Adjustment for time-invariant and time-varying confounders in 'unexplained residuals' models for longitudinal data within a causal framework and associated challenges.

'Unexplained residuals' models have been used within lifecourse epidemiology to model an exposure measured longitudinally at several time points in relation to a distal outcome. It has been claimed that these models have several advantages, including: the ability to estimate multiple total causal effects in a single model, and additional insight into the effect on the outcome of greater-than-expected increases in the exposure compared to traditional regression methods. We evaluate these properties and prove mathematically how adjustment for confounding variables must be made within this modelling framework. Importantly, we explicitly place unexplained residual models in a causal framework using directed acyclic graphs. This allows for theoretical justification of appropriate confounder adjustment and provides a framework for extending our results to more complex scenarios than those examined in this paper. We also discuss several interpretational issues relating to unexplained residual models within a causal framework. We argue that unexplained residual models offer no additional insights compared to traditional regression methods, and, in fact, are more challenging to implement; moreover, they artificially reduce estimated standard errors. Consequently, we conclude that unexplained residual models, if used, must be implemented with great care.

Improving the accuracy of birth notification data: lessons from the Birth to Ten study.

The aim of the present study was to evaluate the accuracy of birth notification data collected during Birth to Ten, a longitudinal birth cohort study based in the Soweto-Johannesburg Metropole. Photocopies of birth notification forms were obtained from three local health authorities (Soweto, Diepmeadow and Johannesburg) for 5 448 of the 5 460 singleton births that occurred during seven weeks between April and June 1990, to women resident in Soweto-Johannesburg. By comparing the data recorded on the three different types of notification forms used by delivery centres within the Metropole, it was possible to assess the consistency of data collected during birth notification. For 539 of the 2 120 births that occurred at Baragwanath Hospital in Soweto, it was possible to locate the original records of maternal age, gravidity, infant sex, date of birth, birth weight and gestational age at birth, contained in obstetric and neonatal hospital files. By comparing these records with information contained in the birth notification forms it was possible to assess the accuracy of birth notification data submitted for deliveries at Baragwanath Hospital. Each of the different notification forms contained a different selection of variables and failed to specify the precision with which continuous variables should be recorded. For 12 selected variables, the proportion of missing records ranged from 0.0% to 40.9%, and was highest for those variables (such as APGAR scores and parity) that were not required on all four forms. The percentage agreement between information recorded on these forms and the original hospital records was highest for the categorical variable infant sex (99.1%), while the accuracy of notification data for continuous variables ranged from 95.2% (maternal age) to 29.7% (gestational age at birth). The upper 95% confidence intervals for the mean absolute errors in gestational age at birth and birth weight were two to three times the units of measurement, at 2.4 weeks and 165 g, respectively. When these extremes of error were applied to data for all 539 children, the proportion classified as premature or post-term varied by up to 25.7%, while those classified as macrosomic, low or very low birth weight varied by 10.5%. This analysis illustrates the potential consequences of imprecise birth notification data on the apparent prevalence of premature and low birth weight babies, both of which are key indicators in maternal and child health. Improving the process of birth notification and standardising the format of birth notification forms would increase the consistency of birth notification data. Selecting variables that are established indicators of health status, and can be reliably measured, would help improve the utility and accuracy of birth notification data.