Participant population
Comparisons of the outcomes between the trial arms will use the all-randomised population, under the intention-to-treat (ITT) principle with all participants analysed according to the trial arm they were randomised to. Adherence will be assessed based on the percentage of operations that were performed using the approach that was allocated at randomisation, and percentage of follow up assessments that were completed within the pre-specified data collection windows. All deviations from the study protocol will be reported. The number of participants with protocol deviations will be reported descriptively by trial arm. The pre-specified data collection window for the early post-operative follow up is POD 3 or POD 4. For the POD 120 follow up, the data collection window is between POD 110 and 130.
Outcome measurements collected outside the pre-specified windows will be excluded from the main analyses. Additional analyses will be carried out including these measurements. Main analyses of all primary and secondary outcomes at POD 3 and POD 120 will be on the complete case data and will be repeated to include imputed data, using multiple imputation.
Where an outcome can be completed by proxy (i.e. OHS and EQ-5D-5L), these proxy data will be included together with participant-reported data in the main analyses. Those participants for whom outcome data collection was by proxy will be excluded in an additional analysis.
Levels of confidence and p-values
All statistical tests and confidence intervals (CIs) will be two-sided. All between-group comparisons will be presented as the estimate with two-sided 95% CI and p-value. Statistical significance will be set at the 5% level. Results of all between-group comparisons of continuous outcomes will be checked for validity using bootstrap methods.
Unadjusted and adjusted analyses
Unless stated otherwise, analyses of all outcomes will be adjusted for the stratification variables hospital site and cognition level, and pre-fracture characteristics age (continuous), gender, place of residence (categorical) and co-morbidities (American Society of Anesthesiologists (ASA) score, grouped into 1 or 2, 3, 4+). Unadjusted analyses will also be reported. The adjusted analyses will be considered to be the main analyses.
Multiple testing
No adjustments will be made for multiple testing, and the secondary outcomes will be considered exploratory.
Missing data
For the primary outcome and secondary outcomes, at each follow up time point, the percentage of missing data will be reported, by trial arm and overall.
Every effort will be made to collect outcome data within the pre-specified data collection windows. However, data collection will still be attempted outside these windows and the data will be used for the purposes of additional analyses.
Where participants have missing data on a subset of items for a given measure a decision will be made on the minimum number of items that should be responded to for a total score to be obtained. This will be done in advance of the database being locked and blind to knowledge of which arm the participants are allocated to. Any measure-specific rules for obtaining total scores when items are missing will be used. Specifically, for the primary outcome OHS, if one or two items are missing these will be replaced with the mean of the remaining items, and a total score calculated. If more than two items are missing, a total score will not be calculated and that participant will not be included in the main analysis [14].
The main analyses will be based on complete case data. Additional analyses will be carried out for all outcomes, based on multiple imputation. Multiple imputation will be used to impute missing data on outcomes, under the assumption that data are missing at random according to Rubin’s rules, i.e. that missingness is accounted for by other variables within the dataset [15]. Missing data will be imputed using the chained equation approach. Predictive mean matching, in which imputed values are sampled only from the observed values, will be used [16]. A total of 50 imputed datasets will be generated. Variables used to impute missing data will include all outcomes at all follow up time points, trial arm status, stratification variables, and variables included as adjustment factors in the regression models fitted to outcomes. While all participants will be included in the imputation process, no outcomes will be imputed for participants who die before the outcome could have been assessed. That is, if a participant dies before POD 3, no outcomes will be imputed for that participant; if a participant dies between POD 3 and POD 120, POD 3 outcomes will be imputed if missing; if a participant survives beyond the POD 120 follow up all missing outcomes will be imputed. Note: this principle is used in the process of obtaining multiple imputed datasets but does not apply to the composite approach used in additional analyses, described later in this article, in which those participants who have died before the outcome can be assessed are given a ‘worst case’ score for that outcome.
The multiple imputation will be carried out using the mi suite of commands in Stata. All statistical analyses will be carried out using Stata version 17.0 or higher [17].
Presentation of comparative analyses
For continuous outcomes (including the primary outcome), results will be presented as means and SDs in the two trial arms, crude (unadjusted) mean differences, and adjusted mean differences with 95% CIs and p-values. In addition, the effect size (adjusted mean difference divide by pooled SD) will be reported for the OHS at POD 120, for the main analysis only. For time to event outcomes, frequencies of events in the two arms will be presented, crude hazard ratios (HRs), and adjusted HRs with 95% CIs and p-values.
For binary outcomes, numbers and percentages in the two arms will be presented, with crude odds ratios (ORs) and adjusted ORs with 95% CIs and p-values.