Sequential drug decision problems in long-term medical conditions: a case Study of Primary Hypertension Eunju Kim ba, ma, msc

8.2.1Methodological implication of Hypertension SDDP modelling

The hypertension SDDP model is a novel cost-effectiveness model, which involves the real drug switching rules for treatment success/failure, maintenance and contraindication. Whereas most previous CEAs in primary hypertension assumed that a drug of interest was continuously used over the follow-up period, this model attempts to represent clinical practice that includes switching the drug treatment option depending on the patient’s SBP level and other relevant health states over time. This was achieved by the surrogate outcome modelling based on the relationship between SBP lowering effect and CVD prevention effect. Rather than relying on the conventional RR approach to estimate the CVD prevention effect, the hypertension SDDP model used QRISK2, which is a validated CVD risk engine in the UK, to predict the CVD risk based on the SBP level. QRISK2 also includes other key risk factors of CVD such as age, gender, ethnicity, height, weight, Townsend score, TC, HDL, CHD, family history, smoking, treated hypertension, type 2-DM, RA, AF and RD. Utilising the functionality provided by the modelling software Matlab, Monte Carlo sampling was used in the underlying evaluation model to consider the population variation in the baseline SBP and SBP lowering effect. Changes in patients’ SBP levels and relevant risk factors over time were stored and used to calculate the CVD risk in the next period. Parallel computing with 12 cores in a fast and high performance computer facilitated the enumeration of 4,128 sequential treatment policies (including Monte Carlo sampling of SBP and SBP lowering effects within the underlying evaluation model, and 100 replications to obtain mean estimate for each option).

Whereas the studies by Richter et al^[257] and Martikainen et al^[380] tried to identify the most cost-effective sequential treatment policy in a short-term model, the SDDP hypertension model tried to consider the long-term impact of sequential treatment strategies. In the initial stage of model development, the hypertension SDDP model was divided into two sub-models – the short-term drug switching model and the long-term CVD model, which had different structures and employed different mechanisms to estimate the cost and effectiveness of sequential treatment policy. The drug switching rules were only modelled in the short-term drug switching period, whereas the long-term CVD model followed a standard Markov-based economic evaluation based on the result. Clinically and economically relevant events were selected based on a widely accepted underlying disease process of primary hypertension and the causal linkages between major CVD and antihypertensive treatment. The potential impact of DM and other AEs on the long-term costs and effectiveness was also considered.

The short-term drug switching model resembles a form of successive decision tree that allows drug switching depending on SBP level, CV events and AEs in each period. Each state in the decision tree has a ‘memory’, which includes the information about SBP, CVD and AE after treatment, and then uses them to determine the transition probabilities to the next state. The main concern of the short-term model was how to store and utilise historical information efficiently, which inevitably involved the extension of decision tree branches. To reduce the number of health states in the drug switching period, the short-term drug switching model used the aggregated health states (e.g., the uncontrolled state including the patients with high blood pressure, CVD or AEs) and then applied different CVD risks based on the medical history for the patients who were controlled and uncontrolled. Once the patients in the aggregated uncontrolled state have a CV event or DM, different transition probabilities were applied to them in the long-term CVD model. By doing so, any impact, which could be missed by aggregating health states, was minimised.

From the health economic modeling perspective, the hypertension SDDP model made a contribution by introducing a new type of dynamic and stochastic optimisation approach and applying the approximate optimisation techniques to solve a real decision problem. Several studies have developed a comprehensive policy model that can simultaneously evaluate the cost and benefit associated with the prevention and therapeutic interventions across the entire disease pathway. The hypertension SDDP model moves a step forward from the comprehensive modelling approach by incorporating an additional outer loop, which identifies the optimal or near optimal solutions through optimisation searching methods, which has not been studied before.