Statistics clinically important difference

Trial 4 is different, however. We do not have statistical significance, but the confidence interval suggests that there could still be something of clinical importance with potential differences of 5 mmHg, 10 mmHg, even... 

A statistically significant difference, no matter how small the P, does not mean that the difference is clinically important. A P value of <0.0001, if it emerges from a well-designed study, conveys a high degree of confidence that a difference really exists but says nothing about the magnitude of that difference or its clinical importance. 

The beneficial effects of spironolactone in congestive cardiac failure and hypertension are additive to those of ACE inhibitors. In the RALES, patients taking an ACE inhibitor who were randomized to spironolactone 25-50 mg/day had only a 0.3mmol/l increase in median potassium concentration (1). Although the difference between the spironolactone and placebo groups was statistically significant, the mean increase was not clinically important, and the incidence of serious hjrperkalemia was minimal in both groups of patients. 

As we have seen, several summary measures of central tendency can be used for continuous outcomes. The most common of these measures is the mean. In clinical trials we calculate sample statistics, and these serve to estimate the unknown population means. When developing a new drug, the estimated treatment effect is measured by the difference in sample means for the test treatment and the placebo. If we can infer (conclude) that the corresponding population means differ by an amount that is considered clinically important (that is, in the positive direction and of a certain magnitude) the test treatment will be considered efficacious. 

The a priori identification of specific risk hypotheses leads to the ability to take some of the methodology normally applied to hypothesis testing of efficacy results and apply it to safety assessment. The most important difference between these two analysis domains is that in the case of efficacy, outcomes of interest are identified a priori with great specificity, and clear statistical hypotheses are laid out in advance with complete analysis and decision rules documented in the clinical protocol. In practice, only a few efficacy variables are identified as primary, and only a few others as second-ary. Sensitivity, statistical power, and sample size are all carefully analyzed in advance to assure the trial will have a high probability of detecting differences of interest in these few critical variables. 

When our most recent - and most definitive - meta-analysis was published, the headlines in many newspapers blazoned that antidepressants don t work .1 The Daily Telegraph headline phrased it more specifically, clarifying that antidepressants are no better than dummy pills ,2 but even this headline was not entirely accurate. What our analyses actually showed was that antidepressants work statistically better than placebos, but that this statistical difference was not clinically meaningful. It was too small a difference to be of much importance in the life of a severely depressed person. 

The clinical significance of these CT scan alterations is unknown, and it is important to note that two similar studies found no statistically significant differences between the CT scans of long-term BZD users and matched control subjects ( 212, 213). Poser et al. (212) did find, however, that patients with combined BZD—alcohol dependence showed some degree of cerebral atrophy. Uhde and Kellner (211) also observed that the apparent association between VBR and duration of BZD exposure may be secondary to alcohol consumption, because long-term BZD users also may be more frequent users, but not necessarily abusers, of alcohol. 

It is noted several times in this book that the goal of experimental methodology is to provide optimum quality data for subsequent statistical analysis. This is true, but there is also a very important intermediary between data acquisition and data analysis this is the field of clinical data management. In many cases, Data Management and Statistics fall under the same division within a company, and in some cases these tasks are handled by different divisions. Whichever is the case, it is vital to have statisticians involved in all discussions regarding database development and use. 

Three comments are appropriate here. First, consideration of the traditional clinical trial design that has been the focus of attention up until this chapter is extremely worthwhile and instructive It has facilitated the introduction of fundamental design, methodology, and statistical concepts, and it will be an influential player in pharmaceutical drug development for many years to come. Second, the simple observation that the adaptive design may seem different does not in itself make it less valid, less valuable, or less important. Third, statistical approaches that are suitable for adaptive designs are, as yet, less well developed than they are for other study designs. 

Different types of evidence can be ranked in term of importance when decisions about clinical interventions are made (Figure 19.1).13,14 For example, the confidence from randomized controlled trials gives stronger evidence for treatment effects than open studies. Moreover, apparently conflicting results between studies may be compatible when a statistical meta-analysis of the data has been performed. This chapter will give a brief summary of evidence on the treatment effects of common dry skin disorders with urea-formulations. Furthermore, data on the influence of urea on the skin barrier function will be reviewed. 

---