Wednesday, May 1, 2019

Survival analysis or Time-to-Event datasets

Question:
I have a graduate student wondering if there are any microdata sets from Statistics Canada that might be useful for survival analysis. The topic or focus of the data is not important, so long as the dataset would lend itself to this type of analysis. He already has datasets from hospital discharges, but is looking for others. Basically the dataset would need to contain microdata (not aggregated) and provide information on the time elapsed before an event happens. Common subjects are disease onset or death, mechanical failure of equipment, etc., but again any topic would work.

I hope I have explained this well enough to be understood (I only grasp it at a basic level myself!). Any help would be greatly appreciated.

Answer:
I’ve received quite a lengthy response from subject matter that will hopefully help some:

“To clarify, by “time-to-event”, is the student looking for longitudinal data? For example, are they looking for data points regarding what happens in between two things (i.e. what happens to my health from the point when I am diagnosed with a  disease to when I die?)

To my knowledge, there are no longitudinal Health Public Use Microdata Files on any topic (deaths, disease onset, et cetera).  The National Population Health Survey (or NPHS) was longitudinal, but, by the very nature of longitudinal data, it does not lend well to PUMF analysis. As such, we only offer NPHS data for reference years 1994 – 1995, 1996 – 1997, and 1998 – 1999 and there is no longitudinal element to the PUMF. I have attached the NPHS PUMF questionnaire here in case the student might see some data points of interest. We also have PUMFs for the Canadian Community Health Survey (CCHS). These offer a greater range of reference periods – dating back to 2000 and in two-year intervals up to and including 2016. I have attached that questionnaire as well. If these options might work - we can offer the access for free.

Normally, to acquire data like this, some researchers apply for record linkages. This enables them to link one thing (hospital records) to another thing (death certificates) in order to make meaning about how the two records might work together regarding outcomes. However, this is not a dataset and requires an application, approvals, and is usually done on a record by record basis. I have attached a copy of the application form for record linkage to this message just in case, though I honestly do not think this is what would best meet the needs of the student. 

CIHI (https://www.cihi.ca/en/access-data-and-reports ) might also be able to offer some information relating to patient experiences, hospital stays, health outcomes, emergency care, et cetera.”