The Use of Probability Values in Statistical Testing

¶ … Suitable P-Value for a Clinical Trial

Statistical testing to determine whether results are significant is extremely useful in all types of research. In most cases, where a significant level, or p-value, is being chosen, a p-value of .05 is deemed to be sufficiently accurate. However, while this may be suitable for many types of research, it may be argued that in clinical trials from drugs, a lower p value may be more appropriate, due to the nature of the research. To understand this, it is necessary to understand what the p-value is, what it signifies.

The p value gives a probability, but is easy to misunderstand, as it indicates the level of support for the null hypothesis, with the probability level used to determine whether to accept or reject the null hypothesis. The p-value provides the probability of gaining an effect at the same level if the null hypothesis is true (Cowen, 1997). For example, if undertaking a basic hypothesis that there are at least twenty sweets in a packet of M&M's, with a null hypothesis that there are not twenty sweets in a packet, and the p-value is 0.4, this would mean there is only a 4% chance of the null hypothesis being true in that sample, and as the cut-off was .05, the null hypothesis could be rejected and the hypothesis accepted (Cowen, 1997).

This may be acceptable when applying the p-value to marketing, business studies, or other areas which are unlikely to have a potentially high human cost. However, in drug trials there is the potential for human costs as well as financial costs, and it is essential to ensure that drugs are effective at treating conditions are designed for, as well as assessing the potential of undesired side-effects. It was as a result of flawed research, and the way in which results were presented, that the drug Vioxx made it onto the market, and subsequently needed to be withdrawn, after finding detrimental side effects, with more than 23,000 people suffering from heart attacks (Berenson, 2006). This indicates a high potential human cost associated with drug trials. Therefore, the cut-off rate may vary depending upon the hypothesis being tested, but should be at least .01, so that the result is only deemed to be statistically significant if there is only a 1% or less chance of the proving the null hypothesis. For example, if a new drug is developed to treat cancer the difference the adoption of a lower p value can significantly reduce the potential for harmful human effects. However, it may also make it harder for the drug to get to market.

The important element associated with the significance level, and reduce it from the standard .05 to 01, is a reduction of potential harm. For example, if testing a new drug for dangerous side-effects, at a .05 p-value, a drug could be seen as potentially safe, but if given to 1 million people, can still end up harming 5% of them, or 50,000 people. However, if only allowed on the market if the p value for the test is below .01, then theoretically only 10,000 people, or 1%, may suffer the harmful side effects would be tested for. However, this should also be balanced with the need for treatment, the potential benefit it can create. For example, if the test were for the efficacy of cancer treatment, moving the p value from .01 up to .05 could result in an additional 1 million people having an effective cancer treatment, although it may not work for 50,000 of them. This is a difficult decision, and indicates the other types of information which should be considered.