View Article PDF

Cardiovascular disease (CVD) is the leading cause of potentially preventable global health loss and demand for health and disability services for older people.1 There is evidence that reducing smoking,2 blood pressure (BP)3 and lipids4 is associated with reduced fatal and non-fatal CVD for adults at any age, and the benefits are largely determined by patients’ pre-treatment CVD risk. As older people are more likely than younger people to be at high CVD risk, they are also likely to benefit most from CVD risk-reducing medications.1 A recent individual person meta-analysis of 28 statin trials found that treatment produced a similar reduction in major vascular events per mmol/L reduction in low-density lipoprotein (LDL) cholesterol irrespective of age, although findings were more attenuated for those over 75 years without established CVD.4 In the case of BP-lowering, the benefits of treatment accrue even among the very old. In a meta-analysis involving only octogenarians from seven clinical trials, BP-lowering medications were associated with lower rates of stroke (34%), heart failure (39%) and major CVD events (22%) than those not receiving treatment.5 Despite these findings, most CVD risk management guidelines are vague about how to manage older people. In New Zealand, for example, CVD risk assessment and management guidelines recommend a formal quantitative CVD risk assessment for people aged 30–74 years,6 but once a person turns 75 years of age, risk assessment is ‘at the discretion’ of the clinician. General practitioners (GPs) are given general advice to use clinical judgement taking into account the results of a risk assessment, the likely benefits and risks of treatment and patient preferences.

We have identified an increasing number of New Zealanders aged 75 years and over receiving formal quantitative CVD risk assessments in routine primary care and have generated a cohort of these older people as part of our ongoing PREDICT-CVD risk prediction cohort study.7 Given the clinical relevance of accurate assessment of CVD risk in older people, we aimed to investigate how well the recently developed PREDICT-CVD equations8 (derived in people aged 30–74 years), performed in older people. As a comparison, we also assessed the performance of the relatively similar US Pooled Cohort Equations (PCEs),9 derived in people aged 40–79 years. We hypothesised that these CVD risk equations should perform reasonably well in the subgroup of older patients who GPs considered suitable for routine preventive CVD risk assessment.

Methods

Study population

The PREDICT study has been described in full elsewhere.7 In brief, patients are automatically recruited to this prospective open cohort study when primary care clinicians undertake standard quantitative CVD risk assessments using the PREDICT web-based decision support programme. Over one-third of GPs use PREDICT software, which is integrated into their electronic patient management systems. An encrypted version of each person’s unique national health identifier (National Health Index number, eNHI) is used to anonymously link patients’ risk profiles to national and regional health datasets, including all community pharmaceutical dispensing, all community laboratory testing, state-funded hospitalisations and all deaths.10 Over 98% of New Zealanders have an NHI number,10 allowing identification and linkage of multiple health contacts, augmentation of risk factor data (prior hospital admissions, pharmaceutical dispensing, laboratory test results) as well as health outcome ascertainment (fatal CVD events in-hospital and out-of-hospital and non-fatal hospital admissions for acute CVD events) during follow-up. Over 95% of CVD hospitalisations occur within our state-funded health services.11

For the purposes of this study, eligible patients were those who had a first (baseline) CVD assessment from 31 October 2004 to 30 December 2016 and were aged 70 years or older, unless they met any of the following exclusion criteria: prior history of ischaemic CVD, heart failure, renal disease or missing risk factors needed for CVD risk prediction models. In addition, to emulate the cohort used to develop the PREDICT CVD equations,8 patients were excluded if their self-identified ethnicity was recorded as Middle Eastern, Latin American, African or recorded as ‘other’ or ‘unknown’. The rationale for this exclusion was that these ethnic groups were too heterogeneous to combine into one category and too few in numbers to disaggregate into meaningful subgroups.

A history of prior CVD was classified according to an International Classification of Diseases, version 10 Australian Modification (ICD-10 AM) for hospitalisations or primary care clinical diagnosis at the time of CVD risk assessment for angina, myocardial infarction (MI), percutaneous coronary intervention (PCI), coronary artery bypass graft (CABG), ischaemic stroke, transient ischaemic attack (TIA), or peripheral vascular disease (PVD). The capture of a patient’s history of a hospitalised event used data available from 1 January 1988. (Appendix contains full list of ICD-10 codes) Patients who were dispensed anti-anginal medications on at least three occasions up to five years prior to their baseline visit were also excluded. Renal disease was determined either by an estimated glomerular filtration rate (eGFR) of less than 30ml/min per 1.73m2, an ICD-10 AM hospitalisation for renal dialysis, prior renal transplantation or a recording of diabetes with nephropathy at the time of CVD risk assessment. Heart failure diagnoses were based on ICD-10 AM hospitalisation code for heart failure or if participants had been dispensed a loop diuretic three or more times in the preceding five years.

Ethnicity classification was based on a nationally agreed prioritisation algorithm when individuals identified with more than one ethnicity12 in the following order; Māori, Pacific, Indian, Chinese/other Asian and European. Socio-economic status was assessed using the NZ Deprivation Index (NZDep), a measure assigned to patients according to the deprivation score of their area of residence.13 For these analyses, NZDep was divided into quintiles from 1 (least deprived) to 5 (most deprived).

Smoking status was defined as either current smoker (including recently quit in the last 12 months) or non-smoker. Diabetes status was classified according to ICD-10 hospitalisation with diabetes and/or dispensing of at least one diabetes medication in the last six months and/or recorded as such by their primary care clinician at the time of CVD risk assessment.

The Charlson comorbidity index is a weighted scoring system that assesses the degree of previously hospitalised comorbidity burden. It is based on 12 conditions that predict one-year survival and has been adapted for use with hospitalisation data using a well-validated ICD-10 coding algorithm.14 Comorbidities were identified from hospitalisations up to five years prior to the first CVD risk assessment.

The pharmaceutical collection (PHARMS) is a national database of community pharmaceutical dispensing. Reliable identification of dispensing episodes by eNHI has increased over the last decade from 64% in 2004, to 92% in 2006 and over 96% from 2009 onwards.7 PHARMS was used to identify patients who were dispensed one of the following medications on at least one occasion in the six months prior to the baseline CVD risk assessment: BP-lowering, lipid-lowering and antiplatelet/anticoagulant medications (henceforth termed antithrombotic medications). All these medications are government subsidised. (CVD medications are listed in the Appendix).

Outcomes during follow-up

CVD outcomes for the PREDICT-CVD equations8 were defined as ICD-10-AM coded hospitalisation or death from ischaemic heart disease, ischaemic or haemorrhagic cerebrovascular events (including TIA), PVD or heart failure. CVD outcomes for the American PCE Equations9 (termed hard atherosclerotic CVD) are a subset of the former and include fatal or nonfatal MI, fatal or nonfatal stoke, or CHD death (Appendix contains ICD-10-AM codes for both sets of outcomes). Time on study was the time from baseline CVD risk assessment to the first of the following: hospital admission or death related to CVD, death from other causes or end of follow-up.

Statistical analysis

The distributions of CVD risk factors, event rates and follow up were investigated for the total population and by five-year age groups; 70–74 years, 75–79 years, 80–84 years and 85 years and over. Calibration was tested separately for PREDICT-CVD8 and the PCE models9 by five-year age groups using equation-specific outcomes. The PREDICT-CVD risk models include age, ethnicity, deprivation, diabetes status, history of atrial fibrillation, smoking status, systolic blood pressure (SBP), the ratio of total cholesterol (TC) to high-density lipoprotein (HDL) cholesterol (TC/HDL) and prior dispensing of BP lowering, lipid lowering and antithrombotic medications. The PCEs include age, TC, HDL, smoking and diabetes status, SBP and treated SBP. Calibration was assessed graphically by categorising participants into deciles of predicted five-year CVD risk and plotting mean predicted five-year CVD risk against observed CVD events at five years of follow up, obtained by the Kaplan-Meier method.15 For PCEs we used recalibrated models where the baseline survival values were estimated by fitting Cox models with the prognostic index from the PCE model (offset term) in the PREDICT-CVD dataset.16 Discrimination was assessed using Harrell’s C statistic and Royston and Sauerbrei’s D statistic.17,18 The proportion of outcome variation explained by PREDICT-CVD and PCEs was assessed using Royston and Sauerbrei’s R2 statistic.18 All analyses were performed using Stata 15.0 software.19

Ethics approval

Approved by the Northern Region Ethics Committee Y in 2003 (AKY/03/12/314) with subsequent annual approval by the National Multi Region Ethics Committee since 2007 (MEC07/19/EXP).

Results

After applying exclusion criteria, 40,161 participants aged 70 years or over had a baseline CVD risk assessment between 31 October 2004 and 30 December 2016 (Figure 1, Table 1).

Figure 1: Study exclusions and incidence of CVD events during follow up.

*Excluded if ethnicity recorded as Middle Eastern, Latin American, African or recorded as ‘other’ or ‘unknown’ ‘MELAA’, ‘other’ or ‘unknown’ (in 70+ n=542).

Table 1: Description of the PREDICT cohort aged 70 years and over.

*IQR: interquartile range; SD: standard deviation; CVD: cardiovascular disease; NZ: New Zealand; TC/HDL: Total Cholesterol to HDL Cholesterol ratio.

During 185,150 person-years follow-up (mean follow-up time 4.6 years), 5,948 (15%) experienced an incident CVD event of which 1,065 (18%) were fatal. The Appendix describes the number and type of CVD event; mostly due to MI (1,690 events; 28.4%), ischaemic stroke (1,154 events; 19.4%) and heart failure (1,107 events; 18.5%). An additional 3,932 people (10%) died from non-CVD causes. The incidence of CVD and fatal non-CVD events increased markedly with increasing five-year age bands.

The majority of the cohort were women (57%), European (76%) and non-smokers (74%). Just over a third were resident in the two most deprived quintiles. In terms of comorbidity, 6% had a history of atrial fibrillation, 18% had diabetes and 18% had a Charlson comorbidity index of one or more. Overall, 51% of the cohort were on BP-lowering medications, 32% on lipid-lowering and 29% on antithrombotic medications. With increasing age there was an increase in the proportion of women, those of European ethnicity, non-smokers, those dispensed BP-lowering medications and people with atrial fibrillation, diabetes and a comorbidity score of one or more, as well as an increase in mean SBP. Only the mean TC/HDL and proportion dispensed lipid-lowering medications decreased with age.

Calibration graphs by decile of predicted risk versus observed five-year CVD event risk for the New Zealand PREDICT-CVD equations and recalibrated PCEs are shown for women (Figure 2) and men (Figure 3) by five-year age groups. In the over 80-year age groups, some deciles could not be plotted according to observed event rate due to insufficient numbers with follow-up at five years.

Figure 2: Calibration plots by decile of predicted risk and observed CVD event risk at five years according to PREDICT-CVD, PCE and recalibrated PCE for women by age group.

A diagonal line with intercept of 0 and slope of 1 represents perfect calibration. Plotted risk deciles below the diagonal represent an underestimate of predicted risk, above the diagonal, an overestimate.

Figure 3: Calibration graphs by decile of predicted risk and observed CVD event risk at five years according PREDICT-CVD, PCE and recalibrated PCE for men by age group.

A diagonal line with intercept of 0 and slope of 1 represents perfect calibration. Plotted risk deciles below the diagonal represent an underestimate of predicted risk, above the diagonal, an overestimate.

For women, PREDICT-CVD was well calibrated for those aged 70–74 years but underestimated CVD risk for women aged 75–84 years with the exception of those in the highest deciles of risk. Among those aged 85 years and over, underestimation of risk was also observed as well as suggesting poor discrimination by decile of risk (ie, risk deciles clustered over a similar observed five-year event rate). In contrast, the recalibrated PCEs overestimated CVD risk across all age groups. For those over 80 years, while predicted risk deciles were well separated, they were also stacked above the same observed event rate, indicating poor discrimination as well as poor calibration.

For men, PREDICT-CVD was reasonably well calibrated for those aged 70–79 years but underestimated CVD risk for men aged 80–84 years with the exception of those in the three highest deciles of risk, and underestimated CVD risk for all aged 85 years and over as well as showing poor discrimination. The PCEs overestimated CVD risk across all age groups from 70–84 years. However, in the 85+ age group, in the deciles with sufficient participants at five years of follow-up, overestimation with the recalibrated PCEs was less marked than at younger age groups.

Tables 2 and 3 summarise the discrimination metrics by age group and sex for PREDICT-CVD and PCEs. The discrimination and overall performance were generally poor for both equations in people over age 75 years.

Table 2: Performance metrics of PREDICT-CVD models, by sex and age group.

Table 3: Performance metrics of PCE models, by sex and age group.

Discussion

This study investigated the performance of contemporary CVD risk prediction equations in a cohort of 40,161 ambulatory people aged 70 years and over who had a heart health check, while visiting their GP. Over a third of the cohort were aged over 74 years. While both the calibration and discrimination performance of the equations varied, in general they performed increasingly poorly in people over 75 years of age, particularly the PCEs.

This is the first study comparing the performance of contemporary CVD risk equations in a cohort of older people whose GPs had decided to risk assess them in a routine practice setting. Previous studies have been based on either total general practice population samples,20 population-based health surveys21 or combined cohort studies populations.9,22 Our study reported comorbidity and non-CVD deaths across our total cohort and in five-year age bands. While some comorbidities have been reported in previous evaluations of some equations,20 none have reported how these factors, that are likely to influence the accuracy of risk assessment in the elderly, change with increasing age. Furthermore, previous evaluations of equation performance in older people have used wide age bands, which can mask significant differences by age, given the diminishing numbers of people in increasingly older age bands.20,22

There are other CVD risk prediction equations recommended for use in older patients. These include QRISK3 developed by UK researchers for people aged 25–84 years;20 Systematic COronary Risk Evaluation in older people (SCORE O.P.) developed by European researchers for people aged 65–80 years;22 and the Canadian CVD Population Risk Tool (CVDPoRT) for ages 20–105.21 The QRISK3 model includes over 20 predictor variables,20 many of which we were unable to incorporate given our more limited CVD profile data. Similarly, the Canadian CVDPoRT equation, derived from large population health surveys includes many lifestyle factors also not captured in our primary care-derived dataset.21 We considered assessing the performance of SCORE O.P. equations, derived in people aged 65–80 years, but it only predicts CVD mortality and many patients are also concerned about the impact of non-fatal major CVD events. Moreover, diabetes is not included as a predictor, yet a quarter of the PREDICT cohort over 75 years had diabetes.

It has previously been reported that CVD risk prediction equations developed and validated in younger age cohorts may not perform well when applied to populations aged 75 and older due to competing risks.23 Most equations do not incorporate the effect of other comorbid conditions or polypharmacy on CVD risk and competing risks of death from cancer or dementia. These competing risk events (non-CVD death events, which preclude an individual from experiencing a CVD event) become increasingly important with age. Indeed 10% of our study cohort died from non-CVD causes. For older age cohorts, where mortality rates are comparatively high, equations that treat non-CVD events as competing risks are likely to be needed to achieve more accurate risk prediction.23

The major strength of this study is also its major limitation. Participants were those older people, who, in the clinical judgement of primary care clinicians, were suitable for a routine preventive CVD risk assessment. While 90% of all New Zealanders aged 30–74 years have had a CVD risk assessment,24 many of the older people in our cohort have been risk assessed largely at the discretion of their primary care provider. Therefore they are not representative of all older people in the study region, because risk assessment would not be clinically appropriate for many of those not included (eg, those with dementia or requiring palliative care). We estimated that 50% of people aged 75–79 years, 30% aged 80–84 years and 25% of those aged 85 years and over, who did not have prior CVD, were included in our study. A further limitation is the use of the Charlson comorbidity index, modified to the extent that only 9 of the 12 comorbidities could be present in the study cohort (heart failure, stroke, renal disease being excluded). While the index has been validated in a New Zealand population, it suffers from including a very limited range of hospitalised-only long-term conditions and therefore underestimates the true multimorbidity burden in primary care. Indeed the prevalence of multimorbidity in the 65–84 year age group has been found to be as much as 65% in Scottish general practices.25 This is particularly relevant as most CVD risk prediction equations, and CVD risk management guidelines, tend to take a narrow disease-focused approach.26

This paper poses a series of clinical implications to current CVD risk prediction practice. Many older people are still engaged in the workforce and physically active. In the current Ministry of Health CVD guidance,6 healthy people over 75 years with few comorbidities and an estimated life expectancy of more than five years, CVD risk assessment using the PREDICT equations are recommended as well as discussing the same management options as for people under 75 years of age. However, although CVD risk factors have similar effects in those under and over 75 years,22,27 risk assessment and management is more complex for older age groups as health status, physical and cognitive functioning varies greatly.28 The risk of other long-term conditions increases with age25 and this in turn is associated with polypharmacy and complicated medication regimens.29,30 In this context, the risks and benefits of CVD risk-reducing interventions is accompanied by less certainty. Some treatment-related risks will increase, such as bleeding with aspirin, requiring clinicians to vary their advice.30 Furthermore, general health and functioning such as frailty, cognitive impairment, quality of life and personal preferences need to be taken into account. Older people may be more concerned about the risk of stroke than MI, as stroke may result in mental and physical disability and loss of independence, so single CVD outcomes (eg, stroke), as well as composite CVD outcomes, may be useful to guide discussions. The emerging guidance on how best to manage multimorbidity might offer a way forward here, with its focus on realistic treatment goals shared between clinician and patient and the need to recognise preference sensitive decisions (eg, medication that may benefit one condition but may make another worse).31–33

From this study of presumed healthy older people being risk assessed in general practice, we have found that the performance of the CVD equations derived mainly in people under age 75 years need to be improved to support clinical decision-making for people aged 75 years and over. We recommend that CVD equations used in people over 75 years incorporate factors such as multimorbidity and competing risks with additional risk-benefit tools taking into account physical and cognitive functioning and patient preferences.

Appendix

Number and type of incident CVD events in the PREDICT-1° patients, 70+ years old men and women

a If a patient had more than one type of CVD event, only the first was counted.

Medications available in New Zealand according to the CVD treatment categories of interest

*Alpha blockers, loop diuretics (bumetanide, frusemide), metolazone and spironolactone excluded as primary indication not usually to reduce blood pressure.

International Classification of Disease-10-Australian Modification (ICD-10-AM) codes for the PREDICT-1° CVD events outcome, from hospital discharge and mortality records

International Classification of Disease-10-Australian Modification (ICD-10-AM) codes for the PCEs hard atherosclerotic CVD outcome,* from hospital discharge and mortality records.
*Both fatal and nonfatal events with myocardial infarction and stroke codes were included but only fatal events with ‘Other coronary heart disease’ codes.

ICD 10 codes included in the definition of prior CVD

Summary

Abstract

Aim

To investigate how well the New Zealand PREDICT-CVD risk equations, derived in people aged 30–74 years and US Pooled Cohort Equations (PCEs) derived in people aged 40–79 years, perform for older people.

Method

The PREDICT cohort study automatically recruits participants when clinicians use PREDICT software to conduct a CVD risk assessment. We identified patients aged 70 years and over, without prior CVD, renal disease or heart failure who had been risk assessed between 2004 and 2016. Equation performance was assessed in five-year age bands using calibration graphs and standard discrimination metrics.

Results

40,161 patients (median 73 years; IQR 71–77) experienced 5,948 CVD events during 185,150 person-years follow-up. PREDICT-CVD equations were well calibrated in 70–74 year olds but underestimated events for women from 75 years and men from 80 years. Discrimination metrics were also poor for these age groups. Recalibrated PCEs overestimated CVD risk in both sexes and had poor discrimination from age 70 years for men and from age 75 years for women.

Conclusion

While PREDICT-CVD equations performed better than PCEs, neither performed well. Multimorbidity and competing risks are likely to contribute to the poor performance and new CVD risk equations need to include these factors.

Author Information

Sue Wells, School of Population Health, University of Auckland, Auckland; Romana Pylypchuk, School of Population Health, University of Auckland, Auckland; Suneela Mehta, School of Population Health, University of Auckland, Auckland; Andrew Kerr, Cardiology Department, Middlemore Hospital, Auckland; Vanessa Selak, School of Population Health, University of Auckland, Auckland; Katrina Poppe, Department of Medicine, University of Auckland, Auckland; Corina Grey, School of Population Health, University of Auckland, Auckland; Rod Jackson, School of Population Health, University of Auckland, Auckland.

Acknowledgements

The authors thank the primary health care organisations, affiliated primary care physicians, nurses and patients for their contributions to this study.

Correspondence

A/P Sue Wells, School of Population Health, University of Auckland, Private Bag 92019 Auckland Mail Centre, Auckland 1142.

Correspondence Email

s.wells@auckland.ac.nz

Competing Interests

All authors report grants from Health Research Council of New Zealand during the conduct of the study; Drs Wells, Poppe, Jackson and Corina report grants from Heart Foundation of New Zealand during the conduct of the study; Drs Wells and Jackson report grants from New Zealand Ministry of Business, Innovation and Enterprise during the conduct of the study; Dr Wells reports grants from Stevenson Foundation during the conduct of the study.

1. Prince MJ, Wu F, Guo Y, et al. The burden of disease in older people and implications for health policy and practice. Lancet 2015; 385(9967):549–62.

2. Hermanson B, Omenn GS, Kronmal RA, et al. Beneficial six-year outcome of smoking cessation in older men and women with coronary artery disease. Results from the CASS registry. New England Journal of Medicine 1988; 319(21):1365–9.

3. Blood Pressure Lowering Treatment Trialists C, Sundstrom J, Arima H, et al. Blood pressure-lowering treatment based on cardiovascular risk: a meta-analysis of individual patient data. Lancet 2014; 384(9943):591–8.

4. Cholesterol Treatment Trialists C. Efficacy and safety of statin therapy in older people: a meta-analysis of individual participant data from 28 randomised controlled trials. Lancet 2019; 393(10170):407–15.

5. Gueyffier F, Bulpitt C, Boissel JP, et al. Antihypertensive drugs in very old people: a subgroup meta-analysis of randomised controlled trials. INDANA Group. Lancet 1999; 353(9155):793–6.

6. Ministry of Health. Cardiovascular disease risk assessment and management for primary care. Wellington: Ministry of Health, 2018.

7. Wells S, Riddell T, Kerr A, et al. Cohort Profile: The PREDICT Cardiovascular Disease Cohort in New Zealand Primary Care (PREDICT-CVD 19). International Journal of Epidemiology 2015; doi: 10.1093/ije/dyv312 doi: doi: 10.1093/ije/dyv312

8. Pylypchuk R, Wells S, Kerr A, et al. Cardiovascular disease risk prediction equations in 400,000 primary care patients in New Zealand: a derivation and validation study. Lancet 2018; 391(10133):1897–907.

9. Goff DC, Jr., Lloyd-Jones DM, Bennett G, et al. 2013 ACC/AHA guideline on the assessment of cardiovascular risk: a report of the American College of Cardiology/American Heart Association Task Force on Practice Guidelines. Journal of the American College of Cardiology 2014; 63(25 Pt B):2935–59.

10. Ministry of Health. National Health Index Wellington2015 [Available from: http://www.health.govt.nz/our-work/health-identity/national-health-index accessed February 5 2015.

11. Chan WC, Wright C, Tobias M, et al. Explaining trends in coronary heart disease hospitalisations in New Zealand: trend for admissions and incidence can be in opposite directions. Heart 2008; 94(12):1589–93.

12. Ministry of Health. Ethnicity data protocols for the Health and Disability Sector. Wellington: Ministry of Health, 2004.

13. Salmond C, Crampton P, Atkinson J. NZDep2006 Index of Deprivation User’s Manual. Wellington: ; 2007. Wellington: Department of Public Health, Wellington School of Medicine and Health Sciences, 2007.

14. Quan H, Li B, Couris CM, et al. Updating and validating the Charlson comorbidity index and score for risk adjustment in hospital discharge abstracts using data from 6 countries. American Journal of Epidemiology 2011; 173(6):676–82.

15. Kaplan E, Meier P. Nonparametric estimation from incomplete observations. J Am Stat Assoc 1958; 53:457–81.

16. Crowson C, Atkinson E, Therneau T. Assessing Calibration of Prognostic Risk Scores. Stat Methods Med Res 2016; 25(4):1692–706. doi: . doi:10.1177/0962

280213497434.

17. Harrell F, Califf R, Prior D, et al. Evaluating the yield of medical tests. Jounal of American Medical Association 1982; 247:2543–6.

18. Royston P, Sauerbrei W. A new measure of prognostic separation in survival data. Stat Med 2004; 23(5):723–48.

19. StataCorp. Stata Statistical Software. Release 15. College Station, TX: StatCorp LP, 2015.

20. Hippisley-Cox J, Coupland C, Brindle P. Development and validation of QRISK3 risk prediction algorithms to estimate future risk of cardiovascular disease: prospective cohort study. BMJ 2017; 357:j2099. doi: http://dx.doi.org/10.1136/bmj.j2099

21. Manuel DG, Tuna M, Bennett C, et al. Development and validation of a cardiovascular disease risk-prediction model using population health surveys: the Cardiovascular Disease Population Risk Tool (CVDPoRT). CMAJ Canadian Medical Association Journal 2018; 190(29):E871–E82.

22. Cooney MT, Selmer R, Lindman A, et al. Cardiovascular risk estimation in older persons: SCORE O.P. European Journal of Preventive Cardiology 2016; 23(10):1093–103.

23. Wolbers M, Koller MT, Witteman JC, et al. Prognostic models with competing risks: methods and application to coronary risk prediction. Epidemiology 2009; 20(4):555–61. doi: 10.1097/EDE.0b013e3181a39056 [published Online First: 2009/04/16]

24. Ministry of Health. How is My PHO performing? 2015/16 Quarter 4 (January to March) results summary. ed. Wellington: Ministry of Health, 2015.

25. Barnett K, Mercer SW, Norbury M, et al. Epidemiology of multimorbidity and implications for health care, research, and medical education: a cross-sectional study. Lancet 2012; 380(9836):37–43.

26. Forman DE, Maurer MS, Boyd C, et al. Multimorbidity in Older Adults With Cardiovascular Disease. Journal of the American College of Cardiology 2018; 71(19):2149–61.

27. Benetos A, Thomas F, Bean KE, et al. Role of modifiable risk factors in life expectancy in the elderly. Journal of Hypertension 2005; 23(10):1803–8.

28. Lowsky DJ, Olshansky SJ, Bhattacharya J, et al. Heterogeneity in healthy aging. J Gerontol A Biol Sci Med Sci 2014; 69(6):640–9. doi: 10.1093/gerona/glt162 [published Online First: 2013/11/20]

29. Calderon-Larranaga A, Poblador-Plou B, Gonzalez-Rubio F, et al. Multimorbidity, polypharmacy, referrals, and adverse drug events: are we doing things well? British Journal of General Practice 2012; 62(605):e821–6. doi: http://dx.doi.org/10.3399/bjgp12X659295

30. Wolff JL, Starfield B, Anderson G. Prevalence, expenditures, and complications of multiple chronic conditions in the elderly. Archives of Internal Medicine 2002; 162(20):2269–76.

31. American Geriatrics Society. Guiding principles for the care of older adults with multimorbidity: an approach for clinicians: American Geriatrics Society Expert Panel on the Care of Older Adults with Multimorbidity. J Am Geriatr Soc 2012; 60(10):E1–E25. doi: http://dx.doi.org/10.1111/j.1532-5415.2012.04188.x

32. Muth C, van den Akker M, Blom JW, et al. The Ariadne principles: how to handle multimorbidity in primary care consultations. BMC Med 2014; 12:223. doi: http://dx.doi.org/10.1186/s12916-014-0223-1

33. Stokes T. Multimorbidity and clinical guidelines: problem or opportunity? New Zealand Medical Journal 2018; 131(1472):7–9.

Contact diana@nzma.org.nz
for the PDF of this article

View Article PDF

Cardiovascular disease (CVD) is the leading cause of potentially preventable global health loss and demand for health and disability services for older people.1 There is evidence that reducing smoking,2 blood pressure (BP)3 and lipids4 is associated with reduced fatal and non-fatal CVD for adults at any age, and the benefits are largely determined by patients’ pre-treatment CVD risk. As older people are more likely than younger people to be at high CVD risk, they are also likely to benefit most from CVD risk-reducing medications.1 A recent individual person meta-analysis of 28 statin trials found that treatment produced a similar reduction in major vascular events per mmol/L reduction in low-density lipoprotein (LDL) cholesterol irrespective of age, although findings were more attenuated for those over 75 years without established CVD.4 In the case of BP-lowering, the benefits of treatment accrue even among the very old. In a meta-analysis involving only octogenarians from seven clinical trials, BP-lowering medications were associated with lower rates of stroke (34%), heart failure (39%) and major CVD events (22%) than those not receiving treatment.5 Despite these findings, most CVD risk management guidelines are vague about how to manage older people. In New Zealand, for example, CVD risk assessment and management guidelines recommend a formal quantitative CVD risk assessment for people aged 30–74 years,6 but once a person turns 75 years of age, risk assessment is ‘at the discretion’ of the clinician. General practitioners (GPs) are given general advice to use clinical judgement taking into account the results of a risk assessment, the likely benefits and risks of treatment and patient preferences.

We have identified an increasing number of New Zealanders aged 75 years and over receiving formal quantitative CVD risk assessments in routine primary care and have generated a cohort of these older people as part of our ongoing PREDICT-CVD risk prediction cohort study.7 Given the clinical relevance of accurate assessment of CVD risk in older people, we aimed to investigate how well the recently developed PREDICT-CVD equations8 (derived in people aged 30–74 years), performed in older people. As a comparison, we also assessed the performance of the relatively similar US Pooled Cohort Equations (PCEs),9 derived in people aged 40–79 years. We hypothesised that these CVD risk equations should perform reasonably well in the subgroup of older patients who GPs considered suitable for routine preventive CVD risk assessment.

Methods

Study population

The PREDICT study has been described in full elsewhere.7 In brief, patients are automatically recruited to this prospective open cohort study when primary care clinicians undertake standard quantitative CVD risk assessments using the PREDICT web-based decision support programme. Over one-third of GPs use PREDICT software, which is integrated into their electronic patient management systems. An encrypted version of each person’s unique national health identifier (National Health Index number, eNHI) is used to anonymously link patients’ risk profiles to national and regional health datasets, including all community pharmaceutical dispensing, all community laboratory testing, state-funded hospitalisations and all deaths.10 Over 98% of New Zealanders have an NHI number,10 allowing identification and linkage of multiple health contacts, augmentation of risk factor data (prior hospital admissions, pharmaceutical dispensing, laboratory test results) as well as health outcome ascertainment (fatal CVD events in-hospital and out-of-hospital and non-fatal hospital admissions for acute CVD events) during follow-up. Over 95% of CVD hospitalisations occur within our state-funded health services.11

For the purposes of this study, eligible patients were those who had a first (baseline) CVD assessment from 31 October 2004 to 30 December 2016 and were aged 70 years or older, unless they met any of the following exclusion criteria: prior history of ischaemic CVD, heart failure, renal disease or missing risk factors needed for CVD risk prediction models. In addition, to emulate the cohort used to develop the PREDICT CVD equations,8 patients were excluded if their self-identified ethnicity was recorded as Middle Eastern, Latin American, African or recorded as ‘other’ or ‘unknown’. The rationale for this exclusion was that these ethnic groups were too heterogeneous to combine into one category and too few in numbers to disaggregate into meaningful subgroups.

A history of prior CVD was classified according to an International Classification of Diseases, version 10 Australian Modification (ICD-10 AM) for hospitalisations or primary care clinical diagnosis at the time of CVD risk assessment for angina, myocardial infarction (MI), percutaneous coronary intervention (PCI), coronary artery bypass graft (CABG), ischaemic stroke, transient ischaemic attack (TIA), or peripheral vascular disease (PVD). The capture of a patient’s history of a hospitalised event used data available from 1 January 1988. (Appendix contains full list of ICD-10 codes) Patients who were dispensed anti-anginal medications on at least three occasions up to five years prior to their baseline visit were also excluded. Renal disease was determined either by an estimated glomerular filtration rate (eGFR) of less than 30ml/min per 1.73m2, an ICD-10 AM hospitalisation for renal dialysis, prior renal transplantation or a recording of diabetes with nephropathy at the time of CVD risk assessment. Heart failure diagnoses were based on ICD-10 AM hospitalisation code for heart failure or if participants had been dispensed a loop diuretic three or more times in the preceding five years.

Ethnicity classification was based on a nationally agreed prioritisation algorithm when individuals identified with more than one ethnicity12 in the following order; Māori, Pacific, Indian, Chinese/other Asian and European. Socio-economic status was assessed using the NZ Deprivation Index (NZDep), a measure assigned to patients according to the deprivation score of their area of residence.13 For these analyses, NZDep was divided into quintiles from 1 (least deprived) to 5 (most deprived).

Smoking status was defined as either current smoker (including recently quit in the last 12 months) or non-smoker. Diabetes status was classified according to ICD-10 hospitalisation with diabetes and/or dispensing of at least one diabetes medication in the last six months and/or recorded as such by their primary care clinician at the time of CVD risk assessment.

The Charlson comorbidity index is a weighted scoring system that assesses the degree of previously hospitalised comorbidity burden. It is based on 12 conditions that predict one-year survival and has been adapted for use with hospitalisation data using a well-validated ICD-10 coding algorithm.14 Comorbidities were identified from hospitalisations up to five years prior to the first CVD risk assessment.

The pharmaceutical collection (PHARMS) is a national database of community pharmaceutical dispensing. Reliable identification of dispensing episodes by eNHI has increased over the last decade from 64% in 2004, to 92% in 2006 and over 96% from 2009 onwards.7 PHARMS was used to identify patients who were dispensed one of the following medications on at least one occasion in the six months prior to the baseline CVD risk assessment: BP-lowering, lipid-lowering and antiplatelet/anticoagulant medications (henceforth termed antithrombotic medications). All these medications are government subsidised. (CVD medications are listed in the Appendix).

Outcomes during follow-up

CVD outcomes for the PREDICT-CVD equations8 were defined as ICD-10-AM coded hospitalisation or death from ischaemic heart disease, ischaemic or haemorrhagic cerebrovascular events (including TIA), PVD or heart failure. CVD outcomes for the American PCE Equations9 (termed hard atherosclerotic CVD) are a subset of the former and include fatal or nonfatal MI, fatal or nonfatal stoke, or CHD death (Appendix contains ICD-10-AM codes for both sets of outcomes). Time on study was the time from baseline CVD risk assessment to the first of the following: hospital admission or death related to CVD, death from other causes or end of follow-up.

Statistical analysis

The distributions of CVD risk factors, event rates and follow up were investigated for the total population and by five-year age groups; 70–74 years, 75–79 years, 80–84 years and 85 years and over. Calibration was tested separately for PREDICT-CVD8 and the PCE models9 by five-year age groups using equation-specific outcomes. The PREDICT-CVD risk models include age, ethnicity, deprivation, diabetes status, history of atrial fibrillation, smoking status, systolic blood pressure (SBP), the ratio of total cholesterol (TC) to high-density lipoprotein (HDL) cholesterol (TC/HDL) and prior dispensing of BP lowering, lipid lowering and antithrombotic medications. The PCEs include age, TC, HDL, smoking and diabetes status, SBP and treated SBP. Calibration was assessed graphically by categorising participants into deciles of predicted five-year CVD risk and plotting mean predicted five-year CVD risk against observed CVD events at five years of follow up, obtained by the Kaplan-Meier method.15 For PCEs we used recalibrated models where the baseline survival values were estimated by fitting Cox models with the prognostic index from the PCE model (offset term) in the PREDICT-CVD dataset.16 Discrimination was assessed using Harrell’s C statistic and Royston and Sauerbrei’s D statistic.17,18 The proportion of outcome variation explained by PREDICT-CVD and PCEs was assessed using Royston and Sauerbrei’s R2 statistic.18 All analyses were performed using Stata 15.0 software.19

Ethics approval

Approved by the Northern Region Ethics Committee Y in 2003 (AKY/03/12/314) with subsequent annual approval by the National Multi Region Ethics Committee since 2007 (MEC07/19/EXP).

Results

After applying exclusion criteria, 40,161 participants aged 70 years or over had a baseline CVD risk assessment between 31 October 2004 and 30 December 2016 (Figure 1, Table 1).

Figure 1: Study exclusions and incidence of CVD events during follow up.

*Excluded if ethnicity recorded as Middle Eastern, Latin American, African or recorded as ‘other’ or ‘unknown’ ‘MELAA’, ‘other’ or ‘unknown’ (in 70+ n=542).

Table 1: Description of the PREDICT cohort aged 70 years and over.

*IQR: interquartile range; SD: standard deviation; CVD: cardiovascular disease; NZ: New Zealand; TC/HDL: Total Cholesterol to HDL Cholesterol ratio.

During 185,150 person-years follow-up (mean follow-up time 4.6 years), 5,948 (15%) experienced an incident CVD event of which 1,065 (18%) were fatal. The Appendix describes the number and type of CVD event; mostly due to MI (1,690 events; 28.4%), ischaemic stroke (1,154 events; 19.4%) and heart failure (1,107 events; 18.5%). An additional 3,932 people (10%) died from non-CVD causes. The incidence of CVD and fatal non-CVD events increased markedly with increasing five-year age bands.

The majority of the cohort were women (57%), European (76%) and non-smokers (74%). Just over a third were resident in the two most deprived quintiles. In terms of comorbidity, 6% had a history of atrial fibrillation, 18% had diabetes and 18% had a Charlson comorbidity index of one or more. Overall, 51% of the cohort were on BP-lowering medications, 32% on lipid-lowering and 29% on antithrombotic medications. With increasing age there was an increase in the proportion of women, those of European ethnicity, non-smokers, those dispensed BP-lowering medications and people with atrial fibrillation, diabetes and a comorbidity score of one or more, as well as an increase in mean SBP. Only the mean TC/HDL and proportion dispensed lipid-lowering medications decreased with age.

Calibration graphs by decile of predicted risk versus observed five-year CVD event risk for the New Zealand PREDICT-CVD equations and recalibrated PCEs are shown for women (Figure 2) and men (Figure 3) by five-year age groups. In the over 80-year age groups, some deciles could not be plotted according to observed event rate due to insufficient numbers with follow-up at five years.

Figure 2: Calibration plots by decile of predicted risk and observed CVD event risk at five years according to PREDICT-CVD, PCE and recalibrated PCE for women by age group.

A diagonal line with intercept of 0 and slope of 1 represents perfect calibration. Plotted risk deciles below the diagonal represent an underestimate of predicted risk, above the diagonal, an overestimate.

Figure 3: Calibration graphs by decile of predicted risk and observed CVD event risk at five years according PREDICT-CVD, PCE and recalibrated PCE for men by age group.

A diagonal line with intercept of 0 and slope of 1 represents perfect calibration. Plotted risk deciles below the diagonal represent an underestimate of predicted risk, above the diagonal, an overestimate.

For women, PREDICT-CVD was well calibrated for those aged 70–74 years but underestimated CVD risk for women aged 75–84 years with the exception of those in the highest deciles of risk. Among those aged 85 years and over, underestimation of risk was also observed as well as suggesting poor discrimination by decile of risk (ie, risk deciles clustered over a similar observed five-year event rate). In contrast, the recalibrated PCEs overestimated CVD risk across all age groups. For those over 80 years, while predicted risk deciles were well separated, they were also stacked above the same observed event rate, indicating poor discrimination as well as poor calibration.

For men, PREDICT-CVD was reasonably well calibrated for those aged 70–79 years but underestimated CVD risk for men aged 80–84 years with the exception of those in the three highest deciles of risk, and underestimated CVD risk for all aged 85 years and over as well as showing poor discrimination. The PCEs overestimated CVD risk across all age groups from 70–84 years. However, in the 85+ age group, in the deciles with sufficient participants at five years of follow-up, overestimation with the recalibrated PCEs was less marked than at younger age groups.

Tables 2 and 3 summarise the discrimination metrics by age group and sex for PREDICT-CVD and PCEs. The discrimination and overall performance were generally poor for both equations in people over age 75 years.

Table 2: Performance metrics of PREDICT-CVD models, by sex and age group.

Table 3: Performance metrics of PCE models, by sex and age group.

Discussion

This study investigated the performance of contemporary CVD risk prediction equations in a cohort of 40,161 ambulatory people aged 70 years and over who had a heart health check, while visiting their GP. Over a third of the cohort were aged over 74 years. While both the calibration and discrimination performance of the equations varied, in general they performed increasingly poorly in people over 75 years of age, particularly the PCEs.

This is the first study comparing the performance of contemporary CVD risk equations in a cohort of older people whose GPs had decided to risk assess them in a routine practice setting. Previous studies have been based on either total general practice population samples,20 population-based health surveys21 or combined cohort studies populations.9,22 Our study reported comorbidity and non-CVD deaths across our total cohort and in five-year age bands. While some comorbidities have been reported in previous evaluations of some equations,20 none have reported how these factors, that are likely to influence the accuracy of risk assessment in the elderly, change with increasing age. Furthermore, previous evaluations of equation performance in older people have used wide age bands, which can mask significant differences by age, given the diminishing numbers of people in increasingly older age bands.20,22

There are other CVD risk prediction equations recommended for use in older patients. These include QRISK3 developed by UK researchers for people aged 25–84 years;20 Systematic COronary Risk Evaluation in older people (SCORE O.P.) developed by European researchers for people aged 65–80 years;22 and the Canadian CVD Population Risk Tool (CVDPoRT) for ages 20–105.21 The QRISK3 model includes over 20 predictor variables,20 many of which we were unable to incorporate given our more limited CVD profile data. Similarly, the Canadian CVDPoRT equation, derived from large population health surveys includes many lifestyle factors also not captured in our primary care-derived dataset.21 We considered assessing the performance of SCORE O.P. equations, derived in people aged 65–80 years, but it only predicts CVD mortality and many patients are also concerned about the impact of non-fatal major CVD events. Moreover, diabetes is not included as a predictor, yet a quarter of the PREDICT cohort over 75 years had diabetes.

It has previously been reported that CVD risk prediction equations developed and validated in younger age cohorts may not perform well when applied to populations aged 75 and older due to competing risks.23 Most equations do not incorporate the effect of other comorbid conditions or polypharmacy on CVD risk and competing risks of death from cancer or dementia. These competing risk events (non-CVD death events, which preclude an individual from experiencing a CVD event) become increasingly important with age. Indeed 10% of our study cohort died from non-CVD causes. For older age cohorts, where mortality rates are comparatively high, equations that treat non-CVD events as competing risks are likely to be needed to achieve more accurate risk prediction.23

The major strength of this study is also its major limitation. Participants were those older people, who, in the clinical judgement of primary care clinicians, were suitable for a routine preventive CVD risk assessment. While 90% of all New Zealanders aged 30–74 years have had a CVD risk assessment,24 many of the older people in our cohort have been risk assessed largely at the discretion of their primary care provider. Therefore they are not representative of all older people in the study region, because risk assessment would not be clinically appropriate for many of those not included (eg, those with dementia or requiring palliative care). We estimated that 50% of people aged 75–79 years, 30% aged 80–84 years and 25% of those aged 85 years and over, who did not have prior CVD, were included in our study. A further limitation is the use of the Charlson comorbidity index, modified to the extent that only 9 of the 12 comorbidities could be present in the study cohort (heart failure, stroke, renal disease being excluded). While the index has been validated in a New Zealand population, it suffers from including a very limited range of hospitalised-only long-term conditions and therefore underestimates the true multimorbidity burden in primary care. Indeed the prevalence of multimorbidity in the 65–84 year age group has been found to be as much as 65% in Scottish general practices.25 This is particularly relevant as most CVD risk prediction equations, and CVD risk management guidelines, tend to take a narrow disease-focused approach.26

This paper poses a series of clinical implications to current CVD risk prediction practice. Many older people are still engaged in the workforce and physically active. In the current Ministry of Health CVD guidance,6 healthy people over 75 years with few comorbidities and an estimated life expectancy of more than five years, CVD risk assessment using the PREDICT equations are recommended as well as discussing the same management options as for people under 75 years of age. However, although CVD risk factors have similar effects in those under and over 75 years,22,27 risk assessment and management is more complex for older age groups as health status, physical and cognitive functioning varies greatly.28 The risk of other long-term conditions increases with age25 and this in turn is associated with polypharmacy and complicated medication regimens.29,30 In this context, the risks and benefits of CVD risk-reducing interventions is accompanied by less certainty. Some treatment-related risks will increase, such as bleeding with aspirin, requiring clinicians to vary their advice.30 Furthermore, general health and functioning such as frailty, cognitive impairment, quality of life and personal preferences need to be taken into account. Older people may be more concerned about the risk of stroke than MI, as stroke may result in mental and physical disability and loss of independence, so single CVD outcomes (eg, stroke), as well as composite CVD outcomes, may be useful to guide discussions. The emerging guidance on how best to manage multimorbidity might offer a way forward here, with its focus on realistic treatment goals shared between clinician and patient and the need to recognise preference sensitive decisions (eg, medication that may benefit one condition but may make another worse).31–33

From this study of presumed healthy older people being risk assessed in general practice, we have found that the performance of the CVD equations derived mainly in people under age 75 years need to be improved to support clinical decision-making for people aged 75 years and over. We recommend that CVD equations used in people over 75 years incorporate factors such as multimorbidity and competing risks with additional risk-benefit tools taking into account physical and cognitive functioning and patient preferences.

Appendix

Number and type of incident CVD events in the PREDICT-1° patients, 70+ years old men and women

a If a patient had more than one type of CVD event, only the first was counted.

Medications available in New Zealand according to the CVD treatment categories of interest

*Alpha blockers, loop diuretics (bumetanide, frusemide), metolazone and spironolactone excluded as primary indication not usually to reduce blood pressure.

International Classification of Disease-10-Australian Modification (ICD-10-AM) codes for the PREDICT-1° CVD events outcome, from hospital discharge and mortality records

International Classification of Disease-10-Australian Modification (ICD-10-AM) codes for the PCEs hard atherosclerotic CVD outcome,* from hospital discharge and mortality records.
*Both fatal and nonfatal events with myocardial infarction and stroke codes were included but only fatal events with ‘Other coronary heart disease’ codes.

ICD 10 codes included in the definition of prior CVD

Summary

Abstract

Aim

To investigate how well the New Zealand PREDICT-CVD risk equations, derived in people aged 30–74 years and US Pooled Cohort Equations (PCEs) derived in people aged 40–79 years, perform for older people.

Method

The PREDICT cohort study automatically recruits participants when clinicians use PREDICT software to conduct a CVD risk assessment. We identified patients aged 70 years and over, without prior CVD, renal disease or heart failure who had been risk assessed between 2004 and 2016. Equation performance was assessed in five-year age bands using calibration graphs and standard discrimination metrics.

Results

40,161 patients (median 73 years; IQR 71–77) experienced 5,948 CVD events during 185,150 person-years follow-up. PREDICT-CVD equations were well calibrated in 70–74 year olds but underestimated events for women from 75 years and men from 80 years. Discrimination metrics were also poor for these age groups. Recalibrated PCEs overestimated CVD risk in both sexes and had poor discrimination from age 70 years for men and from age 75 years for women.

Conclusion

While PREDICT-CVD equations performed better than PCEs, neither performed well. Multimorbidity and competing risks are likely to contribute to the poor performance and new CVD risk equations need to include these factors.

Author Information

Sue Wells, School of Population Health, University of Auckland, Auckland; Romana Pylypchuk, School of Population Health, University of Auckland, Auckland; Suneela Mehta, School of Population Health, University of Auckland, Auckland; Andrew Kerr, Cardiology Department, Middlemore Hospital, Auckland; Vanessa Selak, School of Population Health, University of Auckland, Auckland; Katrina Poppe, Department of Medicine, University of Auckland, Auckland; Corina Grey, School of Population Health, University of Auckland, Auckland; Rod Jackson, School of Population Health, University of Auckland, Auckland.

Acknowledgements

The authors thank the primary health care organisations, affiliated primary care physicians, nurses and patients for their contributions to this study.

Correspondence

A/P Sue Wells, School of Population Health, University of Auckland, Private Bag 92019 Auckland Mail Centre, Auckland 1142.

Correspondence Email

s.wells@auckland.ac.nz

Competing Interests

All authors report grants from Health Research Council of New Zealand during the conduct of the study; Drs Wells, Poppe, Jackson and Corina report grants from Heart Foundation of New Zealand during the conduct of the study; Drs Wells and Jackson report grants from New Zealand Ministry of Business, Innovation and Enterprise during the conduct of the study; Dr Wells reports grants from Stevenson Foundation during the conduct of the study.

1. Prince MJ, Wu F, Guo Y, et al. The burden of disease in older people and implications for health policy and practice. Lancet 2015; 385(9967):549–62.

2. Hermanson B, Omenn GS, Kronmal RA, et al. Beneficial six-year outcome of smoking cessation in older men and women with coronary artery disease. Results from the CASS registry. New England Journal of Medicine 1988; 319(21):1365–9.

3. Blood Pressure Lowering Treatment Trialists C, Sundstrom J, Arima H, et al. Blood pressure-lowering treatment based on cardiovascular risk: a meta-analysis of individual patient data. Lancet 2014; 384(9943):591–8.

4. Cholesterol Treatment Trialists C. Efficacy and safety of statin therapy in older people: a meta-analysis of individual participant data from 28 randomised controlled trials. Lancet 2019; 393(10170):407–15.

5. Gueyffier F, Bulpitt C, Boissel JP, et al. Antihypertensive drugs in very old people: a subgroup meta-analysis of randomised controlled trials. INDANA Group. Lancet 1999; 353(9155):793–6.

6. Ministry of Health. Cardiovascular disease risk assessment and management for primary care. Wellington: Ministry of Health, 2018.

7. Wells S, Riddell T, Kerr A, et al. Cohort Profile: The PREDICT Cardiovascular Disease Cohort in New Zealand Primary Care (PREDICT-CVD 19). International Journal of Epidemiology 2015; doi: 10.1093/ije/dyv312 doi: doi: 10.1093/ije/dyv312

8. Pylypchuk R, Wells S, Kerr A, et al. Cardiovascular disease risk prediction equations in 400,000 primary care patients in New Zealand: a derivation and validation study. Lancet 2018; 391(10133):1897–907.

9. Goff DC, Jr., Lloyd-Jones DM, Bennett G, et al. 2013 ACC/AHA guideline on the assessment of cardiovascular risk: a report of the American College of Cardiology/American Heart Association Task Force on Practice Guidelines. Journal of the American College of Cardiology 2014; 63(25 Pt B):2935–59.

10. Ministry of Health. National Health Index Wellington2015 [Available from: http://www.health.govt.nz/our-work/health-identity/national-health-index accessed February 5 2015.

11. Chan WC, Wright C, Tobias M, et al. Explaining trends in coronary heart disease hospitalisations in New Zealand: trend for admissions and incidence can be in opposite directions. Heart 2008; 94(12):1589–93.

12. Ministry of Health. Ethnicity data protocols for the Health and Disability Sector. Wellington: Ministry of Health, 2004.

13. Salmond C, Crampton P, Atkinson J. NZDep2006 Index of Deprivation User’s Manual. Wellington: ; 2007. Wellington: Department of Public Health, Wellington School of Medicine and Health Sciences, 2007.

14. Quan H, Li B, Couris CM, et al. Updating and validating the Charlson comorbidity index and score for risk adjustment in hospital discharge abstracts using data from 6 countries. American Journal of Epidemiology 2011; 173(6):676–82.

15. Kaplan E, Meier P. Nonparametric estimation from incomplete observations. J Am Stat Assoc 1958; 53:457–81.

16. Crowson C, Atkinson E, Therneau T. Assessing Calibration of Prognostic Risk Scores. Stat Methods Med Res 2016; 25(4):1692–706. doi: . doi:10.1177/0962

280213497434.

17. Harrell F, Califf R, Prior D, et al. Evaluating the yield of medical tests. Jounal of American Medical Association 1982; 247:2543–6.

18. Royston P, Sauerbrei W. A new measure of prognostic separation in survival data. Stat Med 2004; 23(5):723–48.

19. StataCorp. Stata Statistical Software. Release 15. College Station, TX: StatCorp LP, 2015.

20. Hippisley-Cox J, Coupland C, Brindle P. Development and validation of QRISK3 risk prediction algorithms to estimate future risk of cardiovascular disease: prospective cohort study. BMJ 2017; 357:j2099. doi: http://dx.doi.org/10.1136/bmj.j2099

21. Manuel DG, Tuna M, Bennett C, et al. Development and validation of a cardiovascular disease risk-prediction model using population health surveys: the Cardiovascular Disease Population Risk Tool (CVDPoRT). CMAJ Canadian Medical Association Journal 2018; 190(29):E871–E82.

22. Cooney MT, Selmer R, Lindman A, et al. Cardiovascular risk estimation in older persons: SCORE O.P. European Journal of Preventive Cardiology 2016; 23(10):1093–103.

23. Wolbers M, Koller MT, Witteman JC, et al. Prognostic models with competing risks: methods and application to coronary risk prediction. Epidemiology 2009; 20(4):555–61. doi: 10.1097/EDE.0b013e3181a39056 [published Online First: 2009/04/16]

24. Ministry of Health. How is My PHO performing? 2015/16 Quarter 4 (January to March) results summary. ed. Wellington: Ministry of Health, 2015.

25. Barnett K, Mercer SW, Norbury M, et al. Epidemiology of multimorbidity and implications for health care, research, and medical education: a cross-sectional study. Lancet 2012; 380(9836):37–43.

26. Forman DE, Maurer MS, Boyd C, et al. Multimorbidity in Older Adults With Cardiovascular Disease. Journal of the American College of Cardiology 2018; 71(19):2149–61.

27. Benetos A, Thomas F, Bean KE, et al. Role of modifiable risk factors in life expectancy in the elderly. Journal of Hypertension 2005; 23(10):1803–8.

28. Lowsky DJ, Olshansky SJ, Bhattacharya J, et al. Heterogeneity in healthy aging. J Gerontol A Biol Sci Med Sci 2014; 69(6):640–9. doi: 10.1093/gerona/glt162 [published Online First: 2013/11/20]

29. Calderon-Larranaga A, Poblador-Plou B, Gonzalez-Rubio F, et al. Multimorbidity, polypharmacy, referrals, and adverse drug events: are we doing things well? British Journal of General Practice 2012; 62(605):e821–6. doi: http://dx.doi.org/10.3399/bjgp12X659295

30. Wolff JL, Starfield B, Anderson G. Prevalence, expenditures, and complications of multiple chronic conditions in the elderly. Archives of Internal Medicine 2002; 162(20):2269–76.

31. American Geriatrics Society. Guiding principles for the care of older adults with multimorbidity: an approach for clinicians: American Geriatrics Society Expert Panel on the Care of Older Adults with Multimorbidity. J Am Geriatr Soc 2012; 60(10):E1–E25. doi: http://dx.doi.org/10.1111/j.1532-5415.2012.04188.x

32. Muth C, van den Akker M, Blom JW, et al. The Ariadne principles: how to handle multimorbidity in primary care consultations. BMC Med 2014; 12:223. doi: http://dx.doi.org/10.1186/s12916-014-0223-1

33. Stokes T. Multimorbidity and clinical guidelines: problem or opportunity? New Zealand Medical Journal 2018; 131(1472):7–9.

Contact diana@nzma.org.nz
for the PDF of this article

View Article PDF

Cardiovascular disease (CVD) is the leading cause of potentially preventable global health loss and demand for health and disability services for older people.1 There is evidence that reducing smoking,2 blood pressure (BP)3 and lipids4 is associated with reduced fatal and non-fatal CVD for adults at any age, and the benefits are largely determined by patients’ pre-treatment CVD risk. As older people are more likely than younger people to be at high CVD risk, they are also likely to benefit most from CVD risk-reducing medications.1 A recent individual person meta-analysis of 28 statin trials found that treatment produced a similar reduction in major vascular events per mmol/L reduction in low-density lipoprotein (LDL) cholesterol irrespective of age, although findings were more attenuated for those over 75 years without established CVD.4 In the case of BP-lowering, the benefits of treatment accrue even among the very old. In a meta-analysis involving only octogenarians from seven clinical trials, BP-lowering medications were associated with lower rates of stroke (34%), heart failure (39%) and major CVD events (22%) than those not receiving treatment.5 Despite these findings, most CVD risk management guidelines are vague about how to manage older people. In New Zealand, for example, CVD risk assessment and management guidelines recommend a formal quantitative CVD risk assessment for people aged 30–74 years,6 but once a person turns 75 years of age, risk assessment is ‘at the discretion’ of the clinician. General practitioners (GPs) are given general advice to use clinical judgement taking into account the results of a risk assessment, the likely benefits and risks of treatment and patient preferences.

We have identified an increasing number of New Zealanders aged 75 years and over receiving formal quantitative CVD risk assessments in routine primary care and have generated a cohort of these older people as part of our ongoing PREDICT-CVD risk prediction cohort study.7 Given the clinical relevance of accurate assessment of CVD risk in older people, we aimed to investigate how well the recently developed PREDICT-CVD equations8 (derived in people aged 30–74 years), performed in older people. As a comparison, we also assessed the performance of the relatively similar US Pooled Cohort Equations (PCEs),9 derived in people aged 40–79 years. We hypothesised that these CVD risk equations should perform reasonably well in the subgroup of older patients who GPs considered suitable for routine preventive CVD risk assessment.

Methods

Study population

The PREDICT study has been described in full elsewhere.7 In brief, patients are automatically recruited to this prospective open cohort study when primary care clinicians undertake standard quantitative CVD risk assessments using the PREDICT web-based decision support programme. Over one-third of GPs use PREDICT software, which is integrated into their electronic patient management systems. An encrypted version of each person’s unique national health identifier (National Health Index number, eNHI) is used to anonymously link patients’ risk profiles to national and regional health datasets, including all community pharmaceutical dispensing, all community laboratory testing, state-funded hospitalisations and all deaths.10 Over 98% of New Zealanders have an NHI number,10 allowing identification and linkage of multiple health contacts, augmentation of risk factor data (prior hospital admissions, pharmaceutical dispensing, laboratory test results) as well as health outcome ascertainment (fatal CVD events in-hospital and out-of-hospital and non-fatal hospital admissions for acute CVD events) during follow-up. Over 95% of CVD hospitalisations occur within our state-funded health services.11

For the purposes of this study, eligible patients were those who had a first (baseline) CVD assessment from 31 October 2004 to 30 December 2016 and were aged 70 years or older, unless they met any of the following exclusion criteria: prior history of ischaemic CVD, heart failure, renal disease or missing risk factors needed for CVD risk prediction models. In addition, to emulate the cohort used to develop the PREDICT CVD equations,8 patients were excluded if their self-identified ethnicity was recorded as Middle Eastern, Latin American, African or recorded as ‘other’ or ‘unknown’. The rationale for this exclusion was that these ethnic groups were too heterogeneous to combine into one category and too few in numbers to disaggregate into meaningful subgroups.

A history of prior CVD was classified according to an International Classification of Diseases, version 10 Australian Modification (ICD-10 AM) for hospitalisations or primary care clinical diagnosis at the time of CVD risk assessment for angina, myocardial infarction (MI), percutaneous coronary intervention (PCI), coronary artery bypass graft (CABG), ischaemic stroke, transient ischaemic attack (TIA), or peripheral vascular disease (PVD). The capture of a patient’s history of a hospitalised event used data available from 1 January 1988. (Appendix contains full list of ICD-10 codes) Patients who were dispensed anti-anginal medications on at least three occasions up to five years prior to their baseline visit were also excluded. Renal disease was determined either by an estimated glomerular filtration rate (eGFR) of less than 30ml/min per 1.73m2, an ICD-10 AM hospitalisation for renal dialysis, prior renal transplantation or a recording of diabetes with nephropathy at the time of CVD risk assessment. Heart failure diagnoses were based on ICD-10 AM hospitalisation code for heart failure or if participants had been dispensed a loop diuretic three or more times in the preceding five years.

Ethnicity classification was based on a nationally agreed prioritisation algorithm when individuals identified with more than one ethnicity12 in the following order; Māori, Pacific, Indian, Chinese/other Asian and European. Socio-economic status was assessed using the NZ Deprivation Index (NZDep), a measure assigned to patients according to the deprivation score of their area of residence.13 For these analyses, NZDep was divided into quintiles from 1 (least deprived) to 5 (most deprived).

Smoking status was defined as either current smoker (including recently quit in the last 12 months) or non-smoker. Diabetes status was classified according to ICD-10 hospitalisation with diabetes and/or dispensing of at least one diabetes medication in the last six months and/or recorded as such by their primary care clinician at the time of CVD risk assessment.

The Charlson comorbidity index is a weighted scoring system that assesses the degree of previously hospitalised comorbidity burden. It is based on 12 conditions that predict one-year survival and has been adapted for use with hospitalisation data using a well-validated ICD-10 coding algorithm.14 Comorbidities were identified from hospitalisations up to five years prior to the first CVD risk assessment.

The pharmaceutical collection (PHARMS) is a national database of community pharmaceutical dispensing. Reliable identification of dispensing episodes by eNHI has increased over the last decade from 64% in 2004, to 92% in 2006 and over 96% from 2009 onwards.7 PHARMS was used to identify patients who were dispensed one of the following medications on at least one occasion in the six months prior to the baseline CVD risk assessment: BP-lowering, lipid-lowering and antiplatelet/anticoagulant medications (henceforth termed antithrombotic medications). All these medications are government subsidised. (CVD medications are listed in the Appendix).

Outcomes during follow-up

CVD outcomes for the PREDICT-CVD equations8 were defined as ICD-10-AM coded hospitalisation or death from ischaemic heart disease, ischaemic or haemorrhagic cerebrovascular events (including TIA), PVD or heart failure. CVD outcomes for the American PCE Equations9 (termed hard atherosclerotic CVD) are a subset of the former and include fatal or nonfatal MI, fatal or nonfatal stoke, or CHD death (Appendix contains ICD-10-AM codes for both sets of outcomes). Time on study was the time from baseline CVD risk assessment to the first of the following: hospital admission or death related to CVD, death from other causes or end of follow-up.

Statistical analysis

The distributions of CVD risk factors, event rates and follow up were investigated for the total population and by five-year age groups; 70–74 years, 75–79 years, 80–84 years and 85 years and over. Calibration was tested separately for PREDICT-CVD8 and the PCE models9 by five-year age groups using equation-specific outcomes. The PREDICT-CVD risk models include age, ethnicity, deprivation, diabetes status, history of atrial fibrillation, smoking status, systolic blood pressure (SBP), the ratio of total cholesterol (TC) to high-density lipoprotein (HDL) cholesterol (TC/HDL) and prior dispensing of BP lowering, lipid lowering and antithrombotic medications. The PCEs include age, TC, HDL, smoking and diabetes status, SBP and treated SBP. Calibration was assessed graphically by categorising participants into deciles of predicted five-year CVD risk and plotting mean predicted five-year CVD risk against observed CVD events at five years of follow up, obtained by the Kaplan-Meier method.15 For PCEs we used recalibrated models where the baseline survival values were estimated by fitting Cox models with the prognostic index from the PCE model (offset term) in the PREDICT-CVD dataset.16 Discrimination was assessed using Harrell’s C statistic and Royston and Sauerbrei’s D statistic.17,18 The proportion of outcome variation explained by PREDICT-CVD and PCEs was assessed using Royston and Sauerbrei’s R2 statistic.18 All analyses were performed using Stata 15.0 software.19

Ethics approval

Approved by the Northern Region Ethics Committee Y in 2003 (AKY/03/12/314) with subsequent annual approval by the National Multi Region Ethics Committee since 2007 (MEC07/19/EXP).

Results

After applying exclusion criteria, 40,161 participants aged 70 years or over had a baseline CVD risk assessment between 31 October 2004 and 30 December 2016 (Figure 1, Table 1).

Figure 1: Study exclusions and incidence of CVD events during follow up.

*Excluded if ethnicity recorded as Middle Eastern, Latin American, African or recorded as ‘other’ or ‘unknown’ ‘MELAA’, ‘other’ or ‘unknown’ (in 70+ n=542).

Table 1: Description of the PREDICT cohort aged 70 years and over.

*IQR: interquartile range; SD: standard deviation; CVD: cardiovascular disease; NZ: New Zealand; TC/HDL: Total Cholesterol to HDL Cholesterol ratio.

During 185,150 person-years follow-up (mean follow-up time 4.6 years), 5,948 (15%) experienced an incident CVD event of which 1,065 (18%) were fatal. The Appendix describes the number and type of CVD event; mostly due to MI (1,690 events; 28.4%), ischaemic stroke (1,154 events; 19.4%) and heart failure (1,107 events; 18.5%). An additional 3,932 people (10%) died from non-CVD causes. The incidence of CVD and fatal non-CVD events increased markedly with increasing five-year age bands.

The majority of the cohort were women (57%), European (76%) and non-smokers (74%). Just over a third were resident in the two most deprived quintiles. In terms of comorbidity, 6% had a history of atrial fibrillation, 18% had diabetes and 18% had a Charlson comorbidity index of one or more. Overall, 51% of the cohort were on BP-lowering medications, 32% on lipid-lowering and 29% on antithrombotic medications. With increasing age there was an increase in the proportion of women, those of European ethnicity, non-smokers, those dispensed BP-lowering medications and people with atrial fibrillation, diabetes and a comorbidity score of one or more, as well as an increase in mean SBP. Only the mean TC/HDL and proportion dispensed lipid-lowering medications decreased with age.

Calibration graphs by decile of predicted risk versus observed five-year CVD event risk for the New Zealand PREDICT-CVD equations and recalibrated PCEs are shown for women (Figure 2) and men (Figure 3) by five-year age groups. In the over 80-year age groups, some deciles could not be plotted according to observed event rate due to insufficient numbers with follow-up at five years.

Figure 2: Calibration plots by decile of predicted risk and observed CVD event risk at five years according to PREDICT-CVD, PCE and recalibrated PCE for women by age group.

A diagonal line with intercept of 0 and slope of 1 represents perfect calibration. Plotted risk deciles below the diagonal represent an underestimate of predicted risk, above the diagonal, an overestimate.

Figure 3: Calibration graphs by decile of predicted risk and observed CVD event risk at five years according PREDICT-CVD, PCE and recalibrated PCE for men by age group.

A diagonal line with intercept of 0 and slope of 1 represents perfect calibration. Plotted risk deciles below the diagonal represent an underestimate of predicted risk, above the diagonal, an overestimate.

For women, PREDICT-CVD was well calibrated for those aged 70–74 years but underestimated CVD risk for women aged 75–84 years with the exception of those in the highest deciles of risk. Among those aged 85 years and over, underestimation of risk was also observed as well as suggesting poor discrimination by decile of risk (ie, risk deciles clustered over a similar observed five-year event rate). In contrast, the recalibrated PCEs overestimated CVD risk across all age groups. For those over 80 years, while predicted risk deciles were well separated, they were also stacked above the same observed event rate, indicating poor discrimination as well as poor calibration.

For men, PREDICT-CVD was reasonably well calibrated for those aged 70–79 years but underestimated CVD risk for men aged 80–84 years with the exception of those in the three highest deciles of risk, and underestimated CVD risk for all aged 85 years and over as well as showing poor discrimination. The PCEs overestimated CVD risk across all age groups from 70–84 years. However, in the 85+ age group, in the deciles with sufficient participants at five years of follow-up, overestimation with the recalibrated PCEs was less marked than at younger age groups.

Tables 2 and 3 summarise the discrimination metrics by age group and sex for PREDICT-CVD and PCEs. The discrimination and overall performance were generally poor for both equations in people over age 75 years.

Table 2: Performance metrics of PREDICT-CVD models, by sex and age group.

Table 3: Performance metrics of PCE models, by sex and age group.

Discussion

This study investigated the performance of contemporary CVD risk prediction equations in a cohort of 40,161 ambulatory people aged 70 years and over who had a heart health check, while visiting their GP. Over a third of the cohort were aged over 74 years. While both the calibration and discrimination performance of the equations varied, in general they performed increasingly poorly in people over 75 years of age, particularly the PCEs.

This is the first study comparing the performance of contemporary CVD risk equations in a cohort of older people whose GPs had decided to risk assess them in a routine practice setting. Previous studies have been based on either total general practice population samples,20 population-based health surveys21 or combined cohort studies populations.9,22 Our study reported comorbidity and non-CVD deaths across our total cohort and in five-year age bands. While some comorbidities have been reported in previous evaluations of some equations,20 none have reported how these factors, that are likely to influence the accuracy of risk assessment in the elderly, change with increasing age. Furthermore, previous evaluations of equation performance in older people have used wide age bands, which can mask significant differences by age, given the diminishing numbers of people in increasingly older age bands.20,22

There are other CVD risk prediction equations recommended for use in older patients. These include QRISK3 developed by UK researchers for people aged 25–84 years;20 Systematic COronary Risk Evaluation in older people (SCORE O.P.) developed by European researchers for people aged 65–80 years;22 and the Canadian CVD Population Risk Tool (CVDPoRT) for ages 20–105.21 The QRISK3 model includes over 20 predictor variables,20 many of which we were unable to incorporate given our more limited CVD profile data. Similarly, the Canadian CVDPoRT equation, derived from large population health surveys includes many lifestyle factors also not captured in our primary care-derived dataset.21 We considered assessing the performance of SCORE O.P. equations, derived in people aged 65–80 years, but it only predicts CVD mortality and many patients are also concerned about the impact of non-fatal major CVD events. Moreover, diabetes is not included as a predictor, yet a quarter of the PREDICT cohort over 75 years had diabetes.

It has previously been reported that CVD risk prediction equations developed and validated in younger age cohorts may not perform well when applied to populations aged 75 and older due to competing risks.23 Most equations do not incorporate the effect of other comorbid conditions or polypharmacy on CVD risk and competing risks of death from cancer or dementia. These competing risk events (non-CVD death events, which preclude an individual from experiencing a CVD event) become increasingly important with age. Indeed 10% of our study cohort died from non-CVD causes. For older age cohorts, where mortality rates are comparatively high, equations that treat non-CVD events as competing risks are likely to be needed to achieve more accurate risk prediction.23

The major strength of this study is also its major limitation. Participants were those older people, who, in the clinical judgement of primary care clinicians, were suitable for a routine preventive CVD risk assessment. While 90% of all New Zealanders aged 30–74 years have had a CVD risk assessment,24 many of the older people in our cohort have been risk assessed largely at the discretion of their primary care provider. Therefore they are not representative of all older people in the study region, because risk assessment would not be clinically appropriate for many of those not included (eg, those with dementia or requiring palliative care). We estimated that 50% of people aged 75–79 years, 30% aged 80–84 years and 25% of those aged 85 years and over, who did not have prior CVD, were included in our study. A further limitation is the use of the Charlson comorbidity index, modified to the extent that only 9 of the 12 comorbidities could be present in the study cohort (heart failure, stroke, renal disease being excluded). While the index has been validated in a New Zealand population, it suffers from including a very limited range of hospitalised-only long-term conditions and therefore underestimates the true multimorbidity burden in primary care. Indeed the prevalence of multimorbidity in the 65–84 year age group has been found to be as much as 65% in Scottish general practices.25 This is particularly relevant as most CVD risk prediction equations, and CVD risk management guidelines, tend to take a narrow disease-focused approach.26

This paper poses a series of clinical implications to current CVD risk prediction practice. Many older people are still engaged in the workforce and physically active. In the current Ministry of Health CVD guidance,6 healthy people over 75 years with few comorbidities and an estimated life expectancy of more than five years, CVD risk assessment using the PREDICT equations are recommended as well as discussing the same management options as for people under 75 years of age. However, although CVD risk factors have similar effects in those under and over 75 years,22,27 risk assessment and management is more complex for older age groups as health status, physical and cognitive functioning varies greatly.28 The risk of other long-term conditions increases with age25 and this in turn is associated with polypharmacy and complicated medication regimens.29,30 In this context, the risks and benefits of CVD risk-reducing interventions is accompanied by less certainty. Some treatment-related risks will increase, such as bleeding with aspirin, requiring clinicians to vary their advice.30 Furthermore, general health and functioning such as frailty, cognitive impairment, quality of life and personal preferences need to be taken into account. Older people may be more concerned about the risk of stroke than MI, as stroke may result in mental and physical disability and loss of independence, so single CVD outcomes (eg, stroke), as well as composite CVD outcomes, may be useful to guide discussions. The emerging guidance on how best to manage multimorbidity might offer a way forward here, with its focus on realistic treatment goals shared between clinician and patient and the need to recognise preference sensitive decisions (eg, medication that may benefit one condition but may make another worse).31–33

From this study of presumed healthy older people being risk assessed in general practice, we have found that the performance of the CVD equations derived mainly in people under age 75 years need to be improved to support clinical decision-making for people aged 75 years and over. We recommend that CVD equations used in people over 75 years incorporate factors such as multimorbidity and competing risks with additional risk-benefit tools taking into account physical and cognitive functioning and patient preferences.

Appendix

Number and type of incident CVD events in the PREDICT-1° patients, 70+ years old men and women

a If a patient had more than one type of CVD event, only the first was counted.

Medications available in New Zealand according to the CVD treatment categories of interest

*Alpha blockers, loop diuretics (bumetanide, frusemide), metolazone and spironolactone excluded as primary indication not usually to reduce blood pressure.

International Classification of Disease-10-Australian Modification (ICD-10-AM) codes for the PREDICT-1° CVD events outcome, from hospital discharge and mortality records

International Classification of Disease-10-Australian Modification (ICD-10-AM) codes for the PCEs hard atherosclerotic CVD outcome,* from hospital discharge and mortality records.
*Both fatal and nonfatal events with myocardial infarction and stroke codes were included but only fatal events with ‘Other coronary heart disease’ codes.

ICD 10 codes included in the definition of prior CVD

Summary

Abstract

Aim

To investigate how well the New Zealand PREDICT-CVD risk equations, derived in people aged 30–74 years and US Pooled Cohort Equations (PCEs) derived in people aged 40–79 years, perform for older people.

Method

The PREDICT cohort study automatically recruits participants when clinicians use PREDICT software to conduct a CVD risk assessment. We identified patients aged 70 years and over, without prior CVD, renal disease or heart failure who had been risk assessed between 2004 and 2016. Equation performance was assessed in five-year age bands using calibration graphs and standard discrimination metrics.

Results

40,161 patients (median 73 years; IQR 71–77) experienced 5,948 CVD events during 185,150 person-years follow-up. PREDICT-CVD equations were well calibrated in 70–74 year olds but underestimated events for women from 75 years and men from 80 years. Discrimination metrics were also poor for these age groups. Recalibrated PCEs overestimated CVD risk in both sexes and had poor discrimination from age 70 years for men and from age 75 years for women.

Conclusion

While PREDICT-CVD equations performed better than PCEs, neither performed well. Multimorbidity and competing risks are likely to contribute to the poor performance and new CVD risk equations need to include these factors.

Author Information

Sue Wells, School of Population Health, University of Auckland, Auckland; Romana Pylypchuk, School of Population Health, University of Auckland, Auckland; Suneela Mehta, School of Population Health, University of Auckland, Auckland; Andrew Kerr, Cardiology Department, Middlemore Hospital, Auckland; Vanessa Selak, School of Population Health, University of Auckland, Auckland; Katrina Poppe, Department of Medicine, University of Auckland, Auckland; Corina Grey, School of Population Health, University of Auckland, Auckland; Rod Jackson, School of Population Health, University of Auckland, Auckland.

Acknowledgements

The authors thank the primary health care organisations, affiliated primary care physicians, nurses and patients for their contributions to this study.

Correspondence

A/P Sue Wells, School of Population Health, University of Auckland, Private Bag 92019 Auckland Mail Centre, Auckland 1142.

Correspondence Email

s.wells@auckland.ac.nz

Competing Interests

All authors report grants from Health Research Council of New Zealand during the conduct of the study; Drs Wells, Poppe, Jackson and Corina report grants from Heart Foundation of New Zealand during the conduct of the study; Drs Wells and Jackson report grants from New Zealand Ministry of Business, Innovation and Enterprise during the conduct of the study; Dr Wells reports grants from Stevenson Foundation during the conduct of the study.

1. Prince MJ, Wu F, Guo Y, et al. The burden of disease in older people and implications for health policy and practice. Lancet 2015; 385(9967):549–62.

2. Hermanson B, Omenn GS, Kronmal RA, et al. Beneficial six-year outcome of smoking cessation in older men and women with coronary artery disease. Results from the CASS registry. New England Journal of Medicine 1988; 319(21):1365–9.

3. Blood Pressure Lowering Treatment Trialists C, Sundstrom J, Arima H, et al. Blood pressure-lowering treatment based on cardiovascular risk: a meta-analysis of individual patient data. Lancet 2014; 384(9943):591–8.

4. Cholesterol Treatment Trialists C. Efficacy and safety of statin therapy in older people: a meta-analysis of individual participant data from 28 randomised controlled trials. Lancet 2019; 393(10170):407–15.

5. Gueyffier F, Bulpitt C, Boissel JP, et al. Antihypertensive drugs in very old people: a subgroup meta-analysis of randomised controlled trials. INDANA Group. Lancet 1999; 353(9155):793–6.

6. Ministry of Health. Cardiovascular disease risk assessment and management for primary care. Wellington: Ministry of Health, 2018.

7. Wells S, Riddell T, Kerr A, et al. Cohort Profile: The PREDICT Cardiovascular Disease Cohort in New Zealand Primary Care (PREDICT-CVD 19). International Journal of Epidemiology 2015; doi: 10.1093/ije/dyv312 doi: doi: 10.1093/ije/dyv312

8. Pylypchuk R, Wells S, Kerr A, et al. Cardiovascular disease risk prediction equations in 400,000 primary care patients in New Zealand: a derivation and validation study. Lancet 2018; 391(10133):1897–907.

9. Goff DC, Jr., Lloyd-Jones DM, Bennett G, et al. 2013 ACC/AHA guideline on the assessment of cardiovascular risk: a report of the American College of Cardiology/American Heart Association Task Force on Practice Guidelines. Journal of the American College of Cardiology 2014; 63(25 Pt B):2935–59.

10. Ministry of Health. National Health Index Wellington2015 [Available from: http://www.health.govt.nz/our-work/health-identity/national-health-index accessed February 5 2015.

11. Chan WC, Wright C, Tobias M, et al. Explaining trends in coronary heart disease hospitalisations in New Zealand: trend for admissions and incidence can be in opposite directions. Heart 2008; 94(12):1589–93.

12. Ministry of Health. Ethnicity data protocols for the Health and Disability Sector. Wellington: Ministry of Health, 2004.

13. Salmond C, Crampton P, Atkinson J. NZDep2006 Index of Deprivation User’s Manual. Wellington: ; 2007. Wellington: Department of Public Health, Wellington School of Medicine and Health Sciences, 2007.

14. Quan H, Li B, Couris CM, et al. Updating and validating the Charlson comorbidity index and score for risk adjustment in hospital discharge abstracts using data from 6 countries. American Journal of Epidemiology 2011; 173(6):676–82.

15. Kaplan E, Meier P. Nonparametric estimation from incomplete observations. J Am Stat Assoc 1958; 53:457–81.

16. Crowson C, Atkinson E, Therneau T. Assessing Calibration of Prognostic Risk Scores. Stat Methods Med Res 2016; 25(4):1692–706. doi: . doi:10.1177/0962

280213497434.

17. Harrell F, Califf R, Prior D, et al. Evaluating the yield of medical tests. Jounal of American Medical Association 1982; 247:2543–6.

18. Royston P, Sauerbrei W. A new measure of prognostic separation in survival data. Stat Med 2004; 23(5):723–48.

19. StataCorp. Stata Statistical Software. Release 15. College Station, TX: StatCorp LP, 2015.

20. Hippisley-Cox J, Coupland C, Brindle P. Development and validation of QRISK3 risk prediction algorithms to estimate future risk of cardiovascular disease: prospective cohort study. BMJ 2017; 357:j2099. doi: http://dx.doi.org/10.1136/bmj.j2099

21. Manuel DG, Tuna M, Bennett C, et al. Development and validation of a cardiovascular disease risk-prediction model using population health surveys: the Cardiovascular Disease Population Risk Tool (CVDPoRT). CMAJ Canadian Medical Association Journal 2018; 190(29):E871–E82.

22. Cooney MT, Selmer R, Lindman A, et al. Cardiovascular risk estimation in older persons: SCORE O.P. European Journal of Preventive Cardiology 2016; 23(10):1093–103.

23. Wolbers M, Koller MT, Witteman JC, et al. Prognostic models with competing risks: methods and application to coronary risk prediction. Epidemiology 2009; 20(4):555–61. doi: 10.1097/EDE.0b013e3181a39056 [published Online First: 2009/04/16]

24. Ministry of Health. How is My PHO performing? 2015/16 Quarter 4 (January to March) results summary. ed. Wellington: Ministry of Health, 2015.

25. Barnett K, Mercer SW, Norbury M, et al. Epidemiology of multimorbidity and implications for health care, research, and medical education: a cross-sectional study. Lancet 2012; 380(9836):37–43.

26. Forman DE, Maurer MS, Boyd C, et al. Multimorbidity in Older Adults With Cardiovascular Disease. Journal of the American College of Cardiology 2018; 71(19):2149–61.

27. Benetos A, Thomas F, Bean KE, et al. Role of modifiable risk factors in life expectancy in the elderly. Journal of Hypertension 2005; 23(10):1803–8.

28. Lowsky DJ, Olshansky SJ, Bhattacharya J, et al. Heterogeneity in healthy aging. J Gerontol A Biol Sci Med Sci 2014; 69(6):640–9. doi: 10.1093/gerona/glt162 [published Online First: 2013/11/20]

29. Calderon-Larranaga A, Poblador-Plou B, Gonzalez-Rubio F, et al. Multimorbidity, polypharmacy, referrals, and adverse drug events: are we doing things well? British Journal of General Practice 2012; 62(605):e821–6. doi: http://dx.doi.org/10.3399/bjgp12X659295

30. Wolff JL, Starfield B, Anderson G. Prevalence, expenditures, and complications of multiple chronic conditions in the elderly. Archives of Internal Medicine 2002; 162(20):2269–76.

31. American Geriatrics Society. Guiding principles for the care of older adults with multimorbidity: an approach for clinicians: American Geriatrics Society Expert Panel on the Care of Older Adults with Multimorbidity. J Am Geriatr Soc 2012; 60(10):E1–E25. doi: http://dx.doi.org/10.1111/j.1532-5415.2012.04188.x

32. Muth C, van den Akker M, Blom JW, et al. The Ariadne principles: how to handle multimorbidity in primary care consultations. BMC Med 2014; 12:223. doi: http://dx.doi.org/10.1186/s12916-014-0223-1

33. Stokes T. Multimorbidity and clinical guidelines: problem or opportunity? New Zealand Medical Journal 2018; 131(1472):7–9.

Contact diana@nzma.org.nz
for the PDF of this article

Subscriber Content

The full contents of this pages only available to subscribers.

LOGINSUBSCRIBE