Paperonium

← Back to the plain-language case

Abstract calls a non-significant subgroup difference at seroconversion significant; Results and Table 3 report none

Diabetologia (2025) · doi:10.1007/s00125-025-06573-6 · PMC12686032

Summary

This DIPP cohort analysis compares 1334 children with islet autoimmunity in three groups: an affected first-degree relative (FDR) at birth (FDR+, n = 145), an FDR diagnosed during follow-up (FDR− FDR+, n = 87), and no affected FDR (FDR−, n = 1102). The abstract's Results state that at seroconversion both FDR+ and FDR− FDR+ children were "significantly more often positive for GADA and multiple autoantibodies" than FDR− children. The paper's own Table 3 gives non-significant results for the FDR− FDR+ group: GADA OR 1.3 (95% CI 0.9, 2.1), p = 0.201, and multipositivity OR 1.5 (0.9, 2.4), p = 0.106. The Results text and the Discussion both say this group did not differ from FDR− children at seroconversion, and a recomputation from the Table 2 counts agrees (p = 0.19 and 0.11). The abstract is the only place where the claim appears. A reader who sees only the abstract could reasonably conclude that a family history arising after birth already marks a distinct autoantibody profile at seroconversion, which the body of the paper does not show.

What the paper reports

Text is quoted from the PubMed Central archive, which writes the group label as "FDR- FDR+" with a hyphen; the version of record typesets it with a minus sign (FDR− FDR+).

Abstract, Results:

At seroconversion, both FDR+ and FDR- FDR+ children were significantly more often positive for GADA and multiple autoantibodies than children without affected FDRs (FDR-).

Results, fourth paragraph (beginning “We compared the islet autoantibody profiles”):

At seroconversion, there was a statistically significant difference between the three groups for GADA positivity and multipositivity (Table 2). In pairwise comparisons, at seroconversion, FDR+ children had a higher risk for GADA positivity and multipositivity than FDR- children, but no difference was observed between FDR- FDR+ and FDR- children (Table 3).

Table 3, "At seroconversion" rows (columns: FDR+ vs FDR− | p value | FDR− FDR+ vs FDR− | p value; footnote: "Values are OR (95% CI) by logistic regression analysis"):

GADA+ | 1.5 (1.1, 2.2) | 0.020 | 1.3 (0.9, 2.1) | 0.201

Multipositive | 1.8 (1.2, 2.6) | 0.002 | 1.5 (0.9, 2.4) | 0.106

Table 2, "At seroconversion" rows (columns: FDR+ (n=145) | FDR− FDR+ (n=87) | FDR− (n=1102) | p value; data are % (n); footnote: "The exact χ2 test was used for categorical variables"):

GADA+ | 62.8 (91) | 59.8 (52) | 52.5 (578) | 0.036

Multipositive | 35.2 (51) | 31.0 (27) | 23.3 (257) | 0.004

Discussion, seventh paragraph (beginning “In terms of the timing of type 1 diabetes in the FDRs”):

FDR- FDR+ children did not differ from FDR- children at seroconversion, but showed a significantly higher risk of being positive for IAA, GADA, IA-2A or multiple autoantibodies during the follow-up period than FDR- children (Table 3).

Methods, "Statistical analyses":

The standard normal distribution (SND) test was used for pairwise comparisons of categorical variables. Logistic regression analysis was used to estimate the risk of various islet autoantibody profiles and progression to type 1 diabetes in children with and without an FDR with the disease.

Abstract, Conclusions/interpretation (for comparison; this sentence does not mention the FDR− FDR+ group):

At seroconversion, genetically susceptible children who had one or more FDRs with type 1 diabetes, especially an affected father, were more often positive for GADA and multiple islet autoantibodies.

What the paper's own numbers show

Method. From the Table 2 counts I rebuilt each 2×2 table (group vs FDR−; positive vs not positive) at seroconversion. For each I computed the unadjusted odds ratio with a Woolf 95% CI and Wald p value. These are exactly what a logistic regression of the outcome on a single group indicator gives, and the Methods name no covariates. I also computed the pooled two-proportion z test (the "SND test" named in the Methods), Fisher's exact test, the likelihood-ratio test and the Yates-corrected χ² test, and repeated the three-group χ² tests from Table 2. Every number below is printed by recompute.py.txt.

Quantity Location Reported Recomputed from the paper's numbers
Group sizes Abstract; Table 2 145 / 87 / 1102; total 1334 145 + 87 + 1102 = 1334
GADA+ at seroconversion, FDR− FDR+ vs FDR− Table 2; Table 3 59.8% (52) vs 52.5% (578); OR 1.3 (0.9, 2.1), p = 0.201 52/87 vs 578/1102; OR 1.35 (0.86, 2.10); Wald p = 0.189; SND p = 0.188; Fisher p = 0.220; LR p = 0.186
Multipositive at seroconversion, FDR− FDR+ vs FDR− Table 2; Table 3 31.0% (27) vs 23.3% (257); OR 1.5 (0.9, 2.4), p = 0.106 27/87 vs 257/1102; OR 1.48 (0.92, 2.38); Wald p = 0.106; SND p = 0.104; Fisher p = 0.117; LR p = 0.114
GADA+ at seroconversion, FDR+ vs FDR− Table 3 OR 1.5 (1.1, 2.2), p = 0.020 91/145 vs 578/1102; OR 1.53 (1.07, 2.18); p = 0.020
Multipositive at seroconversion, FDR+ vs FDR− Table 3 OR 1.8 (1.2, 2.6), p = 0.002 51/145 vs 257/1102; OR 1.78 (1.23, 2.58); p = 0.002
Three-group test, GADA+ at seroconversion Table 2 p = 0.036 (exact χ²) Pearson χ² = 6.71, df 2, p = 0.035
Three-group test, multipositive at seroconversion Table 2 p = 0.004 (exact χ²) Pearson χ² = 11.31, df 2, p = 0.004
Abstract statement on FDR− FDR+ vs FDR− at seroconversion Abstract, Results "significantly more often positive for GADA and multiple autoantibodies" not significant on either measure under any of the tests above; both 95% CIs include 1

The two disputed cells of Table 3 reproduce from Table 2: the odds ratios and intervals match to the printed precision, and the p values are close (printed 0.201, recomputed 0.189 for GADA; printed 0.106, recomputed 0.106 for multipositivity). The body of the paper is internally consistent. Only the abstract sentence departs from it.

Side observation, not part of the error. When every Table 3 cell is recomputed in the same way from the Table 2 counts, six other cells do not reproduce (rule applied in recompute.py.txt, section 7: the recomputed OR, rounded to one decimal, differs from the print by more than 0.1, or the p value differs by more than 0.05). Five are in the FDR− FDR+ vs FDR− column: IAA+ at seroconversion (printed 1.3 (0.8, 2.1), p = 0.181; recomputed 1.52 (0.97, 2.39), p = 0.070), IA-2A+ at seroconversion (printed 1.0 (0.5, 1.6), p = 0.937; recomputed 0.95 (0.52, 1.75), p = 0.869), ZnT8A+ at seroconversion (printed 0.8 (0.3, 1.5), p = 0.484; recomputed 0.89 (0.43, 1.81), p = 0.739), GADA+ during follow-up (printed 2.8 (1.1, 3.7), p = 0.021; recomputed 2.05 (1.12, 3.76), p = 0.020) and ZnT8A+ during follow-up (printed 1.2 (0.8, 1.8), p = 0.361; recomputed 1.48 (0.96, 2.30), p = 0.077). One is in the FDR+ vs FDR− column: ZnT8A+ during follow-up (printed 1.2 (0.8, 1.7), p = 0.227; recomputed 1.21 (0.86, 1.71), p = 0.278). For GADA+ during follow-up, the printed estimate 2.8 is far from the geometric midpoint of its interval (1.1, 3.7), which is 2.0, while the recomputed OR is 2.05; this suggests a typographical slip in the point estimate. None of these six changes which side of p = 0.05 a result falls on, and none bears on the abstract sentence.

Figure

The abstract and the body of the paper give opposite verdicts on the same comparison

Figure 1. The left panel quotes the abstract's claim that FDR− FDR+ children were significantly more often GADA-positive and multipositive at seroconversion than FDR− children. The right panel quotes the Results sentence and the Table 3 estimates for the same comparison, which are non-significant. Values the paper reports are in orange. The blue panel gives the estimates recomputed from the Table 2 counts (unadjusted logistic regression, equivalent to the 2×2 odds ratio), with Fisher exact p values for comparison. Both recomputed intervals include OR = 1.

Reported and recomputed odds ratios for the four pairwise comparisons at seroconversion

Figure 2. Odds ratios with 95% confidence intervals for the four pairwise comparisons at seroconversion, on a log scale with a thin reference line at OR = 1 (no difference). Orange circles show Table 3 as printed; blue diamonds show the values recomputed from the Table 2 counts, with the p value for each printed at the right. The two shaded rows are the FDR− FDR+ comparisons, which the abstract calls significant but Table 3 and the Results text do not. In both, the printed and the recomputed intervals cross 1. The two unshaded FDR+ rows, which the abstract also mentions, are significant and reproduce.

Why it matters

Explanations considered

Status

Suggested correction

In the abstract's Results, replace

At seroconversion, both FDR+ and FDR− FDR+ children were significantly more often positive for GADA and multiple autoantibodies than children without affected FDRs (FDR−).

with

At seroconversion, FDR+ children were significantly more often positive for GADA and multiple autoantibodies than children without affected FDRs (FDR−), whereas FDR− FDR+ children did not differ significantly from FDR− children (GADA OR 1.3 [95% CI 0.9, 2.1]; multipositivity OR 1.5 [0.9, 2.4]).

No change to the Results, Tables or Discussion is needed for this point. Separately, the authors may wish to check the six cells of Table 3 listed above (five in the FDR− FDR+ column, one in the FDR+ column) that do not reproduce from the Table 2 counts. One example is GADA+ during follow-up, printed as 2.8 (1.1, 3.7), where the counts give 2.05 (1.12, 3.76).