Flexibility of a large blindly synthetized avatar database for occupational research: Example from the CONSTANCES cohort for stroke and knee pain.

OBJECTIVES: Though the rise of big data in the field of occupational health offers new opportunities especially for cross-cutting research, they raise the issue of privacy and security of data, especially when linking sensitive data from the field of insurance, occupational health or compensation claims. We aimed to validate a large, blinded synthesized database developed from the CONSTANCES cohort by comparing associations between three independently selected outcomes, and various exposures. METHODS: From the CONSTANCES cohort, a large synthetic dataset was constructed using the avatar method (Octopize) that is agnostic to the data primary or secondary data uses. Three main analyses of interest were chosen to compare associations between the raw and avatar dataset: risk of stroke (any stroke, and subtypes of stroke), risk of knee pain and limitations associated with knee pain. Logistic models were computed, and a qualitative comparison of paired odds ratio (OR) was made. RESULTS: Both raw and avatar datasets included 162,434 observations and 19 relevant variables. On the 172 paired raw/avatar OR that were computed, including stratified analyses on sex, more than 77% of the comparisons had a OR difference ≤0.5 and less than 7% had a discrepancy in the statistical significance of the associations, with a Cohen's Kappa coefficient of 0.80. CONCLUSIONS: This study shows the flexibility and the multiple usage of a synthetic database created with the avatar method in the particular field of occupational health, which can be shared in open access without risking re-identification and privacy issues and help bring new insights for complex phenomenon like return to work.

Patient-centric synthetic data generation, no reason to risk re-identification in biomedical data analysis.

While nearly all computational methods operate on pseudonymized personal data, re-identification remains a risk. With personal health data, this re-identification risk may be considered a double-crossing of patients' trust. Herein, we present a new method to generate synthetic data of individual granularity while holding on to patients' privacy. Developed for sensitive biomedical data, the method is patient-centric as it uses a local model to generate random new synthetic data, called an "avatar data", for each initial sensitive individual. This method, compared with 2 other synthetic data generation techniques (Synthpop, CT-GAN), is applied to real health data with a clinical trial and a cancer observational study to evaluate the protection it provides while retaining the original statistical information. Compared to Synthpop and CT-GAN, the Avatar method shows a similar level of signal maintenance while allowing to compute additional privacy metrics. In the light of distance-based privacy metrics, each individual produces an avatar simulation that is on average indistinguishable from 12 other generated avatar simulations for the clinical trial and 24 for the observational study. Data transformation using the Avatar method both preserves, the evaluation of the treatment's effectiveness with similar hazard ratios for the clinical trial (original HR = 0.49 [95% CI, 0.39-0.63] vs. avatar HR = 0.40 [95% CI, 0.31-0.52]) and the classification properties for the observational study (original AUC = 99.46 (s.e. 0.25) vs. avatar AUC = 99.84 (s.e. 0.12)). Once validated by privacy metrics, anonymous synthetic data enable the creation of value from sensitive pseudonymized data analyses by tackling the risk of a privacy breach.