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

8.2.3Limitations

Although the hypertension SDDP model is based on the best available date, systematic reviews, which were recognised as the gold standard in evidence-based decision making, were not fully conducted to inform the values of parameters used for the underlying evaluation model. Therefore the results can be updated if the newer or systematically pooled date is available.

Internal validity was checked during the process of building the model. However, the comparison of the final results against those of previous models was a challenge because there was no directly comparable existing study. For this reason, the enumeration result was reviewed by two clinical experts (their names and affiliation are provided in section 7.2), who confirmed that the results are plausible and can be explained at an intuitive level. Furthermore, comparing with the current recommendations from major clinical guidelines, the results of the hypertension SDDP model appear reasonable.

The SDDP hypertension model tried to capture the essential part of the dynamic relationship between sequential treatment choices and the underlying disease process in primary hypertension. However, model simplifications in the drug switching period and the transitions between health states were inevitable because of the computational complexity and data availability. Clinically plausible, but rather arbitrary, assumptions were used to combine the short-term drug switching model and the long-term CVD model. For example, after the pre-defined drug switching period, the controlled patients were assumed to continue the currently used treatment, and uncontrolled patients were assumed to switch to a randomly selected drug until a CV event or DM happened. The patients who have used all four lines of treatments but were never controlled during the drug switching period were assumed to have resistant hypertension and were assumed to see a specialist to correct the underlying cause. This study did not specify the impact of resistant hypertension because treatments can be varied depending on the exact causes and may involve drug therapy, surgery or renal dialysis. For the simplification of search space, the hypertension SDDP model did not consider drug switching within the same class: this decision was based on the evidence of equivalent efficacy within the same class^{[381, 382]}.

For Ds, BBs and CCBs, which may have different efficacy depending on dosage, the patients were assumed to take the equivalent dosages of the drugs included in the same class^[383-385]. Dose-titration was not considered as a separate treatment option, but it was assumed that the patients take the equivalent dose of each drug. If dose titration of a single drug is considered as a separate treatment option and only low dose drugs are combined, there will be eight single treatment options, six two-drug combinations and four-three drug combinations. The size of the search space will be increased from 4,128 to 20,256:

Z(SS) = 8*(14-1)*(18-2)*(18-3)-4704 = 20256 Equation 8.1.
For the same size of health state space, the decision space will be expanded to 5,023,488:

Z(DS) = 31*20256*8 = 5023488 Equation 8.2

Assuming the same computational time per policy with the hypertension SDDP model, it will take 59.87 hours to enumerate all the possible treatment sequences:

10.64 seconds * 20256 = 215523.84 seconds (≈59.87 hours) Equation 8.3

If a fast computer and parallel computation is not possible, the computational time will take more than 12 times as long as the hypertension SDDP model would take for enumeration.

As the structure of the long-term CVD model is based on the NICE hypertension model, the limitations of the NICE hypertension model also apply to this model. For example, the NICE hypertension model excluded some CVDs (such as stable angina, peripheral vascular disease and transient ischemic attacks) and also excluded the health states for patients having more than two health states (e.g., the patient having several CVDs or DM together) because of inconsistently reported data in the trials^[63].

Some of the limitations were driven by the lack of data availability. Although an absence of data does not justify the simplification in itself, it commonly happens in CEAs. For example, there was no meta-analysis or systematic review in the treatment effectiveness of combination treatment with antihypertensive drugs. Data from RCTs were not directly comparable with each other because of the heterogeneity in participants, follow-up period and treatment regimen used in different trials. Due to this, the hypertension SDDP model assumed that the SBP lowering effect is additive and the RR of CVD is multiplicative when two or three single drugs were combined. Therefore, clinically positive or negative interactions among different drugs, and safety and tolerability issues may be missed where two or three drugs are combined^[369].

There may be an impact of drug switching on health outcome and resource use that is not accounted for in the hypertension SDDP model; for example, baseline risk may be higher in patients who are not controlled with the initial drug^[386]. Drug switching could disturb the therapeutic consistency, which could delay blood pressure control and increase the risks of mortality and morbidity associated with hypertension^{[3, 347, 351, 386, 387]}. Drug switching may also cause patients concern or dissatisfaction with the process and might reduce QoL^[387]. Some studies highlighted the additional costs of health care incurred by drug switches. The additional costs could incur either in the process of switching such as costs for additional clinic visits and extra laboratory tests; or as a consequence of switching such as hospitalisation due to adverse-events^{[6, 387, 388]}. From the patients’ perspective, change in treatment regimen was one of the main reasons for decreasing medication compliance^[346]. Low compliance increases the risk of treatment failure, and leads to further change in treatment regimen. The vicious circle between increased drug switching, low compliance and higher risk of treatment failure can be perceived intuitively, but was difficult to incorporate practically in the hypertension SDDP model. If an IBM is used as an alternative to the cohort-based structure of the hypertension SDDP model, the correlations between drug switching, low compliance and higher risk of treatment failure at an individual level could be incorporated easily within the mathematical structure of such a model.