Statistical analyses of ordinal outcomes in randomised controlled trials: a scoping review

Table 4 Summary of the ordinal outcome

Design	Studies―n (%)
Type of ordinal scale used―n (%)
- Single-state scale	128 (89%)
- Transition-state scale	16 (11%)
Type of outcome―n (%)
- Primary outcome	59 (41%)
- Secondary outcome	85 (59%)
Type of outcome measure―n (%)
- Clinical outcomes	130 (90%)
- Life impact	11 (8%)
- Physiological outcome	1 (1%)
- Adverse event	1 (1%)
- Other	1 (1%)
Target parameter(s)―n (%)^a
- Odds ratio	78 (54%)
- Risk difference	31 (22%)
- Non-parametric procedure used	19 (13%)
- Risk ratio	13 (9%)
- Difference in means	11 (8%)
- Difference in medians	8 (6%)
- Other	8 (6%)
- Descriptive analysis only	6 (4%)
- Unknown	4 (3%)
A priori sample size based on the ordinal outcome―n (%)^a	32
- Analytical	26 (81%)
- Simulation^†	6 (19%)
Scale dichotomised for analysis―n (%)	47 (33%)
No. of categories in ordinal outcome―median (IQR, range)	7 (6–7; 3–23)

^aThe sample size (N = 98) here is not the same as the number of studies that analysed using an ordinal scale since one study planned to analyse the ordinal outcome in its original scale, but the assumptions of the statistical model were violated and was thus dichotomised for analysis. We retained a sample size of N = 98 instead of focussing on primary outcomes only as the sample size was determined from the secondary ordinal outcome in one study. There were 32 studies that estimated the sample size based on the ordinal outcome
^†Of those studies that used simulation to determine the sample size, four of these studies used an adaptive design (either early stopping or adaptive enrichment); four studies used frequentist inference, one used Bayesian inference alone and another study used a combination of both

ISSN: 1745-6215