Background Hip axis length (HAL) has been proposed as an important risk factor for hip fracture, independent of bone mineral density. Significant differences of HAL have been demonstrated in relation to gender, age and ethnic origin. The latter have been proposed to play a role in the different incidence of osteoporotic hip fractures among different countries.

Objectives Our objective consisted of studying the normal distribution of HAL in the Portuguese population and try to identify clinical parameters capable of predicting HAL, therefore reinforcing indication of bone densitometry.

Methods A randomly selected sample of 1705 inhabitants from the centre of Portugal, (474 males; 1231 females), aged 19 to 89 were included in the study. HAL was evaluated by DEXA (Hologic) using inbuilt software. Patients were asked and examined for age, present height and weight and at age 25, age at menarche and arm span. Following correlation studies, significant variables were included in stepwise regression analysis (SPSS).

Results HAL was significantly longer in males than in females (12,44 ± 0,69 vs 10,84 ± 0,69; p < 0.001). In men, significant correlations (p < 0.001) were found between HAL and height at age 25 (r:0.483), weight at age 25 (r:0.176) and arm span (r:0.0272). Using stepwise regression only height at age 25 was retained as significant (Model r: 0.490; Adj. R2: 0,238 – p < 0.001). In women, age (r: 0,06), height at 25 (0,304), present height (0,314), weight at 25 (0,214), present weight (0,058) and arm span (0,267), all showed significant correlations. Stepwise regression showed significant contributions for armspan, height and weight at age 25. The best fitting model explained 16,5% of the variance. In our population no major difference in HAL has been observed between the young (20–30 years old) and the elderly (>70), the difference being less than 1%.

Conclusion Although HAL is correlated with height at age 25 for both sexes, and also with arm span and weight at age 25 in females, the percent variance explained by the best fitting model using these parameters is too small to allow for reasonable prediction and inherent implications.