You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Copy file name to clipboardExpand all lines: NEWS.md
+92-2Lines changed: 92 additions & 2 deletions
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -1,3 +1,93 @@
1
+
CohortMethod 6.0.0
2
+
==================
3
+
4
+
This major release introduces many changes. The three most important ones are (1) changing the settings objects, (2) supporting nesting cohorts, and (3) covariate balance significance testing.
5
+
6
+
7
+
## Changes related to the settings objects
8
+
9
+
1. All settings objects have been changed to R6 classes, and are now used both when calling functions individually and when using `runCmAnalyses()`. The main rationale is that allows 3rd parties to more easily generate valid settings.
10
+
11
+
2. Dropped the `cdmVersion` argument in `getDbCohortMethodData()` and `runCmAnalyses()`. The version will be identified in the `cdm_source` table.
12
+
13
+
3. Dropped the `trimByIptw()` and `trimByPsToEquipoise()` functions. Added `equipoiseBounds` and `maxWeight` arguments to `createTrimByPsArgs()` so functionality remains.
14
+
15
+
4. Dropped the `matchOnPsAndCovariates()` function and added a `stratificationCovariateIds` argument to `createMatchOnPsArgs()` so functionality remains.
16
+
17
+
5. Dropped the `stratifyByPsAndCovariates()` function and added a `stratificationCovariateIds` argument to `createStratifyByPsPsArgs()` so functionality remains.
18
+
19
+
6. Renamed `createStudyPopArgs` argument of `createCmAnalysis()` to `createStudyPopulationArgs` for consistency.
20
+
21
+
7. Dropping the deprecated `attritionFractionThreshold` argument of `createCmDiagnosticThresholds()`. The amount of attrition is not a good measure of generalizability. Use the generalizability diagnostic instead, which measures the similarity between the target and analytic cohort characteristcs.
22
+
23
+
8. Changed the default outcome model type from 'logistic' to 'cox'.
24
+
25
+
9. Set the defaults of `createGetDbCohortMethodDataArgs()` to those most often used.
26
+
27
+
10. Dropped the `firstExposureOnly`, `restrictToCommonPeriod`, `washoutPeriod`, and `removeDuplicateSubjects` arguments from `CreateStudyPopulationArgs`. These were duplicated from `getDbCohortMethodData()`, and we'll keep them only there from now on.
28
+
29
+
30
+
## Changes related to nesting cohorts
31
+
32
+
11. Added ability to restrict to a nesting cohort (e.g. restricting drug exposures to a specific indication). See the `nestingCohortId` argument in the `createGetDbCohortMethodDataArgs()` and `createTargetComparatorOutcomes()` functions and the `nestingCohortDatabaseSchema` and `nestingCohortTable` arguments in the `getDbCohortMethodData()` function.
33
+
34
+
12. The results schema now includes the `target_comparator` table that combines the `target_id`, `comparator_id`, and `nesting_cohort_id` into a single unique `target_comparator_id`. This new ID is a hash of its components, allowing results from multiple runs to be combined into a single database.
35
+
36
+
13. In addition to restricting to a nesting cohort the population can now also be restricted by age and gender using the `minAge`, `maxAge`, and `genderConceptIds` arguments of `createGetDbCohortMethodDataArgs()`.
37
+
38
+
39
+
## Changes related to the new covariate balance diagnostic
40
+
41
+
14. Added optional significance testing to covariate balance. This avoids failing the balance diagnostic on smaller databases just because of random chance, and was found to be superior in our methods research. This introduces the following changes to the interface:
42
+
43
+
- Added the `threshold` and `alpha` arguments to the `createComputeCovariateBalanceArgs()` function. These do not impact blinding when running `runCmAnalyses` but do add columns to the balance files, for when running single studies.
44
+
- Added the `sdmAlpha` argument to the `createCmDiagnosticThresholds()` function.
45
+
46
+
This adds the `sdm_family_wise_min_p` and `shared_sdm_family_wise_min_p` fields to the `cm_diagnostics_summary` table when exporting to CSV.
47
+
For now, the default is not to use significance testing, but the family-wise min P can help understand if one would have passed when using it.
48
+
49
+
50
+
## Other important changes
51
+
52
+
15. Added a new option for the `removeDuplicateSubjects` argument: "keep first, truncate to second". This is similar to "keep first", but also truncates the first exposure to stop the day before the second starts.
53
+
54
+
16. Now performing empirical calibration *after* removing estimates that fail diagnostics. In general this should lead to narrower calibrated confidence intervals.
55
+
56
+
17. If high correlation is detected when fitting a propensity model, but `stopOnError = FALSE`, the export will show the highly correlated covariates in the model with extreme coefficients (1e6 * correlation).
57
+
58
+
18. Added the ability to use bootstrap for computing confidence intervals. See the `bootstrapCi` and `bootstrapReplicates` arguments of `createFitOutcomeModelArgs()`.
59
+
60
+
19. All restrictions on the study populations performed by `getDbCohortMethodData()` are now step-by-step recorded in the attrition table.
61
+
62
+
20. Completely updated of all unit tests to increase coverage of functional tests, while also increasing speed.
63
+
64
+
21. Renamed the `showEquipoiseLabel` argument of `plotPs()` to `showEquipoiseLabel`.
65
+
66
+
67
+
## Minor changes
68
+
69
+
22. Added support for grid-with-gradient likelihood profiles. Use the following arguments in `createFitOutcomeModelArgs()` to use:
0 commit comments