Risk of bias for analysis 1.7 HRQoL (mental component) at maximum follow‐up.
| Study | Bias | |||||||||||
| Randomisation process | Deviations from intended interventions | Missing outcome data | Measurement of the outcome | Selection of the reported results | Overall | |||||||
| Authors' judgement | Support for judgement | Authors' judgement | Support for judgement | Authors' judgement | Support for judgement | Authors' judgement | Support for judgement | Authors' judgement | Support for judgement | Authors' judgement | Support for judgement | |
| Pressler 2016 | Low risk of bias | The triallists stated that “patients were randomly allocated 1:1 according to a computer‐generated allocation program code”. Although there was no mention of allocation sequence concealment, the fact that a computer‐generated program code was used, leads the review authors to judge that this would prevent the enrolling investigator and the participant from having any knowledge of the forthcoming allocation. The triallists state no differences for TG and UC with respect to age, proportion of females, the Society of Thoracic Surgeons mortality risk score, NYHA class, pre‐existing conditions, or medication (see Table I). | Low risk of bias | The nature of an exercise‐based CR intervention means that participants are aware of their intervention and it might not be possible to blind them, their carers, or the people delivering this intervention. Although the triallists did not state specifically if there were any deviations from the intended intervention, the flow of participants through the study (Figure 1 of 2016 paper) shows that no participant changed groups during the study and all changes to the intervention were either consistent with the intervention (i.e. drop out due to unrelated causes), or consistent with what could occur outside the trial context (i.e. dropout due to unwillingness to continue participation). Triallists reported: data were tested for their distribution and the primary end point and key secondary outcomes were normally distributed; thus it was decided to present data as mean (SD). Baseline differences between groups were analysed using the Student t test for independent samples for continuous variables and the Fisher exact test for categorical variables. Between‐group differences in changes of the primary and secondary end points were analysed using the Student t test for independent samples. For relevant effect sizes (mean differences between study groups), 95% CIs are presented. Because of the low number of dropouts that were all unrelated to the intervention and the primary intention of the study to evaluate efficacy and safety of exercise, a per‐protocol analysis was performed. To account for observed differences in baseline parameters between the study groups, linear regression models using the change in the measure of interest as dependent variable and the study group as well as the baseline value of the measure of interest as independent variable were fitted to the data. |
Some concerns | The appropriate study population for an analysis of the intention‐to‐treat effect is all randomised participants; in this trial less than 95% of the participants in both the exercise group and the control group provided data for this outcome. The review authors judged that there was no evidence to suggest that the result was not biased by missing outcome data. The review authors judged that although reported reasons for missing outcome data were similar between both groups, there was a minimal difference in the proportions of missing outcome data between the two groups overall with data for 5 participants missing from the exercise group and data for 8 participants missing from the control group. Most of the missing data were missing because of death or other health conditions that were not necessarily related to the intervention. The circumstances of the trial did not make it likely that missingness in the outcome depends on its true value as more participants dropped out of the control group than from the exercise group. |
High risk of bias | HRQoL was measured using SF36 and KCCQ, both of which are validated questionnaires. The review authors noted that measurements were made in the same way for both intervention and control. The person reporting this outcome was the participant; participants were aware of the intervention they received. Collation and analysis of the results was most likely carried out by study personnel who were blinded to the intervention received by study participants. The review authors judged that the assessment by the participants could have been affected by knowledge of the intervention as it is subjective. Exercise is generally believed to be a beneficial intervention in this patient group and it is possible that participants might have considered their HRQoL would be better if they had the intervention. | Low risk of bias | The trial was registered on clinicaltrials.gov before it started (NCT01935297), with researchers’ prespecified intentions available in sufficient detail. There was no reason to believe that the data which produced this result were not analysed in accordance with the prespecified analysis plan. The triallists reported results for all time points at which this outcome was assessed with all extra analysis intended as exploratory analysis to provide additional information on the group‐specific impact of the intervention. The triallists reported all results. There was no reason to believe that this result might have been selected from multiple eligible analyses of the data. | High risk of bias | At least one risk of bias domain with judgement of 'High'. |
| Sibilitz 2016 | Low risk of bias | The triallists stated that “Participants were allocated 1:1 to intervention or control using computer‐generated allocation sequence with varying block sizes of 8, 6 and 4, concealed to the investigators by central telephone correspondence with the Copenhagen Trial Unit.” “There was no evidence of baseline imbalances.” The review authors judged that table 1 shows a good balance in patient characteristics between groups. |
Low risk of bias | The nature of an exercise‐based CR intervention means that participants are aware of their intervention and it might not be possible to blind them, their carers, or the people delivering this intervention. Although the triallists did not mention expressly if there were any deviations from the intended intervention, their CONSORT flow diagram of patients (Figure 1 in article) shows that all participants were analysed in their initial intervention group. The triallists reported: the analysis was based on intention to treat using a mixed model with repeated measures (MMRM) and adjusted for the stratification variable of LVEF, ensuring that missing data will not create bias as long as the values are missing at random. The MMRM was used for continuous outcomes (both primary and secondary outcomes). This model assumes normally (Gaussian) distributed residuals. However, the residuals were not normally distributed for the primary outcome, and therefore transformed. We used log‐transformation, and when log‐transformation did not result in normally distributed residuals, we used Box–Cox transformation. In the MMRM we assumed correlation within the individual patient, but no correlation between patients. |
Some concerns | The appropriate study population for an analysis of the intention‐to‐treat effect is all of the randomised participants; in this trial less than 95% of the participants in both the exercise group and the control group provided data for this outcome. The triallists stated that the dropouts were due to complications after surgery and withdrawal of consent; 9 people dropped out of the study before completing 1 month of exercise training or usual care, no suggestion was made that the intervention played a part in these withdrawals. The review authors judged there was no evidence to show that the results was not biased by missing outcome data. The review authors noted that there was a minimal difference in the proportions of missing outcome data between the two groups. Reported reasons for missing outcome data were similar between both groups. The circumstances of the trial did not make it likely that missingness in the outcome would depend on its true value as more participants dropped out of the control group than from the exercise group. | High risk of bias | HRQoL was measured using SF36, a validated questionnaire. The review authors noted measurements were made in the same way for both intervention and control. The person reporting this outcome was the participant and participants are aware of the intervention they received. Collation and analysis of the results was most likely carried out by blinded study personnel . “The trial applied central, stratified randomisation securing against selection bias, and blinded outcome assessment and blinded statistical analyses, reducing detection and interpretation bias”. The review authors judged assessment by the participants could have been affected by knowledge of the intervention, as it is subjective. Exercise is generally believed to be a beneficial intervention in this patient group and it is possible that participants might have considered their HRQoL would be better if they had the intervention. |
Low risk of bias | Trial was registered on clinicaltrials.gov before it started (NCT01558765), with researchers’ prespecified intentions available in sufficient detail. The review authors judged that there is therefore no reason to believe that the results would not have been analysed in accordance with the prespecified analysis plan. The triallists reported results for all time points at which this outcome was assessed within the scope of this article. The review authors noted that there was no evidence of multiple eligible analyses for this outcome. | High risk of bias | At least one risk of bias domain with judgement of 'High'. |