Version 9

Interviews were conducted in French and audio recorded, transcribed verbatim and
stripped of identifying information before analysis. Data were analyzed with a
conventional content method^{17 (link)} in an inductive
way.¹⁸ Two female researchers (MG, master-level psychologist and
LS, master-level medical student) conducted the initial coding independently.
Both researchers met to merge their codes and create a codebook that would
include the common codes; idiosyncratic or redundant codes were grouped or
removed. A third senior female researcher (VG, psychologist, PhD-level senior
researcher) tested out the codebook by coding independently 2 interviews, and
the codebook was adapted according to her feedback. Two researchers
(master-level psychologist and master-level medical student) double-coded 10% of
the interviews independently with the final codebook; any discrepancies were
addressed until a consensus was found. This process was conducted until
obtaining an adequate intercoder concordance of more than 80%.¹⁹ A single
coder (master-level medical student) independently coded the remaining
interviews and explored the overreaching themes. The software used was Atlas.ti
version 9 (ATLAS.ti Scientific Software Development GmbH).

Patient demographic information was collected by review of the electronic health record and via survey questions within the interview. Our multidisciplinary study team with expertise in implementation science, geriatrics, and critical care created the interview guide. All questions were open-ended and designed to stimulate conversation on cognitive screening following critical illness (Appendix A for our interview guide, http://links.lww.com/CCX/B195). When available, we invited caregivers or additional family members to provide input as well. The interviews were conducted via telephone by author (A.K.), a medical student with training in qualitative research methods who was naive to the research participants. Interviews were audio recorded and transcribed verbatim with identifying data removed. Transcripts were compared with the audio recordings and edited for accuracy, and then imported to ATLAS.ti Version 9 (ATLAS.ti Scientific Software Development GmbH, Cologne, Germany) for coding and analysis.

The quantitative data analyses were univariate, bivariate, and multivariate. Univariate analysis was attempted to calculate the frequency distribution of each variable. The chi-square test was used for bivariate analysis, which involved creating a 2×2 table between each independent variable and the dependent variable to calculate the prevalence ratio (PR). In a stepwise backward elimination, all variables with p-values less than 0.25 were included. A multivariate logistic regression model was constructed by adjusting for all previously identified covariates from the stepwise backward analysis. A p-value <0.05 was considered to indicate statistical significance. The outcomes of the regression analysis were presented as adjusted PRs with 95% confidence intervals. For the statistical analysis of quantitative data, SPSS version 25 (IBM Corp., Armonk, NY, USA) was used.
Qualitative data analysis was conducted using Atlas.ti version 9 (ATLAS.ti Scientific Software Development GmbH), which is a tool for qualitative research content analysis. The results of the in-depth interviews with informants were transcribed into the Atlas.ti program. The data were identified and grouped using a main code (predisposing, enabling, or reinforcing factor) as well as subcategories (infrastructure, punishment, policy, COVID-19, prevention and control team, and their duties and responsibilities).