Data-analysis
No interim data-analysis will be carried out. The data-analysis will be conducted by a biostatician at the end of the follow-up period.
Patients randomised to the non-surgical management group but who cross-over and undergo secondary surgery will primarily be analysed as they were randomised. Similarly, patients randomised to osteosynthesis but due to surgical technical difficulties are treated with primary hemiarthroplasty, will be analysed as randomised.
Patients that drop-out after the six months evaluation will be included in the analysis based on six months data ('last observation carried forward'). Drop-out of patients before the six months evaluation cannot be included in the analysis.
Descriptive statistics (mean and standard deviations, as well as median and 10 and 90 percentiles for continuous variables, and frequency and 95 percent confidence interval for binary variables) will be calculated for each of the three groups, and their relevant subgroups, at both time points. Relevant subgroups include:
-
Valgus-impacted versus classical four-part fractures
-
Cross-overs from non-surgical treatment to secondary hemiarthroplasty versus non cross-overs in non-surgical treatment group
-
Cross-overs from osteosynthesis to primary hemiarthroplasty versus non cross-overs in osteosynthesis group.
Continuous and binary baseline characteristics will be tabulated according to the three treatment groups. No statistical testing will be performed, but relevant imbalances will be noted and reported. Baseline characteristics include: age, ASA Score, patient's treatment preference, surgeon's treatment preference and fracture pattern (valgus-impacted or classical four-part).
The primary analyses of effect will be two analyses of variance (ANOVA) based on the overall Constant scores, and the Constant pain subscale score, measured at 12 months. We will use the last observation carried forward approach, excluding patients only because of missing data, and analyse all patients according to their randomisation (intention-to-treat analysis).
If one or both of the ANOVA analyses are statistically significant, we will subsequently perform a pair-wise testing based on t-tests:
-
Osteosynthesis versus non-surgical management.
-
Hemiarthroplasty versus non-surgical management.
-
Osteosynthesis versus hemiarthroplasty.
Thus, our protection against the risk of type 1 errors consists on a clear definition of the two primary outcomes and a single analysis ANOVA approach.
We will also conduct the following secondary analyses:
-
The above analysis will be conducted for the following two subgroups: valgus-impacted and classical fractures respectively.
-
In the osteosynthesis group we will conduct a subgroup analysis comparing patients with intraoperative change to hemiarthroplasty with those without.
-
We will repeat the primary analysis as a mixed model for repeated measurements (including data from six months and the predefined covariates: age, ASA-Score, and treatment preference at baseline).
-
The number of patients with Constant score ≥ 70 at month 12 for each group (and subgroup) will be compared with Chi-square tests.
The same analyses (except the third) will be performed for the other outcomes (Oxford Shoulder Score, Short Form-36).
Finally, we will perform a per protocol effect analysis in which only patients that were compliant with physiotherapy and training (attending > 75% physiotherapy sessions judged by physiotherapist) will be included.
The same analyses will be performed for the other outcomes (Oxford Shoulder Score, Short Form-36).
A three year follow-up analysis will be reported in a separate publication. The analyses will compare number of patients in each group that died, received shoulder surgery, etc.