O-6 Why missing data matter-a guide for palliative care practitioners. Issue Volume 8: Issue (2018)Supplement 2 (November 2018)

• Record Type: Journal Article
• Authors: Hussain, Jamilla; Currow, David; White, Ian; Bland, Martin; Johnson, Miriam
• Is Part Of: BMJ supportive & palliative care. Volume 8: Issue (2018)Supplement 2
• Page Start: A2
• Page End: A3
• Publication Date: 2018-11
• Subjects: Palliative treatment -- Periodicals; Terminal care -- Periodicals; 616.029
• DOI: 10.1136/bmjspcare-2018-hospiceabs.6
• Languages: English
• ISSNs: 2045-435X

Abstract

Abstract : Evidence-based palliative care is essential to improving the value of palliative care clinical practice. In order to provide this, we not only need more research in palliative care but also, crucially, we need better quality research. Missing data (MD) present a particular challenge to clinical studies as they reduce the power, precision and generalisability of study findings. Importantly, they can also introduce bias. Aims: This mixed-methods study aimed to (i) quantify the extent and risk of bias of MD in palliative care trials; (ii) explore the factors associated with MD, (iii) develop MD recommendations for patients, clinicians, researchers and policymakers. Methods: (i) Systematic review of MD in 108 palliative care trials (meta-analyses and meta-regression); (ii) individual participant-level data (IPD) meta-analysis of factors associated with MD in 10 palliative care trials (multi-level modelling); (iii) in-depth interviews with research personnel (thematic analysis). Results: (i) Systematic review: 23% of primary outcome data were missing at the primary end-point (95% CI 19, 27). Larger MD proportions were associated with increasing numbers of questions/tests asked of participants (odds ratio (OR) 1.2 per-doubling, 95% CI 1.1, 1.4) and longer study duration (OR 1.1 per-doubling, 95% CI 1.0, 1.2). There was evidence of differential rates of MD between trial arms, thus potentially introducing bias. (ii) IPD: Participants with previous MD (OR 17.2, 95% CI 8.6, 34.5) and poorer Karnofsky Performance Status (10-unit increase: OR 0.8, 95% CI 0.7, 0.9) were more likely to have MD. (iii) Interviews: themes included 'clinical vs research-role tension' and 'attention-to-detail vs attention-to-person'. Conclusions: MD pose a significant risk to palliative care trial findings but can potentially be reduced. It is therefore essential that multi-disciplinary team members can (i) critically assess the MD risk of bias and (ii) help to reduce MD – guidance on how to do these will be discussed.