Collaboration between explainable artificial intelligence and pulmonologists improves the accuracy of pulmonary function test interpretation.

BACKGROUND: Few studies have investigated the collaborative potential between artificial intelligence (AI) and pulmonologists for diagnosing pulmonary disease. We hypothesised that the collaboration between a pulmonologist and AI with explanations (explainable AI (XAI)) is superior in diagnostic interpretation of pulmonary function tests (PFTs) than the pulmonologist without support. METHODS: The study was conducted in two phases, a monocentre study (phase 1) and a multicentre intervention study (phase 2). Each phase utilised two different sets of 24 PFT reports of patients with a clinically validated gold standard diagnosis. Each PFT was interpreted without (control) and with XAI's suggestions (intervention). Pulmonologists provided a differential diagnosis consisting of a preferential diagnosis and optionally up to three additional diagnoses. The primary end-point compared accuracy of preferential and additional diagnoses between control and intervention. Secondary end-points were the number of diagnoses in differential diagnosis, diagnostic confidence and inter-rater agreement. We also analysed how XAI influenced pulmonologists' decisions. RESULTS: In phase 1 (n=16 pulmonologists), mean preferential and differential diagnostic accuracy significantly increased by 10.4% and 9.4%, respectively, between control and intervention (p<0.001). Improvements were somewhat lower but highly significant (p<0.0001) in phase 2 (5.4% and 8.7%, respectively; n=62 pulmonologists). In both phases, the number of diagnoses in the differential diagnosis did not reduce, but diagnostic confidence and inter-rater agreement significantly increased during intervention. Pulmonologists updated their decisions with XAI's feedback and consistently improved their baseline performance if AI provided correct predictions. CONCLUSION: A collaboration between a pulmonologist and XAI is better at interpreting PFTs than individual pulmonologists reading without XAI support or XAI alone.

Predicting Need for Escalation of Care or Death From Repeated Daily Clinical Observations and Laboratory Results in Patients With Severe Acute Respiratory Syndrome Coronavirus 2.

We compared the performance of prognostic tools for severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) using parameters fitted either at the time of hospital admission or across all time points of an admission. This cohort study used clinical data to model the dynamic change in prognosis of SARS-CoV-2 at a single hospital center in the United Kingdom, including all patients admitted from February 1, 2020, to December 31, 2020, and then followed up for 60 days for intensive care unit (ICU) admission, death, or discharge from the hospital. We incorporated clinical observations and blood tests into 2 time-varying Cox proportional hazards models predicting daily 24- to 48-hour risk of admission to the ICU for those eligible for escalation of care or death for those ineligible for escalation. In developing the model, 491 patients were eligible for ICU escalation and 769 were ineligible for escalation. Our model had good discrimination of daily risk of ICU admission in the validation cohort (n = 1,141; C statistic: C = 0.91, 95% confidence interval: 0.89, 0.94) and our score performed better than other scores (National Early Warning Score 2, International Severe Acute Respiratory and Emerging Infection Comprehensive Clinical Characterisation Collaboration score) calculated using only parameters measured on admission, but it overestimated the risk of escalation (calibration slope = 0.7). A bespoke daily SARS-CoV-2 escalation risk prediction score can predict the need for clinical escalation better than a generic early warning score or a single estimation of risk calculated at admission.

Biologic therapy practices in severe asthma; outcomes from the UK Severe Asthma Registry and survey of specialist opinion.

BACKGROUND: several biological treatments have become available for management of severe asthma. There is a significant overlap in the indication of these treatments with lack of consensus on the first-line biologic choice and switching practice in event of treatment failure. AIMS: to evaluate outcomes of biologic treatments through analysis of the UK Severe Asthma Registry (UKSAR), and survey of the UK severe asthma specialists' opinion. METHODS: patients registered in the UKSAR database and treated with biologics for severe asthma in the period between January 2014 and August 2021, were studied to explore biologic treatments practice. This was complemented by survey of opinion of severe asthma specialists. RESULTS: a total of 2,490 patients from 10 severe asthma centres were included in the study (mean age 51.3 years, 61.1% female, mean BMI 30.9kg/m2 ). Biologics use included mepolizumab 1,115 (44.8%), benralizumab 925 (37.1%), omalizumab 432 (17.3%), dupilumab 13 (0.5%), and reslizumab 5 (0.2%). Patients on omalizumab were younger and had earlier age of onset asthma than those prescribed mepolizumab or benralizumab. Patients prescribed mepolizumab and benralizumab had similar clinical characteristics. Those on benralizumab were more likely to continue treatment at approximately one year follow up (93.9%), than those on mepolizumab (80%), or omalizumab (69.6%). The first choice biologic differed between centres and changed over the study time period. Experts' opinion also diverged in terms of biologic initiation choice and switching practice. CONCLUSION: We observed significant variation and divergence in the prescribing practices of biologics in severe asthma that necessitates further research and standardisation.

Dynamic early warning scores for predicting clinical deterioration in patients with respiratory disease.

BACKGROUND: The National Early Warning Score-2 (NEWS-2) is used to detect patient deterioration in UK hospitals but fails to take account of the detailed granularity or temporal trends in clinical observations. We used data-driven methods to develop dynamic early warning scores (DEWS) to address these deficiencies, and tested their accuracy in patients with respiratory disease for predicting (1) death or intensive care unit admission, occurring within 24 h (D/ICU), and (2) clinically significant deterioration requiring urgent intervention, occurring within 4 h (CSD). METHODS: Clinical observations data were extracted from electronic records for 31,590 respiratory in-patient episodes from April 2015 to December 2020 at a large acute NHS Trust. The timing of D/ICU was extracted for all episodes. 1100 in-patient episodes were annotated manually to record the timing of CSD, defined as a specific event requiring a change in treatment. Time series features were entered into logistic regression models to derive DEWS for each of the clinical outcomes. Area under the receiver operating characteristic curve (AUROC) was the primary measure of model accuracy. RESULTS: AUROC (95% confidence interval) for predicting D/ICU was 0.857 (0.852-0.862) for NEWS-2 and 0.906 (0.899-0.914) for DEWS in the validation data. AUROC for predicting CSD was 0.829 (0.817-0.842) for NEWS-2 and 0.877 (0.862-0.892) for DEWS. NEWS-2 ≥ 5 had sensitivity of 88.2% and specificity of 54.2% for predicting CSD, while DEWS ≥ 0.021 had higher sensitivity of 93.6% and approximately the same specificity of 54.3% for the same outcome. Using these cut-offs, 315 out of 347 (90.8%) CSD events were detected by both NEWS-2 and DEWS, at the time of the event or within the previous 4 h; 12 (3.5%) were detected by DEWS but not by NEWS-2, while 4 (1.2%) were detected by NEWS-2 but not by DEWS; 16 (4.6%) were not detected by either scoring system. CONCLUSION: We have developed DEWS that display greater accuracy than NEWS-2 for predicting clinical deterioration events in patients with respiratory disease. Prospective validation studies are required to assess whether DEWS can be used to reduce missed deteriorations and false alarms in real-life clinical settings.

Detecting asthma exacerbations using daily home monitoring and machine learning.

OBJECTIVE: Acute exacerbations contribute significantly to the morbidity of asthma. Recent studies have shown that early detection and treatment of asthma exacerbations leads to improved outcomes. We aimed to develop a machine learning algorithm to detect severe asthma exacerbations using easily available daily monitoring data. METHODS: We analyzed daily peak expiratory flow and symptom scores recorded by participants in the SAKURA study (NCT00839800), an international multicentre randomized controlled trial comparing budesonide/formoterol as maintenance and reliever therapy versus budesonide/formoterol maintenance plus terbutaline as reliever, in adults with persistent asthma. The dataset consisted of 728,535 records of daily monitoring data in 2010 patients, with 576 severe exacerbation events. Data post-processing techniques included normalization, standardization, calculation of differences or slopes over time and the use of smoothing filters. Principal components analysis was used to reduce the large number of derived variables to a smaller number of linearly independent components. Logistic regression, decision tree, naïve Bayes, and perceptron algorithms were evaluated. Model accuracy was assessed using stratified cross-validation. The primary outcome was the detection of exacerbations on the same day or up to three days in the future. RESULTS: The best model used logistic regression with input variables derived from post-processed data using principal components analysis. This had an area under the receiver operating characteristic curve of 0.85, with a sensitivity of 90% and specificity of 83% for severe asthma exacerbations. CONCLUSION: Asthma exacerbations may be detected using machine learning algorithms applied to daily self-monitoring of peak expiratory flow and asthma symptoms.

Applications of artificial intelligence and machine learning in respiratory medicine.

The past 5 years have seen an explosion of interest in the use of artificial intelligence (AI) and machine learning techniques in medicine. This has been driven by the development of deep neural networks (DNNs)-complex networks residing in silico but loosely modelled on the human brain-that can process complex input data such as a chest radiograph image and output a classification such as 'normal' or 'abnormal'. DNNs are 'trained' using large banks of images or other input data that have been assigned the correct labels. DNNs have shown the potential to equal or even surpass the accuracy of human experts in pattern recognition tasks such as interpreting medical images or biosignals. Within respiratory medicine, the main applications of AI and machine learning thus far have been the interpretation of thoracic imaging, lung pathology slides and physiological data such as pulmonary function tests. This article surveys progress in this area over the past 5 years, as well as highlighting the current limitations of AI and machine learning and the potential for future developments.

Asthma admission rates and patterns of salbutamol and inhaled corticosteroid prescribing in England from 2013 to 2017.

Asthma exacerbations are a common reason for hospital admission. We sought to identify whether patterns of inhaler prescribing are significantly associated with regional asthma admission rates. Asthma admission rates were obtained for English Clinical Commissioning Group (CCG) regions from 2013/2014 to 2016/2017. Raw prescribing data were obtained from OpenPrescribing.net, based on monthly general practice-level data published by the National Health Service Business Services Authority. Data were analysed using a linear mixed effects model. The ratio of salbutamol to inhaled corticosteroid prescriptions within a CCG was positively associated with asthma admission rates, independently of median age, asthma prevalence and socioeconomic deprivation.

Effects of indoor particulate matter exposure on daily asthma control.

BACKGROUND: Household dust often contains aeroallergens, such as the house dust mite antigen Der p 1. It has been proposed that overnight exposure to particulate matter from bedding and other sources may be an important driver of atopic asthma. Whether variability in overnight particulate matter exposure is a significant determinant of asthma control is unknown. OBJECTIVE: To test the hypothesis that overnight particulate matter exposure is associated with day-to-day symptoms, lung function, and airway inflammation in patients with asthma who are sensitized to house dust mite. METHODS: We undertook a prospective, single-center panel study in 28 adults with asthma and house dust mite sensitization. Overnight exposure to particulate matter was measured using a commercially available indoor air quality monitor. Symptom scores, peak expiratory flow, and exhaled nitric oxide were measured and electronically recorded daily. Participants were followed up for 12 weeks and attended study visits every 4 weeks, at which they underwent spirometry and completed the Asthma Control Questionnaire and Asthma Quality of Life Questionnaire. Data were analyzed using cross-correlation and linear mixed-effects models. RESULTS: No significant associations were observed between overnight particulate matter exposure and clinical outcomes measured daily or at study visits. CONCLUSION: Natural variability in overnight particulate matter exposure does not appear to be a major determinant of daily asthma control in patients with asthma and house dust mite sensitization.

Associations in asthma between quantitative computed tomography and bronchial biopsy-derived airway remodelling.

Airway remodelling in asthma remains poorly understood. This study aimed to determine the association of airway remodelling measured on bronchial biopsies with 1) lung function impairment and 2) thoracic quantitative computed tomography (QCT)-derived morphometry and densitometry measures of proximal airway remodelling and air trapping.Subjects were recruited from a single centre. Bronchial biopsy remodelling features that were the strongest predictors of lung function impairment and QCT-derived proximal airway morphometry and air trapping markers were determined by stepwise multiple regression. The best predictor of air trapping was validated in an independent replication group.Airway smooth muscle % was the only predictor of post-bronchodilator forced expiratory volume in 1 s (FEV1) % pred, while both airway smooth muscle % and vascularity were predictors of FEV1/forced vital capacity. Epithelial thickness and airway smooth muscle % were predictors of mean segmental bronchial luminal area (R2=0.12; p=0.02 and R2=0.12; p=0.015), whereas epithelial thickness was the only predictor of wall area % (R2=0.13; p=0.018). Vascularity was the only significant predictor of air trapping (R2=0.24; p=0.001), which was validated in the replication group (R2=0.19; p=0.031).In asthma, airway smooth muscle content and vascularity were both associated with airflow obstruction. QCT-derived proximal airway morphometry was most strongly associated with epithelial thickness and airway smooth muscle content, whereas air trapping was related to vascularity.

Characterization of acinar airspace involvement in asthmatic patients by using inert gas washout and hyperpolarized (3)helium magnetic resonance.

BACKGROUND: The multiple-breath inert gas washout parameter acinar ventilation heterogeneity (Sacin) is thought to be a marker of acinar airway involvement but has not been validated by using quantitative imaging techniques in asthmatic patients. OBJECTIVE: We aimed to use hyperpolarized (3)He diffusion magnetic resonance at multiple diffusion timescales and quantitative computed tomographic (CT) densitometry to determine the nature of acinar airway involvement in asthmatic patients. METHODS: Thirty-seven patients with asthma and 17 age-matched healthy control subjects underwent spirometry, body plethysmography, multiple-breath inert gas washout (with the tracer gas sulfur hexafluoride), and hyperpolarized (3)He diffusion magnetic resonance. A subset of asthmatic patients (n = 27) underwent quantitative CT densitometry. RESULTS: Ninety-four percent (16/17) of patients with an increased Sacin had Global Initiative for Asthma treatment step 4 to 5 asthma, and 13 of 17 had refractory disease. The apparent diffusion coefficient (ADC) of (3)He at 1 second was significantly higher in patients with Sacin-high asthma compared with that in healthy control subjects (0.024 vs 0.017, P < .05). Sacin correlated strongly with ADCs at 1 second (R = 0.65, P < .001) but weakly with ADCs at 13 ms (R = 0.38, P < .05). ADCs at both 13 ms and 1 second correlated strongly with the mean lung density expiratory/inspiratory ratio, a CT marker of expiratory air trapping (R = 0.77, P < .0001 for ADCs at 13 ms; R = 0.72, P < .001 for ADCs at 1 second). CONCLUSION: Sacin is associated with alterations in long-range diffusion within the acinar airways and gas trapping. The precise anatomic nature and mechanistic role in patients with severe asthma requires further evaluation.

Lung clearance index in adults with non-cystic fibrosis bronchiectasis.

BACKGROUND: Lung clearance index (LCI) is a measure of abnormal ventilation distribution derived from the multiple breath inert gas washout (MBW) technique. We aimed to determine the clinical utility of LCI in non-CF bronchiectasis, and to assess two novel MBW parameters that distinguish between increases in LCI due to specific ventilation inequality (LCIvent) and increased respiratory dead space (LCIds). METHODS: Forty-three patients with non-CF bronchiectasis and 18 healthy control subjects underwent MBW using the sulphur hexafluoride wash-in technique, and data from 40 adults with CF were re-analysed. LCIvent and LCIds were calculated using a theoretical two-compartment lung model, and represent the proportional increase in LCI above its ideal value due to specific ventilation inequality and increased respiratory dead space, respectively. RESULTS: LCI was significantly raised in patients with non-CF bronchiectasis compared to healthy controls (9.99 versus 7.28, p < 0.01), and discriminated well between these two groups (area under receiver operating curve = 0.90, versus 0.83 for forced expiratory volume in one second [% predicted]). LCI, LCIvent and LCIds were repeatable (intraclass correlation coefficient > 0.75), and correlated significantly with measures of spirometric airflow obstruction. CONCLUSION: LCI is repeatable, discriminatory, and is associated with spirometric airflow obstruction in patients with non-CF bronchiectasis. LCIvent and LCIds are a practical and repeatable alternative to phase III slope analysis and may allow a further level of mechanistic information to be extracted from the MBW test in patients with severe ventilation heterogeneity.

Validation of a photoacoustic gas analyser for the measurement of functional residual capacity using multiple-breath inert gas washout.

BACKGROUND: The respiratory mass spectrometer is the current gold-standard technique for performing multiple-breath inert gas washout (MBW), but is expensive and lacks portability. A number of alternative techniques have recently been described. OBJECTIVES: We aimed to validate, using an in vitro lung model, an open-circuit MBW system that utilises a portable photoacoustic gas analyser, with sulphur hexafluoride (SF6) as the inert tracer gas. METHODS: An acrylic glass lung model was utilised to assess the accuracy of functional residual capacity (FRC) measurements derived from MBW. Measurements were performed in triplicate at 20 combinations of simulated FRC, tidal volume and respiratory rate. FRC measured using MBW (FRCmbw) was compared to FRC calculated from the known dimensions of the model (FRCcalc). MBW was also performed in 10 healthy subjects and 14 patients with asthma. RESULTS: The MBW system measured FRC with high precision. The mean bias of FRCmbw with respect to FRCcalc was -0.4% (95% limits of agreement of -4.6 and 3.9%). The mean coefficient of variation of triplicate FRC measurements was 4.0% in vivo and 1.0% in vitro. MBW slightly underestimated low lung volumes and overestimated high lung volumes, but this did not cause a significant error in lung clearance index except at lung volumes below 1,500 ml. CONCLUSIONS: The open-circuit MBW system utilising SF6 as the inert tracer gas and a photoacoustic gas analyser is both accurate and repeatable within the adult range of lung volumes. Further modifications would be required before its use in young children or infants.

Real-world implementation of the National Early Warning Score-2 in an acute respiratory unit.

INTRODUCTION: The National Early Warning Score-2 (NEWS-2) is used to detect deteriorating patients in hospital settings. We aimed to understand how NEWS-2 functions in the real-life setting of an acute respiratory unit. METHODS: Clinical observations data were extracted for adult patients (age ≥18 years), admitted under the care of respiratory medicine services from July to December 2019, who had at least one recorded task relating to clinical deterioration. The timing and nature of urgent out-of-hours medical reviews (escalations) were extracted through manual review of the case notes. RESULTS: The data set comprised 765 admission episodes (48.9% women) with a mean (SD) age of 69.3 (14.8). 8971 out of 35 991 out-of-hours observation sets (24.9%) had a NEWS-2 ≥5, and 586 of these (6.5%) led to an escalation. Out of 687 escalations, 101 (14.7%) were associated with observation sets with NEWS-2<5. Rising oxygen requirement and extreme values of individual observations were associated with an increased risk of escalation. 57.6% of escalations resulted in a change in treatment. Inpatient mortality was higher in patients who were escalated at least once, compared with those who were not escalated. CONCLUSIONS: Most observation sets with NEWS-2 scores ≥5 did not lead to a medical escalation in an acute respiratory setting out-of-hours, but more than half of escalations resulted in a change in treatment. Rising oxygen requirement is a key indicator of respiratory patient acuity which appears to influence the decision to request urgent out-of-hours medical reviews.

Airway impedance entropy and exacerbations in severe asthma.

Variability of peak flow measurements has been related to clinical outcomes in asthma. We hypothesised that the entropy, or information content, of airway impedance over short time scales may predict asthma exacerbation frequency. 66 patients with severe asthma and 30 healthy control subjects underwent impulse oscillometry at baseline and following bronchodilator administration. On each occasion, airway impedance parameters were measured at 0.2-s intervals for 150 s, yielding a time series that was then subjected to sample entropy (SampEn) analysis. Airway impedance and SampEn of impedance were increased in asthmatic patients compared with healthy controls. In a logistic regression model, SampEn of the resistance at 5 Hz minus the resistance at 20 Hz, a marker of the fluctuation of the heterogeneity of airway constriction over time, was the variable most strongly associated with the frequent exacerbation phenotype (OR 3.23 for every 0.1 increase in SampEn). Increased airway impedance and SampEn of impedance are associated with the frequent exacerbation phenotype. Prospective studies are required to assess their predictive value.

Advances in the management of severe asthma.

Severe asthma affects 5 to 10% of the asthma population but consumes a disproportionate amount of the global asthma budget (~50%) due to unscheduled health care utilization in primary care, hospitalizations due to severe exacerbations, and the costs of pharmacotherapy. A key challenge in managing severe asthma is to identify appropriate groups of patients that will respond best to existing and evolving therapies. Recent advances in our understanding of how to classify severe asthma using multivariate taxonomical approaches have provided a unique model of a stratified medicines approach. For example, patients with inflammation-predominant asthma with eosinophils benefit from both inhaled and oral corticosteroids as well as targeted biologics such as anti-interleukin (IL)-5, all of which significantly reduce asthma exacerbations. On the other hand, patients with noneosinophilic neutrophilic asthma may be more suitable for steroid down-titration and therapeutic trials of antineutrophilic agents such as macrolide antibiotics. A similar paradigm can be applied to other domains in severe asthma such as airway hyperresponsiveness, which may now be treated with the first mechanical therapy in airways disease (bronchial thermoplasty). At the same time it is important for the clinician to recognize and treat comorbid factors that make asthma difficult to manage such as poor adherence to medication, rhinosinusitis, and psychological comorbidity. Therefore it is of vital importance to develop a multidisciplinary approach to the management of severe asthma that is best applied within specialist centers with experience and wider access to national and international severe asthma networks.

Predicting outcome in acute respiratory admissions using patterns of National Early Warning Scores.

AIMS: Accurately predicting risk of patient deterioration is vital. Altered physiology in chronic disease affects the prognostic ability of vital signs based early warning score systems. We aimed to assess the potential of early warning score patterns to improve outcome prediction in patients with respiratory disease. METHODS: Patients admitted under respiratory medicine between April 2015 and March 2017 had their National Early Warning Score 2 (NEWS2) calculated retrospectively from vital sign observations. Prediction models (including temporal patterns) were constructed and assessed for ability to predict death within 24 hours using all observations collected not meeting exclusion criteria. The best performing model was tested on a validation cohort of admissions from April 2017 to March 2019. RESULTS: The derivation cohort comprised 7,487 admissions and the validation cohort included 8,739 admissions. Adding the maximum score in the preceding 24 hours to the most recently recorded NEWS2 improved area under the receiver operating characteristic curve for death in 24 hours from 0.888 (95% confidence interval (CI) 0.881-0.895) to 0.902 (95% CI 0.895-0.909) in the overall respiratory population. CONCLUSION: Combining the most recently recorded score and the maximum NEWS2 score from the preceding 24 hours demonstrated greater accuracy than using snapshot NEWS2. This simple inclusion of a scoring pattern should be considered in future iterations of early warning scoring systems.