How many patients are needed? Variation and design considerations in bone histomorphometry

In osteoporosis research, bone histomorphometry plays an important role in documenting the biological effects and possible side-effects of new drug tr eatments. To ensure that the study is properly scaled, it is important to b e concerned with the risk of type II error; that is, the risk of failing to detect a real difference. We therefore calculated the necessary sample siz e in bone histomorphometric studies according to a specified difference of 15% between two groups. The calculations were based on variance components estimated from three different studies: women with a distal fracture of the forearm (n = 22); patients with pituitary insufficiency (n = 21); and pati ents with primary hyperparathyroidism (n = 21), Using a significance level of 0.05 and a risk of type II error of 0.20, the statistical power of two d ifferent designs was compared: a single biopsy design comparing the respons es in two groups after the treatment; and a paired biopsy design in which i ndividual differences (posttreatment minus baseline) were calculated before the comparison of the two groups. We found that the mineral apposition rat e, wall thickness, and erosion depth are statistically powerful indites tha t, in the single biopsy design, require no more than n = 25 in each group t o detect differences of 15% between the groups. Bone volume, erosion surfac e, osteoid surface, mineralizing surface, and activation frequency need gro up sizes of 100-600 individuals to find a 15% difference to be statisticall y significant, However, the effect of bisphosphonate treatment, for instanc e, is Large enough to reduce the group size to 20 individuals concerning ac tivation frequency. The remodeling balance reaches extreme group sizes of s everal thousand for a 15% difference to be statistically significant, but f or a 5 mum (approximately 150%) improvement, about 100 individuals are requ ired in the single biopsy design. An analysis of the components of variance showed that the variation between individuals is small and often negligibl e compared with the variation within individuals, and sample sizes needed f or the paired biopsy design are therefore larger than those for the single biopsy design. In conclusion, the most rest-effective histomorphometric stu dy design within a randomized clinical trial appears to be a single biopsy design comparing posttreatment biopsies with scaling performed according to the statistical power of the indices of interest. (C) 2001 by Elsevier Sci ence Inc. All rights reserved.