Jeong H. Oh writes: >For the past years we have started to use in our studies statistical >methods that try to adjust for differences between experimental/control >groups, etc. > >Most of those studies are reported in RR or OR from logistic regression. > >Can you use the formula for NNT based on OR for systematic reviews in >these therapy studies? I'm not sure I understand your question, but I'll try to answer anyway. The Relative Risk (RR) and Odds Ratio (OR) both measure RELATIVE change in the probability of some event (remission, cure, survival, etc.). Relative changes involve division. The Number Needed to Treat (NNT) requires that you know the ABSOLUTE change in probability. Absolute changes involve subtraction. >From what I understand, clinicians are more interested in absolute changes than relative changes, because an absolute change is directly proportional to the number of lives you will save or the number of diseases you will prevent. If you know the OR/RR and if you also know the baseline rate (the rate in the control group), then you can figure out the absolute change. But if you don't know the baseline rate, then you are out of luck. A large OR may seem quite impressive, but if the baseline probability is small, it leads to a disappointingly large NNT. For example, if you cut the risk of an event from .3% to .1%, then the OR is approximately 3 (or 1/3 depending on what you put in the numerator and denominator), but the NNT would be 500. If you cut the risk from 25% to 10%, the OR is still 3, but the NNT is a much nicer value of 6.7. It's important to understand the distinction between relative and absolute changes. A small relative change in a common event may have more impact than a large relative change in a rare event. Many systematic reviews and meta-analyses summarize data from studies where the baseline probabilities vary all over the place. In this case, it's unclear which baseline probability to use. Perhaps you could compute the NNT for a range of baseline rates. For example, compute NNTs for low, medium, and high risk populations. There's a related question about whether we should use meta-analysis when there is a wide variation in baseline rates, but that's the topic for another email. Steve Simon, [log in to unmask], Standard Disclaimer. STATS - Steve's Attempt to Teach Statistics: http://www.cmh.edu/stats %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%