Do meta-analyses of randomized controlled trials estimate the true population effects of multiple risk factor interventions?
More details
Hide details
Western University, Department of Epidemiology & Biostatistics, Western Centre for Public Health & Family Medicine, 1465 Richmond Street, London Ontario, N6G 2M1, Canada
University of Toronto Mississauga, Studies in Life Sciences Program, 3359 Mississauga Rd., Mississauga Ontario, L5L 1C6, Canada
Publication date: 2023-04-26
Popul. Med. 2023;5(Supplement):A32
Large reductions in cardiovascular disease (CVD) mortality since 1950 are a significant public health triumph. Explanatory models of declining CVD mortality trends in high-income countries attribute just under 50% each to population trends in medical/surgical interventions and risk factor reductions, leaving <10% unexplained. However, meta-analyses (MA) of multifactorial trials yield modest effect sizes that explain only a tiny fraction of the population declines. Arguably, entire population data are superior to MAs of Randomized Controlled Trial (RCT) samples given the methodological and practical problems of experimentally estimating the effect of simultaneously modifying several factors in real-world settings. Might MAs of multifactorial RCTs also underestimate the true population effects of multiple risk factor modifications in other common outcomes in older adults?

For cognitive impairment (CI) and unintentional falling (UF) in older adults, literature searches were conducted for i) MAs of multifactorial RCTs and ii) attempts to explain population trends in terms of changes over time in risk factors and clinical interventions.

For UF, while some well-done RCTs show clinically significant comparative reductions in both fall occurrence and the number of risk factors in intervention groups, MAs tend to show modest or even null results. While fewer multifactorial RCTs have been completed for CI, early MAs also show modest effects. No attempts were found to replicate the CVD trend analysis for UF or CI, possibly because neither UF nor CI outcomes, or trends in risk factors or clinical interventions, are available in population data like for CVD. This data incompleteness might be overcome with statistical models that interpolate partial time-series data from numerous sources to estimate the true population effects of multifactorial interventions.

Estimating the true population effects of UF and CI interventions may require more sophisticated methods than MAs of samples studied in complex and problematic multifactorial RCTs.