Covariate adjusted meta-analytic predictive (CA-MAP) prior for historical borrowing using patient-level data.

Utilization of historical data is increasingly common for gaining efficiency in the drug development and decision-making processes. The underlying issue of between-trial heterogeneity in clinical trials is a barrier in making these methods standard practice in the pharmaceutical industry. Common methods for historical borrowing discount the borrowed information based on the similarity between outcomes in the historical and current data. However, individual clinical trials and their outcomes are intrinsically heterogenous due to differences in study design, patient characteristics, and changes in standard of care. Additionally, differences in covariate distributions can produce inconsistencies in clinical outcome data between historical and current data when there may be a consistent covariate effect. In such scenario, borrowing historical data is still advantageous even though the population level outcome summaries are different. In this paper, we propose a covariate adjusted meta-analytic-predictive (CA-MAP) prior for historical control borrowing. A MAP prior is assigned to each covariate effect, allowing the amount of borrowing to be determined by the consistency of the covariate effects across the current and historical data. This approach integrates between-trial heterogeneity with covariate level heterogeneity to tune the amount of information borrowed. Our method is unique as it directly models the covariate effects instead of using the covariates to select a similar population to borrow from. In summary, our proposed patient-level extension of the MAP prior allows for the amount of historical control borrowing to depend on the similarity of covariate effects rather than similarity in clinical outcomes.

Bayesian Additive Regression Trees (BART) with covariate adjusted borrowing in subgroup analyses.

It is crucial in clinical trials to investigate treatment effect consistency across subgroups defined by patient baseline characteristics. However, there may be treatment effect variability across subgroups due to small subgroup sample size. Various Bayesian models have been proposed to incorporate this variability when borrowing information across subgroups. These models rely on the underlying assumption that patients with similar characteristics will have similar outcomes to the same treatment. Patient populations within each subgroup must subjectively be deemed similar enough Pocock (1976) to borrow response information across subgroups. We propose utilizing the machine learning method of Bayesian Additive Regression Trees (BART) to provide a method for subgroup borrowing that does not rely on an underlying assumption of homogeneity between subgroups. BART is a data-driven approach that utilizes patient-level observations. The amount of borrowing between subgroups automatically adjusts as BART learns the covariate-response relationships. Modeling patient-level data rather than treating the subgroup as a single unit minimizes assumptions regarding homogeneity across subgroups. We illustrate the use of BART in this context by comparing performance from existing subgroup borrowing methods in a simulation study and a case study in non-small cell lung cancer. The application of BART in the context of subgroup analyses alleviates the need to subjectively choose how much information to borrow based on subgroup similarity. Having the amount of borrowing be analytically determined and controlled for based on the similarity of individual patient-level characteristics allows for more objective decision making in the drug development process with many other applications including basket trials.

Bayesian semiparametric meta-analytic-predictive prior for historical control borrowing in clinical trials.

When designing a clinical trial, borrowing historical control information can provide a more efficient approach by reducing the necessary control arm sample size while still yielding increased power. Several Bayesian methods for incorporating historical information via a prior distribution have been proposed, for example, (modified) power prior, (robust) meta-analytic predictive prior. When utilizing historical control borrowing, the prior parameter(s) must be specified to determine the magnitude of borrowing before the current data are observed. Thus, a flexible prior is needed in case of heterogeneity between historic trials or prior data conflict with the current trial. To incorporate the ability to selectively borrow historic information, we propose a Bayesian semiparametric meta-analytic-predictive prior. Using a Dirichlet process mixture prior allows for relaxation of parametric assumptions, and lets the model adaptively learn the relationship between the historic and current control data. Additionally, we generalize a method for estimating the prior effective sample size (ESS) for the proposed prior. This gives an intuitive quantification of the amount of information borrowed from historical trials, and aids in tuning the prior to the specific task at hand. We illustrate the effectiveness of the proposed methodology by comparing performance between existing methods in an extensive simulation study and a phase II proof-of-concept trial in ankylosing spondylitis. In summary, our proposed robustification of the meta-analytic-predictive prior alleviates the need for prespecifying the amount of borrowing, providing a more flexible and robust method to integrate historical data from multiple study sources in the design and analysis of clinical trials.

Propensity-score-based meta-analytic predictive prior for incorporating real-world and historical data.

As the availability of real-world data sources (eg, EHRs, claims data, registries) and historical data has rapidly surged in recent years, there is an increasing interest and need from investigators and health authorities to leverage all available information to reduce patient burden and accelerate both drug development and regulatory decision making. Bayesian meta-analytic approaches are a popular historical borrowing method that has been developed to leverage such data using robust hierarchical models. The model structure accounts for various degrees of between-trial heterogeneity, resulting in adaptively discounting the external information in the case of data conflict. In this article, we propose to integrate the propensity score method and Bayesian meta-analytic-predictive (MAP) prior to leverage external real-world and historical data. The propensity score methodology is applied to select a subset of patients from external data that are similar to those in the current study with regards to key baseline covariates and to stratify the selected patients together with those in the current study into more homogeneous strata. The MAP prior approach is used to obtain stratum-specific MAP prior and derive the overall propensity score integrated meta-analytic predictive (PS-MAP) prior. Additionally, we allow for tuning the prior effective sample size for the proposed PS-MAP prior, which quantifies the amount of information borrowed from external data. We evaluate the performance of the proposed PS-MAP prior by comparing it to the existing propensity score-integrated power prior approach in a simulation study and illustrate its implementation with an example of a single-arm phase II trial.

Application of Machine Learning Techniques to High-Dimensional Clinical Data to Forecast Postoperative Complications.

OBJECTIVE: To compare performance of risk prediction models for forecasting postoperative sepsis and acute kidney injury. DESIGN: Retrospective single center cohort study of adult surgical patients admitted between 2000 and 2010. PATIENTS: 50,318 adult patients undergoing major surgery. MEASUREMENTS: We evaluated the performance of logistic regression, generalized additive models, naïve Bayes and support vector machines for forecasting postoperative sepsis and acute kidney injury. We assessed the impact of feature reduction techniques on predictive performance. Model performance was determined using the area under the receiver operating characteristic curve, accuracy, and positive predicted value. The results were reported based on a 70/30 cross validation procedure where the data were randomly split into 70% used for training the model and the 30% for validation. MAIN RESULTS: The areas under the receiver operating characteristic curve for different models ranged between 0.797 and 0.858 for acute kidney injury and between 0.757 and 0.909 for severe sepsis. Logistic regression, generalized additive model, and support vector machines had better performance compared to Naïve Bayes model. Generalized additive models additionally accounted for non-linearity of continuous clinical variables as depicted in their risk patterns plots. Reducing the input feature space with LASSO had minimal effect on prediction performance, while feature extraction using principal component analysis improved performance of the models. CONCLUSIONS: Generalized additive models and support vector machines had good performance as risk prediction model for postoperative sepsis and AKI. Feature extraction using principal component analysis improved the predictive performance of all models.