Every outcomes paper has to answer one question before any analysis: how did you get from the patients you started with to the patients you actually analysed? The CONSORT diagram is that answer drawn as a flow chart. It is the box-and-arrow figure near the front of the methods section: a top box of everyone screened, a side box at each step listing how many were dropped and why, and a bottom box of the final analysed cohort. A reviewer reads it to check that your exclusions are defensible and that the numbers add up.
Reach for it whenever you have built a study cohort by applying a sequence of exclusion rules and you need to account for every patient between the registry pull and the analysis set. hv_consort() (Ehrlinger 2026) builds the diagram through a multi-step API: hv_consort_start() initializes a patient-level tracker, hv_consort_exclude() adds an exclusion stage, and hv_consort() renders the result.
One thing sets this figure apart from the rest of the book. The diagram is drawn by the consort package (Dayim 2026) using grid graphics, not ggplot. So unlike every other plot here, you do not finish it with + theme_*() or any other ggplot layer. There is nothing to add a + to; the object renders itself. If you want to change its look, you do it through the constructor’s own arguments, not through ggplot. The tracking helpers below give you the exclusion-count audit you would otherwise build by hand.
The tracker starts from a data frame with one row per patient and whatever columns your exclusion rules will test. sample_consort_data() simulates a cardiac-surgery cohort and returns a finished three-stage hv_consort_tracker, which is the quickest way to see the shape of the object.
tracker <- sample_consort_data(n = 300, seed = 42)
tracker<hv_consort_tracker>
Patients : 300
ID column : patient_id
Stages : 3
[screened] Screened -- N = 300
-> excl [excl_screen]: 73
[eligible] Eligible -- N = 227
-> excl [excl_eligible]: 41
[analyzed] Analyzed -- N = 186To build a tracker from your own data, start with hv_consort_start(), which records the patient identifier and marks every patient as screened. Each hv_consort_exclude() call then adds one exclusion stage. The rules are two-sided formulas, <condition> ~ "<reason>", evaluated against the data. The first matching rule wins, and patients dropped in an earlier stage are skipped in later ones, so each stage operates only on the survivors of the last. That last point is what keeps the counts honest: a patient is removed once, for the first reason that applies, and is not counted again.
set.seed(42)
cohort <- data.frame(
mrn = paste0("P", 1:300),
age = sample(12:85, 300, replace = TRUE),
had_surgery = sample(c(TRUE, FALSE), 300, replace = TRUE, prob = c(0.9, 0.1)),
echo = sample(c(TRUE, FALSE), 300, replace = TRUE, prob = c(0.85, 0.15))
)
tracker2 <- hv_consort_start(cohort, patient_id = mrn, label = "Screened") |>
hv_consort_exclude(
label = "Eligible",
col = "excl_screen",
age < 18 ~ "Age < 18",
!had_surgery ~ "No qualifying surgery"
) |>
hv_consort_exclude(
label = "Analyzed",
col = "excl_eligible",
!echo ~ "Missing echocardiogram"
)
tracker2<hv_consort_tracker>
Patients : 300
ID column : mrn
Stages : 3
[screened] Screened -- N = 300
-> excl [excl_screen]: 59
[eligible] Eligible -- N = 241
-> excl [excl_eligible]: 32
[analyzed] Analyzed -- N = 209hv_consort() derives the box layout from the tracker’s stage metadata, and plot() draws it. There is no bare-then-decorated pair here, because there is nothing to decorate: the diagram comes out finished. By default every exclusion column becomes a side box; pass a character vector to side_box if you want to show only some of them.
fig <- hv_consort(tracker)
plot(fig)
A CONSORT diagram is read top to bottom, with a glance sideways at each step. Look for:
The tracker is auditable, which is the point of building the diagram through it rather than drawing boxes by hand. hv_consort_summary() returns a per-stage count table you can drop straight into a methods section.
hv_consort_summary(tracker) label include_col n_included excl_col n_excluded
1 Screened screened 300 excl_screen 73
2 Eligible eligible 227 excl_eligible 41
3 Analyzed analyzed 186 <NA> NAhv_consort_patients() returns the patient IDs active at a stage, or the subset excluded for a specific reason. This is how you go from a box on the diagram back to the actual patients behind it, to spot-check an exclusion or pull a list for a sensitivity analysis.
# IDs still in the analysed cohort
head(hv_consort_patients(tracker, "Analyzed"))[1] "PT0001" "PT0002" "PT0003" "PT0004" "PT0005" "PT0007"# IDs dropped at screening for being under 18
head(hv_consort_patients(tracker, "Screened", reason = "Age < 18"))[1] "PT0008" "PT0018" "PT0022" "PT0025" "PT0035" "PT0037"plot(fig) + theme_hv_manuscript() (or any + layer) will error or be ignored. Adjust it through hv_consort()’s own arguments, such as side_box, not through ggplot.hv_consort_exclude() call decides which reason a patient is logged under when more than one applies. Put the rule you want to be the recorded reason first, and read the summary table to confirm the counts land where you expect.hv_consort_summary() and hv_consort_patients() to find the patients in question rather than trusting the visual.