Missing observations in bone histomorphometry on osteoporosis: implications and suggestions for an approach

Crucial bone histomorphometric indices, i.e., turnover-related indices, are based on tetracycline double labeling. However, these indices are particularly exposed to loss of information because of missing readings on double labels. If the failure to make the observation is related to its magnitude, then selection bias may invalidate the conclusions. Therefore, ignoring missing double labels may lead to a selection of high-turnover patients. The aim of this study was to analyze the dimension and the impact of excluding iliac crest bone biopsies with missing readings in women with spinal crush fracture osteoporosis (n = 158, median 68 years, range 49–80 years). Furthermore, two different lower limits of the mineral apposition rate (MAR) were examined to explore their usefulness as a biological minimum that can be used for cases with missing readings, i.e., recoding of missing values. The average MAR (calculated as the mean of all interlabel widths measured in each individual) shows a lower limit of 0.3 μm/day, suggesting an apparent minimum for the interlabel width (Ir.L.Wi) of 3–4 μm. Identifying the smallest interlabel width measured in each individual and calculating the minimal MAR shows that 77% of the minimal MAR values are below 0.3 μm/day and reach a minimum of 0.1 μm/day, corresponding to an interlabel width of about 1 μm. Therefore, the minimal MAR presents a biological minimum of 0.1 μm/day. This value is used for our recoding: if no labels are sampled (2% of our population), Ir.L.Wi is assigned the value 0; if none or an insufficient number of double labels are sampled (29% of our population), then Ir.L.Wi is assigned the value 1 μm. Excluding cases with missing readings on any dependent variable increases the mineralizing surface (MS/BS) by 60% (2p< 0.01); other indices show no significant change. The suggested recoding decreases the average MAR by 4% (2p< 0.01), prolongs the remodeling period by 19% (2p< 0.01), and tends to decrease the activation frequency (2p = 0.09). Furthermore, the number of excluded biopsies tends to be larger among the older (2p = 0.09) and more severely osteopenic individuals (2p = 0.09). We conclude that ignoring missing double labels leads to selection bias; therefore, specific measures such as recoding procedures are needed to allow proper representation of low turnover patients. There is also a risk of bias caused by the exclusion of the older, osteopenic patients in bone histomorphometric osteoporosis trials.