OBJECTIVE To develop a performance indicator for acute myocardial infarction which would reliably measure success of treatment and which might provide an alternative to case fatality as an audited outcome.
DESIGN A two year audit of all cases of acute myocardial infarction and resuscitated cases of out of hospital cardiac arrest from coronary heart disease in patients under 75 years of age. Behaviour of patients in calling for help, performance of the ambulance services in treating out of hospital arrest, and of the hospitals in providing resuscitation and thrombolytic treatment are audited separately.
SETTING Four district general hospitals.
AUDITED INTERVENTIONS Resuscitation from cardiac arrest and thrombolytic treatment.
MAIN OUTCOME MEASURES Hospital case fatality and lives saved/1000 patients treated.
RESULTS Overall, the lives of 83/1000 patients were saved (95% confidence interval 70 to 96). Of these, 29 (35%) were saved by out of hospital resuscitation and 38 (46%) by in hospital resuscitation from cardiac arrest. It was estimated that 16 lives (19%) were saved by thrombolytic treatment. There were no significant differences in case fatality among the hospitals.
CONCLUSIONS Lives saved/1000 patients treated is an easily measurable index and assesses performance of the ambulance service as well as of the hospital. Because it is relatively insensitive to diagnostic definitions, it may provide a robust alternative to case fatality as a performance indicator.
- acute myocardial infarction
- case fatality
- outcome indicators
Statistics from Altmetric.com
There is a growing perception that delivery of health care must be audited1 and outcome indicators for acute myocardial infarction have been proposed.2 3 However, for any audit system to be meaningful, outcome measures must be unequivocal and as far as possible insensitive to variations in case mix and differences in diagnostic definitions.
The present two year study was carried out in four district general hospitals to establish a system for audit of acute myocardial infarction which was cost-effective and which might measure success of treatment as an outcome, in addition to case fatality. Data recorded were based on experience from the earlier UK heart attack study (UKHAS),4-6 but omitting patients who died outside hospital (recorded by UKHAS). Data from the in-hospital course were reduced to those items which, from the results of UKHAS, were found to be essential for reliable assessment of outcomes. In addition to the audit programme, a previously reported public educational campaign (Heart Attack Action!)7 was extended during the second year of the study to the districts served by all four hospitals. Results of this latter project, which was aimed at reducing patient delay in calling for help, will be reported separately.
At three of the participating hospitals a senior nurse with coronary care experience worked for one day a week on the project, while at the fourth hospital a member of the audit staff with nursing experience worked for one day a week. Another senior nurse acted as co-coordinator. Data collection started on 1 January 1997 in Brighton, on 1 April 1997 in the other hospitals, and finished on 31 March 1999 at all hospitals.
As in the earlier study, identification of cases was carried out by a combination of “hot” and “cold” pursuit,8 and events were recorded only in patients under 75 years of age. Cases of myocardial infarction showed at least two of the following three features: a typical or compatible clinical history; sequential ECG changes; and a rise in creatine kinase activity to at least twice the upper limit of normal for the hospital laboratory. Cases of out of hospital cardiac arrest were included if the cause was ischaemic heart disease and they survived with an independent cardiac output to reach hospital. Cases of myocardial infarction or sudden cardiac death occurring in patients who had been admitted for another reason, and had already been in hospital for 24 hours or more, were excluded. As a check on the completeness of inclusion, lists of patients in the appropriate age range with a coded diagnosis of acute myocardial infarction were obtained periodically from the clinical coding departments of the participating hospitals. Case notes of omitted patients were obtained and scrutinised for possible inclusion if study criteria were met. If study criteria were not met the cases were excluded, even though they had been coded as myocardial infarction.
Selection of cases was thus without reference to the final diagnosis as recorded using the 10th revision of theInternational Classification of Diseases(ICD 10) by the hospitals' clinical coding department. In order to discover possible discrepancies between coding diagnoses and our own designation of acute myocardial infarction made using predefined criteria, we searched at one of the hospitals (hospital A) after completion of the study for study cases which had not been classified as acute myocardial infarction by clinical coding, and for coded cases not recorded by the study. This was achieved with the help of the information technology department using the hospital number and the computer operated patient administration system.
Data were recorded on standard forms at each hospital and entered on computer at the coordinating centre using a Paradox database. As in the previous study, we recorded the times of onset of symptoms, call for help, arrival of ambulance at home and at hospital, arrival in the coronary care unit or medical ward, administration of thrombolytic treatment (if given), and discharge from hospital or death. We also recorded details of the diagnostic ECG, whether or not the patient had cardiac failure (diagnosed clinically or radiologically), and the final diagnosis (Q wave or non-Q-wave infarction). For patients who had had a cardiac arrest, whether resuscitation was attempted or not, we recorded the date and time, place (out of hospital, ambulance, accident and emergency, coronary care unit, ward, or elsewhere in the hospital), the witness, the initial rhythm (if known), and survival or not from the initial resuscitation to the time of hospital discharge. Details of first arrests only were recorded. If the patient was resuscitated but later had a further cardiac arrest and died, resuscitation was deemed to have been unsuccessful.
Progress of the project was discussed at investigators' meetings which were held three to six monthly. Because we were aiming, through the public educational campaign, to bring patients under care more quickly, we analysed patient behaviour in calling for help from the ambulance service or from their general practitioner, and consequent delays in coming under care. We next examined ambulance performance in speed of response to the call and in bringing patients to hospital, and in the number of patients successfully resuscitated from out of hospital cardiac arrest. Finally we studied performance of hospitals in delivering both resuscitation from cardiac arrest and thrombolytic treatment.
Because lives saved/1000 patients treated is highly sensitive to delay5 and measures ambulance as well as hospital intervention, we used this as an alternative to case fatality as a performance indicator. As in the earlier study, we assumed that the lives of all hospital survivors from cardiac arrest had been saved by treatment. We used a figure of 30 lives saved/1000 patients treated9 as the measure of success of thrombolytic treatment; calculations based on delay in administration of thrombolysis10 gave almost identical results to the mean figure of 30/1000. The formula used for calculation of lives saved per 1000 treated was thus: Thus if 50% of patients received thrombolytic treatment, the number of lives saved by thrombolysis was 15/1000. For the purpose of this report, we designate the four hospitals with their associated ambulance services A, B, C, and D.
Comparisons were made using the χ2 test.
We recorded 1759 cases, of which 640, 322, 405, and 392 had been treated, respectively, at hospitals A, B, C, and D.
In all, 81% of patients arrived at hospital by ambulance, and of these the majority called the ambulance directly (table 1). More patients did this in centre A than in the other centres (63%v 50%, 37%, and 51%; p < 0.001). Patients who called the ambulance directly came under ambulance care (median delay from symptom onset 1.2 hours v4.1 hours) and arrived at hospital (median delay 1.7 hoursv 4.8 hours) much more quickly than those who consulted their general practitioner before the ambulance was called. Surprisingly, however, patients in centre A, who were more likely than those in the other centres to call the ambulance directly, did not as a group come under ambulance care (median delay 1.6 hoursv 1.7, 1.7, and 1.6 hours) or hospital care (median delay 2.5 hours v 2.5, 2.6, and 2.3 hours) earlier than those from the other centres. The main reason for the longer delay for patients calling their general practitioner was delay by the patients themselves in calling for help (data not shown). In particular, centre A patients who called their general practitioner appeared to have delayed much longer than the larger proportion who called their general practitioner from the other centres. The minority of patients who used their own transport arrived nearly as quickly (median 1.9 hours from symptom onset) as those who called the ambulance directly (median delay 1.7 hours). Only 15% of patients arrived at hospital within one hour of onset of symptoms and 42% within two hours of onset.
AMBULANCE SERVICE PERFORMANCE
Performance of the ambulance services in treating out of hospital cardiac arrest is compared in table 2. The proportions of patients successfully resuscitated are biased towards success, because patients who did not reach hospital alive were not included in the study. Overall, 78 patients (5% of those who were transported by ambulance) had had restoration of cardiac output following a verified arrest, and in a little over one third of these the arrest had been witnessed by ambulance personnel. Fifty three per cent of these patients survived to be discharged from hospital, and success was more likely (p < 0.001) for witnessed arrests (86%) than for unwitnessed arrests (34%). Significantly fewer cases of successfully resuscitated cardiac arrest were transported by ambulance service D than by the other services.
Case mix, incidence and outcomes from cardiac arrest, use of thrombolytic treatment, and case fatality for the four hospitals—together with the total results from all the hospitals—are shown in table 3. Indicators of case mix were similar at each of the hospitals except that hospital D recorded significantly fewer cases of out of hospital cardiac arrest (1.3%) compared with the other hospitals (5.9%, 5.6%, and 4.2%). There was no significant difference among the hospitals in the proportion of patients who suffered a first cardiac arrest in hospital (overall 13%), the proportion in whom resuscitation was attempted (72%), or the proportion of arrests which were recorded to have been caused by ventricular fibrillation (37%); neither was there any difference in the proportion of patients with arrest who were successfully resuscitated and discharged from hospital (39% of all treated arrests and 70% of arrests known to have been caused by ventricular fibrillation). Use of thrombolytic treatment was also similar (given to 55% of patients on average), as was the proportion of patients given thrombolysis who had the conventional ECG indications of ST elevation or left bundle branch block (94%). The “door to needle” time (mean 45 minutes) was also not significantly different among the hospitals, although the door to needle time was less than 25% of the total pain to needle time (median 3.3 hours). Only 2% of the patients who were given thrombolytic treatment received it within the “golden hour”9 from the onset of symptoms, and 24% within two hours. Case fatality in hospital D (8.9%) was lower than in hospital A (13.2%), and this was of borderline significance (p < 0.05). However, this difference became non-significant (8.7%v 10.5%) when cases of out of hospital arrest were excluded.
Although indicators of case mix were similar at each of the hospitals (table 3), case mix had the expected powerful effect on case fatality (table 4). We considered age in the three decades of < 55 years, 55–64 years, and 65–74 years. Fatality was more than three times as great in the highest than in the lowest age decade, and was significantly higher in women than in men in the lowest and highest decade. Left ventricular failure increased fatality nearly 10-fold (37% v 4%) and out of hospital resuscitation nearly fivefold (47% v 10%). However, the proportion of patients resuscitated outside hospital was the only case mix variable that differed significantly among the hospitals (table 3).
LIVES SAVED/1000 PATIENTS TREATED
Lives saved/1000 patients treated by resuscitation from cardiac arrest performed by the ambulance service, by resuscitation performed in hospital, and by thrombolytic treatment (in all cases given in hospital) are shown in fig 1. Because well under 1000 cases were treated by each individual hospital or ambulance service, 95% confidence intervals (CI) of the proportions for each hospital were wide, and there were no significant differences between them. Overall, the lives of 83 patients (95% CI 70 to 96) per 1000 were saved. Thirty five per cent of the salvage was attributable to out of hospital resuscitation, 46% to in hospital resuscitation, and 19% to thrombolytic treatment.
CORRESPONDENCE BETWEEN STUDY AND CLINICAL CODING DIAGNOSES
Of the 640 cases which we had recorded at hospital A, 135 had not received a first coding diagnosis for acute myocardial infarction (ICD 121 or 122). An additional 71 cases had been coded ICD 121 or 122 but had not been recorded or had been rejected for study inclusion. Thus the sensitivity of clinical coding for identification of study cases from the number of cases coded divided by the number identified by the study was 505/640, or 79%. The positive predictive value of the coding diagnosis to predict the study diagnosis was the number of study cases coded divided by the total number coded, 505/[505 + 71], or 88%. The hospital fatality rate for the 135 study cases not coded ICD 121 or 122 was 18%, and of the 71 coded cases not included in the study, 28%.
Coded diagnoses of the 135 non-coded cases of acute myocardial infarction included in our study are shown in table 5. The most common cause of disagreement—applicable to nearly half the cases—was confusion between myocardial infarction and unstable angina. Case notes of 55 of the 71 cases coded as myocardial infarction which we had not included had been reviewed during the study and had been rejected for inclusion. The notes of the other 16 patients had been unobtainable by the end of the study. Reasons for rejection had been acute myocardial infarction or sudden death of patients already in hospital for another reason (excluded by the study protocol), necropsy evidence of recent infarction in patients who appeared clinically to have died from another cause, and cases of prolonged chest pain in which we considered that study criteria had not been met.
Because between two thirds and three quarters of all deaths, comprising one third of all cases of acute myocardial infarction, occur outside hospital,4 8 we considered it important—as far as was possible in these hospital treated patients—to separate the prehospital from the hospital phase of the illness. We therefore considered patient delay in calling for help and patient behaviour in calling their general practitioner or dialling the emergency number as the first variable for audit, followed by performance of the ambulance services in resuscitation from cardiac arrest, and performance of the individual hospitals in delivering resuscitation and thrombolytic treatment.
Although it is well known that patients who use an emergency number to telephone the ambulance service directly arrive at hospital earlier than those who contact their general practitioner, the difference of nearly three hours was even greater than previously described.5 11 Nearly all of the extra delay in coming under care was attributable to delay by patients who clearly had a lesser sense of urgency, presumably because they had less severe symptoms.12 The reason why centre A patients were more likely to call the ambulance directly may have been a pilot of the Heart Attack Action! public educational campaign, which had been promulgated through general practitioner surgeries in centre A, but not the other centres, since the beginning of 1995.7 The message of Heart Attack Action! was “Chest pain lasting longer than 15 minutes. Call 999 for an ambulance.” It appeared that the second part of the message had been acted upon, but the urgency implied by the first part of the message had not.
Because hospital fatality rates must be determined to a large extent by the proportion of high risk patients which are brought to it by the ambulance service, ambulance performance needs to be considered among the explanatory variables for the assessment of hospital performance. Hospital D, which serves a largely rural area and is situated about 15 miles distant from two other hospitals which were not included in the study, received significantly fewer cases of out of hospital arrest than the other three participating hospitals. This may have contributed to the lower fatality rate at hospital D. The importance of the ambulance in saving lives is emphasised by the fact that 41 (39%) of the 105 patients successfully resuscitated and discharged from hospital owed their lives primarily to the ambulance service.
Apart from the difference just noted, case mix—with its expected major effect on outcomes (table 4)—was similar at the four hospitals. Likewise hospital performances in resuscitation, delivery of thrombolytic treatment, and hospital fatality rates (if the differing proportions of out of hospital arrests were excluded) were also similar. The low proportion of hospital arrests verified to have been caused by ventricular fibrillation (37% overall) was surprising, although this figure included all first cardiac arrests, whether resuscitation was attempted or not. Although the proportion of patients given thrombolytic treatment (55%) corresponded exactly with current guidelines13 and 94% of patients given thrombolysis had the accepted ECG indications, delays to administration were still far from ideal. Only 2% of patients were treated within the golden hour, during which period thrombolysis—in terms of saving lives—has been estimated to be twice as effective as when given later.10Only half the patients achieved the recommended target for thrombolytic treatment to be started within 90 minutes of calling for help.14 Door to needle time, although an important indicator of hospital staff performance, accounted for less than 25% of the total delay; the median door to needle time of 45 minutes was similar to that reported recently from 15 UK hospitals.11Administration of thrombolysis in the accident and emergency department, as was the practice in hospital A, did not materially alter the hospital delay time compared with administration after direct admission to the coronary care unit, as was the policy in hospital B. The only realistic strategy for substantial reduction in delay appears to be administration of thrombolysis during the prehospital phase.
LIVES SAVED VERSUS CASE FATALITY AS PERFORMANCE INDICATORS
The present figure of 83 (70–96) lives saved/1000 patients treated was higher than that previously reported from UKHAS5 (64 (54–74)). However, lives saved by out of hospital resuscitation were omitted from the previous calculation. Inclusion of ambulance resuscitated cases in the UKHAS results gives a figure of 83 (75–91) lives saved, which is the same as the present figure.
Even with the use of similar protocols for case detection and selection, however, hospital fatality in the present survey (11.5%) was 20% lower than in the earlier UKHAS study (14.2% when deaths after discharge from hospital were excluded).5 We are unable to account for this discrepancy, as strenuous efforts were made in both studies to identify all deaths, and in particular all those that occurred in accident and emergency departments or in tertiary referral centres, to which some patients had been transferred for further investigation or treatment. Although it would not explain the above discrepancy, much of the variation of case fatality between published studies and “official” figures15 may reflect differences of definition between clinical coders and auditors on the one hand and clinicians on the other. In hospital A in the present study, sensitivity and positive predictive value for clinical coding to predict study diagnoses were 79% and 88%, respectively, and both of the omitted groups of patients had fatality rates (18% and 28%) which were higher than that which we found in study patients (13.2%). It can be calculated that the removal of 135 cases with 18% fatality and the addition of 71 cases with 28% fatality would have increased the fatality rate which we found from 13.2% to 13.9%. Because the numbers of fatal cases added and removed in hospital A were approximately balanced, this difference is small, but the difference may have been larger at the other three hospitals in our study.
As a performance indicator, lived saved/1000 treated is attractive because it measures the success of treatment rather than its failure. Moreover the numerator can be measured unambiguously as patients who have been resuscitated or treated with thrombolysis can easily be identified. Inevitable disagreements between clinical and coding diagnoses, which we found to involve a large proportion of fatal cases, should not affect calculation of this index unduly, because such cases affect only the denominator and not the numerator of the indicator. “Lives saved” also emphasises the importance of advanced life support—both in and out of hospital—as the cornerstone of treatment for acute myocardial infarction.
LIVES SAVED BY THROMBOLYTIC TREATMENT
The indicator does not acknowledge the life saving antithrombotic role of aspirin, which was found in the ISIS-2 trial16 to be similar in degree to that of streptokinase (about 24/1000 lives saved by aspirin v about 28/1000 saved by streptokinase). Because the use of aspirin is now near universal, we thought it unnecessary to include it in the indicator, but it should be noted that the figure of 16 lives saved/1000 patients treated underestimates the efficacy of thrombolytic treatment when combined with an antithrombotic drug.
Neither does the indicator assess theprocess of thrombolysis, and in particular delays from onset of symptoms to the start of treatment. Trials comparing different thrombolytic regimens show an important adverse effect of delay.17 18 However, this effect was almost abolished—at least for delays of less than six hours—by correction for baseline prognostic variables.18 The most reliable method for estimation of the effects of delay comes from meta-analyses of the clinical trials9 10 in which survival after thrombolysis was compared with survival with no treatment.
The fibrinolytic therapy trialists (FTT) collaborative group9 concluded that the loss of efficacy with delay in treatment was linear, with an extra 1.6 per 1000 lives lost for each hour of delay. For patients with ST elevation or bundle branch block treated within six hours of onset, a figure of 30 lives saved/1000 treated could be assumed.9 Boersma and colleagues,10 by adding results from smaller studies of prehospital thrombolysis to the meta-analysis, concluded that loss of efficacy with time was curvilinear, with efficacy almost doubled during the first hour, at 65 (38–93) per 1000 v 37 (20–55), 26 (14–37), and 29 (19–40) at one to two, two to three, and three to six hours, respectively. Comparison of the FTT estimates with those of Boersma and colleagues shows that differences between the two estimates are essentially confined to the first hour. Calculated differences in efficacy between one to two hours and three to six hours, according to Boersma and colleagues, were little different from the 1.6/1000 per hour estimate of the FTT collaborators, particularly when the wide confidence intervals are taken into account. In the present survey—as in the previous UKHAS study5—only 2% of treated patients started their treatment within the first hour, so differences in times from onset of symptoms to start of treatment between the four hospitals did not materially affect the overall figure of 30/1000.
The above does not deny the importance of reducing hospital delay in starting thrombolytic treatment to improve the process of care, but results of the meta-analyses do suggest that such an effect on process is unlikely to improve outcomes significantly because the major delay has already occurred. This estimate might need revision if restoration of blood flow to the infarct related coronary artery could be accomplished more quickly than at present. The thrombolytic drug used for most patients in our hospitals—as in the meta-analyses9 10—was streptokinase. Administration of a more rapidly acting plasminogen activator to a substantial proportion of patients during the prehospital phase and within one hour of onset of symptoms, or primary angioplasty performed immediately after arrival at hospital, might make the figure of 30/1000 lives saved an underestimate. If a larger proportion of patients were to come under care earlier, many more lives would also be a saved by resuscitation from cardiac arrest.
THE PRESENT METHOD AS A POSSIBLE MODEL FOR FUTURE AUDIT PROGRAMMES
We believe that the present audit method provides potential advantages over other hospital based methods for audit of acute myocardial infarction. First, by separating patient behaviour in calling for help and ambulance performance in treating out of hospital arrest from hospital management, it gives due weight to the all important prehospital phase of infarction. Second, lives saved/1000 patients, considered in conjunction with delays to coming under care, may provide an alternative to case fatality as a possibly more robust indicator of patient behaviour and ambulance and hospital performance.
- Brighton—Coordinating centre, Royal Sussex County Hospital: RM Norris (honorary consultant cardiologist, study director), R Vincent (consultant cardiologist), Gaynor Dixon (research sister responsible for coordination and database management), Christopher Edwards (staff nurse)
- Eastbourne—Eastbourne District General Hospital: N Sulke (consultant cardiologist), Janet Large (sister).
- Hastings—Conquest Hospital: R Wray (consultant cardiologist), Michele Parks, Hayley Coxon (sisters).
- Canterbury—Kent and Canterbury Hospital: A Owen (consultant cardiologist), Janet Husk (audit officer).
This project could not have been performed without the enthusiasm and dedication of the research nurses. I am grateful for the expert secretarial assistance of Mrs Janet Stevens. Supported by a project grant from the British Heart Foundation and by grants from the PPP Medical Trust and the Brighton Heart Support Trust.