Licorice Gargle Trial: Analysis Notebook and Report

Summary

The questions

The PI asked, at the intake meeting (Chapter 1):

“did it reduce sore throat afterwards?”

“does it last past the recovery room?”

Every comparison is licorice against sugar water, among patients randomized to them, with any sore throat at rest (a score above zero on the zero-to-ten scale) as the outcome.

Did it reduce sore throat?

Yes. Licorice reduced sore throat in the recovery room. Thirty minutes after arrival in the recovery room, 42 of 116 (36%) patients who gargled sugar water had a sore throat at rest, against 22 of 117 (19%) who gargled licorice: a risk difference of -17.4 percentage points (95% CI -28.7 to -6.2). The marginal odds ratio is OR 0.41, 95% CI 0.22 to 0.74 (p = 0.003).

Adjusting for baseline covariates gives a conditional odds ratio of OR 0.37, 95% CI 0.19 to 0.71. Filling in the missing outcomes of patients in the file in the way least favourable to licorice gives OR 0.43, 95% CI 0.24 to 0.78. Also counting the 1 enrolled patient absent from the file as a licorice patient with a sore throat (post hoc) gives OR 0.45, 95% CI 0.25 to 0.81. None of these checks moves the confidence interval to include one. Details are in Chapter 6, with the bounds in Chapter 2.

Does it last past the recovery room?

The data are consistent with a benefit that lasts past the recovery room. The marginal odds ratio is OR 0.21, 95% CI 0.10 to 0.42 at ninety minutes, OR 0.32, 95% CI 0.18 to 0.57 at four hours and OR 0.39, 95% CI 0.22 to 0.70 on the first postoperative morning, when the risk difference is -19.1 percentage points (95% CI -30.7 to -7.6). At every assessment the confidence interval for the odds ratio lies below one and the interval for the risk difference below zero. These are descriptive estimates with no formal test across time and no multiplicity adjustment (D-003, D-013).

Summarising all four assessments in one number, which assumes a common effect and does not itself answer this question, the population-averaged (marginal) odds ratio is OR 0.34, 95% CI 0.22 to 0.53. The patient-level (conditional) odds ratio from the mixed model is OR 0.14, 95% CI 0.06 to 0.35. It is further from one because it compares the same patient under each gargle, not because the effect is larger (D-008). Details are in Chapter 7.

What is still open

These questions for the client could not be answered from the data (Chapter 1).

• Enrolled but not in the file. The codebook counts 236 enrolled; the file has 235. Why is not yet known. If a randomized patient is missing, the analysis is not fully intention to treat.
• No post-operative data. 2 patients in the file have no post-operative data. Why is not yet known; they are in the bounds above.
• Swallowing pain. Not analysed. The PI was undecided. Adding it now, after the results have been seen, would be a post hoc analysis and would be reported as one.
• Stratified randomization. Not yet known. If randomization was stratified, the unadjusted analysis gives conservative standard errors.
• Preoperative sore throat in the protocol. Whether the protocol excluded these patients is not yet known. The analysis keeps them (D-005).

How to read this book

It is both the lab notebook and the report. The Plan part was written before the results. Additions made after it was locked are marked as such, and the one made after results were seen is labeled post hoc. The Results part follows it, and the Record part holds every analysis decision and the research log. Every number in the Plan, Results and Summary prose is computed from the data when the book renders, so the text cannot drift from the analysis; make numbers checks this. The decision record and research log are hand-written logs, and say so.