errors in the interpretation of statistical results include confusion between clinical and statistical significance, incorrect application of P values, and the belief that statistically significant association provides direct evidence of a causal relationship between the variables concerned. Finally, common data presentation errors and some special problems of interest to clinical scientists are examined (reference intervals and method comparisons). We conclude that the increasing availability of powerful statistical methods is not matched by an equivalent level of understanding in their appropriate application. These increasingly prevalent errors can mislead other investigators, lead to inferior or delayed patient treatments, and perpetuate similar errors in the literature.
|Number of pages||9|
|Journal||New Zealand Journal of Medical Laboratory Science|
|Publication status||Published - 1999|
- Inappropriate statistics
- Method comparison
- Reference interval
- Statistical errors