By default speedtest only reliably measures last mile speed, since it defaults to the nearest server it can find. That data has its uses, but it doesn't tell the whole story, and is often misleading. Backbone, stability and practical latency tend to be much bigger issues, which are not represented here at all (last mile latency is represented but not even accounted for in the ranking, nor is upload bandwidth).

You could make a much better index using speedtest by taking several measurements for each ISP using near and far nodes, and using weighted averages accounting for upload, download and latency.