Bayesian Meta-Analyses Could Be More: A Case Study in Trial of Labor After a Cesarean-section Outcomes and Complications

TL;DR

The paper addresses bias in medical meta-analyses when a key decision variable is unmeasured, using TOLAC and Bishop score as a case study. It develops a Bayesian hierarchical meta-analysis that treats the Bishop score as a hidden mediator with provider-discretion thresholds, informed by clinician priors, and compares it to standard fixed-effects. In 16 studies with 7049 patients, the Bayes estimate for the cesarean risk ratio is $RR = 1.04$ (95% CI $[0.93, 1.18]$), indicating no difference after accounting for the mediator, whereas a naive fixed-effect analysis suggested harm. The approach demonstrates how incorporating a known mediator can correct biased inferences, guiding clinical decisions and trial design, and underscores the need for reporting Bishop scores in primary studies.

Abstract

The meta-analysis's utility is dependent on previous studies having accurately captured the variables of interest, but in medical studies, a key decision variable that impacts a physician's decisions was not captured. This results in an unknown effect size and unreliable conclusions. A Bayesian approach may allow analysis to determine if the claim of a positive effect is still warranted, and we build a Bayesian approach to this common medical scenario. To demonstrate its utility, we assist professional OBGYNs in evaluating Trial of Labor After a Cesarean-section (TOLAC) situations where few interventions are available for patients and find the support needed for physicians to advance patient care.

Figures (7)

• Figure 1: Cervical ripening is achieved with the application of Pitocin or mechanical dilation (a catheter). Whether one method has a higher rate of successful vaginal delivery after a C-section is of clinical relevance to the decision, as a failed TOLAC has higher rates of complications than a scheduled repeat C-section. The choice is made by computing a Bishop Score in clinical practice but creates a shared dependency (i.e., in the same Markov blanket), and this factor was ignored in previous studies. This creates an unaccounted causal impact between the two groups.
• Figure 2: The Plate Diagram for our model. Grey nodes are observed variables, white are sampled, and dashed lines are subgraphs that repeat together. The key observation is that there is a bifurcation of variables modeling the outcome given a mediating decision variable (in this study, a Bishop score) for the control (Pitocin) and intervention (catheter). This shows how our algorithm accounts for a single mediator (Bishop score) and threshold (outside the num_samples plate) that is altered by a provider's discretion $\delta_i$ and then causes the two groups to have different underlying Bishop scores.
• Figure 3: Forest Plot of mechanical dilation (intervention) vs. Pitocin when ignoring the unaccounted Bishop score as outlined in Fig. \ref{['fig:bishopWhatIs']}. Because a Fixed-Effect meta-analysis fails to account for this mediating variable, and fails assumption tests, we do not recommend using its conclusion. It is presented to show why a Bayesian approach with a hidden variable for the Bishop score is needed.
• Figure 4: An improved meta-analysis that accounts for the hidden variable of the Bishop score, which is used to determine whether mechanical dilation (intervention) or Pitocin (control) is used. This shows that current evidence is not sufficient to conclude there is any difference in the outcome of C-section rates.
• Figure 5: A fixed-effect model is used due to the lack of a causal relationship between Bishop scores and adverse uterine events. The result is included for complete clinical relevance of our results. No significant effect between Pitocin and mechanical dilation is detected.