Introduction to precision medicine in COPD

Precision medicine versus “evidence-based” care

Currently, clinicians are encouraged to provide “evidence-based” care for their patients, which was originally defined as “the conscientious, explicit and judicious use of current best evidence in making decisions about the care of individual patients” [8]. The highest level of evidence is believed to be derived from multiple large randomised controlled trials (RCTs) that generate narrow confidence intervals [9]. Accordingly, over the past two decades there has been a rise in very large RCTs in COPD that have enrolled thousands (and in some cases tens of thousands) of patients to generate p-values (on primary outcomes) that fall below the “magic” threshold of 0.05 (two-tailed). Thus, a “positive” study is one that meets this threshold, whereas those that do not are considered “failures”.

An obvious weakness of this evidence-based approach is that these RCTs generally do not consider the heterogeneity of disease or patients. There is mounting evidence that COPD is not a single disease entity but rather a collection of abnormalities, driven by different molecular processes or pathophysiologies [10]. Thus, for practicing clinicians, implementing evidence-based data from large therapeutic trials for individual patients in their everyday practice is an enormous challenge. Use of inhaled corticosteroids (ICS) is a prime example of this conundrum. Whereas evidence-based data would suggest that most COPD patients do not benefit from ICS therapy, clinicians in “real-world” practice often use ICS therapy for their patients with COPD. One important reason is that ∼10–25% of patients with COPD in the real world also have asthma [11]. Although these patients are generally excluded in therapeutic RCTs, they cannot be ignored in clinical practice.

The weakness of the one-size-fits-all care approach has given rise to the concept of precision medicine [12]. Although there is no universally accepted definition of precision medicine, the National Institutes of Health (NIH) defines it as “treatments targeted to the needs of individual patients on the basis of genetic, biomarker, phenotypic, or psychosocial characteristics that distinguish a given patient from other patients with similar clinical presentations” [12] (box 1). Implicit in this definition is the goal of improving clinical outcomes for individual patients and minimising unnecessary side effects for those less likely to have a response to a particular treatment [13]. Simply, precision medicine is “prevention and treatment strategies that take into account individual variability in genes, environment and lifestyle” [12]. An extreme version of precision medicine is personalised medicine, in which the entire management is structured for (or catered to) an individual and thus may not be generalisable beyond that patient.

Biomarkers: an essential component for precision medicine

One key component of enabling precision medicine at the bedside is biomarkers. Although many consider biomarkers synonymous with “blood tests”, an NIH expert panel defined them more broadly “as factors that are objectively measured and evaluated as indicators of normal biological or pathological processes, or pharmacological responses to therapeutic intervention” [14]. In practice, biomarkers can be sputum or blood tests, imaging modalities, prediction rules or, more broadly, treatable traits [15]. Biomarkers can be sub-classified as diagnostic, response (further divided into pharmacodynamic biomarkers and surrogate endpoint biomarkers), prognostic (correlated with clinical outcome but not necessarily directly related to specific mechanisms) or predictive (predict response to specific targeted drug interventions) [16].

The ratio of the forced expiratory volume in 1 s (FEV₁) to the forced vital capacity, based on post-bronchodilator spirometry, is an example of a diagnostic biomarker. The body mass index, airflow obstruction, dyspnoea and exercise (BODE) index is an example of a prognostic biomarker, and provides information on the future risk of mortality in a given patient. Blood eosinophil count is an example of a predictive biomarker in determining therapeutic guidance for ICS. Serum concentration of α1-antitrypsin in those with emphysema related to SERPINA1 deficiency is an example of a response biomarker.

In the context of precision medicine, predictive biomarkers are of prime importance because they guide therapeutic choices and make therapies safer and more cost-effective. There are several biomarkers in practice that are commonly used to target specific therapies for specific subgroups of COPD patients (figure 2). For example, according to the National Emphysema Treatment Trial (NETT) study, the overall number of patients needed to treat (NNT) for lung volume reduction surgery (LVRS) to prevent one death over 5 years is ∼246 [17]. However, by targeting LVRS to those patients with predominantly upper lobe disease (as determined by thoracic computed tomography (CT)) and low exercise capacity (defined as a maximal workload of 25 W in women and 40 W in men post-rehabilitation), the NNT improves to seven [17]. Cazzola et al. [18] showed in their meta-analysis of RCTs that provision of “triple” therapy consisting of an ICS, a long-acting β2 agonist (LABA) and a long-acting muscarinic antagonist (LAMA) was associated with an NNT of 38 to prevent at least one exacerbation over 1 year compared with dual LABA/LAMA combination therapy. However, this NNT improved to nine when triple therapy was given only to those with “high” blood eosinophils (≥300 cells·µL⁻¹). Treatment with continuous oxygen therapy is associated with an NNT of five to prevent one death over 3 years in those with resting hypoxaemia (arterial oxygen tension (Pa_O2) <60 mmHg) [19], whereas the NNT inflates to 56 for provision of continuous oxygen therapy for those with just nocturnal or exertional hypoxaemia [20]. In the Prevention of Exacerbations with Tiotropium in COPD (POET) trial, the use of a LABA (tiotropium) was associated with an 18% relative reduction in the risk of exacerbations compared with a LABA (salmeterol) in patients with moderate to severe COPD (corresponding to an NNT of 24 over 1 year). A subgroup analysis showed that in patients with a Global Initiative for Chronic Obstructive Lung Disease (GOLD) severity grade of 4 (FEV₁ <30% of predicted) [7], there was a 36% relative risk reduction (corresponding to an NNT of seven) with LAMA use [21]. More recently, use of bronchoscopy to ascertain collateral ventilation was demonstrated to be an effective biomarker in predicting therapeutic responses to endobronchial valve therapy in COPD patients with emphysema [22].

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 2

A comparison of number needed to treat (NNT) between non-biomarker and biomarker-based approaches to chronic obstructive pulmonary disease (COPD) therapies currently available in most jurisdictions. x-axis denotes therapies currently available for COPD patients, with the therapeutic trial or meta-analysis from which NNTs were derived indicated above the bars. y-axis denotes NNT to prevent one event of clinical importance (over a fixed period of time) for each intervention with or without the use of biomarker(s). LVRS: lung volume reduction surgery; O₂: supplemental (domiciliary) oxygen therapy; LAMA: long-acting muscarinic antagonist; LABA: long-acting β2 adrenergic agonist; Triple: a combination of inhaled corticosteroids and LABA and LAMA; NETT: National Emphysema Treatment Trial; LOTT: Long-term Oxygen Treatment Trial; MRC: Medical Research Council Oxygen Therapy Trial; POET: Prevention of Exacerbations with Tiotropium in COPD; GOLD: Global Initiative for Chronic Obstructive Lung Disease; Eos: blood eosinophil count in absolute scale (cells·µL⁻¹). NNT is calculated based on persons.

Biomarkers as a gateway for future drug development

Although there have been notable improvements in the care of COPD patients over the past several decades, the pace of new developments in respiratory diseases has been extremely slow and fraught with repeated failures. The cumulative probability of respiratory drugs reaching the clinic is only 3% (from phase I to regulatory approval), whereas it is 14% for HIV/AIDS drugs and 7% for cancer therapeutics [6]. The greatest attrition occurs during phase II and III studies, mostly owing to lack of efficacy (accounting for ∼60% of failures) [6]. Traditionally, pharmaceutical companies have relied on preclinical animal models to assess potential therapeutic targets and compounds. However, because they are poorly predictive of the human condition, they have been largely abandoned in favour of genomics-based approaches to drug discovery and development in COPD [23].

There are several common reasons for the repeated failures of drug development in COPD (table 1), including inadequate target engagement of the drug, poor patient selection and use of clinical endpoints that are insensitive or inaccurate for detecting adequate treatment responses [24]. Drug companies have found that discovery and implementation of biomarkers to phenotype patients and select only those who are likely to experience benefit from the drug dramatically increases the probability of success of novel drugs in phase IIa trials from ~29% (pre-biomarker implementation) to 82% (with biomarker implementation) [24].

Although it is widely acknowledged that biomarkers are urgently needed to address gaps in drug development, biomarker implementation has been difficult owing to multiple barriers (table 2). Academics often work in silos. However, as with therapeutic products, biomarker development requires academics to work closely with industrial partners to “push” discovery into validation and ultimately into clinical utility studies for regulatory qualification and clinical implementation. Because this process usually takes many years, often requiring millions of pounds in investment (with no guarantee of success), very few companies are willing to take these risks by themselves. To accelerate biomarker development, novel public–private funding models will be needed in the future to de-risk these efforts. Further, similar to genomics data, biomarker data should be publically available on searchable and easy-to-find databases with adequate annotations and on high-quality platforms to enable “data mining”, reproduction and validation of results and to prevent unnecessary duplication of efforts.

Other enablers of precision medicine

Beyond biomarkers, clinicians may use clinical features to guide therapeutic choices. For instance, symptoms of chronic bronchitis (e.g. daily productive cough) significantly modify the therapeutic efficacy of roflumilast, a PDE4 inhibitor [25]. These features are also being used to pursue investigational treatments in COPD, including mucolytics and radiofrequency as well as cryoablation therapies of the airways (clinicaltrials.gov). Pharmacogenomics, which combines pharmacology and genetics to study how genes affect patient responses to particular drugs, is another enabler of precision medicine. Although there are no clear examples of pharmacogenomics in COPD, in cystic fibrosis, genotyping (G551D) is currently being used to guide therapeutic choices for ivacaftor [13]. Given the heterogeneity of COPD, precision medicine would also be enabled by targeting therapies based on disease endotypes. For example, anti-IgE therapy may be highly effective in COPD patients whose disease is driven largely by a severe allergic endotype [26].

Scale of estimation versus scale of interest in precision medicine

In addition to the above, in precision medicine the benefits as well as the risks of therapeutic products need to be accurately assessed. Treatment effects in clinical trials and observation studies are often estimated and reported on a relative scale, such as per cent reduction in the risk or rate of events (scale of estimation). However, in precision medicine, what matters more is the absolute risk or rate that is modified by the therapy of interest (scale of interest). This approach lends itself to metrics that are easily understood in clinical practice and thus have clinical relevance: the “number needed to treat” (NNT) and the “number needed to harm” (NNH). The original concept of NNT (or NNH) is based on the following formula: 1 divided by the absolute risk reduction between the experimental therapy and placebo over the study's duration [27].

In most therapeutic trials in COPD, the primary outcomes are derived from a Cox (proportional hazard) survival model (to compare time to first exacerbation between treatment arms), logistic regression (to compare the proportion of patients with at least one exacerbation during follow-up between treatment arms) or count models such as Poisson or negative binomial regression (to compare the rate of exacerbations during follow-up between treatment arms). These regression models produce treatment effects, which are represented as hazard ratios, odds ratios or rate ratios, respectively. These parameters all generate numbers on a relative scale and not on an absolute scale, which can cause confusion for the reader.

An example to illustrate the major difference in risks between relative and absolute scales can be found in the Macrolide Azithromycin for Prevention of Exacerbations of COPD (MACRO) study. The MACRO study was a landmark clinical trial that demonstrated a 27% reduction in the risk of exacerbations with daily low-dose azithromycin therapy (rate ratio 0.73, 95% CI 0.63–0.84, p<0.001) [28]. However, it is widely known that there is tremendous variability in the background rate of exacerbations across patients. A recent study [29] estimated that the range (containing 95% of the study population) of annual background rate of moderate to severe exacerbations (without azithromycin therapy) was between 0.47 (i.e. one exacerbation every 564 days) and 4.22 (i.e. one exacerbation every 87 days). For the individual with a background exacerbation rate of 0.47, a 27% relative reduction in the exacerbation rate with azithromycin would translate to a prevention of 0.13 exacerbations per year, corresponding to an event-based NNT of 1/0.127=7.88. For those with a background exacerbation rate of 4.22, azithromycin therapy would prevent 1.14 exacerbations per year, corresponding to an event-based NNT of 1/1.14=0.88. In this context, the provision of azithromycin is likely justifiable clinically and economically for the latter group of patients, whereas for the former group long-term azithromycin therapy may not be justifiable owing to its side effects and costs. One limitation of this therapeutic approach is that it is difficult (if not impossible) to predict the background rate of exacerbation for a given patient. Efforts are underway to develop simple and reasonably accurate clinical prediction tools using easily verifiable patient characteristics and traits, e.g. age, sex, number of exacerbations in the previous year and FEV₁, that can be used clinically to estimate the future risk of exacerbations in a given patient [30].

Event-based versus patient-based NNTs

In the MACRO example, an event-based NNT was used to illustrate the importance of the background exacerbation rate of an individual patient. This approach may be reasonable given that patients can experience repeated exacerbation events in a given year. However, some endpoints are distinct, non-repeated events. Death would be one such example. The Towards a Revolution in COPD Health (TORCH) trial was powered on total mortality [31]. In this trial, 193 of the 1533 patients (12.6%) in the fluticasone/salmeterol arm died, compared with 231 of the 1524 patients (15.2%) in the placebo arm over 3 years of treatment. Thus, the NNT was 1/(0.126–0.152)=38. In other words, on average, the provision of fluticasone/salmeterol for 3 years would prevent one death among 38 patients who were treated with this therapy (compared with placebo). The TORCH trial also showed that fluticasone/salmeterol combination therapy was associated with an increased risk of pneumonia. The risk was 19.6% in the treatment arm compared with 12.3% in the placebo arm. The NNH was 1/(0.196–0.123)=14. Thus, there was one extra case of pneumonia for every 14 patients taking a fluticasone/salmeterol combination compared with placebo over 3 years.

A patient-based NNT (e.g. death or pneumonia NNT calculations for TORCH) is calculated by first dividing the number of patients who experience at least one event over the study period by the number of patients at risk for each of the study groups (i.e. absolute risk) and then taking the difference in the absolute risk between the study groups. For repeated events such as exacerbation, an event-based NNT can be calculated by first dividing the total number of exacerbations by the follow-up time for individuals in each of the study groups (thus generating absolute rates) and then taking the difference in the absolute rates between the study groups. Some have criticised this approach because the results are hard to interpret and, clinically, physicians treat patients and not events [32]. An alternative approach is to dichotomise the numerator of event-based NNTs by considering only the first exacerbation (or time to first exacerbation) and using the number of at-risk patients in the denominator (rather than person-time). This renders the following simple clinical interpretation: the number of patients needed to treat over a given unit of time (e.g. a year) to prevent at least one exacerbation over the study period [32]. In general, event-based NNTs are lower than patient-based NNTs. This is illustrated by the Informing the Pathway of COPD Treatment (IMPACT) trial for which the event-based NNT for exacerbations was approximately four, whereas the person-based NNT was approximately 34 comparing triple versus dual bronchodilator therapy over 1 year [33].

NNTs and precision medicine

In general, precision therapies are those that are associated with low NNTs. Indeed, the ideal precision medicine is one that has an NNT of one [34]. Even before the advent of the term “precision medicine”, physicians were employing N of 1 trials (a randomised controlled crossover trial in a single patient designed to establish optimal treatment for that patient) to determine the efficacy of therapeutics. One notable example in COPD is theophylline [35]. Several decades ago, it was a widely held belief that theophyllines improved outcomes in COPD by inducing bronchodilation and providing anti-inflammatory effects. In 1999, Mahon et al. [35] performed a series of N of 1 trials to determine whether theophylline improved health status and exercise tolerance in patients with moderate to severe COPD. They found that theophylline had similar effects to placebo for these outcomes, a finding that was replicated in a recent large RCT [36] consisting of 1567 patients (compared with n=34 for the study of Mahon et al. [35]). Given the enormous costs of bringing new drugs to the market, estimated to be $2.6 billion per pill [37], in the era of precision medicine, N of 1 trials should be considered early on in the drug developmental process [34].

Emerging predictive biomarkers of therapy

Summary and conclusions

Precision medicine is the future of COPD care and is enabled by the advent of biomarkers that can clearly identify subgroups of patients who will either benefit from novel therapies or experience only harm. Biomarkers are also required to gauge therapeutic responses during all phases of drug development to reduce failure rates and make drugs more affordable. To enable precision medicine, the benefits (as well as the harm) of therapeutics should be reported in both relative and absolute scales, which will enable calculation of metrics such as NNT and NNH. Finally, to reduce the cost of therapeutic trials (especially early on in development), drug companies should consider developing companion diagnostic tests to predict therapeutic and pharmacodynamic responses, and deploy N of 1 trials for therapies that target symptoms and modify risk (of exacerbations and/or disease progression) for committing to large-scale phase III trials.