Intervention bias: presence and impact in trials of socially-complex health care interventions

Article type
Authors
Lindsay B
Abstract
Background: Systematic reviews of socially-complex interventions rely on trial reports. However, there is a risk that many such trials suffer from intervention bias which is not adequately identified by reviewers’ assessment tools and quality measures. This poster reports on intervention bias in a study of 47 randomised controlled trials (RCTs) of socially-complex nursing interventions. Objectives: This study evaluated four key characteristics of RCTs of nursing interventions. Three characteristics are reported here: the ‘intervention sample’, the ‘intervention definition’ and the involvement of intervention nurses in other aspects of the research process. Methods: Forty-seven papers (reporting on 43 RCTs) met the inclusion criteria. Each paper was assessed for information relating to intervention characteristics. Data analysis was undertaken using Microsoft Excel. Results: In 11 papers, it was not possible to identify how many intervention nurses were included in the RCT. Where the intervention sample size was identifiable (36 papers), it varied from 1 nurse to 20 nurses (mean of 3.28 nurses per RCT, median of 2). Thirty-nine papers provided insufficient information about qualifications, experience and preparation to replicate these aspects of the intervention. No paper provided sufficient information about the personal characteristics of the intervention nurses (e.g. age, gender, ethnicity) to enable future studies to replicate these aspects. This is in stark contrast to the detailed information provided in most reports of RCTs regarding the characteristics of members of the intervention and comparison groups. Six papers clearly showed involvement of one or more intervention nurses in other aspects of the RCT: as co-authors of the paper, as study designers, and/or in recruitment or randomisation. No paper stated overtly that the intervention nurses had not taken part in any aspect of trial design, data analysis or report writing. Conclusions: Small numbers of intervention nurses, and lack of transparency regarding intervention characteristics and intervention nurse involvement in data collection or analysis, result in potentially high levels of intervention bias in this group of studies. This bias has major implications for reliability and validity of trials and, consequently, for systematic reviews that include these trials.