Is the high-risk strategy to prevent cardiovascular disease equitable? A pharmacoepidemiological cohort study

Data source and participants

From nationwide Danish registers maintained by the National Board of Health and Statistics Denmark, we retrieved individual-level information on dispensed prescription drugs, hospital discharges, dates of death or emigration, and socioeconomic indicators. Data were linked by means of a unique encrypted person-identifier, allowing authorised researchers to follow individuals in multiple individual-level registries hosted in Statistics Denmark. Register-based studies in Denmark do not require approval by an ethics board.

From the Danish National Patient Registry, we retrieved information on patient discharge from non-psychiatric hospitals since 1977. Records include the admission and discharge dates, discharge diagnoses according to the International Classification of Diseases (ICD), 8^th revision until 1993, and 10^th revision thereafter (9^th revision has not been used in Denmark) along with codes for diagnostic and surgical procedures. We included main and secondary diagnoses for admitted patients and patients in ambulatory care. From the Registry of Causes of Death, we retrieved date and cause of death (primary and contributory).

Information on dispensed prescription drugs was retrieved from the Danish National Prescription Registry (DNPR), containing full information since 1996 on all out-of-hospital purchases of prescription drugs at Danish pharmacies including those of nursing home residents [43]. Records include the person identifier, date of dispensing, and the Anatomical Therapeutic Chemical (ATC) classification code of the dispensed drug [44]. From the DNPR we retrieved information on dispensed cardiovascular drugs (ATC groups C and B01) and antidiabetics (A10).

To identify asymptomatic individuals, we applied historical register data on in/out-patient diagnoses and procedures along with dispensed prescription drugs as register-markers for a range of CVD conditions, including ischemic heart disease (IHD) with or without myocardial infarction (MI), stroke, a range of other atherosclerotic conditions, and diabetes. We define asymptomatic individuals as individuals without register-markers of CVD or diabetes, as defined in a recent publication [45].

Study design

While measures such as the Gini-coefficient of inequality, concentration index and the slope index of inequalities provide means for quantifying the degree of for example income-related inequality in health or health care delivery [46], a measure combining potential inequalities both in health care delivery and health care needs is indispensable to quantify inequities in health care delivery – if needs also are unequal across strata. However, measuring the need for preventive health care is a challenge, as such needs not may be captured by for example self-rated health scales. We opted to apply a need-proxy analogous to the underlying presumption of the risk-score chart, namely a measure of CVD- incidence in the background population of asymptomatic individuals, i.e. without CVD, diabetes or statin therapy - stratified by gender, 5-year age groups and SEP indicator.

Due to the high validity of the diagnosis of MI in the Danish registries [47], we applied the incidence of MI as need-proxy, using two alternative need-proxies in a sensitivity analysis: first stroke or MI as combined CVD endpoint and CVD as cause of death (MI, stroke or aortic aneurysm). Stratum specific MI-incidence rates were calculated, corresponding to number of incident MI cases (hospitalised or fatal) per 10,000 person-years at risk (PYR) during 2002–2006, censoring at death, emigration and register-markers of CVD, diabetes or statin therapy. Analogously, we calculated the ‘observed’ incidence of statin therapy (number redeeming the first statin prescription/10,000 PYR – censoring at death, emigration and register-markers of CVD or diabetes) and the combined MI-stroke endpoint. In order not to confine CVD-mortality to sudden CVD-death, CVD-mortality was calculated without censoring for new events of CVD or diabetes, covering also a longer span of time (2002–2008).

We applied a fixed SEP level corresponding to the beginning of the observation period (primo 2002). In order to capture income fluctuations over time, we calculated the average annual income between 1996 and 2001 (inflated to 2000 purchasing power), divided into income quintiles within gender and age group. The highest attained educational levels as of 2002 were divided into four groups according to length of formal education, cf. Table 1. As the analysis covers a time span of five years, individuals were considered to belong to a fixed 5-year age-group (i.e., that of 2002).

Based on the need-standardized statin incidence parameters (observed statin events (counts), need-standardized PYR (exposure)), Poisson regression analyses were applied to test the overall horizontal equity across SEP. With the lowest SEP-group (and age group) as reference, a need-standardized statin-IRR > 1 translates into horizontal inequity favouring the higher SEP-groups. The null hypothesis, horizontal equity, corresponds thus to statin-IRR = 1. We estimated a ‘horizontal inequity gradient’ (the overall linear trend) reflecting the increase in need-standardized statin-IRR for each increase in SEP. Owing to a gender and age specific pattern of both MI-incidence (need) and incidence of preventive statin therapy, we stratified the analyses according to gender and ages +/−65, cf. Figure 1.

Nonparametric bootstrapping [48] was applied to incorporate the precision of the need-weights in the confidence intervals (CI) of the need-standardized statin IRR. Based on 10,000 bootstrap replications, need-weights were calculated and applied in the Poisson regression analyses of need-standardized statin incidence parameters. Normal based 95% CI from the bootstrapping procedure were applied as CI for the point estimate for statin IRR calculated from the original data.

All analyses were performed using Stata Release 11.1 (StataCorp, College Station, TX, USA). Access to data was provided and secured through collaboration between the University of Copenhagen and Statistics Denmark. Register-based studies in Denmark do not require approval by an ethics board.

Principal findings

Applying indirect standardisation and MI-incidence as a proxy for need, we developed a pharmacoepidemiological method to explore horizontal equity in initiation of preventive statin therapy across SEP-groups with unequal needs, adjusting the observed statin incidence according to relative needs across socio-demographic groups. Our study indicates that the high-risk strategy (as implemented in Denmark) to prevent CVD by initiating preventive statin therapy is inequitable, reaching primarily high-risk individuals in low-risk SEP-groups. The favouring of more advantaged groups holds for both genders, independently of applying income quintiles or educational level as SEP indicator. In men aged 30–64, the need-standardized statin incidence increased by 17% for each increase income quintile (the horizontal inequity gradient) – in women the increase was 21%.

Only among men, the observed incidence statin therapy tended to increase with increasing SEP, but due to a steeper social gradient in MI-incidence among women, the horizontal inequity gradient of initiating preventive statin therapy was steepest in women.

Strengths and limitations

Given the inverse relationship between SEP and CVD, the challenge of this study examining equity in the medicamental high-risk strategy to prevent CVD was twofold: firstly, to operationalize need and equity in CVD preventive drug therapy across SEP-groups with unequal needs, and secondly, to develop appropriate pharmacoepidemiological methods for testing horizontal equity.

Needs

We opted to apply nationwide register data on MI-incidence in the statin free and asymptomatic background population stratified by gender, age and SEP as need-proxy, instead of calculating individual-level CVD-risk based on survey information on CVD-risk factors and risk scoring,

Pharmacoepidemiological method for testing equity

Analogous to studies within economic equity research [19, 22, 42], we applied indirect standardisation to evaluate horizontal equity in health care delivery. In a study on equity in US ambulatory care [42], the number of ambulatory visits was adjusted according to differential self-rated health. Applying the need-standardized counts of the dependent variable (ambulatory visits) and a continuous income variable as the explanatory variable, a ‘horizontal inequity index’ (i.e., a need-standardized concentration index) was estimated. In our pharmacoepidemiological approach, we calculated, instead, a need-standardized incidence rate of statin therapy. Applying need-standardized statin incidence parameters as a dependent variable and a SEP indicator as an ordinal explanatory variable (e.g., income quintiles), we estimated a horizontal inequity gradient. We consider this methodological analogy to be a strength. Yet, while it is intuitively reasonable to ‘adjust’ for differential health conditions when evaluating horizontal equity in ambulatory visits, it may be less obvious that incidence of preventive CVD-drug therapy should be proportional to the risk of disease for equity to be met.

Interpretation and comparison with other studies

While a range studies have demonstrated inequality in prescription of CVD preventive drugs [23, 59–61], no studies have examined and quantified inequities, including both prescribing patterns and needs in a nationwide perspective. In contrast to a Norwegian health survey study [40] showing a decreasing trend of incidence of statin treatment by increasing education in individuals without reported CVD or diabetes at baseline, we found almost the same incidence across educational groups among asymptomatic individuals, censoring for new onset of CVD/diabetes. The lack of censoring for onset of disease in the Norwegian study most likely explains the discrepancy between the studies, as lower SEP individuals are at higher risk of developing disease and may thereby be misclassified as free of CVD or diabetes when initiating statin treatment. Our finding that the high-risk strategy as implemented in Denmark seems to be inequitable may reflect both the poor predictive value of the applied risk-score charts and a selective uptake. The latter being an inherent consequence of applying an opportunistic screening strategy, where uptake depends on the client’s participation and the physician’s general judgement of her/his client. A so-called ‘healthy user effect’ has been shown in pharmacoepidemiological studies, indicating that preventive measures (e.g., vaccinations and drugs) tend to be used by population segments with a broad spectrum of healthier behaviours [62]. With the consistently shown social gradient in CVD in most Western countries, our findings are likely to be applicable in other settings applying an opportunistic screening strategy. Several studies have demonstrated a socioeconomic gradient in screening uptake [63], indicating both financial (due to drug co-payment) and psychosocial barriers in socially deprived groups [64]. Psychosocial barriers to CVD screening may include negative perceptions about screening tests, risk perceptions and the social stress associated with talking about unhealthy lifestyles with the GP – of higher SEP. Our findings may also reflect that high CVD risk in lower SEP at first hand is ‘attacked’ by encouraging individual lifestyle modifications.

In line with other studies [61, 65] our study indicates that the high-risk strategy may widen the socioeconomic gradient in CVD owing to the inequitable uptake. However, any widening of the CVD incidence gradient depends on the outcome of therapy and not merely on initiation of therapy. Here two other issues are important: Differential adherence to therapy and differential outcome of therapy. In fact, long-term adherence to statin treatment is disappointing [66] – and is likely to depend on SEP, indication and experienced adverse effects (potentially weighted against the perceived beneficial effect). While the risk of life-threatening adverse effects is low (e.g., rhabdomyolysis), various degrees of muscle side effects are not unusual, ranging from muscle weakness to rhabdomyolysis [67]. If both incidence and duration of therapy are lowest among less advantaged groups – the social gradient in prevalence and outcome of therapy is likely to be even steeper than the gradient found as to initiation of therapy. However, being exposed to multiple risk factors (known and unknown) acting in concert, socially disadvantaged groups may be more vulnerable to high LDL levels than the better-off. Hence, the outcome or beneficial effect of lifelong preventive statin therapy may be greater in less advantaged groups – provided adherence to therapy.

Unanswered questions and future research

Various incentives have been proposed to enhance adherence, requiring often GPs to be more actively involved [68, 69]. In a forthcoming study on the same nationwide Danish data we explore potential socioeconomic differences in adherence to statin treatment in asymptomatic individuals.

The incidence of preventive statin treatment in this study was found to peak around the age of 65, and to decrease steeply hereafter. This pattern may reflect the widespread use of the risk-score charts, covering the age range of 40–65, potentially representing an issue of ageism. The finding that statin incidence is considerably higher in asymptomatic women than men although MI-incidence is higher in men (cf. Figure 1) may both reflect a consequence of an opportunistic preventive screening strategy and an overestimation of CVD-risk in Danish women, corresponding to the finding in a Norwegian study [70]. Both matters will require further research.

In contrasts to the opportunistic screening strategy applied in Denmark, a universal screening programme to prevent CVD is actually being implemented in the UK [71]. Here, all asymptomatic individuals aged 40–74 are invited for risk-scoring and potential preventive statin therapy. Yet, there is an ongoing discussion about the programme covering both resource implications [72] and the dilemma between a targeted high-risk strategy with a high CVD risk threshold for initiating therapy with potentially non-generic more potent and effective statins versus a “population strategy” in which all invited individuals should be prescribed low-price preventive treatment (generic statins) [71] or eventually a Polypill (a combination of four CVD-preventive drug categories) [73, 74],bypassing the risk-scoring charts of poor predicative value and potentially also subsequent GP visits. While recent cost-effectiveness reviews indicate a very high cost effectiveness of Polypill strategies [75], a pilot project may uncover equity concerns before implementation of a general screening strategy. Another strategy could be to focus the high-risk drug strategy on middle-aged asymptomatic men in whom the beneficial effect of preventive statin treatment is best documented, testing various settings in order to reach lower SEP groups before implementation - potentially also adjusting the reimbursement system accordingly. Yet, by not controlling the causes of high CVD incidence (e.g., low cigarette prices; high levels of saturated and trans-fats, sugars, salt hidden in processed foods, and environmental factors) this ‘population’ strategy will be palliative and not radical as structural/population strategies tends to be [55]. Proposing a range of actions to be taking, a newly published Danish report “Health inequality - determinants and policies” [76] demonstrates that reducing health inequality is not primarily a health care task, but a complex task requiring coordinated efforts from different sectors (e.g. the education, social, healthcare, and employment sectors).