Commentary: Sources of bias must be controlled

BMJ 1997; 315 doi: (Published 26 July 1997) Cite this as: BMJ 1997;315:220
  1. Leila Duley, senior research fellowa
  1. a National Perinatal Epidemiology Unit, Radcliffe Infirmary, Oxford OX2 6HE

    Taha and colleagues address an important clinical issue, postpartum and neonatal infection. They conclude that infection is reduced by a more thorough cleansing procedure before vaginal examination during labour, combined with wiping the baby after delivery. The authors claim that alternative explanations of their results are unlikely—but, as they used a study design which could not ensure that the comparison was fair, the reported effects may be due to bias.


    The most reliable way of comparing different forms of care is in a properly conducted randomised trial.1 Non-randomised studies tend to report larger estimates of treatment effects than those using random allocation2; even among randomised trials, those in which there is poor concealment of allocation are associated with bias.3 This study was not randomised and so there is no way of knowing, or proving, that there were no systematic differences (biases) between the two groups. The observed differences in outcome may be due to differences between the women in each group or the care they received, rather than a differential effect of the cleansing procedure on risk of infection. Simple, practical ways of randomising large numbers of people are available4; rather than being dismissed as impractical, these need to be developed for use within a wider variety of settings, particularly in developing countries.

    Consent was given during labour for the new vaginal cleansing with chlorhexidine, but not the traditional perineal wash with Savlon. As Savlon was removed during the experimental period, it is unclear what happened if a woman refused. Consent for enrolment into the study was sought after delivery. The decision about eligibility and the woman's willingness to consent may therefore have been influenced by events related to the intervention. For example, women whose babies died “before enrolment” were excluded and there is no information about time or cause of death. If these women are included in the analysis of neonatal mortality the relative risk is smaller and no longer statistically significant (0.87; 95% confidence interval 0.71 to 1.05). The potential for bias would be less if all women delivering during the study period had been included in the analysis, and if only those not needing a vaginal examination (if delivered before admission, for example) or with a stillbirth before the first examination had been excluded.

    Another potential source of bias is in how outcome is decided. The diagnoses of neonatal and maternal sepsis and cause of death were made by hospital clinicians who were “independent” of the study. In an open study such as this, however, clinicians could very easily have found out, or guessed, which cleansing procedure was used each month.

    In view of these limitations, it is difficult to assess the true value of vaginal cleansing and wiping the baby in comparison to perineal cleansing. Also, data from a single study are rarely conclusive and should be put in the context of all available evidence.5 A brief search of the Cochrane Controlled Trials Register6 yielded citations of six possibly randomised studies evaluating vaginal cleansing with chlorhexidine.7 8 9 10 11 12 The most reliable guide for clinical practice and future research would be a systematic review of all relevant studies with adequate control of bias.


    1. 1.
    2. 2.
    3. 3.
    4. 4.
    5. 5.
    6. 6.
    7. 7.
    8. 8.
    9. 9.
    10. 10.
    11. 11.
    12. 12.
    View Abstract