Figure 2 Adjusted* hazard ratios with 95% CI for first hospitalisation by causes and age groups in relation to baseline second-hand smoke (SHS) exposure (compared with no SHS exposure). *Adjusted for sex, birthweight, gestational age, breastfeeding history, maternal age, highest parental education level, type of hospital at birth, household income per head, mother’s smoking and mother’s SHS exposure in pregnancy.
Here is Figure 2 from the Kwok report. I primarily choose it as it gives a very good immediate visual impression of the true outcome of the statistical manipulation.
If you just gaze at the whole image you will see that the X's and O's are pretty close to the baseline RR of 1, despite the expanded x-axis scale. You will also notice that the universally accepted level of RR to hint at an association of 3 is not met by any of the values except in the accident group.
No information as to the actual number of hospital admissions in each group (infection, accidents etc) is given, which is essential to make any conclusions.
Apart from the accident and all causes groups the distribution of values across the baseline is equaly shared between the increased and the decreased RR areas. (Above and below the baseline). You may also notice that virtually all the lines would overlap if they were superimposed.
These graphs, far from supporting any association between the proximity of the smoker to the infant and illness in these infants, irrefutably show that this study has demonstrated no link between "passive smoking" and hospital admission of infants for any cause.
This type of analysis is what is lacking in the media and can be applied to all studies on passive smoking.