Are ‘healthy cohorts’ real-world relevant? Comparing the National Child Development Study (NCDS) with the ONS Longitudinal Study (LS)

Abstract

<jats:p>Comparisons between cohort studies and nationally representative ‘real-world’ samples are limited. The NCDS (1958 British birth cohort) follows those born in Britain in a single week in March 1958 (n=18,558); and the ONS Longitudinal Study (LS) contains linked census data and life events for a 1% sample of the population of England and Wales (&gt; 1 million records; allowing for sub-samples by age, ethnicity, or other socio-demographic factors). Common country-and age-matched socio-demographic variables were extracted from the closest corresponding time-points, NCDS 55-year survey in 2013 (n=8107) and LS respondents aged 55 in 2011 (n=7052). Longitudinal associations between socio-demographic exposures (from the NCDS 46-survey in 2004 and LS respondents aged 45 in 2001) and long-term limiting illness (from NCDS 2013 and LS respondents 2011, aged 55) were assessed using logistic regression. The NCDS 55-year sample had similar characteristics to LS respondents aged 55 for sex and marital status, but the NCDS sample had lower levels of long-term limiting illness (19.7% vs 22.8%), non-white ethnicity (2.1% vs 11.7%) and living in South England (46.9% vs 50.1%), and higher levels of full-time employment (61.2% vs 55.2%), working in professional/higher managerial occupations (35.7% vs 29.2%), and living with a spouse (69.1% vs 64.9%), all p&lt;0.001. Nevertheless, longitudinal associations between socio-demographic exposures and long-term limiting illness were similar in the NCDS and LS samples (all tests of between-study heterogeneity in mutually adjusted models p&gt;0.09) suggesting these NCDS findings are largely generalisable to the population of England and Wales.</jats:p>

Reply to ‘Letter to the editor: Don’t forget survey data: “healthy cohorts” are “real world” relevant if missing data are handled appropriately’ by Richard Silverwood et al

“…In our paper, we noted that restoring representativeness may also become problematic if the target population is fluid compared to the study's original sample, for example, due to emigration and immigration (13.7% of the UK population was foreign-born in 2020; Office for National Statistics, 2020). We should underscore that there is little evidence to suggest that this is an issue for the NCDS, as demonstrated by Silverwood et al (2022) and others (Archer et al, 2020;Mostafa et al, 2021). We stand by our concluding remark that administrative data could plausibly be used to enhance the accuracy of estimates intended to approximate national populations (as other research is demonstrating, Douglas et al, 2018;Eggleston and Westra, 2020;Büttner et al, 2021), but agree the extent of which is yet to be established.…”

Childhood adversity, life course outcomes and a diamond jubilee

“…The Finnish dataset used here, though with a much higher sampling fraction, has features in common with the ONS Longitudinal Study of England and Wales. The latter 1% record linkage study was used by Archer et al (2020) in the previous issue of this journal to investigate possible response bias in NCDS.…”