Article Text

  1. JANET RENNIE, Deputy Editor

Statistics from

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.

Chronic lung disease; prevention and outcome

Large doses of oral vitamin A (5000 IU per day) failed to reduce the incidence of CLD when evaluated in a randomised controlled trial in Liverpool by Wardle et al (page 9). Previously, benefits have been claimed for high dose intramuscular vitamin A, and the Cochrane database reports a reduction in CLD from 62% to 55% after 5000 IU IM three times a week.

In this month's Archives, researchers in Melbourne managed to persuade over 200 14 year olds to complete a complex protocol of lung function tests (Doyle et al. Arch Dis Child2001;84:40–4). Eighty six had weighed less than a kilo at birth in 1977, and many had been diagnosed with CLD in infancy. Asthma was equally common in all the groups (about 20%). There was no difference between the ex-premature group and the control children, apart from the finding that those who had been born preterm had low air flow. The authors speculate that this may herald the earlier onset of obstructive airways disease in later life.

Hypocarbia in trigger ventilation

Babies who were enrolled into the multicentre trial of patient triggered ventilation in Plymouth were studied carefully for the degree and duration of hypocarbia (page 14). In the post antenatal steroid and surfactant era hypocarbia is a more common complication of artificial ventilation, and there is an association between hypocarbia and PVL. Given the potent effects of change in carbon dioxide tension on cerebral blood flow, this association is not surprising. Hypocarbia was defined as an arterial PaCO2 < 3.33 kPa, and was surprisingly common in babies ventilated conventionally or on PTV. Over a third of babies experienced inadvertent hypocarbia on at least one occasion during the first three days (mainly the first), but there was no difference between the two modes of ventilation. Hypocarbia was more likely if the baby was ventilated in low oxygen concentrations with low pressure; in other words if the baby did not really need artificial ventilation. The authors propose more aggressive weaning to avoid hypocarbia in this group.

Finger whorls and blood pressure

The Barker group have proposed that fingertip whorls can be used as markers of impaired fetal development and that some patterns are associated with high blood pressure in adult life. The Liverpool group challenge this hypothesis with their findings in ex preterm children (page 18). Whilst the blood pressure was higher in 15 year olds who had been born preterm when compared with a control group, there was no difference in their dermatoglyphic patterns.

A statistical snare (from Professor Michael Healy)

One of the commonest stastistical problems is that of assessing the significance of the difference between two proportions1 ,2—for example, the proportion of subjects responding to a new treatment and to a placebo control. This is usually best done by way of the χ2 test with 1 degree of freedom, incorporating a so called correction for continuity (Yates's correction). This is actually an approximate procedure and the approximation may be suspect when the numbers involved are small. The exact test appropriate under these circumstances is due to Fisher and is provided by most of the standard statistical computer packages.

What is not always appreciated is that Fisher's test calculates a one sided probability—the chance probability of a deviation as large as that observed but only in the observed direction. Some packages provide this probability without comment and these one sided p values often appear in papers submitted to the Archives. They are not usually what the authors intended, and as statistical advisor I try to call attention to them, but some slip through the net.

Regrettably, there is not total agreement among statisticians as to the method of obtaining a two sided p value from Fisher's test. The generally accepted solution (that advocated by Fisher himself) is simply to double the single tail probability.

When using Fisher's exact test it is a good discipline to calculate the χ2 p value as well. It will be found in practise that the two agree remarkably well, even when quite small numbers are involved. If one p value is twice the other, suspicions should be aroused.