Statistical analyses

All analyses were conducted at the patient level and were based on individual items and on scales. Item-based analyses consisted of the calculation of the number and percentage of patients answering each of the response categories in each item. Scale scores were calculated as the percentage of the maximum score achievable on all items, with scores ranging from 0 to 100. For all the scales, higher scores suggested higher levels of patient safety. For multi-item scales, where responses were missing for more than 50% of the items, the whole scale was scored as missing; otherwise a score was derived using the available items without any imputation. Scale-based analyses were restricted to the scales showing the best psychometric properties in each of the five PREOS-PC domains (Appendix 1), consisting of the calculation of weighted scores’ mean and standard error (SE).

Inverse probability weights, related to likelihood of response, were applied in analysis to produce results more representative of the full practice populations, not just the patients who participated. For each participating practice, data were extracted on the sex and age distributions of the patients registered. Subsequently, separate sex and age probability weights were computed for each practice. For example, if data were received from 30 male patients from a practice with 3000 male registered patients, the weight was calculated as 3000/30 = 100 (so each male in the sample would represent 100 males at that practice). The sex and age weights were then multiplied and rescaled for the weighted samples to match the practice list sizes.

Tables report both unweighted and weighted (in square brackets) results for questionnaire items and scales; results in the main text are weighted. In general, weighted results did not substantially differ from unweighted results.

All data manipulation and analysis were carried out in Stata 12.1.