Clinical analytics · used by doctors
One patient. Every blood, urine and stool marker they have ever had, mapped against researched reference ranges, known disease clusters and the trends running the wrong way, across nineteen clinical specialties, on every visit.
The job is to see what one person cannot.
It reports what is present, what is genuinely ruled out, and what has never been looked at. It holds pharmacology, botanicals and nutrition at the same time, reads the record longitudinally rather than one panel at a time, and every claim it makes carries a source you can open.
What it reports
NeXXum runs 305 cited condition patterns against the whole record, every draw rather than just the panel in front of it, and reports what it finds with the values, the dates and the source behind each one.
Including the negatives. Where the deciding criteria are declared, it also says which conditions were genuinely ruled out, which rule-outs are resting on a value too old to trust, and which were never checked at all. A documented negative is a clinical finding in its own right, and it is the one thing a lab report never gives you.
What it reads
A haematologist reads the blood count. A hepatologist reads the liver. NeXXum reads all of it at once, each result against every other, on every visit. That is where the findings that only appear in combination live.
Functional medicine
Read the whole person, over time, look upstream, intervene early. That is the premise, and NeXXum is built to carry it. Every range it applies is sourced, every supplement it names is graded, every connection it draws states the finding it rests on.
Same ambition, with the working shown. Each claim arrives with its threshold, its source and what would rule it out.
Longitudinal
Most software reads the panel in front of it. NeXXum reads every draw a patient has ever had: the direction, the rate, the moment a value started drifting while still inside the reference range, and whether two numbers were even measured close enough together to be compared.
That is where early detection actually lives: not in a red flag, but in a slope nobody plotted and a constellation nobody assembled.
Borderline high · 0.23 mmol/L above the 5.2 healthy maximum (4% of the normal span).
Reference: ≤ 5.2 mmol/L · ≤ 200 mg/dL (US)
Range source: Canada reference range (researched)
Borderline high · 0.01 g/L above the 1.05 healthy maximum (1% of the normal span).
Reference: ≤ 1.05 g/L
Range source: Canada reference range (researched)
Borderline high · 0.49 mmol/L above the 3.4 healthy maximum (14% of the normal span).
Reference: ≤ 3.4 mmol/L · ≤ 130 mg/dL (US)
Range source: Canada reference range (researched)
Borderline high · 0.04 mmol/L above the 4.2 healthy maximum (1% of the normal span).
Reference: ≤ 4.2 mmol/L · ≤ 130 mg/dL (US)
Range source: Canada reference range (researched)
A real record, published with the consent of the person whose it is · four years, drawn to scale. Each of these is a percent or two over on its own. Read together, across every draw, they are the finding.
Unedited, from the assessment generated for that same record on 24 August 2026. It is written to one clinician by another, the reasoning is shown, the numbers carry their dates, and the sources are named so they can be opened and checked.
The three things he has been most worried about are, on close reading, the three most reassuring: the “iron retention” is a single borderline transferrin saturation sitting on top of a normal ferritin; the “UTI” was adjudicated as a contaminated specimen; and the “gut inflammation” rests on one stale zonulin of uncertain validity.
The signals that genuinely deserve attention are quieter: a persistent atherogenic particle burden he is under-treating because he is under-dosing the tools already prescribed; a fasting glucose climbing steadily across four years; a PSA that is both pharmacologically suppressed and drifting upward with a low free-to-total ratio; and, added since this panel, three self-started injectable peptides whose safety profile intersects with the prostate question.
This is sound and I agree with it on the evidence, not just deferentially: nitrite plus leukocyte esterase in an asymptomatic non-pregnant adult does not meet a threshold to treat, and treating asymptomatic bacteriuria drives resistance without benefit (Nicolle LE et al., IDSA 2019). Urine blood, protein and glucose were all negative on 2026-07-28, which is consistent with contamination rather than infection. Nothing to do.
Where the knowledge came from
The library was built from guideline bodies, classification criteria and primary papers, each one recorded against the entry it supports. This is the short version; the app carries 112 reading sources and a citation on every claim it makes.
What that is, and what it isn’t. Nineteen fields, none of them thinner than a dozen conditions, every one carrying its source. That is not the depth a haematologist has in blood or a hepatologist has in liver, and it does not replace either. It is the breadth of all nineteen held at the same time, on every patient, on every visit, which is the part no one person does.
Evidence
Not a model’s recollection. Guidelines, meta-analyses and primary papers, each one resolved against PubMed and the answer written down: author and year against what we wrote, in a ledger that ships with the code. A citation that fails is kept and marked, not quietly dropped, because a row that vanishes and a row that was checked look identical afterwards.
That check exists because it caught things. Twelve citations named one paper while their link pointed at another. That is the hardest kind to see, since the sentence reads correctly. One cited a phase 2 trial registration for a myopathy risk the trial never studied.
Keeping it current means checking where the answer changes first. PubMed shows what has been published; the trial registry shows what is underway. Three compounds this library graded “no published human evidence” turned out to have phase 2 trials recruiting: true of the literature, and out of date about the world. Every one of those entries now carries the registration and the date the registry was checked, so the claim to be current is a falsifiable one.
Interactions
The patient is already on medication before anyone recommends anything. So every drug entry leads with what that drug does to a lab result, then what it interacts with, including the botanicals and nutrients on the same shelf, which is the half most software does not hold at all.
A 5-alpha-reductase inhibitor HALVES the PSA, so the number must be doubled before it is read. An ACE inhibitor raising creatinine up to 30% is the drug working, not kidney injury. High-dose biotin is an assay artifact that can falsely lower a troponin.
St John’s wort induces the enzymes that clear a contraceptive, a transplant drug and an anticonvulsant. Calcium, magnesium, iron and zinc bind a fluoroquinolone into treatment failure. DIM lowers endoxifen in a woman on tamoxifen.
And it can subtract. Reading that one list lets it say what should come off: an over-the-counter product whose active constituent is the prescription drug beside it, several agents stacking one hazard, the same nutrient arriving from two bottles, a dose above the licensed maximum. A patient on eighteen things is the common case, and almost no tool will tell a clinician to remove one.
How it cannot invent
A language model will produce a plausible number. In a clinical document a plausible number is the whole problem, and a disclaimer does not fix it because nobody reads disclaimers. So the two jobs are separated: deciding what is true, and saying it well.
A report is de-identified on the clinician’s own machine, and she is shown exactly what was removed before anything is sent. On confirmation the server does not trust that result: it goes back to the stored original, re-extracts and re-de-identifies independently, and discards what the browser produced. If an identifier survives, the send is refused outright rather than flagged. A scanned report is read by an OCR service inside Canada, so the image itself never crosses a border. Only anonymous text does.
Limits
Written down, because a tool that only lists its strengths is asking to be taken on faith.