Cardiovascular polypharmacy is not associated with unplanned hospitalisation: evidence from a retrospective cohort study

Statistical analysis was performed using Stata v11.2 (StataCorp, Texas, USA). Mixed-effect binary logistic regression was used to model hospital admission for a primary non-cardiovascular problem, with count of cardiovascular medicines as the exposure of interest. Adjustment was made for sex, age group (20-year age ranges), an area-based socioeconomic deprivation measure (quintiles of the Scottish Index of Multiple Deprivation, SIMD), count of non-cardiovascular conditions, count of non-cardiovascular medicines, and the presence of certain cardiovascular conditions (ischaemic heart disease, heart failure, stroke including transient ischaemic attack, peripheral vascular disease, hypertension and atrial fibrillation). We also included a random effect for GP surgery to account for potential clustering of prescribing and admission characteristics by this factor.

One could argue that the effect of cardiovascular conditions amongst patients not receiving appropriate medicines may be different from that in those patients being treated. If so there may be some residual confounding in the model described above. In order to overcome this we conducted an additional more complex analysis, including interactions between the specific cardiovascular conditions and the cardiovascular medicine count. Exploratory analysis suggested that using interaction terms with the full 5-category drug count variable (as used in the non-interaction model) provided no further information than using a simple binary variable indicating presence or absence of any cardiovascular drug. We therefore present the results of a model using the more parsimonious parameterisation for the interaction term only (i.e. a model containing a single interaction between a binary variable indicating the presence of at least one cardiovascular drug and 6 binary variables representing the presence of each of the cardiovascular conditions – a total of 6 interaction terms).

We conducted a number of sensitivity analyses to check the robustness of our findings. We substituted cardiovascular medicine count in the main model with a count of eight different cardiovascular drug classes (ACE inhibitor or angiotensin receptor antagonist, beta-blocker, calcium channel blocker, diuretic, nitrate, anticoagulant or antiplatelet agent, lipid lowering drug, and any other cardiovascular medicine), to ensure that the means of classifying cardiovascular polypharmacy did not affect the results. To explore the impact of using an alternative approach to classifying cardiovascular morbidity, we included only the two commonest cardiovascular diseases (ischaemic heart disease and stroke) and the two commonest risk factors (hypertension and diabetes) as separate binary fixed effects, and included the other cardiovascular morbidities (heart failure, atrial fibrillation, peripheral vascular disease) in our “non-cardiovascular” morbidity count. Finally, to exclude the potential that excessive death might result in an apparent reduction in hospitalisation, we repeated the analysis excluding those with death within 6 months of admission (death data was unavailable for the full 12 months).

Our study has a number of important strengths. These include the large study population, the use of high quality data, and the linkage to secondary care data which helps improve capture of hospital events. There are, of course, a number of limitations. Firstly, we used a crude count of cardiovascular medicines, and did not take into account different drug classes or medication dose. However, our sensitivity analysis employed an alternative measure of polypharmacy based on drug classes and found similar results. Secondly, we were unable to factor in clinical indication for particular medicines. Nonetheless, we did include a number of cardiovascular diagnoses in our analysis (with an alternative approach used in the sensitivity analyses) including an interaction term in the secondary analysis. Thirdly, we used simple counts of non-cardiovascular long-term conditions and medicines, and it may be that results may differ if one looks at potential high-risk or complex groups, such as heart failure patients receiving non-steroidal anti-inflammatory drugs, or patients with severe renal impairment unable to tolerate ACE inhibitors or diuretics. More sophisticated measures, taking into account appropriateness of drugs and clinical context, should certainly be considered when evaluating polypharmacy, but to date consensus has not been reached around appropriate definitions, and one needs to balance the transparency and pragmatism of simple measures of polypharmacy against more complex approaches. Fourth, the analyses are based on data from 2006 to 2007, yet rates of medication use may have increased in the interim. It is possible that if this increase is also associated with more inappropriate use, then we may have underestimated the potential for adverse consequences. Finally, it could be argued that a more specific medicine-related outcome might be more relevant, such as hospital admission for adverse drug reactions. However, inappropriate polypharmacy can be considered to reflect poorer care more generally (e.g. non-adherence to clinical guidance, lack of continuity of care), and as such might be expected to manifest in broader measures of adverse outcome, of which unplanned hospitalisation is one. We specifically did not examine the association between cardiovascular medicines and cardiovascular admissions as we believe that this would be impossible to usefully interpret. For example, a positive association between the two may be due to patients on multiple cardiovascular medicines having poorer health, whereas a negative association may be due to more cardiovascular medicines reflecting better care; a combination of these effects might also be possible.

Previous work has found associations between increasing unplanned hospitalisation and increasing degrees of polypharmacy [3‐5]. However, this has not previously been examined in the specific context of cardiovascular medicines. Although non-cardiovascular admissions were observed more frequently in those receiving more cardiovascular medicines, this association was reversed once key confounding factors had been adjusted for. This negative association was attenuated once the interaction between cardiovascular disease and cardiovascular therapy was accounted for, suggesting some of the observed effect is driven by people with cardiovascular disease who are not receiving appropriate cardiovascular treatment, and are less well managed in general. Nonetheless, the effect still persisted to a lesser degree, and there are a number of potential explanations for this observation. Firstly, it is possible that more cardiovascular medicines may be associated with better clinical care generally. This might be due to better attentiveness to evidence-based practice for other clinical areas, that increased follow-up for cardiovascular disease provides an opportunity for clinicians to address non-cardiovascular morbidities, or that more frequent medication reviews are likely to occur with consequent optimisation of non-cardiovascular therapy. A second possibility is that more cardiovascular medicines may be associated with different patient behaviours; patients on these medicines may have a higher awareness of preventative strategies against admission and be more proactive about contacting their GP at other times. Finally, it may well be that some admissions are associated with cardiovascular disease, but are not coded as such [15], with examples including acute kidney injury, complications of diabetes, and smoking-related illness. It is seems unlikely that cardiovascular polypharmacy is primarily responsible for reducing non-cardiovascular admissions. It is also not possible to exclude the possibility that cardiovascular polypharmacy may be harmful in terms of unplanned admissions, but it would appear that overall any harm is more than outweighed by the associated better clinical care. It is of course possible that any such benefit would no longer be observed if there were improvements in holistic care not simply limited to those with specific long-term conditions such as cardiovascular disease. We can contrast the current findings with related work we have carried out examining the general case, where we found that increasing numbers of medicines was indeed associated with increased unplanned admissions, albeit an effect tempered in the context of increasing co-morbidity [5]. It is possible that our current findings differ from others’ work due to differences in the study populations and statistical models employed, although previous work has corrected for similar factors such as age, gender and comorbidity [3, 4]. The differences between cardiovascular and general cases seem more likely to be accounted for by the more systematic and evidence-based nature of prescribing for cardiovascular disease, compared with medication use in other clinical areas. The current findings confirm that it is essential to consider the clinical context of polypharmacy, as well as providing clinicians with some confidence that the considerable burden of cardiovascular medications is in general not contributing to undesirable hospital admissions.

We corrected for a number of important confounding factors in our analysis. Older people were more likely to be admitted, reflecting the increased prevalence of long-term conditions and frailty. Socioeconomic deprivation was also associated with increased admissions. This is probably due to general higher levels of ill-health, and poorer resources in terms of coping strategies and support networks [16]. With the exception of stroke, cardiovascular morbidity was not particularly strongly associated with non-cardiovascular admissions. The weak associations observed may reflect general frailty, and the stronger effect found with stroke likely reflects complications of this condition less commonly found with other cardiovascular conditions, such as poor mobility and falls, or stasis pneumonia [17]. The negative association with hypertension has been noted previously, and could be suggestive of improved clinical monitoring in such patients [18]. It is worth noting that the effect of cardiovascular morbidity generally goes up in those patients not receiving any cardiovascular drugs. Non-cardiovascular morbidity and medicine count are both unsurprisingly associated with increased admissions. Medicine count may in this circumstance either reflect increasing disease severity, or illness not captured by the simple morbidity count. Of course, there are numerous other factors related to prescribing that we were unable to account for, beyond simply number of medications. These include physician adherence to clinical guidelines, patient experience of care, potential adverse effects and aspects of the doctor-patient relationship. Some of these issues might be addressed through the use of more sophisticated metrics of polypharmacy, and some may themselves be influenced by treatment burden; such factors certainly have the potential to modulate the effect of polypharmacy that we observed.

This work was approved by the Privacy Advisory Committee of the Information Services Division of NHS National Services Scotland.