Projecting the 10-year costs of care and mortality burden of depression until 2032: a Markov modelling study developed from real-world data

Abstract

Background: Based on real-world data, we developed a 10-year prediction model to estimate the burden among patients with depression from the public healthcare system payer's perspective to inform early resource planning in Hong Kong.

Methods: We developed a Markov cohort model with yearly cycles specifically capturing the pathway of treatment-resistant depression (TRD) and comorbidity development along the disease course. Projected from 2023 to 2032, primary outcomes included costs of all-cause and psychiatric care, and secondary outcomes were all-cause deaths, years of life lived, and quality-adjusted life-years. Using the territory-wide electronic medical records, we identified 25,190 patients aged ≥10 years with newly diagnosed depression from 2014 to 2016 with follow-up until 2020 to observe the real-world time-to-event pattern, based on which costs and time-varying transition inputs were derived using negative binomial modelling and parametric survival analysis. We applied the model as both closed cohort, which studied a fixed cohort of incident patients in 2023, and open cohort, which introduced incident patients by year from 2014 to 2032. Utilities and annual new patients were from published sources.

Findings: With 9217 new patients in 2023, our closed cohort model projected the 10-year cumulative costs of all-cause and psychiatric care to reach US$309.0 million and US$58.3 million, respectively, with 899 deaths (case fatality rate: 9.8%) by 2032. In our open cohort model, 55,849-57,896 active prevalent cases would cost more than US$322.3 million and US$60.7 million, respectively, with more than 943 deaths annually from 2023 to 2032. Fewer than 20% of cases would live with TRD or comorbidities but contribute 31-54% of the costs. The greatest collective burden would occur in women aged above 40, but men aged above 65 and below 25 with medical history would have the highest costs per patient-year. The key cost drivers were relevant to the early disease stages.

Interpretation: A limited proportion of patients would develop TRD and comorbidities but contribute to a high proportion of costs, which necessitates appropriate attention and resource allocation. Our projection also demonstrates the application of real-world data to model long-term costs and mortality, which aid policymakers anticipate foreseeable burden and undertake budget planning to prepare for the care need in alternative scenarios.

Funding: Research Impact Fund from the University Grants Committee, Research Grants Council with matching fund from the Hong Kong Association of Pharmaceutical Industry (R7007-22).

Keywords: Burden projection; Comorbidities; Cost; Depression; Health policy; Mortality; Real-world data; Real-world evidence; Time-varying Markov model; Treatment-resistant depression.

Fig. 1

Schematic presentation of the Markov model structure. Yellow oval represents the initial state, grey ovals represent the absorbing states and blue texts represent the time-varying transition probabilities. “C” represents the complement of other probabilities from the same state. The definitions of the health states are as follows. Non-treatment-resistant depression (NTRD): Patients with depression who were yet to develop TRD or further clinical characteristics. Treatment-resistant depression (TRD): Patients who took at least two antidepressant regimens for an adequate duration and had the third regimen to confirm refractoriness in the first two regimens. New-onset comorbidities (NTRD-comorbid): Patients with new-onset somatic comorbidities included in the list of diseases used to calculate Charlson Comorbidity Index, or pre-specified psychiatric comorbidities before TRD, and the new-onset condition(s) did not occur before depression diagnosis. New-onset post-TRD comorbidities (TRD-comorbid): Similar to NTRD-comorbid but condition(s) occurred only after TRD. Low-intensity service user (absorbing state): Patients with minimal care need and free of further depression-related diagnosis records and antidepressant prescriptions. The health state acts as a proxy for recovery from depression, rather than relapse or recurrence manifested during the development of TRD. Patients in this state are not considered as active or living with depression. All-cause death (absorbing state): Observable deaths regardless of causes.

Fig. 2

Modelled trajectories of 10-year patient distribution and flow between health states. All health states are mutually exclusive. The closed cohort model shows the 10-year flow among the incident patients diagnosed in 2023, with the tenth cycle equivalent to the year 2032. The open cohort model shows the annual snapshot of patient distribution between states among the patients diagnosed in the recent 10 years counting from the corresponding calendar year. The pandemic impact was not accounted. Abbreviations: NTRD, Non-treatment-resistant depression; TRD, Treatment-resistant depression.

Fig. 3

Projected annual costs of all-cause and psychiatric care from 2023 to 2032. All costs are undiscounted and valued in 2023 U.S. Dollars (USD). The closed cohort setting shows the projected annual costs of care among the incident patients diagnosed in 2023, with the tenth cycle equivalent to the year 2032. The open cohort setting shows the projected annual costs of care among the patients diagnosed in the recent 10 years counting from the corresponding calendar year.

Fig. 4

Projected costs and mortality burden by 2032 stratified by subgroups in the closed cohort model. The cumulative burden refers to the projected total collective costs of care or number of deaths accumulated in 10 years in one subgroup. The annualized burden per patient refers to the projected annual individual costs of care or the annual mortality rate, calculated by dividing the cumulative burden (nominator) by the years of lives lived (denominator) accumulated in 10 years in one subgroup. All costs are undiscounted and valued in 2023 U.S. Dollars (USD).

Fig. 5

One-way deterministic sensitivity analyses for costs of all-cause and psychiatric care. Each bar represents the corresponding change in the costs of care when each parameter listed on the left side changes to its lower and upper bound. All costs are undiscounted and valued in 2023 U.S. Dollars (USD). Abbreviations: NTRD, Non-treatment-resistant depression; TRD, Treatment-resistant depression.