Commensurate Priors on a Finite Mixture Model for Incorporating Repository Data in Clinical Trials

DHA supplementation for early preterm birth prevention: An application of Bayesian finite mixture models to adaptive clinical trial design optimization

BACKGROUND

Early preterm birth (ePTB) - born before 34 weeks of gestation - poses a significant public health challenge. Two randomized trials indicated an ePTB reduction among pregnant women receiving high-dose docosahexaenoic acid (DHA) supplementation. One of them is Assessment of DHA on Reducing Early Preterm Birth (ADORE). A survey employed in its secondary analysis identified women with low DHA levels, revealing that they derived greater benefits from high-dose DHA supplementation. This survey's inclusion in future trials can provide critical insights for informing clinical practices.

OBJECTIVE

To optimize a Phase III trial design, ADORE Precision, aiming at assessing DHA supplement (200 vs. 1000 mg/day) on reducing ePTB among pregnant women with a low baseline DHA.

METHODS

We propose a Bayesian Hybrid Response Adaptive Randomization (RAR) Design utilizing a finite mixture model to characterize gestational age at birth. Subsequently, a dichotomized ePTB outcome is used to inform trial design using RAR. Simulation studies were conducted to compare a Fixed Design, an Adaptive Design with early stopping, an ADORE-like Adaptive RAR Design, and two new Hybrid Designs with different hyperpriors.

DISCUSSION

Simulation reveals several advantages of the RAR designs, such as higher allocation to the more promising dose and a trial duration reduction. The proposed Hybrid RAR Designs addresses the statistical power drop observed in Adaptive RAR. The new design model shows robustness to hyperprior choices. We recommend Hybrid RAR Design 1 for ADORE Precision, anticipating that it will yield precise determinations, which is crucial for advancing our understanding in this field.

Empirical bayes approach for dynamic bayesian borrowing for clinical trials in rare diseases

Application of Bayesian methods is one the tools that can be used to face the multiple challenges that are met when clinical trials must be conducted in rare diseases. We propose in this work to use a dynamic Bayesian borrowing approach, based on a mixture prior, to complement the control arm of a comparative trial and estimate the mixture parameter by an Empirical Bayes approach. The method is compared, using simulations, with an approach based on a pre-specified (non-adaptive) informative prior. The simulation study shows that the proposed method exhibits similar power as the non-adaptive prior and drastically reduce type I error in case of severe discrepancy between the informative prior and the study control arm data. In case of limited discrepancy between the informative prior and the study control arm data, then our proposed adaptive prior does not reduce the inflation of the type I error.

Early and late preterm birth rates in participants adherent to randomly assigned high dose docosahexaenoic acid (DHA) supplementation in pregnancy

BACKGROUND

Intention-to-treat analyses do not address adherence. Per protocol analyses treat nonadherence as a protocol deviation and assess if the intervention is effective if followed.

OBJECTIVE

To determine the rate of early preterm birth (EPTB, <34 weeks gestation) and preterm birth (PTB, <37 weeks gestation) in participants who adhered to a randomly assigned docosahexaenoic acid (DHA) dose of 1000 mg/day.

STUDY DESIGN

Eleven hundred women with a singleton pregnancy were enrolled before 20-weeks' gestation, provided a capsule with 200 mg/day DHA and randomly assigned to two additional capsules containing a placebo or 800 mg of DHA. In the Bayesian Adaptive Design, new randomization schedules were determined at prespecified intervals. In each randomization, the group with the most EPTB was assigned fewer participants than the other group. Adherence was defined a priori as a postpartum red blood cell phospholipid DHA (RBC-PL-DHA) ≥5.5%.and post hoc as ≥8.0% RBC-PL-DHA, the latter after examination of postpartum RBC-PL-DHA. Bayesian mixture models were fitted for gestational age and dichotomized for EPTB and PTB as a function of baseline RBC-PL-DHA and dose-adherence. Bayesian hierarchical models were also fitted for EPTB by dose adherence and quartiles of baseline RBC-PL-DHA.

RESULTS

Adherence to the high dose using both RBC-PL-DHA cut points resulted in less EPTB compared to 200 mg [Bayesian posterior probability (pp) = 0.93 and 0.92, respectively]. For participants in the two lowest quartiles of baseline DHA status, adherence to the higher dose resulted in lower EPTB (≥5.5% RBC-PL-DHA, quartiles 1 and 2, pp = 0.95 and 0.96; ≥8% RBC-PL-DHA, quartiles 1 and 2, pp = 0.94 and 0.95). Using the Bayesian model, EPTB was reduced by 65%, from 3.45% to 1.2%, using both cut points. Adherence also reduced PTB before 35, 36 and 37 weeks using both cut points (pp ≥ 0.95). In general, performance of the nonadherent subgroup mirrored that of participants assigned to 200 mg.

CONCLUSION

Adherence to high dose DHA reduced EPTB and PTB. The largest effect of adherence on reducing EPTB was observed in women with low baseline DHA levels.

CLINICALTRIALS

gov (NCT02626299).

Docosahexaenoic acid (DHA) intake estimated from a 7-question survey identifies pregnancies most likely to benefit from high-dose DHA supplementation

BACKGROUND

Two randomized trials found women with low blood docosahexaenoic acid (DHA; an omega 3 fatty acid) had fewer early preterm births (<34 weeks gestation) if they were assigned to high dose DHA supplementation, however, there is currently no capacity for clinicians who care for pregnancies to obtain a blood assessment of DHA. Determining a way to identify women with low DHA intake whose risk could be lowered by high dose DHA supplementation is desired.

OBJECTIVE

To determine if assessing DHA intake can identify pregnancies that benefit from high dose DHA supplementation.

STUDY DESIGN

This secondary analysis used birth data from 1310 pregnant women who completed a 7-question food frequency questionnaire (DHA-FFQ) at 16.8 ± 2.5 weeks gestation that is validated to assess DHA status. They were then randomly assigned to a standard (200 mg/day) or high dose (800 or 1000 mg/day) DHA supplement for the remainder of pregnancy. Bayesian logistic regressions were fitted for early preterm birth and preterm birth as a function of DHA intake and assigned DHA dose.

RESULTS

Participants who consumed less than 150 mg/day DHA prior to 20 weeks' gestation (n = 810/1310, 58.1%) had a lower Bayesian posterior probability (pp) of early preterm birth if they were assigned to high dose DHA supplementation (1.4% vs 3.9%, pp = 0.99). The effect on preterm birth (<37 weeks) was also significant (11.3% vs 14.8%, pp = 0.97).

CONCLUSION

The DHA-FFQ can identify pregnancies that will benefit most from high dose DHA supplementation and reduce the risk of preterm birth. The DHA-FFQ is low burden to providers and patients and could be easily implemented in obstetrical practice.

The value of a two-armed Bayesian response adaptive randomization trial

We investigate the value of a two-armed Bayesian response adaptive randomization (RAR) design to investigate early preterm birth rates of high versus low dose of docosahexaenoic acid during pregnancy. Unexpectedly, the COVID-19 pandemic forced recruitment to pause at 1100 participants rather than the planned 1355. The difference in power between number of participants at the pause and planned was 87% and 90% respectively. We decided to stop the study. This paper describes how the RAR was used to execute the study. The value of RAR in two-armed studies is quite high and their use in the future is promising.

Bayesian EMAX model with a mixture of normal distributions for dose-response in clinical trials

When a dose-response relationship is monotonic, the EMAX model has been shown to provide a good empirical fit for designing and analyzing dose-response data across a wide range of pharmaceutical studies. However, the EMAX model has never been applied to a finite mixture distribution. Motivated by a proposal investigating DHA dose effect on preterm birth (PTB, <37 weeks gestation) rate, we developed a Bayesian EMAX mixture model incorporating the three normal components finite mixture model into the EMAX framework. The proposed Bayesian EMAX mixture model analyzes gestational age as a continuous variable, which allows for statistically efficient estimates of PTB rate using various cut point with the same parsimonious model. For example, we can estimate the rate of early PTB (ePTB, <34 weeks gestation), PTB (<37 weeks gestation), and late-term birth (>41 weeks gestation) using the same model. We compared our proposed EMAX mixture model with an EMAX logistic model and an independent doses logistic model for a dichotomized endpoint using extensive simulations. Across the scenarios under consideration, the EMAX mixture model achieved higher power than the EMAX logistic model and the independent doses logistic model in detecting the effect of DHA supplementation on the PTB rate. The EMAX mixture model also resulted in smaller mean squared errors (MSE) in PTB rate estimates.

Higher dose docosahexaenoic acid supplementation during pregnancy and early preterm birth: A randomised, double-blind, adaptive-design superiority trial

BACKGROUND

Several meta analyses have concluded n-3 fatty acids, including docosahexaenoic acid (DHA), reduce early preterm birth (EPB, < 34 weeks), however, the amount of DHA required is unclear. We hypothesized that 1000 mg DHA per day would be superior to 200 mg, the amount in most prenatal supplements.

METHODS

This randomised, multicentre, double-blind, adaptive-design, superiority trial was conducted in three USA medical centres. Women with singleton pregnancies and 12 to 20 weeks gestation were eligible. randomization was generated in SAS® by site in blocks of 4. The planned adaptive design periodically generated allocation ratios favoring the better performing dose. Managing study personnel were blind to treatment until 30 days after the last birth. The primary outcome was EPB by dose and by enrolment DHA status (low/high). Bayesian posterior probabilities (pp) were determined for planned efficacy and safety outcomes using intention-to-treat. The study is registered with ClinicalTrials.gov (NCT02626299) and closed to enrolment.

FINDINGS

Eleven hundred participants (1000 mg, n = 576; 200 mg, n = 524) were enrolled between June 8, 2016 and March 13, 2020 with the last birth September 5, 2020. 1032 (n = 540 and n = 492) were included in the primary analyses. The higher dose had a lower EPB rate [1.7% (9/540) vs 2.4% (12/492), pp=0.81] especially if participants had low DHA status at enrolment [2.0% (5/249) vs 4.1%, (9/219), pp=0.93]. Participants with high enrolment DHA status did not realize a dose effect [1000 mg: 1.4% (4/289); 200 mg: 1.1% (3/271), pp = 0.57]. The higher dose was associated with fewer serious adverse events (maternal: chorioamnionitis, premature rupture of membranes and pyelonephritis; neonatal: feeding, genitourinary and neurologic problems, all pp>0.90).

INTERPRETATION

Clinicians could consider prescribing 1000 mg DHA daily during pregnancy to reduce EPB in women with low DHA status if they are able to screen for DHA.

FUNDING

The National Institutes of Health Child Health and Human Development (NICHD) funded the study. Life's DHA™-S oil, DSM Nutritional Products LLC, Switzerland provided all capsules.

Dose-response relationship between docosahexaenoic acid (DHA) intake and lower rates of early preterm birth, low birth weight and very low birth weight

As previously reported, intention-to-treat findings from our phase III randomized clinical trial found that a supplement of 600 mg docosahexaenoic acid (DHA)/day during the last half of pregnancy reduced the incidence of early preterm birth (ePTB, <34 weeks gestation) and very low birth weight (VLBW < 1500 g) offspring. Given the potentially immense clinical significance of these findings, the goal of this secondary analysis was to (1) identify maternal characteristics related with capsule intake (i.e. DHA dose exposure) and (2) determine if DHA dose was associated with low (<2500 g) and very low birth weight after controlling for any relevant maternal characteristics. Three hundred forty-five pregnant mothers were recruited from hospitals in the Kansas City metropolitan area between 2006 and 2011. Most participants (n = 299) were from the phase III trial mentioned above, but we also included 46 participants from a second smaller, randomized trial that utilized an identical intervention design and was conducted concurrent to the larger trial. Both trials assigned participants to either 3 daily capsules of vegetable oil without DHA (n = 169) or 3 daily capsules of 200 mg DHA each (n = 176). Total capsules consumed was recorded by pharmacy supervised capsule count or participant self-report when needed. Maternal age, education, race and gestational age at delivery as well as infant birth weight were available for both trials. A Bayesian linear model indicated capsule intake increased with maternal age (p = 0.0100) and years of education (p = 0.0002). A Bayesian bivariate mixture-model associated capsule intake with simultaneous lower probability of ePTB, low birth weight (LBW, <2500 g) and VLBW (p = 0.0327). This, in conjunction with the positive findings in the clinical trial, support the need for future research to examine intervention methods to improve capsule compliance strategies in younger and less educated mothers.

Predicting the effect of maternal docosahexaenoic acid (DHA) supplementation to reduce early preterm birth in Australia and the United States using results of within country randomized controlled trials

The DHA to Optimize Mother Infant Outcome (DOMInO) and Kansas DHA Outcomes Study (KUDOS) were randomized controlled trials that supplemented mothers with 800 and 600mg DHA/day, respectively, or a placebo during pregnancy. DOMInO was conducted in Australia and KUDOS in the United States. Both trials found an unanticipated and statistically significant reduction in early preterm birth (ePTB; i.e., birth before 34 weeks gestation). However, in each trial, the number of ePTBs were small. We used a novel Bayesian approach to estimate statistically derived low, moderate or high risk for ePTB, and to test for differences between the DHA and placebo groups. In both trials, the model predicted DHA would significantly reduce the expected proportion of deliveries in the high risk group under the trial conditions of the parent studies. Among the next 300,000 births in Australia we estimated that 1112 ePTB (95% credible interval 51-2189) could be avoided by providing DHA. And in the USA we estimated that 106,030 ePTB (95% credible interval 6400 to 175,700) could be avoided with DHA.

Comparison of Dichotomized and Distributional Approaches in Rare Event Clinical Trial Design: a Fixed Bayesian Design

This research was motivated by our goal to design an efficient clinical trial to compare two doses of docosahexaenoic acid supplementation for reducing the rate of earliest preterm births and/or preterm births. Dichotomizing continuous gestational age data using a classic binomial distribution will result in a loss of information and reduced power. A distributional approach is an improved strategy to retain statistical power from the continuous distribution. However, appropriate distributions that fit the data properly, particularly in the tails, must be chosen, especially when the data are skewed. A recent study proposed a skew-normal method. We propose a three-component normal mixture model and introduce separate treatment effects at different components of gestational age. We evaluate operating characteristics of mixture model, beta-binomial model, and skew-normal model through simulation. We also apply these three methods to data from two completed clinical trials from the USA and Australia. Finite mixture models are shown to have favorable properties in preterm births analysis but minimal benefit for earliest preterm births analysis. Normal models on log transformed data have the largest bias. Therefore we recommend finite mixture model for preterm births study. Either finite mixture model or beta-binomial model is acceptable for earliest preterm births study.