Development and implementation of databases to track patient and safety outcomes

Abstract

Purpose of review

Recent advancements in big data analytical tools and large patient databases have expanded tremendously the opportunities to track patient and safety outcomes.

We discuss the strengths and limitations of large databases and implementation in practice with a focus on the current opportunities to use technological advancements to improve patient safety.

Recent findings

The most used sources of data for large patient safety observational studies are administrative databases, clinical registries, and electronic health records. These data sources have enabled research on patient safety topics ranging from rare adverse outcomes to large cohort studies of the modalities for pain control and safety of medications. Implementing the insights from big perioperative data research is augmented by automating data collection and tracking the safety outcomes on a provider, institutional, national, and global level. In the near future, big data from wearable devices, physiological waveforms, and genomics may lead to the development of personalized outcome measures.

Summary

Patient safety research using large databases can provide actionable insights to improve outcomes in the perioperative setting. As datasets and methods to gain insights from those continue to grow, adopting novel technologies to implement personalized quality assurance initiatives can significantly improve patient care.

INTRODUCTION

The field of anesthesiology was one of the first to pioneer the concept of patient safety. Since the development of the anesthesia patient safety movement more than 35years ago, a multitude of initiatives has resulted in a dramatic decline in adverse perioperative patient outcomes [1]. While the specialty of anesthesia has achieved tremendous success in decreasing perioperative mortality [2], much work remains to be done in improving patient outcomes and increasing anesthesia safety, especially in the developing world [3,4^■■]. The most recent developments of novel tools to gain insights from large data in healthcare hold the promise to generate new actionable insights which can lead to a new era in patient safety.

Given the increasing availability of large datasets, conducting anesthesiology safety and outcomes research that is generalizable and widely applicable is becoming more accessible than ever before. At present, the most used sources of data for large observational studies are administrative databases, clinical registries, and electronic health records (Table 1). These data sources have enabled research in patient safety topics ranging from rare adverse outcomes, such as perioperative ischemic optic neuropathy [5^■] to studies of the nonopioid modalities for pain control in a large cohort of pregnant patients who underwent cesarean delivery [6]. In the past year, two comprehensive reviews of the data sources in anesthesiology research were published [7^■■,8^■■]. In this review, we discuss the most common data sources used in patient safety retrospective anesthesiology research, the strengths and limitations, implementation in practice, and future directions of the field. Rather than an exhaustive review of all database research, we focus on the current opportunities to use technological advancements to improve patient safety.

Table 1.

Types of large databases used in anesthesiology patient safety studies

Database	Type	Examples of studies	Reference
PHD	Administrative private	Safety of administration of tranexamic acid and lack of heightened risk of thrombotic events	[14]
NRD	Administrative governmental	Readmission rates following spinal cord stimulator implantation are 7.7% and are most often due to infection; open surgical compared to percutaneous implantation was associated with both extended initial hospitalization and a higher rate of readmission	[17]
MPOG	Anesthesia Registry	Risk of difficult intubation of 1 : 49 and risk of failed intubation of 1 : 808 in cesarean delivery performed under general anesthesia	[21]
BIDARDR	EHR	Preoperative administration of gabapentinoids reduces the risk of hospital readmission and postoperative respiratory complications in the noncardiac surgical patient population	[25]
CCPHDS	EHR	The use of brachial artery cannulation for hemodynamic monitoring in a cardiac surgical population is safe and rarely causes complications.	[26]
ICES	Registry from Ontario, Canada	Care by high-volume anesthesiologists is independently associated with lower odds of a composite of 90-day major morbidity and readmission and lower unplanned intensive care unit admission	[28■]

BIDARDR, Beth Israel Deaconess Anesthesia Research Data Repository; CCPHDS, Cleveland Clinic Perioperative Health Documentation System; EHR, electronic health records; ICES, Institute for Clinical Evaluative Services; MPOG, Multicenter Perioperative Outcomes Group; NRD, National Readmission Database; PHD, Premier Healthcare Database.

ADMINISTRATIVE DATABASES

Administrative databases are large repositories of medical information that are collected and maintained by hospitals, health maintenance organizations, and health insurers. These databases generally include information, such as medical claims for reimbursement, records of services, procedures, prescriptions, and diagnoses. Unlike electronic health records (EHRs), these data are not meant to be granular records of a patient’s health information, rather they are intended to be used for financial bookkeeping and other administrative endeavors. However, despite the lack of detailed clinical information in these databases they have become a powerful research tool in recent years given the large volume of information they can provide [9].

Administrative databases are particularly useful for longitudinal questions (given that they often contain information about patients spanning multiple interactions with the healthcare system), incidence calculation (in the population included in the database), and adequate sample size for identification of rare events and the requisite power needed to investigate them with techniques such as multivariate adjustment for risk factors that simply is not possible with smaller studies [9]. For example, perioperative ischemic optic neuropathy is a rare, devastating complication of spinal fusion surgery. Using a large longitudinal medical claims database, of 65 cases and 106,871 controls was used to develop an accurate predictive model for the risk of ischemic optic neuropathy which allowed more accurate determination of the temporal sequence of the adverse event and the associated factors [5^■].

Despite being robust tools for observational research, the use of administrative databases has its limitations as the data included are not optimized for research purposes. The coding of disease entities may differ across institutions and information not pertinent to billing is often excluded. Surveillance bias may be introduced if data are systematically missing and these potential threats to validity need to be addressed with a rigorous study design [10]. To account for these challenges, newer methods like estimating causal treatment effects with propensity score methods are developed; these methods can also increase transparency and reproducibility of the results [11]. Another problem that is unique to insurance databases is that patients are only included so long as they are covered by the entity collecting data. This means, for example, for an insurance database if a patient’s insurance coverage wereto lapse or change, data in that period would be missing. Similarly, if a patient receives care at multiple facilities only some of which are contained in a database, data will be missing. Administrative databases also may have limited generalizability given that some of them are geographically limited such as Premier Healthcare, which skews towards the American South [12] or limited to patients of a certain age or socioeconomic status, such as Medicare and Medicaid [13]. Some of these limitations can be addressed with data linkage where patient data in one database is linked to their data in another database for a more complete record. An example of this would be an American patient who is both elderly and low-income and may have data stored both in Medicare and Medicaid databases, which researchers can link [13].

The Premier Healthcare Database is a privately owned database maintained by a group purchasing organization with over 1000 contributing hospitals. It contains information collected upon inpatient discharges mostly from nonprofit, community, and teaching hospitals that submit administrative, healthcare utilization, and financial data. The database contains over 121 million unique visits with over 10 million per year since 2012, which represents approximately 25% of USA inpatient admissions annually [12]. While it is not anesthesia-specific, it has been used for multiple patient safety studies pertinent to anesthesiologists. Recently, data from 765,011 total hip/knee arthroplasties demonstrate the safety of administration of tranexamic acid and lack of heightened risk of thrombotic events like myocardial infarction, ischemic stroke, or transient ischemic attack [14]. In another study, Reed et al. demonstrated that in 804,752 parturients who delivered via cesarean delivery with neuraxial anesthesia from 2008 to 2018, 81.3% received acetaminophen-opioid combination medications, whereas only 6.1% received the recommended multimodal combination [15] of neuraxial morphine with nonopioid analgesics in the postpartum setting [6].

The National Readmission Database is a database maintained by the Agency for Healthcare Research and Quality with data from 30 states and contains data from patients of all ages and any payer [16]. It captures approximately 35 million hospitalizations annually accounting for approximately 60% of USA hospitalizations [16]. Using a cohort of 3737 patients admitted for spinal cord stimulator placement, Goel et al. found that readmission rates following stimulator implantation are 7.7% and are most often due to infection. They also demonstrated that open surgical implantation was associated with both extended initial hospitalization and a higher rate of readmission when compared to percutaneous placement [17].

CLINICAL REGISTRIES

The clinical registries are large repositories of data for specific patient populations often provided by multiple contributing institutions. They can be broad in scope, such as the Society of Thoracic Surgeons Database, which captures patients undergoing thoracic surgery in the USA or narrow such as the Malignant Hyperthermia Association of the USA [18]. They tend to contain more granular information than administrative databases including standard definitions of given diagnoses, details about treatments delivered, and specific patient outcomes that are of interest to researchers [7^■■]. However, clinical registries are often labor-intensive and expensive to maintain especially if they are multi-institutional [8^■■]. As with administrative databases, clinical registries can be prone to bias if the data they contain do not represent a diverse sample of the population.

The Multicenter Perioperative Outcomes Group (MPOG) is a registry of data on patients undergoing anesthesia collected from a consortium of over 60 USA hospitals. It contains data from over 18 million encounters [19]. Recent notable studies include research on postoperative outcomes. In a cohort of patients at increased risk for pulmonary complications, compared with neostigmine, the use of sugammadex was independently associated with a reduced risk of subsequent development of pneumonia or respiratory failure, adjusted odds ratio 0.39, P<0.0001 [20^■]. In another example, MPOG data was used to study 14,748 cases of cesarean delivery performed under general anesthesia and the results demonstrated an overall risk of difficult intubation of 1:49 and risk of failed intubation of 1:808 [21]. The risk factors for difficult intubation were mostly nonobstetric in nature. These safety insights are relevant to the daily practice of the anesthesiologist.

It should also be noted that many studies have been conducted using surgical clinical registries. For example, the American College of Surgeons National Surgical Quality Improvement Program was developed with the goal to improve the quality of surgical care. In the past year, topics were published ranging from an evaluation of the likelihood of postoperative complications in patients undergoing ankle fracture repair based on anesthesia type [22] to an investigation of which intraoperative data can be used for risk prediction of mortality after intra-abdominal surgery [23].

INSTITUTION-SPECIFIC REGISTRIES AND ELECTRONIC HEALTH RECORD

Multiple institutions across the USA maintain their own internal databases including some that are specific to the perioperative period. The challenges of using EHR data are well described and include technical barriers to access, lack of standardization, and varying quality of the data [24]. Nevertheless, EHR data are a valuable source of real-world practice insights and can aid machine learning, cross-sectional and longitudinal studies. Two such examples are the Beth Israel Deaconess Anesthesia Research Data Repository and the Cleveland Clinic Perioperative Health Documentation System. Recently, the Beth Israel Deaconess Medical Center database was used to demonstrate that in the noncardiac surgical patient population, the preoperative administration of gabapentinoids reduces the risk of hospital readmission and postoperative respiratory complications, which was in part due to reduced intraoperative opioid use [25]. In the past year, Cleveland Clinic researchers have demonstrated that in a cardiac surgical population the use of brachial artery cannulation for hemodynamic monitoring is safe and rarely causes complications [26].

INTERNATIONAL REGISTRIES

Multiple impactful patient safety studies have also been published using data from outside of the United States. One notable example is ICES (for-merly known as the Institute for Clinical Evaluative Services), a province-wide database with access to data from all patients who receive medical care in Ontario, Canada from public hospitals. Using this database, the safety of peripheral nerve blocks in elderly patients with hip fractures has been demonstrated [27]. In another study, this database was used to gain insight into the expertise of anesthesiologists. The results demonstrate that care by high-volume anesthesiologists was independently associated with lower odds of a composite of 90-day major morbidity and readmission (adjusted odds ratio [aOR], 0.85; 95% confidence interval, CI, 0.76–0.94), and unplanned intensive care unit admission (aOR, 0.84; 95% CI, 0.76–0.94) [28^■]. Other large databases originate from Denmark [29], Italy [30], and South Korea [31]. Using a large, linked, national health administrative database would allow investigating the incidence of events like opioid use after hospital discharge and the associated predictors on a national level [32]. In another study, a cardiovascular mortality prediction algorithm applied to administrative databases from Canada and UK had moderate validity when compared to vital statistics data [33]. Studies like these provide insights into current safety practices and healthcare utilization patterns outside of the USA.

IMPLEMENTATION OF PATIENT SAFETY INSIGHTS

The success of any patient safety initiative depends on implementing and tracking measures on a provider, hospital, national, and global level (Fig. 1). Entities at each of those levels can both provide data to large databases to be used for patient safety research and also to be used for monitoring the implementation of quality assurance initiatives.

FIGURE 1.

Patient safety research from large databases can inform practice at multiple levels in healthcare: provider, institutional, national, and global levels. In turn, each of those levels can provide data on the implementation of patient safety.

Technology can play a leading role in enhancing the monitoring and reporting of the quality improvement initiatives on the individual provider and institutional level. In one institution, automating the patient perioperative quality assurance tracking by use of the embedded outcome reporting feature in the anesthesia information management system increased the rate of reporting compared to the existing paper reporting system by 389.8% [34]. Further integration of these systems with the American Society of Anesthesiologists’ Anesthesia Quality Institute’s National Anesthesia Clinical Outcomes Registry can facilitate reporting and gaining more insights at a national level. In this fashion, the MPOG Anesthesiology Performance Improvement and Reporting Exchange (ASPIRE) aims to improve patient care by providing individual feedback to the anesthesia providers [35]. Each participating site develops safety goals and selects a number of quality measures based on previously defined priorities. Subsequently, the individual providers receive monthly emails with information about their performance and how it compares to the department average; moreover, they can also obtain additional safety training. Currently, one of the ASPIRE measures is avoiding hypotension in patients presenting for noncardiac surgery independent of other risk factors [35]. Adjusting for risk factors based on the individual provider’s patient mix may further identify opportunities for quality improvement [36]. Moreover, capturing data from the EHR may further improve this measure in the future.

On an institutional level, tracking the safety initiative implementation will allow for an assessment of the effectiveness and identifying areas in need of improvement. Comparing the incidence of adverse events before, during, and after interdisciplinary communication training to healthcare professionals with the goal of enhancing safety demonstrates measurable mitigation of the incidence of adverse outcomes after the training [37]. A more practical approach that can be utilized in near real-time is by creating dashboards and other visualization tools that can allow hospitals to better track outcomes and compare performance in relation to state and national benchmarks. Using a large administrative database, a dashboard was created based on identifying cardiac surgery admissions using a large national administrative dataset, and mortality rates are provided at 30, 60, 90days, and 1year postoperatively [38]. Users can filter results by state, hospital, and individual provider and visualize summary data comparing these filtered results to national metrics. Moreover, using machine learning and the institution’s EHR, models can be deployed that predict operating room occupancy, efficient resource management, bed allocation, discharge planning, and staffing needs.

At the national and global level, evidence-based insights from large database research should be included in the most current clinical guidelines. The leaders of national and international societies and patient safety committees should continue to combine efforts and lead patient safety initiatives on a societal level [1,4^■■]. As databases and tracking of patients across the healthcare continuum continue to expand, protecting patient privacy should be a high priority [39^■]. Providing the regulatory framework for large database research would allow further expansion of the available data and the development of more accurate methods and tools, that, in the long run, will greatly enhance safety while protecting patient confidentiality.

FUTURE DIRECTIONS

The landscape for the development and implementation of patient safety initiatives based on insights from large databases looks very promising in the near future. Linking multiple databases and registries would allow gaining comprehensive information on large groups of patients and aid the development of highly accurate safety tools. In addition, as our abilities to collect and analyze large data expand, new insights about the perioperative period will be gained. For example, data from preoperative wearable electronics, when added to a surgical risk calculator, can improve the prediction of postoperative complications [40]. Harnessing the intraoperative physiologic data, including waveforms, will allow gaining further insights. Recently, a large vital signs repository of perioperative patient data became publicly available [41]. This dataset contains high-resolution multiparameter data from 6388 cases, including waveform and numeric data of 196 intraoperative monitoring parameters, 73 perioperative clinical parameters, and 34 time-series laboratory results. Using these types of data can generate highly accurate predictions.

For example, using high-resolution physiologic signals, the risk for adverse events such as hypoxemia, hypocapnia, and hypotension can be predicted more accurately than any currently existing methods [42]. In the near future, integrating models like this in real-time patient care would allow the anesthesiologists to intervene earlier and more efficiently to prevent adverse outcomes. In addition, large repositories of genomic data are becoming available. For example, AllOfUs is a program that started in 2018 with the goal of enrolling a diverse group of at least 1 million persons in the USA to accelerate biomedical research and improve the health [43]. In addition to genomic data, the elements of the program include health questionnaires, EHRs, physical measurements, the use of digital health technology, and the collection and analysis of biospecimens. Investigation of the genetic factors associated with adverse perioperative outcomes will allow personalizing patient care to achieve the best results. For example, patients with genetic variations of the cholinergic system are at a higher risk for postoperative delirium [44^■]. In this way, an anesthetic technique associated with lower risk for those patients could be selected. As genomic data becomes more available in clinical care, implementing those insights would allow further personalization of perioperative management.

CONCLUSION

Despite their shortcomings, large databases of clinical data can be useful for anesthesiologists seeking to improve patient safety and clinical outcomes. Many retrospective observational studies using big data have findings that are of great importance to anesthesiologists and could be clinically actionable to improve patient safety and outcomes in the perioperative setting. We expect that as datasets and methods to gain insights from those continue to grow, there will be a multitude of opportunities to expand patient safety research and implement novel technologies with the ultimate goal of significantly improving patient care.

KEY POINTS.

• Patient safety is at the center of anesthesiology practice; recent large database research provides new actionable insights to improve patient care.

• Implementing patient safety initiatives in clinical practice can be augmented by recent technological advances which provide automated outcome tracking on the individual provider, institutional, national, and international level.

• Opportunities to expand the available patient data by including wearable electronics, waveform data, and genomics hold great promise to provide personalized patient safety goals.