What the systematic search produced, and how much of it can be pooled

Published

November 11, 2026

Code 
source("_setup.R")
cand <- read.csv("data/panel_candidates.csv")
ind  <- read.csv("data/cohort_independence.csv")

Counts in the search, tier and orientation sections are from the working status document as at 16 October 2026, unless a table is computed from a file in data/. The source checks and corrections below are current to 19 October 2026. The user reports that the human round-2 checks confirmed the extraction against the source PDFs. The record of those checks is not yet in the repository, so the numbers below are still provisional.

From search to extracted rows

flowchart TD
  A["831 unique records screened<br/>(PubMed 263, Embase 568)"] --> B["81 included"]
  B --> C["63 extracted from a usable source"]
  B --> D["12 with no PDF supplied"]
  B --> E["6 voided: wrong source document"]
  C --> F["2,008 variant rows<br/>after supplementary files"]

Several records without a PDF look like conference abstracts of papers already included in full. If confirmed, they are duplicate reports of the same cohort and not missing data, which changes the PRISMA numbers.

Rows that can be pooled

A row is usable if it carries a complete effect estimate with an orientable effect allele, or a complete genotype table with both alleles named.

Code 
tibble(
  Tier = c("estimate", "counts_2x3", "counts_2x2"),
  Rows = c(823, 115, 32),
  Meaning = c(
    "Effect estimate, both confidence limits and the effect allele",
    "All six genotype cells with both alleles named",
    "Complete 2x2 under a stated dominant, recessive, carrier or allelic model; weaker, kept as a separate sensitivity stratum"
  )
) |> gt() |> fmt_integer(Rows)
Table 1: Usable rows by tier (16 October 2026).
Tier Rows Meaning
estimate 823 Effect estimate, both confidence limits and the effect allele
counts_2x3 115 All six genotype cells with both alleles named
counts_2x2 32 Complete 2x2 under a stated dominant, recessive, carrier or allelic model; weaker, kept as a separate sensitivity stratum

Variants reported by three or more studies

Counts of reporting studies overstate independent evidence, because the Spanish registry series and the Jena AML cohorts overlap. The last column collapses same-group series.

Code 
panel_tab <- cand |>
  filter(studies_inclusive >= 3) |>
  left_join(ind |> select(rsid, independent = effective_independent_cohorts_worst_case), by = "rsid") |>
  arrange(desc(studies_inclusive), rsid)

panel_tab |>
  transmute(
    rsID = rsid, Gene = gene, `Studies reporting` = n_studies,
    `Via estimate` = studies_via_estimate, `Via 2x3 counts` = studies_via_counts_2x3,
    `Via 2x2 counts` = studies_via_counts_2x2,
    `Meta-analysable` = studies_inclusive, `Independent cohorts` = independent
  ) |>
  gt()
Table 2: Panel candidate variants: studies reporting, how each contributes, and independent cohorts.
rsID Gene Studies reporting Via estimate Via 2x3 counts Via 2x2 counts Meta-analysable Independent cohorts
rs7309123 CLEC7A 10 7 1 0 8 6
rs16910526 CLEC7A 12 6 3 1 7 7
rs2305619 PTX3 9 5 1 1 7 7
rs3816527 PTX3 7 4 2 1 6 6
rs4986790 TLR4 9 1 4 0 5 5
rs4804800 CD209 6 4 0 0 4 3
rs4986791 TLR4 9 1 3 0 4 4
rs7248637 CD209 6 4 0 0 4 3
rs1800896 IL10 3 3 0 0 3 3
rs1840680 PTX3 5 3 0 0 3 3
rs2069705 IFNG 3 2 1 0 3 3
rs5743708 TLR2 5 1 2 0 3 3
rs5743810 TLR6 3 2 1 0 3 3

13 variants reach three or more meta-analysable studies.

Code 
panel_tab |>
  select(rsid, Reporting = n_studies, `Meta-analysable` = studies_inclusive,
         `Independent cohorts` = independent) |>
  pivot_longer(-rsid, names_to = "measure", values_to = "n") |>
  mutate(measure = factor(measure, c("Reporting", "Meta-analysable", "Independent cohorts")),
         rsid = factor(rsid, levels = rev(panel_tab$rsid))) |>
  ggplot(aes(n, rsid, fill = measure)) +
  geom_col(position = position_dodge(width = 0.8), width = 0.75) +
  scale_fill_manual(values = c("Reporting" = jama_cols[["pale"]], "Meta-analysable" = jama_cols[["steel"]], "Independent cohorts" = jama_cols[["navy"]])) +
  labs(x = "Studies or cohorts", y = NULL, fill = NULL) +
  theme_jama(grid = "x")
Figure 1: Reporting studies, meta-analysable studies and independent cohorts for each candidate variant.

Allele orientation

Reported alleles were compared with the alleles segregating in Europeans in dbSNP. A flipped strand inverts the risk allele and so the sign of the modelled effect.

Code 
tibble(
  Flag = c("fwd_ok", "strand_flip", "palindromic_unresolvable (A/T or C/G)", "single_allele_only",
           "no alleles reported", "mismatch (allele absent from dbSNP)", "no valid rsID"),
  Rows = c(701, 188, 130, 234, 444, 4, 306)
) |> gt() |> fmt_integer(Rows)
Table 3: Orientation flag across all 2,008 variant rows (16 October 2026).
Flag Rows
fwd_ok 701
strand_flip 188
palindromic_unresolvable (A/T or C/G) 130
single_allele_only 234
no alleles reported 444
mismatch (allele absent from dbSNP) 4
no valid rsID 306

Within the panel, the affected variants are rs16910526, rs1800896, rs2069705 and rs5743810 (10 study by variant cells). All are handled by complementing the alleles.

rs7309123 is a C/G palindrome, so its strand cannot be settled from the allele letters. It was resolved study by study from control allele frequencies, primer sequences and full texts (addenda 16, 19 and 21).

Code 
read.csv("data/rs7309123_strand_resolution.csv") |>
  transmute(Study = paste(first_author, year), Ancestry = ancestry, Strand = gsub("_", " ", strand_conclusion),
            Grade = grade, `In primary analysis` = ifelse(use_in_primary, "yes", "no")) |>
  arrange(Grade, Study) |>
  gt()
Table 4: Strand of rs7309123 by study (A: resolved, high confidence; B: low to medium; C: weak or by inference; D: undetermined). From data/rs7309123_strand_resolution.csv.
Study Ancestry Strand Grade In primary analysis
Chen 2019 Chinese Han same as forward A yes
Fisher 2017 European same as forward A yes
Kalkanci 2020 Turkish same as forward A yes
Schnetzke 2024 European same as forward A yes
Tanpaibule 2020 Thai same as forward A yes
Sainz 2012 European (Spanish) same as forward B yes
Fischer 2015 European (Jena AML cohort) same as forward C no
Fischer 2016 European (Jena AML cohort) same as forward C no
White 2018 European (Welsh) same as forward C no
Zhao 2022 Chinese Han undetermined D no

Source checks and corrections

Every row read from a comparison text with no reference marker was checked against its source table. Each correction is logged with its evidence.

Code 
read.csv("data/direction_check_log.csv") |>
  transmute(Study = study, `Rows without a reference marker` = rows_without_reference_marker, Verdict = verdict,
            Corrections = corrections) |>
  gt()
Table 5: Direction check of the text-parsed rows, by study (data/direction_check_log.csv).
Study Rows without a reference marker Verdict Corrections
Tang 2020 60 first-named group is the numerator in every row
Herrero-Sanchez 2018 41 reference label reversed in the source: corrected 30 rows (data/extraction_corrections.csv)
Sainz 2012 19 one row reversed: corrected; others correct V0261
Fischer 2016 9 reference group listed first: all rows corrected 12 rows (V1681-V1684, V1690-V1697)
Chen 2019 2 first-named genotype is the numerator in every row
Tanpaibule 2020 9 first-named group is the numerator in every row
Fisher 2017 3 carriers of the minor allele (dominant) and minor-allele homozygotes (recessive) are the numerator
Lupianez 2015 2 not checked
Code 
corr <- read.csv("data/extraction_corrections.csv")
n_dir <- sum(corr$study_id %in% c("S035", "S073", "S071"))

data/extraction_corrections.csv holds 71 corrected rows across 6 studies. 43 correct a reversed direction: Herrero-Sanchez 2018, Fischer 2016 and one row of Sainz 2012. The remaining 28 makes a comparison text readable (Tanpaibule 2020). Only Lupiañez 2015’s two IFNG rows were left unchecked, and IFNG is no longer in the panel.

What extraction could not recover

As at 16 October 2026. Since then, the full texts of Kalkanci 2020, Chen 2019 and Fisher 2017 were supplied and read (addenda 19 and 20).

Gap Cells Route
Counts in a supplement not supplied 9 Retrieve supplement
Estimate printed but no nucleotide allele anywhere 10 Author contact, or drop
Reported as a null with no numbers 3 Author contact
Data only inside a figure 2 Author contact

TLR4 is usable only through genotype counts. Every contributing cohort has zero minor homozygotes in cases and controls, so only a carrier model is estimable and a continuity correction is needed.