I’ve noticed that in the characterization tab in Atlas, it says “In addition, covariates during a period may be stratified into temporal units of time for time-series analysis such as fixed intervals of time relative to cohort_start_date (e.g. every 7 days, every 30 days etc.)” and I wonder how to make this happen.
For context, I’m interested in monitoring the the glycemic control of diabetic patients in my cohort and would like to know the control rate of diabetes every month. What I did so far is in the “Incidence rate analysis” tab, I stratified the analysis by the time of observation, but it has only a few options (as attached).
Hi @Shanshan4Q33
This is really interesting! I never thought of this potential in ATLAS. I will give it a try and let you know if I find anything. Please, post here if you figure it out in the meantime. Maybe creating subgroups by observation period?
Hi, @Shanshan4Q33 . I apologize for the confusion, and this is something that slipped my attention, but there’s a disconnect between that statement in the UI and what Atlas functionality actually is doing. For clarity, this is the complete statement:
Cohort characterization is defined as the process of generating cohort level descriptive summary statistics from person level covariate data. Summary statistics of these person level covariates may be count, mean, sd, var, min, max, median, range, and quantiles. In addition, covariates during a period may be stratified into temporal units of time for time-series analysis such as fixed intervals of time relative to cohort_start_date (e.g. every 7 days, every 30 days etc.), or in absolute calendar intervals such as calendar-week, calendar-month, calendar-quarter, calendar-year.
The background of that Characterization description is more high-level about what Cohort Characterization means in OHDSI, and references functionality that is avaialble out of HADES specifically FeatureExtraction. It probably came directly from the Book of OHDSI.
However, in Atlas, in some cases we have to limit the exposed functionality of these Underlying HADES packages in order to simplify the UI and make it more approachable. In the case of custom time-series analysis, we didn’t expose any ways to specify those parameters. There are other ‘defaults’ that we specify such as the Short/Medium/Long term type of analyses (ConditionOccurrenceShortTerm vs. ConditionOccurrenceLongTerm) have hard coded 30, 180, 365d prior lookbacks to make less complicated to specify (and make these settings consistent across Cohort Characterization results).
So, to answer your question, we should adjust the messaging in Atlas to reflect the functionality that is available. That would change the above statement into something like:
Cohort characterization is defined as the process of generating cohort level descriptive summary statistics from person level covariate data. Summary statistics of these person level covariates may be count, mean, sd, var, min, max, median, range, and quantiles. These covariates are calculated standard analyses from FeatureExtraction or using custom covariates from SQL or ‘criteria features’. In addition, Target populations can be sub grouped using criteria expressions.
With this, we can get the cumulative numbers of diabetic patients, cumulative and new cases of patients identified as poor control, and the corresponding prevalence, time at risk and rate per person year.
Hi Chris - thanks for the correction! I found ‘createTemporalCovariateSettings’ in FeatureExtraction maybe helpful! But I’m currently only able to use Atlas to do the analysis. I might wanna try this function if I can connect to the data server via JDBC protocol in the local environment.
