Bridging the gap between efficacy trials and model-based impact evaluation for new tuberculosis vaccines

Abstract

In Tuberculosis (TB), given the complexity of its transmission dynamics, observations of reduced epidemiological risk associated with preventive interventions can be difficult to translate into mechanistic interpretations. Specifically, in clinical trials of vaccine efficacy, a readout of protection against TB disease can be mapped to multiple dynamical mechanisms, an issue that has been overlooked so far. Here, we describe this limitation and its effect on model-based evaluations of vaccine impact. Furthermore, we propose a methodology to analyze efficacy trials that circumvents it, leveraging a combination of compartmental models and stochastic simulations. Using our approach, we can disentangle the different possible mechanisms of action underlying vaccine protection effects against TB, conditioned to trial design, size, and duration. Our results unlock a deeper interpretation of the data emanating from efficacy trials of TB vaccines, which renders them more interpretable in terms of transmission models and translates into explicit recommendations for vaccine developers.

Fig. 1

Equal prevention readouts from vaccine efficacy trials can map to multiple vaccine mechanisms and expected vaccine impacts. a Elementary M.tb. transmission model. S=susceptible, F=infected, fast progression to disease, L=infected, slow progression to disease (LTBI), D=active TB. The epidemiological parameters (black) can be modified by the vaccine effects (blue). b From the distributions of transition times between the beginning of the trial (green dots) until end-point infection (orange arrows), survival curves are built for the control and vaccine cohorts, and from their analysis, ${\mathrm{VE}}_{\inf }$ is estimated. c Equivalent schematics for the estimation of ${\mathrm{VE}}_{\mathrm{dis}}$ from survival analysis of transition times from trial’s beginning (green) to the end point associated with active TB (red arrows). d Curve of values of (${\epsilon }_{\mathrm{p}}$,${\epsilon }_{\mathrm{r}}$) compatible with a measurement of POD of ${\mathrm{VE}}_{\mathrm{dis}}=0.5$ after 4 years of follow-up (assuming no POI, i.e. ${\epsilon }_{\mathrm{\beta }}=0$). We have marked five different points in this curve, with different balances between ${\epsilon }_{\mathrm{p}}$ and ${\epsilon }_{\mathrm{r}}$, to be used in the next example. e Foreseen impacts obtained after introducing the vaccines highlighted in d in Ethiopia, at the end of 2025. Blue bars: vaccine impacts. Grey bars: difference in impact estimated between each vaccine and the least impactful case of a vaccine acting enterily through ${\epsilon }_{\mathrm{r}}$ (lightest blue). Impacts are estimated by using a large-scale transmission model as the number of TB cases prevented in the country by the vaccine during the period 2026–2050. Error bars (black bars) represent the 95% confidence interval.

Fig. 2

Methods for characterizing vaccine mechanisms from the analysis of clinical trials conducted on naive cohorts. a Inference of ${\epsilon }_{\mathrm{r}}$. From the distribution of times from infection (orange) to disease (red), we obtain the rates of fast progression to TB in each cohort: either $r^{\mathrm{c}}=r$ (control), or $r^{\mathrm{v}}=r\left(1-{\epsilon }_{\mathrm{r}}\right)$ (vaccine). b Transition times of control (left) and vaccinated cohort (right: vaccine acting through ${\epsilon }_{\mathrm{r}}$) between IGRA conversion and disease onset. By using likelihood maximization, we infer within-cohort transmission rates to disease $r^{\mathrm{c}}$ an $r^{\mathrm{v}}$, which are associated with expected values for the transition times (blue, continuous lines) that closely resemble the a priori-known analytical predictions (dashed lines). From these estimates, ${\epsilon }_{\mathrm{r}}$ is estimated as $1-r^{\mathrm{v}}∕r^{\mathrm{c}}$. c Schematic representation of the computational pipeline used in this work for the analysis of clinical trials conducted on IGRA-negative cohorts, structured in three modules. Module I: trial simulation: From a given vaccine (${\epsilon }_{\mathrm{\beta }},{\epsilon }_{\mathrm{r}},{\epsilon }_{\mathrm{p}}$) and trial dimensions ($N,T$) we simulate 500 equivalent trials. Module II: vaccine characterization: then, we analyze the outcomes of the simulated trials to estimate ${\epsilon }_{\mathrm{\beta }}$, ${\epsilon }_{\mathrm{r}}$ and ${\epsilon }_{\mathrm{p}}$. Module III: impact evaluation. We use the comprehensive transmission model developed in ref. to evaluate the impact associated with the characterized vaccines. d Fraction of valid realizations of a trial yielding epidemiologically plausible vaccine parameterizations, (excluding failed attempts). e Vaccine characterization of ${\epsilon }_{\mathrm{\beta }},{\epsilon }_{\mathrm{r}},{\epsilon }_{\mathrm{p}}$. Error bars represent the 95% confidence interval. f Estimated probability of obtaining a trial result, leading to a successful characterization of ${\epsilon }_{\mathrm{p}}$ (up) or ${\epsilon }_{\mathrm{r}}$ (bottom) (CI not crossing 0 at a 95% confidence level for the parameter whose ground-truth value is non-zero).

Fig. 3

Impact evaluation of empirically characterized vaccines: mechanism effects on expected impact and uncertainty. a, b, c Vaccines characterized from an efficacy readout of ${\mathrm{VE}}_{\mathrm{dis}}=25\%$, ${\mathrm{VE}}_{\mathrm{dis}}=50\%$ and ${\mathrm{VE}}_{\mathrm{dis}}=75\%$, respectively. Blue bars: impact estimates for different vaccines. Two different contributions to the overall impact uncertainty are distinguished: Gold bars: intrinsic contribution coming from general inputs of the transmission model. Black, dashed bars: extra contribution from uncertain vaccine characterizations. Grey bars: differences in impact between each vaccine and the least impactful case of a vaccine acting 100% through ${\epsilon }_{\mathrm{p}}$ (leftmost, light blue bar in each panel). Error bars represent the 95% confidence interval.

Fig. 4

Vaccine characterization from clinical trials conducted on IGRA-positive individuals. a Section of the transmission chain that is observed during a trial conducted on cohorts of IGRA-positive individuals. Recruiting IGRA-positive participants turns possible to observe a vaccine-mediated protection against fast progression to TB upon reinfection during the trial (i.e. $\widehat{{\epsilon }_{\mathrm{p}}}$), in addition to a delay in the transition rate to disease (${\epsilon }_{\mathrm{r}}$). b Family of curves that bound ${\mathrm{VE}}_{\mathrm{dis}}$, $\widehat{{\epsilon }_{\mathrm{p}}}$ and ${\epsilon }_{\mathrm{r}}$ for different levels of the fraction of individuals recruited within the reservoir $F$: $F∕\left(L+F\right)$. The shaded area represents the whole set of points ($\widehat{{\epsilon }_{\mathrm{p}}}$,${\epsilon }_{\mathrm{r}}$) that are compatible with a single readout ${\mathrm{VE}}_{\mathrm{dis}}=50\%$, when the basal epidemiological parameters of the population are the same used in the previous sections. The dots represent four extreme vaccine examples whose impacts are to be estimated later. c Different impacts foreseen from the four extreme vaccines highlighted in b (blue bars), and differences with respect to the least favourable interpretation (grey bars). Error bars (black bars) represent the 95% confidence interval.