A lesson in embracing uncertainty in early safety signals from the Tsepamo study.

OBJECTIVE

We investigated how randomness may have contributed to the apparent decline in observed risk of neural tube defects (NTDs) following in-utero dolutegravir (DTG) exposure. We aimed to describe statistical approaches to uncertainty using accessible language for nonstatistical audiences.

METHODS

We reanalyzed Tsepamo Study data using frequentist confidence intervals, repeated intervals accounting for group sequential monitoring, and Bayesian posterior and posterior predictive distributions. We estimated the probability of decision reversal using simulation.

RESULTS

The initial Tsepamo analysis reported a large difference in NTD risk between DTG and non-DTG exposures, with point estimates of 0.94 and 0.12%, respectively. This difference diminished with subsequent data, with updated estimates of 0.10% for DTG and 0.11% for non-DTG exposures. Our analyses showed the early finding was statistically compatible with a wide range of effect sizes, including no difference. Due to the large uncertainty in the first analysis, the probability of decision reversal was high under repeated testing frameworks.

CONCLUSION

Early safety signals may reflect statistical noise. Evaluating the range of confidence intervals and estimating decision reversal probabilities provide meaningful insight into early results. Formal frameworks for uncertainty should guide decisions about interim data reporting, especially when findings may influence clinical or regulatory action.