Here we describe the investigations that were collected from the public domain. The data from these investigations make up the training and challenge data for the CMI-Flu prediction challenge. An overview of all the data, including links to more detailed descriptions of the other data, may be found at CMI-x .

The Data

Change log - 20260821

see https://www.kaggle.com/competitions/cmi-flu-first-prediction-challenge/discussion/736185

Following a screen of the training data studies, vaccine_season and vaccine have been updated, based on a manual review of all 152 arms across 59 studies.

Changes per column:

column changed cells
vaccine_season 22
vaccine 12

The majority of studies have had missing data completed, 4 studies have been corrected. The studies affected are: SDY3010 ; SDY3186 ; SDY3193, SDY3427 ; SDY270; SDY272 ; SDY305 ; SDY314 ; SDY404 ; SDY56 ; SDY1393 ; SDY80 ; SDY112 ; SDY212 ; EXT100

The vaccine_strains_per_season.txt reference file has also been updated: H3N2 A/Victoria/361/2011 in seasons 2013-14 and 2014-15 has been updated to A/Texas/50/2012. This information also affects the virus_in_vaccine flag in the serology data; the serology data has also been updated.

Data tables

Investigations
Data dictionary
Column Description
study_accession Study identifier (links to the assay tables). One study may comprise multiple arms.
title Short title of the study.
description Free-text description of the study.
pubmed_ids PubMed IDs of publications associated with the study; semicolon-separated.
arm_id Identifier of the study arm/cohort (a group of participants receiving the same intervention). One row per arm.
arm_name Human-readable name of the arm.
arm_description Free-text description of the arm.
nr_participants Number of participants enrolled in the study.
vaccine_season Influenza season (year) in which the arm was vaccinated. NA where not recorded.
vaccine Vaccine type administered: TIV, QIV, IIV/IIV3/IIV8 (inactivated), or LAIV (live attenuated).
main_pubmed_id PubMed ID of the primary publication for the study.
year Publication year of the primary reference.
first_author First author of the primary publication.

Data preview:

invest = read_tsv('../datasets/260512/train/investigations.tsv')
head(invest)
## # A tibble: 6 × 13
##   study_accession title   description pubmed_ids arm_id arm_name arm_description
##   <chr>           <chr>   <chr>       <chr>      <chr>  <chr>    <chr>          
## 1 2016_UGA        2016_U… Annual vac… 38917104;… 2016_… 2016 UG… 2016 UGA Stand…
## 2 2016_UGA        2016_U… Annual vac… 38917104;… 2016_… 2016 UG… 2016 UGA High …
## 3 2017_UGA        2017_U… Annual vac… 38917104;… 2017_… 2017 UG… 2017 UGA Stand…
## 4 2017_UGA        2017_U… Annual vac… 38917104;… 2017_… 2017 UG… 2017 UGA High …
## 5 2018_UGA        2018_U… Annual vac… 36366544;… 2018_… 2018 UG… 2018 UGA Stand…
## 6 2018_UGA        2018_U… Annual vac… 36366544;… 2018_… 2018 UG… 2018 UGA High …
## # ℹ 6 more variables: vaccine_season <dbl>, vaccine <chr>,
## #   main_pubmed_id <dbl>, year <dbl>, first_author <chr>, nr_participants <dbl>

Data exploration

Each study contributes a number of participants to the training data. The bar for the 2025LJI challenge cohort is highlighted in red; all other studies are shown in blue; the dashed line shows the mean numberof participants per study. We have collected data from 4,805 participants across 59 studies.

study_n <- invest |>
  group_by(study_accession) |>
  summarise(
    nr_participants = first(nr_participants),
    first_author    = first(first_author),
    year            = first(year),
    .groups = "drop"
  ) |>
  mutate(
    label       = ifelse(is.na(first_author),
                         study_accession,
                         paste0(first_author, " (", year, ") - ", study_accession)),
    is_challenge = study_accession == "2025LJI"
  ) |>
  arrange(nr_participants) |>
  mutate(study_accession = factor(study_accession, levels = study_accession))

ggplot(study_n, aes(x = study_accession,
                    y = nr_participants,
                    fill = is_challenge)) +
  geom_col() +
  geom_hline(yintercept = mean(study_n$nr_participants),
             linetype = "dashed", color = "black") +
  coord_flip() +
  scale_x_discrete(labels = setNames(study_n$label, study_n$study_accession)) +
  scale_fill_manual(values = c(`FALSE` = "#2C7FB8", `TRUE` = "#E34A33"),
                    guide = "none") +
  labs(x = NULL, y = "Number of participants", title="Number of participants per study") +
  theme_minimal(base_size = 11)