Professor Julia Hippisley-Cox Professor of Clinical Epidemiology Director ClinRisk Ltd Director QResearch @juliahcox Co-authors QResearch database - EMIS practices, EMIS, Nottingham University EMIS NUG (including screencasts) ClinRisk Ltd (development & software) Office National Statistics (mortality data) HSCIC (pseudonymised HES data) QBleed Algorithm QBleed + QStroke Update on tools integrated into EMIS Web Embargoed until publication Individual assessment Who is most at risk of current or preventable disease? Who is likely to benefit from interventions? What is the balance of risks and benefits for my patient? Enable informed consent and shared decisions Population risk stratification Identification of rank ordered list of patients for recall or reassurance GP systems integration Allow updates tool over time, audit of impact on services and outcomes Anticoagulants used in prevention & treatment of VTE To reduce risk ischaemic stroke with AF Although use of anticoagulants in AF is in QOF uptake is low Legitimate concerns around safety particularly risk of major bleeds Need to quantify absolute risk of bleed to help make informed decision on risk/benefit When discussing benefits and risk of anticoagulation in AF explain that For most people benefit exceeds risk Except for those with increased bleeding risk where careful monitoring required Discuss options and base choice on their clinical features and preferences Only treat after informed discussion on risks & benefits http://www.nice.org.uk/guidance/cg180/chapter/1-recommendations Currently recommends HAS-BLED score Scoring system major bleed in AF Derived from 3978 hospital based patients Not externally validated Risk factors for HAS-BLED v similar to CHADS stroke Simple scoring system not measure of absolute risk Hypertension Renal disease Liver disease Prior Stroke Prior bleed or predisposition Age 65 (yes/no) Medication (antiplatelets, NSAID) Alcohol (> 8 units/week) Labile INR – but supposed to be for new users so INR wont be available! Embargoed until publication Develop new risk algorithm which Predict 1yr & 5yr absolute risk of GI and intracranial bleed new users anticoagulants c.f. non-use Includes clinically relevant variables ameliorable to change Can be implemented in routine GP systems Can be shared with patient to help inform decision making Can be updated regularly Embargoed until publication Developed using QResearch database Very large validated GP database Derived from EMIS (largest GP supplier) Representative ethnically diverse population Linked to Hospital Episode Statistics Linked to ONS cause of death data Embargoed until publication Design: Cohort study Study period: 2008-2013 Patients: 4.4 million aged 21-99 years Baseline: assessment of predictive factors focused on clinically relevant variables primary care Outcome: GI bleed or intracranial bleed on linked mortality or hospital data Embargoed until publication Intracranial bleed 9,040 cases on QResearch linked hospital or mortality records Upper GI bleed 21,614 cases on QResearch linked hospital or mortality records Largest ever such study. Increases reliability of results and generalisability of findings Age, sex, BMI Ethnicity Deprivation Smoking & alcohol Abnormal platelets Medication Antiplatelets NSAIDS Steroids Antidepressants Anticonvulsants Atrial fibrillation Heart Failure Treated hypertension Cancer Liver disease/pancreatitis Oesophageal varices VTE Prior bleed (GI, brain, haematuria,haemoptysis) Gold standard to test performance of risk tool on separate population We used 2 validation samples Different practices in QResearch (from EMIS) Different practices in CPRD (from Vision Practices) Women Men Higher values indicates better discrimination Upper GI bleed ROC 0.77 0.75 R2 40.7 36.9 D statistic 1.7 1.57 0.85 0.81 R2 58 53.3 D statistic 2.4 2.2 Intracranial bleed ROC Similar results CPRD and QResearch Fig 3 Mean predicted risks and observed risks at five years by 10th of predicted risk applying QBleed risk prediction scores to all patients in QResearch validation cohort. Hippisley-Cox J , and Coupland C BMJ 2014;349:bmj.g4606 ©2014 by British Medical Journal Publishing Group Cut off 5 year risk (%) Sensitivity (%) Observed risk (%) Upper GI bleed 1.4% 38% 2.7% Intracranial bleed 0.7% 51% 1.5% For example, using threshold of top 10% at risk will correctly identify 38% of those who get upper GI bleed 51% of those who get intracranial bleed QBLEED 4.4 million GP patients 30,681 events 2 clear outcomes Followed over 5 years Absolute risk Includes more clinically relevant factors Externally validated Easy to update over time HAS-BLED 4,000 hospital patients 53 events Unclear what ‘major bleed’ is Followed over 1 year Simple count only Includes INR which wont have prior to Rx Not externally validated Unclear about updates “Such a model represents a change in our approach to assessing bleeding risk, from simple, point based scores, to a more inclusive, complex model”. “While there may be implications for implementation, this progression may make sense clinically—there are often patient subtleties and characteristics that inevitably increase the risk of bleeding but are not captured in simpler scores”. “While calculating bleeding risk is no longer “simple,” neither is the decision to use long term anticoagulation”. “This is among the largest of the outpatient derivation cohorts used in this specialty to date and provides extra power to develop more robust predictive models using more candidate covariates than other scores”. “A more comprehensive model may adjust for these factors, giving doctors and their patients a more refined estimate of absolute risk”. How should GPs use risk estimates when making decisions about bleeding? What risk is too high? Is threshold same for every patient & every indication? Are there patients for whom extra risk is negligible compared with underling stroke risk? Estimates risk of ischaemic stroke over 1-10 years Includes age, sex, ethnicity, deprivation Smoking, diabetes, AF, CCF, CVD Rheumatoid, chronic renal disease Valvular heart disease Treated hypertension and FH CHD SBP, cholesterol, BMI Integrated into EMIS WEB Embargoed until publication http://qbleed.org/plus-qstroke 75yr old man with AF, light smoker, heavy alcohol, NSAIDS ALREADY IN EMIS WEB QRISk2 QDiabetes QStroke QFracture (QAdmissions) IN PLANNING PHASES QCancer (release 4.11) QKidney QThrombosis QBleed QIntervention http://bmjopen.bmj.com/content/4/8/e005809.abstract EMIS NUG screen casts courtesy of Dr Geoff Schrecker & EMIS NUG http://emisnug.org.uk/video/addingcalculation-template-emis-web http://emisnug.org.uk/video/runningcalculation-eg-qrisk-group-patients-batchadd http://www.emisnug.org.uk/ Currently around 800 practices contributing Would like around 1000 Pseudonymised data with no strong identifiers IG approved EMIS NUG, REC, BMA, RCGP Only used for research All research peer reviewed and published Need to activate QResearch in EMIS Web even if sharing data for many years via LV www.qresearch.org http://emisnug.org.uk/video/add-rbacactivity-emis-web-user-profile http://emisnug.org.uk/video/enablingsharing-agreements-qsurveillance-andqresearch