GSF LITE�IDENTIFY, ENGAGE & MANAGE

Predictive modelling – Discovery Phase

SYSTEM OPPORTUNITY – BENEFITS OF EARLY DETECTION

GRiP Analytics Limited | Confidential between GRiP and PGPA

By focusing on those in greatest need from a 750K population and closing the care gap to ensure people have equitable outcomes we estimate the following system savings could be achieved:

​

• 600+ fewer A&E visits in the last six months of life (with associated costs to the commissioners of £120k)
• 500+ fewer emergency inpatient admissions (with associated costs to the commissioners of £2m)
• 5,000 potential bed days saved
• 100+ people dying in their preferred place of death

PROJECT BRIEF

GRiP Analytics Limited | Confidential between GRiP and PGPA

Data Processing and Analytics Objective

​

What are we aiming for?

​

• To improve the accuracy and reduce variability for identification of our 1% (who will die in the next year) across Calderdale practices

​

• To enable appropriate ACP/RESPECT conversations to take place and have relevant documentation in place to improve quality of care and reduce inappropriate admissions

​

METHODOLOGY

GRiP Analytics Limited | Confidential between GRiP and PGPA

Our approach

​

Using routinely collected data from Central Halifax PCN (no extra collections required) we wanted to learn:

​

• How many people died in calendar year 2023?
• How old were they?
• How many had palliative care codes associated with them?
• How many did we miss that could have been detected prospectively?

​

Through a combination of data engineering, machine learning techniques and clinical engagement we produced a simple risk prediction model that identifies people likely to die in the next twelve months which we validated against historic outcomes for the PCN.

​

This tool was then applied against a recent snapshot (February 2024) of the current registered PCN population which produced a re-identifiable list of people to review who were in the top 1% highest risk category. Where appropriate, these individuals were referred onward for to the most appropriate clinician/ carer for ACP conversations.

​

​

CLINICAL MARKERS

GRiP Analytics Limited | Confidential between GRiP and PGPA

Data Orchestration

​

• There are 55,000+ clinical codes

​

• Representing everything from administrative letters to major diagnoses

​

• Need to structure, organise and aggregate these clinical codes into more manageable markers

​

• We produced a wide range of markers (100+) using a publicly available codesets

​

• Derived from the Cambridge Multimorbidity Score (CMS), QOF and NHS Digital reference sets we have the following clinical markers available for testing (right).

Clinical Marker Examples:

​

• CMS
• Anxiety
• Atrial Fibrillation
• Chronic Liver Disease
• Diabetes T1 & T2
• Epilepsy
• Malignant Melanoma
• Substance misuse
• NHSD
• Acute confusion
• Falls/fractures
• Obese
• Stroke
• Transplant
• QOF
• Chronic kidney disease
• Heart Failure
• Hypertension
• Other
• Polypharmacy

MAKING SENSE OF THE NOISE

GRiP Analytics Limited | Confidential between GRiP and PGPA

Our approach

​

Without processing the data, it can be very difficult to quickly understand someone’s circumstances (below left). Building clinical markers makes it much easier to understand an individual’s clinical position (below right).

74 y/o male – Spring Hall Medical Centre

​

• Asthma
• Dementia
• Hypertension
• Hypothyroidism
• Depression
• Parkinsonism
• Smoker
• Severely frail

​

• Not on palliative care register

FINDINGS – 1% MORTALITY (2023)

• As of 1^st January 2023
• Approximately 55K people registered with Central PCN GP Practices
• Standard population in terms of age and sex distribution (see right)
• We can see below how many died in the following 12 months:

GRiP Analytics Limited | Confidential between GRiP and PGPA

FINDINGS – PALLIATIVE CARE REGISTRATIONS

GRiP Analytics Limited | Confidential between GRiP and PGPA

• 403 people were on the palliative care register as of 1^st January 2023 (0.72%)
• Of these 102 died during the following year (positive predictive value = 25.3%)
• The sensitivity was 18.3% given that 454 people died who were not on the register

PPV = 25.3%

Sensitivity = 18.3%

FINDINGS – PREDICTIVE MODEL PERFORMANCE

GRiP Analytics Limited | Confidential between GRiP and PGPA

Johns Hopkins ACG Solution

​

• As part of the ACG population health management ‘toolkit’ there is the MRS (mortality risk score)
• World class
• International adoption
• Published in academic journals
• Supported and centrally managed
• License fee applicable

​

PPV = 39%

Sensitivity = 29%

PGPA funded bespoke risk model

​

• PGPA sponsored the development of a risk model with Grip Analytics
• Locally owned and managed
• Sensitive to coding practice in PCN
• Explainable
• One off exercise to prove the concept/utility

​

PPV = 40%

Sensitivity = 30%

RESULTS

GRiP Analytics Limited | Confidential between GRiP and PGPA

• Not everyone who is on palliative care is high risk
• Not everyone who is high risk is on palliative care
• This tool is designed to make sure you’re not missing anyone

• 306 people are high risk and are not on the palliative care register
• The Boulevard Medical Practice - 81
• Kings Cross Practice - 54
• Spring Hall Medical Centre - 171

HIGH RISK PERSONA

GRiP Analytics Limited | Confidential between GRiP and PGPA

Mary’s story – 1^st Jan 2023

​

Mary was an 85-year-old lady who was registered with The Boulevard Medical Practice.

​

Mary was a complex, multi-morbid patient who had a range of risk markers that were well documented:

​

• COPD
• T2 Diabetes
• Cancer
• Ischemic Heart Disease
• History of smoking
• History of irregular pulse
• Polypharmacy (10+ unique medications)

​

• The predictive algorithm would have identified Mary as high risk (see her timeline of GP appointments below).

​

​

​

​

​

​

​

​

Mary moved into a care home in April 2023 and later died in September 2023. A quick examination of Mary’s primary care record reveals:

​

• She was not on the Palliative Care register and had never been considered for it
• A ReSPECT form had never been completed
• Mary’s preferred place of death had not been captured

NEXT BEST ACTIONS

GRiP Analytics Limited | Confidential between GRiP and PGPA

• Validation of list and decision of appropriateness and onward referral path for patient.
• Checking using the patient summary view (an output of the tool) and the medical records
• Decide if appropriate for palliative care register – yes/no/not sure
• Status and who referred to eg
• Comments – why above decisions made

​

​

​

​

​

PATIENT SUMMARY VIEW

GRiP Analytics Limited | Confidential between GRiP and PGPA

THANK YOU

GRiP Analytics Limited | Confidential between GRiP and PGPA

Discussion

​

• Thoughts?
• Reflections?
• Suggestions?