All Danish residents benefit from the tax-financed Danish healthcare system that provides free access to hospitals, general practitioners, and private practice specialists, as well as partial reimbursement for prescription medication expenses [6]. Danish residents receive at birth or upon immigration a unique Civil Personal Registration number that functions as a personal identifier across all health services in Denmark, thereby allowing individual-level linkage between Danish health registries [6]. Danish medical databases, i.e., registries and databases that contain health-related data, are broadly categorised as administrative databases, health databases, and clinical quality databases [6]. The content of each category is described in detail elsewhere [6]. Briefly, administrative databases register people according to geographic area, health insurance program, or attendance at specific hospitals or clinics; health databases register people for surveillance and research; and clinical quality databases register people for clinical quality control [6]. Generally, clinical quality databases [7], such as the Danish Heart Failure Registry [8, 9], contain detailed and valid information due to active enrollment and data collection by clinicians. In contrast, information in administrative databases, such as the DNPR [5], is more dubious due to automated inclusion and data collection. Thus, ongoing validation of the DNPR, as done in this study, is essential to assess its applicability for population-based research.

The DNPR is an administrative database with the primary aim of providing a tool for health care planning for the Danish Health and Medicines Authority by monitoring hospital and health service use [5]. However, the registry is also used for other purposes such as for population-based research [5]. The DNPR contains information on all non-psychiatric inpatient contacts in Denmark back to 1977 and all inpatient psychiatric, outpatient clinic, and emergency room contacts back to 1995 [5]. Information on diagnoses, treatments, and procedures from such contacts is routinely reported from all Danish hospitals to the DNPR [5]. Diagnoses are recorded as one primary and optional secondary diagnoses according to the International Classification of Diseases 8^th edition before 1994 and 10^th edition after 1994 [5]. The diagnosis code is registered at the time of hospital discharge or immediately after an outpatient clinic contact by the discharging or treating physician [5]. Treatments and procedures are recorded according to the Danish Classification of Surgical Procedures and Therapies before 1996 and the Nordic Medico-statistical Committee’s Classification of Surgical Procedures after 1996 [5]. The treatment or procedure code is registered immediately after the treatment or procedure by the treating physician [5].

From the DNPR, we randomly sampled patients registered at a Danish University Hospital (Aarhus University Hospital) from 1 January 2017 to 31 December 2022. First, we sampled 200 patients with a first-time in or outpatient, primary or secondary diagnosis code for HF. The 200 patients were selected by sampling 50 patients from each of the following categories: inpatient primary diagnosis, inpatient secondary diagnosis, outpatient primary diagnosis, and outpatient secondary diagnosis. Second, we aimed to sample at least 50 patients with a procedure code for each of eight advanced HF treatments. However, because of the rarity of some treatments, achieving a sample size of 50 for each treatment was not feasible. Consequently, we sampled patients with a procedure code for ICD (n = 100), CRT-pacemaker (CRT-P) (n = 50), CRT-defibrillator (CRT-D) (n = 50), left ventricular assist device (LVAD) (n = 20), intra-aortic balloon pump (IABP) (n = 14), impella (n = 50), cardiopulmonary support (CPS) (n = 50), and heart transplantation (HTx) (n = 56). The sampling of the procedures was conducted exclusively based on the specific procedure codes without the requirement of an HF diagnosis during the same admission. Third, we aimed to sample at least 50 patients with an inpatient, primary diagnosis code for HF during an admission after each of the treatments. Because of a limited sample size, we were only able to identify patients with an HF diagnosis code after ICD (n = 71) and CRT (n = 62). Supplementary Table 1 presents all codes used for the validation. The codes used for defining HF diagnoses were aligned with a previous validation [10]. The codes used for defining the advanced HF treatments were based on clinical practice.

Two adjudicators (KB and CTW) independently compared data on HF diagnoses and advanced HF treatments from the DNPR with the patient’s medical record review as the reference. HF was confirmed if the medical record indicated signs of HF (e.g., reduced left ventricular ejection fraction, clinical signs of congestion, or biochemical markers of HF). Advanced HF treatments were confirmed if the medical record described the specific treatment. The adjudicators first reviewed the discharge summaries (for the diagnosis) or procedure descriptions (for the procedures) and when still unclear the full medical record. Disagreements between the adjudicators were resolved through consensus.

As a measure of validity, we calculated the positive predictive value (PPV) as the number of diagnoses or treatments retrieved from the DNPR confirmed by the medical record review. We used the Wilson Score method to calculate 95% confidence intervals (CIs) [11]. We calculated percent agreement and kappa statistics to estimate the adjudicators’ agreement [12]. We stratified all analyses by sex and age (≤59 years, 60–79 years, or ≥80 years). Comparisons between strata were done using Fisher’s exact test [13]. Lastly, we calculated the PPV for the overall HF diagnosis code (DI50). All analyses were performed using Stata statistical software version 17 (Stata Corporation, College Station, Texas, U.S.).

Our observed PPVs for first-time HF align with those reported in previous studies validating HF within the DNPR. Two studies comparing first-time HF registration with the patients’ medical records reported PPVs of 76% [10] and 100% [14]. Another two studies comparing HF registration during any admission with the patients’ medical records both reported PPVs of 84% [15, 16]. A study comparing HF registration with a clinical evaluation based on the European Society of Cardiology guidelines reported a PPV of 81% [1, 17].

Our finding of no notable differences in the validity of first-time HF based on whether primary or secondary diagnosis codes were used corresponds with a previous Danish validation study [10]. In that study, the PPV for first-time HF was 79% (95% CI: 64–89%) when primary diagnoses were used and 73% (95% CI: 60–83%) when secondary diagnoses were used [10]. The primary reason for misclassification of first-time HF cases was attributed to other cardiac diseases (e.g., atrial fibrillation, atrioventricular block, or valvular heart disease) or related procedures (e.g., implantation of or elective follow-up after ICD or pacemaker).

We found perfect validity of HTx within the DNPR, which is consistent with previous findings [18]. For the other examined advanced HF treatments, our study represents the first validation of their registration. The high validity observed for these treatments, except for IABP and impella, aligns with that reported for other cardiac procedures [18]. The primary reason for the low validity of IABP and impella was misclassification of other advanced HF treatments

We are the first to validate HF registration during admissions after ICD or CRT procedures. The observed low validity in these cases was attributed to the elective management of patients with ICD or CRT. This misclassification may stem from discharging physicians assigning HF as a discharge code, based on the patients’ existing advanced HF condition, even when HF exacerbation was not the primary reason for the current admission.

Denmark holds numerous clinical quality databases covering various disease specialties [7]. Among these databases is the Danish Heart Failure Registry [8, 9]. Patients are actively enrolled in this registry by cardiologists according to the European Society of Cardiology’s definition of HF [1, 9]. The registry’s content is registered during routine clinical care [9] and includes clinical procedures (e.g., echocardiography), medical treatments (e.g., use of angiotensin-converting enzyme inhibitors and beta blockers), and modifiable risk factors (e.g., hypertension status, smoking history, and alcohol intake) [8, 9]. The data in the Danish Heart Failure Registry is considered valid but may have lower completeness than the DNPR and also solely registers first-time HF, limiting its utility for studying HF readmissions [9].

Comparing the coding quality of HF between countries is challenged by differences in healthcare systems and coding standards. Such differences may be smaller among Nordic countries owing to somewhat similar healthcare systems. The PPV for definite HF has been reported as 62% in the Swedish Patient Registry [19, 20], 64% in the Norwegian Patient Registry [21, 22], and 85% in the Finnish Cardiovascular Disease Registry [23, 24]. In Norway, the PPV was higher for primary versus secondary diagnoses (76% versus 57%) and for inpatient versus outpatient diagnoses (66% versus 44%) [21]. The PPV for probable HF has been reported as 32% in the Swedish [19, 20] and 12% in the Norwegian patient registries [21, 22]. In Sweden, 64% of cases classified as probable lacked echocardiographic examination, which prevented them from being classified as definite [19]. Thus, our study’s higher PPV compared with Sweden [19] and Norway [21] could be due to some probable cases actually having HF but missing necessary information for being classified as definite. Unlike the Norwegian study [21], we found no differences in PPV between primary and secondary diagnoses or between inpatient and outpatient diagnoses. To our knowledge, HF readmission after ICD or CRT and the examined HF treatments have not been validated in other Nordic healthcare systems.

Caution should be exercised when applying our findings to healthcare systems that differ from the Danish or earlier study periods, as registration practices may vary across settings and over time. Additionally, our study was restricted to a single university hospital, potentially limiting generalisability to regional hospitals. However, the advanced HF treatments investigated are exclusively conducted at university hospitals, and Denmark is relatively homogenous in terms of demographics, socioeconomic factors, healthcare utilisation, and medication use [25]. Thus, our findings on the validity of advanced HF treatments likely generalise to other Danish regions. Concerning HF diagnoses, registration practices may differ between university and regional hospitals. However, given the consistency of our findings with previous validation studies that sampled patients from various hospitals across Denmark [10, 14, 15, 16, 17, 26], this limitation is likely of minor concern. Also, the results of the validations applied to the chosen codes and cannot be extrapolated to other codes or diseases. Furthermore, the limited sample size prevented us from estimating the validity of HF admissions after LVAD, IABP, impella, CPS, and HTx. Lastly, the negative predictive value could not be estimated as we sampled patients with and not without the specific diagnoses and procedure codes, and estimating sensitivity and specificity within the DNPR is unfeasible as this would require complete information on whether the specific diagnoses (or treatments) are present (or have been performed) for all Danish citizens (≈6,000,000 people) [27].