Paperonium

← Back to the plain-language case

Abstract gives the unknown-IVS and incident-TB mortality estimates as poor IVS and prior TB

Lancet Reg Health Eur (2024) Β· doi:10.1016/j.lanepe.2024.100989 Β· PMC11259909

Summary

This EuroSIDA cohort study of 20,364 people with HIV reports adjusted incidence rate ratios (aIRRs) for all-cause mortality. The abstract gives three of them as headline risk factors: off ART 2.42 (2.14–2.74), poor immunologic-virologic status (IVS) 4.2 (3.39–5.20) and prior TB 3.33 (2.75–4.03). In the paper's own Table 2 and Results text, the triple 4.2 (3.39–5.20) is the estimate for unknown IVS, while poor IVS is 7.61 (6.68–8.68). The triple 3.33 (2.75–4.03) is the estimate for TB diagnosed during follow-up, while TB before baseline is 1.45 (1.26–1.67). The abstract therefore understates the poor-IVS association by 45% and overstates the prior-TB association 2.3-fold. Every risk factor stays above 1 and statistically significant, so the paper's qualitative conclusions hold. The abstract also gives region-interaction p-values (IVS p = 0.02, TB p = 0.5) that differ from the Results (p = 0.001 and p = 0.22).

What the paper reports

Abstract, Findings:

Being off ART (aIRR 2.42; 95% CI 2.14–2.74), poor IVS (aIRR 4.2; 95% CI 3.39–5.20) and prior TB (aIRR 3.33; 95% CI 2.75–4.03) were associated with higher all-cause mortality. For all-cause mortality the effect of ART (test for interaction: p < 0.001) and IVS (p = 0.02), but not TB (p = 0.5) varied across regions.

Abstract, Interpretation:

A poor IVS, being off ART and prior active TB were related to higher mortality.

Abstract, Methods (the IVS categories):

Immunologic-virologic status (IVS) was categorized as poor (CD4-cell count ≀350 cells/mm3 and viral load (VL) > 200 copies/ml), good (CD4 β‰₯ 500 and VL < 200), or intermediate (remaining combinations). Participants missing either CD4-cell count or VL were categorized as unknown.

Methods, "Variable selection" (the TB categories):

history of active TB was classified as no history of TB, TB pre-baseline (or at baseline), and incident TB during follow-up

Results, "Risk factors for all-cause mortality":

The most important risk factors for all-cause mortality in the adjusted multivariable model were poor IVS (vs good IVS; aIRR 7.61; 95% CI 6.68–8.68) or unknown IVS (aIRR 4.20; 95% CI 3.39–5.20), incident active TB during follow-up (aIRR 3.33; 95% CI 2.75–4.03), being off ART (aIRR 2.42; 95% CI 2.14–2.74), and receiving care in Eastern Europe (vs. Western Europe; aIRR 1.55; 1.40–1.71).

Results, "Effect of ART-status, IVS and active TB on mortality across regions":

We also observed significant interactions on the risk of all-cause mortality between region of Europe and ART-status (p < 0.001 for interaction) and IVS (p = 0.001) but not for active TB (p = 0.22).

Table 2 ("Risk factors for all-cause and AIDS-related mortality in univariable and multivariable adjusted Poisson regression models"), all-cause columns, IRR (95% CI) | aIRR (95% CI):

Poor | 8.54 (7.63, 9.57) | 7.61 (6.68, 8.68)

Unknown | 4.26 (3.51, 5.18) | 4.2 (3.39, 5.2)

Off ART | 3.91 (3.5, 4.37) | 2.42 (2.14, 2.74)

History of TB, pre-baseline | 1.76 (1.53, 2.02) | 1.45 (1.26, 1.67)

Incident TB during follow-up | 5.55 (4.63, 6.65) | 3.33 (2.75, 4.03)

Eastern | 1.55 (1.4, 1.71) | 1.5 (1.33, 1.69)

Discussion:

The most important risk factors for all-cause mortality in the adjusted multivariable model were poor or unknown IVS, being off ART, and pre-baseline as well as incident active TB during follow-up.

What the paper's own numbers show

Method. No model needs to be refitted: this is a matching check. For each of the abstract's three estimates I searched Table 2's all-cause columns (adjusted and unadjusted) for a row whose point estimate and both confidence limits agree within rounding (Β±0.005). I then compared each abstract value with the Table 2 row that its label names. To confirm that each Table 2 row is internally coherent, I checked that each estimate equals the geometric mean of its CI limits, as a log-scale Wald interval from a Poisson model requires. To show that the swapped rows are distinct estimates, not near-duplicates, I computed an approximate z for each pair from the standard errors implied by the CIs, SE = (ln hi βˆ’ ln lo)/(2 Γ— 1.96). This ignores the covariance between coefficients and is only illustrative. Every number below is printed by recompute.py.txt.

Quantity Location Reported Recomputed from the paper's numbers
"poor IVS" aIRR Abstract, Findings 4.2 (3.39–5.20) Matches only Table 2 Unknown IVS, 4.2 (3.39, 5.2). Table 2 Poor IVS is 7.61 (6.68–8.68)
"prior TB" aIRR Abstract, Findings 3.33 (2.75–4.03) Matches only Table 2 Incident TB during follow-up. Table 2 pre-baseline TB is 1.45 (1.26–1.67)
"off ART" aIRR (control) Abstract, Findings 2.42 (2.14–2.74) Matches Table 2 Off ART, the row its label names
Row coherence, estimate vs sqrt(lo Γ— hi) Table 2 7.61; 4.20; 1.45; 3.33; 2.42 7.61; 4.20; 1.45; 3.33; 2.42 (all coherent)
Poor IVS: Table 2 / abstract derived β€” 7.61 / 4.20 = 1.81; abstract is 44.8% lower
Poor IVS excess risk (aIRR βˆ’ 1) derived 3.20 (abstract) 6.61 (Table 2); abstract gives 48% of it
Prior TB: abstract / Table 2 pre-baseline derived β€” 3.33 / 1.45 = 2.30-fold
Prior TB excess risk (aIRR βˆ’ 1) derived 2.33 (abstract) 0.45 (Table 2); abstract is 5.2-fold larger
CI overlap, abstract vs correctly labelled row derived β€” none for poor IVS (5.20 < 6.68) or prior TB (2.75 > 1.67)
Approximate z, poor vs unknown IVS derived β€” 4.64
Approximate z, incident vs pre-baseline TB derived β€” 6.86
Supplement all-cause aIRRs (Tables S1, S2, S4) mmc1.docx β€” poor IVS 6.80–10.25; pre-baseline TB 1.21–1.99; none equals 4.20 or 3.33 for these categories
Region Γ— IVS interaction p Abstract vs Results 0.02 (Abstract) 0.001 (Results)
Region Γ— TB interaction p Abstract vs Results 0.5 (Abstract) 0.22 (Results)
Eastern vs Western Europe "aIRR" Results text vs Table 2 1.55 (1.40–1.71) This is the unadjusted IRR in Table 2. The adjusted value is 1.50 (1.33–1.69)

Off ART, the control, maps to its own row, so the abstract was drawn from the same adjusted all-cause column. The other two triples each match exactly one row of that column, but not the row their label names. The Results text, Table 2 and the Discussion agree with each other and disagree with the abstract.

Figure

The abstract's poor-IVS and prior-TB estimates are the Table 2 values for unknown IVS and incident TB

Figure 1. Left: the three all-cause aIRRs in the abstract's Findings, with their labels. Right: the corresponding rows of Table 2's adjusted all-cause column, in table order. The off-ART value (black) matches the Off ART row. The values printed as "poor IVS" and "prior TB" (orange) match, in estimate and both CI limits, the Unknown IVS and Incident TB rows (arrows). The Table 2 values for the rows those labels actually name (blue) are 7.61 (6.68–8.68) for poor IVS and 1.45 (1.26–1.67) for pre-baseline TB. They appear nowhere in the abstract.

Abstract value vs Table 2 value for the same label, on a log scale

Figure 2. For each of the abstract's three all-cause risk factors, the orange interval is the aIRR and 95% CI the abstract gives, and the blue interval is the Table 2 aIRR for the category the label names. The axis is logarithmic; the thin line marks aIRR = 1. For off ART the two coincide. For poor IVS and prior (pre-baseline) TB the intervals do not overlap: the abstract understates the poor-IVS association and overstates the prior-TB association. All six intervals lie above 1, so no association changes direction or significance.

Why it matters

The abstract is the part of the paper that most readers, indexers and guideline reviewers use. Its Findings sentence gives exactly three quantified all-cause risk factors, and two of them are paired with the wrong category.

The study's main conclusions are not affected. Mortality fell over time, it remained higher in Eastern Europe, and poor IVS, being off ART and TB are associated with higher mortality. The Research-in-context panel, the Results, Table 2 and the Discussion all state these correctly. The separate Results statement that Eastern Europe has "aIRR 1.55; 1.40–1.71" uses the unadjusted IRR. The adjusted value, 1.50 (1.33–1.69), differs only slightly and does not change the conclusion.

Explanations considered

Status

Suggested correction

  1. Abstract, Findings: replace "poor IVS (aIRR 4.2; 95% CI 3.39–5.20) and prior TB (aIRR 3.33; 95% CI 2.75–4.03)" with "poor IVS (aIRR 7.61; 95% CI 6.68–8.68), unknown IVS (aIRR 4.20; 95% CI 3.39–5.20), TB before baseline (aIRR 1.45; 95% CI 1.26–1.67) and incident TB during follow-up (aIRR 3.33; 95% CI 2.75–4.03)", or an equivalent wording that pairs each estimate with its Table 2 category.
  2. Abstract, Interpretation: replace "prior active TB" with wording that matches the corrected Findings, such as "active TB before or during follow-up".
  3. Abstract, Findings: reconcile the region-interaction p-values with the Results (IVS p = 0.001, TB p = 0.22), or explain which test each refers to.
  4. Results, "Risk factors for all-cause mortality": give the adjusted Eastern vs Western Europe estimate as aIRR 1.50 (95% CI 1.33–1.69), as in Table 2, or label 1.55 (1.40–1.71) as unadjusted.