Back to Journals » Clinical Epidemiology » Volume 8
Evaluation of algorithms for registry-based detection of acute myocardial infarction following percutaneous coronary intervention
Authors Egholm G, Madsen M , Thim T , Schmidt M , Christiansen EH, Bøtker HE , Maeng M
Received 21 March 2016
Accepted for publication 15 April 2016
Published 19 October 2016 Volume 2016:8 Pages 415—423
DOI https://doi.org/10.2147/CLEP.S108906
Checked for plagiarism Yes
Review by Single anonymous peer review
Peer reviewer comments 2
Editor who approved publication: Professor Henrik Sørensen
Gro Egholm,1,2,* Morten Madsen,2,* Troels Thim,1 Morten Schmidt,2,3 Evald Høj Christiansen,1 Hans Erik Bøtker,1 Michael Maeng1
1Department of Cardiology, 2Department of Clinical Epidemiology, Aarhus University Hospital, Aarhus, 3Department of Internal Medicine, Regional Hospital of Randers, Denmark
*These authors contributed equally to this work
Background: Registry-based monitoring of the safety and efficacy of interventions in patients with ischemic heart disease requires validated algorithms.
Objective: We aimed to evaluate algorithms to identify acute myocardial infarction (AMI) in the Danish National Patient Registry following percutaneous coronary intervention (PCI).
Methods: Patients enrolled in clinical drug-eluting stent studies at the Department of Cardiology, Aarhus University Hospital, Denmark, from January 2006 to August 2012 were included. These patients were evaluated for ischemic events, including AMI, during follow-up using an end point committee adjudication of AMI as reference standard.
Results: Of 5,719 included patients, 285 patients suffered AMI within a mean follow-up time of 3 years after stent implantation. An AMI discharge diagnosis (primary or secondary) from any acute or elective admission had a sensitivity of 95%, a specificity of 93%, and a positive predictive value of 42%. Restriction to acute admissions decreased the sensitivity to 94% but increased the specificity to 98% and the positive predictive value to 73%. Further restriction to include only AMI as primary diagnosis from acute admissions decreased the sensitivity further to 82%, but increased the specificity to 99% and the positive predictive value to 81%. Restriction to patients admitted to hospitals with a coronary angiography catheterization laboratory increased the positive predictive value to 87%.
Conclusion: Algorithms utilizing additional information from the Danish National Patient Registry yield different sensitivities, specificities, and predictive values in registry-based detection of AMI following PCI. We were able to identify AMI following PCI with moderate-to-high validity. However, the choice of algorithm will depend on the specific study purpose.
Keywords: Danish National Patient Registry, registry, percutaneous coronary intervention, validity, sensitivity, specificity
Introduction
First-time ischemic events such as acute myocardial infarction (AMI) are used to study the risk and to improve the prognosis of ischemic heart disease. AMI is often treated with percutaneous coronary intervention (PCI). To monitor the safety and efficacy of this intervention, robust registry-based algorithms are required for the detection of AMI in this population.1
In Denmark, record linkage using the ten-digit civil registration number offers unique possibilities for epidemiological studies.2 As the key registry, the Danish National Patient Registry contains data on all Danish hospital admissions and outpatient clinic visits, starting in 1997.3 Thereby, the Danish National Patient Registry can be utilized for the detection of AMI in the Danish population. However, to what extent the Danish National Patient Registry can be used to identify AMI in patients with existing ischemic heart disease undergoing PCI is unknown.3 In this study, we aimed to create an algorithm for using the Danish National Patient Registry to identify patients with AMI following PCI.
Methods
Study design, setting, and participants
We performed the evaluation in a population of patients treated with drug-eluting coronary stents as a part of clinical drug-eluting coronary stent studies. These patients were enrolled in the Central Region of Denmark, which covers a population of ~1.3 million inhabitants corresponding to 23% of the Danish population. The patients were treated with PCI at the Department of Cardiology, Aarhus University Hospital, Denmark, from January 2006 to August 2012.4–7 Using this cohort with end-point committee adjudication of AMI as reference standard, we compared different algorithms for the detection of AMI in the Danish National Patient Registry following PCI.
Definition of AMI
Clinical end-point committee adjudication of AMI was performed in each trial as previously described.4–7 Briefly, possible AMI events were screened using the Danish National Patient Registry3 and the Western Denmark Heart Registry.8 Possible events were subsequently reviewed by a clinical end-point committee, with reference to the contemporary universal definitions of AMI.9 The end-point committee also reviewed all deaths in order to classify these as cardiac or noncardiac. In case of cardiac death, the end-point committee evaluated whether it was secondary to AMI.
The Danish National Patient Registry
The Danish National Patient Registry contains information on all nonpsychiatric hospital admissions since 1977 and emergency room and outpatient clinic visits since 1995.3 The registry contains data from each admission including the admission and discharge dates, admission type, discharge diagnoses, and procedures performed during the admission.3 The International Classification of Diseases tenth revision (ICD-10) codes have been used since 1994. All admissions have one primary discharge diagnosis reflecting the primary reason for the admission. Additionally, admissions may have one or more secondary discharge diagnoses reflecting coexisting conditions. Discharge diagnoses are determined exclusively by the discharging physician.
The Danish national health care service is tax supported and provides free health care. Mandatory reporting to the Danish National Patient Registry, which is managed by the Danish Health Authority, ensures nationwide coverage of AMI hospitalisations.3
Algorithms for detection of AMI in the Danish National Patient Registry
To establish an algorithm for the detection of AMI in the Danish National Patient Registry, we identified AMI from discharge diagnoses using the ICD-10 code I21. Diagnoses were identified as primary (only) and primary or secondary discharge diagnoses. Furthermore, algorithms were based on patient contact type (inpatient admission), admission type (acute or elective), and hospital type (with or without coronary angiography capability). Table 1 shows the details of the different algorithms.
Statistical analyses
Follow-up of the trial participants started upon discharge after drug-eluting stent implantation.4–7 Patients were followed until a first AMI was detected in the Danish National Patient Registry, by the end-point committee, or in both simultaneously.
For each algorithm for identifying AMI following PCI in the Danish National Patient Registry, we calculated sensitivity, specificity, and predictive values using the end-point committee adjudicated cases of AMI as reference. We stratified the results according to AMI status at the time of PCI (AMI before PCI, AMI at same date of PCI, or PCI without prior AMI) to determine whether recurrent AMI could be detected equally well as first-time AMI. We also stratified according to sex, age (≤65 years vs >65 years), indication for PCI (acute coronary syndrome vs stable angina pectoris), and time from index procedure to AMI. Confidence intervals were calculated with Jeffrey’s method.10
All statistical analyses were performed using SAS software Version 9.4 (SAS Institute Inc., Cary, NC, USA). The study was approved by the Danish Data Protection Agency (Ref no 2012-41-0164) and the Danish Health Authority (Ref no 6-8011-270/2). Registry studies do not require ethical committee approval or patients consent in Denmark.
Results
We evaluated 5,719 patients with a mean follow-up time of 3 years. Of these, 285 had an end-point committee adjudicated AMI. Baseline characteristics of the PCI cohort are presented in Table 2.
The results from different algorithms are reported in Table 3 and Figure 1. Since patients with a detected AMI, either by the algorithm or by the end-point committee, were censored from the time of AMI detection, the number of patients with AMI and the average follow-up period vary between algorithm evaluations. Two-way tables for each algorithm evaluation are provided in Tables S1–S6.
The algorithms with the best performance were the combination of AMI as primary (algorithm D) or primary or secondary (C) discharge diagnosis combined with acute admission. A broader algorithm (A) combining AMI as primary or secondary discharge diagnosis and all inpatients, instead of acute admissions, improved the sensitivity (95%), but decreased the positive predictive value considerably (42%). Restricting the algorithm to admissions at a hospital with coronary angiography capability increased the positive predictive value. However, these narrower algorithms all had a decreased sensitivity (Table 3, Figure 1).
Evaluation of a broad algorithm of AMI diagnosis (code I21) as either primary or secondary diagnosis and inpatient (algorithm A, Table 3) showed that 13 patients with a validated AMI were not detected (Table S1). These AMIs resulted in cardiac arrest (n=6) and were recorded as such with the corresponding ICD-10 code in the Danish National Patient Registry. For the remaining patients, the discharge diagnosis codes covered various ICD-10 codes for ischemic heart disease, examination for angina, and examination for acute coronary syndrome.
Evaluation of a narrow algorithm of AMI diagnosis (code I21) as both primary or secondary diagnosis and acute admission (algorithm C, Table 2) showed that 95 patients were recorded with AMI diagnoses in the Danish National Patient Registry without having an end-point committee adjudicated AMI (Table S3). The majority of these were patients admitted for examination for angina or examination for acute coronary syndrome.
The stratified analyses of algorithm C are reported in Table 4, with corresponding two-way tables provided in Tables S7–S23. Sex and age had no major impact on the parameters. Among patients with acute coronary syndrome, positive predictive value was lower than among patients with stable angina pectoris. Time from index procedure to AMI seemed to influence positive predictive values, which were lowest within the first 30 days after discharge following PCI and improved thereafter.
Table 4 Performance of algorithm C (all acute admissions with acute myocardial infarction as primary or secondary discharge diagnosis) across subgroups Abbreviation: CI, confidence interval |
Discussion
We found that different algorithms yielded different sensitivities, specificities, and predictive values to detect AMI in the Danish National Patient Registry. The choice of algorithm will depend on the specific study purpose. However, combining the discharge diagnosis of AMI (I21) and acute admission yielded a better positive predictive value for patients with prior PCI than use of a discharge diagnosis of AMI alone.
Apart from the diagnosis AMI, our algorithms relied on the variable “acute admission”, which has been shown to have a high validity in the Danish National Patient Registry.11 Previously, the validity of AMI diagnoses in the general population, as registered in the Danish National Patient Registry, has been validated using medical records,12,13 discharge summaries,14,15 or a clinical registry.16 We recorded a lower positive predictive value of first-time AMI in the Danish National Patient Registry than in these earlier studies.12–14 This was expected as our study population consisted of patients with established ischemic heart disease undergoing PCI. These patients are therefore more likely to be given a later discharge diagnosis of AMI, ie, to be misclassified due to their prior medical history. Similar misclassification has also been previously shown for other conditions, eg, venous thromboembolism.16 In agreement with this interpretation, we found a lower positive predictive value of the algorithm among patients with AMI during the index admission or with acute coronary syndrome as indication for stent implantation and within the first 30 days after stent implantation as compared to later.
The choice of algorithm will depend on the specific study purpose. For example, in registry-based randomized clinical trials with end-point adjudication by an end-point committee, it is important to detect as many of the potential events as possible. In this case, a broad algorithm, like algorithm A, seems the optimal choice. The low positive predictive value for this algorithm will be corrected by the end-point committee. In traditional randomized cohort studies relying on registry-based end points, ie, without adjudication by an end-point committee, algorithms C and D are preferable due to the combination of high sensitivity (although lower than algorithm A) and higher positive predictive values. Finally, in case–control studies, a high positive predictive value is preferred to correctly detect cases.
A small number of patients with adjudicated AMI did not have this diagnosis in the Danish National Patient Registry. One half of these patients died from cardiac arrest and were diagnosed with AMI by the end-point committee when the cause of death was reviewed. The other half had various ischemia-related diagnoses and were diagnosed by end-point committee review of all angiographies and coronary interventions during the study period. A composite end point of registry-based AMI and all-cause death, often used in registry-based studies, would thus include half of the missed AMIs, ie, only very few true events would be overlooked by a use of combined end point and thereby improve sensitivity.
Strengths and limitations
We were able to evaluate the described algorithms using a large study population with end-point committee-validated AMIs. In comparison with earlier studies, this gave us an opportunity to evaluate sensitivity and specificity of the algorithms and also the positive predictive values in a subgroup of patients undergoing PCI. Thus, this study included patients treated with drug-eluting coronary stents, and the reported sensitivities and specificities of the different algorithms may not extend to the general population, to patients with ischemic heart disease without stent implantation, or to patients without previous ischemic heart disease.
Conclusion
Different algorithms utilizing additional information from the Danish National Patient Registry yielded different sensitivities, specificities, and predictive values in registry-based detection of AMI following PCI. The choice of algorithm will depend on the specific study purpose. However, it was possible to identify algorithms for AMI detection following PCI in the Danish National Patient Registry with moderate-to-high validity.
Disclosure
The authors report no conflicts of interest in this work.
References
Thuesen L, Jensen LO, Tilsted HH, et al. Event detection using population-based health care databases in randomized clinical trials: a novel research tool in interventional cardiology. Clin Epidemiol. 2013;5:357–361. | ||
Schmidt M, Pedersen L, Sørensen HT. The Danish Civil Registration System as a tool in epidemiology. Eur J Epidemiol. 2014;29(8):541–549. | ||
Schmidt M, Schmidt SA, Sandegaard JL, Ehrenstein V, Pedersen L, Sørensen HT. The Danish National Patient Registry: a review of content, data quality, and research potential. Clin Epidemiol. 2015;7:449–490. | ||
Christiansen EH, Jensen LO, Thayssen P, et al; Scandinavian Organization for Randomized Trials with Clinical Outcome (SORT OUT) V investigators. Biolimus-eluting biodegradable polymer-coated stent versus durable polymer-coated sirolimus-eluting stent in unselected patients receiving percutaneous coronary intervention (SORT OUT V): a randomised non-inferiority trial. Lancet. 2013;381(9867):661–669. | ||
Jensen LO, Thayssen P, Maeng M, et al; Scandinavian Organization for Randomized Trials With Clinical Outcome SORT OUT IV Investigators. Three-year outcomes after revascularization with everolimus- and sirolimus-eluting stents from the SORT OUT IV trial. JACC Cardiovasc Interv. 2014;7(8):840–848. | ||
Maeng M, Tilsted HH, Jensen LO, et al. Differential clinical outcomes after 1 year versus 5 years in a randomised comparison of zotarolimus-eluting and sirolimus-eluting coronary stents (the SORT OUT III study): a multicentre, open-label, randomised superiority trial. Lancet. 2014;383(9934):2047–2056. | ||
Raungaard B, Jensen LO, Tilsted HH, et al; Scandinavian Organization for Randomized Trials with Clinical Outcome (SORT OUT). Zotarolimus-eluting durable-polymer-coated stent versus a biolimus-eluting biodegradable-polymer-coated stent in unselected patients undergoing percutaneous coronary intervention (SORT OUT VI): a randomised non-inferiority trial. Lancet. 2015;385(9977):1527–1535. | ||
Schmidt M, Maeng M, Jakobsen CJ, et al. Existing data sources for clinical epidemiology: the Western Denmark Heart Registry. Clin Epidemiol. 2010;2:137–144. | ||
Thygesen K, Alpert JS, Jaffe AS, et al; Writing Group on the Joint ESC/ACCF/AHA/WHF Task Force for the Universal Definition of Myocardial Infarction, ESC Committee for Practice Guidelines (CPG). Third universal definition of myocardial infarction. Eur Heart J. 2012;33(20):2551–2567. | ||
Brown LD, Cai TT, DasGupta A. Interval estimation for a binomial proportion. Stat Sci. 2001;16(2):101–117. | ||
Vest-Hansen B, Riis AH, Christiansen CF. Registration of acute medical hospital admissions in the Danish National Patient Registry: a validation study. Clin Epidemiol. 2013;5:129–133. | ||
Joensen AM, Jensen MK, Overvad K, et al. Predictive values of acute coronary syndrome discharge diagnoses differed in the Danish National Patient Registry. J Clin Epidemiol. 2009;62(2):188–194. | ||
Coloma PM, Valkhoff VE, Mazzaglia G, et al; EU-ADR Consortium. Identification of acute myocardial infarction from electronic healthcare records using different disease coding systems: a validation study in three European countries. BMJ Open. 2013;3(6):e002862. | ||
Madsen M, Balling H, Eriksen LS. The validity of the diagnosis of acute myocardial infarction in 2 registries: the Heart Registry compared to the National Patient Registry. Ugeskr Laeger. 1990;152(5):308–314. | ||
Thygesen SK, Christiansen CF, Christensen S, Lash TL, Sørensen HT. The predictive value of ICD-10 diagnostic coding used to assess Charlson comorbidity index conditions in the population-based Danish national registry of patients. BMC Med Res Methodol. 2011;11:83. | ||
Madsen M, Davidsen M, Rasmussen S, Abildstrom SZ, Osler M. The validity of the diagnosis of acute myocardial infarction in routine statistics: a comparison of mortality and hospital discharge data with the Danish MONICA Registry. J Clin Epidemiol. 2003;56(2):124–130. | ||
Schmidt M, Cannegieter SC, Johannesdottir SA, Dekkers OM, Horváth-Puhó E, Sørensen HT. Statin use and venous thromboembolism recurrence: a combined nationwide cohort and nested case-control study. J Thromb Haemost. 2014;12(8):1207–1215. |
Supplementary Material
© 2016 The Author(s). This work is published and licensed by Dove Medical Press Limited. The full terms of this license are available at https://www.dovepress.com/terms.php and incorporate the Creative Commons Attribution - Non Commercial (unported, v3.0) License. By accessing the work you hereby accept the Terms. Non-commercial uses of the work are permitted without any further permission from Dove Medical Press Limited, provided the work is properly attributed. For permission for commercial use of this work, please see paragraphs 4.2 and 5 of our Terms.