Data Resource Profile: Regional healthcare information platform in Halland, Sweden
Awais Ashfaq, Stefan Lönn, Håkan Nilsson, Jonny A Eriksson, Japneet Kwatra, Zayed M Yasin, Jonathan E. Slutzman, Thomas Wallenfeldt, Ziad Obermeyer, Philip D. Anderson, Markus Lingman · International Journal of Epidemiology · 2019
Accurate and comprehensive healthcare data coupled with modern analytical tools can play a vital role in enabling care providers to make better-informed decisions, leading to effective and cost-efficient care delivery.1 Moreover, it facilitates quantifying health value—a powerful driver for improving the healthcare system.2 Development of a comprehensive healthcare data infrastructure that captures complete care processes at individual, organizational and population levels remains a challenge, primarily due to ever-increasing complexity in the healthcare ecosystem. Complexity is driven by the advancement in medical knowledge, which in turn has demanded a paradigm shift in healthcare: from care in a single unit to care across multiple units with varying but specialized expertise. Yet, our ability to integrate, prioritize and personalize narrowly construed information has not expanded at a similar pace, which makes it challenging to map the complete care cycle of an individual and related resources on a single platform. In addition to the technical challenge, there exist legal, cultural or political limitations concerning ‘data accesses’ observed by one or more stakeholders involved in the healthcare system.3 The healthcare landscape is very fragmented. Stakeholders include clinics, laboratories, hospitals, rehabilitation units, corporations, government, etc., that often have distinct leaderships, budgets, goals, regulations and tools of operation. The reasons for limited or restricted data access might be legitimate (due to security risks) or illegitimate (to coerce another stakeholder into using a particular system). Another factor is simply a lack of informational interoperability that has developed as a by-product of the fragmented system of healthcare delivery. Either way, it prevents care-providers from accessing complete datasets and thus being unable to understand and safeguard all aspects of the patient health journey. Creating a common information structure on a single platform would enable a step towards individualized, safe and cost-effective medicine. Overcoming the aforementioned challenges involved breaking down internal silos and uniting stakeholders across knowledge domains. This paper introduces a working concept of a comprehensive strategic healthcare analysis and research platform developed in-house by Region Halland (RH) in collaboration with Brigham and Women’s Hospital (BWH). The platform is populated with clinical and administrative information pertaining to every consumer of healthcare with public funding since 2009 in RH. Around 65% of this population is registered in Halland and the remaining population consists of tourists and people from neighbouring counties (Skåne, Västra Götaland, Jönköping and Kronoberg). Tables 1–3 provide an overview of the content of the platform. Of note, 24 private providers support nearly 40 and 6% of total primary care and outpatient specialty visits in Halland, respectively. This information is part of the platform, provided that necessary agreements are in place. Figure 1 reflects the broad representativeness of the persons populating the platform compared with Halland census 2017 in terms of age and gender. Age and gender distribution in the regional healthcare information platform in Halland, Sweden as of December 2017 compared with Halland census data 2017. Population statistics in the regional healthcare information platform in Halland, Sweden, January 2009–October 2018 Population statistics in the regional healthcare information platform in Halland, Sweden, January 2009–October 2018 Distribution of International Classification of Diseases, 10th Edition (ICD-10) codes in the regional healthcare information platform in Halland, Sweden. January 2009–October 2018 Distribution of International Classification of Diseases, 10th Edition (ICD-10) codes in the regional healthcare information platform in Halland, Sweden. January 2009–October 2018 Distribution of anatomical therapeutic chemical (ATC) drug codes among Halland inhabitants in the regional healthcare information platform in Halland, Sweden, January 2009–October 2018 For Antineoplastic agents we observe higher patient pick-up count than prescribed. One reason is that some Halland patients visit bigger hospitals like Sahlgrenska (Gothenburg) and Karolinska (Stockholm) for such specialized cancer-treatment drugs. As of now, prescription data beyond RH care centres is not available in the regional platform. However, the pick-up data are fetched from national registers and thus include information about all RH inhabitants picking up drugs in any part of Sweden. Distribution of anatomical therapeutic chemical (ATC) drug codes among Halland inhabitants in the regional healthcare information platform in Halland, Sweden, January 2009–October 2018 For Antineoplastic agents we observe higher patient pick-up count than prescribed. One reason is that some Halland patients visit bigger hospitals like Sahlgrenska (Gothenburg) and Karolinska (Stockholm) for such specialized cancer-treatment drugs. As of now, prescription data beyond RH care centres is not available in the regional platform. However, the pick-up data are fetched from national registers and thus include information about all RH inhabitants picking up drugs in any part of Sweden. Sweden has a decentralized healthcare system, with 21 county councils responsible for all primary and secondary care delivery in their region. This decentralization allowed each region to invest in an independent information technology (IT) framework—to house their healthcare related data—that is in line with the county’s needs, goals and regulations. The county council RH in Sweden has an electronic health record system entitled VAS that providers have used for many years and into which providers across all healthcare facilities routinely enter the clinical data of patients according to Region guidelines. Each registered individual in Sweden has a unique personal ID that is used to link records, allowing seamless combination of information across different facilities. The VAS data are housed in a single data warehouse along with other patient and administrative data collected from over 20 different IT systems in RH, including data also stored in national registers. In 2016, RH constructed a strategic healthcare analysis and research platform to enable agile management and analysis of this data (Fig. 2). Whereas the different warehouses continue to serve as the backbone for collecting and sharing healthcare data, the platform is a structured, filtered and pseudo-anonymized far-reaching subset of it, designed to facilitate rapid analysis for clinical and management research and evaluation purposes. Data flow in Region Halland, Sweden. The linked data are housed in an SQL Server. It is a visit- and patient-centred platform, containing data tables relating to patient or visit information in different care units. A table is a data storage structure that is like a spreadsheet: each column contains consistent information (e.g. platform identifiers, dates/times, clinical codes) and each row contains an instantiation of that information. The platform has been reported to national authorities according to Swedish regulations. The content is owned by RH and maintained on premise. The frequency of data collection hinges on individual patient needs and is influenced by age and underlying morbidity. The data thus reflect real-world practice with its advantages and limitations. Moreover, linked data are time-variant (updated monthly). There are five types of table in the platform (Fig. 3). The regional healthcare information platform in Halland, Sweden. VisitID, Visit ID; PatID, Patient ID; OrgID, Organisation ID; VardgivID, Care provider ID; GeoID, Geographic ID; ED, Emergency Department; TDABC, time-driven activity-based costing. Visit tables define a unique hospital/clinic encounter. These include primary care, outpatient specialist care, emergency, inpatient, ambulance and pharmacy visits. Detail visit tables explain the aforementioned events in terms of diagnoses, procedures, notes, medications, etc. related to each visit. These can be linked to visit tables through visit IDs or platform IDs and date. Detail patient tables include patient demographics extracted from national registers and include gender, age, primary care units where the person is registered, corresponding geography IDs etc. Resource tables for personnel utilization (nurses, assistant nurses, physicians, medical secretaries) per hour per day from human resource data. It also includes codes for all care delivery units and all care providers (pseudo-anonymized), enabling linkage of care encounters. Cost tables describe costs allocated at the patient-encounter level across all care venues, and take into account resources and utilized capacity. The platform is designed to operate within Swedish and EU laws and regulations respecting confidentiality and privacy pertaining to patient and corporate data. Patient data is pseudo-anonymized by removing identifiable features: name, date of birth, address and telephone numbers. The 10-digit Swedish Personal IDs are replaced with a platform ID. The key to map personal IDs to platform IDs is secured at regional level and used only if a specific study requires an integration of external registers like income or education data with RH-generated data. Access to the key is limited to authorized IT personnel at RH and not to researchers. For studies that do not require data from external registries, preferably new platform IDs are generated for the target population only and the key is deleted so the data cannot be traced back to individual patients. Data are accessed using internal client systems secured within regional IT firewalls. The clients are only accessible to authorized researchers with an approved project according to a pre-specified process. The regional platform supports an agile team of clinicians, policy-makers, lawyers, economists, administrators and data scientists to transform data to information, knowledge and (actionable) insights by looking at a powerful 360-degree data-view of the healthcare system. The over-arching goal is to increase knowledge by applying cutting-edge research methodology and thereby facilitate informed decision-making in a way that improves patient value and simultaneously expands clinical knowledge through research. Core research areas of interest include, but are not limited to, understanding care patterns, predicting outcomes (and adverse events), understanding and quantifying resource utilization (for example by detecting deviations from expected or preferred clinical practices) and operational simulations. Clinical decisions (for admissions, transfers, interventions, etc.) are often based on clinicians’ assessments of likely benefits derived from their personal experience and knowledge of published evidence. However, with an exploding medical knowledge and escalating population, the complexity of medicine now exceeds the capacity of the unaided human mind.4 A data-driven electronic decision support system will serve as a facilitator for clinicians by highlighting relevant health outcome predictions tailored to the specific profile of individual patients. One current focus is leveraging advanced machine learning algorithms that have been shown to capture interesting relationships from huge and complex data structures.5,6 Making sure that patients are treated in line with evidence and recommendations and that the care is cost effective is crucial in any healthcare system, but assuring quality and understanding patient-level costs across the system is a challenge. The RH data platform enables analysis at the cohort and highly granular level to determine compliance with established guidelines, such as those for congestive heart failure as an example. Furthermore, the platform enables near-time collection and analysis of multiple quality metrics to study effects of feedback systems to clinicians and managers. This capability enables analysis at the cohort level (but based on individual data) to determine, for example, the rate of adherence to anticoagulation guidelines in atrial fibrillation. We can measure the risk of stroke for each individual patient via an established function (i.e. CHADS2-Vasc) and determine if each patient appropriately or inappropriately received a prescription for an anticoagulant. With a long-term perspective, and data to match, financial costs and savings could be calculated alongside clinical quality metrics. We have developed a novel patient encounter costing (PEC) methodology that integrates existing financial, operational and clinical data-sources to assign costs to each patient encounter.7 It is based on the principles of time-driven activity-based costing (TDABC), but is substantially less labour intensive, enabling it to be implemented across an entire health system. Our methodology also accounts for the costs of unutilized capacity, which is essential to modelling the impact of interventions. PEC methods can be applied within all care venues (inpatient, emergency, outpatient specialist care, primary care and ambulance) which is not possible using existing traditional accounting methods. Understanding costs of care and having the power to model cost changes with care processes changes, enables coupling of cost and quality assessment for strategic planning. The RH platform allows clinical research questions to be analysed in a real-world evidence (RWE) design. Specifically this is useful in following introductions of new drugs in clinical real-world settings. This could change the way of conducting phase 4 post-marketing studies, but also introduces a concept beyond registry-based randomized clinical trials (R-RCTs) by not being limited to registry data. The key strengths of the platform lie in the highly granular coverage of individual patient and provider level data (both laterally, across multiple care units and longitudinally, over long periods of time) and deep integration of cost data on the visit level. In addition when appropriate approvals are in place data are linked to external national high-quality registries (e.g. containing causes of deaths and drugs picked up at pharmacies) that are abundant in Sweden. Table 4 shows an overview of the platform in relation to two other well-known data structures: MIMIC-III8 and CPRD.9 Regional healthcare information platform in Halland, Sweden: overview compared with CPRD and MIMIC-III Visit-level costs are recorded since 2015 and updated annually. As of 31 October 2018. Basic statistics according to.9 However, we recommend readers visit the following page to get an updated and in-depth understanding of the CPRD resource. https://cprd.com/home Basic statistics according to.8 M, male; F, female. Regional healthcare information platform in Halland, Sweden: overview compared with CPRD and MIMIC-III Visit-level costs are recorded since 2015 and updated annually. As of 31 October 2018. Basic statistics according to.9 However, we recommend readers visit the following page to get an updated and in-depth understanding of the CPRD resource. https://cprd.com/home Basic statistics according to.8 M, male; F, female. The Regional platform provides a comprehensive digital print of patient health status by capturing information related to every single contact a patient makes with any public healthcare unit in the county (as well as private units when necessary agreements are in place) including visits to primary care, specialists, emergency, ambulance transports, admissions and pharmacy (Fig. 4). This allows modelling of personalized disease trajectories and recommending timely interventions to improve patient outcomes. From an administrative perspective, a systems view facilitates operational modelling and simulations of healthcare resources to overcome challenges like resource utilization, work force planning and scheduling, patient flow streamlining, quality variability and more. Moreover, Halland constitutes a fairly stable population with >90% of the inhabitants being local residents. Put differently, migration from Halland is low, which inherently allows and obligates RH to store long-term patient information, and since it is currently updated monthly, it facilitates both retrospective and prospective research. Patient data collection points in the regional healthcare information platform in Halland, Sweden; *hospital transitions refer to in-hospital transfers from one facility to another. The integration of cost data allows understanding of care-processes from an economic perspective and allows quantifying of health value using the recently developed PEC methodology.7 Put differently, the platform not only aims to optimize outcomes, but outcomes with respect to costs and other non-monetary resources, including unutilized capacity. Since the RH information platform is populated with routine collected data (RCD), not with the primary purpose of being used for research and development analyses, all results need to be evaluated and validated carefully. For instance, should there be errors during data entry in any variable, they are reflected in the warehouse, perpetuating to the linked data on the platform. Moreover, the variability of completeness of data among patients over time results in missing data. Data completeness often depends on the underlying condition of patients, care-provider and workplace as well as varying care policies. Additionally, the RH platform (like all RCD) comes with subtle biases that exist when analysing observational data, thus challenging the causal interpretations of clinical studies. RH is exploring several different possibilities of extending the RH platform, such as by including municipality homecare visits and clinical images for research. Additionally, a future possibility is to integrate patient-generated data from home sensors and wearables, as well as patient-reported experience and outcome measures (PREMs and PROMs) through patient apps for interacting with the platform to monitor patient health beyond regional walls.10 Access for operational analyses is approved by administrators for each task according to a pre-specified process. Gaining research access to the platform requires an approved project and analytical plan from an Ethical Review Board in Sweden. Researchers are required to sign a confidentiality and secrecy agreement with RH prior to gaining access. For studies that require data from external sources, data are available with additional approvals. The agreements have been carefully formulated after iterative legal reviews in accordance with the General Data Protection Regulation (2916/679) and other applicable laws. Interest for research collaborations can be addressed to Region Halland, Research and Development (R&D) department at [[email protected]]. The regional healthcare information platform was set up to study the healthcare system as a whole including all knowledge domains and to quantify health value. It encapsulates 360-degree pseudo-anonymized data covering clinical, operational capacity and financial data on over 500 000 patients treated since 2009 across all public care delivery units in the county of Halland, Sweden including 24 primary care units, 3 hospitals and 2 emergency care units. Patients who opted-out from the research setting are not included in the platform. The main categories of data collected include: demographics (patient and visits), care visits (primary care, outpatient specialty, emergency and inpatient), other encounters (ambulance and pharmacy), clinical data (diagnoses, procedures, lab results, radiology diagnostics, Rx prescriptions and collected, waveforms, notes), visit-level costs, resources (doctors, nurses, others) and occupancy (emergency departments, wards). Clinical images, patient-generated data, genetics and visits outside Halland are not included in the platform. Content is owned by Region Halland, Sweden. Queries regarding collaboration can be directed to [[email protected]] Comprehensive deep data from all available aspects of healthcare is the foundation for new insights in the era of new competent analytics including machine learning. We created a platform for such data within the required legal framework to research and development on data in a fragmented healthcare ecosystem. The data platform is a by Region Halland, Sweden. Research are from public and private funding in Sweden This study was part of the of at research and received funding from the Swedish Research of