time horizon - for how long the new intervention will have an effect

https://egon.stats.ucl.ac.uk/static/stat0019/slides/11_VoI/#38

approximation of the impact of each parameter in the EVPPI

https://egon.stats.ucl.ac.uk/static/stat0019/slides/11_VoI/#32

Opportunity loss - Maximum net benefit minus value of the treatment (each row) that is the best in average (overall)

https://egon.stats.ucl.ac.uk/static/stat0019/slides/11_VoI/#15

Minimum: 0 - No gain Maximum: there is not an upper bound

There is no a natural scale.

En la vida real se puede aplicar un analisis para comparar si - what you get is less of what you pay for

https://egon.stats.ucl.ac.uk/static/stat0019/slides/11_VoI/#12

EVSI - help to define the best design of a new study to maximise the benefit.

https://egon.stats.ucl.ac.uk/static/stat0019/slides/11_VoI/#10

EVPPI only in selected parameters. Help to answer - in which parameters should I concentrate if I can collect more information

https://egon.stats.ucl.ac.uk/static/stat0019/slides/11_VoI/#9

what would i do if i get more information compared with what i would do now with data i have available. EVPI maximum amount of value - upper value

https://egon.stats.ucl.ac.uk/static/stat0019/slides/11_VoI/#8

For going deep. Check the last book

https://egon.stats.ucl.ac.uk/static/stat0019/slides/11_VoI/#2

selection model - easy to run although it seem counterintuitive write it in that way.

https://egon.stats.ucl.ac.uk/static/stat0019/slides/10_Missing/#35

2.Analysis of each dataset

https://egon.stats.ucl.ac.uk/static/stat0019/slides/10_Missing/#32

MEI - Model the missing value. For the missing points I create a value from a distribution. You create several versions of the dataset.

https://egon.stats.ucl.ac.uk/static/stat0019/slides/10_Missing/#23

artificially making an assumption. It could be even worst than CCA.

https://egon.stats.ucl.ac.uk/static/stat0019/slides/10_Missing/#21

Inverse... ayuda a que las observaciones que tengo se parezcan mas a la población general

we are going to try to rebalance the differences.

https://egon.stats.ucl.ac.uk/static/stat0019/slides/10_Missing/#13

MAR The blocks are not independent. Ejemplo yi salary xi age https://egon.stats.ucl.ac.uk/static/stat0019/slides/10_Missing/#7

MCAR the blocks are completely independent. Have missingness has nothing to do with anything. It is the best situation.

https://egon.stats.ucl.ac.uk/static/stat0019/slides/10_Missing/#6

Missing completely at random - Expression not completely accurate because I am saying they have a especific distribution.

These are mechanism we assume to describe the reason why we have missing data. "I think" They are not properties of the data but ways that I use to understand the data.

https://egon.stats.ucl.ac.uk/static/stat0019/slides/10_Missing/#6

Partial - intermitent - It is more complicated in terms of modelling

https://egon.stats.ucl.ac.uk/static/stat0019/slides/10_Missing/#5

If you do not have observed data the information criteria become useless.

https://egon.stats.ucl.ac.uk/static/stat0019/slides/10_Missing/#4

PSM- Not ideal I can estimate the pi but not the lambdas.

Make sense if you only move forward, it means you do not go back to the previous state

https://egon.stats.ucl.ac.uk/static/stat0019/slides/09_MM/#35

Markov trace

https://egon.stats.ucl.ac.uk/static/stat0019/slides/09_MM/#25

if the markov cycle is small that approximation is good

Markov is using discret measure while survival curves are continuous.

https://egon.stats.ucl.ac.uk/static/stat0019/slides/09_MM/#23

in situation like this make sense to have a differencial discount between cost and benefits.

https://egon.stats.ucl.ac.uk/static/stat0019/slides/09_MM/#16

cohort model: there is no a track of each individual

https://egon.stats.ucl.ac.uk/static/stat0019/slides/09_MM/#11

Pi: proportion of people in a specific state Lambda: transition probability

m: absolute number of people in each state

https://egon.stats.ucl.ac.uk/static/stat0019/slides/09_MM/#10

Index 1, where you start Index 2, where you end

https://egon.stats.ucl.ac.uk/static/stat0019/slides/09_MM/#7

How to subset from - to from preprogression to progression

https://egon.stats.ucl.ac.uk/static/stat0019/slides/09_MM/?panelset=code#19

hoth-change of dying of everything else

https://egon.stats.ucl.ac.uk/static/stat0019/slides/08_Survival/#41

blue as anchor. try to regularise the analysis to avoid make something stupid.

https://egon.stats.ucl.ac.uk/static/stat0019/slides/08_Survival/#39

you have the prior distributions

https://egon.stats.ucl.ac.uk/static/stat0019/slides/08_Survival/#32

I can use a prior to balance my data and get estimates that make sense.

https://egon.stats.ucl.ac.uk/static/stat0019/slides/08_Survival/#28

censoring and extrapolation part are not observed data but dic is criteria based on observed data.

decision is a bayesian problems

https://egon.stats.ucl.ac.uk/static/stat0019/slides/08_Survival/#24

fit.models - define model (bay, freq) model.fit.plot

psa.plot (green) additional analysis requiere to HTA

https://egon.stats.ucl.ac.uk/static/stat0019/slides/08_Survival/#17

INLA - Integrated nested laplace approximation

https://egon.stats.ucl.ac.uk/static/stat0019/slides/08_Survival/#15

gen gamma and F have more parameters which make them more flexible. Negative side, you need a lot of data. You do not need to try all of them, just those that make sense for the problem you are analysing.

https://egon.stats.ucl.ac.uk/static/stat0019/slides/08_Survival/#14

usually transformate using the log

https://egon.stats.ucl.ac.uk/static/stat0019/slides/08_Survival/#13

t time to event which depend of density that depends of two parameters

mu and alpha can depend of different set of covariates. most of people assumed that covariates affect the scale parameter.

https://egon.stats.ucl.ac.uk/static/stat0019/slides/08_Survival/#12

the difference become smaller (now almost a year versus when we observed the KM)

https://egon.stats.ucl.ac.uk/static/stat0019/slides/08_Survival/#11

median survival time - where half of population have died

https://egon.stats.ucl.ac.uk/static/stat0019/slides/08_Survival/#10

KM - plot of the raw data https://egon.stats.ucl.ac.uk/static/stat0019/slides/08_Survival/#9

parametric models - more used in HTA -Completely specify h and s -once you have the model you can extrapolate it

https://egon.stats.ucl.ac.uk/static/stat0019/slides/08_Survival/#8

you do not specify directly the distribution of the survival time

https://egon.stats.ucl.ac.uk/static/stat0019/slides/08_Survival/#7

observations without a dot - lose of information

-lose contact - you do not know if he experience the event or not -you stop the follow-up or study for different reasons

Variables

t- time-to have the event or censored d-indicator of censoring (1 if fully observed)

https://egon.stats.ucl.ac.uk/static/stat0019/slides/08_Survival/#6

These relationships allow us to define a model using whatever we one

https://egon.stats.ucl.ac.uk/static/stat0019/slides/08_Survival/#5

you can have constant hazard or other type of behaviour (monotonic)

https://egon.stats.ucl.ac.uk/static/stat0019/slides/08_Survival/#4

Probability that you survive until certain time T h hazard H cummulative hazard

h -> gives me an instantaneous risk all these terms are related one to each other

https://egon.stats.ucl.ac.uk/static/stat0019/slides/08_Survival/#3

you can add the KM curve

https://egon.stats.ucl.ac.uk/static/stat0019/slides/08_Survival/?panelset1=plotting-%283%29#19

original-> show results of the original packages in their specific format

https://egon.stats.ucl.ac.uk/static/stat0019/slides/08_Survival/?panelset=estimates-%283%29#18

mod=1, or 2

Indicate the model you want to see

https://egon.stats.ucl.ac.uk/static/stat0019/slides/08_Survival/?panelset=estimates-%281%29#18

1 - No effect of any covariates fit.models formula data distr: which ones I want to fit sequentially

https://egon.stats.ucl.ac.uk/static/stat0019/slides/08_Survival/?panelset=running-the-model#18

now the specific effects come from a distribution

https://egon.stats.ucl.ac.uk/static/stat0019/slides/07_NMA/#19

with random effects with allow the partial pooling

https://egon.stats.ucl.ac.uk/static/stat0019/slides/07_NMA/#18

fixed effects means complete pooling

in black - direct effect

tercero- los resultados se contradicen. esto sugiere heterogeneidad.

ORck es una medida indirecta que se obtiene al relacionar ORc1(intervention c vs 1) con ORk1(intervention k vs 1)

si el OR es consistente con el que obtengo por comparación indirecta I can pool together all the studies.

https://egon.stats.ucl.ac.uk/static/stat0019/slides/07_NMA/#16

<- 0 comparación contra si mismo

https://egon.stats.ucl.ac.uk/static/stat0019/slides/07_NMA/#14

rs quick smoking successfully. Study s in time t. pst. mus and delta s are parametes. It is a regression models but without covariants. Estos pueden ser dificiles de obtener porque tengo los estudios y no siempre van a reportar todas las variables que me interesan ademas de que los datos están agregados porque lo que tengo es la publicación.

Ideal situation to have at least one study with direct evidence.

https://egon.stats.ucl.ac.uk/static/stat0019/slides/07_NMA/#8

Indicative of the interventions considered in each study. three columns because there is just maximum three interventions compared at the same time. Codes are 1,2,3,4 indicate the type of treatment (A,B,C,D). First column indicate the reference of the comparison

https://egon.stats.ucl.ac.uk/static/stat0019/slides/07_NMA/?panelset1=treatment-index#13

then total of population in each study https://egon.stats.ucl.ac.uk/static/stat0019/slides/07_NMA/?panelset1=sample-sizes#13

NA - 3 and 4 were not intervention considered in the study 1 columns maximum number interventions row number of studies

log OR of C(vsA) vs logs OR of B(vsA)

https://egon.stats.ucl.ac.uk/static/stat0019/slides/07_NMA/?panelset=indirect-effects#12