Leveraging Artificial Intelligence and Synthetic Data Derivatives for Spine Surgery Research

Global Spine Journal, 2023 · DOI: 10.1177/21925682221085535 · Published: January 1, 2023

Simple Explanation

Electronic health records (EHRs) could be very helpful for spine surgery research, but patient privacy and data ownership are big concerns. To get around these problems, researchers are trying out 'synthetic' data, which looks and acts like real data but doesn't actually belong to any real people. The researchers made synthetic datasets that mimicked real patient data from two kinds of spine fusion surgeries. They then tested if the synthetic data could accurately predict things like readmissions and complications after surgery. The study found that the synthetic data was very similar to the real data in terms of patient characteristics and how well it could predict surgery outcomes. This suggests synthetic data could be a good way to do more spine surgery research without risking patient privacy.

Study Duration

2010 to 2021

Participants

9,072 real and 9,088 synthetic cervical fusion patients; 12,111 real and 12,126 synthetic lumbar fusion patients

Evidence Level

Retrospective cohort study

Key Findings

1
Descriptive characteristics were nearly identical between the real and synthetic cervical fusion datasets, with only small but statistically significant differences in race and recent hospital/ED visits.
2
Models built using real and synthetic data showed similar discrimination in predicting readmission for both cervical and lumbar fusion patients, with c-statistics ranging from .66 to .71.
3
While there was substantial overlap in influential predictors between real and synthetic data models, some differences were noted, suggesting potential limitations in using synthetic data for identifying the impact of individual predictors.

Research Summary

This study evaluated the use of synthetic data derivatives for spine surgery research by comparing real and synthetic data from anterior cervical and posterior lumbar fusion patients. The results showed that synthetic data closely mirrored real data in terms of descriptive characteristics and predictive performance, suggesting its potential for various applications in spine surgery research. The authors conclude that synthetic data derivatives offer a novel approach for leveraging EHR analytics to support multicenter spine surgery research, with the potential to expand as structured EHR data and organizational buy-in increase.

Practical Implications

Expedited EHR Research

Synthetic data removes the need for IRB approval and data brokers, providing physicians with immediate access to conduct EHR queries.

Multicenter Dataset Creation

Synthetic datasets facilitate the creation and sharing of multicenter datasets supported by multi-institution partnerships, enabling broader research collaborations.

Expanded Research Applications

Synthetic data supports epidemiological analyses, studies of surgical trends, and profiling of quality outcome metrics, with potential for comparative effectiveness analyses as patient-reported outcomes are integrated into EHRs.

Study Limitations

1
Slight descriptive differences between real and synthetic data related to rare categorical variables may impact the accuracy of analyses focused on rare subgroups.
2
Differences in influential predictors between real and synthetic data models suggest that synthetic data may not be best-suited for analyses identifying the impact of individual predictors in complex multivariable analyses.
3
The proprietary nature of the MDClone platform limits the ability to investigate the detailed methods of its underlying algorithm.

Your Feedback

Was this summary helpful?

Back to Surgery