Expanding the All of Us Research Platform Into the Perioperative Domain

What Is the Innovation?

The All of Us Research Program is an extensive National Institutes of Health initiative to improve enrollment in prospective interventional studies.¹ Thus far, the All of Us Research Program has been underutilized by the surgical community. Novel algorithms must be developed if All of Us is to be extended from chronic² to perioperative conditions.

We present 2 tools (Figure) for studying surgical procedures within the All of Us Research Program. These tools are open source and available as Jupyter notebooks with R syntax for integration with the workflow in All of Us Research Program.³ We use these tools to curate a surgical cohort of postoperative acute kidney injury (AKI) and calculate the incidence within a variety of surgical populations, including noncardiac surgery, cardiac surgery, and open cardiac surgery.

Figure. Creation of Surgical Cohorts and Outcomes from the All of Us Research Program.

AKI indicates acute kidney injury; CPT-4, Current Procedural Terminology-4; ICD-10-PCS, International Classification of Diseases, Tenth Revision, Procedure Coding System.

What Are the Key Advantages Over Existing Approaches?

The genetics of perioperative complications have been studied within institutional biobanks; however, these studies were often under-powered and lacked participant diversity to draw conclusions beyond patients of European ancestry.^4,5 Since All of Us targets patients from groups that are historically underrepresented in biomedical research for enrollment, it provides a larger, more diverse population. Furthermore, All of Us includes data from whole genome sequencing, survey responses, physical measurements, wearable technologies, and electronic health records⁶ that could enable discoveries in the perioperative domain not possible using existing data sources.

We demonstrate this process using postoperative AKI, a high-significance complication with interest across clinical domains. Our present study provides an illustrative example identifying surgical cohorts and complications (Kidney Disease: Improving Global Outcomes AKI stage⁷) with novel, national genetic data. Further iterations can build on these tools to define surgical complications and identify the genetic associations that cannot be done using traditional data sources.

How Will This Affect Clinical Care?

Publishing the algorithms we present would democratize access to standardized cohorts and validated complications. This innovation could be applied to determine genetic predictors of postoperative complications, such as kidney injury, in a racially diverse general population. In addition, future studies using All of Us data sources could affect clinical care through improved perioperative risk stratification and clinical decision support tools. At the current time, this innovation is for research purposes only and does not affect clinical care.

Is There Evidence Supporting the Benefits of the Innovation?

Data on 134 831 surgical procedures (12 438 cardiac and 122 393 noncardiac) from 35 227 unique patients were available within All of Us (version 7; data cutoff, July 1, 2022). The mean (SD) patient age was 56 (16) years, and 75 706 participants (56%) were female; 1764 (1.3%) were Asian, 18 305 (13.6%) were Black or African American, 473 (0.4%) were Middle Eastern or North African, 231 (0.2%) were Native Hawaiian or Other Pacific Islander, 83 931 (62.2%) were White, and 1847 (1.4%) reported more than 1 race or ethnicity. Preliminary analysis of the All of Us dataset reveals a larger, more racially diverse population than previously studied in institutional biobanks or surgical registries.⁴

The incidence of AKI, by stage, within the noncardiac surgery population was no AKI, 93.8%; Kidney Disease Improving Global Outcomes (KDIGO) stage 1 AKI, 4.7%; stage 2, 1.2%; and stage 3, 0.3%. Patients with the classifications of no preoperative creatinine, preoperative kidney failure, no postoperative creatinine, and additional surgery with no serum creatinine drawn in-between were excluded from the incidence calculation, which was calculated only using the total number of patients with either no AKI, stage 1 AKI, stage 2 AKI, or stage 3 AKI.

The incidence within the cardiac surgery population was no AKI, 87.9%; stage 1, 10.2%; stage 2, 1.6%; and stage 3, 0.3%. The incidence within the open cardiac surgery population was no AKI, 86.5%; stage 1, 11.1%; stage 2, 1.9%; and stage 3, 0.3%. The incidence of postoperative AKI was notably lower in All of Us compared to cohorts from institutional biobanks, which may reflect an intrinsically lower- risk population. The tools were validated through limited clinician adjudication to confirm that the algorithm was classifying surgical procedures and AKI outcomes as expected.

What Are the Barriers to Implementing This Innovation More Broadly?

Unlike chronic conditions,² perioperative outcomes must account for temporal relationships during data-rich acute care episodes, creating unique algorithmic challenges. Tool 1 enables researchers to create a surgical cohort of interest (for example, open cardiac surgery or kidney transplant). Tool 2 codifies the associations between laboratory values and time of surgery to implement a standardized diagnostic criterion.⁷ Analogous algorithms could enable studies on a variety of standardized outcomes.

Outside of the algorithmic challenges, familiarity with the work environment and bioinformatic techniques for conducting genetic analysis are barriers to broad implementation. To overcome these barriers, the All of Us program offers support resources, including tutorials, seminars, and office hours. Researchers can conduct queries and analysis using either R or python from the Jupyter Notebook–based computing environment. Genotype is available in a variety of standard formats, including VariantDataset and Variant Call Format. There is no cost for registration and initial data access. Limitations of the database are the lack of granular information on surgery and anesthetic-specific characteristics and the inherently low-risk population. A limitation of the technique is that inconsistent data capture could lead to bias, which needs additional characterization in future studies.

In What Time Frame Will This Innovation Likely Be Applied Routinely?

The All of Us Research Program has already enrolled more than 400 000 patients and is starting to yield breakthroughs in chronic disease states.² The paucity of surgical research represents an opportunity to apply this innovation to a new domain. The tools presented provide a framework for studying surgical procedures within All of Us that can be implemented immediately.