1. + Qcancer: symptom based approach to early diagnosis of cancer Julia Hippisley-Cox, GP, Professor Epidemiology & Director ClinRisk Ltd National Institute of Clinical Excellence 12 th December 2012

2. Acknowledgements Co-authors QResearch database EMIS & contributing practices & User Group University of Nottingham ClinRisk (software) Oxford University (independent validation) Macmillan Cancer Support (funding implementation) Copyright ClinRisk Ltd (2013) all rights reserved

3. Overview of talk on QCancer QResearch database Background on QScores QCancer Model development QCancer Model validation QCancer Implementation Pilot work with Macmillan and CRUK evaluation Copyright ClinRisk Ltd (2013) all rights reserved

4. QResearch Database www.qresearch.org www.qresearch.org Over 700 general practices across the UK, 14 million patients Not-for-profit venture between EMIS & Nottingham University EMIS covers 56% of UK GP practices Patient level pseudonymised database for research Data linkage – deaths, deprivation, cancer, HES Available for peer reviewed academic research where outputs made publically available Copyright ClinRisk Ltd (2013) all rights reserved

5. QScores –new family of Risk Prediction tools 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 Availability  All published, publically available as free open source or professionally supported closed source software Copyright ClinRisk Ltd (2013) all rights reserved

6. Qscores and national guidance Risk scoreOutcomeNICE guidanceImplementation Qrisk.org10 year risk of CVD Approved CG68 (2008) QOF & DH vascular screening program All major GP suppliers Qdiabetes.org10 year risk diabetes Approved PH38 (2012)EMIS (largest supplier > 55%) Qfracture.org10yr risk of fracture Approved CG146 (2012) QOF 2013 EMIS 2013 + London CCGs Qthrombosis.orgRisk of VTERelevant to CG92EMIS 2013 Qcancer.orgCurrent cancer risk Relevant to current review Pilot BMJ informatica. Later all suppliers Copyright ClinRisk Ltd (2013) all rights reserved

7. Early diagnosis of cancer: The problem UK has relatively poor track record when compared with other European countries Partly due to late diagnosis with estimated 7,500+ lives lost annually Later diagnosis due to mixture of late presentation by patient (alack awareness) Late recognition by GP Delays in secondary care Copyright ClinRisk Ltd (2013) all rights reserved

8. Why symptoms based approach? Many patients present with symptoms GPs need to decide which patients to investigate and refer Decision support tool must mirror setting where decisions made Symptoms based approach needed (rather than cancer based) Must account for multiple symptoms Must have face clinical validity eg adjust for age, sex, smoking, Family history Need to be able to update to meet changing requirements, populations, recorded data Copyright ClinRisk Ltd (2013) all rights reserved

9. QCancer– what it needs to do Accurately predict individual level of risk of multiple cancers for based on multiple risk factors and multiple symptoms Discriminate between patients with and without cancer Help guide decision on who to investigate or refer and degree of urgency. Educational tool for sharing information with patient. Sometimes will be reassurance. Latest QCancer combined model published online BJGP 12 Dec 2012 and in paper version 01 Jan 2013 Copyright ClinRisk Ltd (2013) all rights reserved

10. Methods – development algorithm Representative cohort from QResearch 2.5 million men & women aged 25-89 years cancer outcome - all new diagnoses on GP record or linked deaths record in 2 years Identify key symptoms Identify key risk factors Established methods to develop risk prediction algorithm Measure of absolute risk of any cancer as well as by cancer type Copyright ClinRisk Ltd (2013) all rights reserved

11. Qcancer (2013) predicts global cancer risk & risk12 types cancers PancreasLung Renal tractOvary Colorectal Gastro Testis Breast Prostate Blood Cervix Uterus These accounts for 85% cancers Plan to extend to rarer cancers when sufficient data Copyright ClinRisk Ltd (2013) all rights reserved

12. Independent risk factors in model Age & sex Smoking status Deprivation score Family history of cancer COPD Endometrial hyperplasia/polyp Chronic pancreatitis Type 2 diabetes Anaemia (HB < 11g/DL) Venous thromboembolism Copyright ClinRisk Ltd (2013) all rights reserved

13. Key symptoms in model (identified from studies including NICE guidelines 2005) Haemoptysis Haematemesis Haematuria Rectal bleeding Haematuria Unexplained bruising Constipation, cough Vaginal bleeding (women) Testicular lump (men) Loss of appetite Unintentional weight loss Indigestion +/- heart burn Dysphagia Abdominal pain or swelling Breast lump, pain, skin Night sweats Neck lump Urinary symptoms (men) Copyright ClinRisk Ltd (2013) all rights reserved

14. QCancer symptoms map women bloodbreastcervixCRCOGlungotherovarypancreasrenaluterineTotal Abdo distension+++3 Abdo pain+++++++++9 Anaemia++++++++8 Appetite loss+++++++7 Breast lump++2 Breast pain+1 Breast skin/nipple changes +1 Bruising+1 Change in bowel++++4 Constipation++2 Cough+1 Dysphagia++++4 Haematemesis+++3 Haematuria++++++6 Haemoptysis+1 Heartburn+1 Indigestion++++++6 IMB ++2 Neck lump+++3 Night sweats+1 PCB+1 PMB+++++++7 Rectal bleed+1 VTE++++++++++10 Weight loss++++++++8 Total105799940875n/a

15. bloodCRCgastrolungotherpancreasprostaterenaltestistotal Abdo distension+++3 Abdo pain++++++++8 Anaemia+++++5 Appetite loss+++++++7 Change in bowel++2 Constipation+++3 Cough+1 Dysphagia+++++5 Frequency+1 Haematemesis+++3 Haematuria++++4 Haemoptysis+++3 Heartburn+1 Impotence+1 Indigestion+++++5 Neck lump++++4 Night sweats+++3 Nocturia+1 Rectal bleed++2 Retention+1 Testicular lump+++3 Testicular pain++2 VTE++++++6 Weight loss++++++++8 Grand Total138911148124382 QCancer symptoms map men

16. Methods - validation is crucial Essential to demonstrate the tools work and identify right people in an efficient manner Tested performance separate sample of QResearch practices external dataset (Vision practices) at Oxford University Measures of discrimination - identifying those who do and don’t have cancer Measures of calibration - closeness of predicted risk to observed risk Measure performance – positive predictive value, negative predictive value, sensitivity, specificity at different thresholds Copyright ClinRisk Ltd (2013) all rights reserved

17. Gold standard validation Independent – not involving original study authors External - test on cohort of patients not involved in original derivation Essential step to check model Identifies right patients Works in other settings – transportability Different from evaluation QCancer validation by Oxford University (Collins & Altman) Collins GS, Altman DG. Identifying patients with undetected gastro-oesophageal cancer in primary care: External validation of QCancer® (Gastro-Oesophageal). European journal of cancer 2012 Collins GS, Altman DG. Identifying women with undetected ovarian cancer: independent and external validation of QCancer((R)) (Ovarian) prediction model. European journal of cancer care 2012 doi: 10.1111/ecc.12015. Collins GS, Altman DG. Identifying patients with undetected colorectal cancer: an independent validation of QCancer (Colorectal). Br J Cancer 2012 Copyright ClinRisk Ltd (2013) all rights reserved

18. Validation Results Women: Discrimination ROC values Combined symptoms model* in separate QResearch cohort Individual cancer models in separate cohort from QResearch independent external validation in Vision data any cancer0.85n/aIn progress lung0.910.92In press colorectal0.89 0.92 gastro0.900.890.93 pancreas0.870.84In press ovary0.84 0.86 renal0.900.91In press breast0.88n/aIn progress blood0.79n/aIn progress uterus0.91n/aIn progress cervix0.73n/aIn progress other0.82n/aIn progress *reference: Symptoms and risk factors to identify women with suspected cancer in primary care; BJGP; 2013 Copyright ClinRisk Ltd (2013) all rights reserved

19. Combined symptoms model* in separate QResearch cohort Individual cancer models in separate cohort from QResearch independent external validation in Vision data any cancer0.89n/aIn progress lung0.92 In press colorectal0.900.91 gastro0.930.920.94 pancreas0.890.87In press renal0.940.95In press prostate0.90n/aIn progress blood0.83n/aIn progress testis0.82n/aIn progress other0.86n/aIn progress *reference: Symptoms and risk factors to identify men with suspected cancer in primary care; BJGP; 2013 Validation Results Men: Discrimination ROC values Copyright ClinRisk Ltd (2013) all rights reserved

20. Validation women calibration mean predicted & observed risk Copyright ClinRisk Ltd (2013) all rights reserved

21. Validation women calibration mean predicted & observed risk Copyright ClinRisk Ltd (2013) all rights reserved

22. Validation women calibration mean predicted & observed risk Copyright ClinRisk Ltd (2013) all rights reserved

23. Validation women calibration mean predicted & observed risk Copyright ClinRisk Ltd (2013) all rights reserved

24. Validation women calibration mean predicted & observed risk Copyright ClinRisk Ltd (2013) all rights reserved

25. Comparison strategies in women based on top 10% risk Risk threshold % Sensitivit y (%) Specificit y (%) PPV (%) NPV (%) lung cancer0.3872.190.11.299.9 colorectal0.3568.290.11.499.9 gastro-oesoph0.1475.090.10.6100.0 pancreas0.1267.990.00.4100.0 ovarian0.1861.690.00.6100.0 renal0.1076.790.00.6100.0 breast0.7268.190.44.699.8 blood0.2244.090.00.599.9 uterine0.1083.790.10.7100.0 cervical0.0554.590.00.2100.0 other0.5519.490.00.1100.0 Copyright ClinRisk Ltd (2013) all rights reserved

26. Comparison strategies in men based on top 10% cancer risk % thresholdSensitivity (%) specificity (%) PPV (%) NPV (%) lung0.6771.590.21.999.9 colorectal0.4569.590.21.899.9 gastro-oesoph0.2976.690.11.3100.0 pancreas0.1070.190.00.4100.0 renal tract0.2082.790.11.6100.0 prostate1.3059.590.22.299.8 blood0.2749.090.10.699.9 testicular0.0267.190.00.2100.0 other0.6627.090.00.1100.0 Copyright ClinRisk Ltd (2013) all rights reserved

27. Web calculator www.qcancer.org Publically available Interactive Gives global cancer risk Risk of different cancer types Visual display of absolute risk to help shared decision making eg “of a 100 patients like 7you. 20 will have cancer” http://www.qcancer.org/2013/male/ http://www.qcancer.org/2013/female/ Copyright ClinRisk Ltd (2013) all rights reserved

28. Using QCancer in practice – v similar to QRISK2 Standalone tools a. Web calculator www.qcancer.org/2013/female/php www.qcancer.org/2013/male/php b. Windows desk top calculator c. Iphone – simple calculator Integrated into clinical system a. Within consultation: GP with patients with symptoms b. Batch: Run in batch mode to risk stratify entire practice or PCT population Copyright ClinRisk Ltd (2013) all rights reserved

29. GP systems integration Batch processing Similar to QRISK2 which is in 95% of GP practices– automatic daily calculation of risk for all patients in practice based on existing data. Identify patients with symptoms/adverse risk profile without follow up/diagnosis Enables systematic recall or further investigation Systematic approach - prioritise by level of risk. Safety netting Copyright ClinRisk Ltd (2013) all rights reserved

30. Use QCancer website by single cancer site (source: google 01 Nov 2012 to 10 Dec 2012) SiteJournalPublication Date Web site visits % from UK (NHS) OvaryBMJJan 20124,46536% ColorectalBJGPJan 20122,79977% Gastro-oesophagealBJGPNov 20112,26565% LungBJGPNov 20114,66259% PancreasBJGPNov 20111,24361% RenalBJGPMarch 201238180% total15,81557% Copyright ClinRisk Ltd (2013) all rights reserved

31. Project update November 2012

32. What are cancer decision support tools? Cancer decision support tools (CDS) are an aid to clinical decision-making, to assist GPs in their decisions about whether or not to refer or request further diagnostic investigation, in patients where they believe there is a risk of cancer The tools in this project display the risk of a patient having a specific type of cancer This risk is founded on analysis of historic population cohorts and their risk of having cancer based on a range of factors including symptoms, medical history and demographic data. In this project we are focusing on rollout of two existing high profile tools: 1.QCancer developed by Professor Julia Hippisley-Cox 2.The Risk Assessment Tool (RAT), developed by Professor Willie Hamilton

33. NovemberDecemberJanuaryFebruaryMarchAprilMay 2012 2013 Assimilation of QCancer and RAT into IT platform June Development of training materials Cancer network training sessions JulyAugust CDS project phase one timeline Phase one end Engagement with GP IT providers Phase one tool in use GP training rollout Data gathering for evaluation Define evaluation methodology GP practices recruited Load software onto GP systems

34. Who is involved? This project will focus on the QCancer and RAT tools Macmillan Cancer Support will lead the project, with an evaluation overseen by Cancer Research UK, and with part-funding from the Department of Health. The project forms part of the National Awareness and Early Diagnosis Initiative (NAEDI) In phase one of the project, thirteen English cancer networks will lead the rollout at a local level, working with over 650 GP practices Practices in Wales and Scotland are also planning to take part In phase one, the tool used in practice will be developed by BMJ Informatica, though it will work across a wide range of native GP IT software In parallel, the project will engage with GP software providers, to promote understanding and ensure widespread use and rollout after phase one

35. What will the tool look like? Cancer decision support tools are designed to assess the risk of a patient having an existing, but as yet undiagnosed cancer, by calculating a risk based on factors such as symptoms, medical history and demographic profile GPs will use in everyday practice an IT-based cancer decision support tool with a simple user interface which can display scores for either RAT or QCancer The tool will work in three ways: i.Reactive prompt ii.Symptom checker iii.Audit function

36. (i) Reactive prompt Working automatically in the background using READ-coded information, the tool will calculate a risk of having cancer for every patient seen in consultation. If the risk is above a certain level, a prompt will appear on screen letting the GP know that they might like to consider whether the patient might warrant a referral or investigation for a suspected cancer.

37. (ii) Symptom checker Used in consultation where READ codes are not already known, a symptom checker can be called up, which allows the GP to enter relevant symptoms, calculate a risk, and then re-enter observed symptoms into the patient’s record as a READ-coded entry.

38. (iii) Audit function An audit function will show calculated risk levels of all registered patients on a practice’s list. This can be sorted to show those calculated to have the highest risk, and used to consider whether any further action should be taken for these patients.

39. Thank you for listening Questions & Discussion This work by Julia Hippisley-Cox is licensed under a Creative Commons Attribution-ShareAlike 3.0 Unported LicenseJulia Hippisley-CoxCreative Commons Attribution-ShareAlike 3.0 Unported License