Paperonium

← Back to the plain-language case

Cohort 2 is reported as 361,258 participants in Table 2 but 32,220 in the flowchart

Psychol Med (2026) Β· doi:10.1017/S0033291726103328 Β· PMC12902173

Summary

The paper analyses two matched cohorts from Taiwan's one-million-person Longitudinal Health Insurance Database (LHID): psychological distress (PD, meaning depression or anxiety) followed for incident IBS (Cohort 1), and IBS followed for incident PD (Cohort 2). For Cohort 2, the Results text and Table 2 describe 361,258 participants (180,629 per arm), while the sample-selection flowchart (Figure 1) gives a final follow-up N of 32,220 (16,110 per arm), 11.2 times smaller. The paper's own follow-up event percentages (2,031 = 12.6% and 1,308 = 8.1%) fit only the flowchart's 16,110 per arm, and the width of the headline hazard-ratio interval fits the roughly 3,300 events in that smaller sample. Table 2 therefore cannot describe the cohort that was analysed. It also reports 45.77% of the IBS arm "With PD" at baseline in a cohort defined by excluding prior PD. Cohort 1 shows the same pattern (Table 1: 234,361 per arm; flowchart: 18,402 per arm).

What the paper reports

Results, "Baseline characteristics: Cohort 2 (IBS predicting psychological distress)":

Table 2 summarizes the baseline characteristics of 361,258 participants, divided evenly into IBS (n = 180,629) and non-IBS controls (n = 180,629).

Table 2, column headers and the first block (rows "Without PD" and "With PD"):

Variable | IBS(n = 180,629) | Without IBS(n = 180,629) | p-value

Without PD | 97,948 (54.23%) | 148,971 (82.47%) |

With PD | 82,681 (45.77%) | 31,658 (17.53%) |

Results, the same section, "Psychological distress outcomes":

The prevalence of psychological distress was substantially higher among individuals with IBS (45.77%) compared with controls (17.53%, p < .001).

Methods, "Cohort 2: IBS predicting psychological distress", on the definition of the cohort:

Those with any depressive or anxiety disorder before the index date were excluded to ensure incident outcome assessment.

A consolidated flowchart including all exclusions (prior IBS, prior psychological distress, and missing data) and final analytic sample sizes is shown in Figure 1.

Figure 1 flowchart, Cohort 2 (list items, rendered as text in PMC and in the publisher's HTML):

Newly diagnosed IBS cases (2002–2012): 16,110

Excluded prior depressive or anxiety disorders: 3,231 excluded

IBS cohort: 16,110

Matched controls: 16,110 Final follow-up N for PD outcomes: 32,220

Results, "Cohort 2: IBS predicting incident depression or anxiety":

In the reverse analysis, during a mean follow-up of 6.9 years, 2,031 IBS patients (12.6%) developed either depression or anxiety, compared to 1,308 (8.1%) in the control group.

Cohort 1 has the same kind of discrepancy. Results, "Baseline characteristics: Cohort 1":

A total of 468,722 participants were included, with 234,361 individuals in each group after matching.

Figure 1 flowchart, Cohort 1:

Newly diagnosed depressive/anxiety disorders (2002–2012): 18,402

Psychological distress group: 18,402

Matched controls: 18,402 Final follow-up N for IBS outcome: 36,804

Results, Cohort 1 follow-up, and the Cohort 1 baseline IBS prevalence:

During a mean follow-up period of 7.3 years, 1,498 individuals (8.1%) in the psychological distress group developed IBS, compared to 921 (5.0%) in the control group.

The prevalence of IBS at baseline was significantly higher in the psychological distress group compared to controls (32.11% vs. 12.04%, p < .001)

The Cohort 1 definition (Methods) and the Discussion:

Individuals with a prior diagnosis of IBS (ICD-9-CM: 564.1) before the index date were excluded to ensure incident outcome assessment.

Our findings apply only to incident cases, as individuals with any history of IBS or prior depressive/anxiety disorders were excluded from both cohorts.

The large sample enabled robust subgroup and interaction analyses.

What the paper's own numbers show

Method. I took every sample size, count and percentage for both cohorts from the Results text, the Figure 1 flowchart text, Tables 1–4 and the Methods description of the LHID. I checked (i) each stated total against twice the per-arm N; (ii) that each block of Table 2 sums to its column header; (iii) which denominators reproduce the printed follow-up event percentages, by finding every per-arm N for which the printed event count divided by N rounds to the printed percentage; (iv) the flowchart's own subtraction; (v) the number of events implied by the width of each overall adjusted hazard ratio's 95% CI. For (v) I used the standard approximation for a 1:1 comparison, var(log HR) β‰ˆ 4/D, and also computed the interval expected from the reported event counts, SE β‰ˆ √(1/d₁ + 1/dβ‚‚), and the interval expected if the Table N had produced events at the printed rates. Finally, I compared the cohort sizes with the size of the LHID. Patient-level data are not available, so the models cannot be refitted. Every number below is printed by recompute.py.txt.

Quantity Location Reported Recomputed from the paper's numbers
Cohort 2 total N Results text; Table 2 headers 361,258 (180,629 per arm) 180,629 Γ— 2 = 361,258, internally consistent
Cohort 2 final follow-up N Figure 1 flowchart 32,220 (16,110 per arm) 16,110 Γ— 2 = 32,220, internally consistent
Ratio of the two Cohort 2 totals β€” β€” 361,258 / 32,220 = 11.21; difference 329,038 (164,519 per arm)
Table 2 block sums (PD status, sex, age group, urbanization, income) Table 2 n = 180,629 per column every block sums to 180,629 in both columns
Depression + anxiety = "With PD" Table 2 82,681 and 31,658 15,806 + 66,875 = 82,681; 7,339 + 24,319 = 31,658
IBS-arm follow-up events Results 2,031 (12.6%) 2,031 / 16,110 = 12.61%; 2,031 / 180,629 = 1.12%
Control-arm follow-up events Results 1,308 (8.1%) 1,308 / 16,110 = 8.12%; 1,308 / 180,629 = 0.72%
Per-arm N that reproduces both printed percentages derived β€” 16,056 to 16,183: contains 16,110, excludes 180,629
IBS cohort after excluding prior PD Figure 1 flowchart 16,110 new cases, 3,231 excluded, cohort 16,110 16,110 βˆ’ 3,231 = 12,879, not 16,110
IBS arm "With PD" at baseline Table 2; Results 82,681 (45.77%) 82,681 / 180,629 = 45.77%, in a cohort that excluded prior PD; 25.6 times the flowchart's 3,231 exclusions and 40.7 times the 2,031 follow-up events
Cohort 2 aHR 1.46 (1.37–1.56): events implied by CI width Results; Table 4 β€” SE(log HR) = 0.0331; 4/SEΒ² β‰ˆ 3,644 events (3,288–4,061 allowing for rounding of the limits); reported events 2,031 + 1,308 = 3,339, inside that range
CI expected from 3,339 events derived 1.37–1.56 1.36–1.57 (matches)
CI expected if 180,629 per arm had the printed rates derived β€” 22,759 + 14,631 = 37,390 events; CI 1.43–1.49 (does not match; the reported interval is 3.1 times as wide)
Cohort 1 total N Results text; Table 1 headers 468,722 (234,361 per arm) 234,361 Γ— 2 = 468,722
Cohort 1 final follow-up N Figure 1 flowchart 36,804 (18,402 per arm) 18,402 Γ— 2 = 36,804; ratio 468,722 / 36,804 = 12.74
Cohort 1 follow-up events Results 1,498 (8.1%) and 921 (5.0%) /18,402 = 8.14% and 5.00%; /234,361 = 0.64% and 0.39%; N reproducing both: 18,381–18,606
Cohort 1 aHR 1.55 (1.42–1.69): events implied by CI width Results; Table 3 β€” β‰ˆ 2,028 (1,885–2,188), slightly fewer than the 2,419 reported (1,498 + 921), i.e. the reported interval is a little wider than the events alone imply; both are far from the 30,701 implied by the Table N. CI from 2,419 events 1.43–1.68; from Table N at printed rates (30,701 events) 1.51–1.59, 3.8 times narrower than reported
PD arm with IBS at baseline Table 1; Results 75,244 (32.11%) 75,244 / 234,361 = 32.11%, in a cohort that excluded prior IBS
Adults in the LHID Methods 1,000,000 βˆ’ 1,982 βˆ’ 2,614 995,404
Cohort 1 eligible population Figure 1 flowchart "990,0xx" 995,404 βˆ’ 5,412 = 989,992, which is not of the form 990,0xx either
Exposed arms of both cohorts Tables 1 and 2 234,361 + 180,629 414,990 = 41.7% of LHID adults
Exposed arms of both cohorts Figure 1 flowchart 18,402 + 16,110 34,512 = 3.5% of LHID adults

The two exposed groups essentially cannot share people. Cohort 1 excludes anyone with IBS before the PD diagnosis, and Cohort 2 excludes anyone with PD before the IBS diagnosis, so each person can be exposed in at most one cohort (apart from anyone first diagnosed with both on the same date). Taken at face value, Tables 1 and 2 would mean that 41.7% of all adults in the database received a first diagnosis of PD or of IBS between 2002 and 2012, with the other condition absent beforehand. The flowchart's figures correspond to 3.5%.

Both totals for each cohort are internally consistent. Table 2's blocks all sum to 180,629, and the flowchart's numbers multiply out to 32,220. The two descriptions cannot both refer to the same matched sample. The follow-up percentages and the CI widths fit the flowchart, so the survival analyses appear to have been run on about 16,110 patients per arm in Cohort 2 (and about 18,402 in Cohort 1). It is not stated what population Tables 1 and 2 describe.

Figure

Per-arm sample sizes printed in the Tables and in the flowchart, with the range implied by the printed event percentages

Figure 1. Participants per arm for each cohort on one linear axis starting at zero. Orange markers are printed in the paper: circles are the Table and Results-text values (234,361 for Cohort 1, 180,629 for Cohort 2), and squares are the Figure 1 flowchart values (18,402 and 16,110). The grey bar and the label on it give the ratio between the two printed values (12.7 and 11.2). The blue diamond marks the range of per-arm N (narrower than the marker at this scale) for which both printed follow-up percentages are reproduced from the printed event counts (Cohort 2: 2,031 at 12.6% and 1,308 at 8.1% give 16,056–16,183; Cohort 1: 1,498 at 8.1% and 921 at 5.0% give 18,381–18,606), computed by recompute.py.txt. In both cohorts this range contains the flowchart value and excludes the Table value.

Overall adjusted hazard ratios with the reported 95% CI and the intervals expected from the reported events and from the Table sample sizes

Figure 2. Overall adjusted hazard ratios on one log-scale axis; the vertical line marks HR = 1. Orange circles are the reported aHRs and 95% CIs (Table 4: 1.46, 1.37–1.56; Table 3: 1.55, 1.42–1.69). Filled blue diamonds are the intervals expected from the reported numbers of events (3,339 and 2,419), with SE(log HR) = √(1/d₁ + 1/dβ‚‚); they agree closely with the reported intervals (for Cohort 1 the reported interval is slightly wider than 2,419 events alone imply, which corresponds to about 2,028 events). Open blue diamonds are the intervals expected if the Table sample sizes had produced events at the printed rates (37,390 and 30,701 events); they are about three to four times narrower than reported (3.1x for Cohort 2, 3.8x for Cohort 1, measured as SE(log HR) and identical for the linear width). The reported precision is what the flowchart-sized cohorts would give, not what the Table-sized cohorts would give.

Why it matters

Cohort 2 is one of the paper's two primary analyses. Its sample size is given in the Results text and in Table 2 as 361,258, about eleven times the 32,220 in the flowchart that the Methods identify as showing the "final analytic sample sizes". The follow-up events and the precision of the hazard ratio fit the smaller number, so the hazard ratios were most likely estimated on about 32,000 people. Their direction and statistical significance are probably not affected by the misstated N.

What is affected is the description of the study population. Table 2, the Results paragraphs that summarize it ("well matched", the income and comorbidity contrasts, the 45.77% vs 17.53% PD prevalence) and the corresponding Table 1 material for Cohort 1 cannot describe the analysed cohorts. They report N values 11–13 times too large and a large prevalence of the very condition that each cohort was defined to exclude: 82,681 IBS patients "With PD" in Cohort 2 and 75,244 PD patients "With IBS" in Cohort 1. A reader cannot use these tables to judge baseline balance, comorbidity burden or generalizability, and the Discussion's "The large sample enabled robust subgroup and interaction analyses" rests in part on the larger figures. Anyone extracting the sample size for a systematic review or meta-analysis would take 361,258 (or 468,722) from the text, overstating the size of these cohorts eleven- to thirteen-fold.

The flowchart has its own problem. It lists 16,110 newly diagnosed IBS cases, then 3,231 exclusions for prior PD, and then an IBS cohort of 16,110 again, where 12,879 would be expected. So it is not clear whether 16,110 is the number before or after the exclusion.

A separate inconsistency adds to the uncertainty. The abstract gives "IBS was associated with increased risks of depression (adjusted hazard ratio [aHR] = 1.55) and anxiety (aHR = 1.68)", but 1.55 and 1.68 are the overall and female HRs for the opposite direction (PD β†’ IBS) in Table 3. The body reports no depression- or anxiety-specific HRs, and the abstract's 1.45 and 1.51 for "depression and anxiety ... developing IBS" do not appear as point estimates anywhere in the body. Table 4's overall HR for IBS β†’ PD is 1.46.

Explanations considered

Status

Suggested correction

  1. State which population Tables 1 and 2 describe. If the analysed cohorts are those in Figure 1 (18,402 and 16,110 per arm, as the event percentages and hazard-ratio precision indicate), replace Tables 1 and 2 and the Results paragraphs that summarize them with the baseline characteristics of those cohorts, and correct "468,722" / "234,361" and "361,258" / "180,629" in the Results text.
  2. Explain why Tables 1 and 2 show 75,244 PD patients with IBS and 82,681 IBS patients with PD at baseline in cohorts defined by excluding those conditions, or remove these rows if they do not describe baseline status.
  3. Reconcile the Cohort 2 flowchart: 16,110 newly diagnosed IBS cases minus 3,231 exclusions should not leave an IBS cohort of 16,110. Give the counts before and after exclusion, and give the Cohort 1 eligible population as a number instead of "990,0xx" (995,404 βˆ’ 5,412 = 989,992, which does not fit "990,0xx" either).
  4. Correct the abstract's hazard ratios so that each one matches a reported estimate in the body for the stated direction and outcome (for IBS β†’ PD, Table 4 gives 1.46 overall), or add the depression- and anxiety-specific estimates to the Results.
  5. Reword statements that rest on the larger sample sizes, such as "The large sample enabled robust subgroup and interaction analyses", if the analysed cohorts are the smaller ones.