Classification of Inpatient and Outpatient Contacts in the Danish National Patient Register: Epidemiological Implications and Future Directions

Abstract

The Danish National Patient Register (DNPR) is an important data source for register-based health-related research, providing detailed information on all hospital contacts in Denmark. With the transition from the second version of the DNPR (DNPR2) to the third version (DNPR3) in early 2019, the variable distinguishing inpatient and outpatient contacts was discontinued, introducing substantial methodological challenges for epidemiological research. This perspective provides a narrative synthesis of current approaches to classifying hospital stays in DNPR3 and highlights key epidemiological considerations. Two main methodological approaches were identified: consensus-based frameworks, which emphasize interpretability and clinical reasoning and data-driven algorithms, which leverage patterns in register data to improve classification performance. Existing methods generally provide reasonable classification for inpatient and elective outpatient contacts. However, classification of acute outpatient visits remains a persistent and unresolved challenge, reflecting heterogeneous registration practices and the absence of a gold-standard reference in DNPR3. Further, current methodological approaches are fragmented and context-dependent, limiting comparability across studies and over time. We therefore call for coordinated efforts to develop transparent, validated, and consensus-driven classification frameworks. Such efforts are essential to ensure valid, reproducible, and interpretable epidemiological research using DNPR data.

Introduction

The Danish National Patient Register (DNPR) has served as a data source for epidemiological and healthcare research for more than four decades, supporting studies of disease incidence, time trends, healthcare utilization, treatment outcomes, and population-level health. Since its establishment in 1977 (Figure 1), the DNPR has contained administrative and clinical information on all somatic inpatients. Information on acute somatic outpatients (known as emergency department contacts until 2014) and elective somatic outpatients has been registered since 1995, as has information on psychiatric patients. Patients at private hospitals were included from 2002.^1–3 Since the early 2000s, the DNPR has achieved near-complete nationwide coverage, and, therefore, its variables play a critical role in ensuring valid and reproducible register-based research.

Figure 1.

Timeline for the content in the Danish National Patient Registry.

In 2019, the DNPR underwent a major structural transition with the introduction of DNPR3.⁴ While the implementation of DNPR3 improved certain aspects of data granularity, it disrupted established research practices and introduced new methodological challenges and possibilities. In DNPR2, two variables were used to determine the type of hospital contact: one distinguishing inpatients and outpatients and one distinguishing acute and elective contacts. In DNPR3, only the latter variable remains, while the explicit inpatient/outpatient indicator was eliminated. Of course, the distinction between inpatient and outpatient is not necessary in all register-based studies, for example, if the population is based on a certain diagnosis or if the outcome is any hospital contact, regardless of the type. However, this distinction is essential in register studies where the epidemiological design depends on differentiating inpatients and outpatients for defining study populations, constructing exposures and outcomes, monitoring disease trajectories and measuring trends in healthcare activity. For example, Winckler et al (2025) conducted a nationwide cohort study of children with heavy prenatal alcohol exposure to quantify their healthcare utilization patterns and defined inpatients as contacts lasting more than 5 hours while von Linstow et al (2024) conducted a nationwide study describing the healthcare resource utilization and costs related to respiratory syncytial virus hospitalization in infants less than 6 months of age and defined inpatients as contacts lasting more than 12 hours.^5,6 Inconsistent definitions like these are likely to bias estimates of disease burden and healthcare utilization.

Studies relying on historical definitions of types of hospital stays (in DNPR (Danish): patienttype) now face challenges in reproducing past methods or maintaining consistency across DNPR versions. Furthermore, several studies have examined the reliability of specific diagnosis codes for various medical conditions in the DNPR2 and consistently found PPVs to be higher after exclusion of acute outpatient contacts.^3,7 However, the omission of the variable “patienttype” in DNPR3 precludes such exclusions in a coherent and reliable manner.

Although guidelines exist for reporting types of hospital stays, such as defining inpatients as patients admitted to a ward and occupying a designated bed and outpatients as those registered at outpatient clinics, these definitions do not address patients admitted to wards without occupying a bed, and they are based on information which is not available in the registers. This limits their usefulness as a gold standard, particularly in the context of DNPR3.

Recently, the Danish Health Data Authority (Danish: Sundhedsdatastyrelsen) launched an algorithm to identify hospital stays, which is available to all researchers and implemented in a new table available via Statistics Denmark. The algorithm classifies hospital stays according to duration. Stays of 12 hours or more are considered inpatient admissions, while stays under 12 hours are considered outpatient contacts. Further, completed physical attendances are linked if they overlap in time or occur within four hours of each other, and the duration of a stay is then calculated from the earliest check-in to the latest check-out of the linked attendances. However, a simple duration-based rule for defining hospital stays is often not an adequate solution for research applications.⁸ While previous studies have proposed and empirically evaluated specific classification algorithms (eg, Skjøth et al, Buchardt et al, Gregersen et al, Duch et al), a broader conceptual synthesis of methodological approaches, their underlying assumptions, and their implications for epidemiological research remains limited.^8–11 In particular, existing work has primarily focused on algorithm performance, with less attention to how classification interacts with study design. This perspective paper provides (1) a conceptual framework distinguishing classification, episode formation, and study-level application, (2) a synthesis of both consensus-based and data-driven approaches, and (3) a discussion of broader epidemiological implications and future directions.

Implications for Epidemiological Research and Public Health

The transition to DNPR3, combined with longstanding variation in coding practices across hospitals and regions, has introduced two central threats to epidemiological validity: 1. misclassification of inpatient, elective outpatient, and out-of-hours contacts; and 2. broken comparability of time trends across DNPR2 and DNPR3.^12,13 These challenges are particularly pertinent to longitudinal research.¹⁴ This is increasingly important as register-based evidence becomes central to clinical decision-making, allocation of healthcare resources, and public health policy. Examples from recent Danish studies illustrate potential implications. For example, Jensen et al conducted a nationwide register-based study of individuals with a hospital contact due to chronic obstructive pulmonary disease (COPD) from January 1, 2015 to December 15, 2021, in Denmark, and found that hospital contacts decreased during the COVID-19 pandemic, reflecting reduced infections in society, while in-hospital care remained unchanged.¹³ In studies like this, ambiguous or inconsistent definitions may bias estimates of disease burden and healthcare utilization among admitted patients. Another example is Lund et al who performed a systematic literature search and identified papers validating cardiovascular variables in the DNPR during 1977 to 2024.¹⁴ They concluded that data quality varied substantially between variables and individual diagnoses depending on the algorithm used to identify them and that key determinants for validity were, among others, type of hospital stays (inpatient and outpatient) and calendar year.

In addition, changes in organizational structures may appear as abrupt increases or decreases in disease incidence or healthcare utilization. For instance, the sharp rise in recorded emergency activity after 2014 primarily reflected administrative changes, with the establishment of the 1813 medical helpline in the Capital Region and emergency department registration of out-of-hours contacts, rather than a true increase in emergency department activity.¹⁵ Separating coding artifacts and system reforms from genuine epidemiological trends is a difficult task and even more so without unified definitions.

Finally, while studies based solely on DNPR2 data may be more robust – though prone to other disadvantages – in terms of retrospective consistency, they may be difficult to interpret prospectively without a clear strategy for aligning past and present definitions.^16–20

Conceptual Challenges of Hospital Stay Classification

Classifying hospital contacts and constructing hospital care episodes in DNPR3 presents several conceptual and methodological challenges which can be understood at three distinct but interrelated levels: hospital stay classification, episode formation, and study-level application.

Hospital Stay Classification

At the most granular level, hospital stay information in DNPR2 distinguishes inpatient contacts from outpatient contacts and acute from elective contacts. In DNPR3, the explicit inpatient/outpatient variable was removed and from a statistical perspective, this constitutes an unsupervised classification problem: no gold-standard variable exists, and no single definition is agreed upon nationally and across research fields. In addition, DNPR2, although informative, lacks minute-level timestamps which would be needed to train fully supervised algorithms for outpatient visits in DNPR2.²¹ As a result, accuracy, interpretability, and transferability of an inferred type of hospital stay remain challenging.

Episode Formation

Beyond individual contacts, researchers often construct hospital care episodes by grouping sequential contacts into a single, clinically meaningful hospital pathway. Episode formation is essential for studying outcomes like length of stay, readmissions, or treatment pathways. Decisions about which contacts to combine based on time intervals, department transitions, and type of hospital stay (inpatient or outpatient) introduce methodological choices that can influence study results. Unlike hospital stay classification, episode construction emphasizes longitudinal continuity and clinical relevance rather than strictly administrative labels.

Study-Level Application

The role of hospital stay information depends on the study design and analytic objective. The types of hospital stay may define the study population (eg, all inpatients admitted with a specific diagnosis), serve as an exposure (eg, comparing subsequent outcomes between inpatients and outpatients or serve as a proxy for describing disease severity), or constitute an outcome (eg, assessing whether an intervention reduces acute outpatient visits). For example, Arulmurugananthavadivel et al used the type of hospital stay as both an outcome and an exposure when investigating temporal trends in Denmark from 1997 to 2017 in inpatient vs outpatient diagnosis of new-onset heart failure and the subsequent risk of death and hospitalization.²² Madsen et al used type of hospital stay as an exposure when comparing 30-day mortality and overall morbidity between outpatient and inpatient elective operations.²³

Data availability considerations also add complexity. Some research groups have access to detailed patient-level or department-level data, while others rely on more limited datasets, which constrain the choice of feasible methods and limit generalizability of any single classification approach.

Together, these conceptual layers highlight why classifying hospital contacts in DNPR3 is not a simple task but a multi-level problem, requiring careful methodological consideration to support valid, reproducible, and interpretable epidemiological research.

Methodological Advances and Current Approaches

The removal of the explicit inpatient/outpatient variable in DNPR3 has spurred the development of a range of methodological strategies to reconstruct the information. This narrative review focused on studies addressing classification of hospital contacts in DNPR, particularly following the transition to DNPR3. Relevant studies were identified through a search in PubMed conducted on 29 April 2026 using the search string “Danish national (patient OR health) regist* (inpatient OR outpatient OR admission OR hospitalization) contact epidemio*” and restricted to the period 1 January 2019 to 1 April 2026. Based on full abstract screening, the resulting 276 studies were categorized into

• 256 papers were irrelevant because they covered only the DNPR2 period, did not differentiate between hospital stay types, did not disclose how they defined hospital stay types, did the classification by department affiliation (eg Dybdal et al (2025, who included only outpatient clinics²⁴), used other study-specific definitions (eg, Danielsen et al (2026) considered day-case versus non-day-case²⁵), or used other sources such as Hyldgaard et al (2026) who used COPD²⁶ and Bernstorff et al (2022) who restrict their data population to the Central Denmark Region where they have maintained the hospital stay type classification across the DNPR2-DNPR3 transition;⁷

• 16 papers were relevant as they differentiated between hospital stay types but without further methodological considerations, eg, Winckler et al (2025), who defined inpatients as contacts lasting more than 5 hours,⁵ Geday et al (2025), who defined inpatients as contacts lasting more than 6 hours,²⁷ von Linstow et al (2024) and Rudolfsen et al (2025), who defined inpatients as contacts lasting more than 12 hours.^6,28

• 4 papers (Skjøth et al, Buchardt et al, Gregersen et al, and Duch et al) were considered directly relevant, proposing or evaluating hospital stay type classification in DNPR3 and presented in the following section.^8–11

The approaches typically combine consensus-based approaches and data-driven algorithms, often supplemented by temporal and trajectory-based models, and supported by validation to ensure reproducibility and comparability.

Algorithms for reconstructing a variable indicating the type of hospital stay may draw on both predefined rules and patterns in the data. Consensus-based approaches are often motivated by clinical considerations, including how hospital care is organized in practice and how different types of contacts are understood by clinicians. While consensus-based approaches provide transparent and interpretable classification criteria, differences across clinics, hospitals, and regions can limit their accuracy in a research setting.

Data-driven algorithms can infer the type of hospital stay from patterns in diagnoses, procedures, length of stay, and department-level characteristics. While supervised methods are not feasible due to the absence of a gold-standard variable in DNPR3, unsupervised clustering and probabilistic models allow researchers to identify latent groups of hospital contacts groups and estimate hospital stay probabilities. Consensus-based and data-driven elements are often combined, with rule-based structures guiding or constraining probabilistic inference.

Previous studies have illustrated these strategies in practice (Table 1). Skjøth et al (2022) proposed one of the first algorithms to characterize hospital stay types in both DNPR2 and DNPR3 following the DNPR3 transition.¹⁰ Their algorithm, based on the proportion (more or less than 50%) of overnight and elective contacts at the department level, was benchmarked against manual department classifications by clinicians. At the contact level, the algorithm achieved high positive predictive values (PPVs) for distinguishing inpatients (91–100%) and outpatients (~99%) across DNPR2 and DNPR3. However, performance was weaker for acute outpatient contacts (PPVs of 33% in DNPR2 and 56% in DNPR3 on the department level; 87% in DNPR2 and 85% in DNPR 3 on the contact level), reflecting substantial misclassification. Building on this, Buchardt et al (2025) extended algorithmic classification (implemented in R and available on GitHub) through department-based, contact-based, and hybrid methods across a nationwide dataset covering 20.9 million contacts in DNPR2 and 27.7 million in DNPR3.⁹ Their hybrid method improved classification for acute and elective outpatients (PPV 97.3%, sensitivity 96.8% for acute outpatients), indicating that the integration of multiple data dimensions can enhance contact-level accuracy.

Table 1.

Approaches to Classification of Inpatient and Outpatient Hospital Contacts in the Danish National Patient Register After the Transition from the Danish National Patient Register Version Two (DNPR2) to Version Three (DNPR3)

Study	Aim	Description	Key Strengths	Main Limitations
Skjøth et al (2022)10	Reconstruct inpatient/outpatient variable from DNPR2 in DNPR3	Consensus-based and data-driven operational rule: Contacts classified by department which are classified as Inpatient if >50% overnight Elective outpatient if ≤50% overnight and >50% elective patterns Acute outpatient if ≤50% overnight and ≤50% elective patterns	Can be implemented in both DNPR2 and DNPR3	Lower performance for acute outpatient contacts Not readily applicable and requires full department-level data (otherwise not reproducible) Sensitive to reorganizations and mixed-use departments
Buchardt et al (2025)9 hybrid department-based	Improve contact-level classification when reconstruct inpatient/outpatient variable from DNPR2 in DNPR3	Consensus-based and data-driven operational rule: Inpatient if contact is elective and department has ≥30% overnight contacts or contact if is acute and department has ≥23% overnight contacts. Elective outpatient if contact is elective and department has <30% overnight contacts. Acute outpatient if contact is acute and department has <23% overnight contacts.	Improved acute/elective outpatient performance Can be implemented in both DNPR2 and DNPR3	Not readily applicable and requires full department-level data (otherwise not reproducible) Sensitive to time-varying cut-offs Sensitive to reorganizations and mixed-use departments
Buchardt et al (2025)9 cluster-proxy	Improve contact-level classification in DNPR3	Consensus-based and data-driven operational rule: Contacts classified by priority, Inpatient if contact is acute and ≥9h or elective and overnight or elective, not overnight and ≥3.5h Elective outpatient if contact is elective, not overnight and <3.5h Acute outpatient if contacts is acute and <9h	Improved acute/elective outpatient performance Readily applicable	Sensitive to time-varying cut-offs Depends on timestamps which are of varying reliability and largely unavailable in DNPR2
Danish Health Data Authority (2021)4	Administrative standardization in DNPR3	Duration-based operational rule: Contact duration ≥12h inpatient; <12h outpatient; merge contacts within 4h	Readily applicable	Duration alone may misclassify many visits Depends on timestamps which are of varying reliability in DNPR2

In many applications, these consensus-based approaches and data-driven algorithms are further strengthened by incorporating temporal patterns and episode formation, which allow researchers to model clinically meaningful episodes, capturing the full trajectory from initial admission through subsequent follow-up contacts. Sequence and trajectory-based approaches capture longitudinal patient pathways, accommodating repeated visits, transfers, or organizational changes.

Gregersen et al (2024) focused on episode formation, developing a SAS macro to group sequential contacts into hospital care episodes.^11,29 Using a 4-hour window, approximately 120 million individual contacts were consolidated into around 106 million episodes, minimizing variation in episode counts relative to shorter or longer cutoffs. Episodes were subsequently classified as either acute or elective based on included contacts (DNPR2) or length of stay (DNPR3). This approach is particularly relevant for studies examining length of stay, readmission, or longitudinal patient trajectories, where single (administrative) contacts are less meaningful than the entire hospital stay, and their method could be combined with any approach mentioned in Table 1.

Further, linking DNPR data with data from other Danish registers containing data on medication at public hospitals, prescription records, laboratory results, and primary care data may further improve classification, particularly when reconstructing study populations or defining exposures and outcomes across research projects. For example, the Danish National Hospital Medication Register (Danish: Sygehusmedicinregisteret), contains a variable indicating type of hospital stay based on duration, and offers an external reference to explore and compare classification strategies in DNPR3. The Danish Health Care Organization Register (Danish: Sundhedsvæsenets Organisationsregister, SOR) contains information on organizational structures within the Danish health care system including unit type (Danish: enhedstype), which could be useful for identifying acute outpatients, since emergency rooms are specifically categorized as unit type “skadestue”.³⁰

Robust validation strategies are critical to ensure transparency and comparability across studies and registry versions. Algorithms can be evaluated against historical DNPR2 data, clinical expert review, or consensus-driven approaches. Duch et al (2025) provided the most comprehensive comparison across DNPR2 and DNPR3, evaluating four classification strategies: the Skjøth algorithm, the 12-hour rule suggested by the Danish Health Data Authority, a hybrid method combining DNPR2 the inpatient/outpatient variable, and the Gregersen episode formation approach.⁸ Applying the same algorithm consistently across both registry versions yielded the most stable longitudinal patterns. While all approaches performed well for inpatient and elective contacts, discrepancies persisted for acute outpatient visits, emphasizing the importance of aligning algorithm choice with the research question, analytic unit (contact versus episode), and the disease or patient group under study.

Collectively, these methodological advances enable researchers to construct information on type of hospital stay at both the contact and episode levels while supporting a wide range of study designs. Despite the absence of an explicit inpatient/outpatient variable in DNPR3, the studies emphasize that robust methodological strategies enable researchers to construct valid hospital stay and episode-level classifications, though, classification of acute outpatient visits remains a persistent and unresolved challenge. The most appropriate method depends on the desired level of classification granularity, the specific research question, and whether the goal is defining populations, modeling exposures, or analyzing outcomes.

Discussion – Performance, Generalizability and Trade‑Offs

The emerging methodological literature demonstrates that reconstructing hospital stay classifications (inpatient/outpatient) in DNPR3 is feasible. Yet, no single approach is optimal across all applications. Algorithmic and machine learning methods accurately distinguish inpatients from elective outpatients. However, performance for acute outpatient contacts remains limited, reflecting both heterogeneous registration practices and the inherent difficulty of capturing emergency care pathways without a gold-standard reference.

Episode-based approaches offer additional advantages for longitudinal and clinically oriented analyses. By grouping sequential contacts into meaningful hospital care episodes, researchers can more reliably study outcomes such as length of stay, transfers, readmissions, or treatment trajectories. Nevertheless, episode formation introduces methodological challenges such as choosing time-window thresholds for grouping that may affect study results, and data availability might also limit episode-based approaches. Trade-offs between granularity, interpretability, and sensitivity to short or complex hospital visits must be carefully considered.

Generalizability is another key consideration. Algorithms validated in one region, department, or clinical specialty may not perform equivalently elsewhere due to differences in hospital organization, coding practices, or local administrative policies. Similarly, changes over time in registration practices or healthcare delivery may limit the comparability of classifications across calendar years. Researchers must therefore assess the applicability of their chosen method for each study population, outcome, and DNPR version.

From a study-level perspective, the choice of classification approach depends on the research question. Contact-level algorithms are appropriate when defining populations or exposures at the visit level, while episode-based methods may be preferable for outcomes spanning multiple contacts. Hybrid approaches that integrate multiple data sources or classification strategies offer improved accuracy but increase complexity and reduce interpretability, which may be a concern for rapid or transparent epidemiological analyses. The methods considered here rely on proxy variables such as overnight stays, department patterns, or time intervals between contacts. Therefore, they may not perfectly reflect clinical reality (eg, same‑day discharges, transfers, outpatient consultations that mimic inpatient patterns). Researchers should explicitly acknowledge these assumptions and interpret results with caution, particularly for analyses focusing on acute outpatient contacts.

In practice, a consistent application of validated algorithms across DNPR2 and DNPR3 would improve temporal comparability and mitigate artificial trends arising from structural changes in data.

This perspective has several limitations. First, it is not based on a dedicated systematic review of the literature, and it was not the primary aim to provide a detailed account of each relevant study. Second, the current evidence base is limited and largely produced by a small number of research groups, reflecting the emerging nature of this field.

An important conceptual question is whether efforts should focus on reconstructing DNPR2 inpatient/outpatient classifications in DNPR3, or whether researchers should instead manage historical DNPR2 and DNPR3 data with the same approach to maximize data comparability. While reconstruction facilitates longitudinal comparability, it may impose outdated categorizations on a system that reflects current clinical workflows. Conversely, adopting DNPR3 definitions may improve clinical relevance but complicate historical comparisons. The appropriate strategy depends on the study objective.

Recommendations and Future Directions

Based on current evidence and methodological developments, several recommendations emerge to guide future research and strengthen the validity of epidemiological studies using the DNPR.

Establish a Consensus-Driven National Framework

A standardized classification of hospital contacts should be developed at the national level, involving researchers, including clinicians, epidemiologists, and informaticians. At minimum, the framework should define a clear taxonomy of type of hospital stay (inpatient, elective outpatient, acute outpatient/emergency) and provide guidance on episode formation, ensuring comparability across DNPR2 and DNPR3. Classification algorithms should be periodically evaluated and updated to reflect changes in hospital organization, registration practices, and healthcare delivery. External validation using auxiliary registers could improve reliability and provide a reference for benchmarking algorithmic performance. Another possible solution would be to reinstate clinician registration of type of hospital stays at the point of care and reintroduce the variable in DNPR. Although such registration practices are also susceptible to variation, this limitation is not unique and applies to many routinely collected register variables.

Emphasize Transparency and Reproducibility

All studies should explicitly report the algorithm or method used, the underlying assumptions (eg, time windows, proxies), and any validation metrics. Providing code, macros, or reproducible workflows will facilitate comparability across studies and reduce misinterpretation of results.

Expand to Broader Hospital Settings

While research specific to classification of hospital stay types has mostly focused on public somatic hospitals, future efforts should explore classification in psychiatric and private hospital contacts. Such expansion will increase the comprehensiveness of epidemiological analyses and support a more complete understanding of healthcare utilization. Sharing best practices, validation results, and standardized algorithms across countries can help harmonize definitions, support cross-national epidemiological studies, and strengthen the generalizability of findings beyond Denmark.

Use Technological Advancements Cautiously

Machine learning, natural language processing, and other AI or hybrid methods offer potential to automate and refine hospital stay classification. However, interpretability and accessibility should remain central priorities, particularly for studies informing policy or clinical decision-making.

By implementing these recommendations, researchers can produce more valid, reproducible, and interpretable findings while maintaining alignment with clinical and policy-relevant questions. Harmonized, transparent classification frameworks will ultimately enhance the utility of DNPR data for evidence-based interventions and public health decision-making.

Conclusion

The transition from DNPR2 to DNPR3 has created both challenges and opportunities for epidemiological research in Denmark. The removal of the explicit type of hospital stay (inpatient/outpatient) variable necessitates innovative approaches to classify hospital contacts to maintain longitudinal comparability over DNPR versions. Evidence from recent methodological studies demonstrates that data-driven, multi-source, and trajectory-based approaches can provide reasonable classification of inpatient and elective outpatient contacts, while performance for acute outpatient visits remains less reliable.

Harmonized definitions, transparent and publicly available algorithms, and rigorous validation are essential to ensure that classification of hospital stays and episodes are reliable, interpretable, and applicable across different hospitals, regions, time periods, and DNPR versions. Aligning methodological choices with study objectives, whether defining populations, exposures, or outcomes, further strengthens the validity and reproducibility of findings.

Establishing a consensus-driven national framework, routinely monitoring classification performance, expanding research to psychiatric and private hospital settings, and fostering international collaboration will support the continued use of DNPR data for robust epidemiological analyses. As digital health tools and register-based infrastructures evolve, a coordinated approach to hospital stay classification will enhance study reliability, inform public health policy, and ultimately improve healthcare decision-making.

Disclosure

Lene Dreyer reports being a Member of the steering committee of the Danish Rheumatology Quality Registry (DANBIO, DRQ), during the conduct of the study; grants from BMS, grants from Abbvie, and being a member of the steering committee of the Danish Rheumatology Quality Registry (DANBIO, DRQ) outside the submitted work. The authors report no other conflicts of interest in this work.