A Bayesian Re-analysis of the GALFLU Trial of High-Dose vs Standard-Dose Influenza Vaccine Against Hospitalization Outcomes

Abstract

The GALFLU trial showed a lower incidence of hospitalization among adults aged 65–79 years randomized to high-dose inactivated influenza vaccine (HD-IIV) vs standard-dose inactivated influenza vaccine (SD-IIV). This Bayesian re-analysis estimated >80% probability of benefit (relative vaccine effectiveness >10%) against influenza- or pneumonia-related hospitalization, supporting the benefit of HD-IIV over SD-IIV.

Clinical Trials Registration. GALFLU ClinicalTrials.gov (NCT06141655); EudraCT (2023-506977-36-00).

Health authorities recommend annual influenza vaccination for older adults (≥65 years) [1, 2]. The results from a phase IIIb–IV randomized controlled trial demonstrated 24.2% (95% CI: 9.7%–36.5%) superior protection of high-dose inactivated influenza vaccine (HD-IIV), compared with standard-dose inactivated influenza vaccine (SD-IIV) against laboratory-confirmed influenza (LCI) cases in older adults [3].

FLUNITY-HD (Pooled Analysis of Methodologically Harmonized Pragmatic Randomized Trials of High-Dose vs. Standard-Dose Influenza Vaccine Against Severe Clinical Outcomes), a prespecified pooled analysis of the harmonized, individually randomized DANFLU-2 (A Pragmatic Randomized Trial to Evaluate the Effectiveness of High-Dose Influenza Vaccine Versus Standard-Dose Influenza Vaccine in Older Adults) and GALFLU (Pragmatic Randomized Trial to Evaluate the Effectiveness of High-Dose Quadrivalent Influenza Vaccine Versus Standard-Dose Quadrivalent Influenza Vaccine in Adults Aged 65 to 79 Years in Galicia, Spain) trials, demonstrated superior protection of HD-IIV compared with SD-IIV against multiple clinical hospitalization endpoints, including influenza-related and cardiorespiratory outcomes in more than 466 000 older adults [4–6]. Results from the GALFLU trial suggested a lower incidence of influenza- or pneumonia-related hospitalization for those receiving HD-IIV vs SD-IIV (23.7% [relative vaccine effectiveness (rVE)]; 95% CI: 6.6%–37.7%) [6].

Unlike the frequentist approach, a Bayesian analysis was used to quantify uncertainty by interpreting probability as a subjective degree of belief, enabling the integration of prior knowledge with new evidence [7]. This Bayesian statistical approach provides a coherent, adaptable, and information-rich framework that can enhance the design, analysis, and interpretation of clinical trials [8]. In contrast to conventional frequentist procedures that rely on hypothesis testing and P values, Bayesian analyses yield direct probability estimates of the magnitude and direction of treatment effects, and the likelihood of such effects exceeding clinically relevant thresholds. This is achieved by combining prior information derived from previous studies or noninformative assumptions with the observed trial data to generate a posterior probability distribution. This framework allows the probabilistic quantification of compliance with certain thresholds for clinical decision-making [9].

METHODS

The design and conduct of the GALFLU trial (NCT06141655), which compared HD-IIV with SD-IIV against severe outcomes, have been detailed elsewhere [6]. GALFLU was a pragmatic, registry-based, open-label, active-controlled, individually randomized trial in community-dwelling older adults aged 65–79 years in Galicia, Spain. The trial was conducted during the 2023–2024 and 2024–2025 influenza seasons, with enrollment from the end of October to the end of November in each season. In each season, the participants were randomized 1:1 to receive HD-IIV or SD-IIV.

A Bayesian log-Poisson regression was used to estimate posterior probabilities of risk ratio (RR) for the analyzed outcomes. Three prior specifications, distributed on the log relative risk scale, were specified to capture different levels of uncertainty and skepticism regarding the expected HD-IIV benefit:

1. Noninformative: Centered at rVE = 0%, SD = 100% (95% CI: −3.3% × 10⁻⁶%; 100%).

2. Evidence-based: Based on previously published rVE results (Supplementary Table 2) [10, 11]. To account for between-study heterogeneity, the SE was inflated using the I² statistic, approximating the variance of the true underlying effect. These priors were based on multiseason meta-analyses of the outcomes.

3. Skeptical: Centered at rVE = 0%, with SD = 0.15%. This prior was specified to constrain a large treatment effect towards no effect (95% CI: −113.6% to 53.2%).

Alternative skeptical SD scenarios were evaluated to assess the impact on posterior distribution estimation (Supplementary Text 1).

Outcomes

Outcomes were posterior rVE ([1 – RR] × 100%) with 95% credible interval (CrI) and rVE probabilities exceeding the range of thresholds, from rVE of greater than 0% (any benefit), greater than 5%, and greater than 10% with corresponding Monte-Carlo SE (MCSE). Convergence statistics R-hat (Gelman-Rubin diagnostic), bulk, and tail effective sample sizes were also presented. The number of divergent transitions were also computed. Statistical analyses were performed using R (version 4.4.1), using the brm function from the brms package (version 2.22.0), considering 6000 iterations per chain, 3000 warmup iterations and 4 Markov chains [12] (see Supplementary File 1). The rstan package (version 2.32.7) was used as the computational backend for model fitting via Stan [13]. The posterior package (version 1.6.1) was considered to extract the distribution from the fitted model [14]. Outcomes assessed included influenza- or pneumonia-related hospitalization, cardiorespiratory disease, and all-cause mortality. Exploratory outcomes included LCI hospitalization [6].

RESULTS

The GALFLU trial enrolled 134 476 individually randomized participants during the 2023–2024 and 2024–2025 influenza seasons, of whom 133 882 were included in the intention-to-treat analysis (see Figure 1 in [6]). The mean (±SD) age of participants was 72.3 ± 4.3 years, and 53.6% of participants were male [6].

The risk of hospitalization for influenza or pneumonia was lower in the HD-IIV group than in the SD-IIV group, with the median posterior rVE estimates ranging from 17.0 to 24.0, with high probabilities (≥83.8%) of substantive clinical benefit (rVE >10%). HD-IIV showed a consistent beneficial effect compared with SD-IIV against cardiorespiratory hospitalizations (posterior median rVE: 7.8%–11.6%) with a high probability (≥76.3%) of a moderate clinical benefit (rVE >5%). The effect of HD-IIV vs SD-IIV against all-cause hospitalizations was moderate and consistent across different priors (rVE: 2.5%–2.8%), with an 88.0%–90.9% probability of any benefit, but with low probability (10.7%–13.5%) of a clinical benefit greater than 5% and 0% probability of an rVE greater than 10%. The estimated rVE against all-cause mortality was 8.6%–12.8%, with high probabilities (>77%) of an rVE greater than 5%.

For influenza hospitalizations, median rVE values ranged from 16.2% to 31.8% with high probabilities (≥74.1%) of substantive clinical effect (rVE >10%). For pneumonia hospitalizations, median rVE values ranged from 9.7% to 22.6% with high probabilities (≥70.1%) of at least a moderate clinical effect (rVE >5%) across all priors. The rVE against LCI hospitalizations ranged from 9.7% to 19.5%, with a probability of greater than or equal to 68.4% of an rVE more than 5% (Table 1 and Supplementary Figure 1).

Table 1.

 Results of Bayesian Analysis, Displaying Posterior Median Relative Vaccine Effectiveness (rVE) and Probabilities of rVE Greater Than 0%, 5%, and 10% (With the Corresponding Monte-Carlo SE [MCSE]), MCSE Diagnosis (R-Hat, Bulk and Tail ESS), Considering Noninformative, Evidence-Based, and Skeptical Priors

						Probability, %
Endpoint, No. of Events (%) per Arm	Prior	rVE Posterior Median (95% CrI), %	R-Hat	Bulk-ESS	Tail-ESS	rVE >0% (MCSE)	rVE >5% (MCSE)	rVE >10% (MCSE)
Hospitalization for pneumonia or influenza	Noninformative	23.7 (7.3; 37.2)	1.0002	9526	7514	99.7 (0.0005)	98.6 (0.0011)	95.2 (0.0020)
HD-IIV: 174 (0.26)	Evidence-based	24.0 (8.1; 37.0)	1.0004	9285	7871	99.8 (0.0004)	99.0 (0.0009)	95.7 (0.0018)
SD-IIV: 227 (0.34)	Skeptical	17.0 (2.6; 29.3)	1.0003	9310	8081	98.9 (0.0009)	95.0 (0.0020)	83.8 (0.0034)
Cardiorespiratory hospitalization	Noninformative	8.4 (0.2; 16.2)	1.0001	9713	6853	97.7 (0.0014)	79.9 (0.0037)	34.2 (0.0043)
HD-IIV: 985 (1.47)	Evidence-based	11.6 (4.8; 17.6)	1.0004	10023	7970	100 (0.0002)	97.2 (0.0015)	68.4 (0.0042)
SD-IIV: 1071 (1.60)	Skeptical	7.8 (−0.1; 15)	1.0001	8850	7781	97.2 (0.0015)	76.3 (0.0039)	28.5 (0.0041)
Hospitalization for any cause	Noninformative	2.5 (−1.7; 6.5)	1.0000	10145	7192	88.0 (0.0030)	10.8 (0.0028)	0.0(0.0001)
HD-IIV: 4336 (6.46)	Evidence-based	2.8 (−1.3; 6.6)	1.0006	11011	7803	90.9 (0.0026)	13.5 (0.0031)	0.0 (0.0001)
SD-IIV: 4427 (6.63)	Skeptical	2.5 (−1.6; 6.5)	1.0001	10287	8087	88.4 (0.0029)	10.7 (0.0028)	0.0 (0.0001)
Death from any cause	Noninformative	12.8 (−1.6; 25.4)	1.0004	10094	7814	96.1 (0.0018)	85.7 (0.0032)	65.0 (0.0044)
HD-IIV: 305 (0.45)	Evidence-based	8.6 (−0.9; 17.4)	1.0001	10347	7918	96.1 (0.0018)	77.2 (0.0038)	37.9 (0.0044)
SD-IIV: 348 (0.52)	Skeptical	10.1 (−2.9; 21.7)	1.0006	11367	8483	93.7 (0.0022)	78.9 (0.0037)	50.4 (0.0046)
Hospitalization for influenza	Noninformative	31.8 (6.4; 50.6)	1.0003	8243	6663	99.1 (0.0009)	98.0 (0.0013)	95.8 (0.0018)
HD-IIV: 63 (0.09)	Evidence-based	22.9 (7.0; 35.8)	1.0002	9239	7839	99.7 (0.0005)	98.7 (0.0010)	94.9 (0.0020)
SD-IIV: 92 (0.14)	Skeptical	16.2 (−3.9; 32.3)	1.0003	10329	8173	94.9 (0.0020)	87.1 (0.0031)	74.1 (0.0040)
Hospitalization for pneumonia	Noninformative	15.6 (−8.3; 33.9)	0.9999	9713	7524	91.0 (0.0026)	82.1 (0.0035)	69.4 (0.0042)
HD-IIV: 116 (0.17)	Evidence-based	22.6 (13.0; 31.0)	1.0004	10140	7558	100 (0.0001)	100 (0.0001)	99.4 (0.0007)
SD-IIV: 137 (0.21)	Skeptical	9.7 (−9.4; 24.9)	1.0002	10078	7780	85.0 (0.0033)	70.1 (0.0042)	48.3 (0.0046)
Laboratory-confirmed influenza hospitalization	Noninformative	19.6 (−9.4; 41.3)	1.0003	9570	7854	91.4 (0.0026)	85.2 (0.0032)	76.2 (0.0039)
HD-IIV: 72 (0.11)	Evidence-based	19.5 (5.0; 31.7)	1.0003	9923	8203	99.4 (0.0007)	97.5 (0.0014)	90.9 (0.0026)
SD-IIV: 89 (0.13)	Skeptical	9.7 (−12.0; 27.3)	1.001	9633	7291	83.1 (0.0034)	68.4 (0.0042)	49.0 (0.0046)

No divergences were observed in any model or any endpoint. The noninformative prior was centered at rVE = 0 with an SD of 100 on the log relative risk scale. The skeptical prior was also centered at 0 with an SD of 0.15. Evidence-based prior are displayed in Supplementary Table 2.

Abbreviations: CrI, credible interval; ESS, effective sample size; HD-IIV, high-dose inactivated influenza vaccine; SD-IIV, standard-dose inactivated influenza vaccine.

Markov Chain Monte Carlo diagnostics indicated good convergence: R-hat less than 1.001, bulk and tail ESS greater than 1000, and MCSE less than 0.01 for all computed models. No divergences were observed for any model and endpoint.

DISCUSSION

In this Bayesian re-analysis of the GALFLU trial, HD-IIV consistently showed high probabilities of improved protection against severe outcomes compared with the SD-IIV in older adults, across noninformative, evidence-based, and skeptical priors. By applying Bayesian methods and considering a range of prior distributions, we have improved the interpretability of the GALFLU trial results while accounting for prior knowledge of the benefit of HD-IIV over SD-IIV. Our findings highlighted the consistently high probabilities of lower incidence of the primary endpoint of influenza- or pneumonia-related hospitalization for HD-IIV compared with SD-IIV, as well as against secondary and exploratory endpoints such as hospitalization for cardiorespiratory disease and hospitalization for LCI, respectively.

Noninformative prior estimates fit accordingly with rVE estimates from the study by Pardo-Seco et al [6] and evidence-based prior estimates are generally well aligned. Evidence-based posterior estimates were well aligned with noninformative estimates when our estimates were close to the prior evidence-based estimates, such as influenza- or pneumonia-related hospitalizations and hospitalization for LCI. Meanwhile, when discrepancies between the evidence-based prior and observed data were present, posterior evidence-based estimates were systematically pulled towards the prior, resulting in higher or lower rVE estimates depending on whether the prior assumed stronger or weaker effects, such as death from any cause.

The most pessimistic scenario corresponds with the skeptical prior (no effect and unlikely large effects). Despite a skeptical prior, probabilities of any benefit (rVE >0%) remained above 83% for all outcomes, giving confidence in the finding of benefit of HD-IIV over SD-IIV against all outcomes assessed. Furthermore, even with the skeptical prior, there were high probabilities of a substantive effect (rVE >10%) for several outcomes including influenza- or pneumonia-related hospitalization or influenza hospitalization, further strengthening those findings.

The strengths of Bayesian analysis include improved reliability of results by incorporating informative priors into GALFLU trial data analysis. This approach enables the continuous updating of evidence, an explicit quantification of uncertainty, and derivation of probabilities that are directly interpretable in clinical terms. A limitation of Bayesian analysis is that it is sensitive to the choice of prior, particularly considering skeptical priors, as they tend to attenuate the estimated effects, or when observed study results differ substantively from historical evidence. In this study, we defined an evidence-based prior using only randomized controlled trials, excluding observational studies to reduce confounding. This allowed integration of GALFLU findings with rigorous historical evidence from randomized controlled trials. The transportability of historical data to the GALFLU 65- to 79-year-old population requires consideration when the prior data come from a multicountry older-adult population that may not fully represent the study cohort. Both the evidence-based prior and the study data were multiseason, helping to overcome season-to-season variability in vaccine effectiveness. Additional confidence in the findings is provided by the consistency across noninformative and informative priors.

The main limitations of the GALFLU trial included the limited number of events, open-label design, nonroutine influenza testing, and lack of clinical outcomes adjudication [6]. While the Bayesian framework enabled a more flexible probabilistic data interpretation, it did not mitigate biases inherent to study design. However, the use of different prior assumptions provided insights into the robustness of the findings under different scenarios, giving further confidence to the findings of protection with HD-IIV vs SD-IIV against various hospitalization and mortality outcomes.

CONCLUSION

This Bayesian re-analysis of the GALFLU trial is consistent with a lower incidence of hospitalizations among adults aged 65–79 years with HD-IIV vs SD-IIV, showing high probability of reduction in influenza- or pneumonia-related hospitalization in older adults with HD-IIV compared with those with SD-IIV, and consistency in the findings with HD-IIV vs SD-IIV across the outcomes and priors explored.

Supplementary Data

Supplementary materials are available at Clinical Infectious Diseases online. Consisting of data provided by the authors to benefit the reader, the posted materials are not copyedited and are the sole responsibility of the authors, so questions or comments should be addressed to the corresponding author.