Personalized Medicine for Chronic, Complex Diseases: Chronic Obstructive Pulmonary Disease as an Example

Josiah E Radder; Steven D Shapiro; Annerose Berndt


Personalized Medicine. 2014;11(7):669-679. 

In This Article

The Importance of Phenotyping Chronic Disease for Personalized Medicine

The human respiratory tract has a total surface area of between 750 and 1000 square feet in direct contact with the external environment, making it the most exposed organ in the human body. Because of this, the pathogenesis of most respiratory diseases is due to a complex combination of environmental and genetic factors. As the most common chronic respiratory disease in adults, COPD epitomizes this interaction. According to the Global Initiative for Chronic Obstructive Lung Disease (GOLD), COPD "is characterized by persistent airflow limitation that is usually progressive and associated with an enhanced chronic inflammatory response in the airways and the lung to noxious particles or gases".[21] Reduced airflow can be a consequence of either airway narrowing or emphysema – the destruction of the peripheral lung tissue leading to decreased elastic recoil. COPD is currently the fourth-leading cause of death in the USA and the world according to the WHO,[19] and unfortunately is predicted to become the third-leading cause of death worldwide by 2020.[22]

Even the basic definition of COPD begs for personalized medicine. Cigarette smoking, the most common cause of exposure to noxious particles in much of the world, leads to inflammation that manifests itself in the airways as physical narrowing (inflammation, mucus, fibrosis and other changes) that obstructs airflow. In the lung parenchyma, inflammatory protease-mediated destruction of elastic fibers reduces elasticity, thereby reducing expiratory airflow. Both of these changes lead to the same gross phenotype (airflow obstruction), but via entirely separate pathways. Lumping these two subpopulations of COPD patients together with a common airflow phenotype such as forced expiratory volume in 1 second (FEV1) will probably impair our ability to find genetic or other measurable attributes for the discreet and separate pathways involved in each pathway.

The improved classification of patients into carefully defined subpopulations will be necessary in order to determine disease susceptibility, understand the clinical course of a disease and prescribe appropriate pharmacological and other therapies. COPD patients have a variety of comorbidities and complications, such as lung cancer, pulmonary hypertension, severe weight loss, osteoporosis and depression. Heterogeneity in imaging, decline in lung function and survival further emphasizes the syndromic nature of this disease. Indeed, a subset of patients that is largely but not entirely related to exacerbations lead to many hospital readmissions. Understanding who these patients are and identifying better ways of treating them will have a major effect on both quality of care and healthcare costs.

Central to this effort is the definition of the term 'phenotype'. It has been proposed that a COPD phenotype should be defined only by attributes that are clinically relevant and that a phenotype should remain a 'candidate phenotype' until validated clinically.[23] While it seems clear that the clinical validation of any changes in therapy or guidelines based on phenotype should occur, the relegation of nonclinically validated phenotypes to a secondary status ignores the fact that there are likely to be clinically relevant attributes that we do not yet recognize. Some of the benefits of avoiding confining a phenotype to attributes that are recognized to be clinically relevant to a disease are outlined in-depth elsewhere,[24] but since we lack a complete understanding of the mechanisms of many chronic disease, it is not a stretch to assume that there may be attributes that define a phenotype that we have not yet recognized.

Although prospective trials following patients through the progression of their disease and collecting comprehensive information on their health would be ideal for identifying more complex phenotypes, the length of progression in chronic disease and the lack of early diagnosis make this extremely difficult. In the ECLIPSE study, 2164 patients were followed longitudinally for 3 years in one of the largest prospective COPD trials thus far conducted.[25] While such studies have greatly contributed to our knowledge of clinically important phenotypes, they are still limited to a short period of observation for a disease that can last for decades. They also inherently fail to capture any information regarding the clinical appearance of the patient prior to diagnosis. Of course, while recent trials, such as COPDGene, have collected large amounts of both clinical and genetic data,[26] such studies are also limited by a priori assumptions regarding which health data should be collected.

Retrospective trials of large cohorts of patients can feasibly solve some of these problems and act as a complement to longitudinal, observational trials of chronic disease. Chronic disease patients frequently utilize the health system, and in many places, these visits are now almost ubiquitously recorded in the electronic health record (EHR). Phenotyping will necessitate more powerful use of this resource. In 2011, it was estimated that 54% of physicians in the USA had implemented an EHR[27] and incentive payments offered by the Health Information and Technology for Economic and Clinical Health (HITECH) Act of 2009 in the intervening years have certainly pushed this number higher. Under the 'meaningful use' requirements of this legislation, the adoption of EHR systems should be accompanied by improvements in care, as determined by measurable standards. Among these are requirements to maintain active records of medications, allergies, smoking status and history, among many other valuable pieces of phenotypic information across a clinician's entire practice.[28] Thus, clinical data is now being collected on an unprecedented level, opening up research possibilities that were previously unachievable.

Studies have demonstrated that diseases such as chronic pain and chronic rhinosinusitis can be accurately phenotyped from EHR data with well-designed algorithms.[29,30] In COPD, carefully considered selection criteria have allowed researchers to identify patients with clinical disease beyond diagnostic code assignment and even allow for initial predictions of severity from administrative data.[31] Impressively, a large study from the UK recently demonstrated that the retrospective analysis of COPD patients is not only capable of identifying patients with COPD, but also identifying when patients were diagnosed and whether this diagnosis could have been made earlier.[32] Once patients are well characterized using EHR data, it becomes possible to ask whether, in patients with a given chronic disease, there are unique attributes that associate together. This approach to defining phenotypes by clustering and principle component analysis has been used in both prospective and retrospective trials, often suggesting that current phenotypes do not accurately represent the variation occurring in chronic disease and supporting the argument that there are attributes of a phenotype that we do not yet recognize as clinically relevant.[33,34]