4 Results
The results of the model obtained by simulating the four studies subsequently are presented in this section.
4.1 UPLIFT
UPLIFT is the longest running and largest RCT to date in a COPD population. It compared 4 years of therapy with either tiotropium or placebo in moderate to very severe patients with COPD. The primary endpoints were the rate of decline in the mean FEV1 before and after bronchodilation beginning on day 30. Secondary endpoints included measures of forced vital capacity (FVC), changes in response on SGRQ, exacerbations of COPD and mortality
The co-variance matrix in the model predicted accurately the smoking status of the patients, although CIs were not available for comparison in the original publication (Table ). Estimates of relative risk (RR) of exacerbations from the model fell within the CI of the study. The model overestimated the annual FEV1 decline rate, which could be attributed to the fact that around 10 % of the UPLIFT trial population were very severe patients. This group typically has a lower rate of decline than patients in other GOLD stages. The estimated mortality RR fell within the CI of the published results , but the absolute rates of both arms are slightly overestimated. Treatment benefit in mortality is not as well captured when using mortality rates associated with GOLD states, as this only allows for a difference in mortality rate when the lung function improves sufficiently for a GOLD state change. Alternative possibilities include mapping the change in mortality risk directly to other clinical indicators, such as lung function decline or change in TDI, or related to exacerbations, which could be made possible with sufficiently
long-term data analysed at the patient level.
4.2 Lung Health Study
The Lung Health Study was a clinical trial that assessed the benefits of smoking cessation over a time horizon of 5 years .
The model was built originally with active treatment arms and a placebo arm only. When this validation was conducted using the Lung Health Study, it was observed that the modelled annual mean rate of FEV1 decline was overestimated. This was attributed to an initial FEV1 boost in the placebo arm based on clinical trial data [, which cause patients discontinuing from placebo to have a significant drop in their FEV1 in line with the assumption described in Sect. of loss of treatment benefit upon treatment termination. In order to address this paradigm, a no treatment arm was added where it was assumed no initial benefit in FEV1 would be observed and patients would only be subject to a constant rate of FEV1 decline.
The model was re-run and it was observed that the mean rates of decline obtained after running the model were aligned with mean rates of decline published by Scanlon et al. , which highlights that the inclusion of a no treatment arm is appropriate and it is correctly used in the model (61.54 ml/year in smokers and 31.22 ml/year in ex-smokers obtained from the model vs. 62 ml/year in smokers and 31 ml/year in ex-smokers from the study publication ). These rates are used in the base case of the model and are not affected when no treatment is selected.
4.3 TORCH
TORCH (TOwards a Revolution in COPD Health) was a randomized, double-blind, placebo-controlled trial where salmeterol and fluticasone in combination was compared with placebo in moderate to very severe COPD patients . The length of follow-up was 3 years and the primary endpoint was death from any cause. Secondary endpoints included frequency of exacerbations, health status using the SGRQ and spirometric values .
The correlation matrix was broken to allow for a higher proportion of smokers as this had an impact on the rate of FEV1 decline. Table shows the results obtained in the model when implementing the TORCH trial data .
The RRs of experiencing an exacerbation are comparable between the model and the trial publication (i.e. 0.72 vs. 0.75, respectively). However, absolute rates observed in the model were slightly lower than those reported in the publication . This could be due to the baseline rate of exacerbations in the source indacaterol trials being extremely low, and could be adjusted with other data sources.
The average mean change of FEV1 over the model time horizon (3 years) was smaller than that reported in the study . This difference can be explained by the trial
patients receiving placebo experiencing a negligible FEV1 improvement. However, this value was not adjusted for placebo patients in the model, who were given a 130 ml improvement in their first cycle. This led to an increase in FEV1 change between the comparators in TORCH compared to that observed in the model Although possible, the use of the no treatment arm in the model was not considered appropriate given that the TORCH trial uses placebo as a comparator. The mortality rates obtained in the model were higher
than those in the published data . Distribution of GOLD states in the TORCH study was not available, but
the mean FEV1% predicted in the population was 44 %, higher than the mean in UPLIFT of 39 %, indicating a
slightly less severe population than UPLIFT .
The current data source for mortality used may overestimate mortality rates associated with the milder health states, but the use of the alternative data source significantly underestimated overall mortality. Further research that correlates different clinical indices and patient characteristics, such as age and smoking status, should improve the accuracy of mortality rates specific to COPD patients.
SGRQ results are the most disparate between the model and the study. Change in HR-QOL over time is not well documented in literature. Change in SGRQ is usually reported over short periods of time, which makes long-term extrapolation of change in SGRQ difficult. Studies utilizing registry data or similar observational studies over the long term may be able to significantly improve the accuracy of SGRQ outputs.
4.4 OLIN
OLIN (Obstructive Lung Disease in Northern Sweden) was a longitudinal study that followed COPD patients (64 % male) over 20 years . The primary endpoint was mortality and the secondary endpoint was the annual decline in FEV1 among survivors .
The correlation matrix was not used to generate patients for this simulation in order to use the published FEV1% predicted of the patient population , which was significantly higher than the original indacaterol trialsnn.
The modelled proportion of patients who have died appears to be overestimated when compared with the average mortality rate reported in the study (70 % [95 % CI 69.61–70.69] as a model result vs. 54 % as reported in the publication The OLIN study cohort has considerably less severe disease than those in the indacaterol studies , UPLIFT or TORCH and bears out the hypothesis that the current data source overestimates the mortality rates associated with the less severe disease states. This is especially likely for the mild diseasennstate , which carries an excess mortality rate of 2.25 % annually but is often considered a pre-COPD state. The
model was run over a time horizon of 10 years in order to compare with the published data for the annual rate of FEV1 decline. The simulated mean decline in FEV1 is found to be marginally higher (27 ml/year [95 % CI
26.1–28] vs. 24 ml/year), which could be due to the higher proportion of smokers generated in the co-variance matrix (64 %) compared with the OLIN study (46 %) .
The validation with the four studies found the model to have generally good predictive ability, although some inputs, such as mortality and SGRQ, could benefit from further research. However, the overall validation yielded a good degree of external validity. It also allowed for refinement of the model through the addition of a no treatment arm.