Chart review of patients with chronic obstructive pulmonary disease, using medical records and artificial intelligence
| ISRCTN | ISRCTN32473131 |
|---|---|
| DOI | https://doi.org/10.1186/ISRCTN32473131 |
| Secondary identifying numbers | BigCOPData |
- Submission date
- 25/11/2019
- Registration date
- 24/01/2020
- Last edited
- 24/01/2020
- Recruitment status
- No longer recruiting
- Overall study status
- Completed
- Condition category
- Respiratory
Plain English summary of protocol
Background and study aims
Chronic obstructive pulmonary disease (COPD) is the third leading cause of death in the world since 2003. Many people suffer from this disease or its complications for many years and die prematurely. In the European Union, the total direct costs of respiratory diseases are estimated to be around 6% of the total healthcare budget, with COPD accounting for 56% (38.6 billion Euros) of the costs of respiratory diseases.
In the natural history of COPD, many patients may experience acute exacerbations (AECOPD) that are described as episodes of sustained worsening of the respiratory symptoms that result in additional therapy. These episodes of exacerbation that often require being seen in the emergency department and/or a hospital admission are associated with significant morbidity and mortality; they are responsible for a significant portion of the economic burden of the disease too. The pharmacological approach used in the management of AECOPD (inhaled bronchodilators, corticosteroids and antibiotics), has the objective to minimize the negative impact of the current exacerbation and to prevent subsequent events. Despite the collaborative effort between the European Respiratory Society, the American Thoracic Society and others to provide clinical recommendations for the prevention of AECOPD, there is still a considerable number of patients that are prone to suffer from recurrent exacerbations and to experience a more severe impairment in health status. Based on all the above, the aim of this study is to identify the factors potentially associated with hospital admission in patients with AECOPD in English-, French-, German-, and Spanish-speaking countries, and to develop a predictive model that predicts the risk of hospitalization in this group of patients, by using artificial intelligence. In this study the researchers propose to take advantage of SAVANA, a new clinical platform, created in the context of the era of electronic medical records (EMRs), to analyse the information included in the electronic medical files (i.e., big data). This clinical platform is a powerful free-text analysis engine, capable of meaningfully interpreting the contents of the EMRs, regardless of the management system in which they operate. In this context, this machine learning analytical method can be used to build a flexible, customized and automated predictive model using the information available in EMRs.
Who can participate?
Adults both genders with Chronic obstructive pulmonary disease
What does the study involve?
For patients there is no intervention, as the data is extracted from their electronic medical records.
What are the possible benefits and risks of participating?
The benefits is to generate an automated predictive model with the use of machine learning that predicts the risk of hospitalization in patients with AECOPD.
Where is the study run from?
In 80 sites distributed in English, French, German and Spanish speaking countries (UK, Canada, USA, France, Belgium, Switzerland, Germany, Austria, Spain)
When is the study starting and how long is it expected to run for?
April 2019 to December 2020
Who is funding the study?
European Commission with a grant Horizon 2020 on research and innovation, Brussels, Belgium
Who is the main contact?
Prof. Rob Stockley
rob.stockley@uhb.nhs.uk
Contact information
Scientific
Queen Elizabeth Hospital
Mindelsohn Way
Edgbaston
Birmingham
B9 5SS
United Kingdom
| Phone | +44 (0)121 3716808 |
|---|---|
| rob.stockley@uhb.nhs.uk |
Study information
| Study design | Data-driven observational retrospective and non-interventional study using secondary data captured in EMRs |
|---|---|
| Primary study design | Observational |
| Secondary study design | Retrospective study |
| Study setting(s) | Hospital |
| Study type | Prevention |
| Scientific title | Chart review of patients with COPD, using medical records and artificial intelligence |
| Study acronym | BigCOPData |
| Study objectives | Chronic obstructive pulmonary disease (COPD) is the third leading cause of death in the World since 2003. Many people suffer from this disease or its complications for many years and die prematurely. In the European Union, the total direct costs of respiratory diseases are estimated to be around 6% of the total healthcare budget, with COPD accounting for 56% (38.6 billion Euros) of the costs of respiratory diseases. In the natural history of COPD, many patients may experience acute exacerbations (AECOPD) that are described as episodes of sustained worsening of the respiratory symptoms that result in additional therapy. These episodes of exacerbation that often require been seen in the emergency department and/or a hospital admission are associated with significant morbidity and mortality; they are responsible for a significant portion of the economic burden of the disease too. The pharmacological approach used in the management of AECOPD (inhaled bronchodilators, corticosteroids, and antibiotics), has the objective to minimize the negative impact of the current exacerbation and to prevent subsequent events. Despite the collaborative effort between the European Respiratory Society, the American Thoracic Society, and others to provide clinical recommendations for the prevention of AECOPD, there is still a considerable number of patients that are prone to suffer from recurrent exacerbations and to experience a more severe impairment in health status. Based on all the above, we aim to identify the factors potentially associated with hospital admission in patients with AECOPD in English, French, German, and Spanish, speaking countries, and to develop a predictive model that predicts the risk of hospitalization in this group of patients, by using artificial intelligence. In this study we propose to take advantage of SAVANA, a new clinical platform, created in the context of the era of electronic medical records (EMRs), to analyse the information included in the electronic medical files (i.e., big data). This clinical platform is a powerful free-text analysis engine, capable of meaningfully interpreting the contents of the EMRs, regardless of the management system in which they operate. In this context, this machine learning analytical method can be used to build a flexible, customized and automated predictive model using the information available in EMRs. Primary objective: To identify factors associated with hospital admission in a population of patients hospitalized for an exacerbation of COPD, and to develop a predictive hospital admission model, using EMRs and artificial intelligence Secondary objectives: 1. To describe the clinical characteristics of COPD patients that require hospital admission 2. To identify the comorbidities associated with hospitalized COPD patients, presented per sex (cardiovascular disease, anxiety, depression, gastroesophageal reflux, etc) 3. To identify and characterise the hospitalizations associated with increased eosinophil blood counts 4. To explore the relationship between hospitalization and inflammatory parameters such as white cell counts, neutrophil count, C-reactive protein (CRP), etc 5. To identify the clinical phenotype of patients with COPD that exacerbate and require hospital admissions 6. To explore the relationship between low adherence to treatment recommendations and hospital admission 7. To determine whether there is a relationship between hospitalization and a change of treatment in the previous 6 weeks 8. To assess stratification risk of patients, using a baseline variable (GesEPOC, the Dyspnoea, Eosinopenia, Consolidation, Acidemia and Atrial Fibrillation [DECAF] Score, or another multicomponent index) 9. To explore whether there are biologic biomarkers (different to eosinophil count) that might predict hospitalization and/or rehospitalizations due to COPD exacerbations |
| Ethics approval(s) | Approved 11/04/2019, Drug Research Ethics Committee of the Princess University Hospital (CEIm La Princesa University Hospital, 62, Diego de León Street, 28006. Madrid, Spain; Tel: +34 (0)91 520 24 76; Email: ceim.hlpr@salud.madrid.org), CEIm Act 07/19 |
| Health condition(s) or problem(s) studied | Chronic obstructive pulmonary disease |
| Intervention | The study is retrospective, non-interventional. It’s expected to collect data from the last 5 years. The study population comprises patients who were admitted in their respective medical centres involved in the study. The methodology data analysis is as follows: Frequency tables will be performed for categorical variables, whereas continuous variables will be described by means of summary tables that may include the mean, standard deviation, median and range of each variable. The number of non-evaluable outcomes and of missing data will also be provided and will not be counted in the percentages. Transformations will be considered where appropriate. Unless otherwise specified, all statistical inference will be performed at the 5% significance level using 2-sided tests or 2 sided CIs. Missing data mechanisms will be evaluated to determine appropriate methods for handling missing data when necessary (e.g. multiple imputation). A comprehensive description of the imputation procedure to ensure the transparency and reproducibility of the analysis will be provided. This is a descriptive and hypothesis-generating study, not a confirmatory one. Therefore, other statistical models can be applied if necessary. A sensitivity analysis will be performed to deal with outliers, should it be necessary. The last phase of the study will build a predictive model to identify those factors associated with hospital admission in a population of patients hospitalized for an exacerbation of COPD. In order to do this, the study will rely on big-data techniques that will combine advanced statistics and machine learning tools in the deep-learning spectrum. The performance of these models will be assessed in terms of precision, recall and F-score, as well as the Area Under Curve (AUC) in some cases. |
| Intervention type | Other |
| Primary outcome measure | Given that this is a Big Data-based study, the potential number of variables that may be included is only limited to the information contained in the EMRs. All mentioned variables will be included if they are found correctly in the text. It is therefore understood that it is impossible to guarantee that all the desired variables will be included in the final study. On the other hand, this technology enables to create new variables, which can neither be described in advance. The following variables will be extracted to meet the objectives of the study: 1. Age 2. Sex 3. Smoking status: current smoker, ex-smoker 3.1. Use of E-cigarettes, iQOS 3.2. Pack-years index 4. History of alcohol and/or drug abuse 5. Exacerbation history: number of exacerbations in the previous 12 months 6. Previous hospital admissions 7. Symptoms on admission: dyspnoea, cough, sputum, chest tightness, or wheezing 8. Clinical phenotypes 8.1. Chronic bronchitis 8.2. Emphysema 8.3. Bronchiectasis 8.4. Asthma-COPD overlap (ACO) 8.5.Frequent exacerbator 9. Pre-existing asthma 10. GOLD stage 11. Airflow obstruction 11.1. FVC 11.2. FEV1 11.3. FEV1/FVC ratio 12. mMRC dyspnea grade, if available 13. COPD Assessment Test 14. Influenza vaccination in the previous year 15. Previous pneumococcal vaccination 16. Previous microbiological isolation in sputum 17. Home oxygen therapy 18. Non-invasive mechanical ventilation (at home) 19. Mechanical ventilation (invasive and/or non-invasive) during hospital stay 20. Medication upon hospital admission, during hospitalization and hospital discharge 20.1. Inhaled corticosteroids (ICS) + LABA + LAMA 20.2. LABA + LAMA 20.3. LABA + ICS 20.4. LAMA + ICS 20.5. LAMA 20.6. LABA 20.7. ICS 20.8. Theophylline 20.9. Roflumilast 20.10. SABA / SAMA 20.11. Systemic corticosteroids 20.12. Mucolytics 20.13. Macrolides 21. Dose of systemic corticosteroids administered during hospital stay 22. Nebulized antibiotic therapy 23. Number of COPD exacerbations requiring hospitalization in the previous 12 months. 24. Number of COPD exacerbations requiring ER visits in the previous 12 months 26. Number of COPD exacerbations seen in Primary Care in the previous 12 months. 27. Blood test at hospitalization admission and sequentially during hospitalization: 27.1. Leucocytes 27.2. Neutrophils (absolute value and %) 27.3. Eosinophils (absolute value and %) 27.4. Basophils (absolute number and %) 27.5. Platelets 27.6. Haemoglobin 27.7. Fibrinogen 27.8. Urea 27.9. CRP 27.10. D-dimer 27.11. Pro-BNP-NT 27.12. Troponin 27.13. Alpha-1 antitrypsin 28. COPD-specific comorbidity test (COTE) 29. DECAF score 30. Associated comorbidities: hypertension, gastroesophageal reflux, diabetes mellitus, CV disease, skeletal muscle dysfunction, metabolic syndrome, osteoporosis, depression, anxiety and lung cancer, and other 31. Blood gas analysis, partial pressure of oxygen in arterial blood (PaO2) at hospital admission and sequentially during hospitalization, partial pressure of carbon dioxide in arterial blood (PaCO2), pH, etc. 32. Length of hospital stay (days) 33. Ward location at hospital: respiratory unit, internal medicine unit, intensive care unit, etc. 34. Discharge location: home, home health care, nursing home, rehabilitation center, short-term hospital, other 35. Mortality during index admission 36. Hospital readmission within 30- and 90-days post-discharge A complete and detailed guidance on the evaluation of the variables and outcomes are presented in the SAP. |
| Secondary outcome measures | There are no secondary outcome measures |
| Overall study start date | 24/04/2019 |
| Completion date | 31/12/2020 |
Eligibility
| Participant type(s) | Patient |
|---|---|
| Age group | Adult |
| Sex | Both |
| Target number of participants | 2,500,000 patients approx |
| Key inclusion criteria | 1. Subjects aged ≥ 35 years old, smokers or former smokers of more than 10 pack-years 2. Had a diagnosis of COPD (a post-bronchodilator ratio forced expiratory volume in the first second [FEV1] / forced vital capacity [FVC] < 0.70, and the presence of respiratory symptoms such as cough, sputum, and dyspnoea) 3. Admitted for ‘‘respiratory disease’’ [respiratory infection or pleural effusion (OR) respiratory failure (OR) right/left heart failure (OR) chronic bronchitis (OR) bronchospasms (AND) [historical diagnosis of COPD (OR) a documented FEV1/FVC < 0.70 in the absence of other obstructive diseases, such as asthma or bronchiolitis] |
| Key exclusion criteria | Patients with a specific diagnosis upon admission of pulmonary oedema, pneumonia, radiological infiltration, pulmonary embolism, pneumothorax, rib fractures, aspiration, or any other associated respiratory or of non-respiratory condition, such as major cardiopathy with chronic heart failure, extended neoplasia, liver or kidney failure. |
| Date of first enrolment | 01/07/2019 |
| Date of final enrolment | 30/09/2020 |
Locations
Countries of recruitment
- Austria
- Belgium
- England
- France
- Germany
- Luxembourg
- Spain
- Switzerland
- United Kingdom
Study participating centre
Edgbaston
Birmingham
B15 2GW
United Kingdom
Sponsor information
Other
108, Provença Street, Bajos 2
Barcelona
08029
Spain
| Phone | +34 (0)934878565 |
|---|---|
| lcampos@separ.es | |
| Website | https://www.separ.es |
Funders
Funder type
Government
Government organisation / National government
- Alternative name(s)
- EU Framework Programme for Research and Innovation, Horizon 2020 - Research and Innovation Framework Programme, European Union Framework Programme for Research and Innovation
Results and Publications
| Intention to publish date | 01/01/2021 |
|---|---|
| Individual participant data (IPD) Intention to share | Yes |
| IPD sharing plan summary | Other |
| Publication and dissemination plan | Final results of the study will be disseminated in the form of a manuscript/s in the peer-reviewed literature. In addition, where relevant, data from potential interim analyses will be presented at (a) relevant congress(es). |
| IPD sharing plan | The datasets generated and/or analysed during the current study during this study will be included in the subsequent results publication. |
Editorial Notes
10/12/2019: Trial's existence confirmed by Ethics Committee for Drug Research of the Hospital Universitario de la Princesa.
