Move text over from Trello board minutes

[lwjohnst86]

Copied over from Trello @Aastedet

Agenda:

• Status on implementations V 1.0: Validated on HICD-2019 data V 1.1: In HICD & DARTER project: Supports LPR3 and ICD-8 (DARTER) codes for inclusion No longer inclusion on semaglutid, dapagliflozin or empagliflozin. DARTER: “E:/workdata/708421/workspaces/diabetes_register_pop/dm_population_1977_2022.rds" T1D: 47.947 T2D: 639.181 Issues for discussion: How to handle glucose-lowering drugs in treatment of non-diabetes DARTER: no “NAME” variable. Handled by VNR-variable instead. How to incorporate ICD-8 codes Current implementation: Insulin-dependent = T1D Insulin-independent = T2D ICD-8 codes only use to classify T1D from T2D in individuals without ICD-10 codes. Medication data 1995-96 (pregnancy register to censor GDM?) Remove extra step for cross-sectional study populations. DARTER: No PATTYPE variable Using the algorithm Interface (script/package), data format (parquet, native R), use-cases for studies Other business & discussion

Summary:

Discussion on major change from v. 1.0 to 1.1:

Ignore purchases of semaglutid, dapagliflozin and empagliflozin for inclusion
    These drugs appear to be a major source of inclusion of cases without any other inclusion events from 2021 onward (likely non-diabetes).

No objections to minor changes from v. 1.0 to 1.1:

Removed reliance on/use of PATTYPE (patient type) variable
Now uses VNR (varenummer) instead of NAME (commercial drug name) variable if not available
Removed reclassification of T1D to T2D in cross-sectional populations based on insulin purchases one year prior to index date (1.7% of T1D cases, majority of reclassified individuals self-reported T1D in HICD-2019).

Discussion on diabetes duration prior to medication data availability:

Diabetes onset prior to medication data is unlikely to be accurately captured by other sources.
    Diabetes onset is considered valid is cases with onset one year after availability of medication data or later.
    Onset of T1D may to some extent be captured by hospital diagnoses. If necessary for a given study design, diabetes duration in T1D cases onset prior to medication data can be used, but is not generally recommended.
    The algorithm will provide two dates on diabetes onset:
        A raw date of onset:
        The second occurance of an inclusion event (the current implementation)
    A valid date of onset:
        Same as the raw date of onset, but in cases with onset prior to one year after availability of medication data, the date of onset is defined as NA/missing.
        Users can choose to exclude these individuals from the study population.

Availability of medication data:

Limited to 1997 onward if GDM censoring is based on patient register data.
Can be extended to 1995 onward if GDM censoring is based on medical birth register.
    This will be implemented in DARTER.

Use of ICD-8 diabetes diagnosis codes:

Insulin-dependent/independent diabetes (249xx/250xx) is considered ICD-8 equivalents of the ICD-10 codes for type-1/type-2 diabetes (DE10/DE11)
ICD-8 codes are used for inclusion/raw date of diabetes onset, but are only used to classify T1D from T2D in individuals with at least 2 ICD-8 diagnoses of diabetes who also have no type-specific ICD-10 diabetes diagnoses.
    A potential edge case arises in individuals with ICD-10 diagnoses who died/emigrated prior to medication data becoming available (potentially 1994 through 1996). These cases will be classified as T2D due to no data on insulin purchases.
    These edge cases could be accounted for by treating ICD-10 codes prior to availability of medication data as ICD-8 codes, so individuals can be classified as T1D purely based on hospital diagnosis data.

Further work:

Explore if DARTER project holds biomarker data to allow the algorithm to potentially identify LADA cases (e.g. C-peptide)
Be explicit about limitations and recommended use in the algorithm's documentation.
Allow users to generate a more detailed table for debugging/exploration of cases (is already generated by the algorithm behind the scenes).