Genome-wide Polygenic Scores for Common Diseases Identify Individuals with Risk Equivalent to Monogenic Mutations
2018 · paper · Biomedicine
Develops and validates genome-wide polygenic scores for five common diseases. The scores identify 8.0% of people at over three-fold risk for coronary artery disease, a share 20-fold higher than carriers of rare mutations conferring comparable risk.
6 findings
Substance Statistical-analysis paragraph gives "the testing dataset of 288,278 participants"; the testing dataset is 288,978 everywhere else (p2 text, p7 Testing cohort, Fig. 1 and Fig. 2 legends, Tables 1 and 2).
- p. 8 Online Methods, Statistical analysis within the testing dataset, first sentence
calculated in the testing dataset of 288,278 participants
- p. 7 Online Methods, Testing cohort
The testing dataset was comprised of 288,978 UK Biobank Phase 2 participants
- p. 8 Statistical analysis within the testing dataset
testing dataset of 288,278 participants
- p. 7 Testing cohort
288,978 UK Biobank Phase 2 participants
Judge's note: The p8 image shows 288,278. The p7 image shows 288,978 twice (Testing cohort paragraph and '222,529 of the 288,978 (77%)'). Table 2 on p15 uses 288,978 as the denominator, and the p2 text and the Figure 1 and 2 legends give 288,978. The validation set of 120,280 plus 288,978 equals the stated total of 409,258; with 288,278 it does not. No separate subset of 288,278 is described anywhere. Digit typo.
Second check: another model agreed
Report a problem with this finding
Substance Table 1, CAD, prevalence in validation dataset: "3,963/120,280 (3.4%)"; the printed count over the printed denominator gives 3.3%.
- p. 14 Table 1, CAD row, Prevalence in validation dataset
3,963/120,280 (3.4%)
Only the authors can say what is right here.
Judge's note: Read Table 1 on the upright page 14 image; the cell prints 3,963/120,280 (3.4%). The ratio is 3.295%, which rounds to 3.3%. The table footnote and the Methods give no alternative denominator or counting rule for this cell. Either the count or the percentage is misprinted, and only the authors can say which.
Second check: another model agreed
Report a problem with this finding
Substance Table 1, type 2 diabetes, prevalence in validation dataset: "2,785/120,280 (2.4%)"; the printed count over the printed denominator gives 2.3%.
- p. 14 Table 1, Type 2 diabetes row, Prevalence in validation dataset
2,785/120,280 (2.4%)
Only the authors can say what is right here.
Judge's note: Read Table 1 on the upright page 14 image; the cell prints 2,785/120,280 (2.4%). The ratio is 2.315%, which rounds to 2.3%. The other cells I recomputed in the table (for example 2,024/120,280 printed as 1.7% and 1,360/120,280 printed as 1.1%) round correctly. No explanation appears in the footnote or Methods. Either the count or the percentage is misprinted.
Second check: another model agreed
Report a problem with this finding
Reference or label Figure 2 legend says panel (a) is "in the UK biobank testing dataset (N=288,978)" but panels (b) and (c) are "versus controls in the UK biobank validation cohort" and "100 groups of the validation cohort binned according to percentile"; Methods place the 100-bin analysis under "Statistical analysis within the testing dataset".
- p. 12 Figure 2 legend, panels (b) and (c)
versus controls in the UK biobank validation cohort
- p. 12 Figure 2 legend, panel (a)
in the UK biobank testing dataset (N=288,978)
- p. 8 Online Methods, Statistical analysis within the testing dataset
Individuals were next binned into 100 groupings according to percentile of the GPS and unadjusted prevalence of disease within each bin determined
- p. 12 Figure 2 legend, panel c
100 groups of the validation cohort binned according to percentile
Only the authors can say what is right here.
Judge's note: Read the Figure 2 legend on the page 12 image, the Figure 3 legend on page 13, and the Methods and results text in the text layer. The paper defines the validation dataset as the 120,280 Phase 1 participants and the testing dataset as the 288,978 Phase 2 participants. The percentile binning is described only under the testing-dataset heading, and the text citing Figure 2 (0.8% to 11.1%, median percentile 69 vs 49) follows the testing-set AUC sentence. Nothing in the article says panels b and c were drawn from the Phase 1 set. The label therefore conflicts with the Methods and with panel a, but only the authors can say which dataset was plotted. The "UK biobank" capitalisation point is a trivial style matter folded into this cluster. The same wording in the Figure 3 legend is handled under C018.
Second check: another model agreed
Report a problem with this finding
Reference or label Figure 3 legend says "100 groups of the validation cohort were derived according to percentile of the disease-specific GPS", whereas Methods describe the 100-percentile binning under "Statistical analysis within the testing dataset".
- p. 13 Figure 3 legend
100 groups of the validation cohort were derived according to percentile of the disease-specific GPS
- p. 8 Online Methods, section heading
Statistical analysis within the testing dataset
Only the authors can say what is right here.
Judge's note: Read the Figure 3 legend on the page 13 image and the Methods text. This is the same kind of label conflict as C017 but in a different figure legend, so it is kept as a separate item. The Methods describe the percentile-bin prevalence analysis only for the testing dataset, and the results text presents Figure 3 as the generalisation of the CAD results. The plot alone cannot show which dataset was used, so this is a question for the authors.
Second check: another model agreed
Report a problem with this finding
Wording Methods sentence reads "the GPS with the best discriminative capacity in the testing dataset was calculated in the testing dataset", whereas the best GPS was selected in the validation dataset.
- p. 8 Online Methods, Statistical analysis within the testing dataset, first sentence
the GPS with the best discriminative capacity in the testing dataset was calculated in the testing dataset
- p. 2 Main text, last paragraph of page
We used an initial validation dataset of the 120,280 participants in the UK Biobank Phase 1 genotype data release to select the GPS with the best performance
- p. 8 Statistical analysis within the testing dataset
the GPS with the best discriminative capacity in the testing dataset was
- p. 7 Methods
the score with the best discriminative capacity was determined
Judge's note: The p8 image shows the sentence as quoted. The p2 text says the validation dataset of 120,280 was used 'to select the GPS with the best performance'. The p7 methods determine the best score by AUC within the validation section, and the Figure 1 legend says the optimal GPS 'was chosen based on ... AUC in the UK Biobank Phase I validation dataset ... and subsequently calculated in an independent ... testing dataset'. The first 'testing' in the p8 sentence should evidently read 'validation'. The design is stated unambiguously elsewhere, so this is rated as a wording slip.
Second check: another model agreed
Report a problem with this finding
7 minor (typos and the like; most checked by the judge only)
Wording Abstract gives the threshold as "greater than three-fold increased risk" while the body ("≥3-fold increased risk for CAD (Table 2)") and Table 2 ("Odds ratio ≥3.0") define the same 8.0/6.1/3.5/3.2/1.5% figures as at least three-fold.
- p. 1 Abstract
greater than three-fold increased risk
- p. 15 Table 2, section header
Odds ratio ≥3.0
- p. 3 Main text
≥3-fold increased risk for CAD (Table 2)
Judge's note: Checked the abstract on p1, the body text on p3 and Table 2 on p15 (upright image). The percentages in the abstract are exactly the Table 2 'Odds ratio ≥3.0' rows. The threshold is worded strictly in the abstract and inclusively elsewhere. For a continuous odds ratio the difference is negligible, so this is a wording inconsistency only.
Second check: another model agreed
Report a problem with this finding
Typo Grammar slips as printed: "an individuals inherited risk" (p5); "a pruning and threshold derivation strategies" (p6); "for each of the five disease" (p8); "conventional risk factors such as hypercholesterolemia was present in 20%" (p3).
- p. 5 Online Methods, Polygenic score derivation, first sentence
an individuals inherited risk
- p. 6 Online Methods, first paragraph on page
a pruning and threshold derivation strategies
- p. 8 Online Methods, Statistical analysis, end of paragraph
for each of the five disease
- p. 3 Main text, paragraph beginning 'GPSCAD has the advantage'
conventional risk factors such as hypercholesterolemia was present
Judge's note: All four are visible on the page images for p3, p5, p6 and p8: a missing possessive apostrophe, a singular article with a plural noun, a missing plural, and subject-verb disagreement. Meaning is unaffected.
Second check: none (judge only)
Report a problem with this finding
Typo Stray period before the parenthetical: "the top 1% had 3.30-fold risk. (Tables 2 & 3)."
- p. 4 Type 2 diabetes paragraph
3.30-fold risk. (Tables 2 & 3).
Judge's note: The p4 image shows 'risk. (Tables 2 & 3).' in the type 2 diabetes paragraph. The parallel sentences for atrial fibrillation, IBD and breast cancer on the same page read 'risk (Tables 2 & 3).' with no period before the parenthesis. Punctuation slip.
Second check: none (judge only)
Report a problem with this finding
Typo Agency name printed as "Office of Population and Censuses Surveys (OPCS-4)"; the word order appears garbled.
- p. 7 Methods, diagnosis
Office of Population and Censuses Surveys
Only the authors can say what is right here.
Judge's note: Rests on outside knowledge: the agency behind the OPCS classification is the Office of Population Censuses and Surveys. The p7 image shows 'Office of Population and Censuses Surveys (OPCS-4)'. The printed word order is ungrammatical on its face and does not match the acronym order, but the correct name cannot be verified from the files.
Second check: none (judge only)
Report a problem with this finding
Typo Acknowledgments print "(S.L., E.L., S.K)" with the period missing after 'S.K'; the 'E.L.' form is not itself an inconsistency.
- p. 9 Acknowledgments
(S.L., E.L., S.K)
Judge's note: The p9 image shows '5UM1HG008895 (S.L., E.L., S.K)'. The acknowledgments drop middle initials throughout (K.A., S.L., P.E. for K.G.A., S.A.L., P.T.E.), so 'E.L.' follows that section's own style. It differs only from the Author Contributions style on p1 ('E.S.L.'). The real slip is the missing period after 'S.K'. Cosmetic.
Second check: none (judge only)
Report a problem with this finding
Typo Reference 23 title reads "national incidence, prevalence, and years live with disability for 310 diseases and injuries"; "live" should be "lived".
- p. 10 Reference 23
years live with disability
Judge's note: Read the crop of the reference list on page 10. The entry prints "years live with disability". This is a grammatical slip evident on its face ("years lived with disability" is the standard term), so no outside lookup of the cited title is needed to see it.
Second check: none (judge only)
Report a problem with this finding
Typo Table 2 prints "10,099 288,978", "9,209 288,978" and "2,020 288,978" without the "/" separator used in every other row.
- p. 15 Table 2, Odds ratio ≥3.0, Type 2 diabetes
10,099 288,978
- p. 15 Table 2, Odds ratio ≥3.0, Inflammatory bowel disease
9,209 288,978
- p. 15 Table 2, Odds ratio ≥5.0, Atrial fibrillation
2,020 288,978
Judge's note: Read Table 2 on the upright page 15 image. The slash is absent on the page itself in three cells: odds ratio ≥3.0 for type 2 diabetes and for inflammatory bowel disease, and odds ratio ≥5.0 for atrial fibrillation. All other rows print n/N, so this is not an extraction artefact. The printed percentages for these rows (3.5, 3.2, 0.7) match the counts, so the defect is purely typographic.
Second check: none (judge only)