[CPA]

"The proportion of variation explained ranged from 0.38 to 0.42, with a wide prediction interval."

My actual background is in statistics and finance. I hated economic analysis studies when they tried to make something fit when it obviously did not. This is a prime example of the same thing in medicine...

What they are describing is R squared - roughly defined as the proportion of the square of the variance actually described by the model... In this case, their model is able to account for less than half of the square of the variance observed. This is a weak prediction in any field of study.

The term wide variance interval is also a danger signal. Think of it this way... If you had two classes, and they each had 5 students. The students in class 1 scored 80, 90, 85, 82, 88 on a test. Their average is 85. The students in class 2 scored 70, 90, 95, 70, 100. Their average is also 85.

Even though both classes averaged the same, the variance between the two is much different. In class 1, nobody was more than 5 points from the average. In class 2 there were multiple people up to 15 points from the average.

My Conclusion: The study can only account for approximately 40% of the square of the variation (weak predictor) and the prediction interval is wide - similar to the second classroom in my example above.

I also find it interesting that they did not include the level of significance used to come to their conclusion that 2 year DFS was a "significant" predictor of 5 yr survival. With such a weak model, and with their use of the wording "moderate strength" I would bet that they used .90 as their cutoff - leaving at least .10 that the results are just random chance. I would expect at least a .95 significance test be required before getting published, but maybe the Journal was desperate for submissions.