Variation in outcome of hospitalised patients with out-of-hospital cardiac arrest from acute coronary syndrome: a cohort study

Ambulance-level variability

In the UK, there is evidence, as shown in Figure 2, of wide variability in ROSC and survival-to-discharge rates across ambulance services. 3 In 2011, the survival-to-discharge rate following OHCA by ambulance service ranged from 2.5% to 12%. 3 Although the number of cases was small, the variation was not reduced through standardisation using the Utstein patient subgroup (witnessed arrest in a shockable rhythm with bystander CPR), and neither was the variation in outcome associated with ambulance response times. 11,12 Such variation in outcome following OHCA across emergency medical service (EMS) systems has also been observed in other countries. 26–28

FIGURE 2.

Variability in OHCA survival rates across English ambulance services. The navy line corresponds to the right vertical axis to describe the number of cardiac arrest calls per 1000 category A emergency calls. The black and green bars correspond to the left vertical axis to describe ROSC (black bar) and survival to hospital discharge (green bar) for services where data were available. Reproduced from Variability in cardiac arrest survival: the NHS Ambulance Service Quality Indicators, GD Perkins and MW Cooke, Emergency Medicine Journal, vol. 29, pp. 3–5, 2011, with permission from BMJ Publishing Group Ltd. 3

Reproduced from Variability in cardiac arrest survival: the NHS Ambulance Service Quality Indicators, GD Perkins and MW Cooke, Emergency Medicine Journal, vol. 29, pp. 3–5, 2011, with permission from BMJ Publishing Group Ltd. 3

Hospital-level variability

In the UK, ambulance service data show variability in outcome in survival to discharge among ambulance services with similar rates of ROSC (e.g. compare Great Western with Yorkshire and East of England Ambulance Services in Figure 2). 3 In the UK, conventional management for OHCA is that the patient will be transferred to the nearest appropriate emergency department (ED) according to locally agreed protocols.

At present, there are no UK data reporting OHCA survival variation between hospitals. International data from Sweden, Australia and North America show survival rates by hospital in OHCA patients admitted alive to hospital, and these range from 14% to 59%. 29–32 However, in the UK, there is evidence of variability in practice. A recent survey of 208 UK intensive care units that treat cardiac arrest patients found that only 28 units could provide all of the interventions [24/7 primary percutaneous coronary intervention (pPCI), ventilator care bundle, targeted temperature management and access to neurophysiology tests] recognised as essential for the effective intensive care management of cardiac arrest patients. 33 This availability is important as the availability and delivery of key interventions is associated with improved hospital outcome. 30,34

One strategy to reduce variability in survival and clinical practice may be the establishment of regional cardiac arrest centres. According to this strategy, the ambulance will, provided that certain criteria are met, bypass the local ED and transfer the patient directly to a regional centre. The rational is that the disadvantage of a longer ambulance transport time is offset by expert care at the regional setting through treatment by clinicians with greater exposure to the condition, improved access to complementary clinical specialities and improved access to imaging and specialist interventions.

Regionalised systems of care are already in place for conditions, such as stroke, STEMI and trauma. However, recent systematic reviews in the clinical areas of stroke and trauma do not show improved outcome when patients are taken directly to a specialist centre, rather to a non-specialist centre. 35,36 In contrast, there is evidence that treatment of STEMI patients with pPCI in a high-volume hospital is clinically effective and cost-effective. 37–39

In the context of cardiac arrest, the American Heart Association released a policy statement in 2010 describing a need to establish regionalised cardiac arrest care in the USA to improve patient outcome following OHCA. 40 In 2015, the International Liaison Committee on Resuscitation (ILCOR) review of evidence led to a treatment recommendation that supported the establishment of regionalised cardiac arrest care systems. 41 However, in making this recommendation, ILCOR acknowledged that the supporting evidence was of a low quality, that much of the evidence was retrospective and that there was inconsistency between studies as to which hospital factors were associated with improved outcome.

Importantly, none of the studies conducted to date has been undertaken in the UK setting. A recently published trial did demonstrate that it was feasible to undertake a randomised controlled trial comparing transfer to a specialist centre and consideration of percutaneous coronary intervention centre (PCI) with standard care in resuscitated OHCA patients without ST elevation on the electrocardiogram (ECG), and an effectiveness trial is now being planned. 42 As such, it is not currently possible to estimate the effects of such a system of care in the UK, where environmental factors (e.g. disease prevalence, response times, distances travelled), EMS systems (e.g. initial response time, ambulance staff skill mix, transfer time to hospital) and hospital configurations vary.

The Myocardial Ischaemia National Audit Project data set

As of 2014, the MINAP data set contained > 1.25 million records, with an additional 90,000 records being uploaded each year. For each patient record, a series of data points are collected. These data points cover the patient journey from the onset of symptoms to hospital discharge. The current data set includes approximately 130 fields. This includes data on patient demographics, past medical history, pre-hospital interventions, in-hospital laboratory results, in-hospital drug therapy, in-hospital interventions, discharge drugs and interventions, and the patient’s status at discharge. A full list of the MINAP fields is included in Appendix 1.

Four fields in the data set relate directly to cardiac arrest, namely the date/time of cardiac arrest (field 3.13), cardiac arrest location (field 3.14), arrest presenting rhythm (field 3.15) and outcome of arrest (field 3.16). One other field (field 3.10 – delay before treatment) has cardiac arrest as a listed response.

The MINAP data set can be linked to Office for National Statistics (ONS) data to provide additional information on long-term mortality and deprivation.

Collection of the Myocardial Ischaemia National Audit data set

The MINAP data set is collected at the hospital level. In 2014, it was reported that all English, Welsh and Northern Irish hospitals that admitted patients with myocardial ischaemia collected and uploaded data, with the exception of Scarborough Hospital. 48 MINAP is part of the national clinical audit programme, such that hospital participation is mandated under section 26.1.2 of the service conditions of the NHS standard contract. 50

The precise process for collecting data, such as methods for identifying eligible cases and the personnel involved in collecting and uploading data, varies between hospitals. This creates the potential for ascertainment bias and is a recognised limitation of the MINAP data set. 48 Data are uploaded using a secure online system. Each patient record must contain, as a minimum, the date and time of hospital admission, the hospital code, the patient’s hospital number and the admission diagnosis. MINAP has developed a handbook that includes definitions of data points to standardise data collection, which is available on the MINAP website. 49

For some data points, there is a real-time data validation check. For example, the system will query the serum cholesterol entry (field 2.15) if it is outside the range 2.5–25 mmol/l. In addition, MINAP performs an annual data validation assessment in which hospitals re-enter data for 20 randomly selected patients with a diagnosis of NSTEMI. The agreement between the original and re-entered data is recorded and reported to the hospital.

Demographic variables

This group included age, sex, ethnicity and the deprivation score. Ethnicity was recategorised, as detailed in Table 3.

The Index of Multiple Deprivation (IMD) is a score of deprivation supplied by the ONS based on postcode data. 55 Geographical areas with an approximate population of 1500 are scored based on seven domains (income, employment, education, health, crime, barriers to housing and services, and living environment). In our analysis, we used the absolute score (rather than rank), so a higher score indicates increased deprivation.

Medical history variables

This group included smoking status, diabetes mellitus status, hypercholesterolaemia, heart failure, cerebrovascular disease, previous AMI, asthma or chronic obstructive pulmonary disease (COPD), chronic renal failure, peripheral vascular disease, previous angina pectoris, previous PCI, previous coronary artery bypass graft (CABG) and hypertension. Diabetes status and smoking status were recategorised as shown in Table 3. Most definitions are based on documented history of the disease. For all variables, a response of unknown was categorised as missing.

Presenting characteristics of out-of-hospital cardiac arrest variables

This group included time point of cardiac arrest (before or after ambulance arrival), cardiac arrest rhythm, serum glucose (mmol/l), creatinine (µmol/l), left ventricular ejection fraction (LVEF), haemoglobin (g/dl), serum cholesterol (mmol/l), admission diagnosis, systolic blood pressure at admission, ECG that determined treatment, time of day of admission (day/night), Killip class, mini-GRACE (Global Registry of Acute Coronary Events) score and year of admission. The variable ECG that determined treatment was recategorised, as shown in Table 3.

For admission diagnosis, we combined two fields (2.01, initial diagnosis, and 2.36, site of infarct) to create a single field to describe both the initial diagnosis and, where appropriate, the site of the infarct. To create the new field, we recategorised ACS, chest pain cause and other initial diagnosis in field 2.01 as a single category of other diagnosis. For participants who were recorded in field 2.01 as having a definite myocardial infarction (MI), we broke down these participants by infarct site from 2.36. The revised field had three categories: definite MI – anterior infarction; definite MI – other infarction site; and other initial diagnosis.

Time of hospital arrival (field 3.06) was used to classify whether the patient was admitted during the day (admission time 08.00–19.59 hours) or at night (20:00–07:59 hours). There is little consistency as to the cut-offs to be used when categorising night and day in studies of OHCA, MI and temporal variability. Some studies dichotomise as night and day, albeit with variability in time cut-off points, whereas some studies add an additional category for evening admissions, and other studies include an additional category for the weekend. 56–64 On this basis, we took the pragmatic decision to categorise as discussed above, which is consistent with a previous OHCA study and similar to the method used in a previous MI study. 56,62 We were unable to analyse the impact of a weekend effect on survival in this study as, despite recent interest in this issue in the UK, NICOR was unable to release these data on the basis that, in combination with other variables, it might enable the identification of individual patients. 65–67

Laboratory values (glucose, cholesterol, creatinine, haemoglobin) are the first recorded value following hospital admission, and these are recorded within the first 24 hours of admission. Systolic blood pressure and heart rate are the first values recorded when the patient is in a stable cardiac rhythm (e.g. sinus rhythm).

Before June 2013, haemoglobin levels in the MINAP data set were reported as g/dl. In June 2013, the unit of measurement was changed to g/l. An analysis of data suggested that different hospitals were using both sets of units during 2013. To ensure consistency, we divided all values from 2014 and 2015 by 10 so that we could report them as g/dl. For 2013, values above 30 were considered to have been reported as g/l, so were also divided by 10.

The GRACE score is a validated score to predict outcome following acute coronary score, derived from key patient presenting characteristics. 68,69 As not all data points may be available in the MINAP data set, a mini-GRACE score has been derived, which has been reported and used by the National Institute for Health and Care Excellence (NICE). 70,71 The score has been tested and validated using the MINAP data set. 71 It is derived from eight data points in the MINAP data set, as described in Table 4.

Care pathway variables

This variable group included hospital volume (OHCA cases per year), hospital pPCI capability, EMS response time, EMS travel distance, admitting consultant, cardiological care during admission, admission ward, time point of aspirin administration, place where first 12-lead ECG performed, in-hospital administration of angiotensin-converting enzyme inhibitor/angiotensin receptor blocker, in-hospital use of loop diuretic, reperfusion treatment and timing, assessment at non-interventional hospital, assessment at intervention centre, intended reperfusion procedure, procedure performed, reason for no angiography, reason for no intervention and reason treatment not given.

The following variables were recategorised, as shown in Table 3: time point of aspirin administration, admitting consultant, place first 12-lead ECG performed and admission ward.

For hospital-level data (distance to hospital, volume and PCI centre), we categorised patients by the hospital to which they were first admitted. We were supplied with both a hospital code and hospital location using northings–eastings data. Initial data review suggested that some hospitals may have been assigned more than one code, so the northings–eastings were used to identify individual hospital locations. Northings–eastings were not available for eight hospitals (91 patients), so the MINAP hospital codes were used to categorise these patients.

Distance to hospital was calculated using the Euclidian distance between the participant’s home address (northings–eastings data to the nearest km) and the hospital to which they were first admitted (northings–eastings data to the nearest kilometre). MINAP does not collect data on event location, so we made the assumption that the cardiac arrest event occurred at the patient’s home. This assumption is supported by UK and international data that show that most cardiac arrests happen at the patient’s home. 5,24 For example, English data show that, where location is recorded, 83.2% of OHCA events occur in the home. 5 However, on the basis that some participants would not have had a cardiac arrest at home (e.g. they had it at work or in a public place), we considered how to identify cases where there was clear evidence that the patient had the cardiac arrest outside the home, so that we could classify these data as missing. Our initial plan to create an upper limit proved to be problematic as the distance was likely to be significantly greater in rural areas than urban areas. Therefore, for each case, we calculated the distance between the home address and nearest hospital. When the distance between the participant’s home address and the hospital they were admitted to was large (> 95th centile) compared with the distance between the participant’s home address and nearest hospital, we concluded that the cardiac arrest did not happen at the participant’s home, so we recorded these data as missing.

For hospital volume, we calculated the number of OHCA cases per year at each hospital and categorised volume as low (1–10 cases), medium (11–24 cases) or high (25–82 cases). Patients were then allocated to a category based on the initial hospital in which they were treated.

Primary percutaneous coronary intervention centres were defined as a hospital that performed at least 100 pPCI procedures per year, based on the British Cardiovascular Intervention Society recommendation for interventional centres. 72 We calculated the number of pPCIs each hospital did per year using the entire MINAP data set, using MINAP field 3.39 (initial reperfusion therapy). If a patient was treated in a hospital in a year that it was designated as a pPCI centre, then the patient was categorised as being treated in a pPCI centre.

Emergency medical service response time was defined as the time (in minutes) from the call for help (MINAP field 3.02) to arrival of ambulance (field 3.04) or, if this field was missing, to the arrival of a first responder (field 3.03).

In the MINAP data set, reperfusion treatment data are collected in four key fields, namely initial reperfusion therapy (field 3.39), additional reperfusion therapy (field 3.40), thrombolytic drug (field 3.36) and date/time of reperfusion therapy (field 3.09). In our analysis, we considered only the initial reperfusion treatment, which was classified as either pPCI or thrombolysis. If field 3.39 was missing but the participant was recorded as having received a thrombolytic drug (field 3.36) then they were classified as having received thrombolysis. For each reperfusion therapy, we classified the timing as either early, late or not recorded. We classified early pPCI as a pPCI started within 90 minutes of hospital arrival, and early thrombolysis as administration of a thrombolytic with 60 minutes of the call for help. Timings outside these windows were considered late. Current NICE and European Society of Cardiology guidelines do not specifically state a timeframe within which thrombolysis and pPCI should be performed, except that the chosen therapy should be delivered as soon as possible and thrombolysis should be considered if pPCI cannot be commenced within 120 minutes. 73,74 As such, our threshold for early thrombolysis was based on the standard described in the Department of Health’s National Service Framework for Coronary Heart Disease. 75 Our threshold for early PCI was based on the European Society Guidelines, which describe a period from first medical contact to PCI of up to 90 minutes as an appropriate target. 73

Discharge care variables

Discharge care variables included discharge diagnosis, echocardiography, coronary angiography, coronary intervention, provision of cardiology follow-up, cardiac rehabilitation, discharge drugs (beta-blocker, angiotensin-converting enzyme inhibitor or angiotensin receptor blocker, statin, aspirin, thienopyridine inhibitor, ticagrelor), provision of smoking cessation advice and provision of dietary advice.

Discharge diagnosis, echocardiography, coronary angiography, coronary intervention and discharge drugs were recategorised as detailed in Table 3.

For echocardiography, coronary angiography and coronary intervention, we acknowledged that these interventions may occur before or following discharge and that when undertaken prior to discharge they may inform in-hospital treatment. However, we considered them to be best categorised in this domain.

In addition, for antiplatelet drugs (aspirin, thienopyridine inhibitor, ticagrelor) we created three categories: no antiplatelet therapy on discharge, monotherapy on discharge or dual antiplatelet therapy on discharge. The category of no antiplatelet therapy was used when a patient received none of the three drugs on discharge. Monotherapy was used when a patient received any one of the three drug groups. Dual antiplatelet therapy was used if the patient was prescribed aspirin plus either a thienopyridine inhibitor or ticagrelor.

Classifying data as missing

To ensure that data were valid, we classified some available data as missing, and these were then imputed.

For some continuous variables, we applied upper and lower cut-off values to some continuous variables beyond which data were considered implausible, and thus were classed as missing. For each variable, we carefully considered the MINAP data definition, as well as the balance between when a variable became implausible rather than just unlikely. This was based on clinical judgement and review of the data distribution. The cut-off points that were used are detailed in Table 5.

For categorical variables, data recorded as unknown or not applicable in individual fields were categorised as missing.

Imputation strategy

Data imputation was undertaken following case identification. No outcomes were imputed.

Our imputation strategy was based on the approach described by Cattle et al. ,76 in relation to the MINAP data set. The imputation method for specific data points is described in Table 6.

Default imputation was used for some variables. For the remaining variables, multiple imputation by chained equations (MICE) was used. 77 This involves fitting a model for each variable included in the imputation model. For binary variables, a logistic regression was fitted, where the binary variable for which missing values were being imputed was the outcome variable and the other variables in the imputation model are used as predictor variables. For categorical variables with more than two categories, polytomous regression, which generalises binary logistic regression to allow more than two categories in the variable for which missing values are being imputed was used. For continuous variables, predictive mean matching (PMM) was used to impute missing values. PMM involves fitting a linear model with the imputed value being an observed (non-missing) value sampled from the values closest to the value suggested by the linear model. The models for different variables are fitted in a chain. To achieve true target distributions, several iterations (each iteration consists of one chain of models for all imputed variables) are required.

For the fields that relate to thienopyridine inhibitors at discharge, we were not supplied with data before 2009. We therefore default imputed for the period before 2009, and incorporated two year slopes (2003–8 and 2009–15) in our adjusted analysis.

The MICE package in the R statistical program (The R Foundation for Statistical Computing, Vienna, Austria) (R is a language and environment for statistical computing) was used to perform the multiple imputation. 77

Default imputation was first performed for the appropriate predictor variables (such as medical history predictor variables). Subsequently, we used the resulting data set as the ‘complete-cases’ data set to impute 25 data sets with 30 iterations. Convergence was assessed by checking whether or not the imputation chains mixed well for all variables.

In-hospital mortality and neurological outcome analyses

Analysis of time to all-cause mortality

Proportional hazards Cox regression RE models were used to analyse time to all-cause mortality for each predictor variable. Unadjusted hazard ratios (HRs) were obtained for a model using both the pre-imputation and the imputed data sets. For each variable, we report the HR and 95% CIs, as well as the p-values, which tests the null hypothesis that the HR is equal to 1. A HR greater than one indicates a higher hazard rate of death and a HR less than 1 indicates lower hazard rate of death.

For analysis after multiple imputation, results were combined using Rubin’s rules,78 as for the analysis for in-hospital mortality, except that HR and log(HR) are used in place of OR and log(OR).

Appendix 1 shows details of which variables were included in the model for time to all-cause mortality.

Model fits for different models were assessed using the AIC. For the analysis using imputed data sets, the median (from separate analyses of the 25 imputed data sets) AICs were used. The AIC reported by the ‘coxme’ package is the difference between the AIC of the fitted model and the AIC of the null model, with a higher AIC value indicating a better fit.

For the adjusted analysis, the final model was developed in a similar way to that described for the in-hospital mortality.

Sensitivity analyses

We performed sensitivity analyses to assess the robustness of the results to the missing data methods used and assumptions made. 80,81 For the in-hospital mortality and neurological outcome analyses, we conducted five analyses, namely patients admitted to hospital between 2003 and 2008; patients admitted to hospital between 2009 and 2015; patients where the ECG determining treatment showed ST elevation or left bundle branch block (LBBB); patients where the ECG determining treatment showed something other than ST elevation or LBBB; and a complete-case analysis. When the defining category was missing in the pre-imputation data set, the case was not included in the sensitivity analysis. The rationale for the cut-off point for the two time periods was that it was between 2008 and 2009 that the use of PCI became more frequent than thrombolysis as a reperfusion treatment. 48 For the time to all-cause mortality analysis, we conducted a complete-case analysis. These complete-case analyses included only patients for whom all relevant data points were available.

Data categorised as missing

As described in the methods, we applied cut-off points to some continuous variables, such that markedly outlying values were removed and categorised as missing. In total, these cut-off points affected 4707 values across 10 data fields. Often, this was because data had been incorrectly recorded as zero. Full details are included as Table 7.

Summary of missingness across the study cohort

The missingness in individual data points varied markedly. In our data set, missingness varied between 0% (age, admission time) and 98.3% (why no angiography), with the missingness for most variables being in the range 10–20%. Table 8 shows the missingness for all variables across all patients (cohort used for the hospital mortality/neurological analyses) and the cohort used for the time to all-cause mortality analysis.

For demographic variables, there were only 46 cases with sex missing. The number of cases missing at least one demographic variable was 4182 (23.8%) as missingness for demographic variables was not simultaneous, so, for example, cases missing ethnicity do not necessarily have IMD score missing. After default imputation for medical history variables, the only predictor variables with missing values were diabetes status (n = 1885, 10.7%) and smoking status (n = 3615, 20.5%). The number of cases missing at least one demographic variable or diabetes status or smoking status value was 7281 (41.4%).

In the OHCA presenting characteristics group, location of cardiac arrest was missing for only 67 cases. Systolic blood pressure and heart rate tended to be missing simultaneously, so the number of cases missing either of them was 3348 (19.0%), which is close to the individual missing proportions. Similarly, haemoglobin and creatinine levels tended to be missing simultaneously, with the number of cases missing either of them being 6173 (35.1%). There are few cases of missing data values for cardiac arrest rhythm (n = 1104, 6.3%) and ECG that determined treatment (n = 612, 3.5%), but there are large proportions of missing values for admission diagnosis (n = 3185, 18.1%) and LVEF (n = 9954, 56.5%). Across the OHCA presenting characteristic variable group, there was at least one missing value in 14,520 (82.5%) cases.

In care pathway variables, EMS response time and EMS distance were not generally missing simultaneously, with the number of cases missing at least one value for these variables being 6269 (35.6%). The proportions of missing values for admitting consultant (n = 321, 1.8%) and admission ward (n = 204, 1.2%) were low. The proportion of missing data for the other care pathway predictor variables included in the modelling were between 10% and 20%. Across all cases, the number of cases with a missing value for a care pathway predictor variable was 10,152 (57.7%).

Across all variable groups, the number of cases with at least one missing value was 16,163 (91.8%).

Performance of the imputation

Multiple imputation chains for all variables mixed well, suggesting that convergence was achieved (see Figures 11–14 in Appendix 2).

Patient demographics

In our patient cohort, most patients were male (n = 13,188, 75.1%) and of white ethnicity (n = 14,343, 93.7%) and the mean age was 65.3 years (SD 13.2 years). The mean IMD score, where a higher score indicates increased deprivation, was 22.3 (SD 15.9; range 0.59–85.6).

Past medical history

Most patients were current or former smokers (n = 8883, 63.5%). The most common comorbidities were hypertension (n = 6389, 41.0%), hypercholesterolaemia (n = 3906, 25.9%), previous AMI (n = 3092, 19.7%), angina pectoris (n = 2758, 17.8%), diabetes mellitus (n = 2158, 13.7%) and asthma/COPD (n = 1814, 11.9%). The incidence of other comorbidities (cerebrovascular disease, heart failure, peripheral vascular disease and chronic renal failure) was < 10% for each disease. A minority of patients had previously received a PCI (n = 1061, 6.9%) or CABG (n = 790, 5.1%). Most patients (n = 10,729, 60.9%) had at least one comorbidity.

Presenting characteristics of out-of-hospital cardiac arrest

Most cardiac arrest events occurred before ambulance arrival (n = 10,533, 60.1%), with a presenting rhythm of ventricular fibrillation (VF) or ventricular tachycardia (VT) (n = 14,778, 89.6%). For the non-shockable presenting rhythms, the breakdown between asystole (n = 885, 5.4%) and pulseless electrical activity (PEA) (n = 837, 5.1%) was similar. Most patients were admitted to the hospital during daytime hours (08.00–19.59 hours) (n = 11,741, 66.7%).

The most common admission diagnosis was definite MI (anterior infarction, n = 3897, 27.0%; other infarction site, n = 3639, 25.2%). ST-segment elevation/LBBB was the most common ECG finding (n = 12,220, 71.9%).

The mean admission systolic blood pressure and heart rate were 125.7 mmHg (SD 29.2 mmHg) and 89.2 beats per minute (SD 24.8 beats per minute), respectively.

Care pathway

The median for EMS response time was 8 minutes (IQR 5–14 minutes). The median distance between the patient’s home address and the hospital to which they were first admitted was 8.1 km (IQR 3.9–15.8 km). The patient distribution between low-volume (≤ 10 OHCA cases per year), medium-volume (11–24 OHCA cases per year) and high-volume (25–82 OHCA cases per year) hospitals was 45.4% (n = 7984), 37.0% (n = 6516) and 17.6% (n = 3104), respectively. The first hospital in which most patients (n = 9804, 55.7%) were treated was classified as a PCI centre if the centre performed at least 100 pPCIs in the year that patient was admitted.

Most patients were admitted under a consultant cardiologist (n = 10,680, 61.8%) and received cardiological care during admission (n = 11,960, 90.7%). Just over half of patients were admitted to the cardiac care unit (CCU) (also referred to as the coronary care unit) (n = 8872, 51.0%) and approximately one-third of patients were admitted to the intensive care unit (n = 6154, 35.4%). Patients typically received aspirin or were already on aspirin (n = 14,126, 87.7%) and received a pre-hospital ECG (n = 11,053, 75.7%).

Reperfusion treatment (pPCI or thrombolysis) was delivered to 62.8% (n = 9540) of patients. Of these 9540 patients, the majority received pPCI (pPCI, n = 6160, 64.6%; thrombolysis, n = 3380, 35.4%). Figures 7–9 show the percentage of patients receiving reperfusion treatment by year for all patients (see Figure 7), STEMI patients (see Figure 8) and patients who did not have a STEMI (see Figure 9). All figures show an increase in use of reperfusion over time, although this increase is notably higher in the STEMI group than in the group of patients who did not have a STEMI. In all groups, there is a move away from the use of thrombolysis to pPCI over the study period. The point at which the use of pPCI overtakes thrombolysis use is around 2008–9, such that by the end of the study period very few patients received thrombolysis.

FIGURE 7.

Percentage of patients receiving reperfusion treatment by year (all patients). Adapted from Couper et al. ,85 with permission from Elsevier. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

Adapted from Couper et al. ,85 with permission from Elsevier. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

FIGURE 8.

Percentage of patients receiving reperfusion treatment by year (patients presenting with STEMI). Adapted from Couper et al. ,85 with permission from Elsevier. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

Adapted from Couper et al. ,85 with permission from Elsevier. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

FIGURE 9.

Percentage of patients receiving reperfusion treatment by year (patients not presenting with STEMI). Adapted from Couper et al. ,85 with permission from Elsevier. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

Adapted from Couper et al. ,85 with permission from Elsevier. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

Discharge care variables

In most patients, the discharge diagnosis was ACS (n = 16,476, 95.1%). The majority of patients underwent echocardiography during their hospital stay, or this was planned to take place after discharge (n = 11,140, 73.8%).

Unadjusted analysis

Unadjusted ORs for in-hospital mortality in relation to variables in the imputed data set are presented in Table 10. For categorical predictor, the column, Patients, n (%), describes the number and percentage of patients in that category who died in hospital.

Adjusted analysis

The adjusted model for in-hospital mortality is shown in tabular form in Table 11 and as a caterpillar plot in Figure 10. The model consists of 37 predictor variables and explains 36.1% (R² = 0.361) of the variability in in-hospital mortality across patients. The RE estimate (0.215) and ICC (0.061) are lower than those for the null model, such that the predictor variables explain some variability in in-hospital mortality across hospitals.

FIGURE 10.

Caterpillar plot for OR of in-hospital mortality (adjusted analysis). NA, not applicable.

Sensitivity analyses

Our sensitivity analyses of the primary outcome included a complete-case analysis, 2003–8 data, 2009–15 data, STEMI patients and patients who did not present with a STEMI. These analyses are included in Table 12.

Across these analyses, results tended to be more variable and CIs wider because of the smaller sample size. Nevertheless, the directions of effects are broadly similar to primary analysis results where the effect was statistically significant. However, there are noteworthy differences across the data sets. In the complete case and 2009–15 analyses, PEA as the presenting rhythm is associated with a statistically significant reduction in mortality compared with asystole. In the complete-case cohort, the ORs for most reperfusion treatments change direction and indicate harm. This association is statistically significant for thrombolysis (late) and thrombolysis (time missing).

In the cases that did not present with a STEMI, admission during the night (between 20.00 and 07.59 hours) was associated with increased mortality. For the year slope between 2003 and 2008, each year was associated with a reduction in mortality. For hospital PCI capability, the direction of OR changed, suggesting benefit to admission to a PCI centre in this patient group, although this did not reach statistical significance. Importantly in this subgroup, there was no evidence of reduced mortality with the use of a reperfusion treatment.

Adjusted analysis

The adjusted analysis for neurological outcome at discharge is presented in Table 13. The results of the analysis are similar to those reported for the primary outcome (in-hospital mortality), although there are a few important differences.

In the baseline demographic category, sex was found not to be predictive of neurological outcome. In the care pathway category, the location where the ECG was performed did influence outcome, such that in-hospital ECG was associated with a poorer outcome. In contrast to the primary outcome analysis, there was no evidence of an association between transfer distance or admission to a pPCI centre and neurological outcome. Finally, in this analysis there was a statistically significant association between late thrombolysis and improved neurological outcome.

Sensitivity analyses

In sensitivity analyses, there was a similar pattern to that observed in the primary outcome sensitivity analyses in that the reduced sample sizes led to greater variability in results and wider CIs (Table 14). However, the results in the sensitivity analyses were broadly similar to those reported in the analysis of the whole cohort.

There were a few results of note. In relation to admission to a PCI centre, this, counterintuitively, was associated with worse outcome for STEMI patients but improved outcome for other cases. Furthermore, reperfusion treatment was found to be of no benefit in the cohort of patients who did not have a STEMI.

Overview of cohort

The analysis of time to all-cause mortality included the 12,483 patients who survived to hospital discharge for whom data were available on outcome.

The characteristics of this cohort are included in Table 15 for both the pre-imputation and the imputed data sets. As was the case for the in-hospital mortality patient cohort, the pre-imputation and imputed data sets for the time to all-cause mortality patient cohorts are similar. Furthermore, the imputed data sets for both hospital mortality and time to all-cause mortality cohorts are broadly similar (Tables 9 and 15).

The key, albeit small, differences between cohorts typically relate to variables that were associated with hospital mortality. For example, patients in the time to all-cause mortality cohort tended to be slightly younger [mean age 65.3 (SD 13.2) years vs. 63.3 (SD 12.8) years], were more likely to be male (75.1% v 78.2% patients), less likely to have a comorbidity (e.g. heart failure, 5.0% vs. 3.5% of patients), more likely to have a cardiac arrest after ambulance arrival (39.9% vs. 47.7% of patients), initial rhythm was more likely to be VF/VT (89.6% vs. 95.9% of patients) and were more likely to have received reperfusion treatment (62.9% vs. 68.4% of patients).

In relation to discharge variables, most patients were discharged on the following drugs: a statin (n = 9339, 95.1%), angiotensin-converting-enzyme (ACE) inhibitor (n = 8898, 91.2%), antiplatelet therapy (dual therapy, n = 5652, 59.3%; monotherapy, n = 3576, 37.5%) and beta-blocker (n = 8844, 90.1%). In addition, patients were usually followed up by a cardiologist (n = 8498, 93.3%) and received cardiac rehabilitation (n = 9683, 90.6%).

Unadjusted analysis

The unadjusted analysis of time to all-cause mortality is included in Table 16.

Among demographic variables, both increased age and female sex were associated with worse outcome. Ethnicity was not associated with outcome. In contrast to other outcomes, although there was a trend to worse outcome as deprivation increased, this did not reach statistical significance in this analysis.

For medical history variables, most medical conditions (previous AMI, heart failure, chronic renal failure, previous angina pectoris, diabetes mellitus, cerebrovascular disease, hypertension, asthma or COPD, peripheral vascular disease, previous CABG and previous PCI) were associated with an increased mortality. For smoking status, the ever smoked category was associated with reduced mortality. Hypercholesterolaemia was not associated with mortality.

For presenting characteristics, most results were similar to the unadjusted in-hospital mortality with factors such as cardiac arrest following ambulance arrival, initial rhythm of VF/VT, and increased haemoglobin levels associated with improved survival. There was an association between admission between 20.00 and 07.59 and improved outcome.

For care pathway variables, results were again similar to the unadjusted in-hospital mortality analysis. Factors such as reperfusion therapy (both pPCI and thrombolysis) and admission under a cardiologist were associated with reduced mortality. However, in this cohort, patient admission to very low volume hospitals was associated with worse outcome, while admission to a PCI centre was associated with improved outcome. Importantly, the need to be admitted to an intensive care unit did not influence long-term outcome.

Discharge interventions such as drugs (antiplatelet therapy, statins, ACE inhibitor, beta-blockers) were all associated with improved outcome. Similarly, follow-up by a cardiologist and cardiac rehabilitation were associated with improved outcome.

Adjusted analysis

The results for the adjusted analysis are included as Table 17.

For demographic variables, increased age and deprivation were predictive of increased mortality, but ethnicity and sex were not associated with mortality.

In the adjusted model, only four medical history variables (previous AMI, heart failure, diabetes mellitus, and asthma or COPD) were associated with worse outcome. In addition, the effect estimate for smoking status changed direction in this model so the ever smoked category became associated with increased mortality. The remaining medical conditions were not associated with outcome.

In the presenting characteristics of OHCA variable section, laboratory values (haemoglobin, cholesterol, glucose, creatinine) were associated with the outcome, although the effect size was small. There was a trend towards improved outcome when the arrest occurred after ambulance arrival, although this did not reach statistical significance. In this model, time of day of admission was not associated with outcome. The year slope suggested worse outcome over time.

None of the care pathway variables was associated with time to all-cause mortality. Although, generally, the effect estimate direction was the same as in previous models, the estimate was closer to one and the result was not statistically significant.

The discharge care variables that do not predict hazard of death are echocardiography, discharge diagnosis, antiplatelet prescription and statin prescription. The provision of coronary angiography, cardiology follow-up and cardiac rehabilitation were associated with reduced the risk of mortality. Similarly, discharge on beta-blocker or ACE inhibitor were associated with improved outcome.

Sensitivity analyses

We undertook a complete-case analysis of 1381 (11.1% of cohort) patients with complete data (Table 17). The results were broadly similar to the adjusted analysis of the imputed data set, although some effects switched direction and/or became non-significant. For example, in the imputed data set, the HR for previous AMI was 1.394 (95% CI 1.2 to 1.6), but the HR was 0.850 (95% CI 0.5 to 1.5) in the complete-case analysis.

Modifiable factors

The decision as to where to transport a patient following successful resuscitation from OHCA is challenging. For pre-hospital clinicians, the decision requires careful consideration of the potential benefits to the patient of access to specialist services and clinical teams at a large centre versus the risk of longer transport times. In our analysis, we used categorisation as a pPCI centre and hospital volume as markers of large, specialist centres. In our analysis, neither of these factors was associated with improved outcome and our primary analysis found that admission to a pPCI centre was associated with worse outcome.

In other health areas, such as trauma, surgery and critical care, increased patient volume is associated with improved outcomes at a hospital level. 86–89 In MI, there appears to be similar evidence of a relationship between increased hospital volume and improved patient outcome. 90,91 Furthermore, in patients treated with pPCI, increased hospital PCI volume is also associated with improved survival, although there is no association in patients treated with thrombolytics. 37,38,92 Based on such data, the British Cardiovascular Intervention Society recommend that hospitals must undertake at least 100 pPCIs per year to function as a pPCI centre. 72

For OHCA, however, there is inconsistency between studies as to the relationship between hospital volume and its effect on patient outcome. 93–98 For example, Schober et al. 94 included 2238 OHCA patients admitted to seven hospitals in Vienna, Austria, and found that the highest volume centre reported the highest survival rate. In contrast, Cudnik et al. 93 analysed data from 4125 patients with OHCA of cardiac origin admitted to 155 US hospitals, but found no association between volume and patient outcome. Instead of volume, it may be that for OHCA patients it is the specialist services (e.g. critical care, PCI) that are available at the treating hospital, rather than its volume, that are most associated with patient outcome. 30,34,98–103 For example, Stub et al. 34 scored 111 North American hospitals based on the availability of coronary angiography, targeted temperature management and use of delayed prognostication, and correlated these performance measures with 3252 OHCA cases from the Resuscitation Outcomes Consortium. In an adjusted analysis, the authors reported that the highest performing hospitals reported the highest rates of hospital survival and survival with good neurological outcome.

This correlates with our finding that, although admission to a pPCI centre may be associated with increased mortality, the actual delivery of reperfusion treatment appeared to be the most important modifiable factor associated with improved survival in this patient cohort, particularly when delivered early. In our patient cohort, over 50% of patients received reperfusion treatment. Both early thrombolysis and primary PCI seemed to significantly reduce the risk of in-hospital death across all cases, although sensitivity analyses suggested that this effect was limited to STEMI cases. This supports current recommendations that resuscitated OHCA patients with a STEMI should be considered for reperfusion. 73,74,104 In contrast, guidelines for OHCA patients who do not have a STEMI recommend that these patients may be considered for reperfusion once non-cardiac causes for the OHCA have been ruled out. 74,104,105 This is seemingly driven by evidence that these OHCA patients with cardiac arrest resulting from a cardiac cause frequently have an occluded coronary vessel. 16,99,106,107 However, observational studies of the effect of PCI following OHCA in patients who have not had a STEMI have produced mixed results, with some finding evidence of a survival benefit while others report that there is no effect on survival. 16,107–110

Over the course of our study, we noted that the use of thrombolysis decreased as the use of pPCI increased, such that overall there was an increase in the use of reperfusion treatment. In 2015, three-quarters of patients presenting with a STEMI in the context of OHCA received pPCI. In contrast, less than 15% of patients who did not present with a STEMI received pPCI. This is consistent with data from across the MINAP data set and correlates with data showing that pPCI is the preferred reperfusion strategy in STEMI. 48,111,112

Following successful resuscitation, the management of OHCA patients may be complex owing to post-cardiac-arrest syndrome. 21 This is evidenced in an observational study of 248 post-cardiac-arrest patients transferred by a critical care transfer team in which a critical clinical event occurred in 23% of patients and 6% suffered a rearrest. 113 There is therefore a possible risk associated with prolonged transportation. Counterintuitively, our primary outcome analysis suggested that for each kilometre travelled, mortality reduced (OR of death 0.994, 95% CI 0.989–0.999). The explanation for this finding is unclear, particularly as it contrasts with a previous observational study of 10,315 critically ill patients transported by English Ambulance Services between 1997 and 2001, which found that each additional 10 km travelled was associated with an absolute increase in mortality of 1%. 114 Importantly, however, this study excluded OHCA patients. International data from OHCA patients suggests that there is no association (benefit or harm) between distance or time travelled and patient outcome. 115–117

There are a number of possible explanations for this unexpected finding. The first is that the analysis was based on the assumption that cardiac arrests included in the analysis occurred in the home. We chose this approach, as MINAP records only the patient’s home postcode rather than the location of the cardiac arrest. This approach seemed reasonable as other studies show that most cardiac arrests occur in the home, although we are unable to test this in our population. 5,24 Interestingly, the median transport distance reported in this study was only slightly longer than that reported in the study by Nicholl et al. 114 (median 5 km vs. 8 km). Second, this may be a type I statistical error, particularly given the small effect size and the associated 95% CI, which was close to 1. Notably, our reported point estimate was similar to the point estimates reported in other studies. 115,116 A third explanation is that this is a true effect due to patients being transferred longer distances to specialist centres for treatment, which were associated with improved outcome, but the effect was confounded by other variables. In Cudnik et al. ’s115 observational study of 7540 patients who had suffered an OHCA resulting from a cardiac cause, there was evidence of an association between being transferred to the nearest hospital, rather than a more distant hospital, and increased mortality. 115

In this study, the use of pre-hospital ECG was not associated with improved survival in the analysis of the primary outcome, but was associated with improved neurological survival in our secondary analysis. The use of pre-hospital ECG is recommended by international guidelines to triage the patient to a specialist centre for pPCI in the event that the ECG shows a STEMI. 73,104 A previous study demonstrated how the use of pre-hospital ECG was associated with increased timeliness of pPCI and reduced mortality. 51

Synthesising these data surrounding pPCI centre, reperfusion treatment, hospital volume, pre-hospital ECG and EMS transport distance is challenging. However, it would seem reasonable to suggest patients in this cohort should receive a pre-hospital ECG following ROSC in order to triage them to appropriate care. If there is an indication for reperfusion treatment, then it would be reasonable to transfer the patient to a specialist centre to receive this treatment. If there is uncertainty as to the cause of the cardiac arrest and the patient’s ECG does not show ST elevation, then the patient may be transferred to the nearest hospital with appropriate facilities to meet the patient’s care needs (e.g. critical care, cardiology services), which may also be beneficial for relatives. Several clinical trials are planned or in progress to address the role of pPCI among patients with NSTEMI, which should help to reduce uncertainty about how best to manage this patient group (e.g. Patterson et al. 42).

In patients who survived to leave hospital, discharge interventions were strongly associated with long-term mortality, such as discharge on a beta-blocker and ACE inhibitor and referral to coronary rehabilitation. These interventions are recommended in current guidelines. 73,105 In contrast, antiplatelet therapy and statins on discharge were not associated with long-term mortality in our study. The reason for this finding is unclear as there is good evidence that they are effective in reducing long-term mortality following MI. 118,119 This may be partly explained by a selection bias, as patients with a better long-term prognosis, as indicated by blood pressure and renal function, may be more likely to be prescribed beta-blockers and ACE inhibitors, whereas prescription of antiplatelets and statins may be less influenced by these factors. We also acknowledge that our antiplatelet analysis may have been affected by the non-availability of pre-2009 thienopyridine inhibitor data. Furthermore, a large proportion of patients received both aspirin and statins on discharge, so our analyses may have had insufficient power to reliably detect a difference between groups.

The findings from this study inform the recent National Framework120 to improve care of people with OHCA in England. The framework recommends that patients who achieve ROSC are transferred to recognised centres of care that have the expertise and facilities to provide round-the-clock access to cardiac catheterisation and an intensive care unit. Although the present study did not find evidence of improved outcomes based on PCI capability or hospital volume, no harm was seen from increasing transfer distance from scene to hospital.

Non-modifiable factors

In our study, a number of demographic and medical history variables were associated with both in-hospital and long-term mortality. Our finding that women were less likely to survive than men conflicts with the results of a recently published systematic review, which compared male and female survival following cardiac arrest. 121 The meta-analysis of 409,323 patients from 13 studies reported that women were more likely to survive cardiac arrest (OR for survival 1.10, 95% CI 1.03 to 1.20), despite having worse baseline characteristics such as increased age and being more likely to have a cardiac arrest at home and be in a non-shockable rhythm. A sensitivity analysis that was limited to OHCA resulting from a cardiac cause produced similar findings.

In contrast, female sex in the MI literature is usually associated with worse outcome. 122–125 Women who present with MI are typically older and have more comorbidities, but there is also evidence that women are less likely to receive evidence-based interventions such as a pre-hospital ECG or PCI, which may explain the apparent discrepancy in outcome. 51,122–124,126,127 In our study, we did not examine sex differences in relation to demographics or treatment delivery.

Our study found that social deprivation was associated with increased hospital mortality and time to all-cause mortality. This finding has been reported previously in both the MI and cardiac arrest literature. 128–134 However, this relationship is likely to be complex given that factors such as race and lifestyle may be associated with both socioeconomic class and risk of cardiovascular disease. 135,136 As is the case with sex, this difference in survival may be partly explained by differences in treatment, such that patients who are more socially deprived are less likely to receive key interventions such as bystander CPR and PCI. 130,132,134,137–140 For example, an observational study of OHCA undertaken in the north-east of England reported that bystander CPR rates ranged from 14.5% in the most deprived areas to 23.2% in the least deprived areas. 138 The studies that describe a relationship between PCI and social deprivation were undertaken in health systems that are markedly different from the UK, such that the generalisability of these data to the UK setting is unclear. Indeed, a Scottish study found no evidence of an association between socioeconomic class and use of coronary angiography and intervention following MI. 141 Our analysis found that social deprivation was associated with mortality even after adjustment for baseline and treatment variables, suggesting that the precise mechanism by which social deprivation is associated with mortality is likely to be complex.