The future of acute and emergency care

Stratification: the new frontier in emergency care

From the earliest days of medicine, doctors have tried to pick the best treatment for each individual patient. However, in emergency care, broad groups of patients are often treated in the same way as insufficient information is available to adopt a more precise approach. A ‘sepsis bundle’, for example, is applied to all patients who fulfil ‘red flag’ criteria. It is unlikely that any single approach is optimal for such a diverse group of patients. Guideline development provides treatment algorithms that are ‘evidence based’, but these are designed for the average patient and so only constitute a best guess for any individual.

Greater stratification allows an increasingly precise approach with more individualised treatment. A large number of emergency point of care diagnostics are being developed, such as novel measures of physiology, novel biomarkers, measures of gene expression, new ways of using large data sets to make individual predictions and, on the horizon, rapid point of care genetic analysis.^1–4

Development of novel stratification requires a series of research steps (Fig 2). This includes a wide range of research skills from technology development to validation studies, and then to pragmatic clinical trials and assessment of cost effectiveness. This pathway requires such a wide range of academic and methodological skills that success is only likely within a large multidisciplinary group. Future emergency care researchers need to develop the skills to play their part in these collaborative efforts as there may be more benefit from precise application of existing treatments than discovery of a novel treatment.

• Download figure
• Open in new tab
• Download powerpoint

Fig 2.

Development pathway for a biomarker in emergency care.

Rapid diagnostics for better antimicrobial use in acute care

Antimicrobial overuse leading to pathogen resistance is a leading global healthcare challenge, with considerable potential to impact on the effectiveness of future care pathways. There is an acknowledged overuse of high-potency, potentially toxic, broad-spectrum antimicrobial agents in acute care. This relates to limitations of time-critical diagnostic information regarding the likelihood of infection and identification of a causal pathogen. Advances in molecular science have led to newer technologies aimed at delivering rapid diagnostic information. Laboratory-based and point-of-care technologies include accurate rapid measurement of circulating host response inflammatory molecules (eg C-reactive protein and procalcitonin) and non-culture based multiplexed pathogen detection. However, it remains uncertain whether these rapid technologies, and the information they provide, can improve antimicrobial use while maintaining patient safety.

An ongoing portfolio of National Institute for Health Research-funded multicentre clinical research seeks to determine whether available technologies can improve antimicrobial use at key clinical decision points for patients with suspected severe infection syndromes.

Precision medicine

Modern medicine should aspire to give the right treatment to the right patient at the right time. RCTs provide the highest quality evidence to inform on the effectiveness of treatments, but they are limited to providing an average effect among a certain population. Critical care has classically used syndromic definitions to identify populations with similar signs and symptoms, but these syndromes contain very heterogenous groups of patients who are likely to have very different treatment needs.¹¹

Attempting to use single biomarkers has not proven useful in guiding therapy.¹² Using multiple biomarkers and identifying patterns or signatures are, however, showing great promise. Using a combination of multiple biomarkers and clinical data, two distinct sub-phenotypes have been consistently identified among patients with ARDS.¹³ Importantly, these two groups of patients appear to respond differently to different treatments, including ventilation, fluid strategies and drug treatments.^13–15

In sepsis, RNA signatures can identify subphenotypes of patients with different outcomes.^16,17 Importantly, these signatures have been associated with different responses to treatment, notably to corticosteroids.¹⁸ This may, in part, explain why different clinical trials have frequently yielded conflicting results.¹⁹ To date, these differential treatment effects have only been demonstrated in retrospective analyses of clinical trials so prospective testing is required. The technology now exists and is being evaluated.²⁰

Nevertheless, even using an ‘-omics’ approach will still result in treating groups of patients, even if they are better defined and more homogeneous. The ultimate goal would be to individualise treatment for each patient. This will require techniques beyond just biomarker-guided clinical trials and will likely require machine learning / artificial intelligence (AI) techniques. Such approaches have already started to show promise to truly individualise clinical medicine within intensive care.²¹

Novel monitoring

While critical care currently uses a wide array of monitoring techniques assessing physiological and biochemical variables, there are some glaring deficiencies that need to be addressed. Prime among these are the inability to optimise antibiotic dosing and the early recognition of organ hypoperfusion and dysfunction.

With regard to antibiotics, abnormal drug handling in critical illness undermines the validity of fixed dosing regimens or simplistic algorithms based on estimates of glomerular filtration. Altered protein binding, distorted volumes of distribution, renal and/or hepatic dysfunction and augmented renal clearance may all conspire to produce significant under- or over-dosing.²² Intriguingly, the so-called ‘obesity paradox’ in critical illness (ie lower mortality in the (morbidly) obese) may relate to excessive weight-adjusted doses of antibiotics and other treatments delivered to normal-weight patients.²³ While an important emphasis is directed towards administering an appropriate antibiotic, little attention has hitherto been paid to ensuring adequate yet not excessive dosing. Other than daily measurement of aminoglycosides and glycopeptides for toxicity, no other antibiotic classes are routinely measured, let alone continuously monitored to assess peak, trough and area under the curve concentrations. Perhaps of greater relevance is the need to assess tissue antibiotic levels as blood levels are poorly reflective; the need is to adequately treat tissue-based infection (such as pneumonia or peritonitis) rather than the scant amount of bacteria present within the circulation.²⁴ The concept of continuous therapeutic drug monitoring is being advanced, not just within blood but within tissues using a variety of techniques including microneedle biosensors, microdialysis and microfluidics.²⁵

Current biomarkers used to detect organ hypoperfusion and dysfunction only become deranged when significant perturbations have occurred; for example, serum creatinine rises when glomerular filtration falls by at least 50%. Other blood biomarkers are often non-specific and reflect a balance of tissue production versus excretion or, as in the case of lactate, its utilisation as an important fuel source in stress states. Earlier, reliable identification of organ hypoperfusion/dysfunction, ideally using continuous monitoring, will prompt faster intervention with optimisation of treatment by titrating to sensitive endpoints. Hopefully, this will lead to prevention of organ dysfunction and better outcomes. A variety of monitoring technologies are being explored, such as tissue oxygen tension, mitochondrial redox status and mitochondrial PO₂ in animal models and early patient studies.^26–28 Their utility does, however, need to be validated in outcome studies.

Artificial intelligence

A discussion of medical research in the Google DeepMind era must soon turn to AI. Yet without resorting to randomisation, these tools can only optimise the observable and unmeasured confounding effects will occasionally lead us astray.²⁹ The rules of Go and the complexities of protein folding are transparent compared with bedside clinical medicine.³⁰

The alternative, the RCT, is slow, expensive and cumbersome. Even with more efficient modern designs, such as those deployed during the COVID-19 pandemic, the answers are only ‘true’ on average.³¹ Subtle endotypes create treatment heterogeneity not exposed by the RCT.³²

Hitherto, the solution depends on the skill of the clinician to take that average answer and decide if it applies to their patient. Inevitably, this results in variation in practice; while we share expertise, each clinician's personal experience differs.

Such variation is normally considered a problem.³³ But it may hold the solution to the tension between observational and experimental research. Instead of just using clinical decision support (CDS) systems to reduce variation where strong evidence is available, they could be used to create randomisation opportunities where evidence is sparse and variation widespread. With both patient involvement to select topics suitable for opt-out consent and a flexible approach to randomisation that nudges rather than mandates treatment allocation, unbiased learning events can be generated without interrupting the delivery of care. If I am uncertain whether to stop antibiotics today or tomorrow, then I could learn from my uncertainty. If I do know, I overrule the nudge, and the system learns from me.

Hundreds of millions of such decisions are made every day. Most are so small that the clinician does not always notice, eg prescribing paracetamol for a fever or setting a target oxygen saturation. It is inconceivable that the aggregate effect of all these small decisions is unimportant. However, to deliver personalised and not ‘average’ medicine, ‘big data’ needs to embrace experimental methods as well as observational ones.

Respiratory

In patients with acute respiratory failure, invasive mechanical ventilation is lifesaving, but also contributes to morbidity and mortality. Research into innovative approaches is needed to limit the harm associated with mechanical ventilation.

In the setting of COVID-19, the RECOVERY-Respiratory Support trial is evaluating the effectiveness of high-flow nasal oxygen or continuous positive airways pressure against standard oxygen therapy with a primary outcome of tracheal intubation or mortality within 30 days of randomisation. This trial should inform on the role of non-invasive respiratory support in COVID-19.

Prone ventilation is well established in the setting of mechanical ventilation patients with severe acute respiratory distress syndrome (ARDS). Although COVID-19 has prompted interest in the role of prone ventilation in awake patients, this concept is not new and the early reports of the use of prone positioning in the 1970s described its use both in patients who were invasively ventilated and also in awake patients to avoid intubation.^35,36 Given the potential for this simple intervention to improve outcomes in patients who are not intubated, trials in patients with acute hypoxaemic failure to avoid intubation as well as in patients following major surgery to avoid post-operative pulmonary complications are needed.

Airway pressure release ventilation (APRV) is a ventilation mode whereby pressure is alternated from a high level (Phigh) applied for a prolonged period (Thigh) to maintain lung volume and alveolar recruitment, to a low level (Plow) for a short period (Tlow). The underlying rationale is to maintain a pressure above the alveolar closing pressure for a sustained time, limiting the release time to allow carbon dioxide removal while avoiding lung de-recruitment. APRV also allows spontaneous breathing during any phase of the respiratory cycle. A systematic review suggested APRV resulted in a higher number of ventilator-free days and lower hospital mortality, although the underpinning evidence was low quality.³⁷

Current approaches to weaning from mechanical ventilation require clinicians to gradually reduce ventilatory support. Emerging technologies can automatically adjust ventilation settings so these systems could both standardise and optimise the weaning process. A recent systematic review found that automated weaning methods may reduce the duration of ventilation.³⁸

In recent years, the technology underpinning extracorporeal support has evolved and this may broaden indications for its use. Extracorporeal membrane oxygenation (ECMO) is increasingly established as an adjunct to protective mechanical ventilation for patients with severe ARDS.³⁹ Whether extracorporeal carbon dioxide removal offers a protective ventilatory strategy with improved patient outcomes remains uncertain.⁴⁰ Other potential indications for extracorporeal support include prevention of intubation, facilitation of extubation, and as an alternative to non-invasive ventilation.⁴¹ Large-scale clinical trials are warranted to determine if these innovative technologies are clinically effective.

Finally, the findings of outcome benefit with steroids and IL-6 blockade in COVID-19 will renew interest in emerging pharmacological (including cell-based) therapies for patients with ARDS.^42–45 It is likely that a precision medicine approach using point-of-care testing will inform the design of future trials of such therapies.⁴⁶

Cardiovascular disease

Cardiovascular disease remains the world's biggest killer.⁴⁷ The worldwide prevalence of heart failure is 64.34 million and increasing, accounting for 9.9 million years lost due to disability and $346 billion expenditure.⁴⁸ Evidence-based interventions have resulted in significant improvements in survival and quality of life in patients with long-term heart failure. However, outcomes remain persistently poor for those with its most severe manifestation, cardiogenic shock, with admission mortality rates of 27%–51%. The reasons are multifactorial and include variable syndrome definition, late recognition, lack of understanding of associated systemic pathophysiology, reliance on global rather than regional circulatory assessment/interventions, poor understanding of mechanisms of critical cardiac dysfunction, failure to tailor/individualise therapies and persistence in using interventions that are, at best, not beneficial and, at worst, harmful.^49,50

Numerous factors have converged to fundamentally influence how research in this field is moving. First, technological evolution has led to a paradigm shift in approaches to circulatory support, using mechanical circulatory support (MCS) to ‘rest’ the heart rather than inotropes to drive contractility and, for non-reversible causes of severe heart failure, potentially providing an alternative to transplantation.^51,52 For acute MCS, research is required to understand the impact on the myocardium and wider circulation, and to determine which patients will benefit, from which support, and when and how to implement. Second, integrating high-content ‘-omics’ to characterise each patient should allow for better understanding of the mechanisms of cardiac dysfunction in critical illness and lead to application of individualised therapies. Third, advanced imaging (including molecular) allows the potential for local drug delivery (‘theranostics’). Finally, shock (as currently defined) is a late manifestation of a catastrophic cardiac insult. Here the rapid advance in technology (wearables, innables, outables, big data and related technologies) should define which metrics indicate the patient is early in their shock pathway and, potentially, how to intervene.

Acute brain injury

Traumatic brain injury (TBI) has been characterised as the most complex disease in the most complex organ.⁵³ Yet, despite this great heterogeneity, treatment guidelines have traditionally used a ‘one size fits all’ approach. Greater use of blood biomarkers, advanced magnetic resonance imaging and genomics, with improved integration of continuous high-resolution physiological monitoring, offers the opportunity for better patient characterisation and stratification.⁵⁴ This will enable improved patient selection and enrichment of clinical trials for subsets most likely to benefit, and avoid including those unlikely to benefit and/or suffer adverse consequences. Perhaps most importantly, these highly granular data offer the potential for more personalised treatment approaches.

Improvements in knowledge and application of invasive neuromonitoring techniques will enable more widespread use outside academic centres with bespoke neuromonitoring programmes. This scale-up ability will facilitate previously not possible clinical trials. Two examples of such trials using invasive brain tissue oxygen-guided therapy are Brain Oxygen Optimization in Severe TBI, Phase 3 (BOOST3) and the Brain Oxygen Neuromonitoring in Australia and New Zealand Assessment (BONANZA).^55,56 Individualised cerebral perfusion pressure management based on cerebral autoregulation indices (such as cerebrovascular pressure reactivity) offer the opportunity to provide personalised dynamic targets to guide management; a feasibility study is now complete.⁵⁷ It is likely that more personalised approaches to TBI management will become the norm over the next few years. Similar approaches in other acute neurological and neurosurgical diseases also offer the potential to improve outcomes.

There is also an increasing recognition that all patients admitted to intensive care, even for primarily extra-cranial illnesses, have a vulnerable brain. Acute neurological dysfunction in the form of delirium is common, and is associated with long-term cognitive dysfunction.⁵⁸ One-third of survivors of acute respiratory failure or shock have significant cognitive impairment 1 year later.⁵⁹ Improvements in technology for non-invasive neuromonitoring and advances in analysis using machine learning (for example with bedside electroencephalography) may help to facilitate cross-over from neurocritical care concepts into general intensive care. This, in turn, may help to improve outcomes.

Long-term consequences of critical illness

Eighty per cent of critically ill patients survive to be discharged from intensive care. This survival is not without cost. Acute muscle wasting leads to profound weakness and significantly impaired function.⁶⁰ One-third of patients remain carer-dependent for basic activities of daily living at 1 year.⁶¹ Compounding this, anxiety, depression and post-traumatic disorder are common. Forty per cent of young patients suffer from persistent cognitive deficit, akin to survivors of TBI or those with early-onset Alzheimer's disease.⁶² At 5 years, one-third of previously employed patients have not returned to any form of work.⁶³ Rehabilitation from critical illness is recognised as a public health issue in National Institute for Health and Care Excellence guidance CG83; the poor understanding of the pathophysiology of muscle wasting was identified as a major therapeutic barrier.⁶⁴

To date, no effective measures for primary or secondary prevention of these newly-acquired functional deficits exist. The field suffers from fundamental knowledge gaps. The underpinning biology of some aspects are being revealed, while others (such as that of cognitive decline) remain unclear. Biological heterogeneity is affected by pre-existing chronic disease states, but the impact of patient heterogeneity and sub-phenotypes on RCT outcomes are only now being realised.^20,65 To address this, translational research needs to be integrated into clinical trials to further reveal mechanisms of differential responses to interventions. Outcome measures in the field of critical illness have traditionally been related to mortality; there is an urgent need for patient-centred functional outcome measures with appropriate clinimetric properties.²⁰ Lastly, social, fiscal and cultural factors are tightly interlinked with patient outcomes.⁶⁶ Data regarding these factors are rarely recorded or used, and this gap is likely to be increasingly important in the post-COVID-19 world.

Conflicts of interest

Virginia Newcombe is supported by an Academy of Medical Sciences / The Health Foundation Clinician Scientist Fellowship. Anthony Gordon is funded by an NIHR Research Professorship (RP-2015-06-18) and support by the NIHR Imperial Biomedical Research Centre. Mervyn Singer and Oxford Optronix are funded by a Wellcome / Deptartment of Health Innovation Grant to assess the role of bladder tissue oxygen monitoring in critically ill and high-risk perioperative patients. Mervyn Singer recently held a MedCity grant to investigate the potential utility of tissue antibiotic monitoring. Mervyn Singer sits on the advisory board of Spiden who are developing continuous monitoring techniques combined with AI. Paul Dark is supported by the NIHR Manchester Biomedical Research Centre.