Sequential drug decision problems in long-term medical conditions: a case Study of Primary Hypertension Eunju Kim ba, ma, msc
Yüklə 10,52 Mb.
|
7.2Model validityThe model structure corresponded to a widely accepted underlying disease process of primary hypertension[63]. Clinically and economically relevant events were selected based on the causal linkages between major CV events and antihypertensive treatment. To ensure that the model structure was well designed on a clinically intuitive level, key modelling assumptions relating to clinical and practical issues (such as the cycle of regular check-up, key considerations to change the treatment regimen, the selection of the next drug and the maximum number of drugs sequenced and maintenance therapy) were advised by two clinical experts: Dr. Paul Morris who is a cardiologist in Sheffield Teaching Hospitals NHS Foundation Trust and Eunhee Lee who is a pharmacist working for the National Health Insurance Review Agency in South Korea. Checks for technical errors were undertaken throughout the entire model development process. In the early stage of model development, two separate models were built in Excel and Matlab and double checked whether the results were agreed each other. The final Matlab model was designed to print key time-dependent parameters on the screen and to alert any errors within the model (e.g., whether the minimum and maximum values of key parameters are acceptable and whether the sum of rows of transition probabilities equals to 1) by popping up an error message box. Validation was also implemented for the randomly generated distributions of key parameters by comparing the mean of the generated distributions against the point estimates. Certain programming errors were automatically checked by ‘Code Analyzer’, which is built in Matlab. The validation process included examining the relationship between key parameters and comparing the outputs with those of previously published models and studies. A validation scenario, which assumed a treatment regimen is continuously used over the follow-up period like the conventional CEAs, was used to test the internal consistency and external validity of the hypertension SDDP model. Table 7. represents the predicted outcomes of 14 treatment options (including no treatment, four single drugs, six two-drug combinations and four three-drug combinations), where they were assumed to be used continuously over the follow-up period. ‘No treatment’ was included as a comparator to test whether the base model makes preconceived expectations of future events. Under the validation scenario the treatment success rate, which was defined as the cumulative percentage of the patients who achieve the treatment goal every three months, was gradually increased over time for all treatment options apart from no treatment (see Table 7.). The treatment regimens using a three-drug combination reached the peak treatment success rate the fastest (in 6 months), followed by the treatment regimens using a two-drug combination and a single drug. For the single drugs, the treatment success rate was 96.1% for ACEIs/ARBs, 70.2% for Ds, 44.0% for BBs and 36.3% for CCBs at one year. Considering that the treatment success rates of single antihypertensive drugs were around 30~60% at one year in the RCTs[317, 368, 369], the estimated treatment success rates of BBs and CCBs were within the bounds, whereas the treatment success rate of Ds was slightly over the upper bound and the treatment success rate of ACEIs/ARBs was considerably over the upper bound. In the hypertension SDDP model, however, this high treatment success rate in ACEIs/ARBs at one year was only applied for a small proportion of patients who consistently achieved the treatment goal for one year with the initial drug being ACEIs/ARBs. The increase in the treatment success rate was mostly due to the time-dependent SBP lowering effects, which were assumed to decrease over time (see Table 5.). This pattern was found in the cumulative treatment success rate and the mean levels of SBP in Table 7.. The mean levels of SBP after using a single drug for one year were 144.98 mmHg for ACEIs/ARBs, 153.23 mmHg for Ds, 155.15 mmHg for BBs and 156.14 mmHg for CCBs. These levels were slightly higher than the SBPs in patients using a single antihypertensive drug after one year in major clinical trials, which were between 135 and 157 mmHg[316, 317, 370, 371]. One possible reason for this difference is that the patient’s initial SBP used in the hypertension SDDP model, which came from the Burke’s retrospective cohort study[283], was higher than the initial SBP of the participants in the clinical trials. Most patients in the Burke’s study had moderate or severe hypertension (32.0% and 32.7%, respectively), whereas patients with mild hypertension or controlled hypertension were only 12.4%. The 10-year risk of CVD was calculated by QRISK2 and then adjusted to a three month basis. The slight increase in the CVD risk between three months and six months was due to the treatment history included as a risk factor in QRISK2. It was assumed that the patients had a treatment history after three months, whereas they had no treatment history at the beginning of simulation. The minor fluctuation in the CVD risk between six months and 12 months was because of the population variation in their baseline SBPs and SBP lowering effect. The annual CVD incidence rate per 100,000 persons was estimated by the model at 1112.49 where no treatment was used. According to the CHD statistics published in 2012[288] and the Stroke Statistics published in 2009[372], the CVD incidence (including acute MI, angina, stroke and HF) per 100,000 was approximately 437.20 for a male population aged between 55 and 64. Considering that the CVD risk more than doubles at SBP≥160 mmHg compared with the optimal BP (SBP<120 mmHg)[373, 374], the estimated annual CVD incidence rate in the hypertension SDDP model seems to be plausible. In the long-term, the percentage of patients without either CVD or DM history steadily decreased and reached 0 at the end of the simulation (see Figure 7.). The slope of the prevalence was steepest for no treatment, followed by single drugs, two-drug combinations and three-drug combinations. The opposite occurred for the percentage of death where patients receiving no treatment died the fastest, and in three-drug combinations the slowest (see Figure 7.). The percentage of patients in a CVD state, which reached the peak at around 75 years old and then gradually declined, was also highest in no treatment across the whole follow-up period, followed by single drugs, two-drug combinations and three-drug combinations (see Figure 7.). The jump between 61 and 62 years old was due to the different mechanism to calculate the transition probability to a CVD between the short-term drug switching model and the long-term CVD model. Whereas the surrogate outcome modelling based on QRISK2 was used to calculate the transition probabilities to a CVD between 60 and 61 years old (i.e., during the drug switching period), the transition probabilities to a CVD afterward came from the NICE hypertension model and propagated through the long-term CVD model. The occurrence of DM was highest in no treatment until 70 years old, followed by single drugs, two-drug combinations and three-drug combinations (see Figure 7.). Although some of antihypertensive drugs were assumed to increase the risk of DM in the hypertension SDDP model, the same pattern was found in Figure 7. and Figure 7. because of another assumption, which doubles the baseline CVD risk in patients with DM risk over patients without DM. The percentage of patients in the DM state gradually declined after 70 years old. Table 7.. Treatment effectiveness predicted from the validation scenarios where the same treatment is applied in the follow-up period
1) Annualised rate when modelled over a lifetime. Figure 7.. The percentage of patients without either CVD or DM history by age Figure 7.. The percentage of death by age Figure 7.. The percentage of patients in a CVD state by age Figure 7.. The percentage of patients in the DM state by age The total costs and effectiveness derived from the validation scenario were compared against the NICE hypertension model[63] (see - Table 7.). For this, the age of the initial population in the hypertension SDDP model was set to 65 years old, which is same as the NICE hypertension model. The model repeated 1,000 times for both men and women. Uncertainty in the costs and effectiveness was reported by SD. PSA was not undertaken because the NICE hypertension SDDP model did not conduct this in the latest version. Please note the objective and structure of the NICE hypertension model was very different from the hypertension SDDP model, so the comparison should be interpreted accordingly. Where a certain drug was assumed to be continuously used over time for the newly diagnosed hypertensive patients aged 65 years, the costs estimated in the hypertension SDDP model were lower than the cost estimated from the NICE hypertension model. While the total costs estimated in the NICE hypertension model ranged from £3,910 to £4,690 in men, the total costs estimated in the hypertension SDDP model were between £2,611 and £3,857 in men (see ). For women, the total costs estimated in the NICE hypertension model ranged from £4,310 to £5,230, while the total costs estimated in the hypertension SDDP model were between £2,985 and £3,857 (see ). The average difference in the total costs between the NICE hypertension model and the hypertension SDDP model was £1,069 in men and £1,320 in women. The QALYs estimated in the hypertension SDDP model were also lower than those from the NICE hypertension model. While the total QALYs estimated in the NICE hypertension model ranged from 9.57 to 10.28 in men, the total QALYs estimated in the hypertension SDDP model were between 7.90 and 8.57 in men. For women, the total QALYs estimated in the NICE hypertension model ranged from 9.96 to 10.71, while the total QALYs estimated in the hypertension SDDP model were between 8.92 and 9.77. The average difference in the total QALYs between the NICE hypertension model and the hypertension SDDP model was 1.8 in men and 0.97 in women. The difference in the total costs and QALYs between the hypertension SDDP model and the NICE hypertension model is possibly because the type 2 DM risk at 65 years old in the hypertension SDDP model (1.38% for men and 1.14% for women in average) was higher than DM risk assumed in the NICE hypertension model (1.1%). Another potential reason is the difference in the annual CVD risk between the hypertension SDDP model and the NICE hypertension model. The annual CVD risk estimated in the hypertension SDDP model was 2.08% for 65 year old men and 1.98% for 65 years old women, whereas the NICE hypertension model assumed 2% for both men and women in the same age. In the NICE hypertension model, CCBs were the most cost-effective initial treatment option, whereas ACEIs/ARBs were the most cost-effective initial treatment option in the hypertension SDDP model. This result arose from the SBP lowering effect used in the hypertension SDDP model. The hypertension SDDP model includes the surrogate outcome modelling based on the time-dependent SBP lowering effect over one year. Although CCBs had a higher SBP lowering effect than ACEIs/ARBs in three months, the cumulative SBP lowering effect was higher in ACEIs/ARBs after three months. This also has an effect on the long-term cost and effectiveness because the transitions between health states in the long-term CVD model depends on the final treatment result in the drug switching period. BBs and no treatment were dominated and ruled out in both studies.
Table 7.. Comparison of the CE results from the hypertension SDDP and the NICE hypertension models in men
1) Willingness-to-pay for a unit of QALY was assumed £30,000. 2) The total costs and effectiveness represents the total costs and QALYs per person over a lifetime. A discount rate of 3.5% was applied to both costs and effectiveness. 3) The treatment options, which were less effective but cost more than the lowest cost option, were excluded from the calculation of the ICERs. 4) SD is presented in parenthesis. Table 7.. Breakdown of the costs and QALYs of the hypertension SDDP model in men
1) SD is presented in parenthesis. 2) BP stands for blood pressure. Table 7.. Comparison of the CE results from the hypertension SDDP and the NICE hypertension models in women
1) Willingness-to-pay for a unit of QALY was assumed £30,000. 2) The total costs and effectiveness represents the total costs and QALYs per person over a lifetime. A discount rate of 3.5% was applied to both costs and effectiveness. 3) The treatment options, which were less effective but cost more than the lowest cost option, were excluded from the calculation of the ICERs. 4) SD is presented in parenthesis. Table 7.. Breakdown of the costs and QALYs of the hypertension SDDP model in women
1) SD is presented in parenthesis. 2) BP stands for blood pressure. Yüklə 10,52 Mb. Dostları ilə paylaş:
| |||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||