Introduction

This document summarises the rationale and details relating to the analysis of MINAP data for input into the cost-effectiveness analysis. Analyses of the MINAP dataset were carried out by John Birkhead. The extrapolation analyses were carried out by the NCC–CC.

The cost-effectiveness model is reported in Appendix C.

Approach

The aim was to obtain contemporary UK event rate data for one of the treatment arms of the cost-effectiveness analysis. Other treatment arms would then be modelled by applying appropriate relative risks from clinical trials to the UK baseline event rates.

Management of UA and non-ST elevation myocardial infarction (UA/NSTEMI) is known to historically vary between countries and revascularisation rates have been lower in the UK than most other Western European countries¹⁷⁸. Therefore baseline event rates from multinational RCTs undertaken to evaluate clinical effectiveness may not provide reliable estimates for UK practice. In addition randomised controlled trials are selective and therefore very high risk patients are often excluded. For these reasons UK specific baseline event rates for the cost-effectiveness model were sought.

The modelling undertaken for the NICE technology appraisal of glycoprotein IIb/IIIa inhibitors (GPIs) ²¹⁰ used registry data from PRAIS UK (1998-1999) with six-month follow-up, supplemented by data from a PCI audit in Leeds 2000 for short-term modelling. Data from the Nottingham Heart attack registry with up to five years follow-up was used for longer term modelling. The GDG felt that obtaining contemporary baseline events from the UK Myocardial Ischaemia National Audit Project (MINAP)⁴⁰ dataset would capture changes in outcome over time due to changes in practice (including increased use of an invasive management strategy) and widespread use of clopidogrel. This includes improved outcomes for patients due to changes in management over time and potentially also an increase in bleeding. It also allowed detailed analysis based on patient risk scores to be undertaken.

The MINAP dataset

The Myocardial Ischaemia National Audit Project (MINAP)⁴⁰ collects information about the hospital management of acute coronary syndrome (ACS). Initially the project focussed on the hospital management of acute ST-elevation myocardial infarction (STEMI) but the dataset has been expanded to cover other ACS (including UA/NSTEMI). All hospitals in England and Wales that admit patients with ACS contribute data. Linkage with the Office of National Statistics allows post-discharge mortality tracking. Examination of readmissions allows estimation of new MI events post-discharge.

The cohort used

A MINAP database for 2005-7 (download 19 Feb 2008) was used for these analyses, and was limited to English hospitals.

Risk stratification

Each patient in the selected MINAP cohort was assigned a risk score based on the GRACE scoring system (the risk scoring methods are described below). This allowed patients to be grouped into risk groups to investigate how cost effectiveness varies with baseline risk.

The GRACE score uses 8 variables: age, systolic blood pressure, heart rate, cardiac arrest, bio-marker elevation, ST deviation, serum creatinine and Killip class. MINAP did not record serum creatinine throughout the period 2005-7 and Killip score is not included in the dataset. A mini-score, without these elements was created using the GRACE scoring system. The six-month risk scoring system was used inline with the other risk work undertaken for the guideline²¹. Patients were split into six risk groups based on their risk score: 1a (~12.5% of patients), 1b (~12.5%), 2a (~12.5%), 2b (~12.5%), 3 (~25%) and 4 (~25%). Group 1a is the lowest risk group and group 4 is highest risk score. See the Risk Chapter for more details about the GRACE scoring system used, the creation of the risk groups and interpretation in the wider guideline context.

36,299 patients receiving the drugs specified above also had sufficient data available to calculate a risk score 2005-2007 (20,021 for 2005-2006 only). For PCI centres this included 7,694 patients.

Acute management stratification

Outcomes were analysed by acute management strategy; that is whether patients underwent PCI, CABG, angiography only or no angiography or revascularisation during their acute UA/NSTEMI episode. This was because some of the clinical trial data being used in the cost–effectiveness model were in a specific subset of the population e.g. those undergoing PCI. As risks of events may vary by acute management strategy it was therefore appropriate to assess outcomes by management strategy. As the interventions being assessed by the model are all used during the acute phase, acute management strategy was determined as the appropriate stratification.

The MINAP record for angiography and revascularisation covers the acute episode including what happens in the admitting hospital and, where the admitting hospital does not have interventional facilities, the hospital they refer to for intervention. Patients were split into the following acute management strategy groups: ‘PCI’, ‘CABG’, ‘Angio only’, ‘No angio’, and ‘Other’. The ‘Other’ group is not utilised in the cost-effectiveness analysis as management strategy is unknown.

The MINAP data fields for coronary angiography and coronary intervention are shown in Figure 1. Patients were first assigned as either ‘yes’ or ‘no’ to angiography. Yes = categories 1-5. No = category 8. Those where coronary angiography is unknown were excluded from the analysis. The ‘no’ category formed the ‘No angiography’ group.

Figure 1

MINAP data fields for coronary angiography and coronary intervention.

Patients who received coronary angiography were then categorised using the coronary intervention field. Patients were assigned to the ‘PCI’ group if they were recorded as ‘1) PCI’, and to the ‘CABG’ group if they were recorded as ‘2) CABG’. Patients who were recorded as ‘8) not performed or arranged’ were assigned to the ‘angio only’ group. Patients who were recorded as ‘9) Unknown’ were assigned to a group designated ‘Other’ (note that the data from this group is not used in the cost effectiveness model). Those recorded as ‘4) PCI planned after discharge’ and ‘5) CABG planned after discharge’ were a slightly complex group to assign. They were however also small in number – PCI planned after discharge = 2%, CABG planned after discharge – 0.5%. This was discussed with the health economic subgroup of the GDG and it was decided that for the purposes of analysing data for the cost-effectiveness analysis patients should be assigned based on what actually happened in the acute admission and so these patients were assigned to the ‘angio only’ group.

Analyses of the MINAP cohort

As described above, all analyses for the purposes of the cost-effectiveness model were restricted to UA/NSTEMI patients receiving aspirin, clopidogrel and heparin (UFH or LMWH) and not receiving an upstream GPI. All analyses were reported stratified by risk group (1a, 1b, 2a, 2b, 3, 4) and acute management strategy group (no angio, angio only, PCI, CABG, other) as far as possible (in some places this was not judged feasible due to low event numbers). This meant that only patients with the information required to assign to these groups were included in the analysis.

The population analysed included all non-interventional and interventional hospitals in England. The advantage of using the entire population is that this more accurately reflects national rates for mortality, but with a relative disadvantage that rates for intervention are understated. This arises because hospitals without interventional facilities may not know if or what intervention was performed after transfer, and may leave this information blank. Where appropriate, data from interventional hospitals only was used, or both were analysed.

Where one-year outcomes were required, the analysis was restricted to 2005/06 patients to ensure availability of one-year follow-up from the cohort.

Using the MINAP cohort described above the following events were analysed for the whole cohort and for each risk group, all split by acute management strategy:

• Mortality up to 1 year (mortality analyses section)
• Readmission up to 1 year (new MI events analyses section)
• In-hospital re-infarction (new MI events analyses section)
• In-hospital bleeding (bleeding analyses section
• Non-fatal MI in those alive at 1 year (inhospital re-infarction or 1 year readmission) (new MI events analyses section)

In addition data was analysed relating to the following:

• In-hospital management strategy (no angiography, angiography only, PCI or CABG) (acute management split analyses section
• Length of stay (overall and with an in-hospital re-infarction or bleed) (length of stay with complications analyses section)
• Demographics: age/sex breakdown by risk group (demographics section)

Details of these analyses follow. The results of the analyses from MINAP were graphed. Apparent anomalies in the data were reviewed to see if they might be accounted for by very low event numbers. Where this appeared to be the case this was discussed with the GDG. If judged likely to be attributable to low event numbers, risk groups were pooled for use in the cost-effectiveness analysis – details are provided below. Note that this mostly only occurred in the CABG group which is a small proportion of the total population.

Where results reported at different time points it is the one-year figures that are generally used in the cost-effectiveness analysis. The cost-effectiveness model report in Appendix C describes which data is used in the analysis in detail.

Mortality analyses

The census date for these analyses was Feb 19^th 2008, using data available to ONS up to 31 Dec 2007. In order to have a complete 365 day follow-up interval, mortality analyses are based on the 2005-6 cohort. 17,280 patients were included in this analysis. The number of deaths at 30 days, 6 months and 1 year were reported.

Results of analyses are presented in Figure 2 and Figure 3. Mortality increased by risk group, ranging from 1.4% at 1 year in risk group 1a to 44.4% in group 4. Anomalies were observed in the lower risk groups for CABG and PCI at one year. Event numbers in these groups were also observed to be very low: in the CABG group there were less than five events in groups 1a, 1b and 2a; in the PCI group there were less than five events in groups 1a and 1b. For these reasons in the cost-effectiveness model group 1a and 1b were pooled for CABG and for PCI, groups 2a and 2b were also pooled for CABG. See Figure 3 for pooled figures.

Figure 2

MINAP mortality analysis: trend over time by risk group.

Figure 3

MINAP mortality analysis: trend by risk group.

New MI events analyses

In-hospital re-infarction

In-hospital re-infarction is recorded by MINAP, and requires new cardiographic changes and new, or further elevation of cardiac markers in the context of new symptoms suggestive of cardiac ischaemia. Clinical trial definition of new MI generally includes all new MIs including those occurring in-hospital. Analyses in the literature have reported that experiencing a re-infarction is independently associated with increased hospital costs^38,274.

Analyses were based on first admissions. The quantity of missing data for re-infarction was noted.

26,291 patients were included in this analysis. The number of re-infarctions in the acute episode were reported.

Results of analyses are presented in Figure 4. In-hospital re-infarction rates were fairly low but generally showed a trend for increasing with risk group in the overall population, ranging from 1.1% to 2.7%. Within acute management strategy groups the trend observed was more erratic. Event numbers in the CABG group were also very low and groups 1a and 1b, and 2a and 2b were pooled for use in the cost-effectiveness model – see Figure 4 for pooled figures.

Figure 4

MINAP re-infarction analysis.

Readmission to hospital up to 1 year

Patients that experience a new ACS event following their acute admission who are readmitted to hospital will have a new MINAP record. This analysis is based on the presence of duplicate records having the same date of birth, patient case record number and hospital. From other MINAP analyses It is known that 85% readmissions after NSTEMI are for further infarction²⁷⁵. Readmission was analysed for admission during 2005/6 in order to have complete data for 1 year readmissions. 19,368 patients were included in this analysis. The number of readmissions at 30 days, 6 months and 1 year were reported.

Results of analyses are presented in Figure 5 and Figure 6. Event numbers in the CABG group were very low and in one risk group no events occurred at all. A pooled event rate across all risk groups was therefore used in the cost-effectiveness analysis for CABG – this was 2.3%.

Figure 5

MINAP readmission analysis: trend over time by risk group.

Figure 6

MINAP readmission analysis: trend by risk group.

Non-fatal MI at 1 year

This analysis is based on the patients who were alive at one year and had had either an in-hospital re-infarction or a new MINAP record (a readmission to hospital). Results were analysed for admissions during 2005/6 inline with the mortality and readmission analyses. 15,888 patients were included in this analysis.

It is noted that using a new MINAP record and not specifically one for MI will slightly overestimate the number of people in the new MI group as it will include UA as well. 85% of readmission following NSTEMI are reported at being for MI²⁷⁵.

Results of analyses are presented in Figure 7. Event numbers in the CABG group were very low and in one risk group no events occurred at all. Events were therefore pooled in group 1a and 1b, and 2a and 2b for use in the cost-effectiveness analysis – see Figure 7 for pooled figures.

Figure 7

MINAP non-fatal MI at 1-year analysis.

Bleeding analyses

Bleeding in relation to intervention can only safely be examined for those hospitals where interventional work is performed as this information is unlikely to be transmitted back to the referring hospital and then be recorded in MINAP. This limits the size of the cohort to those hospitals where intervention takes place. Note that surgery is not performed in all interventional hospitals and this may result in lower reported bleeding rates for CABG.

For the purposes of this analysis major bleeding was defined as the MINAP categories of: intracranial bleed; retroperitoneal bleed; blood loss > 5 G; and blood loss 3-5 G. Minor bleeding was defined as the MINAP category blood loss < 3 G. Patients with ‘unknown’ bleeding complications were excluded from the analysis.

Results could not be cross stratified by risk group and management group as event numbers were very low. Results were therefore presented stratified by each separately. 7123 patients were included in the analysis stratified by risk. 7233 were included in the analysis stratified by acute management strategy. (Note that numbers vary as only patients with sufficient information to allow the necessary stratification can be included in each analysis). Event numbers were also judged too low to split risk groups 1 and 2 into 1a and 1b, 2a and 2b.

Results of analyses are presented in Figure 8 and Figure 9. The number of in-hospital bleeding events was reported. Major bleeding increased by risk group, ranging from 0.2% in risk group 1 (1a and 1b combined) to 2.1% in group 4. Minor bleeding was fairly constant across groups 1-3 at around 1%, although increased in group 4 to 1.7%.

Figure 8

MINAP bleeding analysis: by acute management strategy.

Figure 9

MINAP bleeding analysis: by risk group.

The GDG noted that bleed rates appeared lower than expected based on rates seen in randomised controlled trials. As trials for agents that potentially increase the risk of bleeding may well also exclude patients with high bleed risk, it might be thought that registries would have higher rates of bleed than that observed in clinical trials. It is noted that bleeding forms part of the MINAP validation process.

Management and risk could not be cross tabulated for bleed events as event numbers are very low but both a risk trend and variation by acute management strategy was observed (see Figure 8 and Figure 9). To account for this in the cost effectiveness analysis, a relative risk of a bleed event and confidence interval for each management strategy compared to ‘total’ was calculated. This could then be applied to the risk group rates to calculate a management strategy specific rate for each risk group. In addition, as risk group 1 and 2 could also not be split further into 1a and 1b, and 2a and 2b as event number were very low in the model the rates for 1 will be applied to both 1a and 1b, and the rate for 2 applied to 2a and 2b. The resulting event rates are included in the cost-effectiveness analysis report – see Appendix C.

Length of stay with complications analyses

Complications such as re-infarction and bleeding have been reported as independently associated with increased hospitalisation costs in patients with UA/NSTEMI^{38,160,274,276}. On this basis, length of stay was analysed for patients experiencing these complications.

Length of stay overall and with an in-hospital re-infarction or bleed was analysed for 2007 patients only as analyses suggested that length of stay was falling over time. Length of stay with and without bleeding was analysed in interventional centres only for the reasons described above (Bleeding analyses section).

Results of analyses are presented in Figure 10 and Figure 11. Length of stay was greater in patients that experienced a re-infarction or a bleed complication compared to those that did not.

Figure 10

MINAP analysis of length of stay with bleeding complications. * Classified as major bleed in analysis

Figure 11

MINAP analysis of length of stay with bleeding complications.

Acute management split analyses

The relative percentages of patients undergoing an acute management strategy of no angio, angio only, PCI and CABG is most representative from PCI hospitals only. Intervention is under-represented when the relative percentage is based on all hospitals due to missing data. This arises because hospitals without PCI facilities may not know if or what intervention was performed after transfer, and are likely to leave this information blank.

The acute management split was analysed in both cohorts to verify this. The analyses include 8,299 patients for the PCI centres only and 38,808 patients for all centres. Based on interventional hospital data, 33% received no angiography or intervention, 28% received angiography only, 29% received PCI and 3% received CABG. In 7% the acute management strategy was unknown due to missing data. In comparison in all hospitals this figure rose to 20%.

Acute management strategy was also analysed by risk group. Results are shown in Table 1. Note that patients with missing data have been excluded from this table.

Table 1

MINAP analysis of acute management strategy by risk group (interventional hospital only).

The GDG noted that the CABG rate appeared lower than expected based on BCIS audit data that suggested a 3:1 ratio of PCI to CABG in the UK. It is noted that this may be due a bias in the reporting whereby patients who are transferred for surgery are recorded as ‘unknown’. Alternatively it may due to the fact that CABG patients are often discharged home and scheduled for CABG at a later date.

Demographics

Demographics were reported for each risk group in terms of age breakdown in 10-year bands by gender. See Figure 12 and Table 2. These were used in the extrapolation analysis detailed below

Figure 12

MINAP analysis age breakdown.

Table 2

MINAP analysis age breakdown.

Estimation of life-years for the cost effectiveness model

In order to fully capture lifetime quality-adjusted life-years (QALYs) in the cost–effectiveness model an estimate of life expectancy beyond one year was required. The aim was to extrapolate from the MINAP data to attempt to reflect contemporary mortality rates.

Linear extrapolation estimation

It has been observed that following a UA/NSTEMI event mortality is high but that this rapidly declines over time. After possibly as little as one month and certainly by six months, the mortality rate is at a fairly low level^277,278. In addition long-term studies plotting mortality over time suggest that after 3 months the survival curve is approximately linear^279,280. On this basis it was planned to estimate life expectancy for patients alive at one year by linearly extrapolating the mortality rate between six months and one year from the MINAP cohort. A linear extrapolation implies an increasing mortality rate over time. Separate extrapolations were undertaken for each risk group as mean age varied considerably across risk groups.

The estimated life expectancy in the risk groups 1a and 1b was higher than that predicted using general population life expectancy estimates. This suggested that the linear extrapolation may not be plausible. This may be explained by the very different age profiles across the risk groups – risk group 1a has a mean age of 50 years, while risk group 4 has a mean age of 84 years. Looking at a survival curve for the general population it could be seen that while in older people a linear extrapolation may lead to a reasonable estimation of life expectancy, in younger people, a linear extrapolation may overestimate life expectancy. An alternative approach was therefore sought.

Standardised mortality ratio based estimation

As an alternative to the linear extrapolation, standardised mortality ratios (SMRs) for UA/NSTEMI patients were calculated based on the observed mortality in the MINAP UA/NSTEMI cohort between 6 months and 1 year, and mortality rates for the general population. Separate SMRs were calculated for each risk group.

Mortality rates for each risk group were calculated using the six-month to one-year rates from the MINAP cohort. Comparable mortality rates for the general population were estimated based on the demographic of the risk group in terms of age (in ten-year age bands) and gender, and mortality rates from 2005-2007 life tables for England and Wales²⁸¹. An SMR was then calculated using this information. Formulae for these calculations are shown in Table 3.

Table 3

Formulae for estimation of SMRs.

Life expectancy for each risk group was then calculated using life tables, based on the gender split, mean age and the calculated SMR for the risk group. It was assumed that the SMR past six months is constant over time.

For the cost-effectiveness model we wished to obtain different estimates of life expectancy for people who are: 1) alive at one year and have had a new MI in the past year; and 2) alive at one year but have not had a new MI in the past year. This was in order to reflect the potential prognostic benefit of avoiding MI.

Additional data was obtained from the MINAP cohort in order to do this analysis. Patients who were alive at six months were split into two groups – those that had had a new MI event since their initial UA/NSTEMI event and those that had not. Mortality was then analysed at the one-year time point (that is, 6 months later). Results of this analysis are shown in Table 4. As there were only two events in risk group 1a and none in 1b these events were pooled together and a single SMR calculated.

Table 4

Mortality at one year in those alive at six months.

A new MI event was defined as an in-hospital re-infarction or a new MINAP record (readmission). It is noted that using a new MINAP record and not specifically one for MI will slightly overestimate the number of people in the new MI group as it will include UA as well. However, as 85% of readmission following NSTEMI is reported at being for MI this is considered a reasonable approximation²⁷⁵. The effect of this approximation is likely to be that the mortality rate in each group may be slightly reduced as patients with lower mortality are added to the MI group and patients while concurrently patients with a higher mortality are removed from the no MI group.

SMRs and estimates of life expectancy for those alive at one year are presented in Table 5.

Table 5

SMRs and estimates of life expectancy beyond one year by risk group.

Mortality was higher in the non-fatal MI group than the no event group in each risk group. This translated to a higher predicted life expectancy for those who did not have a new MI compared to those that did. Life expectancy in both UA/NSTEMI groups was lower than that estimated for a comparable group from the general population. These results were plausible and these methods were used to provide estimates of life expectancy for those alive at one year in the cost–effectiveness analysis