A tight interval on a thin base
In March 2024 Kazakhstan moved the entire country to UTC+5, folding 30 degrees of longitude under one administrative clock. Fariza Zharykbassova and Oxana Tsigengagel combined roughly 19 million observations from the national clinical registry with a stratified random survey of 13,648 people. They report that in the first six months cardiovascular consultations in the eastern regions rose 11.4 per cent against the western comparison regions, with a confidence interval running from 9.2 per cent to 13.6 per cent.[1]
A tight interval does not measure the strength of an inference. That interval comes from millions of patient-level observations, while the variation that identifies the effect sits between 15 eastern and 5 western administrative units — twenty clusters in all. The authors read it exactly that way: against the bias a small comparison group creates, they apply cluster-level resampling by the Cameron–Gelbach–Miller method and report a p-value of 0.024. Another reading stays open: if the variation between clusters really is small, the narrow interval is not overstated. Only the clusters themselves can show that; the patient count cannot.[1]
A year later the number came back
The same study also reports that the effect did not last. Monitoring through early 2025 shows consultation rates converging on the modelled baseline, with no significant deviation from expectation in the late period: the p-value is 0.38. In the survey, residents at 82 degrees east had 4.21 times higher odds of reporting sleep disturbance — but that is a separate class of evidence, self-reported and cross-sectional. The authors call the result descriptive evidence and write themselves that they remain limited to observational inference about physiological mechanism.[1]
Another population number that entered circulation the same day makes the difference visible. In Colorado, type 2 diabetes incidence among 10–19-year-olds is reported to have risen from 2.37 per 100,000 in 2002 to 22.30 per 100,000 in 2023, with the annual increase accelerating from 2.92 per cent over 2002–2013 to 4.72 per cent over 2014–2023. The magnitude is striking and the data source is a long-running registry. The difference lies here: the Kazakhstan study's design, clusters and adjustments are written down in a peer-reviewed paper, while the Colorado figures come for now from an abstract to be presented in Milan at the end of September. Two claims of similar size, and not a similar amount of information to test them with.[1], [2]
What to watch next
This study has a testable side. If Kazakhstan's national registry publishes the series past 2025, whether the late-period convergence holds can be seen directly. And the comparison group being limited to five units can be tested with a sensitivity analysis that drops each region in turn; if the result scatters as regions come out, the base really is thin. Until one of the two happens, the 11.4 per cent remains a finding waiting to be repeated.[1]