+ QCancer Scores –a new approach to identifying patients at risk of having cancer Julia Hippisley-Cox, GP, Professor Epidemiology & Director ClinRisk Ltd Pancreatic cancer UK Summit 2012 27th June 2012 + Acknowledgements Co-author Dr Carol Coupland QResearch database University of Nottingham ClinRisk (software) EMIS & contributing practices & User Group BJGP and BMJ for publishing the work Oxford University (independent validation) cancer teams, DH + RCGP+ other academics with whom we are now working + QResearch Database Over 700 general practices across the UK, 14 million patients Joint not for profit venture University of Nottingham and EMIS (supplier > 55% GP practices) Validated database – used to develop many risk tools Available for peer reviewed academic research where outputs made publically available Practices not paid for contribution but get integrated QFeedback tool and utilities eg QRISK, QDiabetes, QFracture. Data linkage – deaths, deprivation, cancer, HES + Clinical Research Cycle Clinical practice & benefit Integration clinical system Clinical questions Research + innovation + QScores – new family of Risk Prediction tools for decision support Individual assessment Who is most at risk of 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 level 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 + Why pancreatic cancer? 11th most common cancer < 20% patients suitable for surgery 84% dead within a year of diagnosis Chances of survival better if diagnosis made at early stage Very few established risk factors (smoking, chronic pancreatitis, alcohol) so screening programme unlikely Challenge is to identify symptoms in primary care particularly hard for pancreatic cancer + Symptoms based approach 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, FH updated to meet changing requirements, populations, recorded data + QCancer scores – what they need to do Accurately predict level of risk for individual based on risk factors and 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. + Methods – development algorithm Huge representative sample from primary care aged 30-84 Identify new alarm symptoms (eg appetite loss, weight loss, abdo distension) and other risk factors (eg age, smoking, smoking, family history) Identify cancer outcome - all new diagnoses either on GP record or linked ONS deaths record in next 2 years Established methods to develop risk prediction algorithm Identify independent factors adjusted for other factors Measure of absolute risk of cancer. Eg 5% risk of pancreatic cancer + ‘Red’ flag or alarm symptoms Haemoptysis Loss of appetite Haematemesis Weight loss Dysphagia Indigestion +/- heart burn Rectal bleeding Abdominal pain Postmenopausal bleeding Abdominal swelling Haematuria Family history dysphagia Anaemia Constipation cough + Incidence of key symptoms vary by age and sex + Currently Qcancer predicts risk 6 cancers Lung Pancreas Ovary Colorectal Kindey Gastro-oesoph + Results – the algorithms/predictors Outcome Risk factors Symptoms Lung Age, sex, smoking, deprivation, COPD, prior cancers Haemoptysis, appetite loss, weight loss, cough, anaemia Gastrooeso Age, sex, smoking status Haematemsis, appetite loss, weight loss, abdo pain, dysphagia, dyspepsia/hearburn Colorectal Age, sex, alcohol, family history Rectal bleeding, appetite loss, weight loss, abdo pain, change bowel habit, anaemia Pancreas Age, sex, type 2, chronic pancreatitis dysphagia, appetite loss, weight loss, abdo pain, abdo distension, constipation, dyspepsia/heartburn Ovarian Age, family history Rectal bleeding, appetite loss, weight loss, abdo pain, abdo distension, PMB, anaemia Renal Age, sex, smoking status, prior cancer Haematuria, appetite loss, weight loss, abdo pain, anaemia + Methods - validation Previous QScores validation – similar or better performance on external data Once algorithms developed, tested performance separate sample of QResearch practices fully 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 – PPV, sensitivity, ROC etc + Results of validation Explained 59-62% of variation R2 ROC 0.84 (women) and 0.87 (men) D statistic high (2.44 for women and 2.61 men) Calibration – close predicted vs observed Good sensitivity : The 10% of patients with highest risk accounted for 62% of all pancreatic cancers diagnosed in next two years + Qcancer.org web calculator PROFILE • 64 yr woman • non smoker • 3+unit alcohol • type2 diabetes • chronic pancreatitis • Loss appetite and weight • Indigestion • Anaemia RISKS • Pancreatic cancer 12% • Gastrooesophageal 7% • Colorectal 4% • Ovarian cancer 2% • Renal cancer 1% • Lung cancer 2% + GP system integration: Within consultation Uses data already recorded (eg age, family history) Stimulate better recording of positive and negative symptoms Automatic risk calculation in real time Display risk enables shared decision making between doctor and patient Information stored in patients record and transmitted on referral letter/request for investigation Allows automatic subsequent audit of process and clinical outcomes Improves data quality leading to refined future algorithms. + GP systems integration Batch processing Similar to QRISK which is in 90% 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. Integration means software can be rigorously tested so ‘one patient, one score, anywhere’ Cheaper to distribute updates + Thank you for listening Any questions (if time)