Critical Care Health Informatics Collaborative

The Critical Care Health Informatics Collaborative (CCHIC) database is an informatics resource for health researchers.

There is currently very little knowledge about the long term outcomes of patients who have been through critical care. Current data is conflicting and there is no national strategy to follow these patients up. Up to this point most outcomes research looks at a single cohort of patients over a defined (and relatively short) period of time. This project aims to create an information technology capability, enabling researchers to investigate the course and outcome of the critically ill. The data, which will be rich both longitudinally and in depth, will enable researchers to answer questions that have been impossible.

The study aims to automatically collect and store routine clinical, demographic and long term outcome data of all patients admitted to participating critical care units. The database will be of interest to Health Services Researchers and Clinical Trial Researchers amongst others.

Linkage of HES and critical care data will enable longer term outcomes for critical care survivors to be tracked.

The broad objectives for this linkage are set out below:

1) Examining long term clinical outcomes. Do survivors of critical illness have significantly reduced life span or increased resource utilisation, compared to the healthy or general hospital population? Surprisingly this is not currently known but this knowledge would allow clinicians and patients alike to make informed care decisions. Equally, ongoing healthcare needs could be predicted and resources appropriately allocated.

2) Looking at predictors of long term outcome. Researchers have previously found a number of variables that may predict patient outcome however, the datasets are frequently limited by the inability to continually re-analyse or examine trends over time. This constantly updating database will enable researchers to analyse how multiple predictive variables interact with each other and over time. The numbers of patients within the database will grow rapidly (The 5 participating NHS trusts admit approximately 10,000 critically ill patients a year) allowing researchers to investigate these predictors in a large population of UK patients over a pro-longed period.

3) The impact of secular or process changes could be measured. The outcome of many Critical Care interventions, whether they are part of research, process change or just over time, are measured in the short term. Updating databases, such as CCHIC, that track longitudinally have the ability to either look for an immediate stepwise change, following implementation (e.g. High Impact Interventions in the Saving Lives campaign) and/or measure that impact over a longer period of time (following HES linkage). The latter is important as the perceived impact is often in reality attributable to secular change, as general levels of care improve rather than a specific intervention

This approach ultimately allows for the use of registry trials of interventions, be it a new process of care, mode of ventilation or the introduction of a new drug. These trials are very efficient (can be run quickly with minimal cost) but until now this approach has been problematic in Critical Care.

4) Clinical Trials: Over the years, approximately 70 drug trials have been performed within critical care, not one has demonstrated a reproducible benefit to the patient. Our understanding of disease has taken huge leaps forward but we seem unable to translate that into patient benefit. The reason for this is multifactorial; however trial design has often been called into question. Outcomes and patient selection is based on small population research and may well not be relevant to the population being studied. Large, all inclusive datasets will allow trialists to examine the course and outcome of their target population and question whether they have chosen the appropriate end points. These end points are often short term survival (usually 28 days, an FDA requirement) which are not highly relevant to the patient. Longer term outcomes could be tracked with HES linkage which would give a far more meaningful outcome thus CCHIC will help design more efficient trials.

This project aims to collect a rich clinical data set during a patient’s admission to a participating critical care unit. This data set (containing identifiable data for linkage) would be encrypted and transferred to the University College London Identifiable Data Handling Solution where it will be pseudonymised on landing and the identifiers split from the clinical data. At regular intervals a data extract from HES will be requested so as to link with the clinical data, thus providing the long term outcomes.

The dataset will be created from merging routine electronic data extracted from several existing databases. These databases are:

1. Critical Care patient information system, ICIP (IntelliVue Clinical Information Portfolio, Phillips). This database integrates data from the hospital Patient Administration System, pathology database, patient monitors and point of care devices. It is also capable of extracting

2. The ICNARC (Intensive Care National Audit and Research Council) pre submission database: Data is routinely collected by NHS Trusts and submitted to ICNARC in order to bench mark and compare performances of critical care units around the country. This database is partly populated by ICIP and through manual entry by a data manager

3. Hospital Episode Statistics. Held by the NHS Digital, this database contains information of hospital admissions

Data flow: -

• Identifiable patient data extracted from NHS Trust clinical systems

• Encrypted and transferred to UCL Safe Haven (legal basis covered by s251)

• Pseudonymised on landing

• Identifiers split from clinical data

• Identifiers sent to NHS Digital to extract requested HES data (Study ID, NHS number, date of birth, sex and postcode)

• HES data linked in UCL Identifiable Data Handling Solution to the clinical data – identifiers removed

• Researcher requests access to data (with appropriate approvals, including information governance concerns addressed, signed data sharing agreement)

• Anonymous, crippled, example dataset of fields requested released to researcher. All data has been randomly adjusted, to give an appearance of what the data will look like

• Researcher works in their own time, with their own tools using a cloned analysis engine (R package or virtual machine)

• Analysis script posted back to safe haven to run on live database

• Analysis results (summary data) returned to researcher

• Primary (patient level) and identifiable data never leaves the Identifiable Data Handling Solution

Data will automatically be extracted to a SQL database within each Trust’s firewall. Integration engine ensure the data is compatible (defined by the dataset) between trusts. Data will then be moved to a SQL database within the University College London IDHS (Identifiable Data Handling Solution) Safe Haven using a secure, encrypted point to point protocol (AS256).

The Safe Haven meets the Information Governance standards required to hold sensitive and identifiable NHS data. It is envisaged that data will be transferred to the UCL Safe Haven every 24 hours. At the point of entry the Research Data Indexing Service within the Safe Haven pseudonymises the data and separates the clinical from the identifier demographics.

All personnel involved with the handling of sensitive and identifiable data will have been trained in information governance. For those involved with handling the data within the participating trusts the training will be provided by the Trusts internal programme. All personnel will comply with local Trust guidelines. Those involved with handling data within the UCL safe haven will have undergone information governance training through the Information Services Division of UCL, it will be a requirement that they comply with all the UCL regulations.

The research team currently consist of investigators from each site involved. Each investigator has a track record in high quality, academic critical care research. The IT support at University College London also has a highly successful research track record and the infrastructure required to support this project. All individuals with access to the record level HES data are substantive employees of UCL.

The Critical Care HIC management team will review all requests to access data. The management team will consist of a representative from each of the five participating NHS Trusts and a member (independent of the investigators) who is skilled in Information Governance and a lay member. This review will look at the both the validity of the research and can the database provide the data required to answer the question and the Information Governance to ensure the requested data subset cannot lead to any patients being identified.

The roll of the Management Committee is to evaluate applications from researchers to access data. The application will be assessed for research validity (by BRC members), information governance and chance of re-identification risk (IG expert) and in the public interest (lay member). Applications will need to show:

1. Submission from bona fide research institutions

2. The data is being acquired solely for valid research, that is non-commercial and intended for patient benefit. It is understood that some of these applications may be from the Bioscience Industry or researchers outside science field.

3. Researchers demonstrate understanding confidentiality and information governance and to abide by the UCL terms and conditions

4. Sign a data sharing agreement that states they will not to try to re-identify the data, will not share the data and will destroy the data within a specified period of time. This data sharing agreement will align with all the data sharing requirements from the HSCIC

5. Have appropriate Research Ethics approval

Any request to the data will need to be from a researcher from a research institute. The research question will need the appropriate HRA ethical approval. The request will need to demonstrate that it is scientifically robust, for healthcare purposes and for public benefit. Providing this is satisfied a ‘dummy’ data set will be released to the researcher. This data set will look very similar to the real data but will not contain any identifiers and the data will be ‘jittered’ (randomly altered). This ‘dummy’ data set will allow the researcher to develop their analysis script, once developed this script will be returned to the UCL Identifiable Data Handling Solution (IDHS) where the IDHS staff will run the script on the live data base. Only aggregate data with small numbers suppressed in line with the HES analysis guide will be released to the researcher.

All researchers will be required to sign an end user license agreement that is modelled on that used by the UK Data service (see https://www.ukdataservice.ac.uk).

The Critical Care HIC Management Team will examine all data access requests. Requests that have the potential for identifying individuals (e.g. narrow date ranges, extremes of age) will be rejected. All data requests will be analysed by the Critical Care HIC management team so as to ensure that small data subsets cannot identify individuals.

UCL will examine data requests classifying data into 4 categories: Direct Identifiers (always removed), Key Variables, Sensitive Fields, Non-identifying variables. The Key Variables and Sensitive Fields could potentially become identifiers in small numbers, UCL employ the concepts of k-anonymity and l-diversity to address this issue. This methodology ensures that data does not contain numbers lower than ten thus minimising the risk of re-identification.

Researchers will only have access to non-identifiable data as described above.

Researchers will come from a range of research institutions:

• Universities

• NHS Institutions

• Charitable Organisations

• Bioscience industries

It is likely interest in data will be from the whole population rather than a specific Trust. However it could be expected that an analysis of geographic variation in practice and outcome would be performed. Currently all Trusts who submit data get an automated data quality report. It is an aim that this data is automatically audited against recognised practice and the results fed back directly (this component of work is underway)

The pharmaceutical, device and diagnostics industry may apply for access to the data as outlined above. UCL believe, that the industry are absolutely vital to progress in this field. The IP developed from the use of the data will remain with the researchers. The access cannot be provided for free as there are significant ongoing costs to hosting and maintaining CCHIC, we would expect such companies to cover these costs. Importantly CCHIC is under the control of the 5 BRCs and is not for profit. All revenue will be re-invested into the project.

UCL are only permitted to approve projects to make use of the data for research where there is a clear benefit to the provision of healthcare or the promotion of health.

As the CCHIC database and the HES linkage should be seen as a core resource for researchers around the UK. The outputs are expected to be broad. However there are specific projects that would use the HES linked data and these include:

Research

A well designed, adaptable and accessible database could help researchers, including the pharmaceutical and medical device industry, to design more accurate, predictable and efficient trials that are relevant to the NHS. This would continue to make the UK an attractive place to run these vital pieces of research. This database will include the clinical, pathological, outcome and demographic data that would allow researchers to construct ‘virtual trials’ examining how altering inclusion/exclusion criteria and endpoints etc. may impact on running the study. The linkage to longer term outcome databases (e.g. HES) will enable more relevant (and patient centred) endpoints to be examined as well as the impact on healthcare resources and the wider health economics. We are using this approach to help identify a patient cohort that may respond to a new treatment for pneumonia. Initial output expected end of 2017, however, longer term outcomes following HES linkage will also be examined and initial output expected end of 2018.

A novel project is also underway to map the new sepsis criteria onto the CCHIC cohort. These results will be submitted for publication. Longer term outcomes of this cohort are unknown, HES linkage will allow us to examine this highly relevant aspect. However, as data is collected prospectively it will take a minimum of a year before any publication becomes relevant.

Audit and Quality Improvement

The data base can be used to audit the performance of individual units in complying with national and international guidelines. For example, we will use the data to examine the ability of ICUs to ventilate their patients within the recognized safe limits. Ventilating above these limits is associated with a poor outcome. There is a complexity in this data that means achieving this locally without specialist data analysts will be difficult, it also enables variation between ICUs to be studied.

It is envisaged that these reports could be automated for participating units. Currently data quality reports are now automated (and returned to the Trusts) as the first step. An initial review of this data (excluding HES) will aim to be submitted for publication by the end of 2017. Linkage to HES will track these outcomes into the future this is important to again analyse whether any differences are sustained

Patient safety: Novel research modelling to identify potential complex signals preceding a clinical deterioration, this could potentially warn clinical staff of impending problems before they become clinically apparent. This approach will then be used to model and predict the longer term outcomes and problems that the HES data will be used for.

CCHIC is being created as a resource for researchers to use. Although the CCHIC team will produce some technical papers around the utility of the database, it is hoped that the majority of the outputs will come for researchers who can use the data.

As with all research UCL would expect the output of the research to be disseminated in the appropriate academic journals and meetings. CCHIC aims to be completely open and transparent. All data releases will be logged on a public facing website. Any coding associated with the database development is freely available on a GitHub repository (no data) and the associated NIHR website is being updated (http://www.hic.nihr.ac.uk/nihr-hic-themes). Any publications stemming from CCHIC will be required to acknowledge the database.

UCL are already presenting the concepts and utility in Critical Care Conferences such as the Intensive Care Society State of the Art meeting and the UK Critical Care Forum. UCL have held ‘datathons’ where jittered and anonymised data can be examined by interested researchers to examine the utility. UCL aim to submit the first paper to a peer reviewed speciality journal such as Critical Care.

Very little is known about the long term outcomes of patients who have been through critical care. Following their recovery, do they have on-going specialist health needs? Despite surviving a critical illness, are they more likely to die earlier than before those who have not? Current data is conflicting and there is no national strategy to follow these patients up. The nature of the combining databases would allow examining the impact that individual, or combinations, of organ failures and severity of illness would have on chronic health. The benefits of this exemplar include:

· Allowing clinicians to make accurate predictions on the course of an illness and the long term outcome

· Allow patients and next of kin to be informed of the condition, empowering them to make choices regarding their treatment with a level of confidence currently unobtainable

· To predict future health care needs and resources

· Education of both patients and healthcare professionals on the sequelae of critical care

· Identify longer term research end points that may be more informative. Cancer treatments often look at 1 and 5 year survivals to measure success or failure of treatment. Critical Care research often looks at day 28 mortality (an FDA requirement for drug trials) or other short term markers, rarely are patients followed up for more than a year that may be relevant for researchers such as 5 year survival.

· The ability to look at regional differences in care and outcome

and endpoints etc. may impact on running the study. Other benefits would include:

· Studying the effect of time on the control group. Clinical trials often run over many years, during this time the introduction of other interventions or processes of care may impact on the trials endpoints, this database could be used to examine this. Providing this data is accurate and up to date, it could conceivably act as a ‘virtual’ control group and allow trials to adapt to secular changes

· Better information would allow more accurate recruitment targets, whilst the database can examine the impact of how recruitment rates may change by:

o altering entry criteria

o altering the time permitted to enrol patients

· Using this database, researchers could look accurately at a variety of different primary or secondary outcomes and endpoints for the trial, enabling each point to be accurately powered.

· Health Service Researchers can use this data to monitor impacts on longer term health care utilisation and long term survival of research interventions

· The potential for an alerting system, such that a recently admitted patient meeting specified criteria could be flagged to the research team. This could give patients the opportunity to take part in clinical trials despite being in a hospital remote to where the research team are based.

· Mapping morbidity relative to disease intervention strategies, thus improving our understanding of disease projection

As an IT resource, rather than a specific research project, UCL expect the outputs and benefits to be broad and ongoing. However, it is envisaged that the benefits will come from several angles. Little is known about the long term survival or outcome of patients who have survived a critical illness. Most previous research follows their acute admission but rarely look beyond hospital discharge. The linking of the proposed databases will bring benefit to several areas that are directly relevant to UK patients.

1) The patient. Long term outcomes and healthcare utilisation is vital if patients, or their proxy, are to make informed decisions over their care and have the appropriate expectations. The only available data is obtained from clinical trials which are not reflective of the general critical care population, often US based and so not very relevant to the UK. As this is about tracking longer term outcomes its value will grow over years. Examples would include:

a. Are those who have survived an episode of severe respiratory failure more likely to be admitted back to hospital with respiratory failure in the future?

b. Are those who require haemodialysis for an episode of acute kidney injury more likely to develop chronic kidney disease?

c. There is an argument that the intense and prolonged inflammation may predispose to cancer and coronary artery disease in the future. Linkage to HES (and disease registries) will help address this argument. It is recognised that this work will take many years to complete.

With all outcome research it is important to try and identify and potentially modify any risk factors?

2) Clinicians and researchers will be able to objectively evaluate the impact of interventions or process change on the longer term health of the critical care survivors. They will also be able to look at potential predictors of outcome enabling them to keep the patient or next of kin fully informed. This will also allow the beginnings of health economic studies to also evaluate these interventions, be it a new drug or a new process

3) Commissioning groups. Knowing these outcomes will enable appropriate planning and anticipation of future healthcare needs.

4) Clinical trialists. All drug trials have failed to produce any reproducible benefit in critical care. The reasons behind this are multi-factorial, but a poor understanding of the patient population is certainly a component. A highly comprehensive database would enable researchers to examine the impact of altering their trial entry criteria and look at different longer term endpoints. This would enable the trial to be appropriately powered and recruitment rates more accurately predicted.

This database will enable:

a) Appropriate powering of studies by examining the properties of the control subjects and how this may change as inclusion/exclusion criteria are altered. This is important, as the nature of the control group appears highly variable between different trials (mortality rate ranging from 15 to 40% within the same therapeutic area). This database will enable trialists to investigate how altering inclusion/exclusion criteria may alter the outcome of their control group and allow appropriate powering of their study.

b) Predict more accurate recruitment rates. The majority of large NIHR sponsored critical care trials have failed to recruit to target; many believe the targets to be unrealistically set. This database will enable accurate modelling of trial recruitment.

c) Critical Care research, unlike cancer research, usually focuses on short term goals such as 28 day mortality or hospital length of stay. This database would enable researchers to examine the impact of interventions on long term outcomes (mortality and health care utilisation).

It could be envisaged that this data will be of value (but not exclusively) to:

• Allow clinicians to make accurate predictions on the course of an illness and the long term outcome.

• Allow patients and next of kin to be informed of the condition, empowering them to make choices regarding their treatment with a level of confidence currently unobtainable.

• Predict future health care needs and resources following critical care discharge.

• Educate both patients and healthcare professionals on the sequelae of critical care.

• Identify longer term research end points that may be more informative. Cancer treatments often look at 1 and 5 year survivals to measure success or failure of treatment. Critical Care research often looks at day 28 mortality (an FDA requirement for drug trials) or other short term markers, rarely are patients followed up for more than a year that may be relevant for researchers such as 5 year survival.

• The ability to look at regional differences in care and outcome.

• Act as a ‘virtual’ control group and allow trials to adapt to secular changes, examine different end-points and inclusion criteria and more accurate recruitment targets.

• Health Service Researchers can use this data to monitor the impact of new changes or interventions longer term health care utilization and long term survival of research interventions.

The nature of the combining databases would allow examining the impact that individual, or combinations, of organ failures and severity of illness would have on chronic health. Data extracted from HES will be linked with clinical data obtained from their critical care admission to track long term mortality and hospital re-admissions.

Patients have not yet been followed up for a sufficient length of time to see any benefit directly from HES lnkage. However the CC-HIC database is beginning to yield benefit. To date, CC_HIC has resulted in 2 published papers, 5 published abstracts, collaboration with 4 major research groups and the employment of 12 researchers (independent to the core group).

It is anticipated that longer term outcome work will be published in 2020.