Non-response bias in prospective studies of patients and health care professionals

Low participation and high attrition rates in research studies may produce socially and psychologically biased study samples which limit the generalizability of findings. Although this problem is often acknowledged, published studies sometimes fail to assess and report non-response bias and make few efforts to prevent it. Two examples, one involving patients and another involving health professionals, are reported to illustrate the problem. The first was a randomized controlled trial of presenting a prenatal screening test, with questionnaires given to participants on four occasions. The final sample differed in ethnic group and use of screening, being more likely to be white and to undergo the test. The second example was a randomized controlled trial of communication skills training of obstetricians and midwives. Those completing the study differed from the others in holding more positive attitudes towards training and having better communication skills. Potential solutions to the problem of non-response bias are addressed.