Strategies for reporting studies with missing patient-reported outcome (PRO) data to minimise the potential for biased interpretation of findings
| Category | Topic | Specific recommendation | N recommendations* | Potential drawbacks | Source/s: first author (year). Full citations are provided as Online Supplementary Appendix C |
|---|---|---|---|---|---|
| Reporting—trial reports enable readers to interpret the possible impact of missing PRO data on findings | Report PRO data collection methods (these may shed light on strategies used to minimise, or potential relationships with, missing PRO data) | Mode of administration of PROs | 1 | None—however, level of detail must be balanced with word limit restrictions. | Revicki (2007) |
| Staff training | 1 | None—however, level of detail must be balanced with word limit restrictions. | Revicki (2007) | ||
| Participant training/education | 1 | None—however, level of detail must be balanced with word limit restrictions. | Revicki (2007) | ||
| Study power calculation and power achieved for the PRO analysis (Has missing data led to substantial loss of power for PRO analyses?) | 1 | None | Revicki (2007) | ||
| Report analysis methods used | PRO analysis methods | 2 | None—however, level of detail must be balanced with word limit restrictions. | Bernhard, Cella (1998), Revicki (2005) | |
| Assumptions of PRO analyses, including assumptions about missing PRO data | 2 | None—however, level of detail provided must be balanced with word limit restrictions. | Bell (2014), Revicki (2005) | ||
| How missing PRO data was handled for the analysis | 9 | None | Calvert (2013), Chassany (2002), Machin (1998), Machin (1998), Noyez (2011), Revicki (2005), Staquet (1996) | ||
| Sensitivity analyses (How has missing data impacted the findings?) | 2 | None—however, level of detail must be balanced with word limit restrictions. | Bell (2014), Revicki (2005) | ||
| Describe the sample | Clinical and demographic characteristics, including baseline PRO scores | 5 | None | Hewitt (2010), Noyez (2011), Revicki (2005) | |
| Compare participants with and without missing data | 4 | None—however, level of detail provided must be balanced with word limit restrictions. | Dumville (2006), Hewitt (2010), Sprangers (2002), Revicki (2005) | ||
| Flow diagram (for PRO study), including rates and reasons for non-completion | 1 | None—however, level of detail must be balanced with word limit restrictions | Revicki (2005) | ||
| Report missing data details | Compliance definitions (What was considered a missing response? How was PRO assessment compliance measured?) | 1 | None—however, level of detail must be balanced with word limit restrictions. | Lee (2000) | |
| Report the expected PRO completion rate (number of participants alive and on the study per time point)70 | 3 | None | Bernhard (1998), Lee (2000), Revicki (2007) | ||
| Report rates of missing PRO data | 26 | None | Bell (2014), Bernhard, Cella (1998), Calvert (2013), Chassany (2002), Fallowfield (2005), Flores (2004), Kaasa (2002), Lee (2000), Luo (2008), Machin (1998), Noyez (2011), Revicki (2005), Revicki (2007), Staquet (1996), Walker (2003) | ||
| Report reasons for/type of missing PRO data | 15 | None—however, level of detail provided must be balanced with word limit restrictions | Bernhard, Cella (1998), Calvert (2013), Chassany (2002), Deo (2011), Fallowfield (2005), Flores (2004), Lee (2000), Macefield (2013), Machin (1998), Noyez (2011), Revicki (2007), Sprangers (2002), Staquet (1996), Walker (2003) | ||
| Potential bias due to non-response/impact on generalisability | Authors should consider and report how missing data may have impacted the generalisability of findings. | 6 | None—however, level of detail must be balanced with word limit restrictions. | Bell (2013), Klee (1999), Machin (1998), Revicki (2005) |
*Some sources may have provided a recommendation more than once.