Data analysis

For quantitative data, we will report means and standard deviations for parametric data, medians, and inter-quartile ranges for non-parametric data. Demographic characteristics (such as age, gender) will be summarized using frequency tables (n (%)) for categorical variables. The patient survey responses will be descriptively analyzed using IBM SPSS Statistics 25.0. We will run descriptive statistics to report means and standard deviations of responses to each TFA construct, and descriptively report categorical data (e.g., demographic characteristics) in terms of frequencies and percentages, for each group. Any open-ended responses will be analyzed qualitatively using directed content analysis based on TFA constructs [30, 31]. Descriptive analysis at months 1, 3, and 6 will inform feasible iterations of the intervention based on these data. If the mean scores on any TFA construct for either population group are below the mid-point of the scale at months 1 and 3, we will use the findings to suggest potential additional strategies and address concerns in the intervention, but we will continue running the program given that it is an entirely new screening program. If the mean scores on any TFA construct are below the mid-point of the scale at month 6 (final feasibility evaluation time point), we will take this to be an indicator that some aspect(s) of the intervention should be further iterated for improved acceptability. If that is the case, we will convene a patient and health system advisory group to decide on whether and how to continue the program beyond the 6-month feasibility pilot and decide on approaches for optimization of the program and its delivery.

The feasibility outcomes obtained from the interviews with health care providers will be analyzed using content analysis guided by the TDF [43] using NVivo. Data codes will be generated by labeling one to two lines of text with a descriptive label and then subsequently sorting these into the TDF domains. Data will be compared within and across codes to assess the similarities, differences, and interrelations and refined accordingly. Codes representing similar thematic topics will be grouped; these will be defined and documented in a codebook. Interview transcript analysis will involve the following four steps: (1) familiarization; (2) coding participant responses to specific domains, as defined by the TDF; (3) generating sub-themes within each domain; and (4) grouping themes across domains [33, 44]. To verify the emerging analysis, a second analyst will review a preliminary set of themes to assess how well the data are represented and the relevance of data within codes and to the associated TDF constructs. Where differences in interpretation arise, the two analysts will discuss until agreement is reached and amendments will be made to the coding and codebook as necessary. The health care provider survey responses from the interviews will be descriptively analyzed and scale scores calculated, and open-ended responses will be summarized based on the TFA constructs and NIH framework.

To document and analyze the costs related with the tele-retinopathy intervention, we will obtain the intervention and implementation costs [21] from the community health center. The intervention cost includes expenses required to purchase and maintain the intervention, salaries for staff, and physician fees. We will use these data along with downstream costs incurred (i.e., patient-related costs, such as travel and parking costs as well as time missed from work), to inform future economic evaluation of the intervention.