Our ultimate goal is to build an end-to-end system that generates longitudinal synthetic health data which captures temporal relations of patient records along with their covariates, preserves utility as well as privacy, and is computationally efficient, scalable and free of cost. Therefore, building models for generating longitudinal synthetic data with such covariates could greatly facilitate healthcare research. In addition, covariates such as age, gender, and race are critical for retrospective observational studies. Due to it’s temporal nature, longitudinal data is conducive for causal analysis studies and in overall offers greater utility when compared to cross-sectional data. In real life, patient data consists of stream of in-patient and out-patient visits and other treatment events through time. Nonetheless, the data generated by these methods is not representative of real medical records, as they contain only one record per patient. have been used successfully to generate health data. ( 2018), Generative Adversarial Networks Choi et al.
Machine Learning models like Bayesian Networks Aviñó et al. Generating synthetic health data which maintains the utility of the data as well as preserves the privacy of the patients is a potential solution. To gain a deeper understanding of synthetic data quality. However, we also demonstrate how stratification by covariates is required Generated data and show close univariate resemblance between synthetic and realĭata. Sleep patterns, from a publicly available dataset. We demonstrate this approach by generating human We then train a generative adversarial network Visit is an event, we transform the data by using summary statistics toĬharacterize the events for a fixed set of time intervals, to facilitateĪnalysis and interpretability. Due to the complexity of the real data, in which each patient It becomes increasingly challenging to extend these methods to medical event
HEALTH DATA GENERATOR SERIES
While there exist seminal methods to model time series data, These events are influenced by patient covariates such asĬomorbidities, age group, gender etc. Patient having multiple health events, non- uniformly distributed throughout In reality, medical data is longitudinal in nature, with a single Generating cross-sectional health data which is not necessarily representative Synthetic medical data which preserves privacy while maintaining utility canīe used as an alternative to real medical data, which has privacy costs and
0 kommentar(er)
