Impact of lowering the risk threshold for statin treatment on statin prescribing: a descriptive study in English primary care

Summary

There was a large reduction in the average coded risk of people initiated on statins, which closely matches the currently available evidence.⁸ When viewed in isolation, the reduction in the average coded risk score (Figure 2) and the change of proportion in each risk category (Figure 3) indicate a significant change in clinical practice. NICE has quoted this evidence in their impact report.¹¹ However, because the coded risk analyses only consider the subgroup of people with a coded risk score, this analysis is at risk of cohort selection bias, as the subgroup may not be representative of all people initiated on statins. This risk is exacerbated by the increasing proportion of people initiated on statins that have a coded risk score (Table 1). As this subgroup increases in size, unless risk scores are recorded at random this will have a significant impact on the average risk of this subgroup. Importantly, the changes in risk are driven by changes in for whom GPs are recording risk scores, rather than a change in who is receiving statins.

No change was found in the average EHR-derived risk of people being initiated on statins after the guideline change, and a small increase in the proportion of people that belonged to the 10–20% risk group was found. This analysis is not affected by the same selection bias, as it considers all people initiated on statins each year. Therefore with the extra data presented in this paper the authors believe the response to the guideline change is not as impactful as first thought.

Comparison with existing literature

To the authors’ knowledge, only one other study has measured the impact of CG181 on clinical practice.⁸ This research is enhanced with an expanded analysis considering all people initiated on statins, their findings validated by replicating their analysis in CPRD. This study’s data indicate recording practices of GPs had a significant impact on the average coded risk. The proportion of people with a coded risk is small and increasing rapidly at this time (26.08% in 2013–2014 and 42.76% in 2014–2015). It is highly likely that the subgroup of people receiving a coded risk score was changing (it is unreasonable to assume that this increase in recording was happening at random), but the typical patient being initiated on statins was not. One hypothesis is that GPs became far more likely to calculate the risk of someone in the 10–20% range using a QRISK tool after the guideline change, but their prescribing behaviour remained the same.

No reduction was found in the average EHR-derived risk after the guideline change (Figure 2). Although this indicates the guideline change had no impact, considering all the results leads to a slightly different conclusion. This constant average risk appears to be caused by a combination of a small increase in 10–20% risk people initiated on statins, and a drop in low-risk (<10%) people. In Figure 3, a steady increase in the proportion of people in the 10–20% risk group can be seen, and a decrease in the other two groups (a larger decrease in the <10% risk group). It is possible that the guideline change has had an equal effect on preventing statin initiation in low-risk (<10%) people as it has on increasing statin initiation in the target 10–20% risk group, resulting in no change to the average EHR-derived risk.

The data agree with a modest impact of the guideline change, but the changes are far more subtle than would be concluded if looking at the currently available evidence. These findings are important, as the numbers in the widely quoted statistics — 4.5 million patients becoming eligible for statin treatment due to the guideline change^17–20 and prevention of up to 28 000 heart attacks and 16 000 strokes each year⁷ — are likely far from being achievable.

Strengths and limitations

This is the first study to evaluate the impact of the NICE guidance CG181 on the risks of all people receiving statins in England. The study cohort is large and results are likely generalisable to the English population as CPRD is representative of the UK in terms of age, sex, and ethnicity.¹²

There are two key limitations in this work. The first is the imperfect agreement between the EHR-derived risks and the coded risks, because the EHR-derived risks should represent the risks of individuals as closely as possible. Potential reasons for the disagreement between these and the coded risks are:

• the EHR-derived risks use the QRISK3 algorithm, whereas QRISK2 will have been used in practice over those years;

• this study used multiple imputation to impute missing data, whereas missing data is imputed using mean imputation when coded QRISK scores are generated by GPs;

• variables have been identified using code lists that may not perfectly match those used by the algorithm in practice; and

• this study considers coded risks within a window of 6 months before statin initiation, while patient data could have changed in that time.

Despite some disagreement, the relationship was strong enough that, if the people with a coded risk score were a random subset of all people initiated on statins, there would have been a large drop in the average EHR-derived risk after the guideline change (as there was in the average coded risk). This was not the case, indicating the likelihood of selection bias in the coded risk analyses. This necessitates the analysis using the EHR-derived risks, even if the estimated risks are not perfect.

The second limitation is that many practices left CPRD towards the end of this study, resulting in a risk of selection bias in the cohort if the drop-out was not at random. There is no reason to believe, however, that people from practices that dropped out were more or less likely to be initiated on statins. Furthermore, the results considering the coded risk scores were comparable with those of Finnikin et al,⁸ a study carried out in The Health Improvement Network database (https://www.the-health-improvement-network.com/en/),²¹ which has not suffered from this limitation.

Implications for research

The change in NICE guidance appears to have had a small effect on statin prescribing by GPs. Given that NICE invests time and resources into developing these guidelines, it would be worthwhile for them to understand why there has been such little response. The authors propose a qualitative study with GPs and patients to assess the barriers to statin initiation for the primary prevention of CVD in people with 10–20% risk. A recent scoping review²¹ of the current literature regarding the use of statins to prevent CVD found only three studies specifically considering primary prevention of CVD, and that ‘it was difficult to interpret how doctors’ or patients’ attitudes would vary according to the risk profile of the individual patients’. A systematic review provided a comprehensive review on patient attitudes towards taking statins;²² however, the majority of studies were looking at long-term adherence, as opposed to statin initiation. No studies have investigated specifically the willingness to initiate at a 10% or 20% threshold for primary prevention of CVD. A debate article published in 2016²³ discusses patient attitudes to taking statins in light of the NICE guidance change, attributing the lack of uptake in lower-risk patients to transferability of evidence from research to practice and the potential for side effects. However, the evidence base^24–27 for their findings pre-dates the large amount of pro-statin research that came about in 2013 that has fuelled the statin debate. The authors of that debate article also noted that ‘there is sparse literature regarding the views of GPs’. Some qualitative research does exist in this area,^{25,26,28–30} but, again, no studies have been carried out in the wake of the NICE guidance, or on prescribing specifically at 10% compared with 20% risks.