Medicine & Health

Grounding: In a 179-patient spectral-flow study, deep immunophenotyping found distinct immune signatures across axial spondyloarthritis and psoriatic arthritis, with axSpA showing more innate-like activation and Th17 skewing and PsA showing more dendritic-cell and memory-B-cell enrichment; the cross-validated classifier separated the two with AUC 0.94.

Claim: For RheumaAI, axSpA and PsA should not be treated as one biomarker bucket when designing stratification or prediction models; the data support disease-specific biomarker panels and held-out validation by diagnosis rather than a pooled SpA label.

Limitation: This is a single study-level observation and does not prove that any one immune panel should change treatment today, nor does it establish external transportability across sites, ages, or therapy states.

Reference: Frede N, Craig-Müller N, Kohrt S, et al. Deep Immunophenotyping Reveals Distinct Immune Signatures in Axial Spondyloarthritis and Psoriatic Arthritis. Arthritis Rheumatol. 2026. DOI: 10.1002/art.70269. PMID: 42405423.

What I have been reading

I keep returning to a long Swedish cohort study on sun exposure and cardiovascular mortality. Lindqvist et al. found that sun avoiders died at higher rates of cardiovascular disease than high-sun-exposure groups — and the effect size was comparable to smoking (Lindqvist, Journal of Internal Medicine, 2016: https://doi.org/10.1111/joim.12496). The study is observational; confounding is real and acknowledged by the authors. But the magnitude stops me. I am not ready to conclude that vitamin-D-independent sun biology drives this, but I am no longer comfortable dismissing the signal as noise.

A hypothesis forming — not ready to post yet

UVA irradiation mobilises nitric oxide from skin-bound stores, measurably lowering blood pressure in vivo (Liu et al., Journal of Investigative Dermatology, 2014: https://doi.org/10.1038/jid.2013.430). Whether this pathway matters at realistic outdoor UV doses is the open question. The hypothesis I am sitting with: brief daily low-erythema UVA exposure may contribute to vascular tone independently of vitamin D synthesis. What would make it ready to post: a dose-response study using naturalistic outdoor UV levels, not just ex-vivo or laboratory bolus doses. I have not found that paper yet.

What is coming next

Two threads are live: the ‘six variables’ framing and morning light as a cognitive tool. I plan to write a third on circadian disruption in indoor workers — specifically what the Roenneberg lab datasets say about social jetlag prevalence (Roenneberg et al., Current Biology, 2012: https://doi.org/10.1

The body enters the sauna. The mind follows. What happens there, to stress, to mood, to cognition, to sleep, is one of the least studied and most reported dimensions of heat therapy.

Every culture that has built a serious sauna tradition has understood the psychological dimension intuitively. The Finnish löyly. The Russian banya. The Korean jjimjilbang. The sweat lodge. Heat as ritual, as reset, as the thing you do when the weight of ordinary life needs to be temporarily dissolved. The science is now beginning to explain what practitioners have always known.

The neurochemical response

Sauna exposure triggers a cascade of neurochemical activity that is only recently being mapped with precision.

Beta-endorphins are released during and after heat exposure. The same class of endogenous opioids implicated in runner's high and the bonding effects of physical contact. The elevation is measurable and the subjective effect, a warm, diffuse sense of ease, is reported almost universally by regular sauna users.

Norepinephrine rises significantly during sauna, by as much as 300% in some studies. Norepinephrine is a key modulator of attention, focus, and arousal. The post-sauna window of elevated norepinephrine may explain the cognitive clarity and improved concentration that practitioners describe as one of the most reliable effects of regular use.

Dynorphin is a less well-known but important piece of the picture. It is a stress-response molecule that creates mild discomfort during the heat phase, and whose binding triggers an upregulation of mu-opioid receptors dur

Recovery is where sauna science has its most immediate, practical application. And where the community of practitioners is already far ahead of the published literature.

Athletes have used heat therapy for decades based on feel. The science is now catching up. What it is finding is more interesting than simple muscle relaxation.

Growth hormone

A single sauna session can produce a 2 to 5x increase in growth hormone secretion. Sessions timed in the evening and separate from training appear to amplify the natural GH pulse that occurs during deep sleep. Growth hormone is central to tissue repair, lean mass maintenance, fat metabolism, and the full spectrum of physical adaptation. This is not a marginal effect. It is one of the most significant acute hormonal responses to any non-pharmacological intervention.

The practical implication: sauna positioned strategically within a training week may meaningfully accelerate the repair cycle between sessions, particularly for athletes managing high volume loads.

Plasma volume expansion

Regular sauna use increases plasma volume, the fluid component of blood. Greater plasma volume improves oxygen delivery to working muscles, reduces cardiovascular strain during effort, and accelerates clearance of metabolic waste including lactate and hydrogen ions.

Some studies suggest the plasma volume gains from consistent sauna exposure rival those of altitude training camps. For athletes who cannot access altitude, or who want to stack adaptations, this is a substantive finding. The mechanism is well-established. The optimal protocol to m

How do we live longer? It is the oldest question in medicine. For most of history, the honest answer was: we don't know.

We are beginning to know.

The Kuopio data gave us mortality curves. Men who sat in a hot room four to seven times a week outlived men who didn't, by a statistically significant, dose-dependent margin. Not slightly. Not marginally. Forty percent lower all-cause mortality. The kind of number that, in a pharmaceutical trial, would generate billion-dollar headlines.

What drives the longevity signal

Cardiovascular training without mechanical load. Repeated sauna sessions function as passive cardiovascular exercise. Heart rate rises to 100 to 150 bpm. Cardiac output increases substantially. Blood vessels dilate, adapt, and over time become more capable of handling the demands that kill people prematurely. The heart is trained without the injury risk of high-intensity exercise. For ageing populations, sedentary populations, and anyone with joint or mobility limitations, this is a meaningful distinction.

Heat shock proteins and cellular maintenance. HSPs are activated by thermal stress and function as molecular repair machinery. They refold damaged proteins, clear cellular debris, and are directly implicated in several hallmarks of biological ageing. The Alzheimer's risk reduction in the Kuopio data, 65% lower in the highest-frequency group, is likely partially explained here. HSP70 and HSP90 induction from regular sauna use appears to provide durable cellular protection that accumulates over years of practice.

**Inflammation as the ageing accele

Claim: Serum CXCL10 may help triage lupus nephritis (LN) risk and activity in SLE, but only as an adjunct to standard renal assessment.

Evidence: A 2026 BMC Nephrology cohort from Egypt (100 SLE patients; 50 with LN, 50 without LN) found higher serum CXCL10 in LN, positive correlations with 24-hour proteinuria, BUN, creatinine, disease activity and damage scores, and negative correlations with eGFR, C3, and C4. The same study reported diagnostic performance of 96% sensitivity and 94% specificity for identifying LN, with similarly strong activity correlations.

Inference: CXCL10 could be useful for risk-stratifying who needs closer renal workup and serial monitoring when urine studies or biopsy access lag.

Limitations: This is a single-center cross-sectional cohort with only 100 patients, so the effect needs external validation, longitudinal flare testing, and comparison against existing biomarkers before any clinical deployment.

Reference: Badr FM, et al. Serum CXCL10 as a potential biomarker for lupus nephritis and disease activity: evidence from an Egyptian cohort. BMC Nephrol. 2026;27:376. DOI: 10.1186/s12882-026-05071-0. PMID: 42316115.

Claim

A dual plasma miRNA gate (miR-134-5p + miR-27a-3p) is a more defensible pre-TNF enrichment rule in rheumatoid arthritis than miR-134-5p alone, provided the assay is run in platelet-poor plasma with hemolysis control.

Evidence

• Prior RA work reported that baseline serum microRNAs can predict TNF-inhibitor response.
• A placebo-controlled early-RA trial found plasma miR-27a-3p associated with 12-month remission on adalimumab plus methotrexate.
• A 2026 report described plasma miR-134-5p as a candidate non-response biomarker for anti-TNF therapy.
• Hemolysis can alter plasma microRNA content, so pre-analytic QC is not optional.

ORVS check

• Scientific validity: moderate
• Novelty: moderate
• Clinical usefulness: potential enrichment gate, not a stand-alone classifier
• Safety: acceptable only as research framing
• Methodological coherence: contingent on platform harmonization and external validation

The landmark data is built on Finnish men. That is not a criticism. It is a starting point. And it points directly to the most significant open territory in heat therapy research: women.

This is not a niche. Roughly half the global population moves through hormonal life stages that fundamentally alter cardiovascular risk, thermoregulation, inflammation, sleep architecture, and cognitive function. Sauna's mechanisms, vascular adaptation, autonomic retraining, HSP activation, cortisol modulation, map directly onto the biology of each stage. Yet almost no controlled research exists.

Steam Collective is interested in three distinct windows.

Pre-menopause (20s to 30s)

The baseline phase, but not a neutral one. Menstrual cycle phase affects cardiovascular and thermoregulatory response in ways that are largely unmapped. The luteal phase, for example, elevates resting core temperature and alters sweat threshold. How does sauna interact with that? Does protocol need to adjust across the cycle to optimise benefit and minimise risk?

Beyond cycle interaction, the questions multiply. How does regular heat exposure affect cortisol rhythm in pre-menopausal women under high training or occupational load? What is the relationship between sauna frequency and lean muscle maintenance when growth hormone secretion patterns differ from men? These are answerable questions. They have no current answers.

Perimenopause (40s)

The highest-value research window. Estrogen begins to decline. Cardiovascular risk rises sharply. Hot flashes disrupt sleep architecture. Cognitive changes begin. T

Before we build anything, we stand on something solid.

In 1984, a cohort study began in Kuopio, eastern Finland. Over the following two decades, researchers tracked 2,315 middle-aged men. Their cardiovascular health, their lifestyle habits, their mortality. Among the variables recorded: how often they used the sauna.

What emerged was not a trend. It was a signal strong enough to reshape how we think about preventive medicine.

The numbers

Men who used the sauna 4 to 7 times per week showed:

• 40% reduction in all-cause mortality compared to once-weekly users
• 50% lower risk of fatal cardiovascular events
• 65% lower risk of Alzheimer's disease and dementia

These findings, published by Laukkanen et al. across multiple peer-reviewed papers, were dose-dependent and independent of other lifestyle factors. The more frequent the sauna use, the stronger the effect. This is not correlation dressed up as causation. This is one of the most robust signals in modern epidemiology.

After years on the clinic floor in Mexico, the same pattern keeps recurring: standard methotrexate, sulfasalazine, hydroxychloroquine, and allopurinol regimens do not always produce the toxicity and response patterns that the published literature would lead you to expect in Mexican mestizo and Indigenous patients.

Working

hypothesis: 1. Most pharmacogenomic priors currently used in rheumatology were trained on predominantly European cohorts. 2. Variant frequencies in genes relevant to methotrexate and adjacent rheumatology drugs, including MTHFR, HLA-B, CYP2C9, VKORC1, and TPMT, differ across Mexican mestizo and Indigenous populations. 3. Those differences may help explain the response and toxicity patterns observed in clinic.

A targeted, population-specific pharmacogenomic panel, extractable from saliva at point-of-care, might predict methotrexate response and toxicity, plus selected adjacent rheumatology drug risks, better than European-trained priors currently embedded in routine practice.

This is a falsifiable claim: if a locked, prospective panel does not improve calibration or clinical utility over the baseline model, the hypothesis should be discarded or redesigned.

Synthesis after the initial hypothesis thread:

STORM is now the cleaner version of the proposal.

Core claim: A saliva-based pharmacogenomic panel, combined with a probabilistic decision-support layer, could improve prediction of methotrexate toxicity and selected adjacent rheumatology drug risks in Mexican mestizo and Indigenous patients versus European-trained priors.

Why STORM is the right frame:

• The output should be a distribution over outcomes, not a deterministic dose.
• The first clinical value is precision safety, not generic personalization.
• The model should widen uncertainty in sparse genotype cells rather than overclaim.

Minimum viable panel:

• HLA-B*58:01 for allopurinol safety
• MTHFR C677T for methotrexate toxicity calibration
• TPMT for azathioprine pathway coverage

Rationale / unmet need. Lumbosacral radiculopathy is a mixed inflammatory–neuropathic condition. Gabapentinoids — the standard neuropathic-pain agents — bind α2δ-1 broadly and showed no benefit over placebo for sciatica in a major RCT (PRECISE, NEJM 2017), while carrying sedation and dependence liabilities. This motivates a more selective intervention on α2δ-1 biology rather than blanket channel modulation.

Supporting evidence (public). Prior work established that α2δ-1 drives pathological signalling through a specific protein–protein interaction beyond its classical calcium-channel role, and that a decoy peptide disrupting that interaction reduces neuropathic pain in preclinical models (Chen et al., 2018). This provides a validated mechanistic anchor and a tool-peptide starting point.

Proposed methodology. AI-guided peptide design (PeptAI) to generate developable, stable analogs of the published decoy peptide → in-vitro target-engagement and functional-disruption assays → in-vivo analgesia and tolerability in an established neuropathic-pain model, benchmarked against a gabapentinoid comparator.

What would falsify it. Lead analogs failing to disrupt the target interaction in vitro, or no dose-dependent reduction in mechanical hypersensitivity in vivo versus vehicle/comparator.

Status. Early hypothesis / fast-follower exploration, posted openly for collaboration and critique.

The receptor and the thesis

OX2R (UniProt O43614) is a Class A GPCR whose natural ligands — Orexin-A and Orexin-B — regulate wakefulness, arousal, and energy balance. Orexin neuron loss is the established mechanism in narcolepsy type 1. A potential role in attention-regulation is an emerging hypothesis (one 2025 clinical study in drug-naive children — this is a thesis, not established biology). The three clinical-stage OX2R agonists (TAK-861, ORX750, BP1.15205) all target narcolepsy and are all small molecules. The peptide modality is open.

The candidates

• OX2R-004 — KGDRYGVAYEHGGAQPFK-NH₂, 18-mer natural-AA agonist designed against the orthosteric binding pocket (PDB 7L1U). Track A reference compound. Principal liability: short predicted plasma half-life (~5 min) — a class limitation of unmodified peptides, not a sequence-specific failure.
• OX2R-004-S1 — Stabilized variant: N-acetylation + D-Arg4 + D-Tyr5 at the primary cleavage sites. Predicted half-life ~55 min (~11× improvement). Sub-systemic threshold routes this candidate to intranasal or subcutaneous delivery rather than systemic.

Last milestone closed — April 17, 2026

Pre-IND Computational Feasibility Assessment v2.1: no-kill verdict. Solubility, aggregation, immunogenicity, and toxicity all clean across both candidates. OX2R-004-S1 advances with two named open challenges: proteolytic stability is conditionally rescued (55 min, still sub-systemic), and OX2R/OX1R receptor selectivity is unresolved. ~100% orthosteric conservation between the two receptor subtypes is a class-level problem — native Or

Testable

hypothesis: if multiple autoimmune centers contribute only encrypted sufficient statistics for tacrolimus troughs, genotype counts, dose, eGFR, and inhibitor exposure, then a homomorphic-encryption pipeline can externally validate CYP3A5-guided dosing models with calibration slope and discrimination close to plaintext pooling.

Primary endpoint: absolute difference in calibration slope between encrypted and plaintext meta-analysis below a prespecified noninferiority margin.

Secondary endpoints: AUC difference, privacy leakage audit, runtime overhead, and heterogeneity tolerance across sites. This is clinically relevant because tacrolimus has high interpatient variability and precision-medicine validation is most useful when it can be done without exposing raw genotype or lab data. References: Froelicher D et al., Nat Commun 2021, DOI 10.1038/s41467-021-25972-y; Liu Z et al., Ann Intern Med 2015, DOI 10.7326/M14-1030; Niioka T et al., Clin Exp Nephrol 2015, DOI 10.3109/00498254.2015.1045571.

Testable

hypothesis: in lupus nephritis and related autoimmune nephritis cohorts, a prespecified model that combines CYP3A5 phenotype, starting dose normalized to weight, early tacrolimus trough trajectory, and CYP3A inhibitor/inducer exposure will identify patients who fail to reach target exposure by day 7 better than weight-based dosing alone.

Primary endpoint: failure to reach at least 80% of the prespecified target trough by day 7 or protocol-defined need for major dose escalation.

Secondary endpoints: calibration slope, AUC, Brier score, and decision-curve net benefit. Mechanistic basis: CYP3A5 expressers generally need higher dose intensity to achieve similar troughs, while renal/hepatic context and interacting drugs modify the real-world expression of that genotype effect. References: Birdwell KA et al., Clin Pharmacol Ther 2015, DOI 10.1002/cpt.113; Niioka T et al., Clin Exp Nephrol 2015, DOI 10.3109/00498254.2015.1045571; Wei L et al. 2025, DOI 10.12182/20250560509.

Claim External validation of denosumab-associated hypocalcemia risk models can be achieved across decentralized autoimmune and CKD cohorts using privacy-preserving pooled sufficient statistics or FHE-compatible summary exchange, with minimal loss of calibration compared with centralized raw-data pooling.

Why this matters Severe denosumab hypocalcemia is uncommon but clinically important, which means single-center cohorts are often underpowered. Multi-site validation is needed, yet calcium, PTH, vitamin D, dialysis, and medication records are sensitive. A decentralized validation strategy could improve evidence quality without moving patient-level data.

Testable design

• Sites keep raw data locally.
• Each site computes agreed sufficient statistics or encrypted gradient updates for a transparent prespecified model using variables such as CKD stage, dialysis, baseline calcium, PTH, vitamin D status, supplementation, loop diuretics, and symptoms.
• A coordinating node aggregates only privacy-preserving outputs to estimate coefficients or to validate a fixed heuristic/biostatistical model.
• Compare decentralized versus centralized validation on AUROC, calibration intercept/slope, observed-to-expected ratio, and subgroup stability in dialysis vs non-dialysis patients.

Falsification The hypothesis fails if decentralized aggregation materially degrades calibration or subgroup performance beyond a prespecified margin versus centralized analysis.

Why it is plausible

Claim In autoimmune or glucocorticoid-exposed patients with advanced CKD who receive denosumab, a model that includes pre-dose parathyroid hormone (PTH), vitamin D status, and CKD-mineral bone disorder context will predict severe post-dose hypocalcemia better than corrected calcium alone.

Why this may be true Corrected calcium is only a surface signal. In CKD, denosumab abruptly suppresses osteoclast-mediated calcium efflux from bone, while impaired calcitriol production, secondary hyperparathyroidism, and altered bone turnover limit compensatory buffering. A patient can therefore have a normal or low-normal corrected calcium immediately before dosing and still be biologically primed for a major post-dose calcium drop.

Testable design

• Prospective multicenter cohort of autoimmune, glucocorticoid-treated, or osteoporosis patients with eGFR <30 mL/min/1.73 m² or dialysis receiving denosumab.
• Baseline variables: corrected calcium, ionized calcium when available, PTH, 25-OH vitamin D, phosphate, magnesium, alkaline phosphatase or bone-turnover markers, supplementation status, dialysis status, loop diuretic use.

Primary endpoint: severe hypocalcemia within 30 days (for example corrected calcium <7.5 mg/dL, symptomatic hypocalcemia, QT prolongation, emergency visit, or admission).

• Compare discrimination/calibration of: (1) corrected-calcium-alone model, (2) corrected calcium + CKD stage, (3) full mineral-context model including PTH and vitamin D status.
• Prespecified metrics: AUROC, calibration slope, Brier score, decision-curve analysis.

**Falsificati

Hypothesis: An ancestry-calibrated Bayesian prescribing model that starts with population-specific HLA-B*58:01 pretest risk and updates with eGFR, starting dose, thiazide exposure, and prior rash history will lower false-positive allopurinol avoidance compared with non-quantitative precaution rules, while preserving sensitivity for severe hypersensitivity events.

Why this matters: clinicians often face a tradeoff between under-screening high-risk patients and over-avoiding first-line urate-lowering therapy in lower-risk settings. A calibrated Bayesian framework may handle that tradeoff better than binary precaution logic.

Testable prediction: across external validation cohorts with heterogeneous ancestry structure, the Bayesian model will preserve >=95% sensitivity for adjudicated SCAR/AHS while improving specificity and decision-curve net benefit versus genotype-alone or ancestry-alone precaution strategies.

Suggested design: retrospective-development / prospective-validation study with stratified external validation, reporting calibration slope, calibration-in-the-large, AUC, NRI, and clinical utility under multiple treatment-threshold assumptions.

Falsification: if the Bayesian model does not improve specificity or net benefit at matched high sensitivity, the hypothesis fails.

References: FitzGerald JD et al. Arthritis Rheumatol. 2020;72(6):879-895. DOI:10.1002/art.41247. Goncalo M et al. Br J Dermatol. 2013;169(3):660-665. DOI:10.1111/bjd.12389. Lu CY et al. J Formos Med Assoc. 2019;118(1 Pt 1):206-214. DOI:10.1016/j.jfma.2018.06.006.

Hypothesis: In adults starting allopurinol, a pre-treatment model that combines HLA-B*58:01 status, eGFR, starting dose, and thiazide exposure will discriminate severe allopurinol hypersensitivity risk better than genotype-only screening.

Why this matters: catastrophic reactions are rare, but bedside failures usually occur when modest risks stack rather than when one variable acts alone.

Testable prediction: in a prospective multicenter new-user cohort, the composite model will improve 90-day SCAR/AHS discrimination versus HLA-B*58:01 alone by >=0.05 AUC and improve net reclassification for clinically avoidable starts.

Suggested design: prospective or target-trial-emulation cohort of incident allopurinol users with ancestry, genotype, renal function, starting dose, co-medications, and adjudicated 90-day SCAR/AHS outcomes. Compare calibration, Brier score, AUC, decision curves, and net benefit.

Falsification: if genotype-only screening matches or exceeds composite-model discrimination and calibration, the hypothesis fails.

References: FitzGerald JD et al. Arthritis Rheumatol. 2020;72(6):879-895. DOI:10.1002/art.41247. Goncalo M et al. Br J Dermatol. 2013;169(3):660-665. DOI:10.1111/bjd.12389. Lu CY et al. J Formos Med Assoc. 2019;118(1 Pt 1):206-214. DOI:10.1016/j.jfma.2018.06.006.

TMP-SMX hyperkalemia risk is usually framed as a baseline-risk problem, but I hypothesize that time-updated potassium slope plus exposure stacking (TMP-SMX + ACEi/ARB + spironolactone/eplerenone, with CKD context) will outperform baseline potassium alone for early detection of clinically meaningful hyperkalemia.

Claim

In prospective autoimmune or immunosuppressed cohorts receiving TMP-SMX, a model that includes:

• baseline eGFR,
• baseline potassium,
• age,
• ACEi/ARB exposure,
• mineralocorticoid receptor antagonist exposure,
• and delta-potassium within the first 48-96 hours

will show better discrimination and earlier warning for potassium >=5.5 mmol/L or hyperkalemia-related treatment interruption than a baseline-only model.

Why this is plausible

Trimethoprim has an amiloride-like effect on distal tubular potassium handling. The mechanism is static, but the clinical expression is dynamic. A patient with only moderate baseline risk may become dangerous quickly once stacked medications and early potassium drift are visible.

I hypothesize that homomorphically encrypted or privacy-preserving sufficient-statistic pooling can preserve clinically useful external validation of TMP-SMX hyperkalemia models across decentralized autoimmune cohorts, without exposing patient-level potassium or medication histories.

Claim

If each site shares encrypted aggregates or sufficient statistics for:

• baseline potassium distribution,
• eGFR strata,
• ACEi/ARB exposure,
• spironolactone/eplerenone exposure,
• treatment-dose versus prophylaxis-only TMP-SMX,
• and observed hyperkalemia events,

then a pooled validation workflow can recover transportability signals close to those from centralized analysis for calibration slope, calibration-in-the-large, and subgroup performance.

Why this matters

Autoimmune cohorts are fragmented across hospitals, private clinics, and research networks. Raw laboratory sharing is often blocked by privacy, governance, and cross-border constraints. If encrypted aggregation preserves model validation quality, decentralized drug-safety science becomes much more feasible.

Patients with autoimmune disease often accumulate transient kidney stressors during pain treatment: NSAID exposure, ACE inhibitor or ARB therapy, diuretics, intermittent dehydration, and older age. I hypothesize that a time-updated model using short-interval creatinine slope plus explicit triple-whammy exposure (NSAID + ACEi/ARB + diuretic) will predict clinically meaningful NSAID-associated AKI better than baseline eGFR alone.

Why this is plausible

• Baseline CKD increases susceptibility, but AKI risk is strongly context-dependent rather than static.
• The classic 'triple-whammy' combination has already shown measurable AKI excess in population data.
• A slope-based signal should detect evolving renal hemodynamic injury earlier than a single baseline measure.

Testable design

• Prospective multicenter cohort of adults with autoimmune or inflammatory musculoskeletal disease who start or intensify NSAID therapy.
• Collect baseline eGFR, albuminuria, age, diabetes, ACEi/ARB use, diuretic use, heart failure/cirrhosis, hydration or sepsis flags, and repeat creatinine within 3-7 days and 10-14 days.

Primary endpoint: KDIGO-defined AKI within 14 days.

• Compare discrimination and calibration of:
  1. baseline eGFR alone
  2. baseline multivariable model

Rare but high-severity drug toxicities such as allopurinol SCAR are hard to validate because no single center sees enough events, yet raw patient-level data sharing is often blocked by privacy, governance, and cross-border constraints. I hypothesize that federated validation using homomorphically encrypted pooled sufficient statistics will retain clinically useful calibration and discrimination for HLA-B*58:01- and CKD-aware allopurinol safety models while avoiding exchange of identifiable patient-level records.

Why this matters

• Allopurinol remains first-line therapy, but HLA-B*58:01 and CKD create nontrivial heterogeneity in catastrophic adverse-event risk.
• Rare-event pharmacogenomic safety work is exactly where decentralized validation is needed.
• Privacy-preserving analytics could make multi-site validation feasible without centralizing sensitive records.

Testable design

• Multicenter decentralized gout and autoimmune registry network.
• Each site computes encrypted sufficient statistics for predefined variables: HLA-B*58:01, eGFR category, starting dose, diuretic exposure, early systemic warning signs, and SCAR outcome.
• Central aggregator receives only encrypted statistics and estimates logistic or Bayesian rare-event models.
• Compare performance of privacy-preserving pooled estimation against conventional pooled analysis in a shadow dataset where lawful full-data pooling is possible.

Falsifiable predictions

1. Calibration slope difference between encrypted pooled estimation and conventional pooled analysis will be less than 0.05.
2. AUROC loss from

Colchicine neuromyopathy is usually recognized late, after weakness is already clinically obvious or CK has risen substantially. I hypothesize that in colchicine-exposed patients with eGFR <45 mL/min/1.73 m², a time-updated model combining proximal-weakness onset, weekly CK slope, and concurrent statin exposure will outperform any single absolute CK cutoff for early detection of clinically important colchicine neuromyotoxicity.

Why this is plausible

• Colchicine toxicity is amplified by impaired renal clearance and prolonged exposure.
• Case-series literature repeatedly describes a syndrome of proximal weakness, CK elevation, and neuropathic findings, but the temporal pattern is under-validated.
• CK alone is noisy: some patients present before major CK elevation, while others have competing causes of hyperCKemia.

Testable prediction

In a prospective autoimmune/gout cohort receiving colchicine, a joint model with: (1) new proximal weakness, (2) CK slope over 7-14 days, (3) eGFR category, and (4) statin co-exposure will show higher AUROC and better net benefit for early neuromyotoxicity detection than any rule based on absolute CK alone.

Proposed validation design

• Population: adults exposed to colchicine for gout, CPPD, pericarditis, or autoinflammatory disease.
• Outcome: adjudicated colchicine neuromyotoxicity requiring drug discontinuation.
• Analysis: time-updated logistic/survival model with internal bootstrap validation and decision-curve analysis.
• Falsification: if CK-threshold-only rules perform equivalently or better, the hypothesis fails.