Cost-effectiveness and cost-utility of a digital technology-driven hierarchical healthcare screening pattern in China

Model overview

A decision-analytic Markov model was constructed using TreeAge Pro 2022 (TreeAge Software; Williamstown, MA, USA) for the economic analysis of different screening strategies for cataracts. The model was built on a simulated cohort of 100,000 residents from 50 years through a total of 30 1-year Markov cycles, which is the common target population based on previous economic evaluations of eye disease screening in elderly individuals^10,11,28. The participants were allowed to enter the model as either healthy (free from cataracts) or unhealthy (with cataracts) and could progress to death from any health states. The primary outcomes were ICURs and ICERs. We assumed that the severity and postoperative visual acuity for bilateral cataract patients were similar. Based on the clinical practice guidelines, the severity of cataracts is assessed by slit-lamp photographs using LOCS grading standards³⁰. Increased cataract severity is strongly associated with a decrease in visual acuity³¹. Therefore, cataract patients’ BCVA is one of the common classification methods in clinical trials^11,32,33. We derived data of mild, moderate and severe cataracts based on patients’ best corrected visual acuity (BCVA) > 0.3, 0.1–0.3, and <0.1 respectively from published research^11,34. Moderate and severe cataracts were identified as referable cataracts^35,36. A Markov model was constructed to simulate the disease progression of mild and moderate to severe cataracts. During each cycle, the participants either stayed in the same stage or transitioned to the more severe phase (Supplementary Fig. 3). Accordingly, we defined three postoperative groups based on patients’ BCVA after surgery, namely the POST-1 group (BCVA > 0.3), the POST-2 group (BCVA 0.1–0.3), and the POST-3 group (BCVA < 0.1) for utility analysis^32,33. Since there was no significant change in postoperative visual outcomes during the long-year follow-up, we assumed that patients’ visual acuity and utility remained stable after surgery³⁶. Severe cataracts and the POST-3 group were combined as bilateral blindness for indirect cost calculations^10,28. We collected data from real-world eye screening programs and a literature search of prevalence, compliance, utility, and other parameters, most of which were specific to China or other LMICs. The costs of screening, examination, and treatment came from real-world eye disease screening programs and the ZOC.

Screening strategies and scenarios

No screening

Cataract patients might be diagnosed and treated upon opportunistically presenting at a hospital for another concern, without routine ophthalmic screening.

Telescreening

Residents over 50 were educated and invited to participate in a cataract telescreening in community-based clinics, including the visual acuity test and slit lamp photography. The data were transmitted to the ZOC telemedicine platform. One certificated ophthalmologist assessed the severity and provided an assessment report back to the primary care settings. The participants returned to collect the reports after one week. Once referable cataracts were detected, patients were referred to the ZOC for comprehensive examinations, diagnosis, and treatments. The others were suggested for follow-up.

AI screening

Residents over 50 were educated and invited to participate in AI screening in community-based clinics, including the visual acuity test and slit lamp photography. The AI models provided a real-time diagnosis and referable advice. Participants with referable cataracts were referred to the ZOC. The others were suggested for follow-up.

DH screening

Residents over 50 were educated and invited to participate in DH screening by using an app for AI cataract screening on smartphones at home. The photographs of ocular anterior segments were captured by themselves or family members as instructed. High-quality images were uploaded for AI diagnosis. Suspected patients were referred to community clinics for visual acuity tests and slit lamp photography assisted by primary eye care staff. Once referable cataracts were detected by AI, patients were referred to the ZOC. The others were suggested for follow-up (Supplementary Fig. 1).

Cataract prevalence, transition probabilities, and screening performances

The prevalence of senile cataracts is 26.66% and 28.79% in urban and rural areas, respectively, based on the systematic review and meta-analysis of large-scale epidemiological surveys of people over 50 years old in China³⁷. The annual transition probabilities were derived from the literature on the natural progression of cataracts in the Chinese population. In studies reporting multiyear incidences, the annual incidence was calculated as r = −log(1- p)/t, where r represents the 1-year incidence and p means the cumulative incidence over interval t³⁸. (Supplementary Table 2).

The model performances of DH smartphone-based screening and community-based AI/DH screening were derived from an ongoing national cataract AI screening investigation launched by the ZOC in 2018 to promote collaborative efficiency and medical resource coverage⁶. The AI cataract screening model involving multilevel clinical scenarios proved to be robust in a real-world evaluation. In the first stage of smartphone-based screening, the AI model achieved a sensitivity of 88.67% and a specificity of 89.33%. Next, in the community-based screening setting, the AI agent distinguished referable cataracts with a sensitivity of 94.80% and a specificity of 97.00%. The performance of telemedicine screening was collected from previous research and had a sensitivity of 95.00% and a specificity of 97.00%³⁹. (Supplementary Table 3).

Screening and treatment costs

Direct and indirect costs were included in the analysis. Direct costs included ophthalmic screening, examination, treatment, follow-up, transportation, food, and accommodation charges for further visits to specialized hospitals. Indirect costs consisted of one accompanying family member’s time and wage loss based on the time spent and per capita daily income in rural and urban areas. The costs of examination, treatment, and follow-up were obtained from the ZOC under the Chinese government’s control and varied little from institution to institution. All costs were expressed in US dollars at the exchange rate as of 2 November 2022 (1 USD = 7.2 CNY), listed in Supplementary Table 4.

Screening costs included equipment, labor, and transportation costs. The annualized cost of fixed assets was calculated by assuming a life span of 5 years, collected from the Finance Department and Procurement Center of the ZOC. Since the participants were over 50, we assumed that they did not produce a wage loss (Supplementary Table 5).

Patients with mild cataracts were suggested for follow-up till next screening. For referable patients, cost computation for examination, treatment and follow-up are listed in detail in Supplementary Table 6. For patients with bilateral blindness, the annual economic burden of indirect costs was assumed to be $3600 per person, including loss of labor resources and productivity of caregivers, based on previous research²⁸.

Other parameters (compliance, utility, mortality rate, and threshold)

We assumed that 98% of residents had access to a smartphone and could use the app for AI cataract screening on their own or with assistance from family members based on the coverage of mobile phones and 5G network in China^40,41. Compliance with telescreening and AI screening in community-based clinics was derived from a previous study that indicated 95% compliance in rural and 90% compliance in urban settings^10,11. Additionally, a randomized controlled trial (RCT) study suggested that the hospital referral adherence of AI screening and traditional screening was 52% and 40%, respectively^7,10,11. Considering that patients in the DH screening group had received two positive results and referral reminders, once home-based self-screening feedbacks more than other groups, a reasonable higher referral adherence rate of 62% was used in this group. Compliance of surgical therapy was 91% and 80% in urban and rural settings, respectively¹¹. For those who fail to participate in the screening program, or don’t adhere to referral or treatment, the possible results can be natural progression of cataracts; otherwise, they can also be diagnosed and treated in opportunistically case finding or next screening cycle^10,42.

The utility of healthy individuals without cataracts was defined as 1²⁸. Patients with mild, moderate, and severe cataracts have utility values of 0.60, 0.45, and 0.26, respectively⁴³. The utility values of the POST-1, 2 and 3 groups were 0.75, 0.55 and 0.53, respectively, based on previous research⁴³.

Age-specific mortality was obtained from the China Population Census Yearbook 2020 from the National Bureau of Statistics⁴⁴. According to previous research, increased odds of mortality for patients with cataracts and no difference after surgery were also accounted for (Supplementary Table 3)^45,46. The discounted cost and utility rate was 3.5% per annum^10,47.

According to the WHO, the definition of being cost-effective refers to interventions that cost less than three times the per capita gross domestic product (GDP). The highly cost-effective strategy refers to interventions that cost less than the per capita GDP⁴⁸. The per capita GDP was calculated for urban ($13,919) and rural ($10,552) China based on the 2022 overall per capita national GDP ($12,741), urbanization rate (0.65), and urban-rural ratio (2.45) of per capita disposable income using the following formulas^10,11,49:

The per capita GDP of urban China

The per capita GDP of rural China

As a result, the thresholds of willingness to pay (WTP) were $41,757 and $31,656 per quality-adjusted life year (QALY) gained for urban and rural China, respectively. Notably, if the ICUR or ICER was negative with fewer costs spent and more benefits gained, the strategy was defined as dominating⁴⁷.

Primary outcomes

The primary outcomes were ICURs and ICERs, calculated using the following formulas:

Sensitivity analysis

We performed extensive deterministic sensitivity analysis and probabilistic sensitivity analyses to assess the robustness of the main outcomes. Fluctuation ranges of 10% (probability data including prevalence, sensitivity, specificity, utility, transition probability, etc.), 20% (costs of examinations, treatments, follow-up, etc.), and 50% (screening costs and indirect costs for blindness) were set for sensitivity analysis¹⁰. Tornado diagrams showed the parameters that had the greatest influence on the ICURs. Probabilistic sensitivity analysis evaluated the impact on the results by taking 10,000 random samples from the probability distribution of each parameter. The methods and results conforming to the Consolidated Health Economic Evaluation Reporting Standards were listed in Supplementary Table 7.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.