Identification of novel phenotypes of acute lung injury using multimodal longitudinal data

点击翻译按钮获取中文摘要

Acute lung injury (ALI) and its severe form acute respiratory distress syndrome (ARDS) are life-threatening conditions commonly seen in the intensive care unit (ICU) where a patient's lungs become damaged comprimising their ability to take in oxygen. The condition is known as a 'syndrome' because it has a variety of causes and can progress in many ways. The condition can happen after a problem in the lung such as a chest infection or alternatively, can be due to inflammation from an illness in another part of the body. Before COVID-19 approximately 10% of critical care beds were occupied by patients with this condition. This has dramatically increased as this is the most common way that COVID-19 patients become critically unwell and die. Approximately 20,000 patients a year suffer with ARDS in the UK, and nearly half die. Patients with this condition require life support on a ventilator. Studies over 20 years ago demonstrated that careful management of ventilator settings is important to protect patients' lungs. The only other therapy that has shown significant benefit is to lay the patient on their front (proning). Except for patients with COVID-19, no medication has been shown to be of benefit in ARDS, despite decades of research.We think that the reason that it has been so difficult to find new treatments is that the ARDS represents a range of different diseases which have been grouped together. Therefore something that works for one subgroup may not work for another. This has been demonstrated with COVID-19: this is a form of ARDS with one specific cause. The focused research on this group has led to several new treatments.My research will focus on using clinical data generated routinely through admission to ICU, as well as the chest x-rays taken, to identify different subgroups within ALI. This has previously been shown possible using specific laboratory tests which are not easily performed. The advantage of this approach is to take data already collected and use a variety of computer-based methods to find patterns that become apparent when hundreds of different variables from thousands of patients are analysed. These methods are summarised as 'machine learning' where the computer learns patterns from the data. To build on previous work I will focus on how a patient changes over time (their 'trajectory') and this will define their subgroup. It is also novel to combine information from x-rays with clinical data. Once these subgroups have been identified, I will assess whether they make sense to doctors and how they compare with previously identified subgroups using biological data. This will improve our understanding of the illness and provide indications as to personalised therapies.I will then analyse whether different treatments are more beneficial in specific subgroups. These treatments will include different settings on a ventilator and medications. For example, I may identify a subgroup of patients that have more inflammation driving their illness. There is evidence for this from previous studies of biological data. This subgroup may benefit from anti-inflammatory medicines such as steroids. Steroids have side effects such as impairing the immune system and therefore giving them in a non-targeted way may be harmful.Ultimately this approach will advance the field of personalised medicine with the eventual goal of providing personalised treatment to every patient by selecting the right treatment for the right patient at the right time. The COVID-19 pandemic has demonstrated that by understanding one specific cause of respiratory illness we can rapidly identify effective treatments. Personalising our approach to this condition is a critical step towards improving survival of this common critical illness.