Paperonium

← Back to the plain-language case

Abstract reports overall-cohort length-of-stay estimates as the results for hypotensive patients

JAMA Netw Open (2025) · doi:10.1001/jamanetworkopen.2025.33266 · PMC12457976

Summary

The abstract states that, among patients with hypotension on admission, splenic angioembolization (SAE) and observation were each associated with a shorter hospital length of stay (HLOS) than open splenectomy (OS): SAE β = -1.44 (95% CI, -1.79 to -1.09; P < .001) and observation β = -1.41 (95% CI, -1.73 to -1.09; P < .001). These six numbers are identical to the paper's Table 2 rows for the total cohort (n = 7567). Table 2's rows for the hypotension subgroup (SBP <90 mm Hg, n = 651) are different: SAE β = -1.65 (-3.24 to -0.06; P = .04) and observation β = -0.13 (-1.58 to 1.32; P = .86). The Results text reports only the SAE figure for this subgroup. For observation, the abstract therefore reports a highly significant benefit of about 1.4 days where the paper's own subgroup analysis, and its inverse-probability-weighted (IPW) sensitivity analysis, show no significant difference. The claim is repeated in words in the main-text Conclusions, and it supports the headline statement that nonoperative management had "favorable outcomes ... even in patients with hypotension on admission".

What the paper reports

Abstract, Results:

Among patients with hypotension, there was no mortality difference, but shorter hospital length of stay was found in the SAE (β = -1.44; 95% CI, -1.79 to -1.09; P < .001) and observation (β = -1.41; 95% CI, -1.73 to -1.09; P < .001) groups.

Results, "Multivariable Analysis for Other Outcomes". This paragraph gives the same two estimates for the total cohort, then reports the hypotension subgroup separately and names only SAE:

Multivariable linear regression analysis of continuous variable outcomes is given in Table 2. SAE (β = -1.44; 95% CI, -1.79 to -1.09; P < .001) and observation (β = -1.41; 95% CI, -1.73 to -1.09; P < .001) showed a negative association with HLOS days.

The same results were observed in normal SBP subgroup. In the hypotension subgroup, SAE was also negatively associated with HLOS days (β = -1.65; 95% CI, -3.24 to -0.06; P = .04) and ICULOS days (β = -1.45; 95% CI, -2.62 to -0.29; P = .01).

Table 2, "SAE or Observation vs OS Outcomes Using Multivariable Linear Regression and IPW". Columns: outcome | No. | mean (SD) | regression β (95% CI) | P value | IPW β (95% CI) | P value. Total block, HLOS days:

SAE | 1547 | 6.35 (4.17) | -1.44 (-1.79 to -1.09) | <.001 | -1.77 (-2.24 to -1.30) | <.001

Observation | 4521 | 6.04 (5.00) | -1.41 (-1.73 to -1.09) | <.001 | -1.60 (-2.05 to -1.14) | <.001

Table 2, SBP <90 mm Hg block, HLOS days (block total 651; OS row 323 | 9.40 (8.31) | 0 [Reference]):

SAE | 124 | 7.48 (4.80) | -1.65 (-3.24 to -0.06) | .04 | -1.67 (-2.89 to -0.46) | .007

Observation | 204 | 8.74 (8.01) | -0.13 (-1.58 to 1.32) | .86 | -0.48 (-2.00 to 1.04) | .54

Main-text Conclusions:

Furthermore, in patients with hypotension on admission, nonoperative management (either with SAE or observation) was associated with no increased risk of complications or mortality and shorter LOS.

Key Points, "Findings" (the abstract's Conclusions and Relevance uses the same closing phrase):

In this cohort study of 7567 patients, nonoperative management (angioembolization or observation) was associated with favorable outcomes when compared with surgery in isolated severe blunt splenic injury, even in patients with hypotension on admission.

What the paper's own numbers show

Method. I compared the abstract's two "hypotension" estimate-and-CI triples with every HLOS row in Table 2 (total, SBP <90, SBP ≥90 and failure blocks, regression and IPW columns). I then checked each row's internal consistency with a Wald approximation from the printed interval: SE = (upper − lower)/(2 × 1.96), z = β/SE, two-sided P = 2Φ(−|z|). I checked that the block sizes add up, and I compared CI widths with the group sizes. As a further check I computed the unadjusted mean differences and Welch t tests from the Table 2 means and SDs. Patient-level data are not available, so the adjusted models cannot be refitted. Every number below is printed by recompute.py.txt.

Quantity Location Reported Recomputed from the paper's numbers
HLOS β, SAE vs OS, "among patients with hypotension" Abstract -1.44 (-1.79 to -1.09), P < .001 Identical to the Table 2 Total row; matches no SBP <90 row
HLOS β, observation vs OS, "among patients with hypotension" Abstract -1.41 (-1.73 to -1.09), P < .001 Identical to the Table 2 Total row; matches no SBP <90 row
HLOS β, SAE, hypotension subgroup Table 2, SBP <90; Results -1.65 (-3.24 to -0.06), P = .04 SE 0.811, z = -2.03, P = 0.042: consistent with itself
HLOS β, observation, hypotension subgroup Table 2, SBP <90 -0.13 (-1.58 to 1.32), P = .86 SE 0.740, z = -0.18, P = 0.861; CI includes 0
IPW HLOS β, SAE / observation, hypotension subgroup Table 2, SBP <90, IPW -1.67 (-2.89 to -0.46), P = .007 / -0.48 (-2.00 to 1.04), P = .54 P = 0.007 / 0.536; observation CI includes 0
HLOS β, total cohort, SAE / observation Table 2, Total -1.44 (-1.79 to -1.09) / -1.41 (-1.73 to -1.09), both P < .001 z = -8.06 / -8.64, both P < 10⁻¹⁵
HLOS β, normotension subgroup, SAE / observation Table 2, SBP ≥90 -1.44 (-1.80 to -1.09) / -1.53 (-1.86 to -1.20) Close to, but not identical with, the abstract (observation -1.53 vs -1.41)
Abstract's failure-subgroup HLOS figures (control) Abstract vs Table 2, Failure 2.50 (1.27-3.73); 0.71 (0.07-1.35) Identical to the Table 2 Failure rows: the abstract's other subgroup result is transcribed correctly
Group sizes, HLOS blocks Table 2 7567; 651; 6916 1499 + 1547 + 4521 = 7567; 323 + 124 + 204 = 651; 1176 + 1423 + 4317 = 6916; 651 + 6916 = 7567
CI width, abstract vs SBP <90 row Abstract; Table 2 0.70 vs 3.18 (SAE); 0.64 vs 2.90 (observation) The subgroup intervals are 4.5 times as wide as the abstract's
Residual SD implied by each CI, SE/√(1/n_arm + 1/n_OS) Table 2 HLOS SD 7.69 (SBP <90), 5.27 (total) SBP <90 rows: 7.68 and 8.27; total rows: 4.93 and 5.48. The abstract's CIs placed at the SBP <90 group sizes would imply 1.69 and 1.83, so covariates would have to explain about 95% of the HLOS variance in that subgroup, while the table's own subgroup rows show almost no such reduction
ICULOS β, observation, hypotension subgroup Table 2, SBP <90 -0.49 (-1.59 to 0.61), P = .38 P = 0.383; CI includes 0
Unadjusted mean HLOS difference vs OS, hypotension subgroup Table 2 means (SD) OS 9.40 (8.31), SAE 7.48 (4.80), observation 8.74 (8.01) SAE -1.92 days (Welch P = 0.003); observation -0.66 days (Welch P = 0.364)

Each Table 2 row is internally consistent: the printed P values follow from the printed intervals, and the block sizes add to the cohort total. The table is not misaligned. The discrepancy lies between the abstract sentence and the subgroup rows it describes. The CI-width check points the same way. The abstract's intervals are as narrow as one would expect from about 3000-6000 patients per comparison, not from the 447 (SAE) and 527 (observation) patients in the hypotension comparisons.

Figure

The abstract's "hypotension" length-of-stay estimates are the Table 2 total-cohort values; the Table 2 hypotension rows differ

Figure 1. The abstract's sentence on patients with hypotension (top, orange outline) quoted next to the two Table 2 blocks it could refer to. All six numbers, for both arms, match the Total-cohort block (n = 7567) exactly. The SBP <90 mm Hg block (n = 651), which is the subgroup the sentence names, gives a different SAE estimate with an interval 4.5 times as wide, and an observation estimate of -0.13 days whose 95% CI includes 0. The blue markers give the P values recomputed from each printed interval with a Wald approximation. They agree with the table's own P values (.04 and .86), so the table rows are internally consistent.

Forest plot of the abstract's hypotension HLOS estimates, the Table 2 SBP <90 rows and the Table 2 total-cohort rows on one axis

Figure 2. The same estimates on one axis, in days of hospital stay relative to open splenectomy. The orange squares are the abstract's "hypotension" values, the blue circles are the Table 2 SBP <90 rows, and the hollow gray circles are the Table 2 total-cohort rows. The orange and gray intervals coincide. For observation, the subgroup interval (blue) crosses the null line at 0, and the abstract's (orange) does not. The right-hand column gives each printed P value, with the Wald P recomputed from the printed CI for the subgroup rows.

Why it matters

Whether nonoperative management, and observation in particular, is reasonable for hemodynamically unstable patients with high-grade splenic injury is a clinically consequential question. The abstract is the part of the paper most readers rely on. In the hypotension subgroup, Table 2 and the Results show no difference in mortality (HR 0.95 for SAE, 1.08 for observation) or in any complication (OR 0.71, P = .30; OR 1.15, P = .62). That leaves length of stay as the only outcome on which the subgroup shows a benefit. For observation, that benefit is not present in the paper's own subgroup analysis: β = -0.13 days (95% CI, -1.58 to 1.32; P = .86), with IPW giving -0.48 days (P = .54). The unadjusted difference is also small and not significant (-0.66 days, P = .36). The main-text Conclusions ("nonoperative management (either with SAE or observation) was associated with ... shorter LOS") and the Key Points' "favorable outcomes ... even in patients with hypotension on admission" both rest partly on the abstract's version.

For SAE the direction is unchanged, but the abstract overstates the precision. The subgroup estimate is -1.65 days with a CI reaching almost to 0 (-0.06) and P = .04, not P < .001 with a 0.7-day-wide interval.

The paper's other main findings do not depend on this sentence. These are lower morbidity and shorter stay in the overall cohort, the same in the normotension subgroup, no mortality difference, and the failure-subgroup results. A null result does not show that observation has no effect on length of stay in hypotensive patients. It shows only that this analysis does not support the significant effect the abstract reports.

Explanations considered

Status

As of 2026-10-23:

Suggested correction

In the Abstract, Results, replace the hypotension sentence with the Table 2 SBP <90 mm Hg values, for example: "Among patients with hypotension, there was no mortality difference; hospital length of stay was shorter in the SAE group (β = -1.65; 95% CI, -3.24 to -0.06; P = .04) but not significantly different in the observation group (β = -0.13; 95% CI, -1.58 to 1.32; P = .86)." The main-text Conclusions should then be revised: in patients with hypotension, shorter LOS was found for SAE only. The phrase "even in patients with hypotension on admission" in the Key Points and in the Abstract's Conclusions and Relevance should be checked against the subgroup results as corrected. In that subgroup, mortality and complications did not differ, and the LOS benefit applies to SAE only.