Nationwide claims data validated for quality assessments in acute myocardial infarction in the Netherlands

To assess validity of national claims data, four representative hospitals are assessed by comparing claims data of each hospital (‘national registry’) to local data, which is obtained by local patient record reviews (‘validation registry’).

Medication use was defined by the registered anatomical therapeutic chemical (ATC) code and the defined daily dose (DDD). The DDD is the assumed average maintenance dose per day for a drug used for its main indication in adults, composed by the World Health Organisation Collaborating Centre for Drug Statistics Methodology [15]. It is a unit of measurement and does not necessarily reflect the recommended or prescribed daily dose. The complete list of the DDD per medication is shown in Table 1. The use of pharmacy records has been validated in previous studies [16].

We assessed baseline data and pharmacological treatment after one-year follow-up of all patients with AMI in the Netherlands in 2012 and 2013.

In every hospital, the patient population was selected by using the hospital’s local claims registration database to find all admitted patients with similar national diagnosis codes for STEMI and NSTEMI. In patients with multiple diagnosis codes within the study period, the first admission code was used for the analysis. All codes were checked for the final diagnosis at discharge in which STEMI is defined as ST-elevation in ≥2 contiguous leads with ≥0.1 mV in all leads other than leads V2–V3 and a typical rise and/or fall of cardiac biomarkers [17]. NSTEMI diagnosis is defined by clinical symptoms suggestive for acute ischaemia, with or without ST-segment depression or T‑wave and presence of cardiac biomarkers [18].

In the validation hospitals, all patient records were screened to assess where follow-up was performed. Patients whose follow-up was performed in another hospital were excluded from follow-up analyses. Additionally, in the non-PCI centre, patients could have been referred from another hospital where the initial AMI was registered, in which case these patients were included for follow-up analysis. Finally, all patients deceased within one year after their first admission were excluded from the follow-up population, similar to the national database, since one-year medication use could not be assessed.

All remaining patient records were reviewed for medication use.

Of all 59,534 admissions for AMI in the Netherlands in 2012 and 2013, 3,980 (7%) patients were registered in one of the four validation hospitals according to the claim database. Half of the diagnosis codes were for STEMI (51%) (95% CI 49–53%), 68% (95% CI 67–70%) was treated with PCI, mean age was 66 ± 13 years and 68% (95% CI 67–70%) were male.

With patient records reviews in the four hospitals, 4,014 patients were identified. Similar to the national registry, 51% of the patients had a STEMI code (95% CI 49–53%), 69% (95% CI 67–70%) was treated with PCI, mean age was 66 ± 13 and 69% (95% CI 67–70%) were male (Table 2). For the direct validation of the diagnosis codes, the final diagnosis at discharge was correlated with the first diagnosis coding. Due to transfers after treatment, a reliable judgement for final diagnosis was not possible in 254 (6%) patients. Of all remaining 1,977 STEMI codes, 88% was coded correctly as STEMI. Of all 1,738 NSTEMI codes, 93% was coded correctly as NSTEMI.

To validate the national algorithm of the location where follow-up had taken place, all patients with outpatient follow-up in another centre were excluded. In the national registry, 3,106 patients had their first follow-up code after their AMI was registered in one of the four selected hospitals. Three hundred and thirty patients died within one year and were excluded, resulting in a follow-up population of 2,776 patients.

In the validation registry of 4,014 patients, 1,157 patients received follow-up in another hospital, 146 patients were followed in one of the hospitals although the first admission of AMI was registered elsewhere, and 302 patients died during the first year, resulting in a validated follow-up population of 2,701 patients.

Analysis of 2,776 patients in the national registry showed aspirin use of 81% (95% CI 79–82%), P2Y12 inhibitor use of 80% (95% CI 78–81%), statin use of 91% (95% CI 90–92%), beta-blocker use of 86% (95% CI 85–87%), ACE inhibitor/AT2 antagonist use of 80% (95% CI 78–81%), VKA use of 17% (95% CI 15–18%) and NOAC use of 1% (95% CI 0–1%) (Table 3).

In the validation registry, one-year medication data were available in 1,986 patients, showing comparable rates between the two registries (Table 3).

Of all 59,534 patients (mean age 67 ± 13 years, 66% male), 6,862 (12%) patients died during one-year follow-up (Table 2). Analysis of the remaining 52,672 patients showed aspirin use of 81%, P2Y12 inhibitor use of 76%, statin use of 85%, beta-blocker use of 82%, ACE inhibitor/AT2 antagonist use of 73% (Table 3).

With the growing importance of quality assessment and efficient data collection, routinely collected claims data are being used and studied more frequently for cardiac outcome measurements. Benefits of claims databases are the automatic and continuous data acquisition which make it less laborious for health care professionals to gather health care outcome measurements. Claims data have the advantage of nationwide coverage, depend less on voluntary hospital participation and can be used for chain of care evaluation, including a connection with external datasets. However, claims data are collected predominantly for billing purposes, not for research. Therefore, it requires validation of data accuracy. Data accuracy is particularly important when using claims data as a quality indicator and when comparing medical centres. As previously shown in the United Kingdom, the current study demonstrates a sufficient accuracy of nationwide-used diagnostic, procedural and medical treatment codes with chart review [19]. Other studies on comparisons of claims data with clinical registries show various results [20‐22]. This is due to differences in medical specialisation, study purpose and database quality. This analysis composes a large sample size (7% of all Dutch AMI patients) and data quality is verified in various stages. Altogether, the use of these nationwide claims data offer an efficient approach in quality measurement on a national scale. However, claims data contain limited detailed clinical information, which must be taken into account. Adjustment for patient characteristics and case mix is possible up to a certain level, but requires additional evaluation [22, 23].

Several large trials have shown the beneficial effects of medication on patient outcome [2‐7]. Therefore, medication poses a robust indicator for quality of care following AMI [24].

Previous studies focussing on medication at discharge show a substantial guideline adherence in patients with AMI (65–69% were prescribed all five indicated drugs) [25, 26]. Additionally, a recent European guideline evaluation by the EUROASPIRE IV Study Group showed comparable secondary prevention rates at a median follow-up time of 16 months after myocardial infarction: 94% antiplatelets, 86% statins, 83% beta-blockers and 75% ACE inhibitors/AT2 antagonists [27]. Another study from the Netherlands evaluated secondary prevention during long-term follow-up with usage of a similar patient linkage system containing drug-dispensing records from community pharmacies and hospitalised patients and showed an improvement in medication compliance over time (from 1999 till 2007) [28]. When comparing these results with the current study, an additional improvement in medication adherence over time can be observed, suggesting an increased focus on adequate medication prescription after AMI. Approximately half of the currently studied patients do not receive all five indicated drugs one year after AMI. The measured suboptimal adherence of all five indicated drugs is partly physician driven (the drugs are not prescribed, whether or not on purpose) and can be partly patient driven (the patient does not want the medication or is intolerant) [4, 18, 29]. In the guidelines of the European Society of Cardiology, American Heart Association and American College of Cardiology, beta-blocker and ACE inhibitor/AT2 antagonist use is indicated for patients with left ventricular dysfunction (Class 1, level A) and recommended if not contra-indicated in the remaining group [4, 18, 29]. A combined measurement of the five indicated drugs might have its limitations as a quality indicator and therefore requests careful interpretation. Comparison per type of medication poses a more robust indicator for the assessment of quality of care.

National claims data offer a range of possibilities for automated and nationwide quality assessments without additional work for professionals, resulting in more time for actual patient care. Furthermore, national claims data serve as an opportunity in the evaluation of the patient’s chain of care. Future research should focus on the development of additional quality indicators and on the connection between currently available databases, such as primary care databases to assess primary care follow-up, or nursery care databases to assess functional status. In the Netherlands, claims data serve as an additional measurement, alongside the National Heart Registry (NHR) [30]. The NHR could function as a national clinical database with more detailed clinical information. National claims data could be used to check the number in the clinical databases. In the long run, when the field of cardiac professionals is ready, cost effective analysis is possible. A development that can already be seen in the field of surgery, with the Dutch Institute for Clinical Auditing (DICA) and the registration of claims data [31]. All in all, this depends on new laws concerning privacy and the will of politicians, hospitals and clinical professionals.

Some limitations should be addressed. Firstly, differences in adherence between local and national databases may also be caused by different adherence measuring methods; national databases are based on pharmacy fulfilment prescription and local databases are based on self-reporting of the patient, deducted from a chart review. This discrepancy between providing and fulfilment of the prescription is not studied and therefore a small difference between the two registries is in accordance with clinical practice and doesn’t relate to data quality. Secondly, patients were lost to follow-up in the validation registry. Reasons were a physician- or patient-driven conservative approach because of the patient’s terminal disease or high age. Patients who were lost to follow-up may be the patients that used less medication and therefore could probably cause a selection bias of the validation registry.