Paperonium

← Back to the plain-language case

Anhedonia–TSPO correlation reported as P < 0.001 at N = 15 corresponds to P ≈ 0.1

Proc Natl Acad Sci U S A (2024) · doi:10.1073/pnas.2406005121 · PMC11363315

Summary

The association named in the title, between anhedonia severity and the LPS-induced change in TSPO availability ([11C]PBR28 VT) in the PTSD group, is reported as r = −0.425, P < 0.001 in the Results and as r = −0.45, P < 0.001 in the Fig. 2 legend, which states N = 15. For a correlation (or the slope of a simple linear regression) on 15 participants, these r values correspond to two-sided P = 0.114 and P = 0.092, and the 95% confidence interval for r includes zero in both cases (about −0.77 to +0.11). P < 0.001 would need |r| > 0.760 at N = 15, or about 51 to 57 independent participants at the reported r. No standard reading of the test that I could find (one-sided, rank-based, R², partial correlation, the N = 12 CAPS-5 subset) brings the P value near 0.001, so on the paper's own numbers the title association is not statistically significant. The main group comparison (PTSD vs. control, P = 0.021) does not depend on this test.

What the paper reports

Quotations are verbatim from the PMC-archived text (the PMC text renders the minus sign as a hyphen; the typeset text uses U+2212).

Title:

Microglia-mediated neuroimmune suppression in PTSD is associated with anhedonia

Results, "PTSD Is Associated with Lower TSPO Response to LPS Administration", paragraph 1:

In the PTSD group, greater baseline severity of anhedonia symptoms on the clinical administered PTSD scale (CAPS) was associated with a more “blunted” neuroimmune TSPO response to LPS (r = -0.425, P < 0.001) (Fig. 2).

Fig. 2 legend:

Individual data for N = 15 participants in the PTSD group demonstrating that a smaller percent change in [11C]PBR28 VT values following LPS administration was significantly associated with worse anhedonia scores from the CAPS (r = -0.45; P < 0.001) using linear regression analysis. Gray shading indicates the 95% CI for the linear regression.

Results, "Demographic and Clinical Characteristics":

Fifteen individuals who met DSM-5 criteria for PTSD participated in the study (9M and 6F).

Table 1, "Symptom scores" row:

CAPS-5 total * N = 12 | 0.6 (1.5) | 32 (11) | 10.95 | <0.001

Materials and Methods, "Trauma History Assessment":

To examine baseline PTSD symptom clusters associated with LPS-induced changes in TSPO availability, we computed sum scores of CAPS measures that assessed the seven symptom dimensions comprising the hybrid model of PTSD symptom structure (47)—intrusions, avoidance, negative affect, anhedonia, externalizing behaviors, anxious arousal, and dysphoric arousal.

Materials and Methods, "Statistical Analysis" (step 4):

Fourth, to determine whether certain PTSD symptom clusters were associated with changes in TSPO availability pre- and post-LPS administration in the PTSD group, we conducted a multivariable linear regression that included intrusion, avoidance, negative alterations in cognitions and mood, and alterations in arousal and reactivity entered as independent variables and LPS-induced changes in TSPO availability as the dependent variable.

Abstract:

Greater anhedonic symptoms in the PTSD group were associated with a more suppressed neuroimmune response.

Discussion, paragraph 3:

Another notable finding in this study is that greater severity of anhedonic symptoms in the PTSD group was associated with a more suppressed neuroimmune response.

Discussion, conclusion paragraph:

They further demonstrate a relationship between anhedonic symptoms of PTSD and neuroimmune suppression, which suggests that restoring neuroimmune function may potentially help mitigate these symptoms and serve as a treatment for PTSD.

What the paper's own numbers show

Method. The Fig. 2 legend describes a linear regression of percent change in VT on anhedonia score across N = 15 PTSD participants, reported with r. For a simple linear regression, the slope t test equals the test of Pearson's r: t = r·√(n − 2)/√(1 − r²) with n − 2 degrees of freedom. I applied it to both reported r values at n = 15 (df = 13), computed the 95% CI for r by the Fisher z transform, found the |r| and the n that P < 0.001 would need, and then repeated the test under each alternative reading listed in "Explanations considered". Every number below is printed by recompute.py.txt.

Quantity Location Reported Recomputed from the paper's own numbers
Anhedonia vs. % change VT, text Results, "PTSD Is Associated with Lower TSPO Response…", para 1 r = −0.425, P < 0.001 t = −1.693, df = 13, two-sided P = 0.114 (one-sided 0.057)
Same association, figure Fig. 2 legend (N = 15) r = −0.45, P < 0.001 t = −1.817, df = 13, two-sided P = 0.092 (one-sided 0.046)
95% CI for r (Fisher z, n = 15) derived not reported −0.770 to +0.112 (r = −0.425); −0.782 to +0.081 (r = −0.45)
|r| needed for two-sided P < 0.001 derived, n = 15 (df = 13) — |t| > 4.221, i.e. |r| > 0.760
n needed for P < 0.001 at the reported r derived — 57 (r = −0.425); 51 (r = −0.45)
Fisher z test instead of t derived, n = 15 — P = 0.116 / 0.093
Spearman ρ = −0.45, exact permutation null derived, n = 15 — P = 0.094; P < 0.001 needs |ρ| ≥ 0.775
Only the 12 participants with CAPS-5 totals Table 1, "CAPS-5 total * N = 12" — P = 0.168 / 0.142 (|r| > 0.823 needed)
0.425 / 0.45 read as R² alternative reading — r = −0.652 / −0.671, P = 0.0085 / 0.0062
Partial r, 3 or 6 other covariates Methods step 4; hybrid model — P = 0.168 / 0.142 (3); 0.254 / 0.224 (6)
Standardized β, uncorrelated predictors (best case) Methods step 4 — P < 0.001 needs model R² > 0.914 (4 predictors) or > 0.957 (7 predictors) for β = −0.425
60 points treated as independent (4 regions × 15) not described anywhere in the paper — P = 0.00071 / 0.00031

The direction and size of the association are unchanged by the recalculation: r is about −0.43 to −0.45. What changes is its significance. At n = 15 the reported P < 0.001 is not attainable for |r| below 0.760, and both reported r values are well below that. The two r values also differ from each other (−0.425 rounds to −0.43, not −0.45), but this smaller discrepancy does not matter for the conclusion, since neither value supports P < 0.001.

Figure

Both reported correlations have recomputed 95% intervals that include zero and lie far from the range that P < 0.001 would require

Figure 1. The anhedonia–TSPO correlation in the PTSD group (N = 15) on the r scale. Orange diamonds are the r values printed in the Results text (−0.425) and in the Fig. 2 legend (−0.45), both reported with P < 0.001. Blue intervals are 95% confidence intervals recomputed from each r with the Fisher z transform at n = 15 (−0.77 to +0.11 and −0.78 to +0.08), with the corresponding two-sided P values (0.114 and 0.092). The shaded orange region (|r| > 0.760) is the set of r values that could produce two-sided P < 0.001 with 15 participants. The thin grey line marks r = 0. Both intervals cross zero, and neither reported r is in the region the reported P value requires.

Under every reading of the test that uses the stated 15 participants, the recomputed P value is far above 0.001

Figure 2. Recomputed P values under each alternative reading of the reported statistic, on a log scale. Filled circles use r = −0.425 (text); open squares use r = −0.45 (legend). The shaded orange region left of the dashed line is the reported P < 0.001; the grey line marks P = 0.05. Every reading that uses the stated 15 participants, or the 12 with CAPS-5 totals, gives P between 0.006 and 0.25. Only treating about 60 data points as independent (for example four regions × 15 participants, which the paper does not describe and which contradicts "N = 15" in the legend) gives P < 0.001.

Why it matters

Explanations considered

Status

Suggested correction

The authors could report the exact test behind the anhedonia association: whether r is Pearson or Spearman, the n analysed, the degrees of freedom, any covariates, and the unit of analysis (one circuit value per participant or region-level values). If the analysis is the bivariate regression on 15 participants described in the Fig. 2 legend, the Results sentence would read

… was associated with a more “blunted” neuroimmune TSPO response to LPS (r = −0.425, P = 0.11) (Fig. 2).

with the Fig. 2 legend and panel changed to match (one r value, the same in both places, and P ≈ 0.1 rather than P < 0.001), and "significantly associated" removed from the legend. The 95% CI for r (about −0.77 to +0.11) could be added. The abstract, introduction, Discussion and conclusion statements that anhedonia, or symptom severity, is associated with a more suppressed neuroimmune response would then need to be described as a non-significant, exploratory trend, and the title revisited accordingly. Separately, the Fig. 3 legend's "P = 0.19" for the PTSD-only CRP correlation would read "P = 0.019", matching the Results.