General outline
SafeGrams are a way to describe the safety characteristics of a treatment - drug, vaccine, food item, whatever - from a passive (‘spontaneous’) reporting dataset. In order to do so, it calculates the proportional reporting rate (PRR) - the rate by which a particular side effect of a particular medication is reported compared to the same side effect of other medications.
Theory
Let \(D\) be an \(m \times n\) matrix that represents the frequency with which each of the \(m\) adverse effects occur for each of the \(n\) drugs, so that \(D_{i,j}\) (\(i \in m\), \(j \in n\)) represents the number of times the adverse effect \(j\) has occurred with the treatment \(i\).
For convenience’s sake, let \(D_{*,j}\) denote \(\sum_{i=1}^{m} D_{i,j}\), let \(D_{i,*}\) denote \(\sum_{j=1}^{n} D_{i,j}\), and let \(D_{*,*}\) denote \(\sum_{i=1}^{m} \sum_{j=1}^{n} D_{i,j}\). Furthermore, let \(D_{* \neg i, j}\) denote \(\sum_{k \neq i}^{m} D_{k,j}\) and \(D_{i, * \neg j}\) denote \(\sum_{k \neq j}^{n} D_{i, k}\).
Then, \(PRR\) can be calculated for each combination \(D_{i,j}\) by the following formula:
\[ PRR_{i,j} = \frac{D_{i,j} / D_{i,*}}{D_{* \neg i, j} / D_{* \neg i, *}} = \frac{D_{i,j}}{D_{i,*}} \cdot \frac{D_{*\neg i, *}}{D_{*\neg i, j}} \] Expanding this, we get
\[ PRR_{i,j} = \frac{D_{i,j}}{\displaystyle\sum_{q=1}^n D_{i,q}} \cdot \frac{\displaystyle\sum_{r=1, r\ne i}^{m} \displaystyle\sum_{s=1}^{n} D_{r,s}}{\displaystyle\sum_{t=1, t\ne i}^{m} D_{t,j}} \] While this looks daunting, it is in fact fairly straightforward computationally - given a 2x2 contingency table
| ADR of interest (\(j\)) | All other ADRs (\(\neg j\)) | TOTAL | |
|---|---|---|---|
| Treatment of interest (\(i\)) | \(a\) | \(b\) | \(a + b\) |
| All other treatments (\(\neg i\)) | \(c\) | \(d\) | \(c + d\) |
\(PRR\) would be calculated as
\[\frac{a/(a + b)}{c/(c + d)}\] This is quite easy to implement computationally.
Implementation
Loading the data
This example presupposes that the VAERS raw data files, from the CDC’s website, have been obtained and downloaded to a single folder named data\, then unzipped. VAERS tables are normalised into three separate tables: XXXXVAERSSYMPTOMS.csv, which contains symptom data, XXXXVAERSVAX.csv, which contains information about the vaccine and XXXXVAERSDATA.csv, which contains information about the recipient and the vaccination event. All are coded by VAERS_ID.
Loading the SYMPTOMS file
The SYMPTOMS file contains one or more rows per VAERS ID, with each row containing up to five symptoms in fields named SYMPTOM1 to SYMPTOM5 for names and SYMPTOMVERSION1 to SYMPTOMVERSION5 for symptom version identifiers. The importer function for the SYMPTOMS file
- imports the CSV file into a data frame,
- melts the data frame down to pair each
VAERS_IDwith a symptom, discarding the symptom sequential, and - filters
NAs.
read_symptoms_df <- function(year, folder = paste(getwd(), '/data', sep = "")){
df <- paste(folder, "/", year, "VAERSSYMPTOMS.csv", sep = "") %>%
read_csv() %>%
dplyr::select("VAERS_ID", "SYMPTOM1", "SYMPTOM2", "SYMPTOM3", "SYMPTOM4", "SYMPTOM5") %>%
melt(id.vars = "VAERS_ID", variable.name = "SYMPTOM_SEQ", value.name = "SYMPTOM_NAME") %>%
dplyr::select("VAERS_ID", "SYMPTOM_NAME") %>%
dplyr::filter(is.na(SYMPTOM_NAME) == FALSE)
return(df)
}Loading the VAX file
The VAX file contains details about the particular vaccine. In the present case, we only read in the VAERS_ID and the vaccine type (VAX_TYPE). In VAERS, a VAX_TYPE is a broader category of vaccines, ranking above the product name - for instance, ANTH is the VAX_TYPE corresponding to anthrax vaccines, while AVA would be the particular anthrax vaccine. Vaccines of unknown types (UNK) are excluded. A single VAERS_ID may have multiple vaccine entries, such as when the incident pertains to multiple coadministered vaccines.
read_vax_df <- function(year, folder = paste(getwd(), '/data', sep = "")){
df <- paste(folder, "/", year, "VAERSVAX.csv", sep = "") %>%
readr::read_csv() %>%
dplyr::select("VAERS_ID", "VAX_TYPE") %>%
reshape2::melt(id.vars = "VAERS_ID", variable.name = "VVAR", value.name = "VAX_TYPE") %>%
dplyr::select("VAERS_ID", "VAX_TYPE") %>%
dplyr::filter(is.na(VAX_TYPE) == FALSE & VAX_TYPE != "UNK")
return(df)
}Loading the DATA file
DATA files contain a record per incident, and as such they contain only one entry for each VAERS_ID.
read_data_df <- function(year, folder = paste(getwd(), '/data', sep = "")){
res <- paste(folder, "/", year, "VAERSDATA.csv", sep = "") %>%
read_csv() %>%
# Age and sex are included as stratifying factors.
dplyr::select("VAERS_ID", "AGE_YRS", "SEX") %>%
# Filter indeterminate or unknown sex
dplyr::filter(is.na(SEX) == FALSE, SEX != "U") %>%
# Add year marker
dplyr::mutate(YEAR = year)
return(res)
}Create multiyear tables
The multiyear table aggregator obtains a table using one of the read_*_df functions, and vertically joins all tables. The start and end arguments specify the start and end years, while the type argument specifies which data source to import. The type argument may take the following values:
datasymptomsvax
create_multiyear_df <- function(start, end, type = "data"){
if (type %in% c("data", "symptoms", "vax")){
start:end %>%
lapply(eval(parse(text = paste("read", type, "df", sep = "_")))) %>%
rbindlist(use.names = TRUE) %>%
return()
} else {
stop(sprintf("'%s' is not an accepted source frame type. Accepted source frames are: data, symptoms, vax.", type))
}
}Merge into a multi-year unitary table
We create a data.frame suitable for openEBGM, in which case identifiers are in the column id, vaccine names are named var1, symptoms var2 and strata variables are of the format strat_*.
create_unitary_df <- function(start, end){
res <- merge(x = create_multiyear_df(start, end, "vax"),
y = create_multiyear_df(start, end, "data"),
by = "VAERS_ID", allow.cartesian = TRUE) %>%
merge(y = create_multiyear_df(start, end, "symptoms"),
by = "VAERS_ID", allow.cartesian = TRUE) %>%
set_colnames(c("id", "var1", "strat_age", "strat_sex", "strat_year", "var2"))
return(res)
}Processing raw data frame and calculating RRs
Create raw-processed data frame
calculate_raw_PRRs <- function(start, end, excluded_issues = NULL, exclude_normals = TRUE, stratify = FALSE, digits = 3, max_cats = 10) {
df <- create_unitary_df(start, end)
if (exclude_normals == TRUE) {
df <- dplyr::filter(df, !grepl(" (normal|negative)", df$var2, ignore.case = TRUE))
}
if (!is.null(excluded_issues)) {
df <- dplyr::filter(df, !var2 %in% excluded_issues)
}
processRaw(df, stratify = stratify, zeroes = FALSE, digits = digits, max_cats = max_cats) %>%
return()
}
raw_PRRs <- calculate_raw_PRRs(2006, 2017)Map vaccines to categories
Sometimes, it’s useful to map vaccines to a particular category, e.g. all influenza vaccines. The following preferred mapping implements a map that categorises vaccines according to pathogens, somewhat arbitrarily, and stores it as a variable vaccine_categories, which can later be reused.
| VAX | CLASS |
|---|---|
| 6VAX-F | Hexavax |
| ADEN_4_7 | Adenovirus |
| ANTH | Anthrax |
| BCG | BCG |
| CHOL | Cholera |
| DPP | DTP and similar |
| DT | DTP and similar |
| DTAP | DTP and similar |
| DTAPH | DTP and similar |
| DTAPHEPBIP | DTP and similar |
| DTAPIPV | DTP and similar |
| DTAPIPVHIB | DTP and similar |
| DTIPV | DTP and similar |
| DTOX | Diphteria |
| DTP | DTP and similar |
| DTPHEP | DTP and similar |
| DTPHIB | DTP and similar |
| DTPIHI | DTP and similar |
| DTPIPV | DTP and similar |
| DTPPHIB | DTP and similar |
| FLU(H1N1) | Influenza |
| FLU3 | Influenza |
| FLU4 | Influenza |
| FLUA3 | Influenza |
| FLUC3 | Influenza |
| FLUC4 | Influenza |
| FLUN(H1N1) | Influenza |
| FLUN3 | Influenza |
| FLUN4 | Influenza |
| FLUR3 | Influenza |
| FLUR4 | Influenza |
| FLUX | Influenza |
| FLUX(H1N1) | Influenza |
| H5N1 | Influenza |
| HBHEPB | H.influenzae b |
| HBPV | H.influenzae b |
| HEP | Hepatitis |
| HEPA | Hepatitis |
| HEPAB | Hepatitis |
| HEPATYP | Hepatitis |
| HIBV | H.influenzae b |
| HPV2 | HPV |
| HPV4 | HPV |
| HPV9 | HPV |
| HPVX | HPV |
| IPV | Polio (injectable) |
| JEV | JEV |
| JEV1 | JEV |
| JEVX | JEV |
| LYME | Lyme |
| MEA | Measles |
| MEN | Meningococcal |
| MENB | Meningococcal |
| MENHIB | Meningococcal |
| MER | Measles |
| MM | Measles |
| MMR | MMR +/- varicella |
| MMRV | MMR +/- varicella |
| MNQ | Meningococcal |
| MNQHIB | Meningococcal |
| MU | Mumps |
| MUR | Mumps |
| OPV | Polio (oral) |
| PER | Pertussis |
| PLAGUE | Plague |
| PNC | Pneumococcal |
| PNC10 | Pneumococcal |
| PNC13 | Pneumococcal |
| PPV | Pneumococcal |
| RAB | Rabies |
| RUB | Rubella |
| RV | Rotavirus |
| RV1 | Rotavirus |
| RV5 | Rotavirus |
| RVX | Rotavirus |
| SMALL | Smallpox |
| SSEV | Summer/spring encephalitis |
| TBE | Tick-borne encephalitis |
| TD | DTP and similar |
| TDAP | DTP and similar |
| TDAPIPV | DTP and similar |
| TTOX | Tetanus |
| TYP | Typhoid |
| VARCEL | Varicella |
| VARZOS | Zoster |
| YF | Yellow fever |
Plot SafeGrams by vaccine category
The base plotting function ingests a data frame and creates a faceted result.
plot_safegram <- function(df, vaccines = NULL, exclude_inf = TRUE, exclude_issues = NULL, N_cutoff = 4, prr_cutoff = 0, prr_significance_margin = 3, wrap_by = "CLASS"){
if (is.null(vaccines)) {
vax <- unique(df$var1)
} else {
vax <- vaccines
}
if (exclude_inf == TRUE) {
df <- dplyr::filter(df, PRR != Inf)
}
if (is.null(exclude_issues) == FALSE) {
df <- dplyr::filter(df, !var2 %in% exclude_issues)
}
plot <- df %>%
dplyr::filter(N > N_cutoff) %>%
dplyr::filter(PRR > prr_cutoff) %>%
dplyr::filter(var1 %in% vax) %>%
ggplot(aes(x = PRR, y = N))
plot +
theme_grey() +
scale_x_log10() + xlab("Proportional Reporting Ratio (PRR)") +
scale_y_log10() + ylab("Number of reports") +
stat_density2d(aes(fill = log(..level..), alpha = ..level..), geom = "polygon") +
scale_fill_gradient2(high = "#ffffbf", mid = "#fdae61", low = "#3288bd", guide = FALSE) +
scale_alpha_continuous(range = c(0.3, 0.6), guide = FALSE) +
geom_vline(aes(xintercept = prr_significance_margin)) +
facet_wrap(eval(parse(text = paste("~", wrap_by))))
}Finally, let’s create a wrap-up call that calculates the PRR table, appends the category mapping from the function above, then plot the results:
raw_PRRs %>%
categorise_vaccines(vaccine_categories) %>%
plot_safegram()
Plot SafeGrams for a particular vaccine or vaccines
The plotting function can be also called to plot a single or limited number of vaccines, by providing the vaccines parameter. If segmentation should be by vaccine type rather than vaccine class (e.g. different HPV subclasses in lieu of HPV collated), the wrap_by parameter should be set to var1, the column name for vaccines. In this case, the four HPV vaccine identifiers (HPV2 (bivalent), HPV4 (quadrivalent), HPV9 (nonavalent) and HPVX (unknown/unreported valency)) are compared:
raw_PRRs %>%
plot_safegram(vaccines = c("HPV2", "HPV4", "HPV9", "HPVX"), wrap_by = "var1")
Plotting with exclusion lists
The preprocessing code automatically excludes ‘normal’ test results, e.g. Biopsy kidney normal. The following are crude but efficient methods to identify three classes of symptom entities that should be excluded:
- Vaccine failure events: these events constitute issues where the adverse event pertains to a reaction that the vaccine was supposed to prevent, e.g. positive HPV tests (
Human papilloma virus test positive) for the HPV vaccine. - Handling, storage and administration issues: in these events, the vaccine was incorrectly administered, and as such the event pertains to the vaccine but rather to the event of administration.
- Non-reactions: these are MedDRA codes that do not describe a reaction but rather a treatment, such as
Resuscitation, or a test (rather than a test result), such asBlood culture. As a rule,negativeornormalresults are routinely filtered at raw PRR calculation (functioncalculate_raw_PRRs(), supra). This leaves only positive test results.
It is practically impossible to comb the over 5,000 individual MedDRA entities mentioned in the filtered list. Therefore, a preferred approach is as follows.
- Create a table comprising the 500 highest-PRR entities and identify those that would constitute errors of group 1.
- Create a table comprising the 500 most frequently used entities overall, and identify from that list the elements that fit into groups 2 (handling, storage & administration) and 3 (non-reactions).
Vaccine failure events
Vaccine failure events can be identified better from the high-PRR events as they are highly specific to the individual vaccine. To filter out low-\(N\) events, only events above a certain minimum (set at a multiplier \(m\) times the number of years) are considered:
identify_top_PRRs <- function(start, end, multiplier = 3) {
calculate_raw_PRRs(start, end) %>%
dplyr::filter(PRR != Inf) %>%
dplyr::filter(N > length(start:end)) %>%
dplyr::arrange(desc(PRR)) %>%
head(500)
}Based on this, the items with the highest \(PRR\)s were reviewed, and from this, the following were isolated as potential vaccine failure markers:
identify_top_PRRs(2006, 2017) %>%
head(500) %>%
dplyr::filter(var2 %in% exclusion_failure_markers) %>%
set_colnames(c("Vaccine", "Symptom", "N", "E", "RR", "PRR")) %>%
dplyr::arrange(desc(N)) %>%
kable("html", escape = FALSE) %>%
kable_styling("striped")| Vaccine | Symptom | N | E | RR | PRR |
|---|---|---|---|---|---|
| VARZOS | Herpes zoster | 5124 | 619.9139675 | 8.266 | 30.479 |
| VARCEL | Varicella post vaccine | 1390 | 171.7652388 | 8.092 | 90.937 |
| VARCEL | Drug ineffective | 648 | 98.0052158 | 6.612 | 23.685 |
| HPV4 | Smear cervix abnormal | 517 | 49.4054237 | 10.464 | 709.481 |
| MMR | Mumps | 356 | 30.6899979 | 11.600 | 550.679 |
| HPV4 | Papilloma viral infection | 348 | 33.2826614 | 10.456 | 668.586 |
| MMR | Rash morbilliform | 344 | 40.9199972 | 8.407 | 26.606 |
| HPV4 | Human papilloma virus test positive | 318 | 30.9255324 | 10.283 | 305.475 |
| RV5 | Rotavirus test positive | 219 | 7.8029454 | 28.066 | 192.720 |
| VARCEL | Varicella virus test positive | 190 | 33.3513684 | 5.697 | 14.361 |
| HEP | No therapeutic response | 174 | 12.7857577 | 13.609 | 30.357 |
| HPV4 | Cervical dysplasia | 150 | 14.5199146 | 10.331 | 360.230 |
| MMR | Measles | 126 | 12.5127264 | 10.070 | 62.015 |
| HPV4 | Anogenital warts | 125 | 13.0113520 | 9.607 | 92.367 |
| VARZOS | Ophthalmic herpes zoster | 122 | 12.4894432 | 9.768 | 81.083 |
| RV5 | Rotavirus infection | 119 | 4.3757694 | 27.195 | 157.080 |
| MMR | Measles antibody positive | 79 | 8.0318176 | 9.836 | 53.463 |
| RV5 | Gastroenteritis rotavirus | 70 | 2.4173831 | 28.957 | 246.400 |
| HEP | Hepatitis B surface antigen positive | 67 | 3.4794029 | 19.256 | 95.704 |
| VARZOS | Herpes zoster ophthalmic | 67 | 7.2019417 | 9.303 | 55.662 |
| MMR | Rubella antibody positive | 65 | 6.9327268 | 9.376 | 41.401 |
| HPV4 | Smear cervix | 59 | 6.1285354 | 9.627 | 94.460 |
| FLUN4 | Influenza A virus test positive | 59 | 1.9483179 | 30.283 | 39.828 |
| RV5 | Culture stool positive | 58 | 2.3255837 | 24.940 | 102.080 |
| MMR | Vaccine breakthrough infection | 58 | 6.1718178 | 9.398 | 41.868 |
| MMR | Mumps antibody test positive | 57 | 5.7490905 | 9.915 | 56.109 |
| FLUN3 | Influenza A virus test positive | 47 | 3.6614550 | 12.836 | 15.719 |
| HEP | Therapeutic response decreased | 43 | 2.7248336 | 15.781 | 44.670 |
| PNC | Blood culture positive | 43 | 4.5929233 | 9.362 | 13.801 |
| FLUN3 | Influenza virus test positive | 41 | 1.7697033 | 23.168 | 35.286 |
| VARCEL | Infection transmission via personal contact | 38 | 6.8296318 | 5.564 | 13.447 |
| MMR | Measles antibody | 37 | 3.8890906 | 9.514 | 44.515 |
| FLUN4 | Influenza virus test positive | 34 | 0.9416870 | 36.105 | 50.661 |
| MMR | Mumps antibody test | 34 | 3.7199997 | 9.140 | 36.815 |
| HEP | Hepatitis B antibody | 33 | 2.0121848 | 16.400 | 50.280 |
| HPV4 | Cervix carcinoma | 32 | 3.1114103 | 10.285 | 307.396 |
| FLUN3 | Influenza serology positive | 32 | 1.7391911 | 18.399 | 25.189 |
| LYME | Lyme disease | 31 | 0.0489647 | 633.110 | 954.346 |
| SMALL | Vaccinia | 28 | 0.2604895 | 107.490 | 1101.396 |
| HEP | Hepatitis B antigen positive | 28 | 1.2576155 | 22.264 | 319.965 |
| RV5 | Rotavirus test | 28 | 1.2239914 | 22.876 | 73.920 |
| SMALL | Vaccinia virus infection | 27 | 0.2520866 | 107.106 | 1062.060 |
| MMR | Rubella | 27 | 2.3672726 | 11.406 | 292.355 |
| HPV4 | Loop electrosurgical excision procedure | 27 | 2.7342696 | 9.875 | 129.683 |
| MMR | Rubella antibody test | 26 | 3.6354543 | 7.152 | 16.560 |
| PNC | Pneumococcal bacteraemia | 25 | 1.4075088 | 17.762 | 49.996 |
| MNQ | Meningitis meningococcal | 25 | 2.0716312 | 12.068 | 35.126 |
| HEP | Therapy non-responder | 24 | 1.3833771 | 17.349 | 60.946 |
| PNC | Meningitis pneumococcal | 24 | 1.6297470 | 14.726 | 31.198 |
| HPV4 | Dysplasia | 23 | 2.2628438 | 10.164 | 220.941 |
| RV1 | Rotavirus test positive | 23 | 1.3084743 | 17.578 | 19.221 |
| LYME | Borrelia burgdorferi serology | 21 | 0.0399168 | 526.094 | 730.297 |
| FLUN(H1N1) | Influenza serology positive | 21 | 1.0053974 | 20.887 | 25.378 |
| FLUN3 | Influenza B virus test positive | 21 | 1.2052290 | 17.424 | 23.371 |
| FLUN4 | Exposure via direct contact | 18 | 0.5439054 | 33.094 | 44.884 |
| FLUN4 | Influenza B virus test positive | 18 | 0.6413213 | 28.067 | 36.054 |
| HEP | Hepatitis B antibody positive | 18 | 1.3414565 | 13.418 | 29.385 |
| VARZOS | Herpes zoster oticus | 18 | 2.9172422 | 6.170 | 12.818 |
| HIBV | Haemophilus infection | 17 | 1.0673577 | 15.927 | 58.221 |
| HIBV | Rotavirus test | 17 | 1.8562743 | 9.158 | 15.188 |
| PNC | Rotavirus test | 16 | 1.4815882 | 10.799 | 17.332 |
| PNC | Pneumococcal infection | 15 | 1.3704690 | 10.945 | 17.726 |
| DTAPIPVHIB | Culture stool positive | 15 | 1.4038207 | 10.685 | 13.067 |
Handling, storage & administration and non-reactions
In these cases, the event either pertains to the handling, storage and administration of the vaccine, or does not denote a reaction (e.g. denoting a test without a result). For this purpose, the 500 most frequently occurring events, regardless of vaccine, will be considered, therefore the events will be aggregated.
all_markers = c(exclusion_handling_markers, exclusion_non_reaction_markers)
identify_top_Ns <- function(start, end) {
calculate_raw_PRRs(start, end) %>%
head(500) %>%
dplyr::filter(PRR != Inf) %>%
dplyr::filter(var2 %in% all_markers) %>%
dplyr::select("var2", "N", "PRR") %>%
plyr::ddply(.(var2), summarize, sum_N = sum(N), avg_PRR = mean(PRR)) %>%
dplyr::arrange(desc(sum_N))
}From a manual review of the top 500 entities by \(N\), two lists were created, one containing handling, storage & administration issues (exclusion_handling_markers) marked in orange in the table below, and of non-reactions (exclusion_non_reaction_markers), marked in green on the table below.
| Symptom/s | N | PRR |
|---|---|---|
| Enema administration | 2 | 12.5830 |
| Blood thyroid stimulating hormone | 1 | 12.3950 |
| Culture stool | 1 | 12.3740 |
| C-reactive protein | 1 | 9.2860 |
| Resuscitation | 1 | 8.7140 |
| Autopsy | 1 | 8.6330 |
| Endotracheal intubation | 1 | 7.7050 |
| Similar reaction on previous exposure to drug | 1 | 7.7010 |
| Culture | 1 | 6.0850 |
| Computerised tomogram | 3 | 5.8300 |
| Lumbar puncture | 3 | 4.8550 |
| Electrocardiogram | 2 | 4.6470 |
| Extra dose administered | 3 | 3.6935 |
| Urine analysis | 2 | 3.2790 |
| Metabolic function test | 2 | 3.1120 |
| Immunoglobulin therapy | 1 | 2.5810 |
| Chest X-ray | 1 | 2.4940 |
| Intensive care | 1 | 2.3800 |
| Full blood count | 5 | 2.2710 |
| Electroencephalogram | 1 | 2.1910 |
| Expired product administered | 1 | 1.6930 |
| Exposure during pregnancy | 1 | 1.5180 |
| Laboratory test | 2 | 1.4360 |
| Blood test | 4 | 1.3515 |
| Immediate post-injection reaction | 3 | 1.0580 |
| No adverse event | 5 | 0.7660 |
| Expired drug administered | 1 | 0.7480 |
| Unevaluable event | 1 | 0.7410 |
| Inappropriate schedule of drug administration | 2 | 0.7240 |
| Drug administered to patient of inappropriate age | 1 | 0.7090 |
Plot with exclusion lists applied
raw_PRRs_with_exclusions <- calculate_raw_PRRs(2006, 2017, excluded_issues = c(exclusion_non_reaction_markers, exclusion_handling_markers, exclusion_failure_markers))
raw_PRRs_with_exclusions %>%
categorise_vaccines(vaccine_categories) %>%
plot_safegram()
SafeGram of major vaccine classes, 2006-2017
Now, we can also plot the adjusted SafeGram for the HPV vaccines by vaccine type:
raw_PRRs_with_exclusions %>%
plot_safegram(vaccines = c("HPV2", "HPV4", "HPV9", "HPVX"), wrap_by = "var1")
SafeGrams of HPV vaccines by valency, 2006-2017
And we can compare specific flu vaccines with different safety records:
raw_PRRs_with_exclusions %>%
plot_safegram(vaccines = c("FLUA3", "FLU(H1N1)", "FLU4", "FLU3"), wrap_by = "var1", N_cutoff = 0)
SafeGrams of injectable flu vaccines, 2006-2017
As well as compare the nasal versus non-nasal flu vaccines:
with different safety records:
raw_PRRs_with_exclusions %>%
plot_safegram(vaccines = c("FLU(H1N1)", "FLUN(H1N1)", "FLU3", "FLUN3", "FLU4", "FLUN4"), wrap_by = "var1", N_cutoff = 0)
SafeGrams of various flu vaccines, 2006-2017
Afterword
SafeGram is a great way to create intuitive comparative visualisations of safety of drugs, vaccines and other products from passive reporting databases. While passive reporting suffers from its own weaknesses, such as underreporting, lack of definite knowledge of the extent of underreporting and other methodological malaises, the \(PRR\) approach is a good start to get useful data out of what many are all too happy to discard as useless at best. With the right methods and intuitive visuals, it can provide clinically actionable intelligence to decision-makers, clinicians and patients alike as to the safety of particular pharmaceuticalss.