838P - Bayesian modeling in the survival analysis of patients with multiple myeloma with emphasis on missing data analysis

Background

In the survival prognosis of patients with multiple myeloma, there are staging models based on Cox survival models. Current staging models do not include patients with missing data, although there are multiple investigations where the problem of missing information is very common. This situation causes results that do not represent the reality of the populations, generating biases in the true results of the research and excluding patients both in training and in the application of the model to new patients.

Methods

To solve this problem, a survival model based on Bayesian statistics was built that not only allowed us to impute the missing data of patients with multiple Myeloma using statistical methods but also allowed us to predict new patients with missing data. This model was built and tuned using data from multiple myeloma patients from a public database from a study conducted in Arkansas.

Results

The results of the Bayesian model allowed us to estimate the survival of patients with missing information. Additionally, when compared with the classic Cox models in patients with complete data, the performance metrics did not deteriorate (concordance index of 72% and 71% respectively) in patients with complete data. In addition, it allows for estimating survival and staging the risk of patients with multiple myeloma with missing data, maintaining the concordance rate at 71%.

Conclusions

A model that performs as well as Cox models in patients with complete data and maintains this performance in patients with missing data is achieved. It also allows the prediction of new patients who are diagnosed with multiple myeloma without the restriction that their complete clinical or genetic data is needed.

Clinical trial identification

Editorial acknowledgement

Legal entity responsible for the study

The authors.

Funding

Has not received any funding.

Disclosure

All authors have declared no conflicts of interest.