Informatics describes the study and use of processes for obtaining and utilizing data. In the clinical context, these data are then used to inform and educate providers to improve patient care. In the current digital age, informatic solutions can help clinicians to understand past or current quality issues (afferent tools), to benchmark personal performance against national averages (feedback tools), and to disseminate information to encourage best practice and quality care (efferent tools). There are countless examples of how these tools can be adapted for use in obstetric anesthesia, with evidence to support their implementation. This article thus aimed to summarize the many ways in which informatics can help clinicians to harness the power of data to improve quality and safety in obstetric anesthesia.
Data are powerful, but only when appropriately captured, cleaned, and interpreted. Given context and meaning, data can be used to both learn from and evaluate past patient encounters and, increasingly in real time, inform future decisions. Each of these aspects is important to obstetric anesthesiologists as they pave the path forward for our subspecialty and the patients we care for.
This article aimed to summarize how informatics and data can assist obstetric anesthesia clinicians in the pursuit of quality and safety goals. These include assimilating data to understand past or current performance (afferent tools), advising users on personal metrics and benchmarking (feedback tools), and disseminating information to encourage best practice and quality care (efferent tools) (Figure 1 and Table 1).
Ongoing collection of clinical and quality data is essential to inform efforts to improve patient care. There are numerous sources of useful data including clinical data from the electronic medical record (EMR), billing data, patient-reported data (eg, patient surveys), and non-EMR provider-reported data (eg, adverse event reporting), among others. Depending on the concerns or questions being addressed, 1 or all of these sources may prove valuable.
Data accuracy should be a primary concern for those accessing it, and a preliminary understanding of clinical data standards, terminologies, and limitations of particular systems should underpin any quality and safety initiatives. Without appropriate data standards, health data cannot be compared between providers or institutions, interchanged between systems, or used secondarily for purposes such as research. Quality efforts can be thwarted by limitations in data collection methods, as evidenced by the lack of International Classification of Diseases codes for anesthetic interventions or complications in the United Kingdom.
Given the quantity of data available, choosing which metrics to monitor can be challenging. Within obstetrics, the Alliance for Innovation on Maternal Health (AIM) has proposed data collection strategies to accompany each of its comorbidity-specific patient safety bundles. Bundles have been developed for obstetric hemorrhage, severe hypertension, cardiac conditions, and sepsis, among others. Each bundle contains 3 types of measures: structure measures assess systems and protocols, process measures assess evidence-based practices, and outcome measures assess patient-specific results (Table 2). For example, the AIM Obstetric Hemorrhage Bundle includes the following measures: presence of a hemorrhage cart (structure), standardized protocols for event debriefs (structure), hemorrhage risk assessment (process), quantified blood loss (process), and severe maternal morbidity (outcome). Through standardized collection and reporting of these measures, participating centers can develop data-driven goals for obstetric hemorrhage management and monitor progress, success, and sustainability. With 8 available bundles and 13 to 19 measures per bundle, automated informatics solutions are necessary to successfully capture, visualize, and manage the appropriate metrics and to ultimately translate that surveillance into improved maternal outcomes.
While many of the AIM metrics listed above exemplify EMR-based data to monitor maternal morbidity, there is also value in capturing provider-reported adverse events related to anesthetic care delivery. One example of a commercially available tool for such reporting is the Anesthesia Quality Institute's Anesthesia Incident Reporting System (AIRS), an online reporting system that collects adverse events related to anesthesia, pain management, and perioperative care. Custom adverse event reporting tools can also be developed by individual institutions and have the benefit of offering more customizable subspecialty-specific measures. An obstetric anesthesia tool, for example, allows users to report obstetric-specific events such as general anesthesia for cesarean delivery or epidural catheter replacement during labor--events that can be missed by more generic reporting tools. Such reporting tools allow for real-time capture and response to adverse events but also facilitate tracking event rates over time.
Data capture for monitoring or reporting of certain quality metrics is also required by regulatory bodies and merit-based government incentive programs. For example, The Joint Commission R Report on Standards for Maternal Safety requires review of severe hypertension cases to evaluate effectiveness of treatment. Such review is facilitated by electronic data capture and informatics solutions. Additionally, Centers for Medicare and Medicaid Services (CMS) uses multiple value-based programs that reward clinicians and hospitals with financial incentives for demonstration of quality care. The CMS Quality Payment Program includes multiple reporting options: Traditional Merit-Based Incentive Payment System (MIPS), Alternative Payment Model Performance Pathways, and MIPS Value Pathways. While the details of each are beyond the scope of this article, all require reporting on quality performance, ongoing quality improvement activities, and promoting interoperability. These facets are supported by robust internal data reporting structures as well as investment in internal informatics infrastructure and resources.
Data Visualization
While programs like AIM are designed for large-scale monitoring of major quality initiatives around comorbidities like cardiac disease or hemorrhage, there are countless other outcomes or metrics that anesthesia providers may want to visualize longitudinally. Understanding that individual clinicians and institutions value access to this type of data, most commercial EMRs now include data visualization tools to allow front-line clinicians to access and display relevant patient outcomes without extensive software training requirements. Examples of such tools include Oracle's Cerner and PowerInsight Explorer and Epic Systems' Reporting Workbench and SlicerDicer. These tools provide impactful information to clinicians rapidly, to support quality improvement initiatives. Users can select base populations to explore, determine quality metrics of interest to measure, and investigate values and trends of these metrics over time. Most platforms have additional filters and refining tools to assist the user in curating a search that is applicable to their specific quality question. As an example, health equity efforts can be supported and monitored through custom reports designed to display quality outcomes, such as general anesthesia for cesarean delivery, by race, ethnicity, or primary language spoken (Figure 2).
More advanced visual analytics platforms, such as Tableau, provide increasingly sophisticated and customized data visualization but require additional upfront investment in the form of software licenses and trained analysts. These tools, however, can provide powerful information to both clinicians and hospital administrators in monitoring clinical care and supporting practice changes.
Quality Research
Finally, in addition to monitoring data at a single institution, informatics can also support assimilation of vast quantities of multicenter data in national databases. These databases collect, maintain, and organize data both for quality reporting and also subsequent query and research. Each database captures different populations and outcomes, and each has its own strengths and limitations. Examples include the Healthcare Cost and Utilization Project's National Inpatient Sample (NIS), National Anesthesia Clinical Outcomes Registry, National Surgical Quality Improvement Program, and Multicenter Perioperative Outcomes Group (MPOG) (Table 3).
These large national data repositories are essential components of obstetric quality research. They provide valuable data to study rare patient outcomes and variability in those outcomes in different hospitals and patient populations. For example, using the NIS, investigators identified significantly higher rates of complicated vaginal and cesarean delivery, blood transfusion, puerperal infection, hysterectomy, vascular complications, thromboembolism, and death at Black-serving hospitals (defined as hospitals with >50% deliveries to non-Hispanic Black women) compared to White-serving hospitals, and also among Black patients overall compared to non-Hispanic White patients. In another recent study, authors attempted to determine variation in obstetric care quality among high- and low-performing US hospitals. They demonstrated that women delivering in low-performing hospitals via vaginal or cesarean delivery experienced a major complication twice as often and 5 times more often, respectively, compared to high-performing hospitals.
With mounting evidence demonstrating significant variation in outcomes across the country, obstetric care quality metrics are desperately needed to track the measurement of disparities in care, which may be contributing to this variation. Regulatory bodies have begun to define obstetric quality metrics, although their usefulness and association with improved outcomes remain unclear. The Joint Commission's standardized performance measures include nonmedically indicated deliveries before 39 weeks' gestation and cesarean deliveries performed in low-risk nulliparous women. However, a previous study showed no correlation between these indicators and maternal or neonatal morbidity.
Thus, future obstetric anesthesia research should focus on development of useful quality performance measures. Glance et al recently proposed metrics assessing severe maternal morbidity, severe neonatal morbidity, and a composite of maternal and neonatal morbidity calculated from administrative and birth certificate data and risk-adjusted for hospital case mix. This and other potential metrics should be assessed to determine the most promising means of assessing and achieving quality and equity.
FEEDBACK SOLUTIONS: BRINGING PERSONALIZED DATA BACK TO USERS
An additional way that informatics tools can improve quality of care is by providing clinicians with real-time feedback of their own clinical performance and their patient's outcomes. The effectiveness of these strategies has been demonstrated in multiple perioperative settings, including for the timely administration of antibiotic prophylaxis in the operating room. After a quality improvement initiative that included biweekly emails to clinicians denoting performance on this quality measure, compliance increased from 69% to 92%.
Provider-specific feedback can further be supplemented with benchmarking to top performers or accepted thresholds. While specific studies demonstrating the benefit of benchmarking in obstetric anesthesia are lacking, numerous publications support the use of "relative social ranking" as an effective means of changing physician behavior. For example, in a randomized controlled trial of physician-specific feedback versus benchmark feedback in primary care, benchmarking significantly increased the odds of compliance with preventive care measures such as vaccination, diabetic foot examinations, and long-term glucose measurements. In the context of obstetric quality improvement, this type of intervention could be used to study a host of easily measurable and reportable process and patient outcomes including antibiotic timing, hypotension, hypothermia, and need for general anesthesia during cesarean delivery. MPOG, mentioned above in the context of their research efforts, also provides individual provider feedback on site-selected quality measures, including those suggested above. Other general measures available for review include acute kidney injury, fluid administration, glucose management, medication overdose, myocardial injury, neuromuscular monitoring, pain management, postoperative nausea and vomiting, pulmonary adverse events, hypothermia, and transfusion.
In addition to benchmarking, comprehensive capture of patient satisfaction with their obstetric anesthesia care undoubtedly improves quality. Informatics systems for survey of patient satisfaction can be customized to include quality metrics of importance to obstetric anesthesia providers. Recent interest in capturing quantitative and qualitative patient-reported data on the experience of intraoperative pain during cesarean delivery can be accomplished via modification of existing postpartum assessment tools to include these questions via a comprehensive and standardized approach, facilitating multicenter quality improvement and research initiatives.
EFFERENT SOLUTIONS: BUILDING INFORMATICS SYSTEMS TO SUPPORT QUALITY
Clinical Decision Support
Capturing and organizing data is only 1 way in which informatics can support quality improvement. Informatics tools that influence and advise clinician behavior have the potential to impact quality to an even greater extent. These interventions are widely termed clinical decision support (CDS). CDS can take many forms, including alerts or reminders to a clinician to perform a specific action, salient and organized display of pertinent patient information (ie, dashboards), diagnostic support, and operationalization of clinical guidelines. Nearly any clinical process can be supported by CDS, including quality initiatives aimed at prevention (preventing an adverse event from ever occurring), identification (identifying patients or populations at risk of an adverse event), or action (expediting appropriate action once an adverse event has occurred).
Alerts and reminders are the most popular form of CDS, and there are numerous published examples of positive changes in clinician behavior based on these types of alerts. An automatic intraoperative reminder system was developed to identify diabetic patients, capture previous glucose measurements and insulin administration, and display an alert to clinicians when glucose check was recommended. After implementation, appropriate intraoperative glucose monitoring increased from 62% to 87%, and recovery room hyperglycemia decreased from 11% to 7%. In another example, an automated point-of-care electronic prompt to encourage antibiotic administration within 60 minutes of surgical incision increased compliance with this metric from 62% to 92%. Surgical site infection also decreased from 1.1% to 0.7%.
Dashboards allow for a more detailed and comprehensive display of situation-specific information but require user engagement for maximum impact. For example, an obstetric transfusion dashboard can provide information on blood type, presence of antibodies, date and time of pretransfusion testing expiration, time required for additional pretransfusion testing, and links to literature-supported recommendations. With appropriate engagement, this type of CDS can assist clinicians in rapidly gathering and interpreting enormous quantities of clinical information at once. This type of salient display of information has been previously shown to improve efficiency and data utilization in the intensive care unit setting and holds promise for similar results in obstetric anesthesia.
Diagnostic support represents a more sophisticated form of CDS, one in which data are not only collected and displayed but also interpreted for the user based on training from previous occurrences. Automated maternal early warning systems, which capture clinical information and suggest patient deterioration, exemplify this type of tool. Commonly, this type of system will be accompanied by an alert, as described above, to ensure that clinicians are aware of the suggested diagnosis.
Finally, CDS can accelerate the adoption of best practice guidelines through EMR integration and operationalization. For example, the Society for Obstetric Anesthesia and Perinatology recommends regular assessment of laboring patients with epidural catheters. A recent study demonstrated that an EMR-based timer that displayed the number of minutes since the last patient assessment improved compliance with this best practice recommendation, decreasing mean time between assessments from a median of 173 to 100 minutes and maximum time between assessments from a median of 330 to 162 minutes. Epidural replacement rates also decreased from 14% to 5%, possibly due to more effective neuraxial management.
Artificial Intelligence
The next frontier of informatics in medicine includes artificial intelligence (AI) applications to improve the quality and safety of daily clinical care. AI is the science of producing computers or machines capable of learning, problem-solving, or achieving goals. In health care, this translates to a computer's ability to analyze large volumes of health care data, reveal knowledge, and support decision-making. Exponential growth of medical data over recent decades now necessitates the adoption of tools with the power to accumulate data from separate sources, synthesize and classify data, give data meaning, and inform subsequent clinician decision-making. CDS tools with clinician-facing alerts, including those described above, represent one form of AI. However, in more recent years, more sophisticated AI methods have been deployed to assist in clinical prediction and diagnosis. In the setting of countless data points for each individual patient, including vital signs, medications, laboratory values, diagnoses, and previous history, predictive models can help to collate and provide summative knowledge on which clinicians can base medical decisions. Within the field of obstetrics, early efforts have focused on the prediction of postpartum hemorrhage, preeclampsia, and severe maternal and neonatal morbidity.
Machine learning techniques are a newer iteration of AI in which computers "learn" to understand data and patterns via training datasets. The computer's learning can be supervised, using labeled data, or unsupervised, using mathematical processes to classify what data may represent. Logistic regression predictive models, such as those described above, are one of the simplest examples of supervised machine learning, in which the relationship between each risk factor and the outcome is assessed mathematically and the final calculation, using basic arithmetic operators, produces a probability of the outcome.
More advanced versions of machine learning include neural networks and deep learning and perform increasing numbers of mathematical operations on multiple layers of data to better define complex relationships. Multiple previous publications have demonstrated the clinical applicability of these new techniques in areas applicable to the obstetric population, including for the prediction of in-hospital mortality and postinduction hypotension. The majority of study, however, has focused on complex radiographic and image classification, such as for computerized detection of diabetic retinopathy from fundal images. An exciting future application of this technology lies in the interpretation of fetal heart rate tracings and prediction of hypoxic-ischemic encephalopathy or other neonatal morbidities, with multiple recent studies investigating this area. Riveros-Perez et al also used machine learning approaches to demonstrate that the use of combined spinal-epidural and total dose of bupivacaine was predictive of decreases in fetal heart rate during labor with neuraxial analgesia. As more applications of machine learning are proposed and studied, clinicians will need to familiarize themselves with the techniques and limitations of machine learning, including the appropriate validation and regularization methods required.
Until recently, AI tools have been limited to the inclusion of discrete data elements; however, natural language processing (NLP), which allows computers to discern meaning from human-authored free text, will soon allow clinicians to exploit even greater amounts of data. A recent study demonstrated that NLP of obstetric hospital admission notes was equally successful in predicting severe maternal morbidity compared to a validated obstetric comorbidity index. In a separate context, NLP has been shown to improve the accuracy of preanesthetic evaluations, where 17% of cases evaluated included relevant medical conditions identified by NLP but not manual anesthesiologist chart review. Only 2% of cases included conditions identified by clinicians but missed by NLP. These new technologies are rapidly evolving and will undoubtedly shape the future of obstetric anesthesia in years to come.
CONCLUSIONS
In the new digital age of health care, informatics solutions are essential to support quality improvement initiatives. At a minimum, providers and institutions must regularly monitor performance and patient outcomes to ensure basic standards are met and identify opportunities for improvement. Providing individualized feedback to clinicians on their own adherence to best practices and the outcomes of their patients can spur further progress. CDS and AI offer additional opportunities to leverage informatics to elevate patient care.
Contribution: This author helped in study conception, article writing, and editing.