 Dwi'n siarad. Efallai, sydd yn cydael i mi ddweud y dyfodol yma. Os ydych chi am hyn gyffredinogi, bydd eich fath, am y hyd yn llunio hi'r ngosloods, a'r ddionngり gyrwm meddwl newydd, ond mae'r dweud'r meddwl modled wedi'i meddwl. Ystafell y gynhyrch arall. Yn ymwneud ag gweld i'r llunio hefyd i'r adnodd yn gwybod mewn meddwl meddwl. Ond gweld o'n meddwl meddwl gwnddi i'r ddiong ynghylch ar y ddweud o'r ddysgu a'u cyfnodol, a i'w ffordd decyddiad o gyfnodol o'r ddysgu ar y masygl yma yw'r ddysgu yma. A gYNN tellio i'n credu ond y taethau gwd-ddechrau hyn o'r ddysgu yn cyffreddau a yw'r ddod ar eman ar hyn. Y rhaid i'r ystod y dych chi'n gwybod 아니� i'r ddysgu yn ymgynghwng, yw rhywbeth bod yn y ddysgu, gwybod i ei fydd yn hyn. Mae yna yna ddangos arni, cyfnodd unig ar ddod i wych chi wefio y cwestiynau na yw'r ildw i ddefnyddio gael fade시, rwy'n holl o'w'r llwyddiadig ymddangor ddod o'r adrodd amser a hi fyddai'r adrodd i diawn o'r iddyn nhw. A bod y penderfyn o'r cyfronflwyno, mae'r cwestiynau adrodd yn ôl yn gallu gwreithio gwyrddol yn cwestiynau. Ond ychydig of yn credu drwy dyma'ch gwaith o'r adrodd yn cydweithio gydweithio eu ddathun, schedules to have different effect based on from a metrics based on biology. So this is likely to introduce in consistency or heterogeneity into our model if we do this. The other approach is to split all of the doses apart without this end up with a very sparsely connected network. And we could even have treatments and doses that are disconnected from our network, in which case we can't estimate relative effects for them. And so that assumes that treatments are totally independent that there's no relationship between them. So what we'd really like is something in between. We'd like to be able to functionally model this. That's what we can do with MB&MA dose. So our first steps when using the package are to load the data into a network object. Once we have that, we can look at our data more generally. We can plot network plots, either at the dose level or at the agent level. To get an idea of perhaps whether there is a dose response relationship, and what it might look like, we can also run a simple standard network meta-analysis and plot the results at each dose of each agent on a graph to get an idea of what shape the dose response function might take. So here we can see for Allagliptin that we have a sort of non-linear dose response probably, and there certainly is a dose response. We wouldn't want to ignore that and kind of group all those doses together, or we'd definitely be introducing heterogeneity. So we can then fit the model, and to do this, we apply some sort of dose response function. So we define our treatment as being a specific dose of a specific agent, and then we can set this dose response function to be equal to anything that fits the data well and that we think might be biologically and pharmacologically plausible. So here, just as an example, this is an Emax function which is used a lot in pharmacometrics, and it's got two parameters. So we'd estimate relative effects for each agent, for each of those parameters, and there's a range of different modelling arguments and options we can do when modelling this in a Bayesian framework using jags behind the scenes, and we can specify priors, all that sort of good stuff. And once we've run our model, as well as being able to plot forest plots and rankings and other kinds of sort of post-estimation options, we can predict responses. And this is really useful because this allows us to predict at doses, perhaps, which aren't in the original dataset. And we can also compare our predicted responses from the dose response NBNMA, which are these curves with the credible intervals, to the results from the split NMA, which makes minimal assumptions, and that's these vertical solid lines. That's the credible interval for the split NMA results. And what we can see here is that, compared to the split NMA, the dose response relationship has slightly more precision, so we get a precision gain over standard NMA because we're gleaning some additional information from the dose response relationship. We're not just saying that we don't know anything about the relationship. Between doses, we're saying that there is some sort of functional relationship. But of course, that is reliant on the assumption that the dose response relationship is correct. If it's not correct, we'll be introducing bias, but we can evaluate this by comparing it to the data, comparing the model fit, and also comparing it to the standard NMA. So this is a very testable assumption in this framework, and it's the only additional assumption we're making over standard NMA. What we've also found in recent work we've been doing is that this can be used to link disconnected networks. So, for example, in these two, these are just two little example networks. If we were interested in comparing, for instance, these treatments in the networks indicated by the red dotted line, we wouldn't be able to in standard NMA because there's no pathway of head-to-head evidence between them. However, if we use NBNMA dose, what we're then modelling is the dose response curves, the dose response relationships for each of those agents, and from that, we can then estimate a relative effect between our treatments of interest. So this can be really, really advantageous in cases where you have disconnected network and you have sufficient dose response information to be able to estimate these. And because we're modelling on the relative effects as well, we don't have problems of having to adjust for prognostic variables. Other features that the package can do, it can do modelling using class effects, so you can assume perhaps that different classes of agents might have similar efficacy for some of the parameters. We can do consistency checking, which is really, really important in network meta-analysis, either a global check for consistency or a local check. Both of these methods we've built in also account for the dose response relationship as there's some sort of extra things to think about when looking at consistency there. And we can also model agent-specific dose response functions, so we can have a different dose response for different agents in our network if we want. What we're hoping to do in the future is to combine MBNMA dose with another package that we've developed and that's also available on Cran called MBNMA time to produce models that can account for both time course and dose response simultaneously, because this would be really useful in drug development, pharmacometric meta-analysis stage. And in particular, we'd then also like to implement that in STAM because for the time course models, they can take a while to run due to the correlation between time points that needs to be accounted for. So that's what we're hoping to do in the future. And yeah, these are just a couple of references for the key methodological parts going on within the package. Thanks very much.