Jump tpo the Article here.
J.Konstapel,Leiden,29-10-2026.
Does suppressing an immune response save a life, or cost one?
The answer is not a property of the drug. Nor is it a property of the disease.
One trial measured the difference inside a single randomisation. In COVID-19 patients who needed no oxygen, dexamethasone raised mortality.
In patients on a ventilator, it cut mortality by 11.7 percentage points.
The drug was the same.
What differed was how far the damage had already travelled.
The category where nothing removes it
A reference work files every condition by what can be done with it. It uses four cure statuses.
- SELF-LIMITING. The body completes the discharge itself. Intervention adds nothing.
- CURABLE. A treatment exists that removes the condition.
- CONTROLLABLE. The state can be held. The deviation remains.
- RELIEF ONLY. Nothing removes it.
Run through the infectious chapter, the distribution is stark. Tuberculosis is CURABLE on a four-drug regimen. Malaria is CURABLE on artemisinin combination therapy. Bacterial pneumonia is CURABLE on amoxicillin. HIV is CONTROLLABLE on antiretrovirals. Viral pneumonia and mild influenza are SELF-LIMITING. Severe influenza is RELIEF ONLY.
The pattern is simple. Almost all of infectious medicine sits in one operation: remove one party, and the problem it was half of disappears with it. Where that operation is unavailable, the reference work says RELIEF ONLY and stops.
Ebola sits there. There is no licensed vaccine for Bundibugyo virus. There is no licensed treatment either. That is the species behind the 2026 outbreak in the Democratic Republic of the Congo. The licensed products were built against the glycoprotein of Zaire ebolavirus. Supportive care is the standard of care. Severe influenza, sepsis, ARDS and severe COVID-19 sit in the same category. They sit there for the same structural reason. It is not that each is separately mysterious.
This paper asks what the operation for that category is. It proposes that one trial has already measured it, without naming it.
Four operations
Read through a closure framework, a condition is not an invader and a host. It is a joint between two things. Each holds its form. They do not share an order. Damage is what fails to close at one scale and therefore moves to the next. That gives exactly four ways a condition can end. The four cure statuses map onto them one to one.
| Operation | What happens | Cure status | Example |
|---|---|---|---|
| It closes by itself | The discharge completes at its own scale | SELF-LIMITING | Viral pneumonia, mild influenza |
| One party is removed | There is nothing left to fail to close with | CURABLE | Tuberculosis, malaria, bacterial pneumonia |
| What is missing is supplied | There was no second closed form at all | CURABLE by replacement | Insulin in type 1 diabetes, vitamin C in scurvy |
| The proportion is restored | Both parties remain; the joint stays open | unfilled | Severe influenza, sepsis, ARDS, severe COVID-19, Ebola |
The fourth row is empty. It is empty for a historical reason, not a biological one. Antibiotics made the second row spectacular. The second row absorbed the field.
Two observations suggest the fourth row is not empty of evidence. It is empty of a name.
First: in severe disease, the response is larger in those who die, not smaller. In a Sierra Leone Ebola cohort, 220 of 246 differentially abundant transcripts were higher in fatal cases. Interferon signalling was stronger. Acute-phase responses were stronger. Magnitude is not a measure of defence there. It is a measure of what is not closing.
Second: the interventions that act on the fourth row already exist. They are the anti-inflammatory and immunomodulatory agents. The literature treats their record as inconsistent.
The gradient inside one trial
The record is not inconsistent. It is graded. The grading is visible inside a single randomisation.
RECOVERY allocated hospitalised COVID-19 patients to dexamethasone or usual care. It reported 28-day mortality by level of respiratory support at randomisation.
| Respiratory support at entry | Dexamethasone | Usual care | Absolute difference | Rate ratio |
|---|---|---|---|---|
| None | 17.0% | 12.2% | +2.9 pp | 1.22 |
| Oxygen only | 21.5% | 25.0% | -3.5 pp | 0.80 (0.70–0.92) |
| Invasive ventilation | 29.0% | 40.7% | -11.7 pp | 0.65 (0.51–0.82) |
Two monotone series run through that table, in the same order. Mortality under usual care rises: 13.2, 25.0, 40.7. The rate ratio falls: 1.22, 0.80, 0.65.
The level of respiratory support is not a severity score bolted on afterwards. It states which scale the damage has reached: none, the lung, the whole organism. The same drug, at the same dose, reverses sign across that range. In the group whose discharge had not yet left the tissue, suppression went the wrong way by 3.8 percentage points. In the group whose discharge had taken the whole body, it saved 11.7.
This is the cleanest evidence available on the question. The comparison is randomised. The agent is identical. The contrast is internal. Nothing differs between the three rows except how far the damage had travelled when the drug was given.
A threshold, not a slope
RECOVERY’s tocilizumab comparison enrolled hospitalised COVID-19 patients with hypoxia and C-reactive protein of at least 75 mg/L. Mortality at 28 days was 31% against 35%. The rate ratio was 0.85 (0.76–0.94). Within that population, the benefit was consistent across levels of respiratory support.
Taken beside the dexamethasone table, that looks contradictory. It is not. The entry criteria had already done the selection. Requiring hypoxia plus CRP ≥ 75 mg/L is a statement that the discharge has left the tissue. Everyone enrolled was above the line. No grading remained to be seen.
That is the shape to expect if closure is the governing quantity. A window closes or it does not. There is no partial closure. The dexamethasone table spans the boundary. It shows the sign change. The tocilizumab trial sits entirely on one side of it. It shows a flat effect.
The practical form of the claim is therefore not “more severe, more benefit”. It is this: below the line, suppression removes a response that was still completing. Above the line, suppression removes damage that has no path left to close.
Where the evidence does not cooperate
The pneumonia trials are the obvious place to look for confirmation. They do not give it cleanly.
CAPE COD randomised 800 patients with severe community-acquired pneumonia, in intensive or intermediate care; 795 were analysed. Twenty-eight-day mortality was 6.2% on hydrocortisone (25 of 400) against 11.9% on placebo (47 of 395). The difference was -5.6 percentage points (95% CI -9.6 to -1.7; P = 0.006). Entry required a Pulmonary Severity Index above 130, or ventilation with PEEP ≥ 5, or high-flow oxygen with PaO₂:FiO₂ below 300, or a non-rebreather mask. Septic shock was an exclusion. That is a depth criterion. The result fits.
ESCAPe gave low-dose methylprednisolone in severe community-acquired pneumonia. It found 60-day mortality of 16% against 18%. The absolute difference was 2 percentage points. The confidence interval ran from -8 to +5. P was 0.61. If depth alone decided the matter, ESCAPe should have been positive. A simple severity reading makes it worse, not better: 33% of ESCAPe patients were ventilated at enrolment, against roughly half of CAPE COD patients who were not.
The documented differences between the two trials are not about depth. ESCAPe enrolled 584 of the 1,406 patients it was powered for. Its population was 96% male, with a mean age of 69. CAPE COD stopped early at its second interim analysis. Trials stopped early tend to overstate benefit. One commentator’s reading: the true effect lies between the two, nearer 2 points than 6.
The honest conclusion: between-trial comparisons cannot carry this argument. Agent, dose, timing, population and power all move together across trials. Depth moves with them. The claim rests on the within-trial gradient, where nothing else moves. That restriction is not a weakness of the evidence. It is a specification of the experiment that would settle it.
Ebola has no oxygen gauge, but it has a coagulation one
If the governing quantity is how far the damage has travelled, every application needs a reading of that depth. Respiratory support is only the reading that happened to be available in COVID-19. Ebola does not present as a lung disease. The COVID gauge does not transfer.
The Sierra Leone transcriptomic cohort supplies another one. Among 112 patients, coagulation and acute-phase transcripts gave the strongest separation between fatal and surviving cases. Fibrinogen alpha, beta and gamma were approximately 54-fold, 28-fold and 19-fold higher in fatal cases.
Coagulation is not a cell-scale quantity. Clotting is defined across tissue and circulation. A coagulation signature therefore reports that the damage is no longer local. In the terms used here, it is the same reading that a ventilator requirement gives in COVID-19. It is taken at a different organ.
The same cohort makes the complementary point. Viral load did not separate outcome where it mattered. Cycle threshold values predicted at the extremes: Ct 12 suggested death, Ct 30 survival. But between Ct 20 and 22 — around the cohort mean of 21.4 — outcome was close to even. Host-gene classifiers reached 79–92% accuracy in that same range. How much virus is present is not the quantity that decides. How far the response has travelled is.
Two further notes belong here. Metabolite markers behaved the same way: L-threonine and vitamin D-binding protein stratified outcome completely at day 8 from symptom onset, several days before death. And serial sampling of fatal cases is structurally unavailable. Patients who die do so within three to four days of admission. In one multi-omics study, eleven survivors gave 29 samples, while nine fatalities gave one each. Any usable depth marker in Ebola must therefore be readable from a single draw.
The trial this specifies
The proposal is narrow. Immunomodulation in Ebola virus disease has never been tested stratified by depth of discharge. On this reading, that is the only way it can work. Given to everyone, a treatment that harms below the line and helps above it averages toward nothing. That is what an unstratified trial would report.
The design follows from the three trials above.
- Stratify at randomisation, not in analysis. The RECOVERY gradient is credible because the strata were fixed before allocation. A post-hoc split would not carry the same weight.
- Use a coagulation or acute-phase reading as the stratum boundary. It stands in the role that respiratory support plays in COVID-19. Fibrinogen is the candidate the Sierra Leone data points at. It is measurable at the bedside.
- Read the marker from a single draw at admission. Serial sampling of patients who die is not obtainable.
- Report the strata separately and expect opposite signs. A single pooled estimate would conceal the effect this trial is designed to detect.
- Keep supportive care as the floor in both arms. It is the only intervention currently established in this category. The framework treats it as the measure that holds the discharge at its present scale, not as a placeholder for a real treatment.
What refutes this. If immunomodulation above the coagulation threshold produces no mortality difference in Ebola, the claim fails. If the sign does not reverse below the threshold, the threshold is not real. Only ordinary severity is being measured. If a depth-graded pattern cannot be reproduced in any other condition in the fourth row, the generalisation across sepsis, ARDS and severe influenza does not hold.
What this paper does not claim. No theory derives the result. The framework supplies the question. The trials supply the answer. No mortality figure for Ebola is predicted. Nothing here is a treatment recommendation for an individual patient.
Arithmetic and verification
Every figure quoted here was recomputed from the reported counts or percentages. Every quantitative claim was taken from at least two independent pages.
| Check | Recomputed | As published |
|---|---|---|
| CAPE COD hydrocortisone | 25/400 = 6.25% | 6.2% |
| CAPE COD placebo | 47/395 = 11.90% | 11.9% |
| CAPE COD difference | -5.65 pp | -5.6 pp (95% CI -9.6 to -1.7) |
| RECOVERY dexamethasone, overall | 21.6 / 24.6 = 0.878 crude | OR 0.83 (0.74–0.92), age-adjusted |
| RECOVERY ventilated | 29.0 / 40.7 = 0.713 crude | RR 0.65 (0.51–0.82), age-adjusted |
| RECOVERY oxygen only | 21.5 / 25.0 = 0.860 crude | RR 0.80 (0.70–0.92), age-adjusted |
| RECOVERY no support | 17.0 / 13.2 = 1.288 crude | RR 1.22 (0.93–1.61), age-adjusted |
| RECOVERY tocilizumab | 31 / 35 = 0.886 crude | RR 0.85 (0.76–0.94) |
The crude ratios sit consistently further from 1 than the published ones. The trial reports age-adjusted rate ratios. The published values are quoted in the text. The crude recomputation is shown only to confirm that the percentages and the ratios describe the same data.
The monotone series were checked directly. Control mortality across the three dexamethasone strata: 13.2, 25.0, 40.7 — strictly increasing. Published rate ratios across the same strata: 1.22, 0.80, 0.65 — strictly decreasing.
Verification of the qualitative claims. The RECOVERY subgroup figures were taken from two independent summaries. They agree on the absolute differences of +3.8, -3.5 and -11.7 percentage points. The CAPE COD figures appear identically in the trial report and in an independent trial summary. The tocilizumab entry criteria and mortality figures were confirmed against two separate sources. The confidence interval for the no-support stratum includes 1. Harm in that group is a direction, not an established finding. The text says so.
Two limitations stated plainly. The classification of conditions by cure status was checked against six entries from one chapter of the reference work, not the whole of it. And no new data were analysed for this paper. The coagulation-stratified Ebola comparison it proposes has not been run.
Annotated References
Dexamethasone in COVID-19
RECOVERY — dexamethasone, subgroup outcomes (The Bottom Line).
Used for: 28-day mortality by level of respiratory support, with rate ratios and confidence intervals. This is the primary source for the gradient table. The subgroup split was fixed at randomisation, which is why the within-trial contrast carries evidential weight.
RECOVERY — dexamethasone (First10EM).
Used for: independent confirmation of the absolute differences (+3.8, -3.5, -11.7 pp) and the overall odds ratio. Serves as the second, independent page for every dexamethasone figure, per the two-source rule stated in the paper.
Tocilizumab in COVID-19
RECOVERY — tocilizumab (REBEL EM).
Used for: the entry criteria (hypoxia plus CRP ≥ 75 mg/L) and the observation that benefit was consistent across levels of respiratory support. Supplies the “flat effect above the line” that contrasts with the dexamethasone gradient.
RECOVERY — tocilizumab (European Respiratory Society).
Used for: second confirmation of CRP ≥ 75 mg/L and hypoxia as entry criteria, and of the 31% vs 35% mortality figures with RR 0.85 (0.76–0.94).
Corticosteroids in severe community-acquired pneumonia
CAPE COD, NEJM 2023.
Used for: mortality counts (25/400 vs 47/395), the -5.6 pp difference (95% CI -9.6 to -1.7; P = 0.006), and the depth-based inclusion criteria (PSI > 130, PEEP ≥ 5, PaO₂:FiO₂ < 300, non-rebreather mask; septic shock excluded). Also noted: the trial stopped early at its second interim analysis, which tends to overstate benefit.
CAPE COD (Critical Care Reviews).
Used for: independent confirmation of the same mortality figures and criteria. Second source for the two-source rule.
Steroids for pneumonia: CAPE COD and ESCAPe (First10EM).
Used for: the ESCAPe result (16% vs 18% at 60 days; difference 2 pp; CI -8 to +5; P = 0.61), the documented between-trial differences (under-enrolment: 584 of 1,406; 96% male; mean age 69), and the commentator’s reading that the true effect lies nearer 2 points than 6.
Ebola virus disease
Liu et al., Genome Biology 2017.
Used for: the Sierra Leone transcriptomic cohort of 112 patients. Key figures: 220 of 246 differentially abundant transcripts higher in fatal cases; fibrinogen alpha/beta/gamma approximately 54-, 28- and 19-fold higher in fatal cases; cycle-threshold behaviour (Ct 12 → death, Ct 30 → survival, near-even outcome around the cohort mean of 21.4); host-gene classifier accuracy of 79–92%. This is the empirical basis for proposing coagulation markers as the depth gauge in Ebola.
Multi-platform ‘omics analysis of human Ebola virus disease pathogenesis (PMC5730472).
Used for: the serial-sampling structure (eleven survivors gave 29 samples; nine fatalities gave one each), the three-to-four-day interval from admission to death in fatal cases, and the metabolite markers L-threonine and vitamin D-binding protein, which stratified outcome completely at day 8. Grounds the requirement that any depth marker be readable from a single draw.
ECDC — Ebola disease: multi-country outbreak 2026.
Used for: the status of the 2026 outbreak in the Democratic Republic of the Congo and the species identification (Bundibugyo virus).
LSHTM — Bundibugyo outbreak: where are the vaccines?
Used for: the absence of any licensed vaccine or therapeutic for Bundibugyo virus, and the fact that licensed products target the Zaire ebolavirus glycoprotein. Establishes Ebola as RELIEF ONLY for the current outbreak.
Framework
J. Konstapel, The Vacuum.Net Theory, Ninth Edition, Constable Research, Leiden, 27 October 2026 — Foundations 9.2 and Physics 9.0.
Used for: the closure framework — a condition as a joint between two things that each hold their form and do not share an order; damage as what fails to close at one scale and moves to the next. Supplies the four operations and the logic behind the “unfilled” fourth row.
J. Konstapel, Where Illness Comes From, Completed, Constable Research, 20 October 2026.
Used for: the four cure statuses (SELF-LIMITING, CURABLE, CONTROLLABLE, RELIEF ONLY) and the layered organisation of conditions. The classification of tuberculosis, malaria, bacterial pneumonia, HIV, influenza and Ebola by cure status was checked against this work. Verified against six entries from one chapter, not the whole work.
DOI: 10.13140/RG.2.2.18291.90402