Article Text

PDF

Chronic condition comorbidity and multidrug therapy in general practice populations: a cross-sectional linkage study
  1. Eyitope R Roberts1,
  2. Daniel Green2,
  3. Umesh T Kadam3
  1. 1Department of Medicine, University of Calgary, Calgary, Alberta, Canada
  2. 2Arthritis Research UK Primary Care Centre, Keele University, Newcastle-under-Lyme, Staffordshire, UK
  3. 3Health Services Research Unit, Keele University, Newcastle-under-Lyme, Staffordshire, UK
  1. Correspondence to Dr Umesh T Kadam; u.kadam{at}keele.ac.uk

Abstract

Objectives The study investigated (1) the association between comorbidity and multidrug prescribing compared with the index condition, and (2) the association between vascular comorbidity and non-vascular condition key drug prescribing.

Design Cross-sectional study linking anonymised computer consultations with prescription records for a 2-year time period.

Setting 11 general practices in North Staffordshire, England.

Participants Study groups aged 40 years and over (N=12 875). Within six conditions, comorbid group with the other five conditions was compared with an ‘alone’ group without them. Additionally, how the ‘vascular’ (one of diabetes, cardiovascular disease and cerebrovascular disease) comorbidity influenced chronic obstructive pulmonary disease (COPD), osteoarthritis (OA) or depression drug prescribing was investigated.

Outcome measures Based on the British National Formulary, five main drug chapters constituted a measure of drug counts, with low count as 2 or less and high multidrug count as 3 or more. Key drugs prescribed for COPD, OA and depression were derived from guidelines.

Results The adjusted associations between the comorbid groups and higher multidrug count compared with their respective ‘alone’ group were: odds ratio (OR) 7.1 (95% CI 5.6 to 9.0) for depression, OR 5.4 (95% CI 4.6 to 6.3) for cardiovascular disease, OR 3.7 (95% CI 2.8 to 5.0) for cerebrovascular disease, OR 3.6 (95% CI 3.1 to 4.3) for OA, OR 3.5 (95% CI 3.0 to 4.2) for diabetes and OR 3.2 (95% CI 2.6 to 4.0) for COPD. In COPD, vascular comorbidity was associated with a significant reduction in key COPD drug treatments (adjusted OR 0.6 (95% CI 0.4 to 0.8). In depression, vascular comorbidity was associated with a reduction in key depression drug treatments (OR 0.6 (95% CI 0.4 to 0.7)).

Conclusions Our findings show that multidrug prescribing for different body systems is higher with comorbidity and may be associated with lower likelihood of prescribing for specific conditions. Further research is required on whether multidrug prescribing influences the outcomes of care for chronic conditions.

  • EPIDEMIOLOGY
  • CARDIOLOGY
  • RESPIRATORY MEDICINE (see Thoracic Medicine)
  • RHEUMATOLOGY
  • MENTAL HEALTH
  • THERAPEUTICS

This is an Open Access article distributed in accordance with the Creative Commons Attribution Non Commercial (CC BY-NC 4.0) license, which permits others to distribute, remix, adapt, build upon this work non-commercially, and license their derivative works on different terms, provided the original work is properly cited and the use is non-commercial. See: http://creativecommons.org/licenses/by-nc/4.0/

View Full Text

Statistics from Altmetric.com

Strengths and limitations of this study

  • The study was based on large-scale data linking common chronic conditions from general practice populations to prescription data over a 2-year time period.

  • The study highlights the innovative approach to multidrug measurement which accounts for vascular condition-specific drugs as well as summarising non-vascular codrug therapy.

  • The study provides the emergent approach to investigating the influence of multidrug therapy on potentially ‘optimal’ drug prescribing in populations.

  • The study uses a specific but limited number of common chronic conditions to illustrate the approach to linking comorbidity and multidrug data within a single large region of the UK.

  • The study used overall broad measures of drug prescribing and further research is required to understand the specific influence of multidrug dose and duration on longer term outcomes.

Introduction

Many older people experience two or more morbidities at the same time which is defined as multimorbidity, and within this comorbidity is defined as other co-occurring diseases in the same individual with an index condition.1 ,2 These are important concepts as the experience of multiple conditions at the same time may influence the progression and treatment of an index condition. Current evidence of the overall implications of chronic diseases has shown that this phenomenon is associated with adverse health, increased healthcare utilisation and increased mortality.3–5 Although the health impact of chronic disease comorbidity has been studied, there have been few studies on how chronic diseases comorbidity might influence drug use and related clinical decisions, especially in general practice. This is a significant evidence gap despite the fact that drug interventions feature routinely in many disease guidelines. Currently, the model for managing chronic diseases focuses on treating individual conditions, and patients may on the one hand benefit from the drug treatment of each of their chronic conditions; however, there is a risk of multiple drug therapy, side effects and drug interactions which could in combination be detrimental.6 ,7

Many national healthcare policies have developed frameworks for chronic disease models of care and specific guidelines for the optimal management of chronic diseases. Examples include policy and guidelines for the common conditions in the general population with diabetes, ischaemic heart disease, stroke, chronic obstructive airways disease and depression.8–12 In addition, these guidelines are beginning to be adapted for the common experience of comorbid conditions, particularly by older people, for each of these individual conditions.13 Since people with one or more chronic conditions are increasing in number, this has increasingly brought in focus the scale and quantity of multiple drug prescribing in general populations. The key questions then become (1) how does multiple drug prescribing for different systems relate to the primary index condition and (2) how does multiple drug prescribing escalate when populations experience multiple conditions which might be directly linked or occur by chance together. The cardiometabolic diseases, such as hypertension, diabetes, heart disease and cerebrovascular disease, share aetiology and common drug treatment pathways, but it is still important to understand the scale of multiple drug therapy that might be associated when these conditions co-occur together in the same individual. Many chronic diseases also have conditions which are related to mechanisms other than pathophysiology. For example, other common chronic conditions include chronic obstructive airways disease and depression, and this epidemiology provides the scale of multiple drug therapies when co-occurring conditions might be unrelated.

In terms of the current evidence in this field, much of it has focused around ‘polypharmacy’ studies.14–16 However, while this might seem an appropriate broad umbrella term, in research and clinical approaches, it has often focused on arbitrarily chosen number of drugs, and linked the term to either inappropriate prescribing or associated adverse events in older populations.16 This lack of consensus defined approach to this problem has led to an argument for less ambiguous terminology,17 and we propose that ‘multidrug’ therapy is used to link in with the standard approach to two or more conditions, which is ‘multimorbidity’. Within this evidence, there is still a clear gap in how morbidity link to drug prescribing, and whether comorbidity influences the drug prescribing for an index disease.

In this study, the focus was on six common chronic conditions in the general population, which included diabetes mellitus, cardiovascular diseases, cerebrovascular diseases, chronic obstructive pulmonary diseases (COPD), osteoarthritis (OA) and depression. The choice of these chronic conditions for the purpose of the study was based on a number of factors including the epidemiology, especially prevalence of the diseases, as well as aetiopathogenesis and impacts on quality of life and psychological well-being. For example, while diabetes mellitus, coronary heart disease and cerebrovascular diseases have a common pathological basis of causation (the ‘vascular group’), and often coexist in one patient, they are also known to have high mortality rates—hence the drive towards measures aimed at optimising the management of these diseases.18 ,19 The other three non-vascular chronic conditions—COPD, OA and depression—are leading causes of morbidity, high cost of care and psychological distress, respectively.20–22 The rationale for our focus on few selected common conditions was also to provide common comorbidity combinations which are potentially treated with drugs as a key intervention.

We investigated two separate issues using the selected group of vascular and non-vascular conditions. First, we wanted to investigate the relative multidrug prescribing for each of six chosen index examples, comparing comorbid groups with prescribing levels in the respective index groups. Second, we wanted to test whether vascular comorbidity influenced key drug prescribing for chosen conditions. The vascular group was likely to be on similar multiple drugs, so the distinct hypothesis was tested, that was drug prescribing in vascular conditions overall may influence key drug prescribing in the individual non-vascular conditions of COPD, OA or depression.

Methods

Design and study population

The cross-sectional study was conducted using two linked databases on patients aged 40 years and over presenting to general practice over a 2-year time period (from 1 January 2002 to 31 December 2003). We wanted to investigate what multidrug prescribing levels were before a national UK performance-based incentive (Quality outcomes Framework) was implemented to test the associations between comorbidity and routine multidrug prescribing.

Settings

The clinical and prescription databases analysed were derived from an anonymised computer-recorded consultations from 11 general practices from the North Staffordshire Keele GP research partnership. The partnership covers a range of practices covering varying socioeconomic groups within rural and urban areas and has been involved in data collection over time for the purpose of epidemiological studies. There is an ongoing process of data validation to improve data quality, and there is evidence that this measure improves data recording by general practitioners (GPs) and their teams.23

Chronic disease data

The Consultation in Primary Care Archive (CiPCA) database focuses on the routinely collected morbidity encounters in actual consultations and coded using a standard clinical classification (READ codes).24 Patients who had a record of a disease-specific READ coded morbidity of interest were included in the study and the main codes were used with all associated ‘daughter codes’. The main READ codes that were used to define the chronic disease groups were: diabetes mellitus (READ codes C10), cardiovascular diseases (ischaemic heart disease (G3); heart failure (G58), excluding hypertension), cerebrovascular diseases (G6), COPD (H30, excluding asthma), OA (N05, excluding arthralgia) and depression (E11, E20, Eu and excluding psychosis).

Comorbidity: definitions

There were two approaches to defining comorbidity. First, comorbidity was defined as the presence of one of the other five selected conditions. So using the diabetes population as an example, the diabetes ‘index’ group was defined as diabetes ‘alone’ and without anyone of the other five conditions, whereas diabetes ‘comorbid’ group was defined as at least one of the other five conditions. The index ‘alone’ group would also enable the capture of the other morbidity that was outside of the ones within the study. This definition was applied to each of the six chronic conditions individually. Second, in the vascular group, comorbidity was defined separately as the individual and specific addition of COPD, OA or depression, and irrespective of whether the latter three occurred together.

Prescribed drug measure: overall multidrug count definitions

The Prescriptions in Primary Care Archive (PiPCA) database focuses on the routinely collected prescribed medications and which were coded using the British National Formulary (BNF) classification.25 The BNF consists of 15 main chapters based on the systems of the body, and within which there are further subsections for specific clinical indications. Only patients on repeat drug prescriptions were selected for defining measures because this gives a better representation of multiple drugs used on a long-term basis for the majority of patients with chronic conditions.

Specific drug treatment chapters for the six chronic diseases of interest in the study were identified and used as a summary of multidrug counts. The BNF chapter for cardiovascular and cerebrovascular drugs was under BNF chapter 2, for COPD drugs under chapter 3, for depression under chapter 4, for diabetes mellitus under chapter 6 and for OA under chapters 4 and 10. This means that overall, there were five main BNF chapters, which could constitute a measure of drug counts of up to a total of 5. The multidrug count definition in this approach would then specifically relate to people prescribed drugs from at least two or more of the five chapters indicated.

Vascular comorbidity and drug prescribing for non-vascular conditions

The key likelihood of receiving drug treatments for the specific conditions of COPD, OA and depression in the study population with vascular comorbidity was also investigated. In this approach, the ‘vascular’ comorbidity was defined as the group any one of diabetes, cardiovascular disease and cerebrovascular disease. The non-vascular groups were then individually compared with and without vascular comorbidity. For example, the COPD group was compared with vascular comorbidity to the COPD without vascular comorbidity, in relation to the likelihood of receiving COPD-specific drug treatment.

While the key drug treatments for COPD, OA and depression can be examined in different ways such as the use of specific drugs, or drug doses and duration of drug therapy, we wanted to first establish the simplest likelihood of a patient given any one of the key group of drugs for COPD, OA or depression. The group of drugs derived from guidelines for COPD10 included bronchodilators, corticosteroids, inhaled steroids and oxygen (BNF sections 3.1, 3.2, 3.5 and 3.6). The group of drugs for OA26 included non-opioid analgesics, opioid analgesics, non-steroidal anti-inflammatories and Cox 2 inhibitors (BNF sections 4.7.1, 4.7.2, 10.1.1 and 10.1.2.2). The group of drugs for depression11 included hypnotics, anxiolytics and antidepressants (BNF sections 4.1 and 4.3).

Analysis

The first analysis was to describe the 2-year period prevalence of the five main BNF chapters in the specified chronic disease population, with a focus on some of the common drugs that were prescribed within each chapter expressed as drug prevalence/10 000 population aged 40 years and over, and differences were assessed using χ2 tests. The five main chapter drug categories prevalence is described by age, gender and deprivation status. Deprivation was measured by the Index of Multiple Deprivation (IMD) which is a composite score that is linked to postal address codes.27 The IMD score was categorised into the bottom 20% (most deprived), middle 60% and the top 20% score (most affluent).

For each of the six chronic conditions, associations between the comorbid groups and higher multidrug counts were compared with the respective reference ‘alone’ group. The ‘outcome’ of higher multidrug therapy was defined as 3 or more of the chapter counts and compared with 2 counts or less. Associations using logistic regression were expressed as ORs with 95% CIs, and also included the ratios comparing prevalence of each drug count category in the comorbid group compared with the ‘alone’ group. Then for the vascular group, associations between each of the comorbid group with COPD, OA or depression were compared with the vascular ‘alone’ alone, and higher multidrug counts were then estimated.

Finally, the data were analysed for the study defined optimal drug treatments for COPD, OA or depression. Three study groups constructed were: COPD with at least one of the vascular conditions; OA with at least one of the vascular conditions and depression with at least one of the vascular conditions. Each group was then compared to their respective vascular group, for example, COPD and vascular group compared with COPD without a vascular condition, by the specific optimal drug treatment. Association estimates using logistic regression are presented as unadjusted and adjusted figures with 95% CIs. Analyses were carried out using SPSS V.17.0 statistical software.

Results

Study population

In the study population of 12 875 aged 40 years and over, the numbers of patients prescribed with cardiovascular system drugs were 9384 (2-year time period prevalence 73%), respiratory system drugs were 2861 (22%), non-opioid analgesia were 5395 (42%), antidepressants were 3241 (25%), antidiabetic drugs were 2916 (23%) and musculoskeletal system anti-inflammatory drugs were 2143 (17%; table 1).

Table 1

Prescribed drug prevalence by BNF main chapter and specific sections

In terms of the sociodemographic distribution, older patients aged 70 years and over and populations in the top 20% most deprived status were significantly more likely to be prescribed all main drug categories, except for the cardiovascular system (χ2 test for trend p<0.001). For women compared with men, there was variation by type of main drug category; the comparative 2-year prevalence figures by gender were significantly higher for men compared with women for the cardiovascular system drugs (76% vs 70%) and diabetes (26% vs 20%), but similar for COPD (p=0.462). Prevalence figures were lower for men compared with women for anxiolytics and antidepressants (49% vs 66%) and anti-inflammatories (15% vs 18%; χ2 test p<0.001; table 2).

Table 2

Sociodemographic characteristics of the main drug categories

Individual chronic condition comorbidity and higher multidrug counts

For all six specified chronic conditions, at lower drugs counts of up to 2, the prevalence numbers were greater for the individual groups without the other five comorbid conditions compared with the numbers for the individual conditions with comorbidity of other five conditions (table 3). For the drug count of 2 different chapters, the comorbid to ‘alone’ ratios ranged from 1.15 for the depression group to 0.5 for the diabetes group. The prevalence ratios were highest for the multidrug count of 4, and these ranged from 13.7 for the depression comorbid group to 2.3 for the diabetes comorbid group.

Table 3

Associations between individual study groups and higher multidrug counts

Adjusting for age, gender and deprivation, the associations between the comorbid groups and higher multidrug count compared with their respective ‘alone’ group ordered by strength of association were: OR 7.1 (95% CI 5.6 to 9.0) for depression, OR 5.4 (95% CI 4.6 to 6.3) for cardiovascular disease, OR 3.7 (95% CI 2.8 to 5.0) for cerebrovascular disease, OR 3.6 (95% CI 3.1 to 4.3) for OA, OR 3.5 (95% CI 3.0 to 4.2) for diabetes and OR 3.2 (95% CI 2.6 to 4.0) for COPD.

Vascular condition comorbidity and higher multidrug counts

The prevalence ratios for the multidrug count of 5 ranged from 3.9 for vascular group comorbid with OA to 1.9 for vascular group comorbid with COPD and 1.0 for the vascular group comorbid with depression (table 4). Adjusting for age, gender and deprivation, the associations between the comorbid groups and higher multigroup count compared with their respective ‘alone’ group ordered by strength of association were: OR 4.6 (95% CI 3.8 to 5.7) for vascular group comorbid with COPD, OR 3.2 (95% CI 2.6 to 3.9) for vascular group comorbid with depression and OR 3.0 (95% CI 2.6 to 3.5) for vascular group comorbid with OA.

Table 4

Associations between vascular comorbidity groups and higher multidrug counts

Comorbid vascular conditions and optimal non-vascular condition prescribing

The three specific non-vascular groups of COPD, OA and depression were compared with comorbid vascular conditions to without such vascular comorbidity in terms of their respective optimal drug treatment (table 5). Adjusting for age, gender and deprivation, the association between the COPD and vascular comorbid groups compared with their respective group without vascular conditions showed a significant reduction in optimal COPD drug treatment with an OR of 0.6 (95% CI 0.4 to 0.8). Adjusting for age, gender and deprivation, the association between the depression and vascular comorbid groups compared with their respective group without vascular conditions showed a significant reduction in optimal depression drug treatment with an OR of 0.6 (95% CI 0.4 to 0.7). Adjusting for age, gender and deprivation, the association between the OA and vascular comorbid groups compared with their respective group without vascular conditions did not show a statistically significant reduction in optimal OA drug treatment with an OR of 0.8 (95% CI 0.6 to 1.1).

Table 5

Key drug treatments of non-vascular conditions in vascular comorbidity

Discussion

Our findings from a large cross-sectional study of nearly 13 000 patients aged 40 years and over with one of six specified and common chronic conditions showed the scale of multidrug prescribing, which was higher in the presence of comorbidity compared with the respective index groups. While previous evidence has shown the high levels of multiple drug prescribing,15 our study findings link the disease and comorbidity status to the measure of multidrug prescribing for different systems.

Depending on whether the chronic conditions were vascular (diabetes, cardiovascular or cerebrovascular) or non-vascular (COPD, OA or depression), the higher levels of multidrug prescribing varied. All six conditions with comorbidity compared with their index condition had much higher multidrug count, even adjusting for age, gender and deprivation. The measure of multidrug count was notably distinct by the use of five different main drug chapter categories which were for different body systems, which means that this ‘outcome’ was not about multiple drug use for the same condition. For example, a diabetic with a higher multidrug count of 4 or 5 in this study relates to different and distinct body systems, and not to the different drugs under the same chapter. The chronic condition of depression comorbidity had the strongest strength of association with higher multidrug counts, followed by cardiovascular disease comorbidity, and the estimates of association for cerebrovascular disease, OA and diabetes were similar. These findings suggest that the index condition and associated comorbidity may influence the range of multidrug prescribing, and generates the interesting hypothesis that potential variation in clinical outcomes of the index conditions is as a result of underlying comorbid drug prescribing.

The study also grouped the vascular-related conditions to investigate the influence of non-vascular drug prescribing compared with vascular conditions ‘alone’ (ie, without any one of COPD, OA or depression). Again, the adjusted associations were significant, with vascular comorbidity being associated with higher multi drug counts compared with the respective ‘vascular index’ group. Here the clinical implication is that vascular comorbidity in populations aged 40 years and over might not only be associated with multiple vascular drugs as routinely suggested by clinical guidelines28 but by a range of conditions such as comorbidity of COPD, OA or depression. It is possible that these conditions and the drug treatments for them may also in the end influence the health and healthcare outcomes of the index vascular conditions.29

In terms of the influence of comorbidity on key drug prescribing, our study findings show that vascular comorbidity in COPD and depression is associated with lower likelihood of drug prescribing for the respective conditions of COPD and depression. Similar findings, particularly for suboptimal depression drug treatment, when depression is comorbid with chronic disease have been shown previously.30 ,31 However, such findings for OA were not found, and here it is possible that the study definition of analgesia was too broad, as analgesia use covers a range of other painful conditions, in addition to OA. Although the key drug definition was simple and broad, our study findings seem to suggest that comorbidity does influence drug prescribing for specific conditions. Whether this is due to some kind of therapeutic inertia or is due to GPs’ reasoned consideration of drug–drug and drug–disease interactions and the overall well-being of the patient is the important question raised by the findings.

The approach taken to look at specific groups and six common conditions was based on a combination of clinical rationale and feasibility. While, one could have investigated any number of combinations of the six conditions, the better and preferred approach taken was to group conditions first at the ‘vascular’ level. As highlighted earlier, diabetes, ischaemic heart disease and cerebrovascular disease have shared pathogenesis and there may be overlapping of drug treatments. However, the ‘non-vascular’ group constitutes individual chronic conditions with distinct and unrelated drug treatments. This approach enabled comorbidity definitions based on (1) group-level, that is, vascular comorbidity with one of the non-vascular conditions and (2) counts, that is, number of other conditions for each of the six index groups. The study focus was also on comorbidity and further research is also required on how multimorbidity, defined as two or more conditions, influences the overall prescribing of multiple drugs and when the unit of analysis for outcome is not the disease but the arguably more important patient-centred outcomes.

The large scale study of specified chronic diseases was conducted using an anonymised database for a 2-year time period. In terms of the cross-sectional associations, the findings on the levels of chronic conditions, comorbidity and multidrug prescribing do offer clinical implications as outlined earlier. However, the implications of the associations between comorbidity and the key drug definitions may be limited in this cross-sectional design and these may be treated cautiously as emergent findings. The chronic disease definitions were also based on routinely collected registers from general practices, which were and are part of a research network dedicated to the collection of clinical data in actual consultation. While these chronic disease registers may be subject to variations in recording,32 the study analyses provide the estimates of association in actual clinical practice across 11 different sites.

The drug definitions were based on routinely coded repeat prescriptions and over a 2-year time period represent an appropriate measure at the simpler but distinct broad system category. Patients however will also have been prescribed other drug categories outside of the five main categories that we had selected and for other less common conditions from the ones selected in the study, which means these drug levels are a specific estimate. The construction of our study defined index or ‘alone’ groups (without the other five conditions) provided the relative multidrug level estimates to when the index condition was comorbid with one of the other five conditions. So the multidrug levels in the ‘alone’ group provide an estimate of main drug system prescribing without the associated condition (ie, for other indications) compared to levels when there is a clear comorbidity record. However, this is time defined by a 2-year time window, so some misclassification may be possible and further research could explore how broad system drug definitions capture the underlying and specific common diagnostic categories. Further research is also required for the arguably more complex assimilation of the range of defined drug categories, other multimorbidity and to investigate specific effect of individual drug categories. Most of these drugs, other than analgesia such as anti-inflammatories, are not available over-the-counter and are usually clinician prescribed. So it is possible that common over-the-counter drugs, particularly in relation to OA, may be an underestimate; however, the selection of repeated prescribing would mitigate against such underestimation. Finally, although a large scale study, these general practices are drawn from one region of England, and while this might limit generalisability, the internal validity of the findings still remains.

In conclusion, our study findings show the links between common chronic conditions, comorbidity and associated multidrug prescribing. The key and distinct finding is that the study shows that multidrug prescribing defined by a range of selected but different systems is high in chronic conditions and higher in comorbidity. The common groups of vascular conditions are not the only ones associated with their ‘own’ guideline driven multidrug therapy, but the addition of non-vascular conditions such as COPD, OA and depression adds to the multidrugs burden in patients. The importance of these findings, in addition to quantifying the scale, is whether such multidrug therapy influences the quality of care for each of the individual conditions. Our findings suggest that the potential for suboptimal drug treatment as a consequence is in line with other evidence33 but further research is required to investigate the impact of disease status, comorbidity, multidrug therapy on prospective and long-term outcomes of clinical care.

References

View Abstract

Footnotes

  • Contributors ERR and DG coordinated the study data collection and contributed to the writing of the manuscript. ER, DG and UTK were involved in study design and developed the statistical approaches. UTK conceived and designed the study, was involved with analysis, interpretation and contributed to the writing of this manuscript. All authors have contributed and approved the final version of this manuscript.

  • Funding This work was funded by a National Institute for Health Research (NIHR) In-Practice Fellowship for ERR and a NIHR School for Primary Care Research Doctoral Training Studentship for DJG. CiPCA and PiPCA databases were funded by the North Staffordshire Primary Care Research Consortium and Keele University Institute for Primary Care and Health Sciences. This report presents independent research commissioned by the NIHR. The views expressed are those of the authors and not necessarily those of the NHS, the NIHR or the Department of Health.

  • Competing interests None.

  • Ethics approval North Staffordshire Ethics Committee.

  • Provenance and peer review Not commissioned; externally peer reviewed.

  • Data sharing statement No additional data are available.

Request permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.