Optimising Resource Allocation Via Prediction of Outcomes for Covid-19

NCT ID: NCT04473105

Last Updated: 2024-04-09

Basic Information

• Recruitment Status: COMPLETED
• Total Enrollment: 11134 participants
• Study Classification: OBSERVATIONAL
• Study Start Date: 2020-05-01
• Study Completion Date: 2022-03-31

Brief Summary

The investigators plan to use all of the information available within their local NHS hospitals Trust to work out what happens to people admitted with both suspected and proven Covid-19 infections. The investigators will use all of the information that they can to provide the most evidence possible to use in their investigation as this will make the results more accurate. This will include information on existing health conditions (e.g. by looking at previous discharge letters, GP summaries), clinical observations recorded in the hospital (e.g. temperature, blood pressure, pulse, oxygen levels) and laboratory measures (e.g. blood markers of infection). The investigators experienced team will then analyse all of this together with information about whether the person has Covid-19 to help work out what any new patients' risk will be. To do this the investigators need to use individual patients' information, however once removed from the hospital records system it will not be identifiable and will be held securely within the hospital at all times. As a result of this work the investigators plan to be able to do two things:

1. When a patient is admitted to hospital with possible or confirmed Covid-19 the investigators will be able to make a highly accurate prediction of what is likely to happen to them (e.g. being admitted to high dependency or intensive care, dying or surviving to discharge) which will help health care professional make decisions about their care.

2. By knowing what is likely to happen to a patient the investigators are able to make informed decisions about how to distribute healthcare resources e.g. which areas are likely to need more ventilators (machines to help with breathing), need for intensive care beds, discharge planning.

Detailed Description

Background.

Relatively simple clinical risk scores based upon easily available clinical information can greatly aid in the triaging of patients to early discharge or more rapid and intensive intervention. One example of this is in upper gastrointestinal bleeding where several such scores (https://www.mdcalc.com/glasgow-blatchford-bleeding-score-gbs4 5; https://www.mdcalc.com/rockall-score-upper-gi-bleeding-complete6) have permitted safe early discharge thereby relieving pressure on hospitals. The investigators believe that similar results might potentially be achievable from data routinely collected in our trust on patients with proven or suspected Covid-19.

Objectives

The investigators aim to answer the following questions

1. What pattern of clinical history and symptoms, observations, blood and other investigative markers best predicts that a patient suspected of Covid-19 or proven to have it will progress to requiring ventilation?

2. What pattern of clinical history and symptoms, observations, blood and other investigative markers best predicts that a patient suspected of Covid-19 or proven to have it will die during their illness?

3. What pattern of clinical history and symptoms, observations, blood and other investigative markers best predicts that a patient suspected of Covid-19 or proven to have it will make a full recovery without requiring supplemental oxygen?

Methods

The investigators propose to two main approaches to the first two of the aims above.

A standard approach assessing baseline characteristics to predict poor outcome, similar to other ongoing studies like ISARIC and PRIEST, but benefitting from an unbiased sample of patients as no additional samples or data are to be collected and so the investigators believe that it will be ethically acceptable subject to due attention to confidentiality to analyse all patients presenting and not only those willing and able to consent.

A multi-level modelling approach that uses the rich repeated daily laboratory and clinical measurements to predict the daily risk of a subsequent deterioration.

This second approach complements other ongoing studies like ISARIC and PRIEST, as it can only feasibly be delivered within a single institution like NUH as it requires detailed longitudinal linked electronic health data. Furthermore, it is an efficient, low cost study design that does not require manual data collection.

The investigators will utilize the data held in the electronic systems of Nottingham University Hospitals (NUH) NHS Trust, to identify all patients either suspected of having Covid-19 infection or in whom the diagnosis is eventually confirmed. This will be performed from the start of the pandemic retrospectively and then in a rolling program prospectively to maximise available data.

For these patients the investigators will gather all relevant data on diagnoses (comorbidities), daily clinical observations and daily laboratory parameters both at presentation and on a rolling basis going forward from that until the point of death or discharge. These data will include repeated measures of, temperature, pulse, blood oxygen saturation, inspired oxygen concentration, respiratory rate, C reactive protein, and white cell count among others.

Data will be analysed, and outcomes modelled within the Nottingham University Hospitals Trust network to ensure that confidentiality is maintained.

Analysis

Characteristics of patients who go on to intubation The investigators will assess the association of any baseline parameters and demographics of patients who require ventilation compared to those who do not by cross tabulating baseline parameters with ventilation, and calculating adjusted associations using a logistic regression model. The investigators will also examine daily risk of intubation using a random effects model with the repeated daily measurements. In order to examine the potential for an easily calculable clinical risk score, logistic models will be prepared, but retaining continuous variables as continuous (to maximise the retention of useful predictive data) and reducing them to categorical data (for ease of clinical calculation). The investigators will select predictors for the model using backward elimination with the Akaike information criterion and alpha = 0.05. For candidate models the investigators will calculate the C-index and receiver operating curves and assess calibration using the Hosmer-Lemeshow test. Bootstrapping and cross validation will be used to avoid overfitting and assess model performance.

Time trends analysis to identify markers leading to intubation The investigators will examine the data for trends over time in repeated measurements where available and describe these using appropriate summary statistics. If there are enough repeated measurements, then each covariate will be assessed with a JoinPoint analysis to assess if there is an obvious inflexion in time to indicate a change in the clinical picture which marks a decline towards ventilation. This process will feed into the rational choice of categories for categorical representation of the data.

Predictors at intubation suggesting death The investigators will assess the association of any parameters and demographics of ventilated patients who die compared to those who do not by cross tabulating parameters at point of ventilation with death, and calculating adjusted associations using a logistic regression model. The investigators will also examine daily risk of death using a random effects model with the repeated daily measurements. In order to examine the potential for an easily calculable clinical risk score, logistic models will be prepared, but retaining continuous variables as continuous (to maximise the retention of useful predictive data) and reducing them to categorical data (for ease of clinical calculation). The investigators will select predictors for the model using backward elimination with the Akaike information criterion and alpha = 0.05. For candidate models the investigators will calculate the c index and receiver operating curves and assess calibration using the Hosmer-Lemeshow test. Bootstrapping and cross validation will be used to avoid overfitting and assess model performance.

Time trends analysis to identify markers leading to death We will examine the data for trends over time in repeated measurements where available and describe these using appropriate summary statistics. If there are enough repeated measurements, then each covariate will be assessed with a JoinPoint analysis to assess if there is an obvious inflexion in time to indicate a change in the clinical picture which marks a decline towards death Baseline characteristics of patients discharged without supplemental oxygen We will assess the association of any baseline parameters and demographics of patients who are discharged without supplemental oxygen compared to those who are not by cross tabulating baseline parameters with ventilation, and calculating adjusted associations using a logistic regression model. In order to examine the potential for an easily calculable clinical risk score, logistic models will be prepared, but retaining continuous variables as continuous (to maximise the retention of useful predictive data) and reducing them to categorical data (for ease of clinical calculation). The investigators will select predictors for the model using backward elimination with the Akaike information criterion and alpha = 0.05. For candidate models the investigators will calculate the c index and receiver operating curves and assess calibration using the Hosmer-Lemeshow test. Each of these steps will be repeated in 10 randomly selected 20% subsamples of the dataset.

Stratified supplemental analyses To identify different patient groups at risk of poor outcomes, interactions with each risk model will be assessed to identify if stratification is needed by age bands, sex, prior co-morbidities and baseline factors such as lymphopaenia, CRP etc.

All analyses will be repeated for groups of PCR proven Covid-19 patients and those only suspected.

For the analysis of discharge without supplemental oxygen the investigators will examine the risk of readmission for those so discharged and repeat the analysis only for those not subsequently readmitted.

Eligibility Criteria

Inclusion criteria • All patients admitted to Nottingham University Hospitals Trust either suspected of having Covid-19 infection or in whom the diagnosis is eventually confirmed and who are over the age of 18 years

Exclusion criteria

• Aged under 18.

Conditions

COVID19

Study Design

• Observational Model Type: OTHER
• Study Time Perspective: PROSPECTIVE

Eligibility Criteria

Inclusion Criteria

• All patients admitted to Nottingham University Hospitals Trust either suspected of having Covid-19 infection or in whom the diagnosis is eventually confirmed and who are over the age of 18 years

Exclusion Criteria

• Aged under 18.

• Minimum Eligible Age: 18 Years
• Eligible Sex: ALL
• Accepts Healthy Volunteers: No

Sponsors

East Midlands Academic Health Sciences Network

UNKNOWN

Sponsor Role: collaborator

Nottingham University Hospitals NHS Trust

OTHER

Sponsor Role: lead

Responsible Party

Responsibility Role: SPONSOR

Principal Investigators

Timothy R Card, FRCP, PhD

Role: PRINCIPAL_INVESTIGATOR

University of Nottingham

Locations

Nottingham University Hospitals NHS Trust

Nottingham, , United Kingdom

Countries

United Kingdom

Other Identifiers

20GA039

Identifier Type: -

Identifier Source: org_study_id