Humans vs. machines
Humans are uniquely skilled at pattern recognition and inductive thinking. This innate skill provides for a significant survival advantage. However, most of us are incapable of explaining how we identify patterns. In health care, for example, we can identify a surgeon’s tug on the abdomen coinciding with a transient rise in heart rate and blood pressure as nociception requiring additional anesthetic agent administration. However, how do we explain our decision-making to the new resident in their first few days in the OR or our internal algorithm for deciding between opioid, volatile agent, and beta-blocker? Further, relying upon pattern recognition can increase risk for safety events. These tendencies toward error reveal opportunities to leverage new technologies to mitigate risk for human and system failures in health care.
Machine pattern recognition is not based on a blank slate. Training datasets are introduced to the computer, creating models to be validated, tested, and implemented. However, these models are prone to biases – whether due to connecting factors that do not share causation or selecting too many factors that share too much causation. Until computer models have a greater “understanding” of the world, they will be unable to independently create algorithms correctly identifying event causation.
As we head toward 2030, we will see an increasing number of hybrid platforms rapidly evaluating human-proposed processes with computer-evaluated outcomes. Some examples of this are already in action in major health systems in our country. For example, at Emory University, the risk of developing a deep vein thrombosis is now based on a simple five-point scoring tool, replacing the 44-item venous thromboembolism risk factor assessment described by Caprini. At UPMC, readmission risk is scored per patient encounter and made readily available within the electronic health record for all care providers to evaluate. And at the University of Texas MD Anderson Cancer Center, enhanced recovery service line teams utilize data to test hypotheses and refine practice based on outcomes. All three models were created using human intuition with a computer’s ability to process large amounts of data to quantify impact rapidly.
Databases and provider metrics
The American College of Surgeon’s National Surgical Quality Improvement Program database uses 21 patient-related variables to estimate the risk of 18 distinct patient outcomes. Data from more than 5 million operations in 855 hospitals undergo rigorous statistical analysis to create a predictive model that allows hospitals to compare their outcomes with other participating centers. Because some of the most consequential outcomes (e.g., stroke, death) are uncommon, the 90% confidence interval for a given complication at a particular facility frequently straddles several deciles on a given six-month report. This highlights one problematic area in quality: it is challenging to separate poor quality from random chance when it comes to rare outcomes. For these reasons, process metrics for individual providers will become increasingly important.
Improving individual provider metrics will enable benchmarking individual anesthesiologist performance to clinical outcomes. Setting these benchmarks is the first step to improving quality of care. It establishes performance expectations and outcomes standards. Currently, establishing these benchmarks for individual providers is challenged by lack of specificity; individual anesthesiologist practice is often influenced by more than clinician judgement – case times, surgeon requests, shiftwork relief, and handover issues all influence outcomes. Establishment of benchmarks is also challenged by opaque and clunky electronic health record systems, where data quality is not always ensured. Data and health record shareability and transparency is necessary to achieve these broader goals.
In January 2019, Seema Verma, the Administrator for the Centers for Medicare & Medicaid Services at the time, made headlines when she openly advocated for increased patient medical record transparency and portability. Today, patients can identify hospitals and providers that manage certain disease states well using tools such as “Care Compare” on the medicare.gov website. By 2030, algorithms will be created allowing a patient to upload their data to identify the hospitals and providers that will provide the best outcomes for all aspects of their individual health.
Quality improvement in 2030 will finally fully integrate toolsets such as Lean and Six Sigma into the workflow. Anesthesiology is in a nascent period of quality improvement and will need to refine the structure of their interventions to achieve hospital and health system quality goals. Toolsets like Lean and Six Sigma will finally be integrated into the quality improvement toolsets by 2030, increasing quality improvement efficiency. Lean and Six Sigma tenets are focused on creating value for consumers by eliminating waste and reducing defects by effectively solving problems. Lean Six Sigma uses the DMAIC Approach to achieve these goals (Figure 2). Even in the earliest stages of implementation, both Lean and Six Sigma have already demonstrated promising outcomes in anesthesiology by improving perioperative care quality while reducing costs. Applying these principles in anesthesiology quality in 2030 will require intentional alignment of efforts with the greater organizational goals and mission – those partnerships between anesthesiology teams and hospital/health system teams need to start now, in 2021.
Anesthesiologists as leaders in quality
Building these processes intentionally – provider benchmarks, hospital/service benchmarks, and process improvement – will eventually transform how we assess quality and ensure that providers have the skill set to deliver value. Ensuring clinical competence throughout one’s career has been challenging, particularly with the escalating scope of practice and knowledge needed to provide appropriate care. To address these challenges, credentialing boards have made major changes in the certification process; mandatory maintenance of certification has become the norm. In addition, continuing medical education requirements will expand over the next decade. Simulation programs will be geared toward improving systemic weaknesses both within hospital and practice structures and across them. Payments to hospitals and practices are likely to incentivize simulation for maintenance of certification or similar quality-based CME activities by 2030, thereby hardwiring quality process improvements within health care systems.
A major focus for addressing quality of care has been the implementation by payors of pay-for-performance plans (P4P). Fortunately for anesthesiologists, the P4P plans will no longer be the norm. There are many reasons why we are experiencing a failure of P4P plans – lack of outcomes metrics for certain specialties, increasing levels of fraud, irreconcilable conflicts of interest, and the realization that incentivizing outcomes worsens health care disparities based on patient race and socioeconomic status. Outcome metrics in anesthesiology are highly problematic. For example, should an anesthesiologist be judged more harshly because a surgeon’s personal preferences result in “excess transfusion?” By tying increased income to the provision of care, regulators incentivize the documentation of the care being given regardless of the appropriateness of that care. Finally, the insurer has an irreconcilable conflict of interest since they cannot simultaneously increase payments to provide better patient care while also maintaining the same rate of return to shareholders.
As we look toward 2030, quality initiatives will have achieved the “Plateau of Productivity.” Physicians finally feel they are receiving the guidance they need to provide truly individualized care to patients, while patients will also have the guidance information necessary to be genuinely informed consumers. The consolidation of medical data into one universal electronic medical record platform will allow for more in-depth analysis of factors important to positive patient outcomes. As the lyrics go for Walt Disney’s Carousel of Progress, “There’s a great big, beautiful tomorrow, shining at the end of every day.”