Joining contact gaps lowered the expected patient return rate under a two-hour count model in two releases of one hospital record

We tested how contact episode rules change a patient return comparison in two public releases from one Lyon hospital ward. We kept each staff member's patient episode counts fixed within two-hour blocks and assigned patient labels in random order to fixed episode slots. Eligible transitions had no overlapping patient episode and a gap greater than 5 and at most 30 minutes. In the original release, the observed same-patient rate was 26.56% and the exact expected rate was 26.13% for 512 transitions. Joining same-pair contacts across gaps of up to 300 seconds gave 28.01% observed and 13.20% expected for 457 transitions. On the same 457 transitions, the observed count stayed at 128 while the expected count fell from 118.67 to 60.33. A second release gave a similar sensitivity. Block length also changed the sign of the excess. The primary comparison was close to its reference. These results show sensitivity of a fixed-slot reference, not a validated model of staff behaviour. Episode labels help define the slots, and arbitrary label exchange need not produce valid reconstructed episodes. We provide source audits, all planned results, code, figures, and repeat instructions. The record has one ward, no health outcomes, and less than five days of activity.

temporal networks · hospital contacts · patient returns · episode construction · reference models · reproducible analysis