The gap between regular health care and research can sometimes be overwhelming. The researchers have the tools, the hospital change routines all the time, but they almost never meet. I’m therefore pleased to see Åstrand et al’s simple article, a follow-up of a screening program for osteoporosis that showed a decrease fracture risk about 50 %.
By using questionnaires at baseline, that is before and after the program started, together with a simple follow-up questionnaire after about 6 years they could see that those that had participated in the screening program were almost half as likely to have another fracture. It’s also interesting that deaths were fewer in the intervention group, 53 in the unscreened group and 34 in the screened group (p = 0.06). Even if not quite significant, it is likely that a fracture might sometimes be an indicator of other illnesses, and a proper follow-up by a GP is always a good idea.
A few epidemiological notes
This study is of course by it’s simple design subjected to several biases, the most important ones probably being:
- Recall bias – the subjects don’t always remember correctly. There is also a risk that those that had a fracture will recall certain details more.
- Loss to follow-up bias – the drop-out rates were slightly over 20 %, and even if their analyses didn’t show any significant reasons this is a type of selection bias and those are usually the most difficult to control and adjust for. The authors also noted that some of the follow-ups occurred when patients had subsequent fractures.
- Self-selection bias – unfortunately it seems that another 20 % disappeared in the first questionnaire, and I can’t find how the authors tried to address this other significant selection bias.
- The Hawthorne effect – the observed subjects improve just by knowing that they are observed
Apart from this I’m slightly surprised by the mean time until fracture – 6 years. This is the same time as to the follow-up!? It seems a little too much of a coincidence and I wonder if they didn’t miss something in the data handling…