TO THE EDITOR: In their national analysis, Robson and colleagues found less favourable outcomes among term singleton babies born in public hospitals than in private hospitals. Most health services research is non-randomised, and it is unrealistic to expect studies such as this to be as internally valid as a randomised controlled trial. As noted by others, in the absence of randomisation, it is difficult to untangle the myriad differences between mothers, babies, and the care provided in private and public hospitals (ie, the results might be subject to confounding). Robson et al statistically accounted for the potential confounding effect of maternal smoking (although these data were only available for about half the mothers), age, Indigenous status, parity, diabetes, hypertension, rurality and method of birth. We replicated their analysis using the Queensland Perinatal Data Collection (July 2005 – December 2007), which included virtually complete data on maternal smoking. Our analysis of 124 300 term singleton babies gave an adjusted odds ratio (AOR) for perinatal mortality of 2.0 (95% CI, 1.4–2.7), which is similar to that reported by Robson et al. Using more detailed data from the Queensland dataset, we statistically adjusted for other potential confounders — including pre-existing and gestational diabetes, pre-existing and pregnancy-induced hypertension, pre-eclampsia, eclampsia, antepartum haemorrhage, anaemia , depression, urinary tract infection, low birthweight (< 2500 g), socioeconomic status (based on area of usual residence), alcohol and drug misuse, and artificial reproductive technology — and obtained an AOR of 2.1 (95% CI, 1.5–2.9). We are not implying that adding more and more variables to a statistical model is an appropriate way to account for confounding. Our aim is simply to show that Robson et al’s result is robust to statistical adjustment using the available data; this is not the same as saying the analysis is robust to confounding. We found that the higher perinatal mortality in public hospitals was greater for neonatal deaths (AOR, 3.1; 95% CI, 1.8– 5.6) than for stillbirths (AOR, 1.6; 95% CI, 1.0–2.4). Excluding lethal congenital anomalies did not materially change the result (AOR, 1.9; 95% CI, 1.3–2.8). We also stratified our analysis by level of hospital — tertiary referral (neonatal intensive care unit), base (special care nursery), and community — and obtained a similar twofold mortality excess in each stratum. Although perinatal mortality is uncommon among term singleton babies (1 in 1000 in private hospitals versus 2 in 1000 in public hospitals), any excess risk should be investigated and the reasons for it understood. The results from Robson et al’s article might be due to confounding, but they should not be dismissed and should be investigated with more detailed clinical data. Even if all the excess risk is due to confounding, explicit confirmation of this would be extremely useful. To this end, the Statewide Maternity and Neonatal Clinical Network in Queensland Health will undertake clinical review and classification of term singleton deaths according to national guidelines and collaborate with the Australian Maternity Outcomes Surveillance System (AMOSS) to enhance prospective monitoring of late gestation perinatal deaths nationally.
Read full abstract