Interviews were based on the use of interview guides with prompts, allowing a participant-led and flexible approach to data collection. Interviews were structured around a core set of topics, including introductions, patient or family members’ experience of MDR-TB, as well as diagnosis and treatment; and their views on recommendations for future treatment provision. Conversation around these topics was participant-led and naturally flowed. Prompts and probes allowed particular topics to be explored in more depth, for example on asking patients if they spoke about having MDR-TB openly, it was then possible to explore their response to ascertain whether there was a fear of negative response or stigma. This approach allows for exploration of particular topics whilst still being participant-led and not risking asking leading questions. Topics used in health care provider and key informant interviews were slightly different, including an examination of the current system of treatment and care, acceptability of home-based care, adherence and stigma. An interpreter was used to translate questions and responses from Luo to English and vice versa in several interviews and discussions. A separate independent interpreter was used to re-translate interview recordings where possible to ensure validity of interpretations. Data were analysed throughout the entire course of the research, in that from the moment data were being generated the “thinking and theorising” began [20]. This iterative process of data collection and analysis enabled adaptation of topic guides and testing of emerging themes. For example, following analysis of initial patient interviews a code around patient adherence to MDR-TB treatment emerged, which led to this being added as a specific topic within interview guides in order for this to be explored further. Data were managed initially through verbatim transcription of all recorded conversational interviews. Systematic analysis of transcripts was conducted to identify codes, relevant themes, patterns and concepts compatible to a deductive grounded theory approach. A framework analysis was used to subdivide the data as well as assign categories to ascertain the most significant themes and patterns, with relationships between constructs, as well as deviant cases, being identified [20,21].