Article Text

PDF

176 Analysis of hierarchical data; Comparing meta-analytical to multilevel estimates
  1. G N Ntani1,
  2. H I Inskip2,
  3. D C Coggon2
  1. 1MRC, Lifecourse Epidemiology Unit, Southhampton, United Kingdom
  2. 2MRC, Southampton, United Kingdom

Abstract

Objectives To compare pooled risk estimates obtained by meta-analytical methods with the corresponding risk estimates derived by analysis of individual data, using hierarchical methods, and to explore explanations for possible differences.

Methods The data were obtained from a study of 12,426 participants from 47 occupational groups (mostly nurses and office workers) in 18 countries. Pain in the low back and wrist/hand that interfered with everyday activities and exposure to possible risk factors had been collected using standardised questionnaires. Unadjusted associations with potential risk factors were explored using logistic regression separately for each occupational group and then the 47 risk estimates for each risk factor were synthesised in a meta-analytical model. Risk estimates were also obtained from the individual data using multilevel logistic models. The multi-level estimates were then compared with the corresponding pooled estimates derived from the meta-analysis.

Results For most risk factors, the odds ratios were similar using the two methods, though the confidence intervals for the odds-ratio estimates obtained from the meta-analysis model were wider than the confidence intervals for those derived from the multilevel model. When disabling low back pain was used as an outcome, the mean of the ratios of the odds ratios derived from multilevel modelling to those derived from meta-analysis was 0.99 (range 0.87 to 1.07). When disabling wrist/hand pain was taken as the outcome, the mean ratio was 0.97 (range 0.84 to 1.11).

Conclusions In the analysis of these data, pooled risk estimates obtained by meta-analysis were very similar to those derived from multilevel analysis of the individual data. However, circumstances in which estimates from the two methods may differ will be discussed.

Statistics from Altmetric.com

Request permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.