Hewgley WP, Griffin RL, Baird E, Hashmi ZG, Baker SJ, Jansen JO, Kerby J, Holcomb JB, Gelbard RB.
Empiric calcium during massive transfusion is guideline-supported. Empiric calcium after a small transfusion has never actually been tested.
The Joint Trauma System recommends 1 g calcium chloride before, during, or after the first blood unit and again every four units, largely to counteract citrate chelation during massive transfusion. Some EMS agencies now give calcium prehospital, before any blood at all. The authors' point: that guidance was built for massive transfusion, and it gets applied reflexively to patients who never end up needing anywhere near that much blood. Prior data cited in the paper show up to 45% of patients who meet clinical triggers for MTP activation ultimately receive 3 units of RBCs or fewer.
This retrospective cohort, drawn from 37,599 trauma patients evaluated at a single Level 1 center from 2018–2024, isolates exactly that group: adults who received 1–3 units of RBCs in the first 24 hours. 3,360 patients met that transfusion criterion; 8 were excluded for missing calcium levels and 672 for a hospital stay under 4 days (too short to reliably capture the outcomes of interest). The final cohort was 2,680 patients.
Patients were stratified two ways: calcium level on presentation, and whether they got calcium within 6 hours of injury.
65% of the cohort presented hypocalcemic on arrival, but only 254 of those 1,742 (14.6%) actually received early supplementation. This matters: the study only has statistical power on a fairly small treated subgroup. Median time to administration was 35 minutes (IQR 15–108 min), so "early" here really does mean early, not a delayed reactive dose hours later.
Groups differed significantly in shock index and lactate. Mean shock index was 0.97 in NormoCa/SUP vs. 0.83 in NormoCa/No-SUP, and 1.08 in HypoCa/SUP vs. 0.89 in HypoCa/No-SUP (both p<0.001). Supplemented patients, hypocalcemic or not, were sicker on arrival. This is real and it's the correct starting caveat for reading everything below.
Adjusted for age, sex, penetrating injury mechanism, ISS ≥16, and shock index >1. Reference group: NormoCa/No-SUP. This is the complete Table 2, not the six outcomes the Instagram post selected.
| Outcome | HypoCa/SUP, n (%) | OR (95% CI) | Significant? |
|---|---|---|---|
| Death | 19 (7.7%) | 2.01 (1.09–3.72) | Yes |
| AKI | 12 (4.7%) | 2.52 (1.14–5.58) | Yes |
| ARDS | 6 (2.4%) | 2.82 (1.04–7.70) | Yes |
| DVT | 29 (11.4%) | 2.44 (1.40–4.26) | Yes |
| PE | 11 (4.3%) | 4.26 (1.67–10.87) | Yes |
| MI | 9 (3.5%) | 5.03 (1.88–13.40) | Yes |
| Sepsis | 3 (1.2%) | 5.44 (1.17–25.37) | Yes |
| Cardiac arrest | 4 (1.6%) | 2.45 (0.71–8.50) | No |
| Infection | 11 (4.3%) | 1.04 (0.51–2.14) | No |
| Stroke | 3 (1.2%) | 0.94 (0.27–3.24) | No |
Sepsis was also significantly elevated (OR 5.44, 95% CI 1.17–25.37) and was left out of the Instagram summary entirely. It has the widest confidence interval of any outcome, driven by very small event counts (3 cases in the HypoCa/SUP group), so it's the least statistically stable finding in the table, but it's a real, significant, and notably large point estimate that didn't make the original six-outcome list.
Not every significant OR in that table means the same thing. The paper's own analysis says three outcomes are the real story, and three others are weaker than they look.
Beyond the odds ratios, the authors calculated an attributable proportion (AP) and interaction contrast ratio (ICR) for each outcome. In plain terms: these measure how much of the harm comes specifically from the combination of being hypocalcemic and receiving calcium, above and beyond what you'd expect from simply adding "hypocalcemia is bad" and "sicker patients get treated" together. A high AP/ICR means the interaction itself is doing real work, not just two separate risk factors stacking additively.
| Outcome | AP | ICR | Interaction strength |
|---|---|---|---|
| AKI | 40.0% | 0.99 | Strong — noted interaction effect |
| ARDS | 57.3% | 4.27 | Strong — noted interaction effect |
| PE | 30.9% | 3.28 | Strong — noted interaction effect |
| Sepsis | 16.9% | 0.90 | Weak |
| Death | 16.3% | 0.32 | Weak |
| MI | 5.8% | 0.29 | Weak |
The authors' own conclusion leads with AKI, ARDS, and PE specifically, not the full six-outcome list the Instagram post presented as equally weighted. For mortality, MI, and sepsis, the "weak interactive effect" language means the elevated odds are more consistent with sicker patients simply having worse outcomes across the board, calcium or not, whereas for AKI, ARDS, and PE, the combination of hypocalcemia plus early supplementation looks like it's doing something specific beyond that baseline severity. That's a real distinction the summary version flattened into six equally-alarming bullet points.
If this were purely sicker patients getting calcium, the association should be weakest, not strongest, in patients least likely to be predicted to get it.
The authors built a supplemental propensity model predicting the likelihood of calcium supplementation (using age, sex, mechanism, shock index, RBC units, lactate, hemoglobin, hypocalcemia, ISS, and region-specific AIS ≥3 injury), then stratified the main analysis by that predicted probability. The logic: if the observed harm is really just "clinicians correctly identify sicker patients and give them calcium," the association should be strongest among patients with a high predicted probability of receiving it, and weak or absent among patients unlikely to receive it.
The significant associations were observed specifically among patients in the 10–14% predicted-probability group, i.e., patients who were relatively unlikely to be flagged for calcium in the first place. In that low-propensity stratum: AKI OR 5.52 (95% CI 1.69–17.98), ARDS OR 8.74 (95% CI 2.29–33.39), DVT OR 9.67 (95% CI 3.27–28.53), and PE OR 4.51 (95% CI 1.17–17.36). Comparisons for cardiac arrest, mortality, infection, MI, sepsis, and stroke were not significant in this stratified analysis.
This is the strongest single piece of evidence in the paper against "this is all just confounding by indication." It doesn't eliminate confounding as a concern, residual confounding from unmeasured variables is still possible, but it does mean the simplest version of that critique (sicker patients get calcium, sicker patients do worse, that's all this is) doesn't fully explain what they found. If anything, the fact that the signal concentrates in patients who weren't expected to need calcium is the more clinically interesting and more concerning version of this finding.
A properly designed answer to this question is already underway: the CAVALIER trial (Calcium and Vasopressin following Injury Early Resuscitation), a randomized, placebo-controlled trial of empiric prehospital calcium, led by Jason Sperry at the University of Pittsburgh with DoD funding. The Hewgley paper specifically flags that CAVALIER should pay close attention to the subset of its own enrolled patients who end up needing only minimal transfusion, since that's exactly the population this retrospective study raises concern about.
The core finding holds up under direct review: in patients receiving only 1–3 units of RBCs, early empiric calcium in hypocalcemic patients associated with significantly higher odds of AKI, ARDS, DVT, PE, MI, sepsis, and death, and the propensity-stratified analysis argues this isn't purely confounding by indication. But the paper itself draws a more precise conclusion than "calcium is harmful": it leads with AKI, ARDS, and PE as the outcomes with genuine interaction effects, treats mortality, MI, and sepsis as weaker signals likely reflecting overall illness severity, and explicitly preserves a role for empiric calcium in non-responders and transient responders during active massive transfusion. The reasonable bedside takeaway is narrower and more specific than the Instagram version suggested: reflexive calcium after a small transfusion, in a patient who is not actively continuing to bleed, deserves real scrutiny; calcium in a non-responder mid-MTP does not carry the same caveat.
This page has now been checked directly against the full published paper rather than a secondhand summary. Two corrections from the original build: sepsis was a significant outcome the Instagram post omitted, and the paper's own interaction-effect analysis (AP/ICR) singles out AKI, ARDS, and PE as the outcomes with the strongest evidence of a true interaction, treating mortality, MI, and sepsis as weaker signals more consistent with baseline severity. One correction to my own prior audit: I previously could not confirm this paper existed in a retrievable form; it does, and the numbers I'd cited secondhand all matched the primary source exactly once verified directly.