Background
This chapter consists of results from modelling the impact different monitoring strategies might have on clinical decisions on statin use. Current UK guidelines recommend that primary prevention patients at high risk of CVD should be prescribed statins.2,56 At the time of writing the project protocol, high CVD risk was defined as a 10-year CVD risk score of ≥ 20%, and the recommended risk scores, such as QRisk2,52 ASSIGN54 and Framingham 199153 all include the TC/HDL cholesterol ratio as a factor. In 2014, guidelines56 changed the definition of high risk to a 10-year CVD risk score of ≥ 10%, calculated using QRisk2.52 Statin therapy is recommended for all secondary prevention patients; guidelines published in 2014 recommend that patients are prescribed 80 mg of atorvastatin, whereas earlier guidelines recommended a lower initiating dose, and that an increased dose should be considered when TC levels of > 4 mmol/l and/or LDL cholesterol levels of > 2 mmol/l are measured.2 Thus, UK recommendations for changes to statin treatment are, at least partially, based upon measured lipid levels. However, as illustrated in Chapter 4, the estimated variation within each lipid measurement is greater than the likely change over 1 year, thus, monitoring at short intervals promotes unnecessary pharmacological treatment with high likelihood. Possible implications are labelling people with ‘normal’ or lower lipid levels as ‘sick’, complacency with regards to lifestyle changes and self-care, and lower adherence to medication later in life when increased age places patients at higher CVD risk and medication adherence may be more beneficial. The key concern of false-negative tests is that patients are receiving lower doses than those recommended, potentially leading to a raised risk of CVD. We applied established methods206,207 to data from the CPRD to estimate the rates of true and false test results, and examine the optimal interval for monitoring dyslipidaemia in primary and secondary prevention patients.
Methods
Model structure
We characterised a population of 100,000 men and 100,000 women for each prevention group (primary, primary on statins, and secondary on statins) through sampling with replacement from the CPRD. The populations were generated with the same distribution of observed lipid measures and baseline cardiovascular risk factors as patients in that prevention group in our CPRD data. Risk factors included those used to calculate 10-year CVD risk using QRisk252 and Framingham 1991,53 and were based on individuals without a history of familial hypercholesterolaemia who had any one lipid measurement (TC, HDL cholesterol, LDL cholesterol or TGs) recorded in 2006, the most recent complete year of data in our sample, for which the minimum age was 40 years for men and 55 years for women. This patient group was selected to represent those attending lipid monitoring. Prevention groups were defined as described in Chapter 4. CVD risk factors were defined in reference to the baseline date, which was the date of the first lipid measurement in 2006. Lipid measurements recorded on this date were used as the baseline observed readings. Age at baseline was defined from year of birth and the date of first lipid measurement. Each condition from diabetes, atrial fibrillation, chronic kidney disease, rheumatoid arthritis, use of BP medication, a family history of angina or heart attack in a first-degree relative of < 60 years of age or left-ventricular hypertrophy, was assumed to be present if there was a medical code relating to the condition in patient medical records up to 1 year after the baseline date. The first measurement recorded up to 1 year after baseline was used for baseline smoking status, BMI and systolic BP. Missing data for TC, HDL, LDL, TGs, BMI, systolic BP and smoking status were imputed using the multivariate chained equations method; lipid values and BMI were log transformed prior to imputing.208 Models assumed that non-lipid CVD risk factors remain constant over time. Separate simulation analyses were run for each lipid, prevention group and gender.
Estimates of lipid variability and the between- and within person trends, derived in UK data in Chapter 4, were used to simulate lipid progression over time (see Tables 23 and 24). The ‘true’ lipid reading for each patient at baseline adjusted for age at measurement, assumed to be unobservable to the patient or clinician, was simulated conditional on the initial observed value, and then an individual annual rate of change was simulated (see Appendix 19). The simulated true lipid reading for each patient at each year of follow-up is then calculated using Equation 1 (see Chapter 4). The observed lipid value in any given sample is simulated from their true value at that time and the model estimate of the within-measurement variability of the lipid (see Equation 2, Chapter 4). For primary prevention patients, the true and observed 10-year CVD risk scores at each time point were also calculated.
The proportion of recommended treatment changes at each time point that are attributable to dyslipidaemia (true-positive results) can be estimated through calculating the proportion of changes for which both the simulated true and observed lipid values in secondary prevention patients, or both true and observed 10-year CVD risk scores in primary prevention patients, were above threshold; likewise the proportion attributable to within-person variability (false-positive results) can be estimated from the proportion for which the simulated observed lipid value/10-year CVD risk on which the decision was based was above threshold, while the true lipid level/10-year CVD risk was below threshold. Similarly, we can estimate the proportions of patients who correctly (true-negative) or incorrectly (false-negative) tested negative. The calculation of these proportions is illustrated in . For CIs, we repeated the simulations changing the model parameters to plausible values given the data, obtained through non-parametric bootstrapping. Random samples of 100,000 patients were drawn with replacement, 100 times from the simulated data set described above; for each, the simulation was then run 50 times and the average proportions of true-positive, false-positive, true-negative and false-negative treatment decisions were calculated. The SD across the 100 averages is used to estimate the SE and calculate CIs.
Proportions of tests that are false
For each patient group, each lipid and both genders, we carried out simulation modelling for annual monitoring, biennial monitoring (every 2 years), triennial monitoring (every 3 years) and quinquennial monitoring (every 5 years). We first consider the proportions of false-positive and false-negative treatment decisions for strategies based on guidelines current, at the time of analysis,2 for which primary prevention treatment guidelines use CVD risk thresholds calculated using QRisk252 and secondary prevention uses lipid thresholds; we then consider how rates might be affected by monitoring using different lipids, different CVD risk scores and thresholds, and different patterns of measurement frequency.
Scenarios based on current guidelines
Primary prevention patients not taking statins
Our simulation assumed that monitoring would follow NICE guidance2 as follows:
Lipid screening for primary prevention patients begins at 40 years for men or 55 years for women.
A 10-year CVD risk score for the observed TC/HDL cholesterol ratio is calculated using the QRisk2
52 algorithm.
An observed risk of ≥ 20% classes a patient as high risk.
Patients not taking statins are eligible for a statin prescription if high risk.
Patients not eligible for a treatment change continue to be monitored and can hence be eligible for a statin prescription in subsequent years.
Therefore, a male aged ≥ 40 years and a female aged ≥ 55 years in the simulated primary prevention population are assumed to be eligible for a statin prescription if their 10-year CVD risk calculated from their observed lipid value is above the 20% threshold. We then classify a treatment change as a ‘true-positive’ if the calculated 10-year CVD risk based on the true lipid reading is above threshold, or a ‘false-positive’ if the calculated CVD risk based on true lipid reading is below threshold (as illustrated in ). Similarly, when a patient is not eligible for a statin prescription because the observed lipid gave a 10-year CVD risk below threshold, we describe this as a ‘true-negative’ if their CVD risk calculated from their true lipid value is below threshold, and ‘false-negative’ if the true CVD risk is above threshold. Patients who are classed as a false-positive can later become true-positives if their ‘true’ CVD risk goes over the threshold. Patients who are true-positives remain classed as true-positives.
Primary prevention patients taking statins
Although NICE guidance suggests that monitoring in primary prevention patients already taking statins is not necessary,2 we examined the consequences of monitoring this group as described above, except that patients whose observed 10-year CVD risk was over threshold were eligible to be prescribed an intensified dose of statin.
Secondary prevention patients taking statins
Total and LDL cholesterol levels rather than CVD risk are used to guide treatment for secondary prevention patients.2 The progression of TC and LDL were considered separately. Our simulation assumed that patients:
were already taking a statin
are eligible for treatment intensification if the modelled lipid reading was over recommended thresholds (4 mmol/l for TC, 2 mmol/l for LDL)
not eligible for an up-titration continue to be monitored and can hence be eligible to be prescribed a higher dose in subsequent years.
Therefore, a patient (male aged ≥ 40 years or female aged ≥ 55 years) in the simulated population is assumed to be eligible to be prescribed an intensified dose if the simulated observed lipid (either TC or LDL) was above threshold, (4 or 2 mmol/l, respectively). The proportions of true and false treatment decisions were calculated as previously described using the true and observed lipid values.
Scenarios based on alternative strategies for monitoring lipids
Framingham 1991 as an alternative 10-year cardiovascular risk score
Until 2014, NICE guidelines2 did not name a specific risk score for calculating 10-year CVD risk, allowing health-care practitioners to use a method best suited to their requirements, such as QRisk2,52 Framingham 199153 or ASSIGN,54 all of which are based upon TC and HDL. Therefore, we carried out simulation modelling for primary prevention men and women not taking statins using the Framingham 1991 risks score,53 on which the Joint British Societies risk charts10 are based, to guide treatment decisions.
Lowering the cardiovascular risk threshold for primary prevention patients
As noted in Chapter 1, NICE guidelines were being reviewed throughout the duration of this project,62 and it was thought that the risk threshold for primary prevention patients would be lowered. We therefore carried out simulation modelling using both 15% and 10% CVD risk thresholds, calculated using QRisk2.52 In July 2014, revised guidelines lowered the treatment threshold to 10%.
Alternative lipid measures
We considered how results would vary if different lipid measures were used in monitoring. Based on the results from Chapter 2, we focused on TC, LDL and HDL only. For primary prevention patients, QRisk252 was edited, to use TC, LDL or LDL/HDL cholesterol instead of TC/HDL cholesterol to calculate a 10-year CVD risk. Simulation models were then run as described above for each lipid type, using a treatment eligibility threshold of 20%.
For secondary prevention patients we also examined monitoring using the ratios of TC/HDL cholesterol and LDL/HDL cholesterol. Thresholds for these lipid measures were selected from existing guidelines as far as possible: summarises the rationale for each selected threshold.
Thresholds by lipid measure for secondary prevention patients
Alternative strategies for lipid measurement
Further simulation modelling was carried out, in which treatment eligibility decisions were based upon the average of repeated lipid measurements taken over a short time period; this strategy aimed to improve the accuracy of the lipid measurement on which treatment was based. Such repeated measurements could either be taken for all patients or limited to those whose first test is ‘near’ the NICE thresholds, for example ± x mmol/l from threshold; however, given the high variability in lipid levels, as described in Chapter 4, it was decided to model this repetition for all secondary prevention patients taking statins and primary prevention patients taking statins. The proportions of true- and false-positive treatment eligibility decisions under three possible alternative strategies were compared:
five repeat tests are used to calculate an average lipid level every 5 years
three repeat tests are averaged every 3 years, and
three repeat tests are averaged every 5 years.
Repeat tests were modelled at 1-month intervals and the mean observed lipid level was calculated. Proportions of true and false treatment decisions were then calculated as described above.
A further alternative monitoring strategy might be to update a subject’s risk score annually using a previous lipid measurement. We therefore also carried out additional simulations in the primary prevention population not taking statins, where the lipid reading used to calculate the 10-year CVD risk score was measured (1) once every 3 years, and (2) once every 5 years, but the risk score was calculated annually using the most recent lipid measure, risk factor data from baseline, and current – i.e. updated – age.
Amendments to protocol
The protocol initially specified that the parameters obtained from models fitted in the St Luke’s cohort and then recalibrated to the UK population using CPRD data would be used to examine clinical decisions on eligibility for treatment. However, as explained in Chapter 4 (see Discussion) and Appendix 17, estimates of the within-person short-term variability were unexpectedly smaller in the St Luke’s cohort than in the UK cohort, and, after comparison with external data from the literature, we proposed to the advisory board a protocol modification to use the direct CPRD estimates in our simulation models, as this adjustment meant that our models were more consistent with those reported in previous literature. This chapter therefore presents results based upon parameters estimated from the CPRD data; however, for completeness, Appendix 18 presents methods and results according to the original protocol.
Our protocol specified that calculations of 10-year CVD risk would be based upon the Framingham 1991 risk score;53 however; QRisk252 has been increasingly favoured in CVD prevention guidelines as the risk score of choice. The annual update to QRisk252 in 2012 allowed calculation of risk over periods of 1–10 years, thus facilitating its incorporation into health-economic models (presented in Chapter 7). Therefore, with the approval of our advisory board, QRisk252 was used as the reference CVD risk score in Chapters 5 and 7, and in Appendix 18 we also present analyses modelling treatment decisions based upon the Framingham risk score as originally specified.
Results
The baseline characteristics for the simulated data based on CPRD patients are shown for each gender/prevention group in .
Baseline characteristics of the simulated data generated with the distribution of the CPRD cohorts: N = 100,000 in each group
Primary prevention not taking statins
Results of modelling the rate of eligibility for statin prescriptions for males and females in this group are shown in and and . shows the percentage of tests that are false-positives (left) and false-negatives (right) in men and women, for annual, biennial, triennial and quinquennial monitoring strategies over 15 years of follow-up. More frequent monitoring resulted in a higher proportion of patients incorrectly recommended for a statin prescription (left). The proportion of patients incorrectly recommended for a statin prescription decreased over time as patients moved from being classed as false-positive to true-positive (as their underlying CVD risk increased). The proportion of patients incorrectly not recommended for a statin was higher when monitoring occurred at longer intervals. Over follow-up, there was a gradual increase in the proportion of false-negative tests for each monitoring interval. As all non-lipid risk factors other than age were held constant and, on average, the change in lipid levels was approximately zero (see Tables 23 and 24), over time, an increasing proportion of the population is classed as at high CVD risk as a result of population ageing, and therefore an increasing proportion of patients who should be eligible for treatment are missed.
Men and women, primary prevention not taking statins, QRisk2 (TC/HDL cholesterol) using ≥ 20% as threshold. Model estimates for the percentage of changes in treatment that are attributable to the within-measurement variability (false-positive) (more...)
Primary prevention men not taking statins: TC/HDL cholesterol ratio, QRisk2 20%
Primary prevention women not taking statins: TC/HDL cholesterol ratio, QRisk2 20%
and show the cumulative number eligible for statin prescriptions, the percentage of those that are false-positives, and the percentage of those repeatedly testing negative that are false-negative. Each outcome is shown for annual, biennial, triennial and quinquennial monitoring strategies. Column 1 of shows the modelled rate of eligibility for statin prescriptions per 1000 men in the first year after monitoring begins, and the cumulative rate of eligibility for statin prescriptions over the subsequent years assuming 1-, 2-, 3- and 5-yearly monitoring. For example, over the first 6 years of follow-up, annual monitoring would result in 610 per 1000 men being eligible for a statin prescription, whereas biennial and triennial monitoring would result in 601 per 1000 men and 597 per 1000 men eligible for prescriptions, respectively. Over 15 years, annual, triennial and quinquennial monitoring would result in 823, 813 and 811 per 1000 men eligible for prescriptions. Column 2 of shows the estimated proportion of individuals who are eligible for statin prescriptions that are incorrectly treated (false-positives), as a percentage of the total number of men who are eligible for a statin prescription, if monitoring is conducted every 1, 2, 3 and 5 years. The percentage of true-positives can be obtained by subtracting the false-positives from 100%. At the first monitoring test, 3.5% of recommended treatment changes can be attributed to variability in the TC/HDL cholesterol measurements (false-positive), whereas 96.5% of recommended statin prescriptions would be attributable to dyslipidaemia (true-positives). This corresponds to about 15 per 1000 men (422 per 1000 prescriptions × 3.5% false) recommended for treatment unnecessarily. At 6 years after first measurement, we estimate that the proportion of false-positive prescriptions under annual, biennial and triennial monitoring would be 4.5%, 3.5% and 3.0%, respectively (corresponding to about 27, 21 and 18 per 1000 men being over-treated), and by 15 years, under annual, triennial and quinquennial monitoring, 2.5%, 1.7% and 1.5% respectively (corresponding to about 21, 14 and 12 men per 1000). The estimated proportion of patients who were incorrectly not recommended for a statin (false-negatives) over follow-up under annual, biennial, triennial and quinquennial monitoring is shown in column 3. At first test 3% of men are incorrectly not recommended for a statin, corresponding to 17 men per 1000 (578 men are not treated × 3.0% false). At 6 years, the proportions of men incorrectly not recommended for a statin are 1.6%, 3.0% or 3.7% when monitoring intervals of 1, 2 or 3 years, respectively. These proportions correspond to 6, 12 or 15 men per 1000. Similar results for women are shown in .
Primary prevention taking statins
Corresponding estimates for modelling the rate of eligibility for an intensified dose of statin under different monitoring schedules in primary prevention patients taking statins are shown in and (men) and (women). Similar trends were seen to primary prevention patients not taking statins.
Men and women, primary prevention taking statins, QRisk2 (TC/HDL cholesterol) using ≥ 20% as threshold. Model estimates for the percentage of changes in treatment that are attributable to the within-measurement variability (false-positive) (more...)
Primary prevention men taking statins: TC/HDL cholesterol ratio, QRisk2 20%
Primary prevention women taking statins: TC/HDL cholesterol ratio, QRisk2 20%
Secondary prevention taking statins
and and show corresponding estimates in secondary prevention patients taking statins when eligibility for treatment change to an intensified dose of statin was based upon a TC threshold. Estimates based upon LDL measurements are presented in and and . The proportions of both false-positive and false-negative decisions with regards to eligibility for treatment were relatively large. The proportion of patients recommended for up-titration unnecessarily (false-positives) was greater when the monitoring interval was shorter. After 3 years of follow-up, 27.1% of men recommended for a higher statin dose were up-titrated unnecessarily when monitoring every year using a TC threshold of 4 mmol/l compared with 21.9% when monitoring at 3-year intervals. This corresponds to 224 per 1000 men and 157 per 1000 men, respectively (825 per 1000 up-titrations × 27.1% false and 718 per 1000 men × 21.9% false). In women, using a TC threshold of 4 mmol/l, the equivalent proportions recommended for unnecessary up-titration were 14.6% and 12.2%, corresponding to 136 per 1000 women and 105 per 1000 women recommended for unnecessary treatment under annual and triennial monitoring, respectively. In contrast, more frequent monitoring resulted in a lower proportion of patients eligible for treatment being missed. At 3 years since first test, the proportion of under-treated men was 3.7% under annual monitoring using a TC threshold of 4 mmol/l, and 12.7% under triennial monitoring, corresponding to 6 per 1000 men and 36 per 1000 men, respectively. The equivalent proportions in women were 7.9% and 24.0% or 5 per 1000 women and 33 per 1000 women under annual and triennial monitoring, respectively. When each of TC and LDL levels in secondary prevention patients were first measured, approximately one-quarter to one-third of men and almost half of women were estimated to be recommended too low a dose of statin; the proportion of patients recommended too low a treatment dose decreased over follow-up. However, the high numbers of patients who test over threshold, and are therefore eligible for an up-titration, mean this corresponds to between 111 per 1000 (men, TC 4 mmol/l) and 134 per 1000 (women, LDL 2 mmol/l) patients being under-treated at first test, dropping to between 35 to 36 per 1000 (men, TC 4 mmol/l) and 38 and 42 per 1000 (women, LDL 2 mmol/l) at second test (depending on the monitoring interval).
Men and women, secondary prevention taking statins, using TC ≥ 4 mmol/l as threshold. Model estimates for the percentage of changes in treatment that are attributable to the within-measurement variability (false-positive) rather (more...)
Secondary prevention men taking statins: TC 4 mmol/l
Secondary prevention women taking statins: TC 4 mmol/l
Men and women, secondary prevention taking statins, using LDL ≥ 2 mmol/l as threshold. Model estimates for the percentage of changes in treatment that are attributable to the within-measurement variability (false-positive) rather (more...)
Secondary prevention men taking statins: LDL 2 mmol/l
Secondary prevention women taking statins: LDL 2 mmol/l
Alternative strategies for monitoring
Lowering the CVD risk threshold for primary prevention patients
We evaluated the impact of a reduction on the CVD risk threshold to 15% and 10%. Prescription eligibility rates and estimates of the proportion of these that are unnecessary (false-positives) for monitoring in primary prevention men and women not currently taking statins and based on a CVD risk threshold of 15% and 10% are shown in and . Trends were similar to those seen with a 20% CVD risk threshold, with higher proportion of false-positive eligibility decisions and lower proportions of false-negative eligibility decisions when monitoring is more frequent, although the proportion of false-positive eligibility decisions decreased as the threshold lowered, whereas the proportion of false-negative treatment eligibility decisions was higher.
Men and women, primary prevention not taking statins, QRisk2 (TC/HDL cholesterol) using ≥ 15% as threshold. Model estimates for the percentage of changes in treatment that are attributable to the within-measurement variability (false-positive) (more...)
Men and women, primary prevention not taking statins, QRisk2 (TC/HDL cholesterol) using ≥ 10% as threshold. Model estimates for the percentage of changes in treatment that are attributable to the within-measurement variability (false-positive) (more...)
Alternative cardiovascular risk measures
Corresponding figures for the following alternative monitoring strategies listed below are presented in Appendix 20:
Primary prevention not taking statins using Framingham 1991 as risk score.
Primary prevention not taking statins using modified QRisk2
52 based on TC, LDL and LDL/HDL cholesterol instead of TC/HDL cholesterol.
Secondary prevention using the ratio of TC/HDL cholesterol instead of TC, and LDL/HDL cholesterol instead of LDL.
Similar trends as described above were seen in all the alternative scenarios.
Alternative strategies for lipid measurement: updating on single or multiple measures
In our primary analysis (above) we assumed that, in primary prevention, if lipids are measured less than annually then CVD risk would also be estimated less than annually (at each lipid measurement). For completeness we now consider strategies in which lipids are measured every 3 or 5 years, but CVD risk is nonetheless recalculated annually, using the most recent available lipid measurement and current age. and compare these strategies with our primary analysis (CVD risk estimated only when lipid is measured). Periodic measurement with annual calculation of CVD risk has led to a non-monotonic trend in estimates of the patients both over- and under-treated, although this is likely an artefact of the simulation, as in practice subjects would not be measured at the same time point. Proportions of false-positives (unnecessary changes to treatment) were higher when the alternative strategy guided treatment than when monitoring occurred at longer intervals but were lower than under annual monitoring. Higher proportions of false-negative tests (undertreatment) were seen when monitoring at longer intervals than when using the alternative strategy; annual monitoring gave the lowest proportion of false-negatives.
Men and women, primary prevention not taking statins, QRisk2 (TC/HDL cholesterol) using ≥ 20% as threshold, Recalculating QRisk2 yearly based on measures every 3 years. Model estimates for the percentage of changes in treatment that are (more...)
Men and women, primary prevention not taking statins, QRisk2 (TC/HDL cholesterol) using ≥ 20% as threshold, recalculating QRisk2 yearly based on measures every 5 years. Model estimates for the percentage of changes in treatment that are (more...)
We explored how using the average of multiple lipid measures could impact on the proportions of incorrect eligibility decisions amongst both primary and secondary prevention populations taking statins. Estimates comparing this alternative monitoring strategy with annual measurement are presented in –. Similar trends were seen in both prevention groups. The proportion of incorrect treatment eligibility decisions was smaller when decisions were based upon the average of a larger number of repeat measurements, with a lower proportion of both false-positive and false-negative decisions when the average of five repeated measurements was used to guide treatment, although annual monitoring with only one measurement still gave a lower proportion of under-treated patients. When strategies with assumed similar total cost were compared (see Appendix 20, –) for primary and secondary patients on statins, respectively, more repeated measurements with a longer interval gave a lower proportion of false-positive eligibility decisions, and estimates of the proportion of false-negative eligibility decisions were similar.
Men and women, primary prevention taking statins, QRisk2 (TC/HDL cholesterol) using ≥ 20% as threshold, impact of using the average of three measurements. Model estimates for the percentage of changes in treatment that are attributable (more...)
Men and women, secondary prevention taking using LDL ≥ 2 mmol/l as threshold, impact of using the average of five measurements. Model estimates for the percentage of changes in treatment that are attributable to the within-measurement (more...)
Men and women, primary prevention taking statins, QRisk2 (TC/HDL cholesterol) using ≥ 20% as threshold, impact of using the average of five measurements. Model estimates for the percentage of changes in treatment that are attributable (more...)
Men and women, secondary prevention taking statins using TC ≥ 4 mmol/l as threshold, impact of using the average of three measurements. Model estimates for the percentage of changes in treatment that are attributable to the within-measurement (more...)
Men and women, secondary prevention taking statins using TC ≥ 4 mmol/l as threshold, impact of using the average of five measurements. Model estimates for the percentage of changes in treatment that are attributable to the within-measurement (more...)
Men and women, secondary prevention taking statins using LDL ≥ 2 mmol/l as threshold, impact of using the average of three measurements. Model estimates for the percentage of changes in treatment that are attributable to the within-measurement (more...)
Discussion
Using parameters presented in Chapter 4, we modelled the rate of eligibility for statin prescriptions, quantifying the proportions of incorrect treatment eligibility decisions when monitoring at intervals of 1, 2, 3 and 5 years. The high variability in each lipid measurement resulted in recommendations for unnecessary initiation or up-titration (false-positive ) and undertreatment (false-negative) in all patient groups and under all scenarios considered. Consistent with previous research on long-term chronic conditions such as hypertension, renal function and hypercholesterolaemia,30,204,206,207 we found that more frequent monitoring resulted in a higher proportion of false-positive tests, which, in these analyses, corresponds to a higher proportion of patients incorrectly recommended for a statin prescription or an up-titration. This is attributable to the low true rate of change over time. Conversely, the proportion of patients incorrectly not recommended for a statin or an increased dose (false-negative results) was lower when monitoring at shorter intervals. However, given the comparatively high numbers of patients over threshold and hence recommended for treatment (in both primary and secondary prevention groups), fewer absolute numbers of patients were undertreated as opposed to unnecessarily recommended for an up-titration.
Given that such lipid measurements guide clinical decisions, this may be detrimental to a patient’s health, and alternative monitoring strategies that aim to estimate the underlying lipid levels more accurately may be preferable. Previous research examining optimal screening intervals for cholesterol has not examined the proportion of under-treated patients.30,31 As statin treatment has been shown to reduce CVD risk, even when patients have low lipid levels, whether or not potential overtreatment in these patients is detrimental requires a full health economics analysis that can account for the potential harms of over and undertreatment; this is the focus of Chapter 7.
Updated guidelines recommended lowering the CVD risk threshold for initiation of statin treatment in primary prevention patients to 10%. As we lower the treatment threshold, a greater proportion of the population will have a true CVD risk that is over the threshold, and this means that more negative test results are likely to be false. However, as the proportion of patients testing positive will increase, fewer patients will therefore test negative. For example, when monitoring every year or 3 years, a threshold of 20% led to 1.6% or 3.7% of men incorrectly not recommended for treatment at 6 years since first test, whereas a threshold of 10% led to 3.1% and 5.4%, respectively. These proportions correspond to 6 per 1000 and 15 per 1000 men incorrectly not recommended for treatment with a 20% threshold, and 5 per 1000 and 8 per 1000 men at a 10% threshold. The proportion of patients incorrectly not recommended for treatment increases with age under both thresholds, however as the proportion of patients recommended for treatment also increases with age, when the threshold is lowered, the absolute number of patients affected is lower.
An alternative to reduce the measurement variability would be the use of the average of a number of measures (e.g. three TC measures taken over time). However, QRisk252 was developed based upon risk factors measured in a primary care database (QResearch), where single measurements of TC and HDL were used to estimate CVD risk. Any change in the variability of a factor included in a risk score (such as systolic BP or TC) would impact on the performance of the score. Therefore, a limitation of these additional analyses is that we did not modify the regression coefficients in the risk equation to account for the potential regression dilution effect. In Chapter 6 we discuss the expected impact of these monitoring alternatives on QRisk2.52
We have used modelling techniques rather than primary data (such as that obtained from a randomised clinical trial design); however, the feasibility of collecting data in the scale required for an experimental design testing diagnostic or monitoring strategies is unlikely, and therefore we have used established methods to quantify the long-term implications of monitoring in data representative of routine general practice in the UK.
One limitation of these analyses is the (unlikely) assumption that CVD risk factors other than age and lipid level remain constant over time; however, this project focuses on lipid levels as a risk factor for CVD, and accurately modelling the changes in each risk factor in QRisk252 was beyond the scope of this project (requiring alternative modelling as seen in Chapter 4 for each of the factors included in the risk score).
In May 2012, the UK patent for atorvastatin expired and this may affect statin prescribing. We found that a high proportion of secondary prevention patients who should have been in the control phase of monitoring (i.e. they should already have been recommended for an up-titration to a therapeutic dose) were prescribed too low a dose of statin according to current guidelines. One possible explanation for this might be availability of appropriate statins. These simulations are based upon patient characteristics from 2006. In 2006, guidance for these patients recommended initial treatment with simvastatin 40 mg, and the higher-intensity statin choices were limited to simvastatin 80 mg,2 a dose of simvastatin that further guidance57,211 suggests limiting to only those patients who have severe hypercholesterolaemia and a high risk of cardiovascular complications. Updated local guidelines59 recommend initial use of atorvastatin 40 mg, with the option of potential up-titration to atorvastatin 80 mg. Thus, patients whose lipid levels are not controlled by simvastatin 40 mg might now be prescribed atorvastatin 40 mg, which is associated with a higher percentage reduction in LDL.59 We were not able to investigate the effects recent potential changes to statin prescribing might have on the proportions of false-positive and false-negative treatment eligibility decisions with the available data. CPRD data was available for the period of May to 21 September 2012; however, the proportion of patients prescribed atorvastatin and simvastatin among secondary prevention patients who were prescribed statins, was similar to that seen in 2006 (atorvastatin: 30% in 2012, 34% in 2006) and it may take time before prescription patterns change.
The threshold for guiding treatment in primary prevention patients was lowered from 20% to 10% in July 2014;56 sensitivity analyses using thresholds of both 15% and 10% showed similar patterns, although the estimated proportion of under-treated patients increased as the threshold decreased. In addition, there is some variation in the interpretation of the 2010 NICE guidelines for secondary prevention; specifically whether both or only one of TC and LDL is over threshold for up-titration to be considered.2,59 Our models do not enable us to examine the progression of both TC and LDL simultaneously so we have instead examined monitoring schedules for each lipid individually. A joint model to guide treatment changes could be part of future research, although the updated guidelines recommend that secondary prevention patients are prescribed 80 mg of atorvastatin,56 and, as such, there is no higher dose for which subjects would need to be monitored for up-titration.
Analyses are restricted to a population attending monitoring. We have not attempted to model the complexity of individual treatment decision-making; however, the analyses presented here aim to form the basis for evaluation of national monitoring strategies. Neither have we attempted to quantify the relative harms of false-positive and false-negative treatment eligibility decisions; this is addressed in Chapter 7, where we extend the work presented here to a full health-economic model that estimates the relative cost-effectiveness of different monitoring strategies.
