Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Update markdown lint issue with headings #86

Merged
merged 1 commit into from
Oct 12, 2024
Merged

Update markdown lint issue with headings #86

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
25 changes: 10 additions & 15 deletions vignettes/cohort_design.Rmd
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -20,11 +20,11 @@ knitr::opts_chunk$set(
library(vaccineff)
```

# Cohort Design
## Cohort Design

In cohort studies, vaccine effectiveness ($VE$) is calculated after following vaccinated and unvaccinated individuals over time. However, VE may depend on individual characteristics (e.g., age and sex) and external factors (e.g., environment, number of doses, virus strain, and calendar period). Typically, VE is estimated in open or dynamic cohorts, where individuals can enter or leave the cohort at any time after the initial study date and change their vaccination status. Thus, $VE$ can be estimated by survival analysis methods, substituting the relative risk for the hazard ratio ($HR$), which considers the person-time at risk in the estimation of VE [@bookvaccine]. `Vaccineff` package estimates VE using $(1-HR) \times 100\%$ and therefore, the outcome variable is the time from the initial study date to the occurrence of the event (e.g., infection, death, hospitalization) or the date of the end of the study. VE is approximated by the Cox proportional hazards model that is usually written by the following expression:

\begin{equation}\label{eq:cox}
\begin{equation}\label{eq:cox}
h(t,X)=h_0(t) e ^{\sum_{i=1}^p \beta_i X_i},
\end{equation}

Expand All @@ -34,20 +34,18 @@ $$HR=\frac{h_0(t) e ^{ \beta_1 X_1|X=1}}{h_0(t) e ^{ \hat \beta_1 X_1|X=0}}=e^{

therefore, the estimated $HR$ does not depend on time $t$, and $h_0 (t)$ can remain unspecified, assuming that the effect of an explanatory variable is proportional over time representing *the proportional hazard (PH) assumption*. When $HR<1$ indicates a reduction in the risk of an event of interest over time and the context of VE, it will be the expected outcome. *Vaccineff* uses the Schoenfeld residual test to determine whether the global PH assumption of the model is violated.

When working with observational data, other potential factors or baseline characteristics besides vaccination status could explain the observed differences in the risk of the event (e.g., death, reinfection) between vaccinated and unvaccinated individuals. `Vaccineff` allows the estimation of VE with and without running a *iterative matching process (IMP)* to emulate the balance achieved by a trial design on potential confounders without involving the calendar period which implies that the disease incidence has not changed during the study period.
When working with observational data, other potential factors or baseline characteristics besides vaccination status could explain the observed differences in the risk of the event (e.g., death, reinfection) between vaccinated and unvaccinated individuals. `Vaccineff` allows the estimation of VE with and without running a *iterative matching process (IMP)* to emulate the balance achieved by a trial design on potential confounders without involving the calendar period which implies that the disease incidence has not changed during the study period.

`vaccineff` package uses the number of days elapsed from the vaccine administration to the occurrence of the event to measure the length of follow-up.


![](https://raw.githubusercontent.com/epiverse-trace/vaccineff/main/inst/images/time-to-event1.png){.rmarkdown-img align="center" style="margin-left: 2.8em; margin-top: 0.8em; margin-bottom: 0.8em;" width="560"}

## Estimating VE without iterative matching process
### Estimating VE without iterative matching process

`vaccineff` performs the conventional estimation of VE based on the Cox regression model using the option `match=FALSE` in `make_vaccineff_data` function. This is the simplest possible model, as it does not take into account that VE can be affected by possible confounding factors, such as age or sex. To begin the analysis, the first step is to transform the original data into a `vaccineff` data by `make_vaccineff_data` function and declare the names of variables containing relevant information such as the date of outcome or event occurs `outcome_date_col` or the censoring date `censoring_date_col`, date of last dose `vacc_date_col`, labels to identify vaccinated `vaccinated_status` and unvaccinated groups `unvaccinated_status`, and the date of last follow up of the cohort `end_cohort`. The `censoring_date_col` option should be used if the study end date varies for some subjects because an event other than the study outcome has occurred, for example, when a subject dies from causes other than the disease of interest.

The crude VE (unadjusted) is then estimated and the survival curves can also be visualized by the following code:


```{r artcohor, include = TRUE, echo = TRUE}
# Load example data
data("cohortdata")
Expand Down Expand Up @@ -80,29 +78,26 @@ plot(ve1, type = "loglog")

The crude VE of death from two doses was 68.9% [95% CI 54.6-78.7]. Furthermore, since the log-log plot shows that the curves do not appear entirely parallel, this indicates that there is a slight violation of the proportional assumption, which in many data sets can be resolved after adjusting for possible confounding factors.


## Estimating VE with iterative matching process
### Estimating VE with iterative matching process

`vaccineff` performs an IMP after the end of the study by dividing the cohort between those who were never vaccinated and those who received the vaccine at some point during follow-up. This method should be seen as an attempt to control for potential confounders that could influence the vaccine status as well as the occurrence of the interesting event. IMP selects pairs of unvaccinated and vaccinated individuals with similar characteristics (potential confounders, e.g., age, sex), making the groups comparable on important confounders variables, as shown in the figure below:

![](https://raw.githubusercontent.com/epiverse-trace/vaccineff/main/inst/images/static_matching.png){.rmarkdown-img align="center" style="margin-left: 2.8em; margin-top: 0.8em; margin-bottom: 0.8em;" width="560"}

IMP runs a nearest neighbor matching using Mahalanobis distance to analyze similarities between a pair of unvaccinated and vaccinated subjects as follows:
IMP runs a nearest neighbor matching using Mahalanobis distance to analyze similarities between a pair of unvaccinated and vaccinated subjects as follows:

$$d^2 (u,v)=(x_u-x_v)^T \Sigma^{-1} (x_u-x_v),$$

where $x$ corresponds to values of confounders variables y $\Sigma$ is the sample variance-covariance matrix to standardize the variables. Thus, IMP selects the unvaccinated individual with the shortest distance for each person in the vaccinated group. However, all these pairings are provisional because the IMP procedure checks for each pairing that the matched unvaccinated individual has not developed the outcome (e.g., death) before the immunization date of the vaccinated partner. This point is important because the follow-up start date for both subjects corresponds to the immunization date of the vaccinated subject plus the number of days required for the vaccine to have a protective effect:

![](https://raw.githubusercontent.com/epiverse-trace/vaccineff/main/inst/images/time-to-event2.png){.rmarkdown-img align="center" style="margin-left: 2.8em; margin-top: 0.8em; margin-bottom: 0.8em;" width="560"}

For this reason, the algorithm iterates until the largest number of matched pairs with equivalent exposure time is obtained. At the end of the procedure, unvaccinated and vaccinated individuals who cannot be matched are eliminated from the analysis. The estimation of VE with IMP using `vaccineff` is performed using a Cox regression model with a robust variance estimator to account for the clustering within matched pairs @austin2014use. In our example, we decided to control for age and sex using the options `match=TRUE` and `exact = c("age", "sex")` in the `make_vaccineff_data` function. Now, this function creates a set of matched data, look at the new dataset using `vaccineff_data_matched[["matching"]]$match`.
For this reason, the algorithm iterates until the largest number of matched pairs with equivalent exposure time is obtained. At the end of the procedure, unvaccinated and vaccinated individuals who cannot be matched are eliminated from the analysis. The estimation of VE with IMP using `vaccineff` is performed using a Cox regression model with a robust variance estimator to account for the clustering within matched pairs @austin2014use. In our example, we decided to control for age and sex using the options `match=TRUE` and `exact = c("age", "sex")` in the `make_vaccineff_data` function. Now, this function creates a set of matched data, look at the new dataset using `vaccineff_data_matched[["matching"]]$match`.

When the `exact` option is used, the IMP searches for each case, a control with the same characteristics. In our example, the IMP matches pairs of the exact same sex and age. If an exact match is not required, a `nearest` option is also available to match similar but not identical pairs. This option could make possible to match pairs with a caliper or distance of 1 on age (e.g., case of 49 years with a control of 50 years). Note also that both options can be used simultaneously for the IMP process.


The following code block estimates a VE using IMP matching only considering the `exact` option.


```{r artcohor1, include = TRUE, echo = TRUE}
# Load example data
data("cohortdata")
Expand All @@ -122,13 +117,14 @@ vaccineff_data_matched <- make_vaccineff_data(
nearest = NULL
)
```

If the `summary` function is applied to a vaccineff object, a report of exact IMP is displayed.

```{r artcohor2, include = TRUE, echo = TRUE}
summary(vaccineff_data_matched)
```

In the output of `vaccineff_data_matched`, the user can check the differences between the unmatched and matched cohorts using standardized mean difference (SMD) to measure the distance between the unvaccinated and vaccinated individuals. In our example, SMD for age and sex is 0 because an exact IMP was run, see the results in the balance matched output. The number of unmatched individuals (removed) from the analysis is also reported. In this example, out of a total of 62743 unvaccinated individuals and 37257 vaccinated individuals, the IMP procedure was able to match 27612 pairs considering the characteristics of the individuals and the time of exposure.
In the output of `vaccineff_data_matched`, the user can check the differences between the unmatched and matched cohorts using standardized mean difference (SMD) to measure the distance between the unvaccinated and vaccinated individuals. In our example, SMD for age and sex is 0 because an exact IMP was run, see the results in the balance matched output. The number of unmatched individuals (removed) from the analysis is also reported. In this example, out of a total of 62743 unvaccinated individuals and 37257 vaccinated individuals, the IMP procedure was able to match 27612 pairs considering the characteristics of the individuals and the time of exposure.
Now, using `the vaccineff_data_matched` object in the `effectiveness` function, the VE is estimated.

```{r artcohor3, include = TRUE, echo = TRUE}
Expand Down Expand Up @@ -165,7 +161,7 @@ vaccineff_data_matched2 <- make_vaccineff_data(
end_cohort = as.Date("2044-12-31"),
match = TRUE,
exact = c("sex"),
nearest = c("age"=2)
nearest = c("age" = 2)
)

summary(vaccineff_data_matched2)
Expand All @@ -182,7 +178,6 @@ plot(ve3, type = "loglog")

# Generate Survival plot
plot(ve3, type = "surv", percentage = FALSE, cumulative = FALSE)

```

## References
10 changes: 5 additions & 5 deletions vignettes/other_designs.Rmd
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -16,23 +16,23 @@ knitr::opts_chunk$set(
)
```

# Test Negative Design (Future release)
## Test Negative Design (Future release)

The test-negative design is a modified case-control design. $VE$ is estimated using the odds ratio ($OR$) of vaccination among cases divided by the OR of vaccination among controls,

$$VE = 1- \frac{\text{Odds cases}}{\text{Odds Control}}\times 100.$$

# Screening Method (Future release)
## Screening Method (Future release)

The screening method estimates $VE$ using the proportion of cases vaccinated ($PCV$) and the proportion of the population vaccinated ($PPV$),

$$VE = \frac{\text{PCV}}{1-\text{PCV}} \times \frac{1-\text{PPV}}{\text{PPV}} \times 100.$$

To calculate the $PCV$, it is assumed that all, or at least a random sample of cases of a disease arising over a given period in a defined population, is available. Additionally, it incorporates the $PPV$ as prior information or an external estimate.

# What type of data is needed?
## What type of data is needed?

## Data for Test-negative design
### Data for Test-negative design

Data disaggregated at the individual level to correlate vaccination and disease status for each case. The dataset must contain the following information:

Expand All @@ -42,7 +42,7 @@ Data disaggregated at the individual level to correlate vaccination and disease

- Individuals' demographic information (e.g., sex, age, health insurance) and other variables that may be specified depending on the confounding variables to be introduced in the model.

## Data for Screening method design
### Data for Screening method design

- Vaccination status for each case or age group. Consider time since vaccination to determine whether individuals are vaccinated or unvaccinated. Exclude patients with incomplete vaccination schedules if the vaccine of interest requires several doses for maximum efficacy.

Expand Down
7 changes: 2 additions & 5 deletions vignettes/vaccineff.Rmd
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -30,10 +30,9 @@ Vaccines are created to offer protection against diseases that affect human heal

`vaccineff` is useful for local, national, and international health agencies looking for a quick implementation to estimate $VE$ based on their available data. It also provides insights to researchers, data analysts, and epidemiology students on how to approach $VE$ using different methods. We believe that `vaccineff` would be specially useful for users without advanced training in statistical methods.


## What is vaccine effectiveness?

In contrast with vaccine efficacy, which is the percentage reduction of disease incidence in a vaccinated group compared with an unvaccinated group under ideal conditions, $VE$ is the percentage reduction of disease incidence in a vaccinated group compared with an unvaccinated group under routine conditions. The reduction attributable to vaccination is usually assessed from data collected in observational studies [@bookvaccine]. Evaluating the effectiveness of vaccines in the field is an important aspect of monitoring immunization programs.
In contrast with vaccine efficacy, which is the percentage reduction of disease incidence in a vaccinated group compared with an unvaccinated group under ideal conditions, $VE$ is the percentage reduction of disease incidence in a vaccinated group compared with an unvaccinated group under routine conditions. The reduction attributable to vaccination is usually assessed from data collected in observational studies [@bookvaccine]. Evaluating the effectiveness of vaccines in the field is an important aspect of monitoring immunization programs.

## For which designs is this package?

Expand Down Expand Up @@ -79,14 +78,12 @@ The current release of the package bases the estimation of $VE$ in the cohort de

The integrated dataset `cohortdata` serves as a minimal example of the package's input. The data is accessed using `data("cohortdata")`.


`vaccineff` has three main functions:


1. `make_vaccineff_data`: This function returns an S3 object of the class `vaccineff_data` with the relevant information for the study. This function also allows to create a matched cohort to control for confounding variables by setting `match = TRUE` and passing the corresponding `exact` and `nearest` arguments.
`make_vaccineff_data` supports the method `summary()` to check the characteristics of the cohort, the matching balance and the sizes of matched, excluded, and removed populations.

2. `plot_coverage`: This function returns a plot of the vaccine coverage or the cumulative coverage. If the population is matched, the plot also includes the resulting count of doses after matching.
2. `plot_coverage`: This function returns a plot of the vaccine coverage or the cumulative coverage. If the population is matched, the plot also includes the resulting count of doses after matching.

3. `effectiveness`: This function provides methods for estimating VE using the $HR$. A summary of the estimation can be obtained using `summary()` and a graphical representation of the methodology is generated by `plot().`

Expand Down
Loading