The aims of this project were to measure the association between treatments for COPD and COPD outcomes among patients not included in RCTs for COPD treatments, and to develop a methodological framework for using non-interventional EHRs. We have met these aims by completing each of the three specific primary objectives outlined in Chapter 1, Aims and objectives: (1) validate methods for measuring COPD medication effectiveness in EHR data by comparing with trial results, (2) use EHR data to measure COPD medication effectiveness in patients excluded from trials (most importantly, those aged > 80 years or those with comorbidities) and (3) determine COPD treatment effectiveness in an understudied disease stage (i.e. mild COPD).
Our results show that routinely collected EHR data can be used to successfully identify the expected beneficial and harmful effects of treatments for COPD when validating against results obtained from randomised trials. Importantly, successful replication was possible when comparing between two active treatments only and could not be achieved for comparisons between active treatment and no treatment. This was because of (1) confounding by indication (an established challenge when trying to compare with no treatment) and (2) the fact that the TORCH trial3 placebo-controlled group were such an unhealthy group in comparison with people with COPD routinely treated in UK primary care (overall much more recently than the TORCH trial3). These conclusions are specific to investigations of the effects of COPD medication and cannot be assumed to replicate in other disease areas. In validating against the results of a large international multicentre randomised trial, it was also clear that, in some instances, some patient characteristics observed in a trial are not always observed in a single-country EHR setting. This raises questions of possible trial result heterogeneity by geographic region, which should be considered in future attempts to replicate trial findings in non-interventional data.
The step of directly comparing findings from non-interventional data with those from the trial provided a methodological validation and template, allowing further work to focus on the types of patients excluded from the original trials.
Analyses involving patients who would have been excluded from, or were under-represented in, COPD treatment trials mostly suggests that treatment effects for FP-SAL are similar in patients aged > 80 years, those with mild COPD and those with both asthma and COPD. However, some potential differences were also suggested. For people with mild COPD, the use of FP-SAL appears to be more beneficial with respect to exacerbations than was seen in the TORCH trial-analogous3 population. By contrast, we observed a small increased risk of mortality when comparing FP-SAL with SAL in the group with both COPD and asthma. These associations should be interpreted with caution, and we recommend that future studies in CPRD GOLD and in other data sets focus on further characterising these associations.
Overall, we have demonstrated the utility of non-interventional data to investigate the expected treatment effects of COPD medications, in both trial-included and trial-excluded patient groups. Analyses largely suggest that COPD treatment effects are consistent across different patient groups, but highlighted a small number of possible differences that need to be investigated further in CPRD GOLD and other data sets before any conclusions that might have an impact on COPD treatment decisions can be reached. Unanswered questions about the effectiveness of currently recommended COPD inhaled combination therapy (other than FP-SAL) in patients excluded from trials should also be investigated using these methods. In addition, further work on advanced technique (e.g. high-dimensional propensity scores) could be performed to investigate whether or not placebo-controlled RCTs can ever be replicated in this therapeutic area (particularly when the target trial is a historical trial that has recruited patients in the placebo group who are much more sick than patients under the care of clinicians in a much more recent primary care setting).
Prioritised list of recommendations for future research
- To further investigate the harmful effect of FP-SAL (vs. SAL) observed with respect to mortality in people with asthma within the CPRD GOLD data set.
- To update the cohort so that (1) patients can be selected up to January 2020 and (2) patients can be selected from both CPRD GOLD and CPRD Aurum databases to increase the power of the subgroup analyses (i.e. for those aged > 80 years, those with mild COPD and those with concomitant asthma) and to facilitate analysis in those groups that we were unable to analyse because of small numbers (i.e. people with a history of lung surgery, people with a history of long-term oxygen therapy, people with substantial comorbidity or people with evidence of drug/alcohol abuse).
- To apply the methodology developed here to the analysis of other COPD treatments that are of current interest and relevance to prescribing practice and now specified in NICE guidance, such as the use of new triple therapy formulations (e.g. LAMA–ICS–LABA) compared with ICS–LABA or LABA–LAMA formulations.5
- To investigate advanced techniques (e.g. high-dimensional propensity scores) to assess whether or not it is possible to replicate placebo-controlled outcomes analyses in this therapeutic area when comparing with historical trials.