164 6 Chapter 6 Blinding The nature of the intervention makes blinding of participants and healthcare providers delivering the intervention impossible. Hence, we judged all studies, except Trappenburg 2011 , which kept patients unaware of the primary study aim (postponed information), as having high risk of performance bias. Although blinding of patients and/or healthcare providers is impossible, outcome assessors in some cases could be blinded to participants’ allocation. Twenty- ve studies were judged as having low risk. These studies had outcome assessors that were adequately blinded for allocation, reported only on outcomes that were objective (i.e. mortality, hospitalisations), or had an outcome committee judging the outcomes. This made risk of detection bias highly unlikely. Outcome assessors were unblinded in 15 studies (Boxall 2005; Cambach 1997; Farrero 2001; Kalter-Leibovici 2018; Khan 2019; Ko 2009; Lenferink 2019; Lilholt 2017; Öztürk 2020; Rea 2004; Smith 1999; Tabak 2014; Theander 2009; Titova 2017; Vianello 2016), posing a high risk of bias. Twelve studies provided insu cient information and were judged as having unclear risk (Bendstrup 1997; Dheda 2004; Fernandez 2009; Gottlieb 2011; Kennedy 2013; Littlejohns 1991; Lou 2015; Mendes 2010; Sridhar 2008; Strijbos 1996; Wijkstra 1994; Wood-Baker 2006). For the remaining 25 studies, outcome assessors were blinded to group allocation. Incomplete outcome data We judged 35 studies as having low risk of bias, as they had low dropout rates, or dropout rates were balanced across groups for similar reasons. We considered 13 studies to have high risk of bias (Bendstrup 1997; Bernocchi 2017; Cambach 1997; Farrero 2001; Gottlieb 2011; Kessler 2018; Lilholt 2017; Lou 2015; Mendes 2010; Smith 1999; Tabak 2014; Vianello 2016; Wang 2017). Four of these 13 studies had larger dropout in the control group than in the intervention group. In Lou 2015, 1217 participants dropped out from the control group compared to 779 from the intervention group. Reasons were death and inability to perform the walking test. In Bernocchi 2017 , larger dropout rates in the control group were due to increased hospitalisations as a result of heart failure. Selective reporting We judged 39 studies to have low risk of reporting bias, meaning that all outcomes mentioned in the protocol or the clinical trial register were reported. Nine studies selectively reported outcomes speci ed in the protocol and/or in the methods section (Bourbeau 2003; Dheda 2004; Freund 2016; Gottlieb 2011; JimenezReguera 2020; Lilholt 2017; Littlejohns 1991; Smith 1999; Tabak 2014), or they changed operationalisation of the outcome (i.e. Physical Component Summary (PCS) subscore instead of SF-36 score to measure QoL) (Lilholt 2017). In three studies (Bourbeau 2003; Dheda 2004; Öztürk 2020), the authors observed no statistically signi cant di erences in outcomes and therefore did not present data. In Tabak 2014, outcomes were reported for only 3 months - not for 6 and 9 months - in contrast to the study protocol. This all points to the risk of selective outcome reporting. With the
RkJQdWJsaXNoZXIy MTk4NDMw