Paperonium

Public health · Int J Epidemiol · 2023

Both main results fall outside their ranges: homeless, 4.0 times the housed death rate, range 2.0–3.3.

Summary and table: homelessoutside its rangeSummary and table: at riskoutside its rangeIts own numbers: homelessif swapped: inside 3.4–4.8Its own numbers: at riskif swapped: inside 2.0–3.34.0×2.0 to 3.3should be inside2.6×3.4 to 4.8should be inside1×1.5×2×3×5×same death rate as housed patients
Summary andtable: homelessoutside its rangeSummary andtable: at riskoutside its rangeIts own numbers:homelessif swapped:inside 3.4–4.8Its own numbers:at riskif swapped:inside 2.0–3.34.0×2.0 to 3.3should be inside2.6×3.4 to 4.8should be inside1×2×5×same death rate ashoused patients

How many times the housed patients’ death rate, each result with its range

The paper says

Among 6,290 people seen at an emergency department in 2003/04, homeless ones died over the next 16 years at 4.0 times the rate of housed ones (range 2.0–3.3); those at risk of homelessness (housed, but insecurely) at 2.6 times (range 3.4–4.8).

Its own data show

A result must sit inside its own range. Neither does, in the opening summary, the results text or the main table: 4.0 is above 2.0–3.3; 2.6 is below 3.4–4.8. Each sits mid-way in the other group’s range, so the two ranges look swapped. The table’s 14 other results, for single causes of death, all fit.

A range, calculated from the data, stretches either side of the result to show how far off it might be. Taken alone, the printed ranges make being at risk look deadlier than being homeless, the opposite of the results. Only the authors can confirm the swap; until they correct it, anyone reusing these ranges gets the wrong ones.

“The summary” means the paper’s abstract: the short summary printed at its top, which is the part most readers see.

Read the paper →Full technical write-up →

Premature mortality 16 years after emergency department presentation among homeless and at risk of homelessness adults: a retrospective longitudinal cohort study · Int J Epidemiol (2023) · doi:10.1093/ije/dyad006