Algorithm

General description

The current implementation of the algorithm is concisely described below. A more complete description of the original implementation of the classifier is found in Anders Aasted Isaksen’s PhD Thesis as well as the validation paper.(1)

In brief, the algorithm first identifies a diabetes population—regardless of diabetes type— based on four types of inclusion events. It then classifies individuals within this population as having type 1 diabetes (T1D) based on patterns in glucose-lowering drug purchases and recorded type-specific diabetes diagnoses. Individuals who do not meet the criteria for T1D are subsequently classified as having type 2 diabetes (T2D).

Inclusion events are:

  1. HbA1c measurements of ≥48 mmol/mol.
  2. Hospital diagnoses of diabetes.
  3. Diabetes-specific services received at podiatrist.
  4. Purchase of glucose-lowering drugs.

Some inclusion events are ignored if any of the following exclusion events apply:

  1. Elevated HbA1c samples are dropped if:
  2. Purchases of glucose-lowering drugs are dropped if:

The diabetes classification date is defined at the date of the second occurrence of any of the inclusion events listed above. For example, an individual with two elevated HbA1c tests followed by a glucose-lowering drug purchase is included at the latest elevated HbA1c test. Had the second HbA1c test not been performed (or had it returned a result below the diagnostic threshold), this person would instead have been included at the date of the first purchase of glucose-lowering drugs. Wherever possible, all available data for each event is used, except for the purchases of glucose-lowering drugs, since the patient register data on obstetric diagnoses necessary to censor glucose-lowering drug purchases is only complete from 1997 onwards.

High-level flowchart

Flow of inclusions and exclusions used for classifying diabetes status using the osdc package. Light blue and orange boxes represent inclusions and exclusions, respectively.

Classifying type 1 diabetes

Diabetes type is classified as either T1D or T2D based on patterns of purchases of insulin drugs (including analogues) and hospital primary diagnoses of T1D and T2D.

For an individual to be classified as having T1D, the available data needs to fulfill the criteria in either of the following branches of logic:

  1. Must only have recorded purchases of insulin drugs and never any other type of glucose-lowering drugs, and have at least one primary diagnosis of T1D from a medical specialty hospital department.
  2. Must have a majority of T1D diagnoses from endocrinology departments (or from other medical specialty departments, in the absence of contacts to endocrinology departments), and a purchase of insulin within 180 days of their initial glucose-lowering drug purchase, with insulin contributing at least two thirds of all defined daily doses of glucose-lowering drugs purchased.

Classifying type 2 diabetes

Individuals not classified as T1D cases are classified as T2D cases.

Detailed and technical description

Below are the technical, exact implementation of the above description. These are the logical conditions and exact variables and registers used that results in the classification of diabetes status and types. They are shown in the form of tables for each register with a table at the end showing the logic that is applied across registers.

lpr_diag

name title logic comments
lpr2_is_needed_code LPR2 codes used throughout the algorithm c_diag =~ ‘^(DO0[0-6]|DO8[0-4]|DZ3[37]|DE1[0-4]|249|250)’ AND (c_diagtype == ‘A’ OR c_diagtype == ‘B’) A c_diagtype of'A' means primary diagnosis, while ‘B’ means secondary diagnosis.
lpr2_is_diabetes_code LPR2 diagnoses codes for any diabetes c_diag =~ ‘^(DE1[0-4]|249|250)’
lpr2_is_t1d_code LPR2 diagnoses codes for T1D c_diag =~ ‘^(DE10|249)’
lpr2_is_t2d_code LPR2 diagnoses codes for T2D c_diag =~ ‘^(DE11|250)’
lpr2_is_pregnancy_code LPR2 diagnoses codes for pregnancy-related outcomes c_diag =~ ‘^(DO0[0-6]|DO8[0-4]|DZ3[37])’ These are recorded pregnancy endings like live births and miscarriages.
lpr2_is_primary_diagnosis LPR2 primary diagnosis c_diagtype == ‘A’

lpr_adm

name title logic comments
lpr2_is_endocrinology_dept LPR2 endocrinology department c_spec == 8L TRUE when the department where the recorded diagnosis was endocrinology.
lpr2_is_medical_dept LPR2 other medical department c_spec %in% c(1L:7L, 9L:30L) TRUE when the diagnosis was recorded at a medical department other than endocrinology.

diagnoser

name title logic comments
lpr3_is_needed_code LPR3 codes used throughout the algorithm diagnosekode =~ ‘^(DO0[0-6]|DO8[0-4]|DZ3[37]|DE1[0-4])’ AND (diagnosetype == ‘A’ OR diagnosetype == ‘B’) AND (senere_afkraeftet == ‘Nej’) A diagnosekode means primary diagnosis and senere_afkraeftet means diagnosis was later retracted.
lpr3_is_primary_diagnosis LPR3 primary diagnosis diagnosetype == ‘A’
lpr3_is_t1d_code LPR3 diagnoses codes for T1D diagnosekode =~ ‘^(DE10)’
lpr3_is_t2d_code LPR3 diagnoses codes for T2D diagnosekode =~ ‘^(DE11)’
lpr3_is_diabetes_code LPR3 diagnoses codes for diabetes diagnosekode =~ ‘^DE1[0-4]’ This is a general diabetes code, not specific to T1D or T2D.
lpr3_is_pregnancy_code ICD-10 diagnoses codes for pregnancy-related outcomes diagnosekode =~ ‘^(DO0[0-6]|DO8[0-4]|DZ3[37])’ These are recorded pregnancy endings like live births and miscarriages.

kontakter

name title logic comments
lpr3_is_endocrinology_dept LPR3 endocrinology department hovedspeciale_ans == ‘medicinsk endokrinologi’ TRUE when the department is endocrinology.
lpr3_is_medical_dept LPR3 medical department hovedspeciale_ans %in% c(‘blandet medicin og kirurgi’, ‘intern medicin’, ‘geriatri’, ‘hepatologi’, ‘hæmatologi’, ‘infektionsmedicin’, ‘kardiologi’, ‘medicinsk allergologi’, ‘medicinsk gastroenterologi’, ‘medicinsk lungesygdomme’, ‘nefrologi’, ‘reumatologi’, ‘palliativ medicin’, ‘akut medicin’, ‘dermato-venerologi’, ‘neurologi’, ‘onkologi’, ‘fysiurgi’, ‘tropemedicin’) TRUE when the department is other medical departments (than endocrinology).

lab_forsker

name title logic comments
is_hba1c_over_threshold HbA1c values over threshold (analysiscode == ‘NPU27300’ AND value >= 48) OR (analysiscode == ‘NPU03835’ AND value >= 6.5) Is the IFCC units for NPU27300 and DCCT units for NPU03835.

lmdb

name title logic comments
is_gld_code ATC codes for glucose-lowering drugs (GLDs) atc =~ ‘^A10’ AND NOT (atc =~ ‘^(A10BJ|A10BK01|A10BK03)’) GLP-RAs or dapagliflozin/empagliflozin drugs are not kept.
is_insulin_gld_code Only insulin glucose-lowering drugs atc =~ ‘^A10A’ AND NOT (atc =~ ‘^A10AE56’) This is used during the classification of type 1 diabetes to identify persons who only purchase insulin or mostly purchase insulin.

Across register logic

name title logic comments
is_within_pregnancy_interval Events that are within a potential pregnancy interval has_pregnancy_event AND date >= (pregnancy_event_date - weeks(40L)) AND date <= (pregnancy_event_date + weeks(12L)) The potential pregnancy interval is defined as 40 weeks before and 12 weeks after the pregnancy event date (birth or miscarriage).
is_podiatrist_services Podiatrist services speciale =~ ‘^54’ AND barnmak == 0 When barnmak == 0, the PNR belongs to the recipient of the service. When barnmak == 1, the PNR belongs to the child of the individual.
is_not_metformin_for_pcos Metformin purchases that aren’t potentially for the treatment of PCOS NOT (koen == 2 AND atc =~ ‘^A10BA02$’ AND (date < (foed_dato + years(40)) OR indication_code %in% c(‘0000092’, ‘0000276’, ‘0000781’))) Woman is defined as 2 in koen.
has_t1d Classifying type 1 diabetes status (from_diabetes_diagnosis OR from_podiatrist_service OR from_gld_purchase OR from_hba1c_over_threshold) AND (has_only_insulin_purchases & has_any_t1d_primary_diagnosis) OR (!has_only_insulin_purchases & has_majority_t1d_diagnoses & has_two_thirds_insulin & has_insulin_purchases_within_180_days) The final classification for type 1 diabetes. Depends on all the previous steps to create these intermediate logical variables.
has_any_t1d_primary_diagnosis Any primary diagnosis for type 1 diabetes (n_t1d_endocrinology + n_t1d_medical) >= 1 This is used to classify type 1 diabetes. Naturally, having any type 1 diabetes diagnosis is indicative of type 1 diabetes.
has_majority_t1d_diagnoses Determine if the majority of diagnoses are for type 1 diabetes if_else(n_t1d_endocrinology + n_t2d_endocrinology > 0, n_t1d_endocrinology > n_t2d_endocrinology, n_t1d_medical > n_t2d_medical) This is used to classify type 1 diabetes. Endocrinology diagnoses are prioritised if available, otherwise other medical department diagnoses are used. If no diabetes type-specific primary diagnoses are available from an endocrinology or other medical departments, this variable is returned as FALSE.
has_two_thirds_insulin Whether two-thirds of GLD doses are insulin doses (n_insulin_doses / n_gld_doses) >= 2/3 This is used to classify type 1 diabetes. If multiple types of GLD are purchased, this indicates if at least two-thirds are insulin, which is important to determine type 1 diabetes status.
has_only_insulin_purchases Whether only insulin was purchased as a GLD n_insulin_doses >= 1 & n_insulin_doses == n_gld_doses This is used to classify type 1 diabetes. If only insulin is purchased, this is a strong reason to suspect type 1 diabetes.
has_insulin_purchases_within_180_days Whether any insulin was purchased within 180 days of the first purchase of GLD any(is_insulin_gld_code & date <= (first_gld_date + days(180L)), na.rm = TRUE) This is used to classify type 1 diabetes. It determines if any insulin was bought shortly after first buying any type of GLD, which suggests type 1 diabetes.

References

1.
Isaksen AA, Sandbæk A, Bjerg L. Validation of register-based diabetes classifiers in danish data. Clinical Epidemiology. 2023 May;Volume 15:569–81.