THE CASCADE ANALYSIS TOOL FOR CONTINUUM OF CARE ANALYTICS: AN APPLICATION IN DIABETES CARE IN UKRAINE

any health programs require people to engage repeatedly with the services in order to benefit maximally—whether it is a diabetes control program, mother and child services, tuberculosis treatment, or cancer care. We are using the terms ‘continuum of care’ and ‘care cascade’ to describe the string of services people in need must utilize, starting from a first contact which might be hearing about the service, or getting screened. Ideally, people then progress through the service chain to get the intended health outcome, whether cure, meeting their treatment target, or even survival.

To capture the journey of health service clients best, we need to understand their movements in and out of the care continuum.
Are they lost because they dropped out, or did they die? Why did they drop out? Do some of the clients who drop out of care re-enter care? If so, do they come back spontaneously or through being sensitized, recalled, or incentivized by the program?
These are important questions and unless the movements of the clients are taken into account, it is challenging to understand what actions the program should take to improve the continuity of care, mitigate bottlenecks and facilitate clients' access. In order to support cascade analytics, a team of program implementers, software developers and data scientists developed the web-based Cascade Analysis Tool (CAT). The intention was to provide service planners with a tool they can use to analyze and improve technical and allocative efficiencies along the service continuum.

THE CASCADE ANALYSIS TOOL
The CAT is a mathematical model with a generic, flexible framework which means different types of service delivery cascades can be analyzed, from various health and development programs where repeat client interactions are required for the outcome to happen. The tool makes practical use of the increasing quantities of data on the costs, coverage, and impact of health and development interventions. By integrating various data types, it adds value to current analytics, and provides additional decision support.
It is designed to address key questions relevant to decision-makers around: ◼ Allocative efficiency (identifying the optimal allocation of funding among available interventions in order to get as many people as possible reach the intended outcome e.g., treatment success) ◼ Technical efficiency (identifying ways to reduce labour or capital input requirements without negatively affecting outcomes), and ◼ Cost-effectiveness (estimating the costs per person reaching an outcome such as treatment success).
The CAT is an open-access web-based app (ui.cascade.tools), described in detail elsewhere. 1 For ease of use, several cascade frameworks have been pre-designed in the CAT software for diabetes, hypertension, HIV, tuberculosis and cervical cancer (figure 2).
Methodologically, the CAT is based on a compartmental model structure which builds on the cascade concept (figure 1). People move between 'care status' compartments with certain probabilities, and the CAT therefore provides methods for setting these transition rates between compartments. There is an inbuilt optimization function to calculate what intervention efforts are The tool makes practical use of the increasing quantities of data on the costs, coverage, and impact of health and development interventions.
required to reach specific programmatic objectives (e.g., the maximum possible number of clients with treatment success).
Practically, CAT users either use a ready-to-use cascade framework or design their own. They can then enter data regarding the interventions, costs and effectiveness of the set of interventions.
Based on these inputs, the user can conduct a variety of analyzes under the 'status quo' and by assuming specific changes: The 'what-if' scenarios look at the effect of specific policy or program changes (e.g., an additional screening intervention). The CAT's 'optimize the cascade' function calculates the ideal coverage and funding levels for the cascade services that would allow the program to meet the set objectives.

APPLICATION OF THE CASCADE ANALYSIS TOOL IN UKRAINE
The Ukraine Ministry of Health, Public Health Centers and development partners have been promoting and using the cascade approach to create 'snap-shots' for various noncommunicable disease (NCD) programs. While these cascades provided an excellent summary of care stages attained by clients, they did not provide an analysis of the service implementation leading to these attainments. Poltava Region in central Ukraine had a particular interest in going further in the analysis of their type-2 diabetes care cascade which showed two major breakpoints: diagnosis and glucose control. 2 At that time, the national health reform brought changes to the diabetes program, most importantly a new cost-sharing for oral diabetes medication. Poltava Region was also a beneficiary of the World Bank Health Sector Support Project and was getting assistance from the World Health Organisation (WHO) and Swiss Agency for Development Cooperation, all aiming to reform NCD care towards better outcomes. The descriptive cascade analysis had provided the analysis team with a good understanding of the diabetes program in Poltava Region and the various prevention and care/treatment interventions. This helped the team in defining the framework in the CAT model. The framework split the diabetic population into two types reflecting morbidity levels: Diabetics who were 'uncomplicated' cases without major vascular damage, and diabetics who already had vascular damage due to the harmful effects of prolonged high blood glucose (left and right pathways in figure 3). In the Poltava CAT model, diabetic individuals move between compartments of care states due to receiving specific services (vertical arrows). Also, uncomplicated cases can progress to vascular damage in the absence of effective treatment (horizontal arrows). Uncomplicated diabetes is treated with oral medicine or non-pharmacological, life-style interventions while cases with significant vascular damage are generally prescribed insulin. outreach services may be required to find people in need of care (intervention 2, figure 3). Qualitative research pointed to the fact that diabetes treatment costs may prevent people from both diagnosis and treatment maintenance, hence the need for co-payment schemes (interventions 9+10). To populate the model, we collected data on the unit costs of diabetes interventions and estimated program expenditure. Our rapid intervention costing included consumables, salaries, overheads and value-added tax. We also made estimates on the coverage of each intervention. We then applied the CAT to estimate the optimal combination of facility-based and outreach screening and investigate how additional funding could best be allocated to improve glucose control.

KEY RESULTS
Of the approximately 65,000 individuals with type-2 diabetes, 68% had been diagnosed and registered, 62% had been linked to diabetes care, 58% were monitored, 42% put on medication, and 16% had evidence of glucose control. We estimated that monitoring costs were higher for those not achieving glucose control (table 1).

Cascade
Unit cost (USD) We found that outreach screening campaigns could play a significant role in improving outcomes: depending on how well-targeted and scalable such campaigns are, we estimated that 10-46% of all screening could be conducted via outreach, at a cost per diagnosed case of USD 7.12-9.63.
Investments in initiatives to improve treatment adherence (medication co-payment, enhanced adherence counseling) are likely to reduce barriers along the care continuum and can lead to savings in care costs. For instance, if the share of patients achieving sustained glucose control was increased by just 1 percentage point, the Poltava diabetes program could see a reduction in annual patient monitoring costs of around USD 10,000 (0.5% of patient monitoring spend). These additional funds would be sufficient to increase coverage of enhanced adherence counselling by six-fold, enabling almost half of patients to have access to these services.

CONCLUSIONS
The Poltava Department of Health appreciated this innovative analytical work on their diabetes cascade and what it means for their decision-making on improving case finding, assessment of 'what works' in treatment maintenance, and budgeting for an effective diabetes response in their region. Key staff were trained in CAT applications and associated activities like generation of unit costs and patient file review for evaluating care outcomes. The Poltava diabetes model is parametrized and can be updated and used for scenario and optimization analyzes.
The modelling study demonstrated that investments to improve case detection and treatment adherence are the most efficient interventions for improved diabetes control in Poltava region. Quantitative tools which capture service delivery and outcomes like the CAT provide decision support and evaluation for targeting investment into services which close the gaps in implementation. Investments to improve case detection and treatment adherence are the most efficient interventions for improved diabetes control in Poltava region.