It was not possible to control for baseline cost in multivariate analysis as these were not recorded.
Other possible allelic configurations for baseline segments, which have equal copy numbers for each allele (e.g. φ, PPMM), are likely to be rare, and currently we do not model them.
In subsequent planned analyses, we adjusted for possible baseline differences relevant for the primary outcome.
For that reason, and because it is not possible for practitioners at baseline to determine with high accuracy which patient will eventually be classified as RA within the first 5 years of follow-up, we performed the rest of the analysis presented here on the entire cohort of 640 patients, without stratification by the classification at 5 years.
Both techniques can adjust for possible baseline imbalance, and potential modifiers (e.g. dementia type and severity, gender caregiver, relation with patient, patient centre/counselor, family type (e.g. conflictions versus harmonious, close versus distant).
After adjusting for possible baseline confounding factors, the odds ratio for treatment success in the switched cohort compared with controls was 0.29 (95% confidence interval [CI], 0.19 to 0.44; p < 0.001) and for unsuccessful treatment was 1.92 (95% CI, 1.47 to 2.56; p < 0.001).
If adjustment for possible baseline differences among patients (as well as for oncologists and centres) is needed, an analysis of covariance will be done.
However, to control for possible baseline differences, we used a randomized crossover design and controlled for the order of interventions in the statistical analysis.
Where possible we will adjust for baseline measurement of patient safety culture.
