Ward simulator: severe hyponatraemia
Admit a fictional patient, write orders, move the clock forward and draw sodium levels. The model follows total exchangeable Na+K over total body water, with urine output set by ADH. It is a teaching model, not a clinical tool.
Runs the toolkit's standard 48-hour protocols on a copy of this patient: rapid intermittent bolus (RIB) or slow continuous infusion (SCI), sodium checked every 6 h, re-lowering once the 24-h limit is reached.
Brain colour tracks the true serum sodium: blue and swollen when low, normal at 135 to 145. The pons glows red once the rise passes the 24-hour limit (osmotic demyelination risk). Kidney droplets show urine flow; fast drops mean a water diuresis.
Drug development: from clinical gap to approval
Eighteen stages take a fictional drug, VH-417, from a real gap in type 2 diabetes (muscle insulin resistance from poor GLUT4 translocation) through the lab, randomised animal studies in rats and dogs, FDA, ethics, bias-control tools, four trial phases, the paper, the approval and a viva, ending in a downloadable clearance report. Every number is synthetic; every rule and reference is real.
A new seed redraws every synthetic dataset. The built-in truth stays the same, so you see how chance moves results around it.
Microbiology bench: culture and identify
Take a patient sample, streak it across four agar plates with the loop, incubate, pick a single colony, Gram stain it, run bedside tests and name the organism. Selective media stop some organisms growing; differential media change colour depending on what the organism ferments or lyses.
Streak quadrant 1 heavily, flame the loop, then drag each new quadrant once through the end of the last one. The loop carries fewer bacteria each time, so quadrant 4 should give single, separate colonies. After incubation, tap a colony to pick it.
Virtual laboratory
Each analyte is the patient's true value plus analytic noise, with reference ranges, H/L and critical flags, turnaround times and the occasional haemolysed sample. Order panels for the ward patient, or build a work-up case of your own.
Results appear when the ward clock passes the reporting time. Advance the clock on the ward tab.
Look at the hyponatraemia work-up, then name the cause.
Clinical study database
Six synthetic datasets from the virtual hospital: the hyponatraemia cohort, the VH-BOLUS trial and the four portfolio studies. Browse records, profile variables and compare groups. Each dataset has a known built-in truth, so you can check whether an analysis finds it.
Crude comparison with Wilson 95% CIs; risk ratio and difference are against the first group. Crude is not causal: the answer key shows where confounding misleads.
Trial toolkit
The command-line trial tools as live calculators: sample size and power, concealed randomisation lists, group-sequential stopping boundaries, and a virtual trial that runs real simulated patients through both arms.
Defaults: SALSA observed overcorrection 24.2% vs 17.2% (Baek 2021). The toolkit's evidence file rounds this to 24% vs 17% and gets 525 per group.
In a real trial this list is held by someone outside the team and released one allocation at a time. Random block sizes stop recruiters from guessing the next arm.
Lan-DeMets alpha spending, exact recursive integration. Check: three equal looks with OBF-type spending give 3.39, 2.41, 2.02, as in the toolkit.
Generates new simulated patients, randomises them in permuted blocks, runs each through the ward model under its arm for 48 hours and analyses overcorrection (rise above 12 mmol/L in 24 h or 18 mmol/L in 48 h, the SALSA definition).
Change the seed and rerun: each run is one possible trial. The spread between runs is what a confidence interval tries to describe.
3D body and cells
Zoom from the whole body to tissue and single cells. Each scene runs a small physiological model, so the cells respond to ADH, serum sodium, oxygen and drugs. Drag to rotate, scroll or pinch to zoom.
3D data explorer
Rotate any three variables of a dataset in space, coloured by a fourth, and see how power changes with sample size and effect size as a surface.
Up to 3,000 points are drawn. Drag to rotate, scroll or pinch to zoom, hover for the record.
Each point is the analytic power of a two-proportion test. The flat 80% contour shows every pair of sample size and treatment risk that meets the usual target.