BackgroundRecent decades have witnessed an increase in mean maternal age at childbirth in most high-resourced countries. Advanced maternal age has been associated with several adverse maternal and perinatal outcomes. Although there are many studies on this topic, data from large contemporary population-based cohorts that controls for demographic variables known to influence perinatal outcomes is limited.MethodsWe performed a population-based cohort study using data on all singleton births in 2004–2008 from the North Western Perinatal Survey based at The University of Manchester, UK. We compared pregnancy outcomes in women aged 30–34, 35–39 and ≥40 years with women aged 20–29 years using log-linear binomial regression. Models were adjusted for parity, ethnicity, social deprivation score and body mass index.ResultsThe final study cohort consisted of 215,344 births; 122,307 mothers (54.19%) were aged 20–29 years, 62,371(27.63%) were aged 30–34 years, 33,966(15.05%) were aged 35–39 years and 7,066(3.13%) were aged ≥40 years. Women aged 40+ at delivery were at increased risk of stillbirth (RR = 1.83, [95% CI 1.37–2.43]), pre-term (RR = 1.25, [95% CI: 1.14–1.36]) and very pre-term birth (RR = 1.29, [95% CI:1.08–1.55]), Macrosomia (RR = 1.31, [95% CI: 1.12–1.54]), extremely large for gestational age (RR = 1.40, [95% CI: 1.25–1.58]) and Caesarean delivery (RR = 1.83, [95% CI: 1.77–1.90]).ConclusionsAdvanced maternal age is associated with a range of adverse pregnancy outcomes. These risks are independent of parity and remain after adjusting for the ameliorating effects of higher socioeconomic status. The data from this large contemporary cohort will be of interest to healthcare providers and women and will facilitate evidence based counselling of older expectant mothers.

Our population-based study suggests that severe stress to a mother during the first trimester may alter the risk of schizophrenia in offspring. This finding is consistent with ecological evidence from whole populations exposed to severe stressors and suggests that environment may influence neurodevelopment at the feto-placental-maternal interface.

C onfounding should always be addressed in studies concerned with causality. When present, it results in a biased estimate of the effect of exposure on disease. The bias can be negative-resulting in underestimation of the exposure effect-or positive, and can even reverse the apparent direction of effect. It is a concern no matter what the design of the study or what statistic is used to measure the effect of exposure.The potential for confounding can be reduced by good study design, but in non-randomised studies this is unlikely to resolve the problem fully. Hence statistical adjustment methods, to reduce the bias caused by measured confounders, are also frequently considered. Such adjustment presupposes that one knows which factors are confounders. However, recent literature on methods for identifying confounders suggest that these are not always obvious. Indeed, in pursuit of guidelines, authors have had to reexamine the meanings of confounding and confounders with some ambiguity and conflict emerging. This literature is reviewed and a recent modification to the traditional definition of a confounder, which emphasises causal rather than statistical relationships, is described and illustrated. Some well known problems in occupational epidemiology, arising from health related selection, are considered in the light of recent ideas.Control of confounding through study design is not addressed, nor is the article concerned with details of statistical methods for adjustment. An overview of design and analysis in relation to confounding by age may be useful additional reading.1 It is assumed that the reader has at least a basic knowledge of epidemiological methods. Unless otherwise stated, definitions and comments apply to all causal study designs including case-control studies. c DEFINITIONS ExampleConsider a study of the relationship between exposure to silica dust and lung cancer where the rate of lung cancer in exposed workers is twice that in unexposed subjects, giving a rate ratio (RR) of two. The RR is a measure of the size of the effect of silica exposure on risk; here it suggests that exposure to silica dust is a cause of lung cancer. However, there might be other explanations for the increased rate among the exposed: if 50% of exposed workers were lifelong tobacco smokers compared to 30% of unexposed subjects, then this difference might explain some of the increase. This would then suggest that the true effect of silica exposure is less than two and that the result, RR = 2, is positively biased; smoking might be labelled a confounder of the relationship between silica and lung cancer. A statistical adjustment method could be used to try to estimate the true, unconfounded effect of exposure.The traditional criteria for identifying confounders are the first three conditions C1-C3 in box 1. In the previous example, tobacco smoking fulfils all the criteria; it is a cause of lung cancer and it is correlated (associated) with silica exposure in this study population. There is unlikely to be a causal pathway (C3) l...

Confounding is a major concern in causal studies because it results in biased estimation of exposure effects. In the extreme, this can mean that a causal effect is suggested where none exists, or that a true effect is hidden. Typically, confounding occurs when there are differences between the exposed and unexposed groups in respect of independent risk factors for the disease of interest, for example, age or smoking habit; these independent factors are called confounders. Confounding can be reduced by matching in the study design but this can be difficult and/or wasteful of resources. Another possible approach-assuming data on the confounder(s) have been gathered-is to apply a statistical ''correction'' method during analysis. Such methods produce ''adjusted'' or ''corrected'' estimates of the effect of exposure; in theory, these estimates are no longer biased by the erstwhile confounders.Given the importance of confounding in epidemiology, statistical methods said to remove it deserve scrutiny. Many such methods involve strong assumptions about data relationships and their validity may depend on whether these assumptions are justified. Historically, the most common statistical approach for dealing with confounding in epidemiology was based on stratification; the standardised mortality ratio is a well known statistic using this method to remove confounding by age. Increasingly, this approach is being replaced by methods based on regression models. This article is a simple introduction to the latter methods with the emphasis on showing how they work, their assumptions, and how they compare with other methods.Before applying a statistical correction method, one has to decide which factors are confounders. This sometimes 1-4 complex issue is not discussed in detail and for the most part the examples will assume that age is a confounder. However, the use of automated statistical procedures for choosing variables to include in a regression model is discussed in the context of confounding. REGRESSION MODELS cAs a means of studying influences on a outcome Most introductions to regression discuss the simple case of two variables measured on continuous scales, where the aim is to investigate the influence of one variable on another. It is useful to begin with this familiar application before discussing confounder control.Suppose we are interested in describing the decline with age of forced expiratory volume in one second (FEV 1 ) in non-smokers and that data on both variables has been gathered from a crosssectional sample of a population. A statistical analysis might begin with a scatter plot of the data (see fig 1); then a model of the relationship in the population would be proposed, where the model is specified by a model form or model equation. The choice of model form should ideally be dictated by subject matter knowledge, biological plausibility, and the data. Suppose a linear relationship is proposed; then the model would have the general form:The three unknown quantities in this model-a, b, r-would then be estim...

Aims: To investigate the lagged effects of cold temperature on cardiorespiratory mortality and to determine whether ''wind chill'' is a better predictor of these effects than ''dry bulb'' temperature. Methods: Generalised linear Poisson regression models were used to investigate the relation between mortality and ''dry bulb'' and ''wind chill'' temperatures in the three largest Scottish cities (Glasgow, Edinburgh, and Aberdeen) between January 1981 and December 2001. Effects of temperature on mortality (lags up to one month) were quantified. Analyses were conducted for the whole year and by season (cool and warm seasons). Main results: Temperature was a significant predictor of mortality with the strongest association observed between temperature and respiratory mortality. There was a non-linear association between mortality and temperature. Mortality increased as temperatures fell throughout the range, but the rate of increase was steeper at temperatures below 11˚C. The association between temperature and mortality persisted at lag periods beyond two weeks but the effect size generally decreased with increasing lag. For temperatures below 11˚C, a 1˚C drop in the daytime mean temperature on any one day was associated with an increase in mortality of 2.9% (95% CI 2.5 to 3.4), 3.4% (95% CI 2.6 to 4.1), 4.8% (95% CI 3.5 to 6.2) and 1.7% (95% CI 1.0 to 2.4) over the following month for all cause, cardiovascular, respiratory, and ''other'' cause mortality respectively. The effect of temperature on mortality was not observed to be significantly modified by season. There was little indication that ''wind chill'' temperature was a better predictor of mortality than ''dry bulb'' temperature. Conclusions: Exposure to cold temperature is an important public health problem in Scotland, particularly for those dying from respiratory disease. M ortality rates for cardiovascular and respiratory disease typically exhibit distinct seasonal variation with the highest rates occurring in the winter months.1 For Scotland, the percentage summer to winter difference in weekly all cause mortality rates is estimated to be in the order of 30%.2 The main factor considered to be influencing the observed seasonal pattern is the relation between mortality and temperature. The association between low temperature and increased morbidity and mortality is well recognised.3 4 What is less clear is the exact nature of the relation. Research has shown that the effect of temperature on mortality can exhibit significant variation from region to region. 5 6 For example, some studies have reported a U or V-shaped relation between temperature and mortality with the maximum number of deaths occurring at each end of the temperature scale 7 8 whereas others have reported a more linear or reverse J-shaped relation, with mortality typically increasing as temperature drops.