Year: 2004
Type: editorial
Abstract: In this issue's report on Australasian longitudinal birth cohort studies since 1970, Nicholson and Rempel1 paint an interesting yet chequered picture of outstanding research contributions interspersed with mediocre publication records. As Australia gears up for its largest and most complex study of Australian children ever, these reflections are timely. The Australian Federal Government recently committed $20.2 million to the Longitudinal Study of Australian Children2 (LSAC); commencing March 2004, a nationally representative sample of 10 000 children (5000 infants and 5000 4-year-olds) will be followed at least biannually over 8 years3. Bigger and better also holds sway internationally, with longitudinal studies such as the recently commenced UK Millennium Cohort Study of nearly 20 000 babies4,5 and the forthcoming US National Children's Study of more than 100 000 children6. Exciting times indeed for longitudinal researchers. See related article, p 87. As Nicholson and Rempel note, Australasian longitudinal studies have directly contributed to major and probably lasting policy and practice changes that have already benefited many children. They cite the standout examples of swimming pool fencing legislation, bicycle safety promotion, environmental lead reduction, and effective sudden infant death prevention. Although such clear-cut ‘action’ examples are relatively uncommon, longitudinal studies are often in the news. In an entertaining study, Bartlett et al. recently showed that this in itself is not necessarily good news, particularly where children are concerned7. Prominent British newspapers, both broadsheet and tabloid, systematically ‘underreported randomised trials, emphasised bad news from observational studies, and ignored research from developing countries’ published in the high-impact journals Lancet and BMJ. Nor is no news good news: compared to studies reporting on women's health, reproduction and cancer, children's studies get very little press − good or bad. Of the 163 research articles published in these journals in 1999 and 2000, 82 (50%) were the subjects of press releases; just 13 (8%) were actually covered by the Times or Sun newspapers. Alas, this is just one way in which longitudinal research may be erroneously reported; far more errors may lie in the published studies themselves. While longitudinal studies can be retrospective or experimental in design, Nicholson and Rempel focus on the classic cohort study − prospective, observational, and with multiple predictor/outcome timepoints. Such studies typically address questions of causality, stability, developmental trajectories, and risk. Since longitudinal studies virtually always collect data in distinct, cross-sectional waves, they are also often used as a source of prevalence data. This is both understandable and questionable. It is questionable, because longitudinal studies typically value retention at the expense of response rates; to reduce attrition (the greatest threat to internal validity in longitudinal studies), subjects may be explicitly (by design) or implicitly (by response rate) not representative of the general population. It is understandable, because the real payoff for the huge monetary investment in cohort maintenance and data collection may be a long time coming: as Nicholson and Rempel show, very low publication rates are typical across the early data collection waves. The very ‘successful’ studies gather pace with time. Prospective studies (unlike cross-sectional studies) can delineate temporal relationships. Many equate this to demonstrating causality. However, temporality is just one of the criteria for inferring causality (Table 1), the most important of which is experimental evidence8. Beta-carotene and lung cancer, hormone replacement therapy and coronary heart disease, Vitamin C and coronary heart disease − strong, consistent, longitudinal data supporting the protective value of each were later debunked in randomized controlled trials9. Davey-Smith, using these examples to conclude that ‘the results from robust experiment and fallible observation are clearly non-compatible’, went on to quote medical journalist James Le Fanu's suggestion that ‘closing down . . . departments of epidemiology could both extinguish this endlessly fertile source of anxiety mongering [i.e. longitudinal studies] while simultaneously releasing funds for serious research’9. This leaves stability, developmental trajectories, and risk − in which the prospective study is king. Longitudinal studies let us understand how certain outcomes develop over time, as well as the relationship between the individual development of an outcome variable and the individual development of other variables8. Multiple predictors and multiple outcomes can be examined, so that a single study can efficiently address questions of interest to diverse research and policy groups − often questions never dreamt of by the study founders. However, this also leaves such studies open to by-problems of data dredging, bias and confounding9. Each of these increases the likelihood that we will accept chance findings as true. We can at least partially avoid this through stringent criteria as to what may represent the ‘truth’. Sterne provides a well-reasoned argument that the ubiquitous P-value of 0.05, and even 0.01, usually does not provide strong evidence against the null hypothesis while P-values of less than 0.001 may usually be accepted as providing such evidence10. This requires a combination of larger studies and more precise measurement tools, fortunately both areas in which great strides have been made. Rigorous attention to both bias and confounding is challenging, but vital. In particular, ‘promiscuous’ exposures that seem to predict multiple outcomes but may actually reflect unknown confounders at work should always be replicated in other populations and preferably different countries9 − though this may work against our wish to also understand cross-cultural differences. On their wish-lists for future longitudinal research, the principal investigators surveyed by Nicholson and Rempel listed an increased focus on child-care, rural-urban differences, the influence of fathers, family factors, peers, social competence and genetics, and the study of ‘policy-relevant’ areas. To this may be added more study of children's time use, advocated by a 1997 workshop on large longitudinal childhood studies by prominent American and international researchers11. The LSAC aims to tackle most of these, despite the challenges each presents. Studying rural-urban differences requires complex sampling strategies and logistics, and sample sizes capable of examining such subgroup effects at a time when the rural population continues to steadily decrease. While a nationally representative sample should build pride and ownership of the study across Australia, examining differential outcomes on an interstate basis is likely to prove disappointing for the smaller states and territories because their share of the sample reflects their own population size. The study of child-care, father, and peer effects are likely to require multiple informants − logistically challenging at the best of times, but likely to be even more so in large, geographically diverse studies. Parents seem to be becoming more cautious (even fearful) of permitting their children to provide biologic − and especially genetic − samples for study. Interestingly, Michael Wadsworth, Director of the 1946 British birth cohort study for the last 30 years, conjectures that a faltering retention at ages 43 and 53 may in part reflect increased requests for biological measures, including DNA12. How do these challenges fit with the structure of ‘successful’ past Australasian studies? The two most-published are the Christchurch Health & Development Study with a median of 11 papers, and the Dunedin Health & Development Study with a median of 69 papers per data collection wave. Both were from New Zealand; both were of medium size (between one and two thousand children), were narrowly geographically circumscribed, culturally uniform, and far from nationally representative. Nonetheless, their findings have resonated internationally. On the other hand, very large national cohorts in other countries, such as the three British birth cohort studies, have also been stunningly successful over many years; these cohorts have been highly focussed, data collection has been woven into the existing health system wherever possible, and respondent burden strictly limited. Now, for the first time ever, insights regarding early life impacts on ageing and intergenerational effects are emerging from truly prospective data collected in these early cohorts. Finally, answering longitudinal questions often requires highly skilled teams of cutting-edge researchers, and the funding to support them − both historically in short supply in Australasia, and nowhere more so than in children's research. In 2002, the Australian National Health and Medical Research Council commenced its program of Capacity Building grants in population health research. These 5-year awards are designed specifically to support the development of researchers at postdoctoral level over a 5-year period, with some already awarded to research groups who study children. Such funding and training schemes offer real opportunities to fully and excellently capitalize on LSAC and other new studies. Further, though it is frequently noted that the gap between what we know and what we do may be larger than that between what we know and don’t know, the gap between what we have collected and what we have yet to analyse may be even greater. A larger, more secure workforce might also reap rewards in the secondary data analysis of existing, unanalysed longitudinal data.
Author: M Wake
Language: English
Subfield: Statistics and Probability
Field: Mathematics
Domain: Physical Sciences
Open Access status: bronze
OpenAlex ID: W2131910688