The study has a cross-sectional design.
In the Netherlands 30 out of the 91 hospitals offer PCI, of which 16 also provide CABG surgery.
The three guideline recommendations monitored in the present study were identified from the European Society of Cardiology guidelines by an expert panel consisting of cardiologists, an emergency department medical resident, an intensive care/cardiac care nurse and health care scientists. Adherence to these three recommendations is measured over 2012, the last year of a national quality improvement program. The program aims to decrease in-hospital mortality caused by ten high-risk patient safety threats , including ACS.
Selection of hospitals
The study is being conducted in 13 hospitals, selected by means of a multi-stage random sampling procedure. Initially six PCI-capable and six non-PCI-capable hospitals with a cardiology department were randomly selected from a pool of 40 randomly selected hospitals. Three PCI-capable hospitals declined participation, for which three additional PCI-capable hospitals were selected. Because the number of STEMI patients was relatively small, an additional PCI-capable hospital was selected. The hospitals are located in 7 of the 12 Dutch provinces, with bed capacities ranging between 200 and 1200 beds (Table 1).
The data are collected by means of retrospective chart review of electronic and/or paper-based medical, nursing and catheterisation laboratory charts of patients discharged between January 1st and December 31st 2012. Monthly, potential study charts are selected from the hospital billing system using diagnosis-treatment combination codes. Charts of patients discharged with a confirmed diagnosis of ACS (indicated in the discharge letter) are considered for inclusion (Fig. 1). When the discharge diagnosis is unclear, the chart is discussed with a cardiologist or other attending physician working in the field of cardiology. Charts of patients without a discharge diagnosis of ACS, a secondary ACS (e.g. due to anaemia), elective procedures, missing or uninformative charts, and charts of patients under the age of 18 years are excluded from the study. Moreover, additional exclusion criteria were defined for each process indicator separately. For timely invasive treatment, charts of STEMI patients not going for acute PCI are excluded. For use of risk scoring instruments, charts of patients transferred from another hospital are excluded. For discharge medication, charts of patients who were transferred to another hospital, patients who died during their admission or received palliative treatment are excluded.
If the monthly number of charts exceeds the screening capacity, screening of the charts is performed in chronological order of discharge representing STEMI and NSTEMI/UA equally, and terminated when the chart abstractors are practically unable to screen additional charts.
In two hospitals the pre-selection procedure based on the hospital billing system is not possible. Therefore in one hospital the pre-selection of charts is performed by requesting all charts of patients with a suspected or confirmed diagnosis of ACS through the cardiology department’s secretariat. In the other hospital, local hospital regulations require that patients with a suspected ACS are informed about the study and asked to give informed consent before their chart can be considered for inclusion. Due to the declined invitations and deviations in inclusion procedures, data collection in 5 hospitals will comprise 9 or 10 months instead of 12 months.
The study has three main outcome measures. First, the percentage of STEMI patients in which the PCI procedure started within 90 minutes from first (para)medical contact. Second, the percentage of NSTEMI or UA patients where use of a validated risk scoring instrument was documented. Finally, the percentage of ACS patients with a prescription of the recommended discharge medication, documentation of a contraindication or other reason for not receiving the recommended medication. Additionally, patient and hospital characteristics associated with guideline adherence will be identified.
From all charts, the following information is abstracted: demographic and clinical information including gender, age, cardiac history, risk factors, biomarker values, electrocardiogram findings, resuscitation, heart failure, cardiogenic shock on arrival and month of discharge (Table 2).
In addition, for the timely invasive treatment indicator, the following variables are recorded: routing of the patient, type of first (para)medical contact, place of first electrocardiogram, type of treatment, and the dates and times of first (para)medical contact, first (ambulance/general practitioner) electrocardiogram and sheath insertion (start of PCI) (Table 3).
To evaluate cardiac risk score adherence, application of a validated risk scoring instrument (e.g. GRACE [18, 19], TIMI , FRISC , HEART  and PURSUIT ), type of instrument, risk score outcome, date of application, and type of treatment are recorded (Table 4).
Finally, for discharge medication, prescription of acetylsalicylic acid, thienopyridine, statin, beta blocker and angiotensin-converting enzyme (ACE) inhibitor and contraindications or other reasons for not prescribing all or some of the medication are recorded (Table 2). Contraindications were derived from an annually updated database containing information about all medication registered in the Netherlands .
Abstraction of data
All data are collected on standard case report forms. Variables are defined in codebooks. Two researchers (JT & JE) developed the codebooks and case report forms based on the European Society of Cardiology guidelines. The case report forms were discussed within the research group, tested in two pilot measurements and adjusted accordingly. The data are collected by six chart abstractors who were introduced to the subject of ACS and instructed in the chart review procedures by JT and JE. Chart reviews were supervised until the quality of the chart reviews was satisfactory. The data are entered into a database using a data entry program with fixed entry fields (BLAISE version 4.7, Statistics Netherlands) and compared with the original case report form by a second researcher.
To ensure reliability of the data and to assess the quality of the codebook, a sample of charts (5–10 %) is independently screened again by one of the five other chart abstractors. The two case report forms are compared, and differences are discussed until consensus is reached. If necessary, changes are made in the original case report form. The reliability between the chart abstractors will be calculated by means of the percentage of agreement for each variable.
Missing data patterns will be analysed by means of missing value analyses. Depending on the pattern , missing values will be imputed by means of a single imputation (missing completely at random) or multiple imputation procedure (missing at random) .
The degree of adherence to the three process indicators will be presented by descriptive statistics. Associations of patient and hospital characteristics with guideline adherence are studied in separate analyses.
Timely invasive treatment
The time to PCI in minutes will be entered as a continuous dependent variable in a generalised linear model taking into account its distribution, as time variables are generally not normally distributed. In univariate analyses, associations of the independent variables, i.e. patient and admission characteristics, are studied. To account for clustering of patient data within hospitals, the variable ‘hospital’ and its significant interactions with any other of the predictor variables will be entered as a covariate in all univariate models . This is because the hospital sample size (7 PCI-capable hospitals) is considered small for multilevel regression analysis . All variables and interactions significantly (p ≤ 0.05) associated with the time to PCI will be included in the multiple generalised linear model. Furthermore, to minimise the probability of making a type II error, all non-significant variables from the univariate models will be added to the multiple generalised linear model one by one. Significant variables (p ≤ 0.05) will be added to the final model.
Use of risk scoring instruments
Associations of independent variables with the use of cardiac risk scoring instruments will be studied by means of a generalised linear mixed model (GLMM). In the analysis the binary dependent variable will be the use of a validated risk score instrument. Independent variables will be patient characteristics, hospital characteristics and month of discharge. To account for clustering of the data, the model will comprise random effects for hospitals. First, independent variables will be tested separately correcting for the random hospital effects. Second, all independent variables with a significance level below p ≤ 0.15 will be selected. Next, pairs of selected independent variables will be tested jointly.
Last, all significant (p ≤ 0.05) variables from the previous steps will be included in the final multivariable model. This final step also comprises a cautious consideration of significant (p ≤ 0.05) interaction terms.
Associations of independent variables with the prescription of the recommended discharge medication will be studied by means of GLMM. In these analyses, prescription of the five guideline-recommended medicines or documentation of contraindications (yes/no) will be the binary dependent variable. The effects of the independent variables including patient, hospital and discharge characteristics will be tested in univariate analyses. All variables with a significant association (p ≤ 0.05) with the dependent variable will be included in a multivariable model. To account for the effects of collinearity, all variables not significantly related to prescription of the recommended discharge medication in the univariate models will be added to the multivariable generalised linear mixed model one by one. Interactions will be tested and added to the multivariable model in case of a significant effect. In all models, hospital will be entered as a random effect variable to account for clustering of the data. As not all medicines are indicated for all patients with ACS according to the European Society of Cardiology guidelines (e.g. ACE-inhibitors are recommended for all patients with ACS, but only indicated for those patients with a reduced cardiac function), additional models will be created to analyse the effects of patient and hospital characteristics on the prescription of ≤3 and ≥4 medicines or documentation of a contraindication.
The data will be analysed in IBM SPSS Statistics (version 20 for Windows) and R (version 3.0.0 for Windows).