Christine MH
05-24-2005, 04:43 AM
I once had a great stats prof who said that he wasn't going to make statistics easy because it already was easy. Let's see if he's right:
Hazard ratio: The lower the figure the better. For HERA the hazard ratio was .54 (54%), meaning that for every one hundred women who would have recurred without herceptin, only an estimated 54 of the recurred when herceptin was given after treatment. The risk reduction is 1 - the hazard ratio (1-.54) which is the same as 100%-54% =46%.
The 95% confidence interval. You will note that I said 'estimate' above, because there is always some uncertainty in statistics, so it's not clear whether the value is exactly 54%. For the HERA results, the 95% confidence interval was .43-.67. This means that it is 95% certain that of one hundred women who would have recurred without herceptin, only between 43 and 67 of them recurred when herceptin was given. So they are 95% certain that the risk reduction is somewhere between 33% (100%-67%) and 57% (100%-43%).
The P-value is the probability that something is purely a coincidence. If the p-value is .01 that means that there is a 1 in a hundred chance that it is a coincidence. If it is .05, there is a 5 in a hundred chance of it being a coincidence. In general, only p-values under .05 or sometimes .01 are considered statistically significant. The smaller the number the better. To demonstrate statistical significance, it is important to have a large number of cases, which brings me to HERA.
The HERA results show that the p-value for disease free survival is less than .0001. This means that the chance of this being a coincidence is less than 1 in 10,000. However, the p-value for overall survival is just .26 which means that the chance of this being a coincidence is 26%. However, HERA was not as far along as the other studies, so it may simply be too early to see a difference in overall survival, since the followup was only one year following women entering the trial. I would caution that HERA included about 1/3 node-negative women, some of whom received really old chemo drugs believed to be less effective against HER2, such as CMF, so it cannot be compared directly with the herceptin-based chemo studies.
The combined studies on herceptin-based chemo seem to have used a two-tailed test, which I am much less familiar with, but again the same rule about p-values values under .05 or 5 in 100 being good news is the same. For distant disease free survival, there is only a 1 chance in 10,000,000,000 that this is a coincidence (I think, I've never seen a p-value this small!). For overall survival, herceptin-based chemo again achieved statistical significance (.015) with there being just a 1.5 chance in 100 of this being a fluke.
Are there any other problems figuring out the stats?
Christine
Hazard ratio: The lower the figure the better. For HERA the hazard ratio was .54 (54%), meaning that for every one hundred women who would have recurred without herceptin, only an estimated 54 of the recurred when herceptin was given after treatment. The risk reduction is 1 - the hazard ratio (1-.54) which is the same as 100%-54% =46%.
The 95% confidence interval. You will note that I said 'estimate' above, because there is always some uncertainty in statistics, so it's not clear whether the value is exactly 54%. For the HERA results, the 95% confidence interval was .43-.67. This means that it is 95% certain that of one hundred women who would have recurred without herceptin, only between 43 and 67 of them recurred when herceptin was given. So they are 95% certain that the risk reduction is somewhere between 33% (100%-67%) and 57% (100%-43%).
The P-value is the probability that something is purely a coincidence. If the p-value is .01 that means that there is a 1 in a hundred chance that it is a coincidence. If it is .05, there is a 5 in a hundred chance of it being a coincidence. In general, only p-values under .05 or sometimes .01 are considered statistically significant. The smaller the number the better. To demonstrate statistical significance, it is important to have a large number of cases, which brings me to HERA.
The HERA results show that the p-value for disease free survival is less than .0001. This means that the chance of this being a coincidence is less than 1 in 10,000. However, the p-value for overall survival is just .26 which means that the chance of this being a coincidence is 26%. However, HERA was not as far along as the other studies, so it may simply be too early to see a difference in overall survival, since the followup was only one year following women entering the trial. I would caution that HERA included about 1/3 node-negative women, some of whom received really old chemo drugs believed to be less effective against HER2, such as CMF, so it cannot be compared directly with the herceptin-based chemo studies.
The combined studies on herceptin-based chemo seem to have used a two-tailed test, which I am much less familiar with, but again the same rule about p-values values under .05 or 5 in 100 being good news is the same. For distant disease free survival, there is only a 1 chance in 10,000,000,000 that this is a coincidence (I think, I've never seen a p-value this small!). For overall survival, herceptin-based chemo again achieved statistical significance (.015) with there being just a 1.5 chance in 100 of this being a fluke.
Are there any other problems figuring out the stats?
Christine