Healthcare Design Systems
The dashboard work behind these: the standard, the system, and the build.
Surgical dressing utilization
Six Sankey studies tracing surgical cases from the dressing used, through patient type, to whether a complication was recorded. Four procedure families, one template, and one decision about missing data that the whole set turns on.
Every case count, complication code and source on these slides is masked. The structure is the work; the figures are not mine to publish.
The six Sankeys
Whether advanced dressings change what happens after surgery
Two categories of advanced wound dressing were in question: silver dressings on orthopedic cases, and negative pressure wound therapy kits on cardiac and skin cases. Both cost more than a standard dressing. The analysis asked how often each was used, and whether cases that used one recorded fewer complications than cases that did not.
The underlying table was a case-level extract with four usable columns: the dressing category, whether the case was inpatient or outpatient, whether a complication was flagged, and which complication codes applied. Complication data existed for inpatient cases only, which turned out to be the fact the whole set had to be built around.
Three versions of the knee chart before the pattern was fixed
Total knee replacement was drawn first and used to settle the format. The first version squared every flow into blocks and put the cohort total in a circle at the left edge. It exposed the problem that shaped everything after it: most of the cases were outpatient and carried no complication data, so the largest rectangle on the slide was also the one with the least to say.
The second version replaced the blocks with curved bands and dropped the circle. Following an individual flow got easier, but the space recovered went into the written insight in the left column instead of into the diagram.
The third kept the curves, filled the outpatient block with diagonal hatching labeled in place, and ranked the complication codes down the right-hand edge. That version became the template.
Two hues, and everything else carried by structure
Color encodes one variable only. Teal marks cases with no dressing and purple marks cases with a dressing or kit, and that pairing holds across all six slides. Later stages use lighter tints of whichever hue the case started in, so an arm can be followed left to right without a legend.
Missing data is drawn rather than omitted. Outpatient cases had no complication records, so those blocks are hatched and labeled in place. Dropping them would have made the inpatient share of each cohort look larger than it was.
Small values were the recurring problem. A handful of cases against a cohort of thousands has no drawable thickness, so those flows leave the diagram as thin curves and terminate in a figure and a label at full size, rather than being rounded away.
One template across four procedure families
The knee chart went through three versions and the third became the pattern. Hip took it without a single change. CABG and skin diverged only where the data forced it, so a reader who learned the first diagram could read all six without starting over. That is the value of settling a format before scaling it.
The outpatient hatching is the decision I would defend hardest. The easy move was to drop the cases with no complication records and let the diagram look cleaner. Drawing them as hatched and labeled keeps the inpatient share honest, and it puts the actual finding on the slide: for most of these procedures, the complication question can only be asked of a minority of the cases.