This was an exploratory model-based economic evaluation using local observational data. The analysis compared VEN+AZA with AZA monotherapy in older patients with newly diagnosed AML who were considered unfit for intensive chemotherapy. The study was designed as an economic model informed by local outcomes, resource use, and quality-of-life inputs.

Data on VEN+AZA were prospectively collected during the period 2022–2024 at the “Specialized Hospital for Active Treatment of Hematological Diseases.” Data on AZA were retrospectively extracted for the period 2018–2022, prior to venetoclax reimbursement. This was performed to eliminate time-related bias in patient study selection and to include patients treated within a modern context, as well as to collect a representative cohort of patients receiving the comparator (AZA).

The study was approved by the hospital’s ethics body, with decision no. PK303/2022. Informed consent was obtained to use patient health data recorded during clinical visits. Data were pseudonymized to protect patient confidentiality and adhere to GDPR standards.

The modeled population reflected local Bulgarian clinical practice and was based on 69 patients in total:

All patients had newly diagnosed AML and were ineligible for intensive chemotherapy because of older age, comorbidity burden, and poor performance status. A patient was considered “unfit” or “ineligible” for intensive chemotherapy if age was ≥75 years, documented pulmonary disease with FEV1 ≤65%, carbon monoxide diffusion rate ≤65%, dyspnea requiring supplemental oxygen therapy, documented heart failure or cardiomyopathy with ejection fraction ≤50%, liver cirrhosis or other documented liver disease, active resistant infection, ECOG functional status ≥3, current psychiatric disorder, kidney failure, or another condition requiring dialysis.

The interventions compared were:

Dosing strategies followed established guidelines and SMPC recommendations, and chemotherapy cycles were applied until disease progression or death.

The analysis was conducted from the perspective of the National Health Insurance Fund (NHIF) and included only direct medical costs in 3 domains-costs for drugs and applications and hospitalization-related costs, including initial diagnostic and laboratory costs.

A 1-year time horizon was used, with monthly cycles. The 1-year horizon was selected because the model was intended as an exploratory local-data analysis and because survival and event data were most reliable within the first year of follow-up. OS was recorded by physicians in months, leading to the preference for monthly state-occupancy cycles instead of weekly transition probabilities. This also reduced reliance on long-term extrapolation.

A 3-state state-occupancy model was developed with the following mutually exclusive health states (Fig. 1):

Although the model originated from a simple Markov-style structure, the final implementation functioned as a state-occupancy/partitioned-survival-style model, where monthly state proportions were informed directly by observed and reconstructed survival quantities rather than by a fully specified transition matrix. This approach was chosen because progression data were limited and because direct estimation of transition probabilities between all states would have required stronger assumptions than were justified by the available data. OS data were available directly, EFS was reconstructed from patient-level data, and PD/RL and death were derived.

Clinical inputs consisted of observed and derived states—overall survival, event-free survival, progressed disease/relapse, and death. Full table properties and confidence levels are available in Suppl. material 1.

Overall survival

Overall survival (OS) for each treatment arm was estimated from local data and expressed as monthly proportions of alive patients during the first year. OS data are pending publication at the time of writing.

Event-free survival

Event-free survival (EFS) was reconstructed from patient-level information using time-to-event logic based on progression and death. In patients with incomplete progression ascertainment, EFS was derived conservatively using the earliest of documented progression, death, or censoring.

Progressed/relapsed disease occupancy

Because direct progression/relapse information was incomplete, monthly PD/RL occupancy was derived indirectly as:

P D R L t = max O S t − E F S t , 0

To preserve coherence, EFS was constrained not to exceed OS when probabilistic draws generated implausible combinations.

Death

Monthly death-state occupancy was derived as:

 Deat  h t = 1 − O S t

Thus, for each cycle:

E F S t + P D / R L t +  Death  h t = 1

This structure was chosen because death and OS were well-captured variables; EFS could be directly estimated because of “date of last contact” information, whereas progression information was more limited.

Patient-level monthly responses to the EORTC QLQ-C10 questionnaire were extracted. Corresponding quality-of-life estimates were derived through the mapping algorithm and regression coefficients published by Hagiwara et al., leading to a monthly utility value. One-year quality-adjusted life-years (QALYs) were derived by summing the utility value of each month for progressed or event-free states and dividing by 12:

Q A L Y t = ∑  Monthly  E F S t × U E F S + PD / RL t × U PD / RL

Life-years were estimated as the sum of monthly alive-state occupancy (EFS + PD/RL) divided by 12 over a 1-year time horizon.

L Y t = E F S t + P D / R L t 12

The model included direct medical costs expressed in Bulgarian lev (BGN) at the time of study completion. Costs were then standardized to euros at a fixed exchange rate of 1.95583 lev for every euro. Included costs were:

Costs were modeled at the annual and monthly level, depending on the structure of the input data. At this exploratory stage, all costs were retained in BGN for internal consistency with local pricing and reimbursement sources. Currency conversion to euros was reserved for later reporting.

Venetoclax annual treatment costs were estimated based on labeled dosing and local acquisition prices. Costs for the first cycle and subsequent 28-day cycles were derived (Table 1). To preserve the model’s monthly logic, a weighted monthly cost of €9,760.31 was calculated.

Azacitidine treatment costs were estimated using a mean body surface area of 1.82 m², labeled dosing, vial-based costing, and local acquisition prices (Table 2). Due to multiple trade names, a weighted average cost per vial was assumed. This was performed to simulate the competitive tendering processes involved with drug acquisition in a hospital setting.

To preserve the model logic, a weighted monthly cost for azacitidine was again calculated at €2,802.03.

Hospitalization data were available for all patients and were used to estimate monthly non-drug medical costs from the perspective of the Bulgarian National Health Insurance Fund (NHIF) (Table 3). The principal reimbursement mechanisms considered are summarized in Table 3. Under the model assumptions, palliative care was only given to patients before death and was therefore excluded for EFS patients. The number of hospitalizations per month during the first year was used as a proxy to estimate monthly resource use.

Average monthly hospitalization-related costs were calculated according to the number of hospitalizations observed in each health state and the corresponding allocation of healthcare resources under these reimbursement routes.

Uncertainty was handled through both deterministic and probabilistic analyses. Monte Carlo simulation of average yearly costs and state occupancy was conducted for both QALYs and LYG.

Deterministic analysis

Base-case results were calculated using mean parameter values and reconstructed first-year state occupancy. A tornado diagram was constructed by varying two parameters of uncertainty-costs and outcomes-in the following ranges: ±5%, ±10%, ±15%, and ±30%.

Probabilistic sensitivity analysis

Probabilistic sensitivity analysis (PSA) was conducted using Monte Carlo simulation. Probability distributions were assigned to selected parameters, including:

Earlier constraints on probability draws were applied to ensure model consistency. Drug acquisition costs were assumed to be static, without a probabilistic draw. The data were based on list prices during 2024.

Internal validity was checked by confirming that:

The model estimated:

Due to the limited time horizon in the model (1 year), no discounting was applied. Discounting would be applicable in models with a longer time horizon, for which sufficient data were not available.

Table 4 displays the calculated 1-year state occupancy probabilities for both venetoclax and azacitidine with the available 95% confidence intervals. First-year survival was non-significantly higher for VEN+AZA; however, VEN+AZA survival was significantly associated with higher EFS probability, indicating improved disease control. Progressed/relapsed occupancy was reconstructed as the difference between OS and EFS, while death was reconstructed as 1 − OS; therefore, confidence intervals are not reported.

The estimated mean monthly cost of additional medical services per patient treated with azacitidine was €4,501.01 (95% CI: €2,999.73 to €6,002.28). For patients treated with venetoclax plus azacitidine, the corresponding mean monthly cost of additional medical services was €4,251.95 (95% CI: €2,242.52–€6,261.38).

All yearly input costs are reported in Table 5 based on 2024 list prices. VEN+AZA costs were €68,887.59 and €25,704.91 for drug acquisition and hospitalization, respectively. Similarly, costs for AZA were €15,130.95 for drug acquisition and €24,272.78 for hospitalization. Both VEN+AZA and AZA patients had additional investigation and cytogenetics costs of €4,940.36. The total cumulative yearly costs for both drugs were €99,532.86 for VEN+AZA and €44,344.09 for AZA, which were the final model output costs.

Monthly utility values were estimated based on EORTC questionnaire responses from patients during follow-up. These values are subject to further validation and subsequent publication but are reported here to adhere to good modeling practices.

In the base case, improved VEN+AZA survival was associated with a higher total cost per patient than AZA monotherapy. However, VEN+AZA also provided greater health benefits, including higher QALYs and life-years. The incremental costs associated with VEN+AZA were €55,188.77, with corresponding gains of 0.19 QALY and 0.23 LYG. The estimated cost-effectiveness ratio exceeded the 3-times-GDP threshold of €70,815 (€289,309.59 and €235,736.28, respectively) (Table 7).

Cost per QALY

The probabilistic sensitivity analysis produced a distribution of incremental costs and health outcomes consistent with the structure of the model. Due to the limited time horizon, QALY gain was also limited; however, at the accepted national threshold of €54,815, only 4.4% of the dispersion cloud lay below the threshold (Fig. 2). When adjusting the WTP threshold, the minimal acceptable cost-effectiveness ratio was €300,000, equivalent to 12.7 times the GDP per capita. Similarly, results plateaued at €400,000, after which cost-effectiveness percentage gain was minimal (Fig. 3). The model correctly captured patients with a high induction ADR burden; thus, in some instances, the QALY gain was negative. For these patients, the hospitalization burden was higher than average, thus leading to increased costs.

Cost per LYG

The heavy monetary burden was also evident in the secondary probabilistic sensitivity analysis. Only 6.7% of points were under the accepted threshold of €54,815; however, the dispersion cloud showed a tighter concentration around the base case, indicating that the survival benefit was more consistent in the VEN+AZA group. The model accurately captured induction fatality events related to VEN+AZA; thus, certain values for LYG were negative.

One-way sensitivity analyses for both outcomes showed that the ICER was most sensitive to negative decreases in VEN+AZA effectiveness. The sharpest increase was observed under the hypothesis that VEN+AZA had 30% lower QALYs and 30% lower LYG, with respective ICERs or ICURs of €1.5 million and more than €1.9 million.