BMJ Open. General practice / Family practice. Keywords: PREVENTIVE MEDICINE, PRIMARY CARE, THERAPEUTICS

Transcription

1 A cohort study investigating the relationship between cardiovascular risk scoring and the prescribing of statins in UK primary care: study protocol Journal: Manuscript ID bmjopen-0-00 Article Type: Protocol Date Submitted by the Author: -Jul-0 Complete List of Authors: Finnikin, Samuel; University of Birmingham, Primary Care Clinical Sciences Ryan, Ronan; University of Birmingham, Institute of applied health research Marshall, Tom; University of Birmingham, Public Health and Epidemiology <b>primary Subject Heading</b>: General practice / Family practice Secondary Subject Heading: Cardiovascular medicine, Public health Keywords: PREVENTIVE MEDICINE, PRIMARY CARE, THERAPEUTICS : first published as 0./bmjopen-0-00 on November 0. Downloaded from on April 0 by guest. Protected by copyright.

2 Page of A cohort study investigating the relationship between cardiovascular risk scoring and the prescribing of statins in UK primary care: study protocol Dr Samuel Finnikin, Dr Ronan Ryan, Prof Tom Marshall Institute of applied health research, University of Birmingham, UK Corresponding author: Dr S Finnikin, Institute of applied health research, Murray Learning Centre, University of Birmingham, Edgbaston, Birmingham, UK, B TT. finniksj@bham.ac.uk Tel: + (0) Fax: not available Key words: Statins, HMG-CoA; Cardiovascular Diseases; Primary Prevention; Primary Health Care Word count: Abstract Introduction The use of cardiovascular risk scoring is an integral part of the prevention of cardiovascular disease (CVD) and forms the basis for the decision to offer medication reduce cholesterol (statins). However, there is considerable evidence that many high risk patients are not being prescribed statins, and many patients are being prescribed statins despite being at low risk. This suggests that the utility of CVD risk scoring is not currently optimised. In order to maximise the benefit and minimise the harms of statin treatment, it is important to investigate the role that CVD risks scoring plays in the decision to prescribe. This research will establish how calculating CVD risk score influences the prescribing of statins for the primary prevention of CVD in primary care. - : first published as 0./bmjopen-0-00 on November 0. Downloaded from on April 0 by guest. Protected by copyright.

3 Page of Methods and analysis The Health Improvement Network (THIN) is a database consisting of coded primary care data from over 00 general practices in the UK. From this resource, a historical cohort of patients who could be initiated on statins for the primary prevention of CVD and had a QRISK score coded will be created. They will be matched to patients who would also be eligible for statin treatment and consulted their GP but did not have a QRISK score coded. The outcome of interest is the subsequent initiation of a statin in the 0 days after consulting. Primary analysis will consist of predictive modelling using multivariate logistic regression. Descriptive statistics will be used to identify trends in prescribing and secondary analysis will explore what other factors may be influencing the prescribing of statins. Ethics and dissemination Ethical approval to extract data from the THIN database for this research has been granted. The results will be published in a peer review journal and presented at national and international conferences. Strengths: The use of a large, validated and generalisable dataset gives an unparalleled insight into current real life medical practice. The use of a large, match cohort will allow the impact of CVD risk estimation to be modelled Limitations in an insightful and novel way. Insight into the decision to prescribe statins is limited to what is coded in the electronic patient record and therefore much of the decision making process will remain unknown. Recording of data is known to be incomplete and therefore some estimates and approximations will have to be made. - : first published as 0./bmjopen-0-00 on November 0. Downloaded from on April 0 by guest. Protected by copyright.

4 Page of Introduction Predicted risk of cardiovascular disease (CVD) is determined from a combination of age, sex and cardiovascular risk factors [, ]. Assessment of predicted risk of cardiovascular disease has been a fundamental part of national clinical guidance on the CVD prevention in primary care since []. In the UK, an absolute risk of CVD greater than a threshold was, and remains, a criterion for offering statin treatment and in, combination with raised blood pressure, a criterion for antihypertensive treatment [-]. Risk assessment is also embedded in clinical guidelines around the world including New Zealand, Europe and America [-]. A variety of different risk calculators are available and whilst QRISK [0] is the calculator of choice in England and Wales, elsewhere, others such as Framingham [], Score [] or Procam [] are recommended. When it comes to the primary prevention of CVD, estimated CVD risk is the best predictor of likely benefits of treatment and therefore this information is a necessary component for effective shared decision making []. Evidence suggests that utilisation of risk scoring improves perceived CVD risk and medical prescribing without causing harms [-]. In the context of primary prevention there is evidence both of undertreatment of high risk patients (above the treatment threshold) and inappropriate treatment of low risk patients (below the treatment threshold) [-0]. Moreover, patients with the highest absolute CVD risk are at greatest risk of being undertreated []. One reason for this is that risk scores may not be consistently used by clinicians to guide decision making []. Reasons for not using risk score calculators include: a belief that risk scoring oversimplifies the decision and may lead to over prescribing [], confusion over how best to use risk score calculators [] or a focus on individual risk factors over absolute risk score []. - : first published as 0./bmjopen-0-00 on November 0. Downloaded from on April 0 by guest. Protected by copyright.

5 Page of In 0, guidelines for England and Wales on lipid modification reduced the predicted risk threshold at which patients should be offered statins from a 0% 0-year risk to 0% [, ]. This recommendation was met with considerable concern from the medical community that it would lead to widespread over prescribing of statins with little clinical benefit but considerable potential for harm [, ]. This widely publicised criticism may have increased concerns that risk scoring leads to oversimplification of clinical decisions and overprescribing and may have, perversely, decreased the use of risk scoring and appropriate prescribing of statins. This research will investigate the relationship between calculating a cardiovascular risk score and subsequent initiation of statins by primary care clinicians. This will give an indication of how the calculated cardiovascular risk score is used by clinicians in routine practice to guide the decision to prescribe statins and how this has changed with time and updates to national guidance. The consistent use of risk scoring can result in more targeted prescribing and fewer patients being prescribed medication overall [] which maximises the efficiency and cost effectiveness of treatment at a population level. Aims and Objectives Aim: To understand how calculating CVD risk score influences the prescribing of statins for the primary prevention of CVD in primary care. Objectives: For primary prevention of CVD. To establish the relationship between calculation of QRISK score and the subsequent initiation of statin therapy in primary care. - : first published as 0./bmjopen-0-00 on November 0. Downloaded from on April 0 by guest. Protected by copyright.

6 Page of To investigate how the relationship between calculation of QRISK score and statin prescribing has changed over recent years.. To investigate the contribution of calculated QRISK score and other patient characteristics to the probability of initiating statin therapy. Methods and analysis Study design A historical open cohort study of patients who have a first ever QRISK score recorded and controls matched by age, sex, consultation date and GP practice. Data source Data will be extracted from anonymised primary care records from practices in England and Wales who contribute to The Health Improvement Network (THIN) database. The THIN database contains routine patient data from over 00 UK general practices which is generalisable to the UK population []. The data has been used in previous studies to validate QRISK and it was shown that the discrimination statistics in THIN are as good as those for the original QRISK cohort [0, ]. Practices that contribute to THIN use the Vision (In Practice Systems) electronic patient records system. Clinical data are coded using Read code clinical classification version [] and drug codes are used which correspond to the British National Formulary (BNF) []. Population Patients will be included if they are aged between 0 and years (the age range recommended by NICE for risk assessment[]) and are suitable for risk assessment for the primary prevention of CVD. Patients will be excluded if they have pre-existing CVD, have a contraindication to statin therapy, are already being prescribed a statin (or other lipid lowering therapy) or have ever been prescribed a statin in the past. For each exposed patient an unexposed control will be selected who is age and sex - : first published as 0./bmjopen-0-00 on November 0. Downloaded from on April 0 by guest. Protected by copyright.

7 Page of matched, registered at the same practice and has consulted a doctor within 0 days of the exposed patient s consultation. All patients will be followed for 0 days following their index consultation. The unexposed cohort have the same exclusion criteria as the exposed cohort with the additional exclusion that they must not have a QRISK score documented in the follow up period. It is foreseeable that some of the unexposed cohort may subsequently go on to have a QRISK score calculated in the study period but after the follow up period. This will be allowable and identified in the analysis. Both exposed and unexposed patient will be required to remain registered and free of CVD and statin contraindications for at least 0 days following their index date. In order to maintain a stable cohort of practices, only THIN practices in which there are data available for the whole study period will be included. The study period will run from the beginning of 0 to the end of 0. Patients will be eligible for inclusion from the earliest of the following dates: study start date, acceptable mortality reporting date (which ensures that patient deaths and deregistrations are being recorded consistently []), Vision date plus one year, registration date plus one year (to ensure time for baseline data to be recorded), and age 0; until the earliest of the following dates: age, study end date, CVD diagnosis, statin initiation, and recording of a contraindication for the prescribing of statins. Study variables Outcome variables The outcome will be the first prescription of a statin within 0 days of the index date (drug codes available on request). Exposure variables The exposure of interest is the coding of a QRISK score. This is taken to indicate that QRISK score was calculated by the clinician. Patients will enter the open cohort the first time their QRISK score is recorded (their index date). If the exposed patients have subsequent QRISK scores coded after - : first published as 0./bmjopen-0-00 on November 0. Downloaded from on April 0 by guest. Protected by copyright.

8 Page of inclusion in the cohort and before the end of the 0 days follow up, the latest score will be assumed to be the most accurate and will be used in analysis in place of the score calculated at the index consultation, but the index consultation date will remain unchanged. Predictor variables Sociodemographic details will be obtained for each patient: age, sex, ethnicity, Townsend quintile and rurality index. Each of these will be determined from the values recorded nearest to the index date. General practice and clinician ID will also be recorded. Patient variables will include the cardiovascular risk factors required to calculate QRISK. These will be obtained from the records nearest to, but prior to, the index date. In addition to sociodemographic variables these include patient measurements recorded up to years prior to the index date, diagnoses recorded at any time prior to the index date. Measurements include blood pressure (BP), total cholesterol, HDL cholesterol, liver transaminase, BMI and smoking status (the most recent record prior to the index date). Diagnoses include diabetes mellitus (type or type ), family history of CHD, atrial fibrillation, rheumatoid arthritis (and other systemic inflammatory conditions), severe mental health conditions, HIV and CKD. Data on medication prescriptions will be extracted. Patients receiving a prescription for antihypertensive medication up to 0 days prior to the index date will be considered to be on treatment for hypertension. Prescriptions of antipsychotics, corticosteroids and immunosuppressant medication will be recorded as these can change the threshold for prescribing lipid lowering medication. Missing data and extreme values Some variables will have missing data, including ethnicity, BMI, total cholesterol, HDL cholesterol, BP and smoking status. If comorbidities are missing they will be assumed to be absent. Missing values arise twice; first when calculating the QRISK score for unexposed patients, and secondly when constructing the multivariable model. - : first published as 0./bmjopen-0-00 on November 0. Downloaded from on April 0 by guest. Protected by copyright.

9 Page of Where data is missing from variables used in QRISK calculations, the default values used by the QRISK calculator will be used []. In the multivariable model missing data may be present in both the unexposed and exposed groups and the fact that some data is missing may affect the decision to prescribe statins. All variables in the model are categorical (apart from age but there will be no missing ages) and a missing category will be used to account for missing data. Biologically implausible values for blood pressure, lipid profile, height and weight will be identified and excluded. Analysis The crude rate of QRISK coding and statin initiation will be calculated and analysed over the study time period. Both exposed and unexposed patients will be stratified into three groups based on QRISK score (<0%, 0%-.%, and 0%). For the exposed group, this will be based on recorded QRISK score (or the latest recorded in the follow up period) and for unexposed controls will be based on QRISK score calculated post-hoc from the data. This will require imputation of missing data and the QRISK algorithm may differ slightly from that used by the clinical system so it is acknowledged that this may not be completely consistent with the CVD risk that would have been calculated by the practice system. Table : Variables used in predictive modelling Variable Categories QRISK <0%, 0%-.%, >0% Age Continuous Sex Male, female Ethnicity White or not stated, Indian, Pakistani, Bangladeshi, Other Asian, Black Caribbean, Black African, Chinese, Other, missing Townsend deprivation quintile st, nd, rd, th, th, missing Urban/rural score Urban, rural, missing BMI <0, 0-., -., 0-., +, missing Smoking status Current smoker, ex-smoker, non-smoker, missing Liver transaminases Normal, abnormal, missing Diabetes Mellitus Yes, no - : first published as 0./bmjopen-0-00 on November 0. Downloaded from on April 0 by guest. Protected by copyright.

10 Page of Family History of CVD Yes, no Atrial Fibrillation Yes, no Rheumatoid Arthritis Yes, no Hypertension Yes, no Chronic Kidney Disease Yes, no HIV Yes, no Severe enduring mental illness * Yes, no Inflammatory conditions ** Yes, no Antipsychotic medication Yes, no Long term corticosteroids Yes, no Immunosuppressant medication Yes, no Year quartile st, nd, rd, th *schizophrenia, bipolar affective disorder, mania or psychosis (unspecified), psychotic depression ** Rheumatoid arthritis, systemic lupus erythematosus, scleroderma, Bechet s syndrome, other inflammatory arthritis The primary analysis will consist of predictive modelling using multivariate logistic regression (variables summarised in table ). Clustering of responses within individual GP practices will be included in the model, and multivariable fractional polynomials will be explored to allow for nonlinear relationships between continuous covariates and outcome. Variation throughout the year due to the seasonality imposed by QOF targets will be accounted for by adding the year quarter as a variable in the model. Assuming a minimum sample size for analysis of 0 per degree of freedom [], and degrees of freedom in the variables, a minimum sample of 0 patients with a QRISK score and subsequent statin initiation should easily be obtained. An estimate of the proportions of the THIN population in the various QRISK categories can be made by extrapolating the results of an analysis of a THIN cohort study from 0 which showed that, in the age range 0-,.% of women and.% of men have a QRISK of >0% []. If the threshold is 0%, these proportions are 0.% and.%. In this cohort, just over million patients were identified in this age range which will result in a potential for 00,000 patients in each risk category before exclusions. It is difficult to predict how many of these patients will have QRISK scores documented, however, at minimum, every patient who attends an NHS health check should have a QRISK score as an outcome and 0% of 0- years olds are invited to have a health check - : first published as 0./bmjopen-0-00 on November 0. Downloaded from on April 0 by guest. Protected by copyright.

11 Page 0 of each year, with approximately 0% taking up this offer []. This would generate 0,000 coded QRISK scores in the target population each year. If one in ten of these individuals is prescribed a statin this means there will be 000 outcome events, sufficient to undertake analysis including variables with 00 degrees of freedom which is more than sufficient for the proposed modelling []. Exploratory analysis will investigate whether the absolute level of cholesterol influences the prescribing of statins over and above its contribution to the QRISK score. Potential limitations The use of electronic patient records, whilst excellent for allowing access to large amounts of reallife data, does not facilitate deeper understanding of the decision making process. We will not be able to explore why a patient with a particular QRISK score does or does not receive a prescription for a statin. We cannot gain insight into what discussion took place as to the risk and benefits of treatment or even whether the clinician discussed the QRISK score with the patient. It is also possible that, because only coded data is available, QRISK score may be calculated in some patients but not recorded in GP records. The Vision EPR system automatically displays a CVD risk score for patients who may benefit from CVD risk reduction [] and the clinician could act on this information without it being coded. This would misclassify exposed patients as unexposed and therefore reduce the strength of the relationship between QRISK calculation and initiating a statin. It is foreseeable that the data available for patients with a recent QRISK score coded will be more complete and up to date than those who have not had a QRISK score coded. This has the potential to introduce some bias which, given that variables such as blood pressure and cholesterol tend to increase over time, is likely to underestimate the calculated QRISK for the unexposed group. It will - : first published as 0./bmjopen-0-00 on November 0. Downloaded from on April 0 by guest. Protected by copyright.

12 Page of not be possible to assess the size of this bias but it will be considered during interpretation of the results. In conclusion, whilst the methodology will not facilitate better understanding of the individual decision making consultations, it will provide new and valuable insight into the way that risk scoring is currently being used in primary care. We will be able to establish whether the use of risk scoring is resulting in more evidence based prescribing and improve understanding of the real-life use of risk scoring which will inform their future utilisation in clinical guidance. Ethics The THIN Data Collection Scheme was approved by the South-East Multicentre Research Ethics Committee (SE-MREC) in 00. Use of the data for research purposes does not required further ethical review but must be approved by the independent Scientific Review Committee (SCR). This research was approved by the SCR on March 0, reference THIN00. Contributorship statement: SF generated the initial idea as doctoral research supervised by RR and TM. All authors contributed to the refinement of the study design and the initial manuscript was drafted by SF. RR and TM provided feedback on the manuscript and all authors approved the final version. Competing interests: None of the authors have competing interests to declare References. Wilson, P.W., et al., Prediction of coronary heart disease using risk factor categories. Circulation,. (): p. -.. Hippisley-Cox, J., et al., Derivation and validation of QRISK, a new cardiovascular disease risk score for the United Kingdom: prospective open cohort study. Bmj, 00. (): p... Wood, D., et al., Joint British recommendations on prevention of coronary heart disease in clinical practice. : BMJ Publishing Group.. NICE, Lipid modification: Cardiovascular risk assessment and the modification of blood lipids for the primary and secondary prevention of cardiovascular disease. 0, National Institute for Health and Care Excellence: London. - : first published as 0./bmjopen-0-00 on November 0. Downloaded from on April 0 by guest. Protected by copyright.

13 Page of NICE, Hypertension: Clinical management of primary hypertension in adults. 0, National Institute for Health and Clinical Excellence: London.. NICE, Lipid modification: Cardiovascular risk assessment and the modification of blood lipids for the primary and secondary prevention of cardiovascular disease. 00, National Institute for Health and Care Excellence: London.. New Zealand Guidelines Group, New Zealand Guidelines Group. New Zealand Primary Care Handbook 0 (updated 0): Cardiovascular disease risk assessment. 0: Wellington.. Perk, J., et al., European Guidelines on cardiovascular disease prevention in clinical practice (version 0). European heart journal, 0. (): p Goff Jr, D., et al., American Heart Association Task Force on Practice G. 0 ACC/AHA guideline on the assessment of cardiovascular risk: a report of the American College of Cardiology/American Heart Association task force on practice guidelines. Circulation, 0. : p. S-. 0. Hippisley-Cox, J., et al., Predicting cardiovascular risk in England and Wales: prospective derivation and validation of QRISK. BMJ (Clinical research ed.), 00. (): p. -.. Conroy, R., et al., Estimation of ten-year risk of fatal cardiovascular disease in Europe: the SCORE project. European heart journal, 00. (): p Assmann, G., P. Cullen, and H. Schulte, Simple scoring scheme for calculating the risk of acute coronary events based on the 0-year follow-up of the prospective cardiovascular Münster (PROCAM) study. Circulation, 00. 0(): p Elwyn, G., et al., Shared decision making and the concept of equipoise: the competences of involving patients in healthcare choices. Br J Gen Pract, (0): p. -.. Sheridan, S.L., et al., The effect of giving global coronary risk information to adults: a systematic review. Archives of Internal Medicine, 00. 0(): p Sheridan, S.L. and E. Crespo, Does the routine use of global coronary heart disease risk scores translate into clinical benefits or harms? A systematic review of the literature. BMC health services research, 00. (Journal Article): p Usher-Smith, J.A., et al., Impact of provision of cardiovascular disease risk estimates to healthcare professionals and patients: a systematic review., 0. (0).. Heeley, E.L., et al., Cardiovascular risk perception and evidence-practice gaps in Australian general practice (the AusHEART study). Med J Aust, 00. (): p. -.. Wu, J., et al., Patient factors influencing the prescribing of lipid lowering drugs for primary prevention of cardiovascular disease in UK general practice: a national retrospective cohort study. PLoS ONE, 0. (): p. e.. Homer, K., et al., Statin prescribing for primary prevention of cardiovascular disease: a crosssectional, observational study. The British journal of general practice : the journal of the Royal College of General Practitioners, 0. (): p. e-. 0. van Staa, T.P., et al., The efficiency of cardiovascular risk assessment: do the right patients get statin treatment? Heart, 0. (): p Barham, A.H., et al., Appropriateness of cholesterol management in primary care by sex and level of cardiovascular risk. Preventive cardiology, 00. (): p Sposito, A.C., et al., Physicians' attitudes and adherence to use of risk scores for primary prevention of cardiovascular disease: cross-sectional survey in three world regions. Current medical research and opinion, 00. (): p. -.. Hobbs, F.D., et al., Barriers to cardiovascular disease risk scoring and primary prevention in Europe. QJM : monthly journal of the Association of Physicians, 00. 0(0): p. -.. Liew, S.M., et al., Cardiovascular risk scores: qualitative study of how primary care practitioners understand and use them. British Journal of General Practice, 0. (): p. e : first published as 0./bmjopen-0-00 on November 0. Downloaded from on April 0 by guest. Protected by copyright.

14 Page of Vagholkar, S., et al., Influence of cardiovascular absolute risk assessment on prescribing of antihypertensive and lipid-lowering medications: a cluster randomized controlled trial. American Heart Journal, 0. (): p. -.. Abramson, J.D., et al., Should people at low risk of cardiovascular disease take a statin? BMJ (Clinical research ed.), 0. (Journal Article): p. f.. Wise, J., Open letter raises concerns about NICE guidance on statins. BMJ, 0. : p. g.. Doust, J., et al., Prioritising CVD prevention therapy: Absolute risk versus individual risk factors. Australian Family Physician, 0. (0): p. 0.. Blak, B.T., et al., Generalisability of The Health Improvement Network (THIN) database: demographics, chronic disease prevalence and mortality rates. Informatics in primary care, 0. (): p Hippisley-Cox, J., et al., Performance of the QRISK cardiovascular risk prediction algorithm in an independent UK sample of patients from general practice: a validation study. Heart, 00. (): p. -.. Collins, G.S. and D.G. Altman, An independent and external validation of QRISK cardiovascular disease risk score: a prospective open cohort study. BMJ (Clinical research ed.), 00. 0(Journal Article): p. c.. Health and Social Care Information Centre. Read codes. //0]; Available from: Joint Formulary Committee, British National Formulary 0th ed. 0, London: BMJ Group and Pharmaceutical PRess.. Maguire, A., B.T. Blak, and M. Thompson, The importance of defining periods of complete mortality reporting for research using automated data from primary care. Pharmacoepidemiol Drug Saf, 00. (): p. -.. INPS, CVD/Stroke risk calculators within Vision, in Vision. 0.. Peduzzi, P., et al., A simulation study of the number of events per variable in logistic regression analysis. Journal of clinical epidemiology,. (): p. -.. Collins, G.S. and D.G. Altman, Predicting the 0 year risk of cardiovascular disease in the United Kingdom: independent and external validation of an updated version of QRISK. BMJ (Clinical research ed.), 0. (Journal Article): p. e.. NHS Health Check, NHS Health Check e-bulletin May : first published as 0./bmjopen-0-00 on November 0. Downloaded from on April 0 by guest. Protected by copyright.

15 A cohort study investigating the relationship between cholesterol, cardiovascular risk score and the prescribing of statins in UK primary care: study protocol Journal: Manuscript ID bmjopen-0-00.r Article Type: Protocol Date Submitted by the Author: 0-Oct-0 Complete List of Authors: Finnikin, Samuel; University of Birmingham, Primary Care Clinical Sciences Ryan, Ronan; University of Birmingham, Institute of applied health research Marshall, Tom; University of Birmingham, Public Health and Epidemiology <b>primary Subject Heading</b>: General practice / Family practice Secondary Subject Heading: Cardiovascular medicine, Public health Keywords: PREVENTIVE MEDICINE, PRIMARY CARE, THERAPEUTICS : first published as 0./bmjopen-0-00 on November 0. Downloaded from on April 0 by guest. Protected by copyright.

16 Page of A cohort study investigating the relationship between cholesterol, cardiovascular risk score and the prescribing of statins in UK primary care: study protocol Dr Samuel Finnikin, Dr Ronan Ryan, Prof Tom Marshall Institute of Applied Health Research, University of Birmingham, UK Corresponding author: Dr S Finnikin, Institute of Applied Health Research, Murray Learning Centre, University of Birmingham, Edgbaston, Birmingham, UK, B TT. finniksj@bham.ac.uk Tel: + (0) Fax: not available Key words: Statins, HMG-CoA; Cardiovascular Diseases; Primary Prevention; Primary Health Care Word count:, Abstract Introduction Risk scoring is an integral part of the prevention of cardiovascular disease (CVD) and should form the basis for the decision to offer medication to reduce cholesterol (statins). However, there is a suggestion in the literature that many patients are still initiated on statins based on raised cholesterol rather than a raised CVD risk. It is important, therefore, to investigate the role that lipid levels and CVD risks have in the decision to prescribe. This research will establish how cholesterol levels and CVD risk independently influence the prescribing of statins for the primary prevention of CVD in primary care. - : first published as 0./bmjopen-0-00 on November 0. Downloaded from on April 0 by guest. Protected by copyright.

17 Page of Methods and analysis The Health Improvement Network (THIN) is a database of coded primary care electronic patient records from over 00 UK general practices. From this resource, a historical cohort will be created of patients without a diagnosis of CVD, not currently receiving a prescription for statins and who had a lipid profile measured. A post hoc QRISK score will be calculated for these patients and they will be followed up for 0 days to establish whether they were subsequently prescribed a statin. Primary analysis will consist of predictive modelling using multivariate logistic regression with potential predictors including cholesterol level, calculated QRISK score, sociodemographic characteristic and comorbidities. Descriptive statistics will be used to identify trends in prescribing and further secondary analysis will explore what other factors may have influenced the prescribing of statins and the degree of inter-prescriber variability. Ethics and dissemination The THIN Data Collection Scheme was approved by the South-East Multicentre Research Ethics Committee in 00. Individual studies using THIN require Scientific Review Committee approval. The original protocol for this study and a subsequent amendment have been approved [THIN00, XXXXX].The results will be published in a peer review journal and presented at national and international conferences. Strengths: The use of a large, validated and generalisable set of general practices gives an unparalleled insight into current real life medical practice. The use of a large cohort will allow the impact of lipid profiles and CVD risk estimation to be Limitations modelled in an insightful and novel way. - : first published as 0./bmjopen-0-00 on November 0. Downloaded from on April 0 by guest. Protected by copyright.

18 Page of Insight into the decision to prescribe statins is limited to information that is coded in the electronic patient record and therefore much of the decision making process will remain unknown. Recording of data may be incomplete and therefore some approximations will have to be made. Introduction Predicted risk of cardiovascular disease (CVD) is determined from a combination of age, sex and cardiovascular risk factors [, ]. Risk of CVD is the best predictor of the benefits of statins because statins have the same proportional effect on people at low and high risk []. Because of this, assessment of predicted risk of CVD has been a fundamental part of national clinical guidance on CVD prevention in primary care since []. In the UK, an absolute risk of CVD greater than a threshold was, and remains, a criterion for offering statin treatment and, in combination with raised blood pressure, a criterion for antihypertensive treatment [-]. Risk assessment is also embedded in clinical guidelines around the world including New Zealand, Europe and America [-0]. A variety of different risk calculators are available and whilst QRISK [] is the calculator of choice in England and Wales, elsewhere, others such as Framingham [], Score [], ASSIGN [] or Procam [] are recommended. Statins are the only class of lipid lowering medications recommended for use in UK primary care [] and are the most widely prescribed medication in England []. A small proportion of patients in England have to pay a prescription charge for their medications but the majority receive their medication free []. Statins are used to lower lipids as part of a CVD primary prevention strategy. As estimated CVD risk is the best predictor of likely benefits of treatment this information is necessary for effective shared decision making when considering the use of statins []. Evidence - : first published as 0./bmjopen-0-00 on November 0. Downloaded from on April 0 by guest. Protected by copyright.

19 Page of suggests that utilisation of risk scoring improves accuracy of perceived CVD risk and medical prescribing without causing harms [-0]. In the context of primary prevention there is evidence both of undertreatment of high risk patients (above the treatment threshold) and inappropriate treatment of low risk patients (below the treatment threshold) [-]. Moreover, patients with the highest absolute CVD risk are at greatest risk of being undertreated []. One reason for this is that risk scores may not be consistently used by clinicians to guide decision making []. Reasons for not using risk score calculators include a belief that risk scoring oversimplifies the decision and may lead to over prescribing [] and confusion over how best to use risk score calculators []. Additionally, one fundamental barrier to using risk scores is a focus on individual risk factors over absolute risk score [, ]. It may appear logical that decisions about initiating cholesterol lowering medication should be based on the patient s cholesterol level, and indeed this was the preferred approach in the pre-cvd risk era [0, ]. However, this approach would not necessarily result in the greatest benefits and is not justifiable based on the available evidence. Despite this, there is evidence that this approach prevails in some areas. After adjusting for other CVD risk factors, lipid profile has been found to predict statin initiation in a US study [] and a randomised theoretical experiment in Australia highlighted clinicians preferences for managing individual risk factors over absolute risk []. To further compound the problem, patients find it difficult to make decisions based on future risks and tend to preferentially focus on cholesterol levels when considering taking statins [, ]. In 0, guidelines for England and Wales on lipid modification reduced the predicted risk threshold at which patients should be offered statins from a 0% 0-year risk to 0% [, ]. This recommendation was met with considerable concern from the medical community that it would lead to widespread over prescribing of statins with little clinical benefit but considerable potential for harm [, ]. This widely publicised criticism may have increased concerns that risk scoring - : first published as 0./bmjopen-0-00 on November 0. Downloaded from on April 0 by guest. Protected by copyright.

20 Page of leads to oversimplification of clinical decisions and overprescribing and may have, perversely, decreased the use of risk scoring and reinforced the use of lipid levels to guide decision making. There is some evidence to suggest there was a decrease in statin initiations around this time due to the negative media attention []. This research will investigate the relationship between lipid profile, QRISK score and subsequent initiation of statins by primary care clinicians. This will give an indication of the patient factors that are influencing the decisions clinicians make in routine practice when considering the use of statins and how this has changed in response to external factors. The consistent use of risk scoring can result in more targeted prescribing and fewer patients being prescribed medication overall [] which maximises the efficiency and cost effectiveness of treatment at a population level. Aims and Objectives Aim: To establish whether UK primary care patients are initiated on statins based more on their lipid levels or on their predicted CVD risk and if this has changed over time. Objectives: For primary prevention of CVD. To evaluate how lipid levels and CVD risk independently predict subsequent prescribing of statins. To determine if the relationship between lipid levels, QRISK score and statin prescribing has changed following recent guideline changes. - : first published as 0./bmjopen-0-00 on November 0. Downloaded from on April 0 by guest. Protected by copyright.

21 Page of To investigate the contribution that calculating a QRISK score has on the probability of initiating statin therapy.. To investigate inter-prescriber variation when initiating statins. Methods and analysis Study design A historical open cohort study of patients who were eligible for statin therapy for the primary prevention of CVD. Data source Data will be extracted from anonymised primary care records from practices in England and Wales who contribute to The Health Improvement Network (THIN) database. The THIN database contains routine patient data from over 00 UK general practices which are generalisable to the UK population [0]. THIN has been used in previous studies to validate QRISK and it was shown that the discrimination statistics in THIN are as good as those for the original QRISK cohort [, ]. Practices that contribute to THIN use the Vision (In Practice Systems) electronic patient records system. Clinical data are coded using Read code clinical classification version [] and drug codes are used which correspond to the British National Formulary (BNF) []. The clinical codes used are available in appendix I and the drug codes are available in appendix II. All data are anonymised and will be stored on secure severs at the University of Birmingham and accessed only through password protected systems. Population Patients will be eligible for inclusion if they are aged between 0 and years inclusive (the age range recommended by NICE for risk assessment []) with no previous diagnosis of CVD and at least one lipid result coded in the study period. The date of the first coded lipid result in the study period - : first published as 0./bmjopen-0-00 on November 0. Downloaded from on April 0 by guest. Protected by copyright.

22 Page of will be considered the index date. Patients must not have been prescribed a statin prior to the index date. In order to maintain a stable cohort of practices, only THIN practices in which there are data available for the whole study period will be included. The study period will run from the beginning of 0 to the end of 0. Patients will be eligible for inclusion from the earliest of the following dates: study start date, acceptable mortality reporting date (which ensures that patient deaths and deregistrations are being recorded consistently []), Vision date plus one year, registration date plus one year (to ensure time for baseline data to be recorded), and age 0; until the earliest of the following dates: age, study end date, CVD diagnosis, other excluding diagnosis (as below), statin initiation, or recording of a contraindication for the prescribing of statins. Patients will also be excluded if they left the database within 0 days of the index date. Patients will be excluded if they had pre-existing CVD, were pregnant, had a contraindication to statin therapy (including allergies and drug interactions), were already being prescribed a statin (or other lipid lowering therapy) or had ever been prescribed a statin in the past. Patients will be considered to have pre-existing CVD if they had a diagnosis of myocardial infarction, angina, peripheral vascular disease or stroke/tia. Patients with type diabetes mellitus, a diagnosis of chronic kidney disease stages - and patients diagnosed with or who were at risk of familial hypercholesterolemia will also be excluded as statin use in these patients should not be based on CVD risk assessment. Patients at risk of familial hypercholesterolemia will defined as those with a total cholesterol of.0 mmol/l as this is the threshold for seeking specialist assessment defined in the NICE guidance []. - : first published as 0./bmjopen-0-00 on November 0. Downloaded from on April 0 by guest. Protected by copyright.

23 Page of Study variables Exposure variables The exposure of interest is the undertaking of lipid testing. Once a patient has their lipids recorded this signifies that a clinician has made a decision about the significance of the lipid levels and will act on this. Individual patients may have multiple lipid results recorded over the study period. Assuming the exclusion criteria are not met, each lipid result will be considered as a new exposure so long as it is not within 0 days of the last recorded lipid level. The number of previous lipid results will be included in predictive modelling. Outcome variables The outcome will be the prescription of a statin within 0 days of a set of lipid results. Predictor variables Sociodemographic details will be obtained for each patient: age, sex, ethnicity, Townsend quintile and rurality indicator (see appendix III). General practice and clinician ID will also be recorded. Patient variables will include the cardiovascular risk factors required to calculate QRISK; age, sex, ethnicity, postcode, smoking status, diabetes status, family history of CVD, CKD (stage or ), atrial fibrillation, hypertension treatment, rheumatoid arthritis, lipids, blood pressure and body mass index (BMI). Other clinical data that may affect the decision to prescribe statins will also be collected comprising of; liver transaminase levels, systemic inflammatory conditions, severe mental health conditions, and HIV. Data on medication prescriptions will be extracted. Patients who received a prescription for antihypertensive medication up to 0 days prior to the index date will be considered to have been on treatment for hypertension. Prescriptions of antipsychotics, corticosteroids and immunosuppressant medication will be recorded as these may change the threshold for prescribing lipid lowering medication. - : first published as 0./bmjopen-0-00 on November 0. Downloaded from on April 0 by guest. Protected by copyright.

24 Page of All included patients will have a QRISK score calculated post hoc from the available data. In addition, any coded QRISK score prior to and including the follow up period will be extracted. The date when predictor variable data are collected will vary depending on whether or not the patient has a QRISK score coded during the follow up period. For patients with a coded QRISK score on the index date or in the follow up period; the date of QRISK recording will be used as the date for collecting data on predictor variables and the most recent data (including the date of QRISK coding) will be recorded. If more than one QRISK score is coded in the follow up period, the latest date will be used for data collection to account for any variables that were updated during the follow up period. For patients who do not have a coded QRISK score; the included data will be the most recent recorded before and including the end of the 0 day follow up period. Missing data and extreme values Missing data is anticipated in some variables; notably BMI, BP and smoking status. If comorbidities are missing they will be assumed to be absent. Missing values arise twice; first when calculating the QRISK score post hoc, and secondly when constructing the multivariable model. These sitations will be accounted for differently. Where data is missing from variables used in QRISK calculations, the imputed values used by the QRISK calculator will be used []. In the multivariable model, all variables will be categorical and a missing category will be used to account for the possibility that missing data is associated with the initiation of statins. Biologically implausible values for blood pressure, lipids, liver transaminases, height and weight will be identified and excluded. Analysis Continuous variables will be categorised. Calculated and coded QRISK scores will be categorised according to risk bands (<0%, 0-.%, -.% and 0%). Age will be categorised into four groups (0-, 0-, 0- and 0). Total cholesterol will be categorised into three bands (<.0, - : first published as 0./bmjopen-0-00 on November 0. Downloaded from on April 0 by guest. Protected by copyright.

25 Page 0 of and.0 mmol/l), and HDL cholesterol into four bands (.,.-.,.-. and. mmols/l) as these cut-offs have been used previously in the literature []. Systolic blood pressure will be split into three groups corresponding to normotensive, stage hypertension and stage hypertension (<0, 0- and 0 mmhg) []. The cohort will be described and prescribing rates and QRISK score trends will be established. Table : Variables used in predictive modelling Variable Categories Total Cholesterol <.0,.0-.,.0 TC/HDL ratio.,.-.,.-.,. Prior cholesterol measurements 0, -, -, Calculated QRISK <0%, 0-.%, -.%, 0% Coded QRISK <0%, 0-.%, -.%, 0%, missing Systolic BP <0, 0-, 0, missing Age 0-, 0-, 0-, 0 Sex Male, female Ethnicity White, Asian, Black, Other, missing Townsend deprivation quintile st, nd, rd, th, th, missing Urban/rural score Urban, rural, missing BMI <0, 0-., -., 0-., +, missing Smoking status Current smoker, ex-smoker, non-smoker, missing Liver transaminases Normal, abnormal, missing Diabetes Mellitus Yes, no Family History of CVD Yes, no Atrial Fibrillation Yes, no Rheumatoid Arthritis Yes, no Hypertension Yes, no Chronic Kidney Disease Yes, no HIV Yes, no - : first published as 0./bmjopen-0-00 on November 0. Downloaded from on April 0 by guest. Protected by copyright.

26 Page of Severe enduring mental illness * Yes, no Inflammatory conditions ** Yes, no Antipsychotic medication Yes, no Long term corticosteroids Yes, no Immunosuppressant medication Yes, no Year quartile st, nd, rd, th *schizophrenia, bipolar affective disorder, mania or psychosis (unspecified), psychotic depression **systemic lupus erythematosus, scleroderma, Bechet s syndrome, other inflammatory arthritis The primary analysis will consist of predictive modelling using multivariate logistic regression (variables summarised in table ). Clustering of responses within individual GP practices will be included in the model. Variation throughout the year due to the seasonality imposed by QOF targets will be accounted for by adding the year quarter as a variable in the model. Assuming a minimum sample size for analysis of 0 per degree of freedom [], and degrees of freedom in the variables, a minimum sample of 0 patients with a lipid profile and subsequent statin initiation should easily be obtained. The potential cohort size can be estimated from the literature. Feasibility data from the THIN database identified over 0,000 patients over two years who were eligible for statin treatment for the primary prevention of CVD and had a lipid result and blood pressure coded on their medical record []. Over a five year period we should expect to identify over double this number. Another cohort identified just over million patients in the THIN database in the age range where NHS health checks are offered []. Given that every patient who attends an NHS health check should have a lipid test undertaken and 0% of 0- years olds are invited to have a health check each year, with approximately 0% taking up this offer []; This would generate around 00,000 coded lipid results in the target population annually. The accuracy of the post hoc QRISK calculations will be tested through comparisons with corresponding coded QRISK scores in patients where these are available. - : first published as 0./bmjopen-0-00 on November 0. Downloaded from on April 0 by guest. Protected by copyright.