Ongoing CRIS projects: psychosis


The COVID-19 pandemic is having a profound effect on people’s lives. Those with a diagnosis of severe mental health problems may be being disproportionately affected. There are many reports of a lack of access to mental health services and resources for people with a diagnosis of severe mental illness at this time Many of the mental health services that continue to operate during the pandemic have had to make considerable changes to the way they are able to provide mental health care, through limiting contact or reducing the amount of activity and treatment they provide. There are also greater social and economic risks for people with a diagnosis of severe mental illness, many of whom are already at high risk of social isolation and exclusion due to their mental illness. Despite this, most of the attention on the impact of COVID-19 on mental health has been on anxiety and depression in the general population.

This study aims to better understand how the COVID pandemic has changed the treatment that people with a diagnosis of severe mental illness receive, and how this might influence their health outcomes including incidents of self-harm, relapse of their illness, and death.

We aim to do this by comparing access to care and health outcomes of people with a diagnosis of severe mental illness during the pandemic in comparison to previous years. We will also look at whether factors such as age, ethnicity, socioeconomic status, the type of illness and any mental or physical comorbidity, or the type of treatment people are receiving influence these differences in access to care and health outcomes during the pandemic.

Our study aims to provide evidence that could support better application of treatments during pandemics in order to protect the health of those with a diagnosis of severe mental illness in order to reduce further health inequalities and associated economic crisis that pandemics may cause in this group of people.

Measuring the fidelity of Early Intervention in Psychosis interventions using Natural Language Processing to improve the prediction of relapse in First Episode Psychosis

When someone starts experiencing a psychotic illness for the first time they will be treated by an Early Intervention in Psychosis (EIP) team. EIP teams try to intervene as soon as people develop signs or symptoms of psychosis. They treat individuals with a combination of medication (if appropriate) and talking therapy, provide advice and education about the illness to both them and their families, and help them manage their physical health and social needs through a case manager. Randomised trials have shown that EIP treatment is the most effective treatment for first episode psychosis and they are now offered throughout England.

The NHS has set out eight key interventions that EIP teams need to deliver to the people they treat to deliver best-evidenced treatment. These are based on the National Institute for Health and Care Excellence (NICE) guidelines for psychosis. However, many EIP teams are unable to deliver these key interventions to their patients, resulting in poorer care. A further problem is that it is difficult to know who is missing out on these treatments, and where in the country it is happening, because many EIP struggle to accurately report this data to the NHS. This is due to the time-consuming nature of collecting this information for each patient from clinical records.

Electronic health records are now the most common method of recording patient clinical data. Most of this information is recorded in what is called ‘free-text’ unstructured notes. This is information entered in a similar way to traditional paper clinical notes, in full sentences, much like a letter, or diary entry. To collect summary data on the NICE interventions given by an EIP service, or all EIP services, someone would have to manually read each individuals’ notes, which is time consuming and not feasible.

In this project we aim to use computer data science to do this data collection automatically. We will do this by using a technique called natural language processing (NLP). NLP is a way to program a computer to process and analyse free-text notes. We will use NLP to identify the eight recommended NICE interventions. We will then test whether the NLP can identify these interventions accurately enough to be confident in using them by comparing the NLP results to results that we have manually collected.

These NLP algorithms can then be used to better understand local and national provision of EIP treatments.

A study estimating long-term rates of relapse, patterns of health outcomes, mental health service use, and moderating factors in patients with psychosis.

In this study, we aim to estimate long term health outcomes & healthcare costs for psychosis patients in CRIS. Using patient characteristics, symptoms and history we will identify the key predictors of health outcomes, such as risk of psychosis relapse. We will also examine the use of mental health services by psychosis patients. The use of services will be costed and we will identify the main predictors of costs. This data will be used to build an economic model to estimate the value for money of new psychosis treatments. The resulting model will help policy makers determine which treatments to implement in the NHS. As a case study, we will apply the model to examine the long-term value of a new virtual reality therapy. The effectiveness of this therapy is currently being evaluated in a 6-month clinical trial.

End-to-end NLP and supervised learning for patient triage

People needing treatment for mental health problems are often first seen by their general practitioners, in hospital Emergency Departments or in educational settings and staff then refer them to mental health services. This network of referral routes converges on community mental health teams who make decisions about suitability for treatment, how urgently to see people and which specialist teams need to be involved. In routine care, all referrals are assessed or ‘triaged’ by a multi-disciplinary team and often require the person is clinically reviewed over one or more contacts to eventually decide on treatment pathways. In 2019 this process cost the NHS £326 million with the referral and triage system often introducing delays to care and this is a common source of frustration for patients and carers.

Electronic health records contain data from clinical encounters between patients and clinicians, over multiple visits and create a trajectory of the patient’s health and outcomes over time.  However, this information ranges from voluminous or very sparse and triage is time consuming even by expert clinicians. Recent advances in artificial intelligence and natural language processing can identity patterns in a patient’s history and their referral documentation which could help make triage more efficient and expedite access to the “right team”.

Aims: We will build clinical-decision tools to support clinicians in making decisions and triaging referrals in routine clinical practice. The tools will identify and improve access to care for people who have traditionally been excluded from, under represented or poorly served by mental health services ensuring they receive equitable care.


Community mental health teams called ‘Early Intervention in Psychosis’ (EIP) services assess and support people with the first signs of psychosis, a group of symptoms that can affect how you feel, think or act. The significant majority of people that suffer with this type of illness never act violently. However, there are certain ‘risk factors’ that might increase the chances of this happening. One thing EIP services have to routinely assess therefore is whether a patient might, due to having some of these risk factors, have any clinical needs related to violence. Importantly this can guide what help someone might need to reduce the risk.

A simple tool called ‘OxMIV’ could support clinicians in EIP services to do this routine assessment of risk in a more structured and consistent manner. OxMIV uses information that is already routinely collected during full clinical assessments, such as about an individual’s past history including of any violent convictions, family history, what treatment they have received and whether there are factors present such as difficulties with drug or alcohol use. Although clinicians already typically gather this information when they get to know someone, having a structured tool can help them be more consistent and confident that they are covering the important information every time. Some of these factors increase risk more than others, so OxMIV also then combines the information in a similar way to how a GP might use a tool to support them thinking about somebody’s risk of having a heart attack or stroke, and whether for example someone might need some support with diet and exercise to reduce this risk. So in a similar way OxMIV can support clinicians to more consistently think about whether there are any clinical needs around reducing violence risk that they can support with. The tool is very quick and simple, and is designed to support the full clinical assessments that clinicians are already doing, without replacing this full assessment or adding to the overall workload. This project is to help look at whether some preliminary local work with clinicians using the tool (which is available alongside the usual risk assessment forms on the standard electronic health record as a form with drop-down boxes to complete the assessment) has helped with how risk is considered and documented.

How do you rate this page?

Thank you for your feedback

Follow us on social media to stay up to date

We are sorry you did not find this page helpful

Tell us how we can improve this page

Page last reviewed: 4 July, 2022