TreeAge Pro (TreeAge Software) was used to build a decision-analytic Markov model to compare the actual cost, effectiveness, and utility of manual grading telemedicine screening and AI-based assessment for DR (Multimedia Appendices 1 -4 ). The incremental cost-effectiveness ratio (ICER) and incremental cost-utility ratio (ICUR) were calculated as the primary results. The effectiveness was defined as years without blindness per 100,000 people with DM, and the utility was evaluated by quality-adjusted life years (QALYs). Although all residents with DM could participate in our community-based screening, the majority were older people [15 (link),16 (link)]; therefore, a hypothetical cohort of community residents with DM was followed in the model from the age of 65 years through a total of 30 one-year Markov cycles [5 (link)]. The characteristics of the simulated cohort were extracted using the Shanghai Digital Eye Disease Screening Program (Table 1 ).
Individuals were enrolled as healthy (free from DR) or unhealthy (experiencing DR) and could die due to any reason. According to the English National Screening Program for Diabetic Retinopathy, a Markov model was constructed that included non-STDR, STDR, and DME [5 (link),15 (link)-17 (link)]. The category was assigned based on the DR grade in the worse eye. During each 1-year cycle, an individual had a risk of progressing to the more severe stage or staying in the same stage. However, the model does not allow returning to an earlier stage even with treatment because of the nature of the disease. Moreover, the treatment can only decrease the probability of progression to the next stage. The prevalence of DR, the incidence of DR (including STDR and DME), transition probabilities, characteristics of DR screening tests, referral and treatment compliance, utility, mortality, and other relevant parameters were collected from published studies specific to Shanghai, other cities in China, and other Asian regions, as well as unpublished data sources (eg, Shanghai Digital Eye Disease Screening Program). The costs of screening, ocular examinations, and treatment were all derived from a real-world eye disease screening program in Shanghai and the unified health care service pricing of the Shanghai Municipal Health Commission. The parameters used in the basic analysis and the ranges used in the sensitivity analyses are listed in detail inMultimedia Appendices 1 -4 .
Individuals were enrolled as healthy (free from DR) or unhealthy (experiencing DR) and could die due to any reason. According to the English National Screening Program for Diabetic Retinopathy, a Markov model was constructed that included non-STDR, STDR, and DME [5 (link),15 (link)-17 (link)]. The category was assigned based on the DR grade in the worse eye. During each 1-year cycle, an individual had a risk of progressing to the more severe stage or staying in the same stage. However, the model does not allow returning to an earlier stage even with treatment because of the nature of the disease. Moreover, the treatment can only decrease the probability of progression to the next stage. The prevalence of DR, the incidence of DR (including STDR and DME), transition probabilities, characteristics of DR screening tests, referral and treatment compliance, utility, mortality, and other relevant parameters were collected from published studies specific to Shanghai, other cities in China, and other Asian regions, as well as unpublished data sources (eg, Shanghai Digital Eye Disease Screening Program). The costs of screening, ocular examinations, and treatment were all derived from a real-world eye disease screening program in Shanghai and the unified health care service pricing of the Shanghai Municipal Health Commission. The parameters used in the basic analysis and the ranges used in the sensitivity analyses are listed in detail in
Free full text: Click here
